WORKI NG PAPER
No. 83
SOE
PL 2009
– An Estimated
Dynamic Stochastic General
Equilibrium Model for Policy
Analysis And Forecasting
Grzegorz Grabek
Bohdan Kłos
Grzegorz Koloch
Warsaw 2011
N a t i o n a l B a n k o f P o l a n d 2
Design:
Oliwka s.c.
Layout and print:
NBP Printshop
Published by:
National Bank of Poland
Education and Publishing Department
00919 Warszawa, 11/21 Świętokrzyska Street
phone: +48 22 653 23 35, fax +48 22 653 13 21
© Copyright by the National Bank of Poland, 2011
http://www.nbp.pl
GRZEGORZ GRABEK, BOHDAN KŁOS, GRZEGORZ KOLOCH, SOE
PL−2009
— An Estimated Dynamic
Stochastic General Equilibrium Model for Policy Analysis and Forecasting , Bureau of Applied
Research, Economic Institute, National Bank of Poland, Warsaw, 2010
(Document compilation: 860.14.2.2011)
Kontakt:
BBohdan.Klos@mail.nbp.pl T(0 4822) 653 15 87
BGrzegorz.Grabek@mail.nbp.pl T(0 4822) 585 41 08
BGrzegorz.Koloch@mail.nbp.pl T(0 4822) 653 21 79
The paper presents the personal opinions of the authors and does not necessarily reﬂect the ofﬁcial position
of the institution with the authors cooperate (have cooperated) or to which they are (or have been)
afﬁliated.
An earlier version of the paper has been published in polish under the title: SOE
PL−2009
— Model DSGE
małej otwartej gospodarki estymowany na danych polskich. Specyﬁkacja, oceny parametrów, zastosowania,
Materiały i Studia NBP, 2010.
Typesetting L
A
T
E
X (T
E
X, format: LaTeX2e, ver. 2009/09/24), pdfT
E
X, ver. 1.40.11 (MikT
E
X 2.9)
L
A
T
E
X2
, pdfT
E
X, BIBT
E
X (etc.) implementations, fonts, packages of macros and the utilities available in the
MikT
E
X collection have been used. Detailed information about the licence conditions, software copyrights,
legal protection of the stored collection, legally protected trademarks (etc.) are provided on the Internet
Web pages of the MikT
E
X http://www.miktex.org and on the site of T
E
X system at http://www.ctan.org.
GRZEGORZ GRABEK, BOHDAN KŁOS, GRZEGORZ KOLOCH, SOE
PL−2009
— An Estimated Dynamic
Stochastic General Equilibrium Model for Policy Analysis and Forecasting , Bureau of Applied
Research, Economic Institute, National Bank of Poland, Warsaw, 2010
(Document compilation: 860.14.2.2011)
Kontakt:
BBohdan.Klos@mail.nbp.pl T(0 4822) 653 15 87
BGrzegorz.Grabek@mail.nbp.pl T(0 4822) 585 41 08
BGrzegorz.Koloch@mail.nbp.pl T(0 4822) 653 21 79
The paper presents the personal opinions of the authors and does not necessarily reﬂect the ofﬁcial position
of the institution with the authors cooperate (have cooperated) or to which they are (or have been)
afﬁliated.
An earlier version of the paper has been published in polish under the title: SOE
PL−2009
— Model DSGE
małej otwartej gospodarki estymowany na danych polskich. Specyﬁkacja, oceny parametrów, zastosowania,
Materiały i Studia NBP, 2010.
Typesetting L
A
T
E
X (T
E
X, format: LaTeX2e, ver. 2009/09/24), pdfT
E
X, ver. 1.40.11 (MikT
E
X 2.9)
L
A
T
E
X2
, pdfT
E
X, BIBT
E
X (etc.) implementations, fonts, packages of macros and the utilities available in the
MikT
E
X collection have been used. Detailed information about the licence conditions, software copyrights,
legal protection of the stored collection, legally protected trademarks (etc.) are provided on the Internet
Web pages of the MikT
E
X http://www.miktex.org and on the site of T
E
X system at http://www.ctan.org.
GRZEGORZ GRABEK, BOHDAN KłOS, GRZEGORZ KOLOCH, SOE
PL—2009
— An Estimated Dynamic
Stochastic General Equilibrium Model for Policy Analysis and Forecasting, Bureau of Applied
Research, Economic Institute, National Bank of Poland, Warsaw, 2010
(Document compilation: 860.14.2.2011)
Contents
WORKING PAPER No. 83 3
Contents
Contents
Introduction 6
I Genesis and anatomy of dynamic stochastic general equilibriummod
els 10
1 Genesis and anatomy of dynamic stochastic general equilibrium models 11
1.1 Methods of the Cowles Commission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.1 Problems with identiﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 LSE and VAR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.1 LSE methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.2 VAR methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Methodology of modern macroeconomics . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Real Business Cycle theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.2 Frictionless economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.3 Representative household . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.4 Representative ﬁrm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.5 General equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.6 Consequences for monetary policy . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4 New neoclassical synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.1 Standard newKeynesian model . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.2 Representative household . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.3 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.4 General equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.5 Consequences for monetary policy . . . . . . . . . . . . . . . . . . . . . . . . 28
2 DSGE model — anatomy 31
2.1 Values of parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1.1 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.2 Maximum likelihood estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.3 Bayesian estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2 Kalman ﬁlter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3 MetropolisHastings algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.4 Model selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3
Contents
N a t i o n a l B a n k o f P o l a n d 4
2.5.1 Structural shocks identiﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5.2 Impulse response analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5.3 Variance decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5.4 Unconditional forecasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
II Speciﬁcation of DSGE SOE
PL−2009
model 47
3 SOE
PL−2009
— general outline 48
3.1 SOE
Euro
model — prototype of SOE
PL
family models . . . . . . . . . . . . . . . . . 48
3.2 Family of SOE
PL
models, SOE
PL−2009
version . . . . . . . . . . . . . . . . . . . . . 50
3.3 Basic features of SOE
PL−2009
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4 Decisionmaking problems, equilibrium conditions, macroeconomic balance of the
model 56
4.1 Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2 Foreign economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3 Producers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3.1 Aggregators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.2 Domestic intermediate goods ﬁrms . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3.3 Importers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3.4 Exporters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4 Households . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.5 Behaviour of other agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.5.1 Central bank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5.2 Government . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 Macroeconomic balance conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.6.1 Proﬁts in economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.6.2 Income and expenditures of households . . . . . . . . . . . . . . . . . . . . . 73
4.6.3 State budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.6.4 Monetary balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.6.5 Balance of payment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.6.6 The aggregate resource constraint . . . . . . . . . . . . . . . . . . . . . . . . 77
III Results of estimation and characteristic features of the DSGE SOE
PL−2009
model 78
5 Forms of model, data, SVAR models 79
5.1 Forms of model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 Observable variables, data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.3 SVAR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3.1 Fiscal SVAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3.2 World’s economy SVAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6 Assessment of parameters — calibration, optimisation, steady state 85
6.1 Calibration of parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.2 Prior distributions and results of estimation . . . . . . . . . . . . . . . . . . . . . . . 87
6.3 Assessment of parameters — conclusions . . . . . . . . . . . . . . . . . . . . . . . . 90
6.3.1 Steady state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.3.2 Structural changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.3.3 Nominal rigidities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4
Contents
WORKING PAPER No. 83 5
7 Variance decompositions, impulse response functions and estimation of distur
bances 95
7.1 Variance decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.1.1 Regime I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.1.2 Regime II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.2 Impulse response functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.3 Smoothing — estimation of structural disturbances . . . . . . . . . . . . . . . . . . 118
8 Historical decompositions and forecasts 121
8.1 Historical decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.2 Forecasting technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.3 Expost forecasting accuracy of the SOE
PL−2009
model . . . . . . . . . . . . . . . . 126
8.3.1 Onedimensional measures of forecasting quality . . . . . . . . . . . . . . . 127
8.3.2 Multidimensional measures of forecasting accuracy . . . . . . . . . . . . . 128
8.3.3 Rolling forecasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Final comments 132
IV Appendix 134
A List of equations, list of variables 135
Forms of the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
List of model variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
List of equations of the structural form of the model . . . . . . . . . . . . . . . . . . . . . 139
Steady state solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
B Global economy SVAR model 148
SVAR model identiﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
Results of estimation of the SVAR model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
C Convergence to the steady state 150
D Analysis of forecasts accuracy 151
Bibliogrphy 158
5
Bibliography
Introduction
N a t i o n a l B a n k o f P o l a n d 6
Introduction
Introduction
The paper documents the effects of work on the dynamic stochastic general equilibrium (DSGE)
SOE
PL
model that has been carried out in the recent years at the National Bank of Poland,
initially at the Bureau of Macroeconomic Research and lately at the Bureau of Applied Research
of the Economic Institute. In 2009, a team consisting of the authors of this paper developed a new
version of the model, called SOE
PL−2009
which in 2010 is to be used to obtain routine midterm
forecasts of the inﬂation processes and the economic trends, supporting and supplementing the
traditional structural macroeconometric model and experts’ forecasts applied so far.
In the recent years many researchers have engaged in the work over a class of estimated
macroeconomic models (of the business cycle) integrating the effects of at least three important
lines of economic and econometric research:
• methods of macroeconomic modelling (gradual departure from the traditional structural
models towards models resistant to Lucas’ and Sims’ critique, strongly motivated with
microeconomics);
• micro and macroeconomic theories (monetary policy issues, with emphasis on the con
sequence of imperfect competition, the role of nominal and real rigidities, as well as
anticipating and optimising behaviours of agents in an uncertain environment, with a
strong shift of point of view towards general equilibrium);
• estimation techniques (reduction in parameters calibration, shift from classical techniques
to Bayesian techniques with Bayesianspeciﬁc risk quantiﬁcation as well as systematic and
controlled introduction of experts’ knowledge, improvement of projections accuracy).
Merger of the three trends has brought about a class of models — DSGE models — with high
analytical and developmental potential. The very potential of the models of this class seems to
be the most important reason for the interest of central banks in that area
1
, research that may
be directly translated into the practice of monetary policy.
1
Attention is being drawn to the fact that in the general case the DSGE models do not have to be based on the new
Keynesian perspective of economy and do not have to be estimated with the use of Bayesian techniques.
6
Introduction
WORKING PAPER No. 83 7
Along with the development of numerical, econometric methods and the theory of economics, a
number of central banks supplement or even replace the traditional structural macroeconometric
models, whose forecasting applications are enhanced with experts’ knowledge, with estimated
DSGE models, namely models which attempt to translate the economic processes in a more
explicit and systematic manner, whereby experts’ knowledge is introduced through Bayesian
methods
2
. It happens although no formal reasons exist for which the ex post veriﬁed accuracy of
forecasts within the DSGE models should be higher than that of classical models
3
. DSGE models
give, however, a chance of structural (internally consistent and microfounded) explanations of
the reasons for the recently observed phenomena and their consequences for the future. DSGE
models present a different image of economic processes than classical macroeconometric models
— they capture the world from the perspective of structural disturbances. These disturbances
set the economy in motion and economic agents respond to them in an optimal way, which
eliminates the consequences of the disturbances, i.e. restores the economy to equilibrium. The
analytical knowledge and experience gathered in contact with the traditional structural models
rather interferes with than helps interpret the results of DSGE models. In econometric categories,
the results of DSGE models are, nevertheless, at least partially compliant with that which may
be achieved with VAR and SVAR models, thus, it is hard to speak about revolution here.
Following the events of 2008–2009 (global ﬁnancial crisis), while searching for the reasons for
the problems’ occurrence, the usefulness of formalised tools constructed on a uniform, internally
coherent (but also restrictive) paradigm for macroeconomic policy tends to be questioned. The
reasons for the global economy problems are searched for in models oversimplifying perception
of the world and burdening the decisions regarding economic policy. We have noticed that the
critique refers to a larger extent to the models as such (i.e. tools) and less to the practice of
applying them (i.e. the user). Therefore, we consider that conclusions from a deeper analysis of
the sources of 2008–2009 crisis, veriﬁcation of the directions of economic research and methods
of the research, which is likely to be held, as well as the analysis of the current policy less
inﬂuenced by its rationalisation shall conﬁrm the legitimacy of building and applying models,
particularly DSGE class models. The issue of applications using the strong sides of the models
remains, however, open. In our opinion, the best we can do is to try to use our model, gather
and exchange experience, develop new procedures and thoroughly verify the results.
The model whose details we shall present further herein derives from the structure developed at
Riksbank — DSGE model for the euro area
4
see Adolfson et al. (2005b). The euro area DSGE
model, knowhow, methods of estimation and applications received within the technical support
of Riksbank enabled us to start several experiments, build different versions of DSGE model
(a family of SOE
PL
models) and develop our own procedures of the model application. Some
of the experiments have been described in separate papers, e.g. Grabek et al. (2007), Grabek
2
For complete image, we wish to add that there are also arguments against the engagement of central banks in
constructing DSGE models, see e.g. Orphanides (2007).
3
The issue of correct measuring of accuracy of forecasts, in which the contribution of experts’ knowledge is
considerable but nonformalised or systematic, has been omitted here. In such a situation it is hard to assess whether it
is the model that failed or the expert. Generally, it may be argued whether the forecasting applications of DSGE models
expose the strongest sides of this class of models.
4
The euro zone model by Riksbank develops the ideas mentioned among others by Christiano et al. (2001, 2003,
2005), Altig et al. (2004a), as well as Smets and Wouters (2004).
7
Introduction
N a t i o n a l B a n k o f P o l a n d 8
and Kłos (2009), Grabek and UtzigLenarczyk (2009). The alternative we present in this paper
summarizes some of the gathered experience.
We pass the DSGE SOE
PL−2009
model for use, with a view to considering and analysing other
interpretation and understanding of economic processes than that proposed by the traditional
models. Additionally, systematic work with the model (preparing forecasts and analyses of their
accuracy, simulation experiments and analytical works) may reveal issues and problems that
will have to be solved. Resulting knowledge shall enable the preparation of a more thorough
future modiﬁcation of the model, taking into account the effects of the parallel research and the
conclusions arrived at during use.
This paper consists of three basic parts. In the ﬁrst part — relatively independent of the other
parts — we have made an attempt to outline the development of the methods of macroeconomic
(macroeconometric) modelling and the economic thought related to monetary policy, which
brought about the creation of dynamic stochastic general equilibrium models, pushing aside
other classes of models — at least in the academic world. The considerations are illustrated with
simple models of real business cycles (RBC) and DSGE model based on new Keynesian paradigm.
The second chapter of the ﬁrst part focuses on the technical aspects of construction, estimation
and application of DSGE models, drawing attention to mathematical, statistical and numerical
instruments. Although it presents only the keynotes, outlines and ideas, the formalisation and
precision of presentation required in that case makes the fragment of the paper slightly hermetic
— a reader less interested in the techniques may omit that chapter.
The further parts of the paper refer to speciﬁcation, results of estimations and properties of the
DSGE SOE
PL−2009
model. We present, therefore, a general nontechnical outline of the basic
features of the model, illustrating at the same time the correlations with other DSGE models
(Chapter 3). The next chapter deﬁnes decisionmaking problems of the optimising agents, their
equilibrium conditions as well as characteristics of behaviours of the non optimising agents. The
description of the model speciﬁcation is completed with balance conditions on a macro scale.
The SOE
PL−2009
model has been estimated with the use of Bayesian techniques. Identically as in
all estimated DSGE models we are aware of, the Bayesian estimation refers solely to some of the
parameters (the rest of the parameters have been calibrated). Although due to the application
of the Bayesian techniques, the number of calibrated parameters has been clearly reduced,
being aware of the consequences of faulty calibration we conducted a sort of sensitivity analysis
(examination of the inﬂuence of changes in the calibration of parameters on the characteristics
of the model). The presented SOE
PL−2009
version takes into account the conclusions we arrived
at based on the analysis. For the purposes of this paper and the ﬁrst forecast experiments we
use only point estimates of the parameters reﬂecting the modal value of posterior distribution,
in other words our reasoning omits — hopefully temporarily — the issue of uncertainty of the
parameters. The results of the estimation of parameters and assumptions made at the subsequent
stages of the work (calibrated values, characteristics of prior distributions) have been presented
in Chapter 6.
A synthetic image of the model characteristics has been presented in Chapters 7–8, which
describes the responses of observable variables to structural disturbances taken into account
in the model (i.e. impulse response functions), variances decompositions (formally — forecast
8
Introduction
WORKING PAPER No. 83 9
error decomposition), thanks to which the structure (relative role) of the impact of shocks on the
observable variables may be assessed, estimation (identiﬁcation) of structural disturbances in
the sample, examples of historical decompositions (counterfactual experiments) and information
about the ex post accuracy of forecasts — this is, thus, a typical set of information allowing
understanding the consequences of the assumptions made at the stage of constructing decision
making problems (model speciﬁcation) and choice of parameters.
The Appendix presents structural form equations, equations used to determine value at a steady
state and a list of variables of the SOE
PL−2009
model.
9
Part I
Genesis and anatomy of dynamic
stochastic general equilibrium
models
10
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 11
1
Chapter 1
Genesis and anatomy of dynamic
stochastic general equilibrium
models
The development of economic models for the purpose of monetary policy analysis has been one
of the most exploited research programs within macroeconomics in the last two decades. A lot of
effort has been paid to make attempts to understand the correlations between monetary policy,
inﬂation and business cycle. The research is deemed to produce a sort of consensus as to the
speciﬁcation of the key elements of a model of economy, within which modern macroeconomic
analysis is being carried out, mainly in the aspects important for the monetary policy applied by
central banks. The speciﬁcation has been named a new Keynesian model. The new Keynesian
model is a dynamic, stochastic model of general equilibrium and, thus, a model deriving from
neoclassical trend. The basis for its architecture is a model of real business cycle, on which
Keynesian elements in the form of real and/or nominal frictions
1
are imposed. Such originating
trend of macroeconomic analysis is called a new neoclassical synthesis, see Goodfriend and King
(1997).
1.1 Methods of the Cowles Commission
In the 1940s and 1950s government institutions of the most important economies started to
collect in a systematic way national statistics regarding economic activity. The economists gained
material which helped them specify quantitative models of national economy and analyze them
in empirical tests. Early works over the econometric models of national economies complied with
1
E.g.: monopolistic competition, frictions in the process of adjusting prices, wages, frictions in the ﬁnancial market.
11
1
Genesis and anatomy of dynamic stochastic
general equilibrium models
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 12
1
the paradigms developed during the works carried out by the Cowles Commission. Empirical
macroeconomic analysis was at that time carried out with often large
2
, dynamic, most often
linear, multiequation econometric models. Their speciﬁcation was based mainly on statistical
tests
3
, while the role of economic theory was limited to preparing a list of regressors to be taken
into account in the particular equations. The choice of variables was based mainly on Keynesian
ISLM type models, i.e. on theories which ignored both the supply side of economy and changes in
relative prices. The models were called ”structural”, as they enabled the consideration of, among
other things, feedback nature relationships between variables. Such relations and the resulting
problem of simultaneity or interrelation, important for model estimation techniques, were the
focus of interest of macroeconometrics and the theory of estimation at that time. Therefore, it
is considered that econometrics practiced in the spirit of the Cowles Commission emphasised
the structural aspects. Nevertheless, according to today’s understanding of structurality in
macroeconomics, it may be concluded that structurality, or practically its absence primarily
accounted for the failure of this trend of modelling. Although model equations were aimed
at presenting the dynamics, which reﬂects the decisions made by economic agents, forms of
equations assumed ad hoc failed to comply with any mechanism of individual choice. Dynamics
of each of the variables would be modelled with a single equation. Groups of variables formed
the blocks of a model and each block was researched by a separate group of experts. The
resultant equations were later on combined in a complete model of economy, additionally taking
into account the interactions occurring among the variables within the different blocks. The
constructed models allowed, seemingly, correct quantiﬁcation of the consequences of controlling
variables that depended on the persons making decisions with regard to economic policy. They
were, however, too large to arrive at a general image of the mechanism according to which
the propagation of shocks in an economic system took place. It was also hard to research the
mechanism of system response to a change in control of economic policy in a longer time
perspective. According to the philosophy of the Keynesian trend, the emphasis was put on short
term analysis of economic aggregates dynamics. The economy was out of (partial) equilibrium
for a short period and model simulations answered the question of how to effectively bring the
economy to equilibrium, i.e. how to stabilise it. Thus, macroeconometrics dealt mainly with
the analysis of variables dynamics in a short time with the use of partial equilibrium model. An
example of a model within the discussed class is provided in Klein and Goldberger (1955).
1.1.1 Problems with identiﬁcation
In the terminology derived from the paper of Spanos (1990), the reasons for failure of the
originators of models maintained in the tradition of the Cowles Commission as regards the
application of the models to economic policy analysis may be divided into two groups. These are
problems with structural identiﬁcation and problems with statistical identiﬁcation.
As mentioned by Hendry (1976) or Qin (1993), in the period of works of the Cowles Commission
and the later development of multiequation models maintained in this tradition, economet
2
Consisting of several hundred or, sometimes, even several thousand equations.
3
A statistical identiﬁcation of model is often being mentioned.
12
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 13
1
rics focused to a large extent on the theory of estimation
4
, and a smaller emphasis was put
on the assessment of the quality of models by virtue of diagnosis of statistical speciﬁcation
errors. Structural identiﬁcation was a priority and as a result, often proved ex post, the models
maintained in this convention were not able to sufﬁciently accurately replicate the statistical
properties of the processes the dynamics of which they were supposed to represent. Although
the Cowles Commission was putting emphasis on structural identiﬁcation of the model, the
developed methods are considered, from the perspective of the present day, to be unsatisfactory.
Problems with structural identiﬁcation were more fundamental. They are presented, among
others, in two papers from the 1970s and 1980s: Lucas (1976) and Sims (1980). Lucas (1976)
criticises the status of exogeneity of variables — the controls of economic policy. He points
out that the model of structural identiﬁcation proposed by the Cowles Commission does not
explicitly take into account the expectations of economic agents. Therefore, the parameters of
the mode
5
, which were deemed to be structural
6
, are actually a mixture of structural parameters
and parameters related to the expectations of economic agents and, thus, may not be considered
to be ﬁxed for various economic policy regimes. The estimations of parameters within a model
estimated on data originating from a speciﬁc economic policy regime shall no longer be valid
if the policy regime changes. Therefore, a model estimated within one regime may not be
extrapolated outside of the regime and, in consequence, may not be applied to analyse the
consequences of a change in the regime. Due to instability of parameters, the traditional
structural macromodels, according to Lucas, are worthless for simulation of the effects of
changes in economic policy, which is exactly the purpose for which they were created. Sims
(1980) only supports the comments by Lucas claiming that no variable may be deemed exogenous
in the world of economic agents that anticipate future events (forwardlooking agents) and
whose behaviour is based on intertemporal optimisation. As a result of endogenous economic
policy macroeconomic variables correlate with variables — the controls of policy. By assuming
erroneously that the policy is exogenous, the endogeneity may be falsely interpreted as a causal
relation, and may seem to identify the channel of policy’s impact on economy.
Finally, the stagﬂation of the 1970s and the related failures of economic policy based on the
traditional macroeconometric models disqualiﬁed, in the academic opinion, the approach of the
Cowles Commission. Pesaran and Smith (1995) conclude that the models represented neither
the data, nor the theory, and therefore were ineffective for practical purposes of forecasting and
policy.
1.2 LSE and VAR models
Problems with statistical and structural identiﬁcation of traditional multiequation econometric
models brought about the development of several trends out of which two had the largest impact
4
Important in the works of the Commission and later works were mainly the issues regarding simultaneity, i.e.
interrelation of the modelled phenomena.
5
This speciﬁcally refers to the so called reduced form of the model.
6
Parameters are deemed to be structural or deep if their value does not change under the impact of a change in the
economic policy regime.
13
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 14
1
on the practice of macroeconometrics: the socalled LSE method (London School of Economics),
see Hendry (1995), and SCVAR method (Structural Cointegrated Vector Autoregression), i.e.
structural vector autoregression for cointegrated variables, see Lütkepohl (2008), which currently
— next to the DSGE model — is the basic tool of macroeconomic analysis.
1.2.1 LSE methodology
There may be numerous possible sources of errors in statistical identiﬁcation of a model. The
errors include, but are not limited to, omitting important variables, erroneous dynamic structure
or illegitimate restrictions regarding exogeneity imposed on the variables. The LSE approach is
an attempt to overcome the problems with statistical identiﬁcation. It is based on the socalled
reduction principle. An econometric model is understood as a simpliﬁed representation of
an unknown and unobservable stochastic process, which generated the researched economic
observations. It originates from a dynamic model of possibly general speciﬁcation in a given
class of models, so as to cover possibly many various processes. Further on, the model is reduced
sequentially to the ﬁnal form. The reduction step entails the elimination of a variable or a
group of variables from the model and is made with the use of statistical tests. For the resulting
representation of a process that generates data to be complete, loss of information by virtue of
reduction must be insigniﬁcant from the point of view of the modelled process. A completeness
of the model is conﬁrmed by the statistical properties of the residuals vector. Any deviations from
the Gaussian white noise certify that speciﬁcation is faulty. Thus, the LSE approach emphasises
the correct statistical identiﬁcation of a model but is not an attempt to solve problems related to
structural identiﬁcation.
1.2.2 VAR methodology
Similar problems are reﬂected in the approach based on the analysis of time series with the use
of nonstructural models of vector autoregression (VAR). Nonstructural VAR models, or VAR
models in reduced form, are in fact the generalisation of the LSE approach to vector time series.
They are models expressing endogenous variables by their lagged values. A VAR model of order
K ≥ 1 has the following form:
y
t
=
K
k=1
A
k
y
t−k
+e
t
, or y
t
=Ay
t−1
+e
t
, (1.1)
where y
t
is a vector of endogenous variables in t period, A
k
matrices for k = 1, 2, ..., K are
autoregressive matrices. Another representation is called a cumulative representation and A
matrix is an autoregressive companion matrix. The e
t
process covers shocks controlling the
dynamics of endogenous variables. The e
t
shocks control the dynamics of endogenous variables
such that the variables may be presented as a function of the history of shocks. This is the
socalled moving average representation of the process (y
t
):
14
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 15
1
y
t
= A
t
y
0
+
t−1
k=0
A
k
e
t−k
, (1.2)
where y
0
is the initial value of the process ( y
t
), or — assuming inﬁnite history of the process of
endogenous variables:
y
t
=
∞
k=0
A
k
e
t−k
. (1.3)
The basic purpose of the macroeconomic analysis carried out in accordance with the concept
of a shock impacting the system of endogenous variables as a source of their dynamics is to
identify shocks of structural nature, i.e. independent shocks with explicit and cohesive economic
interpretation. The e
t
shocks may not, however, be structurally interpreted as they do not need
to be independent, and generally are not, which should be expected from shocks of structural
nature. The e
t
shocks have the nature of errors or regression residuals (forecast errors) and
may form linear combinations of the actual structural shocks which determine the dynamics
of endogenous variables. It may, then, generally happen that e
t
= Bε
t
, where ε
t
are structural
shocks, i.e. independent ones. In particular it is assumed that the covariance matrix of ε
t
shocks
is the identity matrix, i.e. (ε
t
) = I. This leads to the approach giving structural interpretation
to VAR models. The approach is called structural vector autoregression, shortly SVAR (from
structural VAR). Next to the advantages of the LSE approach in the form of representation of
a process generating the dynamics of endogenous variables through a model of rich dynamic
structure, namely VAR, the SVAR methodology makes it possible to provide shocks with structural
interpretation.
The SVAR methodology has been developed not only for stationary processes on which the
approach based on the Cowles Commission paradigms must rely, but may be easily generalised
to nonstationary cointegrated processes. Such approach is called structural cointegrated vector
autoregression method, shortly SCVAR, or more often — structural vector error correction model,
shortly SVECM.
SVAR and SVECM models attempt to solve problems regarding statistical identiﬁcation and
some of the problems regarding structural identiﬁcation, which were identiﬁed in the classical
multiequation models. These advantages explain the popularity of VAR class models in empirical
macroeconomics. In order to emphasise that there have been proposed at least partial solutions
not only for statistical identiﬁcation problems but also for structural identiﬁcation, modern
macroeconomic analysis is often referred to as structural macroeconometrics. Despite the
aforementioned advantages, VAR methods are not free of drawbacks. Mainly, VAR models
speciﬁcation lacks any theoretical basis. The relations between the variables are of clearly
statistical nature and even in the case of structural models they may not be referred to any
economic mechanism on which the modelled process is founded. Thus, a VAR model lacks a
theoretical framework, and upon assessment it may bring about, and in practice often brings,
nonintuitive implications in the form of responses to shocks that may hardly be rationalised and
15
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 16
1
noncohesive forecasts. Due to the foregoing disadvantages, macroeconomic analysis, particularly
at central banks, more and more often refers to models having theoretical foundations, such as
DSGE models.
DSGE models  through economic theory underlying their speciﬁcation — constitute another
step forward to solve the problems with structural identiﬁcation. The step is taken at the cost of
statistical quality of the model, however not so signiﬁcant because the DSGE models, and more
speciﬁcally their approximate solutions, are directly related to VAR class models. Speciﬁcally, the
socalled reduced form of DSGE model, namely the form which sets dynamics of endogenous
variables around the equilibrium, is a VAR model. It may be considered as a VAR model on
the parameters of which, i.e. on elements of matrices A and B, a set of restrictions has been
imposed. The restrictions originate in the theory of economics underlying the speciﬁcation of a
DSGE model. They limit the dynamic structure of the model, yet ensure its internal cohesion and
guarantee structural nature of the identiﬁed shocks. The DSGE model speciﬁcation founded on
the theory of economics ensures that restrictions imposed on the dynamic structure of its reduced
form derive from the optimal decisionmaking rules of rational economic agents operating in
an internally cohesive economic reality. The methodologies of DSGE models are, therefore, at
least theoretically, an answer to the problems related to structural identiﬁcation. They also, to a
major extent, reply to the problems related to statistical identiﬁcation, as the solution of a DSGE
model is a VAR class model. In that scope a DSGE model provides, nevertheless, only partial
solutions, as the structural nature of its speciﬁcation imposes strong restrictions on the dynamic
structure of the reduced form. Thus, a DSGE model expresses a tradeoff between the proper
statistical and structural identiﬁcation of macroeconomic processes.
1.3 Methodology of modern macroeconomics
The views with regard to the method of analysing the dynamics of economic aggregates have
signiﬁcantly changed over the last 30 years. The evolution followed from models speciﬁed in the
tradition of the Cowles Commission, through VAR class models, to dynamic stochastic general
equilibrium models based on the theory of economics.
Modern macroeconomics attempts to explain the dynamics of economic aggregates with the use
of models based on the socalled microfoundations. Hence, as opposed to traditional Keynesian
models or multiequation macroeconometric models where the forms of interrelations between
economic variables were assumed ad hoc, the mechanisms shaping the decisionmaking rules of
economic agents are explicitly modelled. Most often four types of agents are distinguished. They
are households, ﬁrms as well as government and central bank. Whereas the decisionmaking
rules of households reﬂect the process of optimisation of welfare, i.e. a discounted stream of the
expected utility, those of ﬁrms result from maximisation of the expected proﬁt. Optimisation
takes place in a stochastic economic environment and in the case of households the structure of
their preferences — the socalled utility function — is also set. The decisions of agents are, thus,
always optimal a priori, i.e. the best of possible taking into account the available information.
Usually, they refer to three types of economic categories: products and services, work or assets,
both physical such as capital and ﬁnancial such as bonds or money. Households decide about
16
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 17
1
consumption, i.e. demand for a product, supply of labour and change in the structure of the
portfolio of ﬁnancial assets, the level of investments, the level of use of the available capital, etc.
Firms decide about product supply and demand for labour
7
. The government determines the
value of public spending, collects taxes, expends transfers and incurs public debt. The central
bank controls the nominal interest rate and/or the supply of money. The decision rules of the
government and the central bank are usually made ad hoc. Beside of the optimisation criterion,
the process of making decisions by economic agents takes into account several categories of
limiting conditions. These are most often budget constraints, initial conditions, equilibrium
conditions and constraints regarding the available technology and information structure in the
economy
8
.
A central issue in the theory of dynamic general equilibrium is the intertemporal dimension
of the decisionmaking process. The decisions of agents consist of intertemporal allocation of
the available resources. Today’s income may be allocated to future consumption and future
income may ﬁnance today’s consumption expenditures. Intertemporal substitution of resources
is possible by virtue of participation in the market of ﬁnancial assets, e.g. purchase or sale of
bonds. The individual decisions are coordinated by the market, which brings about decentralised
allocation of resources.
Formally, economy is described as a dynamic system. It reﬂects a shortterm equilibrium in
at least two meanings. Firstly, in each point of time economy reﬂects a general equilibrium
as understood by Walras. It is, thus, assumed that prices always clear the markets. Secondly,
economic agents make optimal decisions, i.e. ex post they do not make mistakes in a systematic
manner as to the actions undertaken. In that meaning their decisions are called rational. Should
it appear ex post that the decisions of the agents are not the best possible to be taken, this is only
due to information gap, i.e. due to the fact that after the moment of making the decision an event
occurred that could not have been foreseen by the agent, e.g. an unexpected exogenous growth
in productivity took place. The agents build expectations as to the future values of economic
variables through the operator of conditional expected value. In that sense the mechanism of
building expectations by the economic agents is rational. It is, therefore, assumed that agents
know the complete model of economy, namely that they know (the real) principles governing
the world in which they live and the values of all of its parameters. They are also able, based on
the model, to set out optimal decisionmaking rules for all agents and apply them, which would
require in practice the possibility of making a perfect ﬁltration, i.e. a perfect measurement of the
value of all of the variables and shocks impacting the economy in any period. That omnipresent
transparency and rationality draws critique on the part of alternative paradigms of economic
modelling such as multiagent modelling (agentbased computational economics), see Fagiolo
and Roventini (2008).
Even though the economy is — in the above meaning — assumed to remain at all times in short
term equilibrium, in a short period it may be out of longterm equilibrium. Longterm equilibrium
is also called stationary state or steady state. The names are founded on mathematics.
7
The mentioned division into the decisions of households and ﬁrms is conventional, however, characteristic to the
literature of the subject matter.
8
Ex post there are also imposed the so called transversality conditions of noPonzi game type.
17
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 18
1
Longterm equilibrium is a mathematical concept and refers to the model of economy instead
of the economy in itself. There is no direct equivalent in the real world. Economy reﬂects
longterm equilibrium if all of the variables grow from one period to another according to
ﬁxed growth rates
9
. Economy in a steady state may be knocked out from the state. It happens
because stochastic disturbance, i.e. structural shocks or structural innovations impacts the
economy. Examples of structural shocks are technological shocks (increase or drop in total factor
productivity), shock of preferences or contractionary monetary policy. If the effects of shocks
abate, the economy comes back to longterm equilibrium but not necessarily to its preshock
state. This shall happen if the effect of shock is permanent. If economy is knocked out of long
term equilibrium by a shock of temporary effect, it shall asymptotically come back to the same
equilibrium. As in reality structural shocks happen at any moment, the real economy shall never
come to the same steady state. It shall, however, ﬂuctuate in its environment. The steady state
path may be interpreted as the path taken by the economy would it not be for the shocks. A
model implying the comeback of economy to the steady state upon structural shock occurrence
is called a steadystate or stationary model.
Modern macroeconomic analysis decomposes the time series of economic aggregates into two
basic components, namely longterm trend and shortterm cyclic ﬂuctuations around the trend,
i.e. the socalled business cycle. It shall be emphasised that both the trend and the cycle are
statistical ﬁction and have no equivalent in the real world. Both categories are a product of data
ﬁltration and their form depends on the chosen method of ﬁltration. DSGE models, as ones
deriving from the RBC models family, are used in cycle analysis and due to their construction
are not able to answer any question regarding the forming of the trend. This is the subject of the
theory of economic growth, which is introduced to DSGE models in an exogenous manner.
1.3.1 Real Business Cycle theory
Since the beginning of 1980s, when the ﬁrst papers of the type Kydland and Prescott (1982)
appeared, the theory of Real Business Cycle (shortly RBC) has gained the status of a leading
macroeconomic theory with regard to economic ﬂuctuations analysis — a model that may be
called a standard RBC model from today’s perspective, the basic tool of business cycle analysis.
The inﬂuence of the RBC revolution on understanding short and mediumterm economic
ﬂuctuations has two perspectives — both methodical and conceptual.
From the methodological point of view, the theory of Real Business Cycle made the dynamic
stochastic general equilibrium model a basic tool for macroeconomic analysis. Ad hoc assumed
behavioural equations describing economic aggregates gave way to equations of motion derived
from intertemporal solutions of optimisation problems of the economic agents operating in
perfectly competitive and frictionless markets. Ad hoc assumptions made with regard to the
mechanisms of forming expectations by economic agents gave way to rational expectations.
The creators of the RBC trend emphasised as well the importance of quantitative aspects of
macroeconomic analysis, which was reﬂected in the use of the methods of calibration, simulation
and validation of RBC models.
9
The deﬁnition covers the case in which some or all variables are unchanged from period to period.
18
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 19
1
Equally fundamental proved to be the conceptual implications of the RBC theory. Firstly, it
appears from the RBC theory that a business cycle presents an effective, i.e. optimal path of
economic aggregates. In other words, both expansion and contraction result from optimal
reaction of economic agents to exogenous shocks impacting the real sphere of economy, mainly
technological shocks. Thus, cyclic ﬂuctuations — including recessions — do not bring about
ineffective allocation of resources — they are fully optimal and do not result from market
imperfections. The RBC theory implies, therefore, that the stabilisation policy of a government
may change resources’ allocation but only to less efﬁcient one. This clearly contradicted the
standard Keynesian interpretation of recession as a period in which economic resources are used
ineffectively and the end of which may be advanced by stimulation of aggregated demand.
The second implication of the RBC theory is the conclusion that the main reason for economic
ﬂuctuations are technological shocks, which temporarily improve or deteriorate the total factor
productivity in the economy. RBC models try to replicate the ﬂuctuations of products and other
economic aggregates, even if the only shock impacting economy is the shock of productivity.
Such interpretation of economic ﬂuctuations contradicted the traditional view that changes in
productivity contributed to economic growth, which has nothing to do with a business cycle.
The third fundamental conceptual implication of the RBC theory is the limited, or practically
nonexistent, role of monetary factors in economy. A standard RBC model implies the neutrality
of money even in a short period. Hence, the dynamics of real variables in economy, such as
production, consumption or employment, do not depend on monetary policy. Monetary policy
may only affect the nominal values such as the nominal interest rate or nominal supply of
money, namely inﬂation, whose values do not affect the real sphere of economy. The results
contradicted the general opinion that monetary policy impacts the real economy in a short run,
see Friedman and Schwartz (1963) and Christiano et al. (1998). Even though the RBC theory
had a considerable inﬂuence on the method of understanding business cycles, particularly in
the academia, RBC models — due to the results regarding the neutrality of money — were not
accepted by central banks. The central banks continued to focus on classical multiequation
models and more and more often on LSE type approach, particularly on vector autoregression
methods.
Due to the growing evidence of contradictions between the Real Business Cycle theory and
empirical research, as well as rift between its implications and the practice of economic policy,
RBC methods could not be considered satisfactory. On the other hand, RBC methods proposed
solutions to problems that discredited the traditional approach based on paradigms developed by
the Cowles Commission. RBC models are dynamic stochastic general equilibrium models instead
of partial equilibrium models. They have an internally cohesive theoretical structure. Therefore,
even though they are calibrated (today, estimated) based on the data that may be generated
by nonintuitive statistical artefacts, the results of simulation experiments (e.g. forecasts, shock
response analyses) reﬂect internal cohesion. This is not the case with VAR class models, even the
structural ones. The decisionmaking rules of economic agents in an RBC model are structural,
i.e. derived from microfoundations instead o f being assumed ad hoc. The expectations of agents
are rational, while anticipation of future events inﬂuences their today’s behaviour. RBC models
reﬂect, therefore, the achievements of the revolution of rational expectations. RBC models
19
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 20
1
parameters, or at least some of them, are deep parameters. Thus, the models are at least partially
resistant to Lucas’ and Sims’ critique. In other words, the models allow us to correctly analyse
the effects of changes in economic policy regime. All those features indicate that RBC methods
well manage the problems of structural identiﬁcation. Additionally, the reduced form of an RBC
model is a VAR type model
10
, namely a model covering a wide class of stochastic processes.
This feature shows that problems related to statistical identiﬁcation are not ignored in the RBC
methods.
It is clear that RBC methods could not be simply rejected. Thus, attempts of such modiﬁcation of
the Real Business Cycle theory have been made so that the internally cohesive structure of an
RBC model be preserved, their dynamics be related to VAR class models, while the role of money
be increased in a short time. The introduction of the elements of new Keynesian economics to
the classical RBC model proved to be a solution — such derived model is called a new Keynesian
model. We will start with the frictionless monetary model of the economy and then introduce
some frictions which render money nonneutral.
1.3.2 Frictionless economy
A standard frictionless monetary model is presented below
11
. Despite a simple structure, it
presents the most important components of a DSGE model. In the following paragraph the
model is extended with Keynesian elements, which leads to the basic form of a new Keynesian
DSGE model. A standard DSGE economy consists of a representative household, a representative
ﬁrm and a central bank. The presentation of the standard DSGE model in this paragraph has
been founded on the monograph by Gali (2008).
1.3.3 Representative household
In exchange for nominal wage W
t
,a household provides a representative ﬁrm in each of the
periods t = 0, 1, 2, ... with a homogenous labour supply
12
H
s
t
. The labour market is perfectly
competitive. In exchange, a household receives wage of the worth of H
t
W
t
. The wage, together
with the savings B
t−1
in period t −1, is the income of a household D
t
= H
t
W
t
+B
t−1
, which is
divided between consumption C
t
and savings B
t
. A household consumes homogenous goods
(products) manufactured by the ﬁrm, while the unit price of the product is P
t
. Savings are
located in bonds with no risk. A unit price of bonds in period t amounts to Q
t
=
1
1+i
t
, where i
t
is
the nominal interest rate determined by the central bank. One bond purchased in period t at
price Q
t
is worth in period t +1 one monetary unit. The expenditures of a household in period t
are, thus, P
t
C
t
+Q
t
B
t
. Household’s welfare is measured with a utility function U(C
t
, H
t
), while
consumption increases and labour decreases welfare.
10
The reduced form of a DSGE model is a process of VARMA class, perhaps of inﬁnite order, however, in practice the
process is approximated with a ﬁnite order VAR model.
11
The presented version of the model abstracts from investments and government sector. Beside those simpliﬁcations,
the model corresponds to the standard speciﬁcation of an monetary model which builds on the RBC origins. The assumed
simpliﬁcations are nonsigniﬁcant from the point of view of the presented conclusions.
12
Subscript s stands for the supply of labour offered by a household. Subscript d shall stand for demand for labour
reported by a ﬁrm. In general equilibrium H
s
t
= h
d
t
, therefore, subscripts s or d shall be omitted when it must be
emphasised that we mean the labour in general equilibrium.
20
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 21
1
In period t, a representative household must, then, make a decision as to variables {C
t
, H
s
t
, B
t
},
such as to maximise utility and by virtue of spending P
t
C
t
+Q
t
B
t
not to exceed the budget of
the value of W
t
H
s
t
+ B
t−1
. Decisions of a household are not, however, oneperiod problem, i.e.
their consequences are not important only ”here and now”. The decisionmaking process has a
dynamic, multiperiod nature, as today’s decisions affect the value of resources available tomor
row. A household must, then, decide not only about variables {C
t
, H
s
t
, B
t
} for the determined t
but also for all of the t = 0, 1, 2, ... at the same time. In order to know how to do that, in t = 0
period a household solves the problem of maximisation of the discounted stream of the expected
utility:
max
{C
t
,H
s
t
,t=0,1,2,...}
0
∞
t=0
β
t
U(C
t
, H
s
t
), (1.4)
subject to a sequence of budget constraints:
P
t
C
t
+Q
t
B
t
= W
t
H
s
t
+ B
t−1
, (1.5)
where
t
is the operator of the conditional expected value. Problem (1.4–1.5) may be solved by
using Lagrange’s functional in the form:
=
∞
t=0
β
t
[U(C
t
, H
s
t
) −ω
t
(P
t
C
t
+Q
t
B
t
−B
t−1
−W
t
H
s
t
)]
. (1.6)
First order conditions are as follows:
ω
t
=
∂ U(C
t
, H
s
t
)
∂ C
t
1
P
t
, ω
t
=−
∂ U(C
t
, H
s
t
)
∂ H
s
t
1
W
t
oraz ω
t
=
β
Q
t
t
ω
t+1
, (1.7)
and for CRRA
13
type utility function, the decision rules of a household are the following:
C
t
=
Q
t
β
t
{(C
t+1
)
σ
c
Π
t+1
}
1
σ
c
,
W
t
P
t
= (H
s
t
)
σ
h
C
σ
c
t
. (1.8)
System (1.8) is recursive. The ﬁrst equation, i.e. consumption’s law of motion, determines the
value of consumption C
t
depending on the expected consumption C
t+1
and inﬂation Π
t+1
. The
second equation — labour supply equation — determines, for the assumed real wage value,
the supply of labour that needs to be provided by a household in order to generate income
necessary to cover the costs of consumption. The choice of {C
t
, H
s
t
, t = 0, 1, 2, ...} shall maximise
the expected welfare of the household.
13
The constant relative risk aversion utility function is deﬁned as:
U(C
t
, H
t
) =
(C
t
)
1−σ
c
−1
1 −σ
c
−
(H
t
)
1+σ
h
−1
1 −σ
h
.
21
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 22
1
1.3.4 Representative ﬁrm
A representative ﬁrm hires labour supplied by a household H
d
t
in exchange for nominal wage W
t
.
It manufactures a homogenous real product Y
t
with the use of production function Y
t
= A
t
N
t
,
where A
t
is an exogenous process of total factor productivity. The product market is perfectly
competitive. A decisionmaking variable in the case of a ﬁrm is the demand for labour H
d
t
, which
— for the assumed technology level — determines the value of the product. In order to determine
the H
d
t
value, a ﬁrm solves in each period a static problem of proﬁt maximisation:
max
H
d
t
P
t
Y
t
−W
t
H
d
t
,
subject to the technological constraint:
Y
t
= A
t
H
d
t
,
the solution of which implies equating real wage with marginal productivity of labour:
A
t
=
W
t
P
t
,
which is exogenous.
1.3.5 General equilibrium
Upon determination of optimal decision rules of a representative household and a representative
ﬁrm, market clearing conditions are imposed on all markets, i.e. the economy is assumed to
remain in general equilibrium state in every period.
The model in its existing form is nonlinear. Generally, no satisfactory methods exist to solve non
linear DSGE models. Therefore, the equilibrium conditions are superimposed on the simpliﬁed,
loglinear form of the model, which for the presented frictionless model reads as follows
14
:
c
t
=
t
c
t+1
−
1
σ
c
(i
t
−
t
π
t+1
−ρ),
w
t
− p
t
= σ
c
c
t
+σ
h
h
s
t
,
w
t
− p
t
= a
t
,
y
t
= a
t
+h
d
t
,
(1.9)
where ρ = −lnβ.
In the basic model there are three markets — product market (consumption balancing), labour
market and savings (bonds) market. The product market’s clearing condition states that the
14
Small letters denote in this chapter percentage (logarithmic) deviations from the steady state. Exceptions: real and
nominal interest rates, which are expressed in absolute values, as well as real marginal costs and monopolistic markup
which are expressed in logs. Timeless variables denote steady state values.
22
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 23
1
whole product supply y
t
is subject to consumption:
y
t
= c
t
the labour market clears when:
h
d
t
= h
s
t
= h
t
.
Whereas households trade bonds only between one another, the savings market clears automati
cally:
b
t
= 0.
The condition equating real wage with marginal productivity of labour, after loglinearisation
w
t
− p
t
= a
t
, may be read out as the determination of real marginal cost (denoted by mc
t
) or —
equally — monopolistic markup λ
t
= −mc
t
, equal to one (zero for the logarithm):
mc
t
=−λ
t
= w
t
− p
t
−mpn
t
= 0,
where mpn
t
= a
t
stands for marginal labour productivity.
1.3.6 Consequences for monetary policy
Based on the conditions (1.9) and the market clearing conditions, there may be determined the
dynamics of real categories of product y
t
, employment h
t
, real wage w
t
− p
t
and real interest
rate r
t
= i
t
−
t
π
t+1
:
c
t
= y
t
=
σ
h
+1
σ
h
+σ
c
a
t
,
h
t
=
σ
h
σ
c
1 −σ
c
σ
h
+σ
c
a
t
,
w
t
− p
t
= a
t
,
r
t
= ρ +
σ
c
(σ
h
+1)
σ
h
+σ
c
(ρ −1)a
t
,
(1.10)
where technology
15
is controlled with an exogenous stationary process in the form of:
a
t
= ρa
t−1
+ε
t
,
where ε
t
∼ N(0, σ) is a structural shock of labour productivity. In a standard RBC model this is
the only structural shock. The last of the equations above, i.e. the real interest rate equation
r
t
= i
t
−
t
π
t+1
, results from Euler household equation.
It appears then that the dynamics of real variables in a standard frictionless economy model
depends only on the level of technology a
t
16
. Therefore, from Fisher equation: r
t
= i
t
−
t
π
t+1
,
it results that a change in the nominal interest rate is one to one translated into a change in
15
More precisely, the process of total factor productivity (TFP), here — labour productivity.
16
And in the case of real interest rate — on the expected change in technology
t
∆a
t+1
= (1 −ρ)a
t
.
23
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 24
1
inﬂation expectations. The equilibrium dynamics of real categories does not depend on the
nominal interest rate i
t
, as the interest rate is translated only to nominal category — the expected
inﬂation. Monetary policy does not affect the decision rules of ﬁrms and is neutral for the welfare
of households. Consequently, a standard RBC model implies that implementation of monetary
policy, requiring high monetary and institutional expenditures, is nonproductive.
1.4 New neoclassical synthesis
The conclusion regarding the neutrality of money in a standard frictionless monetary model
of the economy was a decisive factor that the models of that class could not raise serious
interest at institutions such as central banks, whose practice and understanding of economic
categories ﬂuctuations contradicted the implications of Real Business Cycle theory. On the other
hand, microfounded models gained the interest of academic communities because — compared
to previous macroeconometric models — they made a signiﬁcant step ahead from the point
of view of the method of economic modelling. They appeared particularly attractive due to
explicit modelling of the decisionmaking process of economic agents and speciﬁcation of the
model based on structural parameters — unchanging in the analysis of alternative scenarios
of economic policy, including monetary policy applications. Therefore, the normal course of
events seems to be the beginning of construction of a new theoretical trend based on a possibly
simple case, which would, however, reﬂect the central elements of the trend. In the case of
the trend that originated as an effect of merger of the RBC theory with the elements of new
Keynesian economics — the trend that is currently called new neoclassical synthesis — the
starting point was the model of general equilibrium of frictionless economy (or distortionfree
economy). Neutrality of money and the primacy of technological shocks in the course of the
cycle contradicted economic practice and the concept that demand factors and monetary shocks
should play more than marginal role in forming the cycle. This leads to many attempts of
developing the originally frictionless monetary models towards speciﬁcations abiding by the
methodical achievements of that trend, however, implying nonneutrality of money in a short
run.
In the 1980s and 1990s the analysis covered various microeconomic mechanisms, mainly the so
called distortions or noneffectiveness, the purpose of which was to bring the general equilibrium
models closer to economic reality. A number of papers proposed methods to consider nominal
rigidities in the dynamic model of general equilibrium. Initially — analogically to new Keynesian
literature — emphasis was being put on rigidities in the process of adjusting prices. The proposed
models in fact extended the standard frictionless framework to the case of ﬁrms operating in
the market of monopolistic competition which are not always able to determine markups or,
equivalently, prices at the optimal level. The choice of distortions in the process of price or
markup determination offered the simplest solution guaranteeing that monetary shocks shall
have real effects and shall reﬂect adequate persistence. In the following paragraph, we present a
standard new Keynesian model based on a standard frictionless engine, emphasising the reasons
for nonneutrality of money.
24
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 25
1
1.4.1 Standard newKeynesian model
Economy consists of a representative household, ﬁrms — no longer a representative ﬁrm — and
central bank. The economic environment in which a representative household is operating is
identical as in the standard frictionless monetary model. Hence, that decisionmaking problem
of a representative household, as well as its decisionmaking functions are the same as in the
standard frictionless monetary model. The central bank implements monetary policy applying
the monetary policy rule, e.g. Taylor rule. The main difference appears in the sector of ﬁrms
that possess monopolistic power, so they are no longer price takers. It is assumed that there is a
continuum of ﬁrms and each of them is represented by a point on [0, 1]
17
interval. They operate
in the market of imperfect competition, usually monopolistic competition. Thus, ﬁrms have
monopolistic power and may determine the prices of their products. Each ﬁrm manufactures a
product different from the products manufactured by other ﬁrms (a variety), while the elasticity
of substitution between products manufactured by various ﬁrms is ﬁnite. The products of the
particular ﬁrms are distinguishable, e.g. based on brand naming.
1.4.2 Representative household
Whilst consumption of a representative household consists of a continuum of products, it faces
the same decisionmaking problem as in the frictionless model. In other words, households face
the same decision rules as in the frictionless model, i.e. after loglinearisation they take the
following form:
c
t
=
t
c
t+1
−
1
σ
c
(i
t
−
t
π
t+1
−ρ),
w
t
− p
t
= σ
c
c
t
+σ
h
h
s
t
.
(1.11)
In period t, a representative household consumes C
t
(i) of product i, while total consumption in
economy originates as a result of averaging the consumption of the particular products with the
use of CES integral aggregator:
C
t
=
[0,1]
C
t
(i)
η−1
η
di
η
η−1
, (1.12)
where η is the elasticity of substitution between any two products in economy. Budget constraints,
then, take the form of:
[0,1]
P
t
(i)C
t
(i)di +Q
t
B
t
= B
t−1
+W
t
H
t
, (1.13)
where P
t
(i) stands for the price of product i determined by ﬁrm i in period t. If we assume that
17
There is no obstacle for the number of products to be ﬁnite. Integral aggregators are, then, replaced with sumbased
aggregators. Yet, in literature the application of integral aggregators is common.
25
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 26
1
prices P
t
(i) aggregate to the average price P
t
in period t according to:
P
t
=
[0,1]
P
t
(i)
1−η
di
1
1−η
(1.14)
the budget constraint aggregates to:
P
t
C
t
+Q
t
B
t
= B
t−1
+W
t
H
t
(1.15)
i.e. to the constraint identical as in the frictionless model. A household decides the share in the
total consumption C
t
of particular products by virtue of maximisation of consumption
C
t
= (
[0,1]
C
t
(i)
η−1
η
di)
η
η−1
for the assumed value of consumption expenditures
[0,1]
P
t
(i)C
t
(i)di, i.e. for each i ∈ [0, 1],
solving the problem:
max
C
t
(i)
C
t
=
[0,1]
C
t
(i)
η−1
η
di
η
η−1
,
subject to
[0,1]
P
t
(i)C
t
(i)di = Z
t
,
(1.16)
where Z
t
is an auxiliary variable determining the value of consumption expenditures. The
solution to this problem leads to the function of demand in the form of:
C
t
(i) =
P
t
(i)
P
t
−η
C
t
. (1.17)
1.4.3 Firms
A ﬁrm i in period t is assumed to manufacture Y
t
(i) of product i with the use of production
function such as that applied by a representative ﬁrm in the frictionless model:
Y
t
(i) = A
t
H
d
t
(i) (1.18)
The level of stationary technology A
t
is common to all ﬁrms.
As ﬁrms have monopolistic power, they can set the prices of their products. In period t a
ﬁrm i determines the price of a product i at the level of P
t
(i), which maximises the expected
discounted ﬂow of its proﬁts. The process of determining prices is subject to frictions. Price
rigidities of the Calvo (1983) type are most frequently applied. Hence, a ﬁrm i in period t may
determine an optimal price of its product P
t
(i), with probability equal to 1 −ξ. With probability
equal to ξ the ﬁrm is forced to leave the price of its product at the previously determined level.
The probability ξ does not depend on time that elapsed from the last period in which the ﬁrm
had the possibility to reoptimise the price of its product. In other words, all the ﬁrms which in
26
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 27
1
period t may reoptimise prices, solve the same optimisation problem, namely determine the
same price P
t
. The price P
t
solves the following problem:
max
P
t
∞
k=0
(βξ)
k
t
{Z
t,t+k
(P
t
Y
t+kt
−Ψ
t+k
(Y
t+kt
))}, (1.19)
where Z
t,t+k
stands for the stochastic discount factor, Y
t+kt
= (
P
t
P
t+k
)
−η
C
t+k
stands for the
households’ demand in period t + k for the output of a ﬁrm that had the latest opportunity
to reoptimise prices in period t, and Ψ
t+k
(Y
t+kt
) = W
t+k
H
t+k
(Y
t+kt
) stands for the nominal
cost of such ﬁrm. Thus, as opposed to the frictionless model, in the new Keynesian model the
mechanism of inﬂation is modelled explicitly. Firms generate inﬂation by determining an optimal
price at the level different from the average price level in the previous period.
1.4.4 General equilibrium
The condition of clearing the product markets makes demand equal to supply in each market:
C
t
(i) = Y
t
(i)
for every i ∈ [0, 1] and entails the clearing of an aggregated product market:
Y
t
= C
t
providing that product aggregation is carried out with the use of aggregator in the following
form:
Y
t
=
[0,1]
Y
t
(i)
η−1
η
di
η
η−1
.
After loglinearisation, the condition takes the form of:
y
t
= c
t
.
Demand for labour is aggregated with the use of H
d
t
=
[0,1]
H
d
(i)di. aggregator. It may be
shown that such aggregation leads to a relation between product, productivity and employment
in the form of
18
:
y
t
= a
t
+h
d
t
.
The condition of labour market clearing requires that the aggregated demand for labour be
equal to supply of labour by a representative household:
h
d
t
= h
s
t
= h
t
18
This dependence requires the elimination of the effects of wage dispersion in economy.
27
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 28
1
The bonds market clears automatically:
b
t
= 0.
1.4.5 Consequences for monetary policy
The decision rules of households and aggregated production function of a standard new
Keynesian model corresponds to that in the presented frictionless monetary economy:
c
t
=
t
c
t+1
−
1
σ
c
(i
t
−
t
π
t+1
−ρ),
w
t
− p
t
= σ
c
c
t
+σ
h
h
s
t
,
y
t
= a
t
+h
d
t
,
(1.20)
where ρ = −lnβ.
In the frictionless monetary model the decision rules of ﬁrms — making the real wage equal to
marginal productivity of labour or, equivalently, making the nominal marginal cost of production
equal to market price — made the real wage dependent on employment and productivity. Thus,
they force the real marginal cost to be equal to unity
19
:
mc
t
= w
t
− p
t
−mph
t
= w
t
− p
t
−a
t
= 0. (1.21)
Equation which made the real wage dependent on productivity only, along with the conditions of
market clearing, closed the frictionless model in the sense that it was possible to determine the
dynamics of real variables with no need to refer to the rules of monetary policy. Euler equation
determined only the dynamics of real interest rate r
t
= i
t
−
t
π
t+1
. When ﬁrms operate in
monopolistic competition conditions, the average real marginal cost in economy mc
t
does not
have to be equal to unity but it is equal to the inverse of the average monopolistic markup in
economy λ
t
:
mc
t
= w
t
− p
t
−mph
t
= w
t
− p
t
−a
t
=−λ
t
= 0 (1.22)
as ﬁrms that may reoptimise the price determine it above the marginal nominal cost. Equation
1.22 as ﬁrms that may reoptimise the price determine it above the marginal nominal cost.
Equation 1.21 which — making the real wage dependent only on labour productivity — closed
the frictionless model and allowed to determine dynamics of real categories with no need to
deﬁne the method of determining the nominal interest rate i
t
20
. Equation 1.22 relates the real
wage with productivity and average monopolistic markup. Assuming the exogeneity of the
process generating average markups λ
t
, the real variables of the new Keynesian model may be
19
Which means that the real cost logarithm mc
t
shall be zero.
20
Actually, the nominal interest rate was absent in the frictionless model. The interrelation between the real interest
rate and the expected inﬂation was included by virtue of the Fisher rule.
28
Genesis and anatomy of dynamic stochastic general equilibrium models
WORKING PAPER No. 83 29
1
expressed by RBC model variables and by average markups:
y
t
= y
t
−
1
σ
c
+σ
h
λ
t
,
h
t
= h
t
−
1
σ
c
+σ
h
λ
t
,
w
t
− p
t
= (w
t
− p
t
)
−
σ
h
σ
c
+σ
h
λ
t
,
r
t
= r
t
−
1
σ
c
+σ
h
t
(λ
t+1
−λ
t
),
(1.23)
where the values of y
t
, h
t
, w
t
−p
t
and r
t
variables are that resulting from the frictionless model,
namely in the case of absence of price rigidities and in perfect competition. Thus, it may be seen
that nonneutrality of monetary policy in the presented standard model takes place by virtue
of the process of determining optimal margins or, equivalently, by the process of determining
optimal prices. More precisely, the process of determining optimal prices must be disturbed
because if we assume perfect ﬂexibility of prices, i.e. exclude price rigidities, the sector of ﬁrms
shall be reduced to a representative monopolist determining the optimal price in each period at
the level of:
P
t
=
η
η−1
ψ
t
where ψ
t
stands for the nominal marginal cost of a representative monopolist. The markup
shall, then, be ﬁxed and equal to λ
t
= ln
η
η−1
, which would close the considered model so that
the dynamics of real categories would be independent of monetary policy. In the case of price
rigidities of Calvo type, the solution of the decisionmaking problem of ﬁrms leads to a stochastic
difference equation for the optimal price, which upon linearisation has the form of:
p
t
= p
t−1
+βξ
t
{p
t+1
− p
t
} +(1 −βξ) ˆ mc
t
+π
t
, (1.24)
where ˆ mc
t
= mc
t
−mc = −(λ
t
−λ) stands for the deviation from the average real marginal cost
mc
t
in economy from its value in steady state mc or, equivalently, the minus deviation of the
average markup λ
t
in economy from its value in steady state λ = −mc. Based on that equation,
the average markup may be determined as the function of the expected optimal price in the next
period:
λ
t
= λ+
βξ
(1 −βξ)
t
(p
t+1
− p
t
).
Consequently, it may be seen that the model with price rigidities shall not be closed such as the
frictionless model or model with a representative monopolist did. Average markups depend on
the expected increment of the optimal level of prices, which within the model are determined
endogenously. Due to disturbances in the prices formation process changes of the nominal
interest rate are not one to one translated into the expected inﬂation but affect the real interest
rate. In reply to the monetary policy shock not all ﬁrms shall be able to determine markup on
the level of the optimal response. Therefore, as long as the monetary shock has not expired, the
monetary policy shall inﬂuence real categories.
29
Genesis and anatomy of dynamic stochastic general equilibrium models
N a t i o n a l B a n k o f P o l a n d 30
1
Optimal price policy of ﬁrms implies the inﬂation equation in the following form:
π
t
= β
t
{π
t+1
} +κ˜y
t
,
where κ =
(1−ξ)(1−βξ)(σ
c
+σ
h
)
ξ
, and ˜y
t
= y
t
− y
n
t
stands for the output gap, i.e. the difference
between product y
t
and the socalled natural product y
n
t
, namely the product that would be
materialised in the case of perfectly ﬂexible prices, meaning one that may be expressed by the
level of technology a
t
. The equation is known as the newKeynesian Philips curve. Euler equation
brings about the dependence of output gap on the path of differences between the real interest
rate and its natural level, i.e. to the socalled dynamic IS equation:
˜y
t
=
t
{˜y
t+1
} −
1
σ
c
(i
t
−
t
{π
t+1
} − r
n
t
),
where:
r
n
t
= ρ +σ
c
t
1 +σ
h
σ
c
(1 −α) +σ
c
+α
(1 −ρ)a
t
is the natural interest rate. The last three equations and the process deﬁning the technology
a
t
= ρ
a
a
t−1
+η
t
deﬁne the form of the new Keynesian model with regard to π
t
, ˜y
t
, r
n
t
and a
t
.
Nevertheless, as may be expected in the light of the presented motivation, equations includes the
i
t
variable — the nominal interest rate, the level of which is determined by the central bank. In
order to be able to solve the model, an equation for i
t
,shall be introduced, i.e. it shall be decided
how the central bank is to control the nominal interest rate. Most commonly, an equation for i
t
is
assumed in the form of the socalled monetary policy rules. They are ad hoc assumed equations
deﬁning the response of interest rate to the changes in macroeconomic aggregates. If the interest
rate reacts to inﬂation changes and output gap, the form of the rule is following:
i
t
= ρ +φ
y
˜y
t
+φ
π
π
t
.
Such type of rule is known as Taylor rule.
30
DSGE model — anatomy
WORKING PAPER No. 83 31
2
Chapter 2
DSGE model — anatomy
Formally, a DSGE model is a system of ﬁrstorder conditions for economic agents, conditions
constraining their decisions and equilibrium conditions. Should all the variables of the model,
except structural shocks, be gathered in y
t
, vector and structural shocks in vector ε
t
, the DSGE
model shall be a stochastic difference equation in the following form
1
:
t
f ( y
t+1
, y
t
, y
t−1
, ε
t
)
= 0, (2.1)
where {ε
t
I
t−k
} = 0 for k ≥ 1 and I
t
represents the information about the state of the world
possessed by the economic agents in t period, i.e. the so called information set. The form (2.1) is
called a structural form of a DSGE model, i.e. it represents the economic mechanism determining
the dynamic properties of economy. The y
t
vector is called a vector of endogenous variables of
the model, state vector or simply state. The mechanism represented by the structural form of
the model includes the encoded decision rules of economic agents. In order to solve the model,
it is necessary to decode the rules and represent them in operational form, i.e. form enabling
their direct implementation. The operationality of the model solution from the perspective of
t period is understood to be such property that today’s y
t
state may be determined based on
the knowledge of its previous value y
t−1
and the value of ε
t
, shocks that have materialised
today. Both former states and today’s shocks come within the information set I
t
, possessed by
the economic agents in t period. Knowing the solution of the model, economic agents are able
to implement their decisionmaking rules. Formally, the solution of model (2.1) is any g function
in the form of:
y
t
= g( y
t−1
, ε
t
), (2.2)
1
Vector y
t
includes all of the variables of the model except structural shocks. In particular, it may include autoregres
sive variables in the form of: θ
t
= ρθ
t−1
+
t
, where
t
is a structural shock. Variables in the form of θ
t
= ρθ
t−1
+
t
are called model disturbances. From the point of view of model representation in the so called state space, disturbances
are endogenous variables, i.e. states.
31
2
DSGE model – anatomy
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 32
2
which solves the structural model, i.e. shall satisfy the equation (2.1):
t
f
g
g( y
t−1
, ε
t
, ε
t+1
), g( y
t−1
, ε
t
), y
t−1
, ε
t
= 0. (2.3)
The condition implies that the decision rules represented by the g function are actually possible
and optimal. The g function is called a reduced form of a DSGE model, its solution, a decision
rule, the law of motion for the state or policy function. The latter name originates from the
optimal control theory. The g function represents the decision rules of economic agents decoded
from the structural form of the model, describes the laws of motion for all of the variables of the
model, and thus, determines the dynamics of the model. Equations of motion determine the
values of all variables of the model in t period, i.e. the state vector y
t
, based on the previous
state of economy y
t−1
and structural shocks ε
t
that occurred in the current period. Determining
the g function is equivalent to solving the DSGE model.
Solving a DSGE model is a difﬁcult task in general, due to the nonlinear form of the f function
and absence of adequately general analytical or numerical methods. Therefore, instead of con
sidering the nonlinear model (2.1), its linear approximation is being considered. This is the
socalled expansion into the ﬁrstorder Taylor series, (eg. Birkholc, 2002). This considerably
simpliﬁes the procedure of model solving, however, does not lead to determination of the g
function but a linear approximation. Moreover, the model is most often expressed in logarithms
of variables such that after linearisation, the values of their increments
2
d y
t
have the inter
pretation of percentage deviations from the point in the neighbourhood of which the model is
approximated. Such approximation is called a logarithmiclinear (loglinear) approximation. If
the model is loglinearised in the neighbourhood of a steady state, the values of the increments
of variables d y
t
are interpreted as the percentage deviations from the value in the steady state,
i.e. from the values consistent with the longterm equilibrium.
The expansion of model (2.1) into the Taylor series takes place in the neighbourhood of a chosen
point. It has become customary that the point is the point of longterm economy equilibrium, i.e.
model (2.1) is expanded into ﬁrstorder Taylor series in the neighbourhood of a deterministic
steady state, in this chapter identiﬁed as ¯y. The deterministic steady state is deﬁned as each ¯y
point satisfying:
f (¯y, ¯y, ¯y, 0) = 0. (2.4)
In other words, steady state is a state for which selfrepeating shall be optimal, if no structural
shock occur — economic agents in the steady state have no motivation to change their decisions.
Thus, steady state is also a ﬁxed point of g representation, assuming zero values of structural
shocks ε
t
= 0:
¯y = g(¯y, 0). (2.5)
Since y
t+1
= g(g( y
t−1
, ε
t
), ε
t+1
) and f ( ¯y, ¯y, ¯y, 0) = 0, expansion of (2.1) into ﬁrstorder Taylor
2
Formally — the values of the differentials.
32
DSGE model — anatomy
WORKING PAPER No. 83 33
2
series in the neighbourhood of steady state ¯y has the form of:
t
∂ f
∂ y
t+1
(¯y)[
∂ g
∂ y
(¯y)]
2
+
∂ f
∂ y
t
(¯y)
∂ g
∂ y
(¯y) +
∂ f
∂ y
t−1
(¯y)
d y
t−1
+
∂ f
∂ y
t+1
(¯y)
∂ g
∂ y
(¯y)
∂ g
∂ ε
(¯y) +
∂ f
∂ y
t
(¯y)
∂ g
∂ ε
(¯y) +
∂ f
∂ ε
(¯y)
dε
t
= 0.
(2.6)
Denoting matrices
∂ f
∂ y
t+1
(¯y),
∂ g
∂ y
(¯y),
∂ f
∂ y
t
(¯y),
∂ f
∂ y
t−1
(¯y),
∂ g
∂ ε
(¯y) and
∂ f
∂ ε
(¯y) by A
0
, A, A
1
, A
2
, B and
S respectively, the linear approximation of the model takes the form of:
t
(A
0
A
2
+A
1
A+A
2
)d y
t−1
+ (A
0
AB +A
1
B +S)dε
t
= 0. (2.7)
System 2.7 is a stochastic matrix difference equation with unknown matrices A and B. Determi
nation of A and B matrices satisfying equation (2.7), i.e. solution of equation (2.7), enables the
determination of linear approximation of policy function g, because up to the ﬁrstorder Taylor
series expansion around ¯y deterministic steady state reads y
t
= g( y
t−1
, ε
t
) ≈ ¯y +Ad y
t−1
+Bdε
t
or, equivalently:
d y
t
≈Ad y
t−1
+ Bdε
t
.
And the differential d y
t
represents the deviation of state y
t
from the steady state and, if the
variables of a model are expressed in the form of logarithms, the deviation is interpreted as
percentage deviation. The last equation indicates that the reduced form of a DSGE model has
the representation of a VAR model, with accuracy to linear approximation.
As equation (2.7) must be satisﬁed for every d y
t
and dε
t
, both addends of a sum under the sign
of the expected value must be equal to zero, i.e. it is required that:
t
(A
0
A
2
+A
1
A+A
2
)d y
t−1
= 0 (2.8)
for every t and:
t
(A
0
AB +A
1
B +S)du
t
= 0 (2.9)
for every t. The former equation is a stochastic matrix quadratic equation with respect to A, and
the latter — if only A has been determined — shall enable to calculate B from the condition:
B = −(A
0
A+A
0
)
−1
S. (2.10)
Thus, the solution of the model boils down to the determination of matrix A from equation 2.9.
Special numerical algorithms are applied for that purpose, e.g the basic BlanchardKahn method
or the AndersonMoore algorithm.
The solution of linear approximation of a DSGE model or, equivalently, the linear approximation
33
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 34
2
of its reduced form
3
:
y
t
= Ay
t−1
+Bε
t
is represented in the socalled statespace:
Y
t
= Hy
t
+u
t
,
y
t
= Ay
t−1
+Bε
t
,
(2.11)
and:
u
t
∼ N(0, R) oraz ε
t
∼ N(0, I).
The former equation is called the measurement equation and presents the relation between
observations, namely observable variables Y
t
, and endogenous variables y
t
of the model, which
in the context of model representation in the statespace are called state variables or simply
states. Observations may be functions of endogenous variables y
t
, functions of exogenous
variables, and in particular they may include trends. They may also depend on the values of
states in steady state
4
. Random variable u
t
is called a measurement error or disturbance. The
latter equation is called transition equation and describes the dynamics of state variables y
t
,
i.e. the already known linear approximation of the model solution. This equation is a linear
approximation of the reduced form, i.e. solution, of the model. Hereinafter we shall call it a
solution or reduced form. The elements of matrices A, B and H are nonlinear functions of model
parameters, which shall be identiﬁed with θ. Formally, θ ∈ Θ ⊂ R
n
for some n. Representation
of the reduced form of a DSGE model in the state space (2.11) together with the assumptions of
ε
t
shocks distribution and u
t
observation errors shall be abbreviated with the symbol and
called the model.
From the statistical — classical — point of view, the model may be considered a family of
conditional distributions of observables with regard to nonobservable data, i.e. with regard to
states, shocks, measurement errors and parameters:
p(Y y, ε, u, θ), y ∈ Υ, ε ∈ Ξ, u ∈ Ω, θ ∈ Θ
, (2.12)
where Ξ is the space of shocks ε, Υ is the state space y, Ω is the space of measurement errors u,
and Θ is the parameters space. Alternatively, but also from a classical point of view, the model
represents the family of conditional probability distributions of its variables conditional upon
the θ parameters:
p(Y, y, ε, uθ), θ ∈ Θ
. (2.13)
The form, upon integration of states, shocks and disturbances from the distribution p(Y, y, ε, uθ)
3
From now on, to the end of this chapter, we assume that the variables of the model are expressed in the form of
percentage deviations from the steady state.
4
For the simplicity of notation, the representation of model (2.11) makes the observations dependent only on states.
34
DSGE model — anatomy
WORKING PAPER No. 83 35
2
leads to the likelihood function L(θY):
L(θY) = p(Yθ) =
Ω
Ξ
Υ
p(Y, y, ε, uθ)d ydεdu, (2.14)
which is the central object for the estimation of the model parameters.
From the statistical — Bayesian — point of view the model may be considered joint distribu
tion of observations, states, shocks, disturbances and parameters, which gives rise to factorisation
into the conditional distribution of observations, states, shocks and disturbances with regard to
parameters, and unconditional distribution — the socalled prior distribution — of parameters:
p(Y, y, ε, u, θ) = p(Y, y, ε, uθ)p(θ). (2.15)
Upon integration of states, shocks and disturbances from the joint distribution, we receive the
socalled kernel of posterior distribution (posterior kernel) (θY):
(θY) =
Ω
Ξ
Υ
p(Y, y, ε, uθ)p(θ)d ydεdu = p(Yθ)p(θ) (2.16)
being the product of the likelihood function of parameters and density of the probability of
their prior distribution. The kernel of posterior distribution of parameters is the central object of
Bayesian estimation of parameters.
The process of solving a DSGE model requires calculations made on matrices A
0
, A
1
, A
2
i S,
which deﬁne the linear approximation of its structural form. In order to solve the model, the
elements of the matrices — parameters of the model — shall be assigned numerical values.
A solved DSGE model, i.e. its reduced form, may next serve various simulation experiments,
both positive and counterfactual. It is also possible to quantify the response of economy to
structural shocks, estimate the values of shocks in a sample, forecast endogenous variables and
their functions — the socalled observable variables. Nevertheless, all the experiments require
the solution of the model, i.e. prior assignment of numerical values to its parameters. This may
be done in many ways.
2.1 Values of parameters
In order to be able to solve a model and then carry out analyses based thereon, the parameters
of the model shall be assigned numerical values. This may be done at least in two ways — the
values of parameters may be calibrated or estimated. In practice, calibration and estimation are
applied in a complementary manner, i.e. some of the parameters of the model are calibrated
and some of them are estimated. Among the methods of estimation, the most often used is
the maximum likelihood method, method of moments and Bayesian estimation. Currently, the
Bayesian approach becomes more and more popular. This initially resulted from the fact that the
estimation of DSGE models with the maximum likelihood method with the use of deterministic
optimisation algorithms results in numerical problems that prevent the satisfactory use of the
method. Currently, the Bayesian approach to DSGE models estimation becomes more common
35
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 36
2
than only the methods of solving technical problems. This results mainly from the fact that the
approach enables the consideration of knowledge a priori, in an explicit and formal manner.
2.1.1 Calibration
Calibration may entail the assignment of values to the model parameters in an arbitrary manner
or can be based on other research. As an example, there are empirical studies dedicated to the
estimation of elasticity of demand for domestic and foreign products in various sectors of many
economies. The elasticities speciﬁed therein may be assumed to be the one chosen for the DSGE
model. On the other hand, in the calibration process, such values of DSGE model parameters may
be assumed as to make its dynamics (response of economy to structural shocks) and longterm
equilibrium correspond with economic intuition, the results of other studies or indication of
models of more empirical purpose, such as VAR class models. In practice, calibration is a mixture
of both techniques. From the point of view of Bayesian estimation, calibration boils down to
the assignment to the calibrated parameters a possibly informative prior distribution. Thus,
such method of reasoning enables the formal reﬂection of that what is common in the classical
approach but happens ad hoc and in an implicit manner.
2.1.2 Maximum likelihood estimation
Among the classical methods of estimation the most often used is the maximum likelihood
method. It entails the determination of the values of θ, parameters, for which the likelihood
function (θY) obtains the highest value. The method requires the determination of the
analytical form of (θY) function or development of a numerical method of its approximation.
Formally, the likelihood function (θY) is a density function of conditional distribution of
observables Y conditional with respect to θ parameters::
(θY) = p(Yθ).
The likelihood function may be considered a function measuring the likelihood of θ parameters
in the sense that it takes higher value for such conﬁguration of parameters for which the density
of probability — determined by p(Yθ) that data Y was generated by the model whose
parameters have θ value — is larger. The likelihood function (θ, Y) = p(Yθ) originates as
a result of integration of y states, ε shocks and u disturbances from the conditional density
p(Y, y, ε, uθ). Technically, marginalisation of states, shocks and disturbances takes place in the
process of their ﬁltration. Filtration is conditioned with regard to the model. In order to
emphasise this fact, the likelihood function is denoted by (θY, ()), where () stands
for ﬁltration made with the use of the model.
From the operational point of view, likelihood (θY, ()) is measured by using an iterative
formula:
(θY, ()) = p(Yθ, ()) = p(Y
0
)
T
t=1
p(Y
t
I
t−1
),
where I
t
= {Y
0
, Y
1
, ..., Y
t
} denotes the information set in t period. The use of such formula
36
DSGE model — anatomy
WORKING PAPER No. 83 37
2
requires the determination of probability density p(Y
t
I
t−1
) for t = 1, 2, ..., T. Assuming the
normality of shocks and disturbances, the density is the density of normal distribution:
p(Y
t
I
t−1
) =(2π)
−
Tk
2
((Y
t
I
t−1
)
−
1
2
×
×exp
−
1
2
Y
t
−(Y
t
I
t−1
)
((Y
t
I
t−1
))
−1
Y
t
−(Y
t
I
t−1
)
.
In order to calculate the likelihood of parameters, it is vital to know expected values (Y
t
I
t−1
)
and variances (Y
t
I
t−1
) for t = 1, 2, ..., T. For that purpose Kalman ﬁlter is used, the equations
of which are — in the case of normal distribution — identical as the analytically derived formulas
for the values.
In order to determine values of the θ parameters maximising the likelihood function (θY, ())
iterative methods based on NewtonRaphson method are most often used (see
˙
Zak and Chong,
2008). The general sequence of estimation of DSGE model parameters with the maximum
likelihood method is the following:
1. Parameters assume the value θ
k
, k = 0, 1, 2, ..., K. For k = 0 the initial value of parameters
must be assumed.
2. Solution of the model for θ = θ
k
, i.e. determination of matrices A =
∂ g
∂ y

θ=θ
k
(¯y) and
B =
∂ g
∂ ε

θ=θ
k
(¯y).
3. Determination of the value of the likelihood function for θ = θ
k
, i.e. determination of
(θ
k
Y, ()) in one run of the Kalman ﬁlter.
4. Numerical determination of the ﬁrst two differentials of function (θY, ()) for θ = θ
k
,
i.e. the gradient
(θ
k
Y,())
and hessian H
(θ
k
Y,())
.
5. Making an iterative step of the NewtonRaphson method, i.e. assumption that
θ
k+1
= θ
k
−
(θ
k
Y,())
(H
(θ
k
Y,())
)
−1
.
6. Veriﬁcation of algorithm convergence criteria.
DSGE model estimation with the maximum likelihood method, with the use of NewtonRaphson
method is time consuming. This is because NewtonRaphson method is iterative and each
iteration requires the solution of a model (point 2) and running the Kalman ﬁlter in order
to determine the value of the likelihood function (θY, ()) (point 3). Additionally, it
requires numerical determination of a gradient and Hessian of the likelihood function, which is
a numerically instable task. Finally, NewtonRaphson method requires the speciﬁcation of initial
values of θ
0
parameters and is a local method — i.e. it does not guarantee that the obtained
estimation θ
K
maximises globally the likelihood function. Hence, stochastic optimisation methods
have recently become more popular, particularly simulated annealing. Yet, the largest problem
related to the application of the maximumlikelihood method to estimate DSGE model parameters
is insufﬁcient curvature of the likelihood function, namely the presence of the socalled plateau.
Deterministic iterative optimisation methods are then ineffective and usually the obtained θ
K
estimation is not much different from the speciﬁed initial values of θ
0
.Therefore, mainly for
37
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 38
2
that reason, DSGE models are often estimated with Bayesian methods, which allow to increase
the likelihood function curvature by virtue of a change of its shape with the socalled prior
distribution. Maximum likelihood method poses also other difﬁculties.
2.1.3 Bayesian estimation
Although the initial reasons for applying Bayesian estimation to DSGE models were strictly
technical, the Bayesian approach makes the nature of statistical analysis of DSGE models other
than classical. Bayesian inference fundamentally differs from the classical one. Unlike as in
the classical case, in the Bayesian interference the probability of the formulated hypotheses
— and the statement that θ parameters of the DSGE model take a speciﬁc value illustrates
such hypothesis
5
— is assessed not only on the basis of how strongly the Y data conﬁrm the
hypothesis, but also based on subjective assessment of the probability that is not related to the Y
data. Such subjective assessment is called a priori assessment. A priori assessment merged with
classical likelihood of parameters, i.e. with the likelihood function (θY, ()), results in the
a posteriori assessment of probability of the truth of the postulated hypothesis. The application
of the Bayesian approach to estimate parameters equals — from the numerical point of view
— the use of the maximum likelihood method with full information. Hence, it is also possible
to formally include a priori knowledge, consider uncertainty with parameters in the form of
posterior distribution and make formal smallsample inference that does not refer to asymptotic
properties.
Formally, the Bayesian estimation task entails the determination of the posterior distribution:
p(θY) =
p(θ)p(Yθ)
p(Y)
of parameters θ. The analytical determination of posterior distribution is usually impossible, so
simulation approach, e.g. MetropolisHastings algorithm, is applied. For point estimates of θ
parameters one assumes, for example, the maximum of posterior density function in order to
approximate its modal value. Density p(Yθ) is obtained based on the (θY, ()) likelihood,
which is determined in the process of ﬁltration conditionally with regard to the model . Such
procedure leads to posterior distribution:
p(θY, ()) =
p(θ)p(Yθ, ())
p(Y())
, (2.17)
where p(Y()) is a normalising factor — a density function of data subject to ﬁltration
(). Analytical determination of marginal distribution p(Y()) by calculation is usually.
Therefore, while determining the maximum of posterior distribution, the normalising constant
tends to be ignored and maximisation is applied only to the product of posterior distribution
and likelihood function:
(θY, ()) = p(θ)p(Yθ, ())
5
The hypothesis states that the ,model, whose parameters have the θ value, generated the data Y.
38
DSGE model — anatomy
WORKING PAPER No. 83 39
2
i.e. the socalled posterior kernel. Rationalisation of such conduct results from the fact that
posterior distribution is proportional to its kernel:
p(θY, ()) ∝ (θY, ()).
Prior distribution is assumed arbitrarily. This may be both, informative distribution and unin
formative distribution. The informative nature of prior distribution increases along with the
decrease of its variance, i.e. subjective assessment draws a relatively growing attention along
with a drop in variance. If prior distribution is highly informative, it is considered to have a
"strong prior", in contrast to uninformative distribution.
Determination of posterior kernel requires — as in the case of the maximum likelihood method
— the determination of the likelihood function (θY, ()), and strictly speaking — its
logarithm, as instead of maximisation of the kernel (θY, ()), it is possible to equivalently
maximise its logarithm:
ln(θY, ()) = ln p(θ) +ln p(Yθ, ()),
which is a sum of prior distribution logarithmln p(θ) and a loglikelihood function ln p(Yθ, ()).
The value of likelihood function may be determined with the use of Kalman ﬁlter equations.
2.2 Kalman ﬁlter
A DSGE model may be perceived from many perspectives. It may be perceived as a system of
conditions deﬁning the optimal behaviour of economic agents or as a stochastic matrix difference
equation. It may also be viewed as the socalled ﬁlter. Let us assume that we have observable
data, Y = {Y
T
, Y
T−1
, ..., Y
0
}, namely observations from periods t = 0, 1, 2, ..., T. The ﬁlter is
a mechanism processing Y data in order to determine the expected values and variances of
y
t
endogenous variables of the model, Y
t
observations, ε
t
shocks and u
t
disturbances in t period,
under the condition of an informative set I
t
, then we speak of a ﬁlter, or under the condition
of an information set I
T
, then then we speak about a smoother. The task of determination
of the expected values and variances under the condition of set I
t+k
for k = 0, 1, 2, ..., T − t
is called a ﬁltration, while the task of determining the expected values and variances under
the condition of set I
t−k
for k = 1, 2, ..., t is called a prediction. Since we operate on a linear
approximation of a DSGE model, the approximation is a linear ﬁlter. There are many methods
of ﬁltration and prediction. In the context of DSGE models, most often used is the socalled
Kalman ﬁlter. The equations of the Kalman ﬁlter may be derived from the optimisation calculus
by minimisation of oneperiod forecasterrors of observable variables of the model. Assuming the
normality of shocks and measurement errors of the model, ﬁltration and prediction may also be
derived analytically from the probabilistic viewpoint. Obtained derivations coincide with Kalman
ﬁlter equations. Hence the ﬁlter has statistical interpretation and the likelihood function based
thereon is not the approximation which would occur in the case of more general assumptions
with regard to the distribution of shocks and measurement errors, but constitutes an exact result.
The effect of the ﬁlter application has several dimensions. Firstly — the determination of the
39
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 40
2
expected values and variances:
( y
t
I
t
) and ( y
t
I
t
) (2.18)
or:
( y
t
I
T
) and ( y
t
I
T
) (2.19)
of endogenous variables, which enables one to know the estimated values of those among them
which are not perfectly observable, i.e. do not come within set Y. Secondly — determination of
the expected values and variances:
(ε
t
I
t
) and (ε
t
I
t
) (2.20)
or:
(ε
t
I
T
) and (ε
t
I
T
) (2.21)
of structural shocks, which is equivalent to their identiﬁcation. Thirdly — determination of the
expected values and variances:
(Y
t
I
t−1
) and (Y
t
I
t−1
) (2.22)
enables one to determine the value of the likelihood function (θY, ) for θ. Determination of
the value of (θY, ) is an element of estimation both according to the maximum likelihood
method and Bayesian estimation.
Deriving the Kalman ﬁlter that has statistical interpretation requires an assumption that struc
tural shocks of the model have normal distribution. This assumption, provided that it is actually
fulﬁlled, has some advantages. It appears that Kalman ﬁlter — if shocks reﬂect normal dis
tribution  is an optimal ﬁlter with regard to minimisation of variances of the estimations of
states, shocks and observations. If the assumption of normality is not fulﬁlled, the Kalman
ﬁlter is optimal in the class of linear ﬁlters, however, there are more effective nonlinear ﬁlters
in the aforementioned sense. In consequence of the assumptions regarding the normality of
structural shocks, the density function of Y
t
— conditional upon I
t−1
, is a density function of
a multidimensional normal distribution, so its analytical form is known. Kalman ﬁlter enables,
therefore, the determination of the distribution of Y
t
observations in each t period, conditional
with regard to information set of t −1 period. Since it is a multidimensional normal distribution,
an analytical form of the iterative likelihood function of Y observations is known.
For the initial values of ( y
0
) and ( y
0
), the Kalman ﬁlter includes the following equations.
The expected values ( y
t
I
t
) and variances ( y
t
I
t
) of states for t = 1, 2, ..., T are determined
recurrently based on the following relations:
( y
t
I
t
) = A( y
t−1
I
t
) +B (ε
t
I
t
),
( y
t
I
t
) = A( y
t−1
I
t
) A
+B(ε
t
I
t
) B
,
(2.23)
40
DSGE model — anatomy
WORKING PAPER No. 83 41
2
while assuming the normality of structural shocks, the following applies:
( y
t−1
I
t
) = ( y
t−1
I
t−1
) +cov( y
t−1
, Y
t
I
t−1
) (Y
t
I
t−1
)
−1
(Y
t
−(Y
t
I
t−1
)),
( y
t−1
I
t
) = ( y
t−1
I
t−1
) −cov( y
t−1
, Y
t
I
t−1
) (Y
t
I
t−1
)
−1
cov( y
t−1
, Y
t
I
t−1
).
(2.24)
The expected values (ε
t
I
t
) and variances (ε
t
I
t
) of structural shocks for t = 1, 2, ..., T are
determined recursively, assuming the normality of structural shocks, based on the following
relationships:
(ε
t
I
t
) = cov(ε
t−1
, Y
t
I
t−1
) (Y
t
I
t−1
)
−1
(Y
t
−(Y
t
I
t−1
)),
(ε
t
I
t
) = Ψ −cov(ε
t
, Y
t
I
t−1
) (Y
t
I
t−1
)
−1
cov(ε
t
, Y
t
I
t−1
),
(2.25)
The expected values (Y
t
I
t−1
) and variances (Y
t
I
t−1
) of observations for t = 1, 2, ..., T are
determined recursively based on the following relationships:
(Y
t
I
t−1
) = H A( y
t−1
I
t−1
) +A
x
x
t
,
(Y
t
I
t−1
) = H A( y
t−1
I
t−1
)A
H
+R.
(2.26)
2.3 MetropolisHastings algorithm
Bayesian estimation entails the determination of posterior distribution p(θY, ()) of pa
rameters. The distribution is not normal with regard to θ parameters (it is normal with regard
to nonlinear functions of θ parameters). Thus, its analytical form is not known in the gen
eral case. The shape of the density function p(θY, ()), or, equivalently, the shape of the
(θY, ()) function is approximated with simulation methods. For that purpose sampling
methods are used, most often the MetropolisHastings algorithm. The result of the operation
of the algorithm is a sequence of parameters Θ = {θ
0
, θ
1
, ..., θ
M
}, the subsequence of which
θ
m
, θ
m+1
, ..., θ
M
originates — for an adequately large m — from unknown posterior distribution
p(θY, ()). Formally, the Θ sequence is a Markov chain, whose stationary distribution is
the p(θY, ()). distribution. The approximation of posterior distribution p(θY, ())
involves the determination of a histogram of the elements of the Θ set. Based on the members
of the Θ set, also the moments of parameter estimates may be determined.
Let θ
0
be an initial value of parameters, e.g. the modal value of (θY, ()) kernel of
p(θY, ()) posterior distribution, let Σ be a numerically determined second differential of
(θY, ()) in θ
0
, and let γ be a positive constant, and let us assume that Θ = . The kth,
k = 1, 2, ..., M, step of MetropolisHastings algorithm runs as follows:
1. Draw θ from the normal distribution N(θ
k−1
, γΣ) with the expected value θ
k−1
and variance
γΣ.
2. Assume θ
k
=
θ, with probability min(1, r);
θ
k−1
, otherwise.
where acceptance threshold r is speciﬁed by: r =
(θY,())
(θ
k−1
Y,())
=
p(θY,())
p(θ
k−1
Y,())
.
3. Assume: Θ = Θ∪{θ
k
}.
41
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 42
2
Calculation of the value of (θY, ()) (point 2) requires making one run of the Kalman
ﬁlter, thus, the ﬁlter is run in each iteration of the MetropolisHastings algorithm. The Metropolis
Hastings method is highly timeconsuming, as it requires that chain Θ be convergent to stationary
distribution, which happens for very large M values from the practical point of view. There are
analytical tools, i.e. convergence diagnostics, which enable the determination of Θ chain conver
gence. Without convergence veriﬁcation, approximation of posterior distribution p(θY, ())
is not credible.
To verify convergence of Θ we may apply Gelman and Rubin diagnostic or the Geweke diagnostic.
The Gelman and Rubin diagnostic suggests that for, each parameters, the shrink or scale reduction
factor:
r =
(1 −
1
n
)W +
1
n
B
W
should not be too hign (perhaps lower than 1.1 – 1.2), where W and B are estimates of within
and between chain variances of this parameter when length of each chain equals n. The Geweke
test simply applies the difference of means test to two overlapping parts of the chain to check if
the two parts of the chain come from the distribution with the same mean.
2.4 Model selection
In the process of model building one usually works with alternative speciﬁcations or alternative
calibrations of the same model. At the end of the day a single speciﬁcation is used
6
. To select
a single model from a family of models we may want to compare them in a pairwise
manner by ratios of their posterior distributions (the so called posterior odds ratio):
p(
1
)Y
2
Y
=
p(
1
)p(Y(
1
))
p(
2
)p(Y(
2
))
(2.27)
for
1
,
2
∈ where:
p(Y()) =
θ∈Θ
p(θ, Y())dθ =
θ∈Θ
p(θ)p(Yθ, ())dθ (2.28)
is a marginal density of data Y provided that model is used. This integral in most cases is a
difﬁcult one and has to approximated. One way to approximate (2.28) is to assume a functional
form of the posterior kernel. In case of normal distribution this method is known as Laplace
approximation and the estimate of p(Y()) is:
p(Y()) = (2π)
n
2
(Σˆ
θ
)p(
ˆ
θY, ())p(
ˆ
θ
()) (2.29)
where
ˆ
θ
and Σˆ
θ
denote estimate of θ using model (e.g. the posterior mode) and its
covariance.
6
However, in order to reduce the modele selection risk, one could in principle work with many models using methods
of bayesian averaging for policy experiments like stochastic simulations or forecasting.
42
DSGE model — anatomy
WORKING PAPER No. 83 43
2
Comparison of posterior odds ratios is a technical procedure. It should be supplemented by
veriﬁcation of models’ in sample and, if possible, out of sample forecasting power as well as by
qualitative inspection of models’ theoretical structure.
2.5 Applications
With a reduced form of a DSGE model, it is possible to do several standard exercises: to identify
structural shocks, analyse responses of endogenous (observable) variables to structural shocks,
perform variance decomposition of endogenous variables, carry out historical decompositions
and forecasts.
All the exercises are based on the property that the reduced form of a DSGE model, i.e. its
solution, has a structure of a VAR model in reduced form:
y
t
= Ay
t−1
+ Bε
t
Based on Y data all the simulation exercises that have been developed for VAR class models may
be done for a DSGE model. Let us shortly discuss them in the following paragraphs.
2.5.1 Structural shocks identiﬁcation
The basic exercise that may be done on basis of the reduced form DSGE model is identiﬁcation of
structural disturbances ε
t
in a sample, i.e. for periods 0, 1, 2, ..., T. Shocks identiﬁcation consists
in derivation of expected values (ε
t
I
t
) and variances (ε
t
I
t
). As it has been said, this may be
done in the course of one run of the Kalman ﬁlter. In this context, solution or the reduced form
of a DSGE model may be considered not only as a VAR model but also as a linear ﬁlter.
2.5.2 Impulse response analysis
Analysis of response functions (or impulse response analysis) replies to the question of what
happens in the economy after the occurrence of a structural shock. More precisely, the analysis
aims to quantify the responses of endogenous variables in periods t, t + 1,..., i.e. values of
y
t
, y
t+1
, ..., to an impulse from a structural shock ε
t
in period t. The dynamic reaction of the
economy may be determined on the basis of the reduced form of a DSGE model, with the use of
the following relationship:
∂ y
t+k
∂ ε
t
= A
k
Bε
t
,
where:
A =
∂ g
∂ y
(¯y) oraz B =
∂ g
∂ ε
(¯y)
Presentation of the response of endogenous variables to a shock which takes place in period t
depending on index k ≥ 1 is called an impulse response function (shortly IRF). It is assumed,
at the same time, that before t period the economy stays in the longterm equilibrium ¯y, in
t period a structural shock ε
t
, occurs, while in the periods t +1, t +2, ..., t + k no structural
shocks occur, i.e. ε
t+1
= ε
t+2
= ... = ε
t+k
= 0.
43
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 44
2
Based on the above relationship, it may be seen that the necessary and sufﬁcient condition for
stability of the longterm equilibrium is convergence of matrix A
k
along with k to a zero matrix:
lim
k→∞
A
k
= 0.
The condition is satisﬁed if and only if all the eigenvalues of matrix A are smaller in modulus
than 1. If this is the case, all the shocks of the model are of transitory or stationary nature.
Economy being in a steady state ¯y, should it be subjected to shock ε
t
, it shall asymptotically
return to the same steady state ¯y.
Nevertheless, some of the structural shocks do not need to be transitory — they may have
permanent, or longterm, nonstationary effects. If a DSGE model speciﬁcation includes such
shocks, the condition of stability is not fulﬁlled. It is replaced with a more general condition.
Should the economy in a steady state ¯y be subjected to a shock ε
t
with permanent effects, it
returns to a steady state ˜y, different than the one in which it was before the occurrence of the
shock, so it is permitted that ˜y = ¯y. Therefore, it is required that upon the occurrence of a shock
with longterm effect, the model variables move away from the initial steady state ¯y by a ﬁnite
value at the most, i.e. that  ¯y − ˜y < ∞. The condition is satisﬁed if and only if:
lim
k→∞
A
k
=
˜
A
for a some matrix
˜
A with ﬁnite elements, i.e. if and only if the all the eigenvalues of matrix A are
not larger than 1 in modulus. This condition ensures that the response of the model to any type
of structural shock — with stationary or nonstationary effects — shall not be exploding. Shocks
with exploding effects are excluded for lack of economic interpretation.
2.5.3 Variance decomposition
Variance decomposition of endogenous variables answers the question which structural shocks
have the largest importance for the dynamics of given variables. Thus, it is possible to determine
which shocks and to what extent determine the dynamics of the particular variables.
As the reduced form of a DSGE model has the form of an autoregressive equation:
y
t
= Ay
t−1
+Bε
t
the iterative or mechanical point forecasts of the values of endogenous variables y
t+h
for
h = 1, 2, ..., H, read:
ˆy
t+h
=Aˆy
t+h−1
and in period t, since d y
t
is known (estimated), one assumes d ˆy
t
= d y
t
. Thus, the recursive
forecast error may be calculated in period t +h from:
∆
h
= y
t+h
− ˆy
t+h
=
h−1
k=0
A
k
Bε
t+h
44
DSGE model — anatomy
WORKING PAPER No. 83 45
2
as well as the covariance matrix:
(∆
h
) =
h−1
k=0
A
k
BB
(A
k
)
.
We shall identify the elements of matrix (∆
h
) by d
h
i j
, i.e. we shall assume that (∆
h
) = [d
h
i j
].
The diagonal elements d
h
ii
of matrix (∆
h
) are variances of forecast errors of endogenous
variables in the forecast horizon h. By identifying the i j element of matrix A
k
B(B)
(A
k
)
as p
k
i j
,
we get that the contribution of the jth structural shock to the variance of an ith endogenous
variable of the model in the horizon of the forecast h amounts to:
σ
h
i
( j) =
h−1
k=0
(p
k
i j
)
2
d
h
ii
.
2.5.4 Unconditional forecasts
The forecasting process may be of at least dual nature. Firstly, we may be interested in determi
nation of the socalled central forecast path, which usually corresponds to the expected value of
forecast variables or the modal value of their predictive distribution. Additionally, we may be
interested in quantiﬁcation of uncertainty related to the determined central path, namely the
determination of the distribution of forecast variables in the forecast horizon. The distribution is
called a predictive distribution.
Let us assume that the forecast horizon is h ≥ 1 periods, i.e. we are interested in calcula
tion of forecasts for periods T + 1, T + 2, ..., T + h. The forecasting process may refer both
to observable variables, i.e. Y
T+1
, Y
T+2
, ..., Y
T+h
observations, and endogenous variables, i.e.
y
T+1
, y
T+2
, ..., y
T+h
states. If the forecast variables are observable variables, their predictive
distribution is given by:
p
T
(Y
T+1
, Y
T+2
, ..., Y
T+h
) =
Θ
p(Y
T+1
, Y
T+2
, ..., Y
T+h
θ)p
T
(θ)dθ, (2.30)
where: p
T
(Y
T+1
, Y
T+2
, ..., Y
T+h
) = p(Y
T+1
, Y
T+2
, ..., Y
T+h
I
T
, ), and p
T
(θ) = p(θI
T
) is a poste
rior distribution of θ, where I
T
= Y. Determination of the central path of observable variables
involves, usually, the determination of the modal value of predictive posterior distribution, while
the determination of uncertainty related to the forecast based on the central path entails the
determination of the distribution of observable variables in the forecast horizon, i.e. distribution
with the density p
T
(Y
T+1
, Y
T+2
, ..., Y
T+h
). The density is a multidimensional integral with a large
support and may not be calculated analytically. Therefore, simulation methods are applied.
Below we present an algorithm simulating the predictive distribution of observable variables
and endogenous variables of the . Let us assume that y
+
= and Y
+
=.
1. Draw parameters θ parameters from the posterior distribution of the model , ie. θ ∈
p(θY).
2. Draw states y
T
from normal distribution with mean ( y
T
I
T
) and variance
( y
T
I
T
), ie. y
T
∼ N
( y
T
I
T
), ( y
T
I
T
)
.
45
DSGE model — anatomy
N a t i o n a l B a n k o f P o l a n d 46
2
3. Draw a sequence of structural shocks (ε) = (ε
T+1
, ε
T+2
, ..., ε
T+h
), in which every element
has a normal distribution with expected value equal to zero and Ψ, ie. ε
T+i
∼ N(0, Ψ),
i = 1, 2, ..., h. Using the sequence (ε) and equation of motion of states generate a respective
sequence of state variables ( y) = ( y
T+1
, y
T+2
, ..., y
T+h
).
4. Draw a sequence of observation errors (u) = (u
T+1
, u
T+2
, ..., u
T+h
), in which every element
has a normal distribution with expected value equal to zero and variance R, ie. u
T+i
∼N(0, R),
i = 1, 2, ..., h. Using (u) and measurement equation generate a respective sequence of
observable variable (Y) = (Y
T+1
, Y
T+2
, ..., Y
T+h
).
5. Assume y
+
= y
+
∪( y) and Y
+
= Y
+
∪(Y).
Distribution p(θY) in point 1 has been determined with the use of MetropolisHastings algo
rithm. The expected value ( y
T
I
T
) and variance ( y
T
I
T
) have been determined with the use
of the Kalman ﬁlter.
Repeating steps 15, for a large number of times leads to sets y
+
and Y
+
. The set y
+
includes
realizations of forecast paths of unknown values of endogenous variables in periods t + 1,
t +2,...,t +h. These paths come from the predictive distribution of endogenous variables. The
set Y
+
includes the realizations of forecast paths of unknown values of observable variables
in periods t +1, t +2,...,t +h. The paths come from the predictive distribution of observable
variables. By inspecting the statistical values of sets y
+
and Y
+
we may learn the central paths of
forecasts (e.g. by determining the modal values of sets y
+
and Y
+
) and quantify the uncertainties
related to them (e.g. by calculating the variance of forecasts). The approximation of predictive
distributions of states and observations entails the generation of histograms from the obtained
samples of y
+
and Y
+
respectively. Uncertainty of forecast can therefore be modelled by virtue
of consideration of the four sources of uncertainty, namely:
1. structural shocks variance,
2. measurement errors variance,
3. variance of the estimator of the current state y
T
, and:
4. variance of the estimator of parameters θ
7
.
The uncertainty resulting from the variance of measurement errors may be eliminated by
assuming (u) = (0, 0, ..., 0) in step three. The uncertainty resulting from the estimation of the
current state y
T
may be eliminated by assuming y
T
= ( y
T
I
T
) in step two. The uncertainty
resulting from the estimation of parameters θ may be eliminated by assuming in step one that θ
is always the mode of posterior distribution p(θY).
7
The uncertainty shall not be present, if parameters have been calibrated.
46
Part II
Speciﬁcation of DSGE SOE
PL−2009
model
47
SOE
PL2009
— general outline
N a t i o n a l B a n k o f P o l a n d 48
3
Chapter 3
SOE
PL−2009
— general outline
3.1 SOE
Euro
model — prototype of SOE
PL
family models
The family of SOE
PL
models originates directly from the estimated DSGE model of the euro
area developed by the analysts of the Central Bank of Sweden (Sveriges Riksbank), see (por.
Adolfson et al., 2005b, 2007a) — hereinafter the model shall be referred to as SOE
Euro
. The
Riksbank’s DSGE model for the euro area uses the pattern of a small open economy, which
enables its application to the description of the Polish economy. The Swedish analysts acted
similarly. They used the SOE
Euro
elements to construct a model describing the economy of
Sweden — this is how the RAMSES DSGE model of the Riksbank (see Adolfson et al., 2007b)
was created. The SOE
Euro
model was based on the ideas included in the model by L. Christiano,
M. Eichenbaum and Ch. Evans (Christiano et al., 2001, 2003, 2005). Also the inﬂuence of the
model by Smets and Wouters (2002, 2004) may be noticed. The three models deﬁne the line or
school of constructing DSGE models, from which the family of DSGE SOE
PL
models derives. An
important source of ideas we used when modifying the initial speciﬁcation of the models were
the subsequent works by L. Christiano, and in particular Altig et al. (2004a), Christiano et al.
(2007c,d) and Christiano et al. (2007a,b). More complete references are provided at the end of
the paper.
The family of SOE models refer to the economic ideas of the SOE
Euro
model by Riksbank. They
also use the broader understood methods of constructing and applying DSGE models, as well
as Bayesian estimation
1
. The software we use is a modiﬁcation (usually very farreaching) of
the scripts prepared at Riksbank for the purposes of SOE
Euro
. During the several years’ work
1
The authors of the SOE
Euro
model used the work and experience of other researchers, e.g. Schorfheide (2000),
the aforementioned F. Smets and R. Wouters, and also (partially) the authors of the Dynare package (M. Juillard, S.
Adjemian et al.). On the other hand, the experience of Riksbank has been used by the analysts of the European Central
Bank building the NAWM model for the purposes of the ECB (see. Christoffel et al., 2007a). The essence of the methods
applied by the ECB may be reconstructed by virtue of analysis of the construction of the YADA package (a collection of
scripts of the Matlab software), see Warne (2009).
48
3
SOE
PL–2009
– general outline
SOE
PL2009
— general outline
WORKING PAPER No. 83 49
3
and repeated reconstruction of the model and of the computation procedures, the logic of
work with a DSGE model has remained unchanged — upon loglinearisation of equations, the
SOE
PL
model is solved numerically (brought to the reduced form) with the AndersonMoore
algorithm (Anderson and Moore, 1985) and, then, expressed in the state space representation,
which enables — by the application of the Kalman ﬁlter — to determine the value of the
likelihood function and further on to apply the formalised (classical or Bayesian) techniques of
parameters estimation. It must be emphasised in that context that it is vital to construct a block
of measurement equations in the state space model (equation approximating the relations of the
variables of a theoretical model with observable variables). In our opinion, the emphasis put on
that aspect distinguishes the methods applied by Riksbank (and further by the ECB and by us)
from the techniques applied e.g. by a large number of Dynare package users.
As we have mentioned before, the starting point for the formation of the SOE
Euro
model by
Riksbank was the DSGE model of a closed economy by Christiano et al. (to which we shall
hereinafter refer as CEE model) — a model representing the newKeynesian point of view
with regard to the economic processes (see Gali, 2008; Woodford, 2003). J. Lindé and his
team have supplemented the CEE model speciﬁcation with issues related to international
exchange, following the hints included in the literature of the socalled new open economy
macroeconomics, see e.g. Lane (1999). As a result, a model was created in which the optimising
(rational) households maximise the utility originating (among others) from the consumption
of products manufactured from domestic and imported components. An inﬁnite number of
specialised agents manufacture domestic products and import consumption and investment
goods. The specialisation of manufacturers enables them to set prices in a manner characteristic
of imperfect competition. The pricesetting mechanism is related to the appearance of nominal
rigidities (delays in adjustment of prices to market conditions) — a phenomenon with which
the new Keynesian school explains the effectiveness of macroeconomic (monetary) policy in
a short run. The rigidities of the prices of imported and exported products (in the SOE
Euro
approximated with Calvo model (Calvo, 1983)) cause also that the exchange rate passthrough
is incomplete. Another characteristic of the new Keynesian school are real rigidities (an idea
derived from the Real Business Cycle school (RBC)), which along with the stochastic nature of
the technical progress explain the business cycle. In the SOE
Euro
model it has been assumed that
there exists consumption habit persistence, variable capital utilisation, and capital adjustment
costs.
The characteristic feature of the SOE
Euro
model was also the use of a single, nonstationary
disturbance (a stochastic trend) interpreted as the trend of technical progress (see also Altig
et al., 2004b, 2005). By including nonstationary disturbance in the speciﬁcation of a general
equilibrium model, we enable the growth of all of the variables indicating the trend (here e.g.
investments, consumption, GDP, foreign trade turnover, real wages), while the characteristics
of the growth (i.e. characteristics of nonstationary disturbance) are elements of the model
speciﬁcation. Hence, it is possible, at least partially, to exceed the shortterm analysis horizon
characteristic of business cycle models — the SOE
Euro
model had also the potential to explain
49
SOE
PL2009
— general outline
N a t i o n a l B a n k o f P o l a n d 50
3
midterm trends
2
.
Another distinguishing feature of the SOE
Euro
model was a large number of shocks (distur
bances) — larger than in other then constructed models — including structural shocks: several
technological shocks (stationary, nonstationary, stationary investmentspeciﬁc), markup shocks
(domestic products, imported investment goods, imported consumption goods, exported prod
ucts), preference shocks (consumption, labour supply, cash demand), as well as observable
disturbances (ﬁscal and originating from the world’s economy — derived from separately esti
mated SVAR models). The authors of the model have emphasised that thanks to the above, the
assessment of a relative role of the disturbance in shaping business cycles is possible, and the
assessment is more reliable as the parameters of the model are (largely) estimated.
In the group of solutions to which the authors of the SOE
Euro
model attached greater signif
icance was the socalled working capital channel, the solution determining the demand of
manufacturing ﬁrms for money (partial payment of wages in advance, ﬁnanced from a loan,
what translates into direct, positive inﬂuence of the interest rate (the cost of working capital
loan) on the marginal cost of production of intermediate products and inﬂation); the demand
for money by households results — traditionally — from the utility provided by cash to the
households. However, the implemented alternative of working capital (in stochastic version,
when the share of payments made in advance is subject to stochastic disturbance) proved to be
hardly useful in most of the applications. The reason was also a relatively small role of monetary
aggregates in the policy of central banks, therefore, the solution was reduced to deterministic
version or even marginalised. It seems to us now, that the addition of the ﬁnancial sector to
DSGE models would restore importance of this solution.
3.2 Family of SOE
PL
models, SOE
PL−2009
version
In the recent years, the original version of the SOE
Euro
model was subject to a series of
experiments conducted at the National Bank of Poland (NBP). There were attempts to estimate
the model (with Bayesian techniques) on the Polish data. We made experiments with various
collections of observable variables, the ﬁscal rule, the interest rate rule, the construction of
premium for foreign exchange risk and the construction of the tax system (more precisely the role
of taxes and national insurance contributions in the process of generation of the manufacturing
costs). Structural changes were taken into account, as well as stochastic wage markup. Whereas
some of the experiments implemented ideas present in other models (e.g. extended risk premium
— RAMSES), other attempted to better adjust the model to institutional framework of the Polish
economy (taxes, contributions). Experiments with stochastic wage markup were the consequence
of problems with interpretation of the labour supply disturbance identiﬁed in the sample, etc. In
2008 wider studies were conducted of the problems related to Poland’s accession to the euro
area. A special version of a DSGE SOE
PL
€
created for the purposes of that study (see Grabek and
Kłos, 2009) allowed (among others) to compare the method of absorption of disturbances in a
2
We shall remember, however, that technical progress has an exogenous nature, so the conclusions regarding a
severalyear’s horizon shall be very cautious.
50
SOE
PL2009
— general outline
WORKING PAPER No. 83 51
3
small open economy functioning within a monetary union and outside of it. In that version of
the model, the world economy consisted of two areas: the monetary union and the rest of the
world.
The above experience, as well as additional ideas resulting from separate research fed into
the construction of the newest version of the DSGE model named SOE
PL−2009
. An important
role in designing the changes in the speciﬁcation was also played by the analyses of accuracy
of the forecasts received from the earlier, experimental versions. As an example, the analyses
showed that forecasts of the dynamics of investment expenditures and investment deﬂator were
highly imprecise. Although it is very hard to model (and forecast) investment expenditures,
it did not justify such enormous errors. On the other hand, due to limited resources (time,
computational capacity of computers, etc.) we could not implement in the discussed version the
solutions that were the object of our former works (e.g. extension of the ﬁnancial sector). In the
current economic situation these could increase the chances for the model to better explain the
ongoing processes and, therefore, probably improve the accuracy of forecasts. In the hereinafter
presented version of the model there are also many problems that have not been solved — they
have only been outlined. In some cases the applied simpliﬁcations are disputable. An example
is the labour market in which (as in SOE
PL
, and SOE
PLEuro
) unemployment may not occur,
the issue of catching up and the changes in the share of foreign trade in the GDP, midterm
trends (appreciation) of foreign exchange rates, or the issue of indebtedness of households and
governments. At least some of the problems shall be the subject of our work in future but it is
worth mentioning in advance that the growing size of the model is a natural barrier and it may
not be expected that one model shall answer all the questions.
Although the SOE
PL−2009
version was created in consideration of forecasting applications, the
criterion of quality (accuracy) of forecasts has not dominated our choices. In any case, when
the quality of forecasts would force the resignation from logic or coherence — the economic
contents of the model — priority was given to the economic contents. The SOE
PL−2009
model is,
therefore, a dynamic stochastic general equilibrium model to be used for forecasting purposes
and not a ”forecasting model”. The preference for future applications is, however, reﬂected in
omission of interesting threads and deeper research of problems that are not directly related to
forecasting — here an example may be the hypotheses related to structural changes that have
probably occurred at the end of the 20th century and beginning of the 21st century in the Polish
economy.
3.3 Basic features of SOE
PL−2009
model
The speciﬁcation of SOE
PL−2009
is based on a framework typical in the class of DSGE models
derived from CEE model: a representative, forwardlooking and optimising consumer; imperfect
competition in intermediate products and labour markets; perfect competition in the markets of
ﬁnal products and capital services; nominal and real rigidities. Final products are assembled from
domestic and imported intermediate products. Merging of domestic and imported components
into ﬁnal products is made with the use of CES function, according to the logic of DixitStiglitz
aggregator (Dixit and Stiglitz, 1977). Finally two types of goods for domestic use (consumption
51
SOE
PL2009
— general outline
N a t i o n a l B a n k o f P o l a n d 52
3
and investment goods) are created, as well as products to be exported. A departure from the
aforesaid principle refers to the goods consumed by the government — the goods consist only
of the domestic component
3
. The prices of intermediate products (and also wages) are set
under imperfect competition with rigidities in the adjustment processes. The mechanism of
prices rigidity is based on a slightly modiﬁed Calvo model. Upon the solution of the respective
decisionmaking problems, the dynamics of prices (and wages) is described by Phillips curves, in
which inﬂation expectations appear explicitly (the anticipated rate of inﬂation).
It is assumed that there are inﬁnitely many rational (optimising and forwardlooking) households
in the economy. Households maximise utility attained from the consumption (with some habit
persistence), leisure and cash holdings under budget constraint. Maximisation of utility takes
place in the perspective of inﬁnite horizon with classical time discount. Households are the sole
administrator of labour force, while the unique qualiﬁcations of each of the households give
them a monopolistic position in the processes of wage negotiations (see Erceg et al., 2000).
Households are also the owners of ﬁxed capital assets and they receive income on account of
lease of the assets. The possibility of generating income from leasing the capital services causes
that households are interested in increasing this resource — investments. Another source of
income is interest from domestic deposits and deposits in foreign currencies. In the SOE
PL−2009
model we have assumed that households may deposit their savings in domestic currency and
in euro or dollar. Interest on each type of currency deposit is calculated taking into account
risk premium, different for each of the currencies. Additional sources of income are proﬁts of
intermediate goods producers as well as transfers from the budget. Generally, all the households
reﬂect the same consumption pattern, which is ensured by a special type of insurance levelling
the income. The technique enables merging of the pattern of a representative consumer with
differentiation of qualiﬁcation of the labour force supply by households (households are different
but their consumption standards, or more generally — the structure of expenditures, are the
same). All of the income and also consumption expenditures are burdened with a set of taxes.
Next to households, an inﬁnite number of domestic ﬁrms produce heterogeneous intermediate
goods by using the CobbDouglas technology with homogenous capital and labour inputs. The
labour and capital services are purchased in competitive market and set to minimise the cost
of production. Part of the capital lease rent (use of capital services) and wage bill must be
paid in advance and is ﬁnanced with a working capital loan. The marginal costs of domestic
production of intermediate products depend on the costs of capital and labour, the costs of
working credit and national insurance contributions paid be the employers (an additional charge
on the labour costs). Yet, given the speciﬁc nature of manufactured goods, manufacturers may
set prices of their products in a manner characteristic of monopolies, while the applied markup
has a stochastic nature. In order to set the prices of intermediate products, the manufacturers
solve a dynamic (intertemporal) decisionmaking problem, in which the Calvo type rigidity
mechanism is assumed.
3
SOE
Euro
only consumer and investment goods cover the imported component. Pragmatic reasons (avoidance of the
further extension of the model and complication of equations) caused that the goods consumed by the government do
not include imported components. A side effect of such simpliﬁcation is the understating of the steady state share of
foreign trade in GDP.
52
SOE
PL2009
— general outline
WORKING PAPER No. 83 53
3
A special solution applied in SOE
PL−2009
is the inclusion in the model of a disturbance rep
resenting the effects of ﬂuctuations in the prices of energy (e.g. oil). The disturbance has a
structural nature
4
and affects the economic processes through two channels. The ﬁrst one works
through the costs related to capital utilisation rate, such as it is proposed e.g. by Christiano et al.
(2007a), see also Leduc and Sill (2001). The second one is based on the direct inﬂuence of that
disturbance on the marginal costs of domestically manufactured intermediate products and,
thus, also on the prices of all of the ﬁnal products.
Final products (consumption, investment and export goods) consist of domestic intermediate
products and imported components. Importing ﬁrms purchase homogenous goods in the world’s
market (the euro area and the USA), transforming them into heterogeneous products (e.g. by
branding), set their prices and charge a markup (it is assumed that markups have a stochastic
nature). This is, thus, another segment of economy with imperfect competition. The marginal
cost of imports is the function of the price of goods in the world’s markets weighted with the
geographic structure.
Exporting ﬁrms purchase domestic and imported components and produce heterogeneous export
goods, the prices of which are, hence, set by maximising proﬁts under monopolistic competition.
The market in which the exporters sell their products is fully competitive, consumers in both
parts of the world pay identical prices. In other words, the geographical structure of exports has
no importance for the relations described in the model.
We deﬁne in the model two speciﬁc agents who do not have clear objective functions and
their behaviour is described with ad hoc rules. These are the government dealing with the
collection of taxes and expending income and the central bank that controls the interest rate.
The government plays only a passive role, distributes the income from taxes without creating the
budget deﬁcit. No public debt category appears here. The government assigns its expenditures
for public (collective) consumption and lumpsum transfers to households. Theoretically, the
transfers may be considered negative, which would mean the appearance of budget deﬁcit in
periods and repayment of the debt in others. The budget deﬁcit in such situation immediately
reduces the disposable income but the rational, optimising and forwardlooking households may,
in spite of that, maintain the level of expenditures, if they decide to avail of foreign deposits.
The whole reﬂects the Ricardian behaviour. The interest rate of the central bank follows the
interest rate rule. The rule is an effect of the manner in which the rational and forwardlooking
agents — households and ﬁrms — perceive the behaviour of the central bank. In other words,
the model characterises the point of view of economic agents and their perception of monetary
policy, instead of the actual decisionmaking process observed from inside of the bank.
The equilibrium at a micro scale results from ﬁrst order conditions for each group of optimising
agents. Decisions regarding consumption, investments, savings (and their currency structure),
the level of capital utilisation (etc.) bring households to the maximum of their expected utility.
Decisions regarding the production level and proportion of production factors bring ﬁrms to
minimise the costs, while the decisions regarding the prices (wages) lead to maximisation of
4
We have also carried out experiments with a version assuming the observable nature of such shock.
53
SOE
PL2009
— general outline
N a t i o n a l B a n k o f P o l a n d 54
3
expected proﬁts. Equilibrium conditions on every market depend, among others on the type of
competition and the speciﬁc values of elasticity of substitution — generally, it is also required
that all markets are clear. Macroscale balances are satisﬁed, which guarantees that demand for
services of production factors in each period are equal to the total supply of these factors, the
total value of expenditures is equal to the total income and the state budget revenues is equal to
expenditures.
In the SOE
PL−2009
model there are several groups of stochastic shocks (disturbances). The ﬁrst,
basic group, are structural unanticipated shocks. Economic growth is described with a stochastic
trend — nonstationary disturbance characterising technical progress, thanks to which the time
series reﬂecting the trend related to technical progress are modelled in consideration of that
process. More precisely, we assume that the trend is a resultant of two processes (nonstationary
disturbances). The ﬁrst directly impacts labour
5
such as in the original version of SOE
Euro
by
Riksbank. The second process impacts the prices of investment expenditures and ﬁxed assets as
proposed by Christiano et al. (2007c,d), see also Altig et al. (2004a, 2005); Burriel et al. (2009).
The total impact of both types of technical progress gives, however, identical dynamics to all of
the variables growing in steady state. Identical for all of the growing variables is the dynamics
of values and the division of the dynamics into the growth of volume and prices. An exception
are investment expenditures, whose dynamics of value is the same as other variables but the
division into the growth of volume and price is different — accordingly to the characteristics
of the second of nonstationary disturbances and the applied technology. Thus, in a long run
real investments do not have to grow at the same rate as consumption or GDP. Other structural
disturbances (technology, preferences, markups, risk premium, etc.) have stationary nature — it
has been assumed that they shall have the nature of ﬁrstorder autoregressive process (AR(1)).
Another group of disturbances are observable disturbances (also of unanticipated nature). As
we have mentioned before, the world’s economy is heterogeneous in SOE
PL−2009
and consists of
two areas: the euro area and the dollar area (the rest of the world identiﬁed with the USA). The
key characteristics of the world’s economy and their relations are approximated with a structural
vector autoregression (SVAR) model. The SVAR model is estimated separately and is used to
describe the mechanisms generating the shocks (here observable disturbances) from the foreign
environment. In similar manner the fragments of the ﬁscal block have been treated — here also
the SVAR model has been used for approximation of the interdependencies between budget
expenditures and (a part of) budget revenues.
Additionally, beside of the unanticipated structural and observable disturbances, which are sort
of a standard, in SOE
PL−2009
the possibility of occurrence of disturbances anticipated by the
agents has been permitted. Formally, the structure of anticipated disturbances coincides with the
supplementation of the disturbance structure with MA class component — we have used here
the convention proposed by SchmittGrohé and Uribe (2008), a similar version is also proposed
by Christiano et al. (2007a,b). The anticipated disturbances are, however, an option of which
we have not availed in the current version (and, therefore, omit it in the further description).
5
It is also proposed here to extend the interpretation of the disturbance for demographic effects.
54
SOE
PL2009
— general outline
WORKING PAPER No. 83 55
3
Nevertheless, the analytical potential of this solution seems to be considerable, so we shall return
to the issue of the role of anticipated disturbances at further stages of the research.
Looking from another angle, the SOE
PL−2009
speciﬁcation brings the model to neoclassical
responses in the long run, with explicitly and clearly deﬁned steady state
6
, while the short
term effects — thanks to inclusion in the model of real and nominal rigidities — shall have a
more Keynesian nature. For example, ﬁscal stimulation of the economy is effective only in a
short run. Also the stimulation of economic activity with monetary policy instruments may be
effective in a short run however, in the long run the only growth determinant is the (exogenous)
technical progress, and the possible deviations from the longterm trends are absorbed already
in medium term, while the costs of such adjustments may exceed the previous gains. The
characteristics of adjustment processes, i.e. model dynamics, is a consequence of optimising
(forwardlooking) behaviours of agents (the decisionmaking problems of agents formally
derived from intertemporal optimisation problems). The function of institutions managing the
macroeconomic policy has been adjusted to the above logic (i.e. the rules of behaviours derived
by the rational agents based on the observation of the activities of institutions). Their policy
excludes any forms of game with the agents and is timeconsisted. As a result, the model may be
used solely for making analyses in which the condition shall be satisﬁed at least approximately.
The existence and functions of money result from the households’ objective function (holding
some cash resources is useful for households). Additionally, demand for money is also reported
by ﬁrms that need to pay in advance for some share of labour services and of capital
7
. Stochastic
monetary effects (disturbances in cash demand among households and demand among ﬁrms)
— as a result of several experiments made before the year 2009 — proved to be unsuccessful
as the disturbances of this class have not increased the potential of the model and, therefore,
accordingly as in SOE
Euro
they have been excluded. In consequence, the monetary variables do
not appear in the set of observable variables.
6
We use interchangeably the terms of steady state and longterm equilibrium.
7
A competitive method of deriving cash models is the deﬁnition of the costs of transaction. In effect of optimisation
of such costs, agents decide to possess cash resources, see e.g. Coenen et al. (2006).
55
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 56
4
Chapter 4
Decisionmaking problems,
equilibrium conditions,
macroeconomic balance of the
model
4.1 Growth
The SOE
PL−2009
is a model of exogenous, stochastic growth which is driven by changes in the
level of technology (z
t
). The growth rate of technology, µ
z,t
≡
z
t
z
t−1
, is governed by the stochastic
process:
µ
z,t
=
1 −ρ
µ
z
µ
z
+ρ
µ
z
µ
z,t−1
+
µ
z
,t
,
µ
z
,t
∼ N
0, µ
z
σ
µ
z
, µ
z,t
= µ
z
,
where ρ
µ
z
is the persistence coefﬁcient, and µ
z
is a longterm growth rate of technology.
The technological trend has a neutral nature — it refers to all of the macroeconomic categories
characterised by growth. Beside of that, we assume the existence of a technological trend speciﬁc
for capital/investment goods
Ψ
t
, whose changes, µ
Ψ,t
≡
Ψ
t
Ψ
t−1
, are governed by the process:
µ
Ψ,t
=
1 −ρ
µ
Ψ
µ
Ψ
+ρ
µ
Ψ
µ
Ψ,t−1
+
µ
Ψ
,t
µ
Ψ
,t
∼ N
0, µ
Ψ
σ
µ
Ψ
, µ
Ψ,t
= µ
Ψ
.
The presence of an additional technological trend speciﬁc for capital goods, by use of capital as a
factor of production (see Chapter 4.3.2), is translated into other macroeconomic categories and
extends the neutral technological trend. The common technological trend (z
+
t
) for all growing
56
4
Decisionmaking problems, equilibrium conditions,
macroeconomic balance of the model
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 57
4
variables, except capital goods, may be presented as:
z
+
t
= z
t
Ψ
1−
t
, µ
z
+
,t
= µ
z,t
µ
1−
Ψ,t
,
where is the share of capital in production. The level of technology for capital goods
z
+
t
Ψ
t
and all the other categories
z
+
t
allow to express the growing variables in a stationary form
(usually denoted with small letters), i.e.:
y
t
≡
Y
t
z
+
t
, i
t
≡
I
t
z
+
t
Ψ
t
, etc. (4.1)
Additionally, nominal variables are stationarized with the use of the level of prices
P
d
t
, e.g.
nominal wages are translated into stationary real wages:
w
t
≡
W
t
P
d
t
z
+
t
. (4.2)
Therefore, at the ﬁnal stage the whole model may be presented with the use of stationary
variables and explicitly determined steady state.
4.2 Foreign economy
Domestic economy functions in the environment of two foreign economies: the euro area and the
rest of the world. Interactions with those economies entail exchange of goods and ﬁnancial ﬂows.
We assume that the currency of the rest of the world is the dollar and in the euro area — the
euro. We use, therefore, three nominal exchange rates: dollar/zloty, euro/zloty and dollar/euro,
denoted respectively by: S
u
t
, S
e
t
, S
x
t
. The exchange rates satisfy:
S
u
t
= S
x
t
S
e
t
. (4.3)
Additionally, we deﬁne real exchange rates:
x
x
t
≡
S
x
t
P
u
t
P
e
t
, x
e
t
≡
S
e
t
P
e
t
P
c
t
, x
u
t
≡
S
e
t
S
x
t
P
u
t
P
c
t
, (4.4)
where P
u
t
is the level of prices in the rest of the world, P
e
t
is the level of prices in the euro area,
and P
c
t
is the level of domestic consumer prices. To real exchange rates, analogically as for the
nominal rates, the following applies:
x
u
t
= x
x
t
x
e
t
. (4.5)
4.3 Producers
There are ﬁve markets of intermediate goods: domestic goods, imported consumption, investment
and export goods, as well as export goods. In each of the markets there are inﬁnitely many
agents (continuum determined in the [0,1] interval) manufacturing heterogeneous intermediate
57
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 58
4
products of a given type that are aggregated to a homogenous ﬁnal product representing the
production of the given market.
4.3.1 Aggregators
Heterogeneous intermediate products must be aggregated
1
. For each market we assume the
existence of inﬁnitely many ﬁrms (the agents do not consume resources or generate added
value), which operate under perfect competition and use the same production function. They
purchase heterogeneous intermediate products and transform them into a homogenous ﬁnal
product (taking the prices of intermediate products and the price of the ﬁnal product as given).
The production function of the ﬁnal good in each of the markets O (O ∈ {Y, C
m
, I
m
, X
m
, X})
takes the form of the CES function:
O
t
=
1
0
O
1
λ
o
t
i,t
di
λ
o
t
, 1 ≤ λ
o
t
<∞, o ∈ {d, mc, mi, mx, x}, (4.6)
where O
t
is the production of the ﬁnal good, O
i,t
is the production by the ith intermediate goods
producer, λ
o
t
is the markup in the market o, and o identiﬁes market: domestic products (d),
imported consumption goods (mc), imported investment goods (mi), imported goods intended
for export (mx), export products (x). Markups speciﬁc for each of the markets are described
with stochastic processes:
λ
o
t
=
1 −ρ
λ
o
λ
o
+ρ
λ
o λ
o
t−1
+
λ
o
,t
,
λ
o
,t
∼ N
0, λ
o
σ
λ
o
, λ
o
t
= λ
o
, (4.7)
where λ
o
is the value of markup in steady state.
Proﬁt maximisation by the aggregator leads to the demand function for intermediate products
of the ith producer:
O
i,t
=
P
o
i,t
P
o
t
−
λ
o
t
1−λ
o
t
O
t
, (4.8)
where P
o
t
is the price of the homogenous ﬁnal product in market o, P
o
i,t
is the price of the
intermediate product of the ith producer.
Using equations (4.6) and (4.8) we obtain the equation for the price of the homogenous ﬁnal
product in the given market:
P
o
t
=
1
0
P
o
i,t
1
1−λ
o
t
di
1−λ
o
t
, o ∈ {d, mc, mi, mx, x}. (4.9)
1
Aggregation of heterogeneous products into a homogenous product is a technical operation necessary from the
point of view of the model operability, however, we may apply economic interpretation thereto.
58
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 59
4
4.3.2 Domestic intermediate goods ﬁrms
The producers of domestic intermediate goods are the only actual generators of the GDP. Using
the CobbDouglas production function, with production technology identical for all of the
producers, they use individually determined labour and capital inputs to produce:
Y
i,t
=ε
t
z
t
H
i,t
1−
K
i,t
, (4.10)
where H
i,t
and K
i,t
are the inputs of labour (hours) and capital services determined by the i
th
producer. The total factor productivity
ε
t
is described with a stochastic process:
ε
t
= 1 −ρ
ε
+ρ
ε
ε
t−1
+
ε,t
,
ε,t
∼ N
0, σ
ε
, ε
t
= 1. (4.11)
The optimal values for inputs of capital and labour are determined based on the problem of
costs minimisation:
min
K
i,t
,H
i,t
R
f w
t
F
τ
t
W
t
H
i,t
1 +τ
s
t
+R
f k
t
F
τ
t
R
k
t
K
i,t
−λ
t
P
i,t
Y
i,t
−z
1−
t
ε
t
K
i,t
H
1−
i,t
, (4.12)
where W
t
is the nominal wage, τ
s
t
is the rate of national insurance contribution paid by the
employer, R
k
t
is the gross nominal rental rate per unit of capital services, λ
t
is the Lagrange
multiplier. We assume that in each period a fraction of the wage and capital fund, ν
w
and ν
k
,
must be ﬁnanced with a working capital loan hence the presence of effective gross nominal
interest rates, R
f w
t
and R
f k
t
, in the cost function, given by:
R
f w
t
≡ ν
w
R
t−1
+1 −ν
w
, R
f k
t
≡ν
k
R
t−1
+1 −ν
k
, (4.13)
where R
t−1
is the gross nominal interest rate. We assume also that the use of labour and capital
services involves the use of “energy”, whose costs are represented by F
τ
t
function. The F
τ
t
(·, ·)
function, speciﬁed explicitly only at the level of loglinearised form, is a linear function of
structural shock representing the dynamics of the prices of energy (e.g. oil):
π
oil
t
=
1 −ρ
π
oil
π
oil
+ρ
π
oil π
oil
t−1
+
π
oil
,t
,
π
oil
,t
∼ N
0, σ
π
oil
. (4.14)
First order conditions of the decisionmaking problem (4.12) with respect to H
i,t
, K
i,t
and λ
t
are:
R
f w
t
F
τ
t
W
t
1 +τ
s
t
= (1 −) λ
t
P
i,t
z
1−
t
ε
t
K
i,t
H
−
i,t
,
R
f k
t
F
τ
t
R
k
t
=λ
t
P
i,t
z
1−
t
ε
t
K
−1
i,t
H
1−
i,t
,
Y
i,t
= z
1−
t
ε
t
K
i,t
H
1−
i,t
.
Based on the ﬁrst order conditions of the problem of costs minimisation, we arrive at the
equation of real marginal cost of the domestic intermediate goods producers:
59
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 60
4
mc
d
t
≡ λ
t
=
1
ε
t
_
1
_
_
1
1 −
_
1−_
r
k
t
R
f k
t
_
_
w
t
R
f w
t
_
1−
F
τ
t
. (4.15)
The market of domestic intermediate products is characterised with monopolistic competition,
which means that manufacturers produce heterogeneous products and may set their prices. At
the same time, there are some limitations in the spirit of Calvo price setting (Calvo (1983)). In
every period any of the manufacturers, with probability 1 −ξ
d
, may set the optimal price of its
output
_
P
d,new
t
_
. With probability ξ
d
the price cannot be set in the optimal way and it is then
indexed to previous inﬂation (with weight κ
d
) and the current inﬂation target
2
(with weight
1 −κ
d
):
P
d
t+1
=
_
π
d
t
_
κ
d
_
π
c
t+1
_
1−κ
d
P
d
t
. (4.17)
If a producer is allowed to reoptimise its price, it sets its price to maximise the ﬂow of future
proﬁts, assuming that it will not be allowed to reoptimise the price in the future. Thus, the
decisionmaking problem takes the form:
max
P
d,new
t
t
∞
s=0
υ
t+s
_
βξ
d
_
s
_
_
π
d
t
. . . π
d
t+s−1
_
κ
d
_
π
c
t+1
. . . π
c
t+s
_
1−κ
d
P
d,new
t
Y
i,t+s
− MC
d
i,t+s
Y
i,t+s
_
,
where υ
t+s
is the marginal utility of the households’ nominal income
3
, and β is a discount
factor. When solving the proﬁt maximisation problem above, the producer takes into account
the demand for their ouput given by equation (4.8). The solution of the problem takes the form
of the Phillips curve for domestic intermediate goods, in which the main inﬂation determinants
become the real marginal costs
_
mc
d
t
_
, given by (4.15), and markup
_
λ
d
t
_
, described with the
exogenous process (4.7)
4
.
The last problem the manufacturers have to cope with is the determination of the optimal level of
employment (number of full time employees), based on the number of hours worked determined
in the process of costs minimisation. The process of adjusting employment involves Calvotype
2
The inﬂation target
_
π
c
t
_
has a stochastic nature and is given by exogenous process:
π
c
t
=
_
1 −ρ
π
c
_
π
c
+ρ
π
c π
c
t−1
+
π
c
,t
,
π
c
,t
∼ N
_
0, π
c
σ
π
c
_
. (4.16)
In steady state inﬂation target is equal to the steady state level of inﬂation
_
π
c
≡ π
d
_
.
3
Due to the fact that in each period the proﬁt generated by the ﬁrm is transferred to households, the proﬁt in the
particular period is weighted with the marginal utility of the households’ nominal income.
4
Below we present an example of a Phillips curve (binding for each of the markets of intermediate products), already
in a loglinearised but relatively legible form:
´ π
o
t
=
βµ
1 +κ
o
βµ
_
´ π
o
t+1
−
´
π
c
t+1
_
+
κ
o
1 +κ
o
βµ
_
´ π
o
t−1
−
´
π
c
t
_
+
1 +κ
o
βµρ
π
1 +κ
o
βµ
´
π
c
t
+
_
1 −ξ
o
βµ
__
1 −ξ
o
_
ξ
o
_
1 +κ
o
βµ
_
_
´
λ
o
t
+¯ mc
o
t
_
.
(4.18)
Current inﬂation depends on the difference between the inﬂation target and past/expected inﬂation, directly on the
very inﬂation target and the current standing of producers (their markup and marginal costs). Also the impact of price
rigidities can be seen — the smaller the price rigidity (smaller ξ
o
), the larger is the importance of marginal costs and
markup in the market for the current inﬂation.
60
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 61
4
rigidities — with probability 1 −ξ
e
the producer is allowed to set the level of employment in an
optimal way, while with the probability ξ
e
the producer cannot change the level of employment.
When producer is allowed to reoptimise the level of employment, the decisionmaking problem
takes the form:
min
E
new
i,t
∞
s=0
βξ
e
s
n
i
E
new
i,t
−H
i,t+s
2
, (4.19)
where n the number of hours per employee. The solution of the decisionmaking problem
describes the level of employment in the economy.
4.3.3 Importers
The imported consumption, investment and export goods make three separate markets of
imported products. In each of the markets the importers purchase foreign goods (from the
euro area and the rest of the world — we assume the stability of the geographic structure
of the import), and differentiate them. Heterogeneous products are, then, purchased by the
aggregators and transformed into homogenous ﬁnal products. The monopolistic competition
implies that importers may set prices of their products, while the process runs similarly to the
case of domestic intermediate goods producers (with speciﬁc ξ
o
, κ
o
and λ
o
, (o ∈ {mc, mi, mx})
parameters for each market of imported products). Solving the problem of maximisation of
the importers’ proﬁt, we arrive at three Phillips curves in the form compliant with the formula
presented in the footnote — full versions are provided in the Appendix.
The marginal costs, with ﬁxed geographic structure of import, take the form:
mc
mc
t
≡
S
u
t
P
u
t
P
mc
t
ω
mc,u
+
S
e
t
P
e
t
P
mc
t
(1 −ω
mc,u
) ,
mc
mi
t
≡
S
u
t
P
u
t
P
mi
t
ω
mi,u
+
S
e
t
P
e
t
P
mi
t
1 −ω
mi,u
,
mc
mx
t
≡
S
u
t
P
u
t
P
mx
t
ω
mx,u
+
S
e
t
P
e
t
P
mx
t
(1 −ω
mx,u
) ,
(4.20)
where ω
mc,u
, ω
mi,u
and ω
mx,u
determine the share of the rest of the world in the basket of
imported consumption, investment and export goods.
The real marginal costs and markups in the markets, described with exogenous processes (4.7),
determine inﬂation for imported consumption goods, investment goods and goods intended
for export. The total demand for imported consumption and investment goods depends on the
decisions of households (see Chapter 4.4), while the demand for import of goods intended for
export is determined by the exporters.
4.3.4 Exporters
Similarly to the domestic intermediate goods producers and importers, exporters produce, under
monopolistic competition, heterogeneous export goods
X
i,t
, for which they may set prices
P
x
i,t
. Due to the fact that production is intended for the world market, the prices set by
exporters are expressed in dollars. As we assume free ﬂow of products between the euro area
61
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 62
4
and the rest of the world, the price for the euro area market is the same price converted into euro
based on the USD/EUR exchange rate
_
P
x
i,t
S
x
t
_
. The process of setting the price runs similarly as
in the case of domestic goods producers and importers with parameters ξ
x
, κ
x
i λ
x
speciﬁc for
the export market. After solving the problem of proﬁt maximisation we obtain the Phillips curve
for the export market.
Export good of the i
th
exporter is produced with the use of domestic products
_
X
d
i,t
_
and
imported products
_
X
m
i,t
_
. The production function takes the form:
X
i,t
=
_
ω
1
η
x x
x
_
X
m
i,t
_
η
x x
−1
η
x x
+
_
1 −ω
x
_
1
η
x x
_
X
d
i,t
_
η
x x
−1
η
x x
_
η
x x
η
x x
−1
, (4.21)
where η
x x
is the elasticity of substitution between domestic and imported products, and ω
x
is the
share of the imported component. Each exporter must solve the problem of costs minimisation:
min
X
m
i,t
,X
d
i,t
P
mx
t
X
m
i,t
+ P
d
t
X
d
i,t
, (4.22)
where P
mx
t
is the price of imported component, P
d
t
is the price of the domestic component,
subject to (4.21).
The solution of the problem leads to the marginal costs of exporters:
mc
x
t
=
1
S
e
t
S
x
P
x
t
_
ω
x
_
P
mx
t
_
1−η
x x
+
_
1 −ω
x
_
_
P
d
t
_
1−η
x x
_ 1
1−η
x x
.
The marginal cost together with the markup described with the exogenous process (4.7) is
reﬂected in the Phillips curve for the export market.
Assuming that consumption and investments in the euro area and in the rest of the world are
determined based on CES functions with domestic export being one of the inputs, the demand
for domestic export on the part of both economies and both types of products is expressed with
the following equations:
C
x,u
t
=
_
P
x
t
P
u
t
_
−η
f ,u
C
u
t
, I
x,u
t
=
_
P
x
t
P
u
t
_
−η
f ,u
I
u
t
,
C
x,e
t
=
_
S
x
t
P
x
t
P
e
t
_
−η
f ,e
C
e
t
, I
x,e
t
=
_
S
x
t
P
x
t
P
e
t
_
−η
f ,e
I
e
t
,
where C
e
t
_
C
u
t
_
and I
e
t
_
I
u
t
_
are consumption and investments in the euro area (rest of the world),
while η
f ,e
_
η
f ,u
_
is the elasticity of substitution between domestic export products and the euro
area (the rest of the world) products. Assuming that the income of foreign economies is entirely
divided between consumption and investments, we may express the demand for domestic export
as a function of foreign income:
X
t
≡ C
x,e
t
+ I
x,e
t
+C
x,u
t
+ I
x,u
t
=
_
P
x
t
P
u
t
_
−η
f ,u
Y
u
t
+
_
S
x
t
P
x
t
P
e
t
_
−η
f ,e
Y
e
t
. (4.23)
62
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 63
4
Thus, the demand for domestic export depends on the relation of export prices and world prices
and income (output) abroad.
4.4 Households
The households maximise utility consisting of consumption, leisure and cash. Households
provide labour and capital services to the producers of domestic intermediate products. In each
period households divide their income between domestic and foreign deposits, consumption,
investments and purchase/sale of new, installed capital, as well as cover the cost of maintenance
of capital that has not been lent to producers. The income of households consists of domestic and
foreign deposits plus interest, remuneration for the labour and capital services, as well as proﬁts
transferred in the form of dividend. All of the income is adequately taxed, and additionally
direct transfers from the state budget are allowed. Moreover, we assume that ﬁnancial markets
are complete. This enables households to acquire state contingent securities making them
homogenous with regard to the possessed resources and incurred expenditures, thanks to which
the model may be made operational.
Households are characterised with the so called internal habit persistence, which means that
utility is derived not so much from the absolute current level of consumption as from a change
in the level of consumption in reference to the previous period. The utility function of a jth
household takes the form:
t
∞
s=0
β
t+s
_
_
_
_
ζ
c
t+s
ln
_
C
j,t+s
− bC
j,t+s−1
_
−ζ
h
t+s
A
L
h
1+σ
L
j,t
1 +σ
L
+A
q
ζ
h
t+s
_
Q
j,t
z
t
P
t
_
1−σ
q
1 −σ
q
_
_
_
_
, (4.24)
where C
j,t
is the consumption in period t, h
j,t
is the supply of labour (hours), Q
j,t
is cash holdings,
σ
L
is the inverse of the elasticity of labour supply with respect to wage, σ
q
is the elasticity of
demand for cash with respect to interest rate. The preferences regarding consumption, leisure
and cash holdings, ζ
c
t
, ζ
h
t
and ζ
q
t
, are given by exogenous processes:
ζ
l
t
= 1 −ρ
ζ
l +ρ
ζ
l ζ
l
t−1
+
ζ
l ,
ζ
l ∼ N
_
0, σ
ζ
l
_
, ζ
l
t
= 1, l ∈ {c, h, q}. (4.25)
Consumption and investment goods purchased by households consist of domestic and imported
products:
C
t
=
_
_
1 −ω
c
_
1
η
c
_
C
d
t
_
η
c
−1
η
c
+
_
ω
c
_
1
η
c
_
C
m
t
_
η
c
−1
η
c
_
η
c
η
c
−1
,
I
t
=
_
_
1 −ω
i
_
1
η
i
_
I
d
t
_
η
i
−1
η
i
+
_
ω
i
_
1
η
i
_
I
m
t
_
η
i
−1
η
i
_
η
i
η
i
−1
,
(4.26)
where C
d
t
_
I
d
t
_
and C
m
t
_
I
m
t
_
are domestic and import components of consumption (investments),
ω
c
_
ω
i
_
is the share of import in consumption (investments) and η
c
_
η
i
_
is the elasticity of
63
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 64
4
substitution between domestic consumption (investment) goods and imported goods.
We assume that aggregation is made in such a way as to maximise the values C
t
and I
t
subject
to budget constraints:
P
d
t
C
d
t
+ P
mc
t
C
m
t
= P
c
t
C
t
, P
t
I
d
t
Ψ
t
+ P
mi
t
I
m
t
= P
i
t
I
t
, (4.27)
where P
d
t
, P
mc
t
and P
mi
t
are the prices of the domestic components and imported consumption
and investment components. Solving the problems of consumption and investment maximisation,
we arrive at the equation of demand for domestic and imported components of consumption
and investment:
C
d
t
=
_
1 −ω
c
_
_
P
c
t
P
d
t
_
η
c
C
t
, C
m
t
= ω
c
_
P
c
t
P
mc
t
_
η
c
C
t
,
I
d
t
=
_
1 −ω
i
_
_
P
i
t
Ψ
t
P
d
t
_
η
i
I
t
, I
m
t
= ω
i
_
P
i
t
Ψ
t
P
mi
t
_
η
i
I
t
Ψ
t
.
The prices of ﬁnal consumption goods
_
P
c
t
_
and investment goods
_
P
i
t
_
are then expressed with
the following equations:
P
c
t
=
_
_
1 −ω
c
_
_
P
d
t
_
1−η
c
+ω
c
(P
mc
)
1−η
c
_ 1
1−η
c
,
P
i
t
Ψ
t
=
_
_
1 −ω
i
_
_
P
d
t
_
1−η
i
+ω
i
_
P
mi
t
_
1−η
i
_ 1
1−η
i
.
(4.28)
The households’ physical capital stock
_
K
j,t+1
_
evolves according to:
K
j,t+1
= (1 −δ) K
k,t
+Υ
t
F
_
I
j,t
, I
j,t−1
_
+∆
j,t
, (4.29)
where δ is the capital depreciation rate, ∆
j,t
is the purchase/sale of new, installed capital.
Function F is the function of transformation of investment expenditures into physical capital:
F
_
I
j,t
, I
j,t−1
_
=
_
1 −
˜
S
_
I
j,t
I
j,t−1
__
I
j,t
. (4.30)
Function
˜
S is not explicitly speciﬁed, we assume that:
˜
S (x) =
˜
S
(x) = 0 and
˜
S
(x) ≡
˜
S
>0, x = µ
+
z
µ
Ψ
. (4.31)
This means that full transformation of investments into physical capital takes place when invest
ment expenditures grow at the steady state level. In other words, ﬂuctuations in investment
expenditures generate costs, which creates the mechanism of smoothening of investment expen
ditures. An additional factor affecting the effectiveness of transformation of investments into
capital goods is the exogenous process Υ
t
, called the investmentspeciﬁc technology shock or
64
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 65
4
effectiveness of transformation of investment into capital:
Υ
t
= 1 −ρ
Υ
+ρ
Υ
Υ
t−1
+
Υ,t
,
Υ,t
∼ N
0, σ
Υ
, Υ
t
= 1. (4.32)
The physical capital stock is fully or partially leased to the intermediate goods producers in
the form of capital services K
j,t
. With u
j,t
, u
j,t
≡
K
j,t
K
j,t
, we denote utilization rate of capital (in
steady state u = 1). We assume that incomplete use of the capital resource generates cost for
households, depending on the utilitzation rate — F
τ
a,t
a(u
j,t
)
K
j,t
Ψ
t
. Function F
τ
a,t
represents a part
of the cost depending on the changes in the prices of energy and — similarly to the function F
τ
t
— is a function of energy price shock
π
oil
t
(the solution is based on the work by Christiano
et al. (2007a)). Function a(u
j,t
) is not explicitly speciﬁed, we assume only that a(1) = 0 and
a
≥0.
The budget constraint of households takes the form of:
M
j,t+1
+S
e
t
B
e
j,t+1
+S
e
t
S
x
t
B
u
j,t+1
+ P
c
t
C
j,t
1 +τ
c
t
+ P
i
t
I
t
+ P
d
t
F
τ
a,t
a(u
j,t
)
Ψ
t
K
j,t
+ P
k
,t
∆
t
= R
t−1
M
j,t
−Q
j,t
+Q
j,t
+
1 −τ
k
t
Π
t
+
1 −τ
p
t
R
k
t
u
j,t
K
j,t
+
1 −τ
y
t
1 −τ
w
t
W
j,t
h
j,t
+R
e
t−1
Φ
A
e
t−1
z
+
t−1
,
t
s
e
t
s
e
t−1
,
˜
φ
e
t−1
S
e
t
B
e
j,t
+R
u
t−1
Φ
A
u
t−1
z
+
t−1
,
t
s
u
t
s
u
t−1
,
˜
φ
u
t−1
S
e
t
S
x
t
B
u
j,t
+ TR
t
+ D
j,t
−τ
k
t
R
b
t−1
−1
M
j,t
−Q
j,t
+
R
e
t−1
Φ
A
e
t−1
z
+
t−1
,
t
s
e
t
s
e
t−1
,
˜
φ
e
t−1
−1
S
e
t
B
e
j,t
+
R
u
t−1
Φ
A
u
t−1
z
+
t−1
,
t
s
u
t
s
u
t−1
,
˜
φ
u
t−1
−1
S
e
t
S
x
t
B
u
j,t
+ B
e
j,t
S
e
t
−S
e
t−1
+ B
u
j,t
S
e
t
S
x
t
−S
e
t−1
S
x
t−1
+τ
p
t
P
d
t
F
τ
t
a(u
j,t
)
Ψ
t
K
j,t
+τ
p
t
P
t
P
k
,t
δK
j,t
,
(4.33)
where M
j,t
are domestic ﬁnancial assets, B
e
j,t
and B
u
j,t
are assets denominated in euro and dollar,
P
k
,t
is the relative price of capital goods, τ
c
t
is the consumption tax rate, τ
k
t
is the capital tax
rate (on interest from deposits and dividends), τ
p
t
is the corporate income tax rate, Π
t
are the
proﬁts of intermediate goods producers (domestic, exporters and importers), TR
t
are lumpsum
transfers from state budget, D
j,t
is the income from state contingent securities, τ
y
t
and τ
w
t
are the
personal income tax rate and the rate of national insurance contribution paid by an employee.
Foreign assets, B
e
j,t
and B
u
j,t
, bear interest according to the interest rates for the euro area, R
e
t
,
and the rest of the world, R
u
t
, adjusted for risk premium, see e.g. (Adolfson et al., 2007a, page 8)
65
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 66
4
and (SchmittGrohé and Uribe, 2003; Engel, 1996):
Φ
_
A
e
t−1
z
+
t−1
,
t
s
e
t
s
e
t−1
,
˜
φ
e
t−1
_
for assets denominated in euro,
Φ
_
A
u
t−1
z
+
t−1
,
t
s
u
t
s
u
t−1
,
˜
φ
u
t−1
_
for assets denominated in dollar,
(4.34)
where:
A
e
t
≡
S
e
t
B
e
t+1
P
d
t
, A
u
t
≡
S
u
t
B
u
t+1
P
t
, s
e
t
≡
S
e
t
S
e
t−1
, s
u
t
≡
S
u
t
S
u
t−1
,
while
˜
φ
e
t
and
˜
φ
u
t
are the risk premium shocks described with stochastic processes:
˜
φ
e
t
= ρ
˜
φ
e
˜
φ
e
t−1
+
˜
φ
e
,t
,
˜
φ
e
,t
∼ N
_
0, σ
˜
φ
e
_
,
˜
φ
e
t
= 0,
˜
φ
u
t
= ρ
˜
φ
u
˜
φ
u
t−1
+
˜
φ
u
,t
,
˜
φ
u
,t
∼ N
_
0, σ
˜
φ
u
_
,
˜
φ
u
t
= 0.
(4.35)
Risk premium for assets in the given currency depends on the position in those asses at the
scale of the whole economy, while function Φ is strictly decreasing in A
e
t
_
A
u
t
_
. For total foreign
assets, a
t
≡
A
t
z
+
t
=
A
e
t
z
+
t
+
A
u
t
z
+
t
, we assume that in steady state they are equal to 0, while foreign assets
denominated in euro are positive (then a
u
= −a
e
).
Based on the utility function (4.24), budget constraint (4.33) and the law of motion of capital
(4.29) we may formulate an optimisation problem and the Lagrange functional:
max
c
j,t
,i
j,t
,u
j,t
,k
j,t+1
,m
j,t+1
,q
j,t
,
˜
∆
j,t
,b
e
j,t+1
,b
u
j,t+1
0
∞
t=0
β
t
t
,
t
= ζ
c
t+s
ln
_
C
j,t+s
− bC
j,t+s−1
_
−ζ
h
t+s
A
L
h
1+σ
L
j,t
1 +σ
L
+A
q
ζ
q
t+s
_
Q
j,t
z
t
P
t
_
1−σ
q
1 −σ
q
+υ
t
_
R
t−1
_
M
j,t
−Q
j,t
_
+Q
j,t
+ TR
t
+ D
j,t
+
_
1 −τ
k
t
_
Π
t
+
_
1 −τ
p
t
_
R
k
t
u
j,t
K
j,t
+
_
1 −τ
y
t
_
_
1 −τ
w
t
_
W
j,t
h
j,t
+R
e
t−1
Φ
_
A
e
t−1
z
+
t−1
,
t
s
e
t
s
e
t−1
,
˜
φ
e
t−1
_
S
e
t
B
e
j,t
+R
u
t−1
Φ
_
A
u
t−1
z
+
t−1
,
t
s
u
t
s
u
t−1
,
˜
φ
u
t−1
_
S
e
t
S
x
t
B
u
j,t
+τ
p
t
P
d
t
F
τ
a,t
a(u
j,t
)
Ψ
t
K
j,t
+τ
p
t
P
t
P
k
,t
δK
j,t
−τ
k
t
_
_
R
t−1
−1
__
M
j,t
−Q
j,t
_
+
_
R
e
t−1
Φ
_
A
e
t−1
z
+
t−1
,
t
s
e
t
s
e
t−1
,
˜
φ
e
t−1
_
−1
_
S
e
t
B
e
j,t
+
_
R
u
t−1
Φ
_
A
u
t−1
z
+
t−1
,
t
s
u
t
s
u
t−1
,
˜
φ
u
t−1
_
−1
_
S
e
t
S
x
t
B
u
j,t
+ B
e
j,t
_
S
e
t
−S
e
t−1
_
+ B
u
j,t
_
S
e
t
S
x
t
−S
e
t−1
S
x
t−1
_
_
−
_
M
j,t+1
+S
e
t
B
e
j,t+1
+S
e
t
S
x
t
B
u
j,t+1
+ P
c
t
C
j,t
_
1 +τ
c
t
_
+ P
i
t
I
t
+ P
d
t
_
F
τ
a,t
a(u
j,t
)
Ψ
t
K
j,t
+ P
k
,t
∆
t
___
+ω
t
_
(1 −δ) K
k,t
+Υ
t
F
_
I
j,t
, I
j,t−1
_
+∆
j,t
−K
j,t+1
_
66
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 67
4
where υ
t
and ω
t
are Lagrange multipliers. First order conditions of the above decisionmaking
problem create a system of equations determining consumption, investments, physical capital
stock and its utilization rate, cash holdings, foreign assets denominated in euro, marginal utility
of nominal income and nominal exchange rate (USD/PLN). After stationarization, for which we
additionally deﬁne:
υ
t
z
+
t
P
d
t
≡ ψ
z+,t
(4.36)
and upon the consideration of the solution symmetry, the conditions may be presented as
follows:
ζ
c
t
c
t
− bc
t−1
1
µ
z
+
,t
−βb
t
ζ
c
t+1
c
t+1
µ
z
+
,t+1
− bc
t
−ψ
z
+
,t
γ
cd
t
_
1 +τ
c
t
_
= 0,
υ
t
P
d
t
p
k
,t
Ψ
t
=ω
t
,
−ψ
z
+
,t
γ
id
t
+ψ
z
+
,t
p
k
,t
Υ
t
F
1
_
i
t
, i
t−1
, µ
z
+
,t
µ
Ψ,t
_
+β
t
ψ
z
+
,t+1
p
k
,t+1
Υ
t+1
F
2
_
i
t+1
, i
t
, µ
z
+
,t+1
µ
Ψ,t+1
_
= 0,
−ψ
z
+
,t
+β
t
ψ
z
+
,t+1
π
t+1
µ
z
+
,t+1
__
1 −τ
k
t+1
_
_
R
t
−1
_
+1
_
= 0,
−ψ
z
+
,t
p
k
,t
+β
t
ψ
z
+
,t+1
µ
z
+
,t+1
µ
Ψ,t+1
_
_
1 −τ
p
t+1
__
¯r
k
t+1
u
t+1
− F
τ
t+1
a(u
t+1
)
_
+τ
p
t+1
p
k
,t+1
δ + p
k
,t+1
(1 −δ)
_
= 0,
¯r
k
t
= F
τ
a,t
a
(u
j,t
),
ζ
q
t
A
q
q
−σ
q
t
−ψ
z
+
,t
_
1 −τ
k
t
__
R
b
t−1
−1
_
= 0,
−ψ
z
+
,t
+β
t
ψ
z
+
,t+1
µ
z
+
,t+1
π
t+1
_
s
e
t+1
R
e
t
Φ
_
a
e
t
, s
e
t+1
s
e
t
,
˜
φ
e
t
__
1 −τ
k
t+1
_
+τ
k
t+1
_
= 0,
−ψ
z
+
,t
+β
t
ψ
z
+
,t+1
µ
z
+
,t+1
π
t+1
_
s
e
t+1
s
x
t+1
R
u
t
Φ
_
a
u
t
,
t
s
u
t+1
s
u
t
,
˜
φ
u
t
_
_
1 −τ
k
t+1
_
+τ
k
t+1
_
= 0.
The labour market is characterised by monopolistic competition — households provide heteroge
neous labour services
_
h
j,t
_
and set wages
_
W
j,t
_
. In a similar manner to creating homogenous
67
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 68
4
ﬁnal products by the aggregators, also the heterogeneous labour services are aggregated into a
homogenous labour services
_
H
t
_
, which may be then used by the domestic intermediate goods
producers:
H
t
=
_
_
_
1
0
_
h
j,t
_ 1
λ
w
t
d j
_
_
λ
w
t
, 1 ≤λ
w
t
< ∞,
where wage markup is described with an exogenous process:
λ
w
t
=
_
1 −ρ
λ
w
_
λ
w
+ρ
λ
w λ
w
t−1
+
λ
w
,t
,
λ
w
,t
∼ N
_
0, λ
w
σ
λ
w
_
, λ
w
t
=λ
w
. (4.37)
The process of wage setting runs similarly to the process of price setting by the producers (Calvo
model) — in each period, with probability 1 −ξ
w
, a household may set optimal wage; with
probability ξ
w
wage cannot be reoptimised, it may only be indexed to previous inﬂation of
consumer prices (with weight κ
w
), the current value of the inﬂation target (with weight 1 −κ
w
)
and the current technology growth:
W
j,t+1
=
_
π
c
t
_
κ
w
_
π
c
t+1
_
1−κ
w
µ
z
+
,t+1
W
j,t
.
When a household is allowed to set the wage in an optimal way, it maximises the difference
between the utility of income on account of wage and disutility of leisure reduction:
max
W
new
t
t
∞
s=0
_
βξ
w
_
s
_
−ζ
h
t+s
A
L
_
h
j,t+s
_
1+σ
L
1 +σ
L
+υ
t+s
_
1 −τ
y
t+s
__
1 −τ
w
t+s
_
×
×
_
_
π
c
t
. . . π
c
t+s−1
_
κ
w
_
π
c
t+1
. . . π
c
t+s
_
1−κ
w
×
×
_
µ
z
+
,t+1
. . . µ
z
+
,t+s
_
W
new
t
_
h
j,t+s
_
.
The ﬁrst order condition of the above decisionmaking problem leads to the equation of real
wage in economy.
4.5 Behaviour of other agents
Apart from optimising agents (households, ﬁrms), the SOE
PL
model explicitly considers the
existence of two additional agents — the central bank and the government. The agents have not
been assigned any formal, autonomous object functions. It is only assumed that the purpose
of the central bank is to control price dynamics, and the only instrument the bank has at its
disposal is the interest rate. The other agent — the government — fulﬁls a passive function of
managing budget funds, i.e. charges taxes from which expenditures are ﬁnanced.
68
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 69
4
4.5.1 Central bank
The reactions of the central bank
5
are characterised from an external point of view, i.e. rational
and anticipating active participants of the economic processes — ﬁrms and households. From the
actually applied monetary policy, the agents derive characteristics of the rules of behaviour of the
central bank: interest rate persistence, sensitivity of the interest rate to inﬂation deviations from
the reference point (inﬂation target
6
), sensitivity of the interest rate to GDP deviations from its
level in a steady state, etc. The interpretation of activities of the central bank and independent
conclusions regarding the perspectives of the inﬂation processes, as well as further policy of
the central bank must be reliable. In microeconomic decisionmaking problems of households
and ﬁrms there appear values related to the interest rate policy, therefore, optimal decisions
of setting prices and wages depend on the values. On the other hand, it is silently assumed
that the central bank is not trying to carry out any form of game intended for exploitation of
the perception error created by the policy. In other words, we assume that monetary policy is
reliable. Possibly, temporary incoherence or inconsistency is absorbed by the ﬂuctuations of the
inﬂation target.
As a consequence of the above mentioned assumptions, the interest rate rule describing the
activities of the central bank is deﬁned directly in a loglinearised form, without prior reference
to the decisionmaking problem. The form of the rule is sufﬁciently general for covering a
possibly broad spectrum of interest rate policies. During estimation we allow the possibility of a
structural change occurrence, i.e. change in the value of some of the parameters of the rule.
We assume that the real exchange rate present in the rule is the effective rate deﬁned as:
x
ue
t
= x
u
t
+
1 −γ
xux
x
x
t
, (4.38)
where γ
xux
determines the currency structure of international settlements. The standard version
of the interest rate rule has then the form:
R
t
=ρ
R
R
t−1
+
1 −ρ
R
r
π
c
π
c
t
+ r
π
π
c
t−1
−
π
c
t
+ r
y
y
t−1
+ r
x
x
eu
t−1
+ r
∆π
∆ π
c
t
+ r
∆y
∆y
t
+ε
R,t
.
(4.39)
In the current version of the model we assume that r
π
c ≡ 1, therefore,
π
c
t
shall be interpreted as
the perception of the policy of the central bank (inﬂation target) by the agents. The disturbance
of the interest rate (monetary policy, monetary disturbance) ε
R,t
, — contrary to other shocks
appearing in the model — is deﬁned as innovation. The interpretation of the component
5
Depending on the type of interest rate to which the model refers, an extending interpretation of the agent or group
of agents responsible for interest rate change is possible. In a typical situation, when interest rate is derived from the
interbank market, there are grounds for claiming that interest rate ﬂuctuations result at the same time from the activities
of the central bank and the responses to the current events in the interbank market. Such interpretation eliminates the
automatic assignment of any interest rate changes to the decisions of the central bank.
6
We differentiate the inﬂation target declared by the central bank from the inﬂation target being the result of
perception of the inﬂation processes and monetary policy by the optimising agents. The second concept allows for a
stochastic nature of the target oscillating around the stationary value (steady state). The stationary value of the target
(in the second meaning) may but does not have to be compliant with the target declared by the institution responsible
for the monetary policy. In each case when we speak about the inﬂation target, we mean the second concept. This model
does not refer to the ofﬁcially declared targets of the monetary policy.
69
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 70
4
suggesting itself — discretionary component of the monetary policy — is, however, disputable.
Traditionally, it has been assumed that monetary disturbance is not correlated with variables
present in the information set (see e.g. Christiano et al., 1998). Therefore, it should be assumed
that the information set of agents (model) and the information set of the central bank are
different.
4.5.2 Government
In the class of models deriving from the research of L. Christiano, such as in the known models
built for the purposes of central banks, the function of government is reduced to administration
of the state income redistribution. Taxes are not an instrument of ﬁscal policy — their level
is relatively constant, there are only observed temporary deviations from the longterm level
(steady state). The deviations have the nature of disturbances. In the SOE
PL−2009
model the
disturbances are approximated with the AR(1) process or a SVAR model, such as it is made
in DSGE models with exogenous variables. Although the ﬂuctuations of tax rates have been
treated in a slightly simpliﬁed manner, the speciﬁcation of decisionmaking problems of ﬁrms
and households shows the role of national insurance contributions and taxes in determining
the budget of the agents and, therefore, their behaviour. In the SOE
PL−2009
model there are
explicitly present national insurance contributions paid by the employers (τ
s
) and employees
(τ
w
), the capital tax (τ
k
), the tax on consumption (τ
c
), the income tax (τ
y
) and the corporate
income tax (τ
p
). This is, thus, a set of the basic ﬁscal charges imposed on agents. Therefore, the
ﬁscal policy impact is represented on the micro level.
Macroeconomic consequences of ﬁscal policy are treated in a more simpliﬁed manner. The
model is based on the pattern of a representative consumer; one of the functions included
in the ﬁscal policy is, thus, such redistribution of the state revenues that the differentiated
households reﬂect a uniform pattern of consumption behaviour. Formally, in the households
budget constraint the category of special state contingent insurance (D) appears, however, the
system is not characterised in detail, so it must be ﬁnanced by the budget. An element of the
system is also the pension and disabilitypension block. We do not deﬁne separate institutions
dealing with management of such class of funds — in the applied pattern of a representative
agent there is no place for households living only of the pension payments, therefore, national
insurance contributions are treated as the revenues of the government (budget) that are spent
on a current basis.
Within the above outlined redistribution system, the existence of budget deﬁcit and public
debt has not been explicitly provided. The Ricardian logic of optimising and forwardlooking
households rationalises this solution (today’s budget deﬁcit is balanced with the growth of
taxes tomorrow, namely reduction of disposable income, therefore, it requires a reduction in
spending already today). Technically, transfers (TR) received by households may be negative,
so households shall immediately and directly reduce disposable income. If households are not
inclined to reduce their expenditures, they may ﬁnance them with negative foreign deposits.
Thus, the reasoning here is cohesive. Emphasis should be placed on the fact that the aforesaid
approximation may be considered to be sufﬁcient for an economy that only sporadically experi
ences a deﬁcit or surplus in the budget (public ﬁnance). If the budget deﬁcit has a structural
70
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 71
4
nature and the public debt reﬂects no trend of stabilisation, the role of the government sector
may be larger than it appears from the proposed approximation.
The amount of government expenditures (G
t
) has, similarly as the tax rates, a stochastic
nature. The ﬂuctuations of government expenditures (government consumption, collective
consumption) around a steady state are approximated with a (separately estimated) SVAR
model. More precisely, a separate model has been built trying to explain jointly the ﬂuctuations
of government expenditures and the rates of income tax, corporate income tax and tax on
consumption. The model has the following form:
Γ
0
´ τ
t
=Γ
0
Γ (L) ´ τ
t−1
+Γ
0
ε
τ,t
, ε
τ,t
∼ N
_
0, Σ
τ
_
, (4.40)
´ τ
t
≡
_
´ τ
p
t
, ´ τ
y
t
, ´ τ
c
t
,
´
¯
G
t
_
, ε
τ,t
≡
_
ε
τ
p
,t
, ε
τ
y
,t
, ε
τ
c
,t
, ε
τ
¯ G
,t
_
,
Structural decomposition of disturbances is given by:
Γ
0
ε
τ,t
= B
0,τ
u
t
, u · u
= I . (4.41)
The estimated matrices of parameters (Γ
0
) i (B
0,τ
) enable the determination of the matrix of
disturbance covariance (Σ
τ
). The ﬂuctuations (deviations from the steady state) of the tax on
capital and national insurance rates are treated as structural disturbances and are approximated
in the DSGE model with the form of AR(1) process.
Joint reﬂection of budget expenditures and income leads to the following speciﬁcation:
P
t
G
t
+
_
TR
t
+ D
t
_
= R
t−1
_
M
t+1
− M
t
_
+τ
c
t
P
c
t
C
t
+
_
τ
w
t
+τ
s
t
+τ
y
t
_
1 −τ
w
t
_
_
W
t
H
t
+τ
k
t
_
Π
t
+
_
R
t−1
−1
_
_
M
t
−Q
t
_
+
_
R
e
t−1
Φ
e
(a
e
t−1
, ...,
˜
φ
e
t−1
) −1
_
S
e
t
B
e
t
+B
e
t
_
S
e
t
−S
e
t−1
_
+
_
R
u
t−1
Φ
u
(a
u
t−1
, ...,
˜
φ
u
t−1
) −1
_
S
u
t
B
u
t
+B
u
t
_
S
u
t
−S
u
t−1
_
_
+τ
p
t
_
R
k
t
u
t
K
t
−
1
Ψ
t
P
t
F
τ
a,t
a
_
u
t
_
K
t
− P
t
P
k
,t
δK
t
_
.
(4.42)
A new solution introduced to the discussed version of the SOE
PL
model is the use of the above
speciﬁcation in building a consolidated (integrated) balance of total expenditures and income,
which means that the changes e.g. in government expenditures impact e.g. net foreign assets.
The issue is discussed in more detail in the following paragraphs.
4.6 Macroeconomic balance conditions
At macro scale, equilibrium is identiﬁed with simultaneous balancing of income and expenditures
of households and the government, the aggregated supply and demand (factor resources), as
well as the balance of the banking sector and international transactions (payment balance). As a
result, we obtain a system guaranteeing that in each period the total of expenditures is equal to
the total of income, the stream of services of productivity factors is sufﬁcient for manufacturing
71
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 72
4
domestic intermediate products, which in combination with the carried out import and export
ensures the levelling of demand and supply at macro scale. Net export is balanced with foreign
net assets.
The system of macroeconomic balance is considerably much more extended in comparison with
the previous versions of SOE
PL
and SOE
Euro
. In particular, for the ﬁrst time macroeconomic
balance of income and expenditures of households has been introduced, calculated at current
prices, which requires, among others, the derivation of ﬁrms’ proﬁt accounts (being the income of
households). A new solution also consists in the explicit derivation of the balance of expenditures
and income of the government (equivalent to state budget), which upon integration with the
balance of income and expenditures of households covers all of the expenditures at the scale of
the whole economy and the methods of their ﬁnancing (sources of income). The balance enables
the determination of net foreign assets, so it replaces a classical balance of payments.
4.6.1 Proﬁts in economy
We take into account ﬁrms manufacturing intermediate products, importers and exporters, who
transfer their proﬁts to households where the proﬁts are taxed with the tax on dividend (tax on
capital). We assume that total proﬁt in economy is the sum of the proﬁts generated in particular
sectors of economy (domestic production, export and import), i.e.:
Π
t
=Π
d
t
+Π
x
t
+Π
m
t
.
According to the suggestion of Christiano et al. (2007b, page 26–28), proﬁts generated by ﬁrms
may be estimated at macro scale with the use of marginal costs assessment, taking into account
the ineffectiveness of allocation resulting from price setting with the use of the Calvo model.
More precisely, proﬁts result from a difference between the marginal cost and the actual price.
Contrary to the previous versions of the model, we assume that ﬁxed costs are absent in domestic
production, instead proﬁts appear in steady state.
Domestic products manufacturers
In the case of intermediate products manufacturers we have:
Π
d
t
=
_
1
0
P
d
j t
Y
j t
d j − P
d
t
mc
d
t
_
_
1
0
Y
j t
d j
_
= P
d
t
Y
t
− P
d
t
mc
d
t
_
_
_
_
Y
t
_
¯
P
d
t
P
d
t
_
λ
d
t
1−λ
d
t
_
_
_
_
.
The value
_
¯
P
d
t
P
d
t
_
λ
d
t
1−λ
d
t
is the allocation inefﬁciency. Such expression may be approximated with
the function of markup f (λ
d
t
, ...) ≡ f
d
t
, however, in the case of loglinearisation it is also justiﬁed
to treat the price relations as equal one. Taking into account the above, proﬁts at macro scale
may be estimated as:
Π
d
t
= P
d
t
Y
t
− P
d
t
mc
d
t
_
Y
t
f
d
t
_
. (4.43)
72
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 73
4
Making it stationary and then bringing down to ﬁxed prices, we obtain:
d
t
≡
Π
d
t
P
d
t
z
+
t
= y
t
−mc
d
t
y
t
f
d
t
= y
t
1 −mc
d
t
f
d
t
, (4.44)
where the marginal cost mc
d
is expressed with equation (4.15).
Proﬁts in export
Assuming, further, that proﬁts in export (calculated in domestic currency) are subject to domestic
tax, omitting the existence of ﬁxed costs, we have:
Π
x
t
= P
x
t
S
e
t
S
x
t
C
x
t
+ I
x
t
1 −mc
x
t
f
x
t
, (4.45)
where f
x
t
is deﬁned as f
d
t
. After making it stationary, we obtain:
x
t
=
P
x
t
S
e
t
S
x
t
P
d
t
1 −mc
x
t
f
x
t
c
x
t
+i
x
t
= γ
xd
t
1 −mc
x
t
f
x
t
c
x
t
+i
x
t
. (4.46)
Proﬁts in import
Out of the two possible methods of deﬁning proﬁts (on the micro level with further aggregation
or with direct reference to the macro scale), we have used the macroeconomic convention:
Π
m
t
= P
mc
t
C
m
t
+ P
mi
t
I
m
t
+ P
mx
t
X
m
t
−S
e
t
S
x
t
P
t
C
m
t
+ I
m
t
+X
m
t
.
Bringing it to the stationary version, we arrive at:
m
t
=
Π
m
t
z
+
t
P
d
t
= γ
mcd
t
c
m
t
+γ
mid
t
i
m
t
+γ
mxd
t
x
m
t
−γ
,d
t
c
m
t
+i
m
t
+ x
m
t
=
=
γ
mcd
t
−γ
,d
t
c
m
t
+
γ
mid
t
−γ
,d
t
i
m
t
+
γ
mxd
t
−γ
,d
t
x
m
t
.
(4.47)
4.6.2 Income and expenditures of households
We assume that the total of net income and expenditures of households is balanced. Merging
the income side with the side of expenditures, we arrive at the aggregated version of the budget
condition of households:
73
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 74
4
_
1 −τ
k
t
__
R
t−1
−1
_
_
M
t
−Q
t
_
+
_
1 −τ
k
t
_
R
e
t−1
Φ
_
a
e
t−1
, s
e
t
, s
e
t−1
,
˜
φ
e
t−1
_
S
e
t
B
e
t
+τ
k
t
S
e
t−1
B
e
t
+
_
1 −τ
k
t
_
R
u
t−1
Φ
_
a
u
t−1
, s
x
t
, s
e
t
, s
x
t−1
, s
e
t−1
,
˜
φ
u
t−1
_
S
e
t
S
x
t
B
u
t
+τ
k
t
S
e
t−1
S
x
t−1
B
u
t
+
_
1 −τ
k
t
_
Π
t
+
_
1 −τ
p
t
_
_
R
k
t
u
t
K
t
−
1
Ψ
t
P
t
F
τ
a,t
a
_
u
t
_
K
t
_
+τ
p
t
P
t
P
k
,t
δK
t
+
_
1 −τ
y
t
_
_
1 −τ
w
t
_
W
t
H
t
+ M
t
+ TR
t
+ D
t
− M
t+1
−S
e
t
B
e
t+1
−S
e
t
S
x
t
B
u
t+1
−
_
1 +τ
c
t
_
P
c
t
C
t
− P
i
t
I
t
− P
t
P
k
,t
∆
t
= 0.
(4.48)
4.6.3 State budget
The term state budget or ”government budget” means here, approximately, the sector of public
ﬁnance and a fragment of the ﬁnancial sector specialising in pension insurance.
The standard form of the balance of revenues and expenditures of a government stems from the
original version of the SOE
Euro
. Upon adaptation to the domestic tax system and the possibility
of depositing savings in two currency markets (next to domestic deposits), we arrive at (see
equation (4.42)):
P
t
G
t
+
_
TR
t
+ D
t
_
= R
t−1
_
M
t+1
− M
t
_
+τ
c
t
P
c
t
C
t
+
_
τ
w
t
+τ
s
t
+τ
y
t
_
1 −τ
w
t
_
_
W
t
H
t
+τ
k
t
_
Π
t
+
_
R
t−1
−1
_
_
M
t
−Q
t
_
+
_
R
e
t−1
Φ
e
(a
e
t−1
, ...,
˜
φ
e
t−1
) −1
_
S
e
t
B
e
t
+B
e
t
_
S
e
t
−S
e
t−1
_
+
_
R
u
t−1
Φ
u
(a
u
t−1
, ...,
˜
φ
u
t−1
) −1
_
S
u
t
B
u
t
+B
u
t
_
S
u
t
−S
u
t−1
_
_
+τ
p
t
_
R
k
t
u
t
K
t
−
1
Ψ
t
P
t
F
τ
a,t
a
_
u
t
_
K
t
− P
t
P
k
,t
δK
t
_
.
(4.49)
In the above formula we assume that national insurance contributions paid by the employees
and the employer represent the revenues of the government budget.
4.6.4 Monetary balance
As in the original version of the SOE model, we deﬁne the broad money dynamics as:
µ
t
≡
M
t+1
M
t
=
m
t+1
m
t
µ
z
+
,t
π
d
t
. (4.50)
The equation (after loglinearisation) shall be used for explaining the resource of money. We
notice that in the steady state the following applies:
π
d
=
µ
µ
z
+
.
74
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 75
4
The banking system must provide ﬁnancing of ﬁrms with working capital loans and must provide
cash for households. Therefore, we have the dependence:
ν
k
t
F
τ
t
R
k
t
K
t
+ν
w
t
F
τ
t
1 +τ
s
r
W
t
H
t
= M
t+1
−Q
t
, (4.51)
where M
t+1
= µ
t
M
t
.
Making it stationary and expressing in ﬁxed prices, we obtain the following form:
ν
k
t
F
τ
t
r
k
t
k
t
µ
Ψ,t
µ
z
+
,t
+ν
w
t
F
τ
t
1 +τ
s
r
w
t
H
t
=
µ
t
m
t
π
d
t
µ
z
+
,t
−q
t
. (4.52)
4.6.5 Balance of payment
The basic function of the payment balance in the model is the determination of the value of net
foreign assets compliant with the level of activity in foreign trade, expenditures and income in
economy. Foreign assets may be determined at least in two ways: from the classical version of
balance, when we consider net export, and from the integrated balance of expenditures and
incomes. In SOE
PL−2009
both solutions were implemented. The selection was to be done by
virtue of experiments, estimation of parameters, analyses of the model characteristics (impulse
response function) and analyses of forecasting accuracy. Yet, the experiments showed that the
alternative of the income balance increases the potential possibility of the model, therefore, it
was the one ﬁnally chosen. Below we present both alternatives.
Classical version
In the payment balance, as in the SOE
PL
€
model, the geographic structure of net foreign assets
is presented explicitly (see Kłos, 2008). However, we add import to the speciﬁcation, and it
becomes a component of export. Doubts appear at the moment of choosing the price index to be
used for expressing import at current prices. Here we assume that in the balance of payment
import is calculated at global prices, i.e. prices at which the products were purchased. Such
approach complies with the approach assumed in the previous versions of the SOE model.
S
e
t
S
x
t
B
u
t+1
+S
e
t
B
e
t+1
= S
e
t
S
x
t
P
x
t
C
x
t
+ I
x
t
−S
e
t
S
x
t
P
t
C
m
t
+ I
m
t
+X
m
t
+S
e
t
S
x
t
B
u
t
R
u
t−1
Φ
u
a
u
t−1
,
˜
φ
u
t−1
, . . .
+S
e
t
B
e
t
R
e
t−1
Φ
e
a
e
t−1
,
˜
φ
e
t−1
, . . .
.
(4.53)
The classical form of the payment balance (after bringing it to stationary state) has the form:
a
u
t
+a
e
t
=γ
x,d
t
x
t
−γ
,d
t
c
m
t
+i
m
t
+ x
m
t
+s
e
t
s
x
t
a
u
t−1
π
d
t
µ
z
+
,t
R
u
t−1
Φ
u
a
u
t−1
,
˜
φ
u
t−1
, s
e
t
, s
x
t
, s
e
t−1
, s
x
t−1
+s
e
t
a
e
t−1
π
d
t
µ
z
+
,t
R
e
t−1
Φ
e
a
e
t−1
,
˜
φ
e
t−1
, s
e
t
, s
e
t−1
,
(4.54)
75
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
N a t i o n a l B a n k o f P o l a n d 76
4
where:
Φ
i
(a
i
t
, s
i
t+1
, s
i
t
,
˜
φ
i
t
) = exp
_
−
˜
φ
i
a
_
a
i
t
− a
i
_
−
˜
φ
i
s
_
t
s
i
t+1
s
i
t
−1
_
+
˜
φ
i
t
_
, i ∈ {u, e} (4.55)
and
a
u
t
=
S
e
t
S
x
t
B
u
t+1
P
d
t
z
+
t
, a
e
t
=
S
e
t
B
e
t+1
P
d
t
z
+
t
, γ
x,d
t
=
S
e
t
S
x
t
P
x
t
P
d
t
,
γ
,d
t
=
S
e
t
S
x
t
P
t
P
d
t
, s
i
t
=
S
i
t
S
i
t−1
i ∈ {u, e, x}.
Version with the income balance
Merging the income and expenditure balance of households (4.48) and the state budget (4.42),
we arrive at the formula, which upon simpliﬁcation takes the following form:
P
g
t
G
t
+ P
i
t
I
t
+ P
c
t
C
t
+S
e
t
B
e
t+1
+S
e
t
S
x
t
B
u
t+1
= Π
t
+
_
1 +τ
s
t
_
W
t
H
t
+
_
R
k
t
u
t
−
1
Ψ
t
P
t
F
τ
a
a
_
u
t
_
_
K
t
+
_
R
t−1
−1
__
M
t+1
−Q
t
_
+R
e
t−1
Φ
_
a
e
t−1
, ...,
˜
φ
e
t−1
_
S
e
t
B
e
t
+R
u
t−1
Φ
_
a
u
t−1
, ...,
˜
φ
u
t−1
_
S
e
t
S
x
t
B
u
t
.
Taking into account balance of the banking sector, we obtain:
P
g
t
G
t
+ P
i
t
I
t
+ P
c
t
C
t
+S
e
t
B
e
t+1
+S
e
t
S
x
t
B
u
t+1
= Π
t
+
_
1 +τ
s
t
_
_
F
τ
t
_
R
f w
t
−1
_
+1
_
W
t
H
t
+
_
_
F
τ
t
_
R
f k
t
−1
_
+1
_
R
k
t
u
t
−
1
Ψ
t
P
t
F
τ
a,t
a
_
u
t
_
_
K
t
+R
e
t−1
Φ
_
a
e
t−1
, ...,
˜
φ
e
t−1
_
S
e
t
B
e
t
+R
u
t−1
Φ
_
a
u
t−1
, ...,
˜
φ
u
t−1
_
S
e
t
S
x
t
B
u
t
.
(4.56)
Finally, the payment balance obtained by the integration of income and expenditures of house
holds, the income and expenditures of the budget and the balance of the banking sector (after
making it stationary and bringing down to real categories) takes the form:
a
e
t
+ a
u
t
+ g
t
+γ
id
t
i
t
+γ
cd
t
c
t
+
F
τ
a,t
a
_
u
t
_
k
t
µ
Ψ,t
µ
z
+
,t
=
=
d
t
+
m
t
+
x
t
+
_
1 +τ
s
t
_
_
F
τ
t
_
R
f w
t
−1
_
+1
_
w
t
H
t
+
_
F
τ
t
_
R
f k
t
−1
_
+1
_
r
k
t
k
t
µ
Ψ,t
µ
z
+
,t
+R
e
t−1
Φ
_
a
e
t−1
, ...,
˜
φ
e
t−1
_
s
e
t
a
e
t−1
π
d
t
µ
z
+
,t
+R
u
t−1
Φ
_
a
u
t−1
, ...,
˜
φ
u
t−1
_
s
e
t
s
x
t
a
u
t−1
π
d
t
µ
z
+
,t
.
(4.57)
The above equation shows that expenditures calculated at the macro level of economy (consumer
expenditures, investment expenditures, government expenditures, new foreign deposits (net
foreign assets), and capital adjustment are ﬁnanced from the proﬁts, income from labour, income
76
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
WORKING PAPER No. 83 77
4
from capital and revenues from (mature) foreign deposits. Net foreign assets amount in total to
a
t
= a
e
t
+a
u
t
.
4.6.6 The aggregate resource constraint
The starting point for further considerations is the formula in which all of the components
are expressed at ﬁxed prices. In resources constraint we omit the factor characterising the
ineffectiveness of allocation (the effect of Calvo price settings). The obtained inequality has the
form:
G
t
+C
d
t
+
1
Ψ
t
I
d
t
+ F
τ
a,t
a
u
t
K
t
+X
d
t
≤
t
z
1−
t
K
t
H
1−
t
(4.58)
or assuming equality:
G
t
z
+
t
+
C
d
t
z
+
t
+
1
Ψ
t
z
+
t
I
d
t
+ F
τ
a,t
a
u
t
K
t
+
X
d
t
z
+
t
= y
t
(4.59)
and using the solution of the decision problem of exporters (4.22), we may determine:
x
d
t
=
X
d
t
z
+
t
=
1 −ω
x
f
x
t
ω
x
γ
mx,d
t
1−η
x x
+
1 −ω
x
η
x x
1−η
x x
X
t
z
+
t
.
Merging the above and assuming equality of resources and demand for the resources, plus
bringing it to the stationary form, we obtain:
g
t
+c
d
t
+i
d
t
+
1 −ω
x
f
x
t
ω
x
γ
mx,d
t
1−η
x x
+
1 −ω
x
η
x x
1−η
x x
x
t
=
Ψ
t−1
z
+
t−1
z
1−
t
z
+
t
t
H
1−
t
k
t
−
F
τ
a
µ
z
+
,t
µ
Ψ,t
a
u
t
k
t
−φ.
More generally, starting with the identity of GDP calculated in real terms:
g
t
+c
d
t
+i
d
t
+ x
d
t
= ˜y
t
(4.60)
we arrive at an equation that upon loglinearisation will explain the real income:
y
t
= g
t
+c
d
t
+i
d
t
+
1 −ω
x
f
x
t
ω
x
γ
mx,d
t
1−η
x x
+
1 −ω
x
η
x x
1−η
x x
x
t
+
F
τ
a
µ
z
+
,t
µ
Ψ,t
a
u
t
k
t
.
(4.61)
77
Part III
Results of estimation and
characteristic features of the DSGE
SOE
PL−2009
model
78
Forms of model, data, SVAR models
WORKING PAPER No. 83 79
5
Chapter 5
Forms of model, data, SVAR models
5.1 Forms of model
Upon the determination of the ﬁrst order conditions from the decisionmaking problems pre
sented earlier and their loglinearisation, we arrive at the linear structural form of the model
which may be presented in matrix form in the following manner:
t
α
0
˜ z
t+1
+α
1
˜ z
t
+α
2
˜ z
t−1
+β
0
θ
t+1
+β
1
θ
t
=0,
θ
t
= ρ θ
t−1
+
t
(5.1)
where the vector of exogenous variables (θ
t
disturbances) consists of structural disturbances θ
s
t
and observable ﬁscal disturbances θ
τ
t
and observable foreign disturbances θ
t
. The approximation
of the processes controlling observable disturbances is received by separate estimation of two
SVAR models: ﬁscal model and world’s economy model. Fragments of matrix ρ and the matrix
of observable disturbances covariance are equivalent to the respective matrices of the SVAR
model, i.e. they remain ﬁxed during (Bayesian) estimation of the whole DSGE model. In the
Appendix we shall present a list of variables of the structural form and a list of loglinearised
equations which constitute the structural form of the model.
Transition to the reduced form of the model — which eliminates the forwardlooking variables —
is made with the use of AndersonMoore algorithm. Thus, we obtain:
˜ z
t+1
=A˜ z
t
+B θ
t+1
,
θ
t+1
=ρθ
t
+ε
t+1
.
79
5
Forms of model, data, SVAR models
Forms of model, data, SVAR models
N a t i o n a l B a n k o f P o l a n d 80
5
The system may be transformed in the state space form of the model. Then the transition
equation shall be:
˜ z
t+1
θ
t+1
ξ
t+1
=
A B ρ
0 ρ
F
ξ
˜ z
t
θ
t
ξ
t
+
B
I
ε
t+1
υ
t+1
and the state space model takes the form of:
ξ
t+1
= F
ξ
ξ
t
+υ
t+1
, (υ
t+1
υ
t+1
) =Q,
Y
t
= A
x
x
t
+H
ξ
t
+u
t
, (u
t
u
t
) = R,
(5.2)
where Y
t
is a vector of observed variables and u
t
is a vector of measurement errors.
During the construction of the state space model, the state vector is supplemented with lagged
variables. Variables that are not necessary in the measurement equations may be eliminated. In
order to avoid introduction of further symbols, we shall omit the fact that the ˜ z
t
vector present in
the structural form and the ˜ z
t
vector in the general case are different. As results from the notation
of equations, on a standard basis we assume the appearance of measurement errors in the model
(state space representation). The elements of the variancecovariance matrix of measurement
errors R increase the pool of the model parameters. For simplicity, we shall assume that R is
diagonal. The appearance of measurement errors in the equation is explained by inaccuracies
in statistics, approximation errors appearing in measurement equations (loglinearisation of
nonlinear dependencies) and the possibility of errors in model speciﬁcation
1
, therefore, arrival
at (additional) statistical identiﬁcation is a side effect
2
.
Formally, in the presented version of the model there are in total 30 shocks, including 11
observable disturbances (processes estimated with SVAR models). A list of the model variables
and a list of equations of the structural form presented in the Appendix characterise a more
general version of the SOE
PL−2009
model. The version discussed further on originated as an
effect of elimination of the disturbance of national insurance contributions paid by the employer
and the assumption that there is only one interest rate risk premium for the tested economy,
common for both currency markets (
˜
φ
u
t
=
˜
φ
e
t
=
˜
φ
t
). In both cases we deal with a simpliﬁcation
of the model motivated by the problems with shocks identiﬁcation.
5.2 Observable variables, data
The data used in a DSGE model estimation and SVAR models come from the ofﬁcial publications
of the National Bank of Poland (NBP), the Central Statistical Ofﬁce (GUS), the ECB and
1
The subject of the selection of observable variables and the role of measurement errors speciﬁed in DSGE models
were analysed by GuerronQuintara (2009). According to his conclusions, consideration of measurement errors during
a DSGE model estimation allows to increase the model resistance to speciﬁcation errors and incorrect selection of
observable variables; see also Boivin and Giannoni (2005).
2
The issue of statistical identiﬁcation was discussed in more detail by e.g. Ireland (2004), AlvarezLois et al. (2005),
Canova and Sala (2005).
80
Forms of model, data, SVAR models
WORKING PAPER No. 83 81
5
OECD available till November 2009. The observable variables in the Y
t
vector in equation
(5.2) are: GDP deﬂator, investment deﬂator, consumer price index (CPI), GDP, consumption
(calculated together with changes in inventories), investment expenditures, export, import, total
employment, real wages, domestic shortterm interest rate (Wibor 3M), real USD/PLN exchange
rate, GDP deﬂator in the euro area, GDP deﬂator in the USA, shortterm interest rate for dollar
(OECD assessments of threemonth’s dollar rate), shortterm interest rate for euro (Euribor 3M),
GDP in the USA, GDP in the euro area, the nominal cross rate (dollar/euro)
3
— 19 series in total.
The whole sample for the DSGE SOE
PL−2009
model covers the period of 1996:2–2009:3. For the
estimation of the SVAR model of the world’s economy, we have used the longest available series.
The data referring to domestic national accounts stem from the system of measurement at
ﬁxed prices of the year 2000. At the initial stages of preparing the data, the consistency of
the calculations (volume and deﬂators) was veriﬁed, and then the series were subjected to
transformations that referred to the elimination of their seasonality with the X12 method
(only variables that may reﬂect quarterly seasonality), elimination of means (exchange rate),
ﬁnding the logarithm, transformation of variables into quarterly growth rates. Exceptions are
the dynamics of prices, which have an annualised form, interest rate (remaining in the natural
form) and employment. In the case of the last variable, the data regarding the level have been
ﬁltrated (with HodrickPrescott ﬁlter) and deviations from the HP trend act as an observable
variable.
Bearing in mind the solutions assumed by the authors of the SOE
Euro
model, an additional
adjustment of the export and import series was made. Whereas the model does not attempt to
explain why the share of foreign trade in the GDP changes (grows), an adjustment ensuring
constant share of import and export was made. Another arbitrary correction refers to the trend
appearing in the observable exchange rate (real exchange rate of dollar/zloty). The phenomenon
of nominal exchange rate appreciation (in consequence also the real exchange rate) is speciﬁc
for economies catchingup with the development gap and similarly as in the case of trade share
in the GDP, the SOE
PL−2009
model does not undertake to explain the process, therefore, at the
ﬁnal phase of transformation (linear) trend is eliminated from the exchange rate. Although
the practice of elimination of series of various arbitrarily deﬁned components (generally — the
ﬁltration practice) used in estimation of DSGE or (S)VAR models is widespread, it is worth
to remember that the informative content of the series and their possible interrelations are
distorted. As an example, by eliminating a trend from employment or exchange rate, we leave
the consequences of the existence of the trends in other series (e.g. prices, GDP) and such set of
data is further used for searching relations between series.
We wish to point out that foreign variables are present in the set of observable variables, i.e. the
variables fulﬁl a double function in the model: they are disturbances and observable variables.
Thus, we receive a better identiﬁcation of the other disturbances referring to the world (export
3
We present here a set of variables that was used in the estimation of the discussed version of the model. The
version is an effect of many experiments, out of which a part referred to the selection of observable variables. We have
experimented (among others) with monetary aggregates, export deﬂators, import deﬂators, consumption deﬂators
(including changes in inventories), ﬁscal series (government expenditures, revenues from VAT tax), competitive versions
of employment series and change in the real exchange rate of dollar.
81
Forms of model, data, SVAR models
N a t i o n a l B a n k o f P o l a n d 82
5
and import markups, asymmetry in the level of technology, risk premium, etc.) and, additionally,
one of the fundamental assumptions of the model of a small open economy is veriﬁed: the
common steady state (identical inﬂation, interest rate and economic growth in domestic and
world economy). Any possible departures from that principle are compensated with constants in
the block of measurement equations. Figure 5.1 illustrates the graphs of observable variables of
the model after all of the transformations.
Figure 5.1. Data
1998 2000 2002 2004 2006 2008
0
5
10
15
20
25
GDP deflator
1998 2000 2002 2004 2006 2008
4
2
0
2
4
6
GDP
1998 2000 2002 2004 2006 2008
4
2
0
2
4
Consumption
1998 2000 2002 2004 2006 2008
10
5
0
5
10
15
Investment
1998 2000 2002 2004 2006 2008
30
20
10
0
10
20
Exports
1998 2000 2002 2004 2006 2008
15
10
5
0
5
10
15
Imports
1998 2000 2002 2004 2006 2008
20
15
10
5
0
5
10
Real exchange rate
1998 2000 2002 2004 2006 2008
5
10
15
20
25
Interest rate
1998 2000 2002 2004 2006 2008
2
1
0
1
2
Employment
1998 2000 2002 2004 2006 2008
2
1
0
1
2
3
Real wages
1998 2000 2002 2004 2006 2008
40
20
0
20
40
60
Investment deflator
1998 2000 2002 2004 2006 2008
0
5
10
15
20
CPI
1998 2000 2002 2004 2006 2008
2
1
0
1
Euro zone GDP
1998 2000 2002 2004 2006 2008
1
0
1
2
3
Euro zone inflation
1998 2000 2002 2004 2006 2008
1
2
3
4
5
Euro interest rate
1998 2000 2002 2004 2006 2008
1
0.5
0
0.5
1
1.5
US GDP
1998 2000 2002 2004 2006 2008
1
2
3
4
US inflation
1998 2000 2002 2004 2006 2008
1
2
3
4
5
6
USD interest rate
1998 2000 2002 2004 2006 2008
5
0
5
10
Change of dollar/euro exchange rate
5.3 SVAR models
SVAR models are a separate component of the model. The purpose of ﬁscal SVAR model and
world’s economy SVAR model is to receive a description of a stochastic process approximating
the disturbances represented by the variables of the SVAR models.
5.3.1 Fiscal SVAR
Fiscal SVAR, formally represented by equation (4.40), consists of four variables: effective
personal income tax rate (calculated as the revenues of state budget from personal income tax by
the estimates of households income, increased for the health insurance contribution), effective
corporate income tax rate (the revenues of state budget from corporate income tax by the GDP
decreased for households income), consumption tax (state budget revenues from intermediate
taxes referred to individual consumption) and government expenditures (public consumption).
The above described raw time series are subject to transformations. Tax rates are seasonally
82
Forms of model, data, SVAR models
WORKING PAPER No. 83 83
5
adjusted with X12 method and then converted into percentage deviations from the mean in the
sample (according to the logic of loglinearised variables of the model). A logarithm was found
for the series of government expenditures at ﬁxed prices and then the series was seasonally
adjusted with X12 method and converted into percentage deviations from the trend determined
with the use of HodrickPrescott ﬁlter. Such constructed SVAR model has been estimated on a
sample 1996:1–2009:3.
5.3.2 World’s economy SVAR
The task of the VAR model is to determine the dynamics of foreign variables and of dollar/euro
nominal exchange rate. Foreign variables of the model are exogenous and consist of two areas:
the euro area and the dollar area. The dollar area is approximated by the USA economy. Variables
of the euro area shall be identiﬁed with superscript e, while the variables of the dollar area
with subscript u. For modelling each of the areas we have used three variables: real product ( y
e
t
and y
u
t
), inﬂation (π
e
t
and π
u
t
) and nominal interest rate (R
e
t
and R
u
t
). Beside of the variables
characteristic of each of the areas, the VAR model includes the dynamics of the nominal exchange
rate of USD/EUR (∆x
t
). The model consists, thus, of three variables for the euro area and three
variables for the dollar area, as well as the cross rate USD/EUR. Vector autoregression system in
reduced form is given by:
y
t
= A
1
y
t−1
+A
2
y
t−2
+ d +ωt + e
t
where: y
t
= ( y
e
t
, π
e
t
, R
e
t
, y
u
t
, π
u
t
, R
u
t
, ∆x
t
).
Matrices A
1
and A
2
are autoregressive matrices, vector d consists of constants and vector ω
consists of the trend parameters, while the vector of error terms e
t
depends on the vector of
structural shocks through matrix B, i.e. e
t
= B u
t
. The lag order of the system is 2, the model
includes a deterministic trend and a constant, i.e. the reduced form covers 144 parameters.
The sample based on which the model parameters have been estimated covers 166 observations
for each of the variables, starting from the ﬁrst quarter of 1980. The data for the dollar area
come from the OECD and Eurostat databases. The data for the euro area starting from the ﬁrst
quarter 1995 come from the OECD and Eurostat databases, while the data from before 1995
have been reconstructed based on the database of a Area Wide Model (AWM) of the European
Central Bank. A series of the nominal USD/EUR exchange rate comes from the Eurostat database.
The variables have been seasonally adjusted with the ARIMA X12 method. Inﬂation and the
nominal exchange rate are included in the model as quarterly logarithmic growth rates of the
series of GDP deﬂators and nominal exchange rate of USD/EUR. Output is included in the model
as a cyclic component of a quarterly HP ﬁlter applied to the logarithm of an output expressed at
market prices deﬂated with the GDP deﬂator. Interest rate is expressed in percentage points at
quarterly values.
For the purpose of estimation of the structural form of the VAR model within each of the
areas, Cholesky type structuralisation has been applied in the following sequence: production,
inﬂation and interest rate. For the propagation of structural disturbances between the euro area
and the dollar area one quarter delay has been assumed. An exchange rate shock enters the
83
Forms of model, data, SVAR models
N a t i o n a l B a n k o f P o l a n d 84
5
structuralisation orthogonally to the other shocks. The parameters of the reduced form, i.e. the
elements of matrices A
1
and A
2
, as well as the elements of vectors d and ω, have been estimated
on quarterly data with estimated general least squares method (EGLS). On the parameters of
the reduced form dynamic restrictions have been imposed (the matrix structure, the imposed
restrictions and the results of the estimation have been presented in the Annexe). Matrix B
making the forecasts errors e
t
dependent on structural shocks u
t
has been estimated with the
maximum likelihood method.
Figure 5.2 presents the effect of positive structural shock to the interest rate (R
e
for the euro
area and R
u
for the dollar area) of the value of 1 p.p. on the product ( y
e
, y
u
), inﬂation (π
e
and
π
u
) and the interest rate (R
e
and R
u
). After monetary policy contraction in the euro area, output
declines by about 1%, while the maximum drop occurs 9 quarters after the shock and the effect
becomes statistically insigniﬁcant after about 3 years. Inﬂation drops by about 0.25%, while
the effect is signiﬁcant between the 5th and 15th quarter and the maximum effect materializes
in the 12th quarter. Prices, thus, display larger inertia than output. An effect of contractionary
monetary policy of the value of 1 percentage point in the dollar area is similar in terms of quality,
however, the response of inﬂation is much weaker — drop by about 0.05%, insigniﬁcant along
the whole length of the impulse response function. Output responds signiﬁcantly between the
2nd and 5th quarter and drops by about 0.65% in the 6th quarter. Impulse responses seem to be
sensible.
Figure 5.2. Contractionary monetary policy shock in a SVAR model
84
Assessment of parameters — calibration, optimisation, steady state
WORKING PAPER No. 83 85
6
Chapter 6
Assessment of parameters —
calibration, optimisation, steady
state
The procedure of estimation of large DSGE models, motivated with Bayesian ideas
1
, consists
of two basic stages. At the ﬁrst stage we determine with numerical optimisation methods the
approximation of modal value of posterior distribution, i.e. point estimates of the parameters. At
the second stage we estimate the shape of parameters distributions in the neighbourhood of
the posterior mode, by applying MCMC techniques (e.g. Metropolis algorithm versions). Thus
we generate a sample from model parameters posterior distribution
2
. Practice has shown that
already at the stage of optimisation, problems with parameters identiﬁcation appear, while
convergence of numerical procedures is achieved only if prior distributions shall explicitly
determine at least a subset of parameters. In such a situation most of the authors of the DSGE
models decide to divide parameters into two groups: calibrated and estimated parameters.
The estimation procedure (e.g. with Bayesian techniques) is then applied only to some of
the parameters. In principle, one would have to be absolutely certain about the values of the
calibrated parameters and only those parameters to which our knowledge (which we represent
by prior distribution) is uncertain shall be (Bayesian) estimated. Unfortunately, in our case (or
probably not only in our case, see Adolfson et al. (2007a, page 12)), this was not the motivation
for applying calibration to at least some of the parameters — we calibrated parameters which
cannot be estimated because of technical constraints.
1
We refer here to practice, the formulated generalisation refers to the users of the DYNARE package (academic
communities), YADA package (ECB) and Riksbank. Generally, DSGE models do not have to be estimated with Bayesian
techniques and the manner of the Bayesian techniques application may be different than we suggest.
2
We omit here (and further on in the presentation) the fact that some of the parameters of the model are estimated
separately with the use of SVAR models.
85
6
Assessment of parameters – calibration, optimisation, steady state
Assessment of parameters — calibration, optimisation, steady state
N a t i o n a l B a n k o f P o l a n d 86
6
6.1 Calibration of parameters
In the SOE
PL−2009
model, which as for the standards of estimated DSGE models may be deemed
to be large, about 8090 parameters from a set of over 150 parameters are calibrated. The
calibration refers, among others, to 19 variances of the measurement errors of observable data,
27 parameters directly or indirectly characterising the steady state, 22 parameters characterising
the stochastic disturbance processes, 8 parameters centring the variables included in the SVAR
model, as well as 12 other parameters which from economic point of view may be treated as
important. Above enumeration omits zero covariance of measurement errors and a collection
of parameters related to the stochastic processes of observable disturbances (SVAR), yet, it
includes the values of parameters that are subject to structural changes (11) and those that are
determined by the speciﬁcation (economic content) of the model. The percentage of calibrated
parameters is relatively higher, in reference to the estimated DSGE models developed in other
countries, but our model is larger and data sample remains short and nonhomogenous. Thus
we face greater challenges than the authors of models in e.g. the euro area countries or the USA.
Due to the diversity of the parameters excluded from the estimation, no uniform procedure
has been applied to determine their values. In case of characteristics of steady state variables
that have their equivalents among observable data, the average values of the sample were the
main indicator. In a number of cases, the ﬁrst approximations were the values assumed by other
authors of DSGE models, and in particular the assumptions made for the models of SOE
Euro
,
MEDEA (Burriel et al., 2009), NAWM and RAMSES (Adolfson et al., 2009), however, we always
made attempts to verify such assumptions by experimenting with competitive values, i.e. we
tested the consequences of changes in the values of such parameters, making a sort of sensitivity
analysis, such as e.g. the authors of NAWM (Christoffel et al., 2007a, pages 44–45). At the initial
stages of the works over the model, we used Laplace approximation of marginal likelihood as
the criterion of assessment of the model, later we additionally applied the measures of accuracy
of ex post forecasts for the last 45 years of the sample, all the time controlling the dynamic
features of the model (impulse response functions).
The experiments proved that restrictions imposed on a DSGE model by virtue of calibration of
some of the parameters were very strong but in many cases failed confrontation with empirical
material. Seemingly small adjustments of calibrated parameters signiﬁcantly inﬂuenced the
accuracy of forecasts, yet, they are not recognisable for other criteria (e.g. marginal likelihood,
impulse response functions). Small departure from the “typical” calibrated values of parameters
enabled us to improve the quality of the SOE
PL−2009
model, or at least its forecasting features,
without a major breach of its economic content. Despite the efforts, we are aware that there
is large spare capacity in that area and potential consequences of assuming wrong values
are serious. Speciﬁcation of the most important calibrated parameters of the model has been
provided in Table 6.1. The speciﬁcation omits the parameters obtained from estimation of SVAR
models.
86
Assessment of parameters — calibration, optimisation, steady state
WORKING PAPER No. 83 87
6
Table 6.1. Values of the most important calibrated parameters in the SOE
PL−2009
model
Parameter Value Comments Source
0.250 based on literature, veriﬁed by experiment
δ 0.025 based on literature and sample means of the great ratios
β 0.9995 based on literature, veriﬁed by experiment
A
L
7.500 based on literature
A
q
0.135 based on data
σ
L
1.000 based on literature
σ
q
10.620 based on literature
ν
f k
0.400 based on literature, veriﬁed by experiment
ν
f w
0.600 based on literature, veriﬁed by experiment
µ 1.0165 steady state determined by experiment
g
r
0.190 steady state mean from the sample of share of consumption in GDP
τ
k
0.0125 steady state empirically veriﬁed expert assessments
τ
w
0.180 steady state average national insurance contributions
τ
s
0.180 steady state average national insurance contributions
τ
c
0.1772 steady state mean from the sample of effective VAT rate
τ
p
0.0125 steady state mean relation of revenues from CIT to GDP
τ
y
0.1301 steady state mean from the sample of effective PIT rate
ω
c
0.450 determined by experiment
ω
i
0.650 determined by experiment
ω
x
0.330 determined by experiment
ω
mcu
0.460 determined by experiment
ω
miu
0.370 determined by experiment
ω
mxu
0.333 determined by experiment
η
f u
4.500 determined by experiment
η
f e
3.200 determined by experiment
u
y
e, 0.6810 determined by experiment
u
a
e
m
0.0095 determined by experiment
b 0.6401 determined by experiment
˜
S
6.9682 determined by experiment
σ
a
1.1276 determined by experiment
λ
x
1.0000 conclusion from model assumptions
λ
mx
1.0000 conclusion from model assumptions
ρ
ε
0.700 determined by experiment
ρ
λ
d
0.100 based on literature, veriﬁed by experiment
ρ
λ
mx
0.500 based on literature
ρ
ε,R
0.000 based on literature
ρ
¯ π
c 0.900 based on literature, veriﬁed by experiment
ρ
τ
k
0.900 determined by experiment
ρ
τ
s
0.900 determined by experiment
ρ
π
oil 0.900 determined by experiment
cR
d
1.3500 observable variable adjustment determined by experiment
R
kraj
0.0400 variance of measurement error expert assessments
R
ot oczenie
0.0100 variance of measurement error expert assessments
ρ
λ
w
,1
0.6774 change of regime determined by experiment
ρ˜ Φ,1
0.6431 change of regime determined by experiment
σ
λ
w
,1
0.7959 change of regime determined by experiment
σ˜ Φ,1
1.7921 change of regime determined by experiment
σ
ε
R,1
0.1727 change of regime determined by experiment
σ
¯ π
c
,1
0.3482 change of regime determined by experiment
λ
w,1
1.1366 change of regime determined by experiment
ρ
R,1
0.7359 change of regime determined by experiment
r
π,1
1.4497 change of regime determined by experiment
r
x,1
0.0100 change of regime determined by experiment
r
y,1
0.1783 change of regime determined by experiment
Source: Prepared by the authors.
6.2 Prior distributions and results of estimation
As it has been mentioned before, the DSGE model estimation is a twostage process. At the ﬁrst
stage we search for the values of parameters that maximise the value of the function, which
is given by the sum of the logarithms of the likelihood function and prior distribution. These
parameters represent the mode of posterior distribution and the stage is called optimisation.
At the second stage an attempt is made to generate a sample from (posterior) distribution of
87
Assessment of parameters — calibration, optimisation, steady state
N a t i o n a l B a n k o f P o l a n d 88
6
parameters by using e.g. an iterative algorithm in which the starting point is the distribution
modal value, and consecutive values depend on the Hessian obtained during optimisation.
Generation of such sample allows — among others — to better understand the scale of un
certainty related to the estimated parameters. That piece of information may be used to asses
forecasts uncertainty. The formal aspects of Bayesian estimation, MCMC techniques and speciﬁcs
of Bayesian estimation of DSGE models have been shortly outlined in the ﬁrst part of the paper.
Other recommended sources include i.e.: Osiewalski (2001), Koop (2003), Geweke (1999),
Geweke (2005), Hastings (1970), Chib and Greenberg (1995), Brooks (1998), Brooks and
Gelman (1996) and Schorfheide (2000), An and Schorfheide (2005), Canova (2007), Warne
(2009), FernándezVillaverde (2009), providing the details of the use of Bayesian techniques in
DSGE models estimation.
Table 6.2. Results of estimation — characteristics of prior and posterior distributions
Para Prior distribution Optimisation results Sample form posterior distribution
meters Type CVal Std/DF Mode InvH Mean Mode Median 5% perc. 95% perc.
Price rigidities
ξ
w
Beta 0.600 0.100 0.558 0.079 0.558 0.547 0.557 0.427 0.690
ξ
d
Beta 0.600 0.100 0.790 0.026 0.794 0.790 0.795 0.749 0.837
ξ
mc
Beta 0.600 0.100 0.677 0.057 0.675 0.672 0.675 0.573 0.774
ξ
mi
Beta 0.600 0.100 0.687 0.062 0.683 0.690 0.685 0.576 0.783
ξ
mx
Beta 0.600 0.100 0.554 0.081 0.552 0.562 0.553 0.416 0.687
ξ
x
Beta 0.600 0.100 0.527 0.060 0.530 0.530 0.530 0.431 0.631
ξ
e
Beta 0.600 0.100 0.698 0.034 0.701 0.704 0.704 0.641 0.755
κ
w
Beta 0.500 0.140 0.417 0.124 0.423 0.385 0.419 0.231 0.628
κ
d
Beta 0.500 0.140 0.164 0.062 0.171 0.146 0.164 0.079 0.286
κ
mc
Beta 0.500 0.140 0.441 0.144 0.446 0.402 0.440 0.228 0.682
κ
mi
Beta 0.500 0.140 0.291 0.113 0.310 0.278 0.297 0.138 0.529
κ
mx
Beta 0.500 0.140 0.350 0.123 0.357 0.333 0.347 0.172 0.577
κ
x
Beta 0.500 0.140 0.202 0.081 0.215 0.175 0.204 0.095 0.375
Foreign exchange rates and energy prices effects
ν
τ
TNor 0.125 0.075 0.150 0.068 0.125 0.122 0.121 0.021 0.240
˜
φ
u
a
InvG 0.125 2 0.382 0.206 0.610 0.295 0.429 0.184 1.483
˜
φ
e
a
InvG 0.125 2 0.190 0.063 0.196 0.164 0.186 0.109 0.318
˜
φ
u
s
Beta 0.250 0.150 0.217 0.114 0.208 0.133 0.193 0.051 0.421
˜
φ
e
s
Beta 0.250 0.150 0.196 0.082 0.182 0.202 0.180 0.053 0.316
Interest rate rule
ρ
R,2
Beta 0.800 0.085 0.820 0.035 0.825 0.829 0.827 0.768 0.877
r
π,2
TNor 1.700 0.150 1.800 0.124 1.798 1.782 1.795 1.596 2.010
r
∆π
TNor 0.300 0.065 0.237 0.031 0.234 0.231 0.234 0.184 0.286
r
x,2
Norm 0.000 0.065 0.022 0.017 0.020 0.015 0.020 0.048 0.005
r
y,2
Norm 0.125 0.065 0.128 0.057 0.130 0.120 0.128 0.039 0.225
r
∆y
TNor 0.075 0.065 0.128 0.033 0.128 0.124 0.127 0.072 0.187
Markets organisation, technological progress
λ
w
2
TNor 1.100 0.075 1.150 0.060 1.142 1.121 1.137 1.055 1.241
λ
d
TNor 1.200 0.075 1.213 0.063 1.204 1.209 1.203 1.098 1.311
λ
mc
TNor 1.200 0.075 1.299 0.052 1.290 1.286 1.290 1.199 1.383
λ
mi
TNor 1.200 0.075 1.160 0.071 1.146 1.119 1.139 1.033 1.283
η
c
InvG 5.000 2 3.740 0.454 3.861 3.723 3.792 3.146 4.813
η
i
InvG 5.000 2 3.964 0.607 4.454 3.777 4.097 3.271 5.880
η
x x
InvG 5.000 2 4.544 0.861 4.800 4.362 4.622 3.540 6.637
µ
z
TNor 1.008 0.002 1.008 0.001 1.008 1.008 1.008 1.007 1.009
µ
Ψ
TNor 1.008 0.002 1.007 0.001 1.007 1.007 1.007 1.005 1.009
Characteristics of disturbances
ρ
Υ
Beta 0.700 0.100 0.773 0.052 0.774 0.785 0.777 0.683 0.855
ρ
˜ z
Beta 0.900 0.100 0.957 0.044 0.953 0.999 0.971 0.848 0.999
ρ
µ
z
Beta 0.600 0.100 0.592 0.097 0.593 0.607 0.596 0.430 0.748
ρ
µ
Ψ
Beta 0.600 0.100 0.586 0.099 0.583 0.573 0.585 0.412 0.741
ρ
ζ
c Beta 0.750 0.100 0.778 0.053 0.778 0.774 0.782 0.683 0.862
ρ
ζ
h Beta 0.750 0.100 0.625 0.102 0.612 0.629 0.614 0.436 0.782
ρ
λ
mc Beta 0.750 0.100 0.685 0.101 0.678 0.698 0.685 0.498 0.833
Continued on next page
88
Assessment of parameters — calibration, optimisation, steady state
WORKING PAPER No. 83 89
6
Para Prior distribution Optimisation results Sample form posterior distribution
meters Type CVal Std/DF Mode InvH Mean Mode Median 5% perc. 95% perc.
ρ
λ
mi Beta 0.700 0.100 0.544 0.114 0.557 0.555 0.558 0.370 0.739
ρ
λ
x Beta 0.500 0.100 0.633 0.092 0.617 0.631 0.623 0.453 0.761
ρ
λ
w
,2
Beta 0.666 0.100 0.584 0.102 0.561 0.565 0.563 0.396 0.714
ρ˜
φ
e
,2
Beta 0.666 0.100 0.571 0.081 0.582 0.602 0.583 0.449 0.711
σ
ε
InvG 0.750 2 2.043 0.382 2.107 1.971 2.060 1.524 2.843
σ
Υ
InvG 0.750 2 0.663 0.095 0.675 0.639 0.664 0.532 0.856
σ
˜ z
InvG 0.750 2 0.359 0.055 0.372 0.361 0.365 0.285 0.483
σ
µ
z
InvG 0.500 2 0.277 0.047 0.287 0.264 0.282 0.214 0.379
σ
µ
Ψ
InvG 0.500 2 0.425 0.121 0.489 0.396 0.445 0.289 0.794
σ
ζ
c InvG 0.500 2 0.687 0.079 0.707 0.697 0.700 0.582 0.855
σ
ζ
h InvG 0.500 2 0.295 0.058 0.306 0.289 0.298 0.220 0.418
σ
λ
d InvG 0.750 2 0.844 0.102 0.869 0.846 0.860 0.707 1.062
σ
λ
mc InvG 1.000 2 0.773 0.205 0.856 0.757 0.810 0.537 1.322
σ
λ
mi InvG 1.000 2 1.484 0.261 1.547 1.417 1.513 1.104 2.101
σ
λ
mx InvG 1.000 2 5.961 1.531 6.304 5.425 6.055 3.954 9.553
σ
λ
x InvG 1.000 2 2.546 0.436 2.682 2.423 2.610 2.025 3.570
σ
λ
w
,2
InvG 0.750 2 0.380 0.064 0.392 0.384 0.385 0.296 0.515
σ
e
˜
φ,2
InvG 1.000 2 2.129 0.504 2.180 2.088 2.115 1.443 3.139
σ
τ
k InvG 0.010 2 0.012 0.006 0.021 0.010 0.015 0.007 0.051
σ
τ
s InvG 0.025 2 0.031 0.015 0.053 0.025 0.037 0.018 0.134
σ
ν
oil InvG 2.000 2 2.688 1.274 2.994 2.220 2.791 1.440 5.208
σ
R,2
InvG 0.250 2 0.195 0.031 0.205 0.192 0.202 0.155 0.263
σ
¯ π
c
,2
InvG 0.250 2 0.466 0.205 0.467 0.371 0.442 0.216 0.802
Source: Prepared by the authors.
Note: Nrm — normal distribution, InvG — inverse Gamma distribution, TNor — truncated normal distribution; for
the inverse Gamma distribution the modal value and the number of the degrees of freedom have been speciﬁed. HInv
— approximation of standard deviation from the assessment of parameter based on Hessian inverse. CVal — central
value.
Calibration of parameters does not solve all the problems which appear during the optimisation,
e.g. selection of an effective algorithm for optimisation
3
. We also continue to stick to conditional
estimation — optimisation takes place in the presence of the expert knowledge introduced
formally with prior distribution. The type of prior distribution assumed for particular parameters
results from knowledge about the allowed interval of values of the parameters, therefore, on
a standard basis there is assumed Beta distribution for e.g. Calvo probabilities or correlation
coefﬁcient (AR processes of the shocks), an inverse Gamma distribution for variances of distur
bance innovation, normal distribution for parameters that may assume any values, etc. Variances
(degrees of freedom) of prior distributions have been deﬁned such as to prefer information
included in the sample (unless we had a stronger belief regarding the value of the parameters
and the sample proved to be sufﬁciently informative)
4
. The central values of the distributions,
however, proved to be debatable. As in the case of calibrated parameters, extensive literature
describes the experience of the authors of Bayesian estimated DSGE models
5
, of which we have,
of course, availed while verifying our own analyses and beliefs. Also in this case it appears that
relatively small changes in the central values of distributions, completely undistinguishable for an
3
Often applied by researchers, implemented in Dynare and YADA packages, csminwel algorithm by Ch. Sims (provided
by the author on the Internet site in the form of Matlab scripts, also available in the form of C++ function) proved to be
ineffective in our case. After several experiments it appeared that the unconditional optimisation algorithm implemented
in the optimisation toolbox of Matlab deals much more effectively with the extreme’s determination, however, at the
cost of considerable longer time of computation.
4
It is worth to remember that reaching for Bayesian techniques is not always the consequence of the Bayesian point
of view or Bayesian beliefs of the authors. Sometimes, as in this case, the Bayesian procedures are applied instead of the
classical procedures, due to the fact that the available sample does not include a sufﬁcient amount of information to
carry out the traditional estimation (ﬂat likelihood function). Instead of additionally extending the set of calibrated
parameters, we avail of techniques that slightly reduce the arbitrariness of the study. Therefore, pragmatic considerations
are decisive.
5
Grabek et al. (2007, pages 69–70) present several examples drawn from the DSGE models with structure similar to
SOE
PL−2009
. In the speciﬁcation provided there, prior distributions of models built before 2006 are available.
89
Assessment of parameters — calibration, optimisation, steady state
N a t i o n a l B a n k o f P o l a n d 90
6
expert, are important — they impact the characteristics of the model and its forecasting features.
The ﬁnal result of our research — a complete set of assumptions regarding prior distributions,
as well as assessments of parameters and approximations of standard errors arrived at during
optimisation have been provided in Table 6.2.
Table 6.2 provides also selected characteristics of the distribution received as a result of the
application of Metropolis algorithm (in the random walk version), which generates a sample of
distribution convergent toward posterior distribution. During the second stage of the estimation
procedure, Markov chain of the length of 750,000 elements has been generated, while the data
presented in the Table have been taken from a sample of 500,000 elements remaining upon the
rejection of the ﬁrst 250,000 elements. Already a superﬁcial analysis of the values of parameters
for the modal value received from optimisation and determined from the Markov chain reﬂects
a considerable divergence between the corresponding values, suggesting lack of convergence.
More thorough convergence analyses (logarithms of posterior probability for the elements of
the chain, Geweke tests for the particular parameters, test statistics based on cumulated sums,
etc.) provided additional doubts as to whether the convergence is actually present. Due to
limited resources at our disposal we could not further generate the chain
6
. Therefore, during
further works we used only the point estimates of parameters originating from the ﬁrst stage of
estimation — optimisation. Generally, a signiﬁcant issue of the uncertainty of parameters at the
current stage of the works has not been completed — we shall come back to the issue after the
existing technical limitations has been reduced.
6.3 Assessment of parameters — conclusions
6.3.1 Steady state
When all the disturbances (shocks) described in the model expire, the economy reaches a
steady state. Then, all the nominal categories grow at an identical rate determined by the
calibrated parameter µ. The annualised value of the parameter implies the growth rate at the
level of about 6.6%. The division of the value dynamics into the dynamics of volume and prices
is identical for all the growing categories, except investments. The division depends on the
estimated parameters characterising the growth rate of technical progress µ
z
and µ
Ψ
, as well
as the calibrated share of capital in a product ( — parameter of CobbDouglas production
6
As an example, the authors of the SOE
Euro
model generated a chain of the length of 700,000800,000 elements
using 500,000 end elements for inference see (see Adolfson et al., 2005b, page 28). On the other hand, S. SchmittGrohé
and M. Uribe used the end 4,000,000 elements from a chain of 10,000,000 elements for a signiﬁcantly smaller DSGE
model (see SchmittGrohé and Uribe, 2008, page 16). The characteristics of posterior distribution of the NAWM model
(ECB) have been determined based on a chain of the length of 500,000 elements (with ﬁrst 50,000 elements rejected),
however, convergence was tested with the use of four parallel chains of the length of 500,000 elements each (see
Christoffel et al., 2007a, page 42). In the DSGE model built for the Central Bank of Hungary, the Metropolis procedure
was used for generation of two Markov chains of the lengths of 300,000 elements, and for reasoning a sample of 500,000
elements was used, originating from the merger of the last 250,000 elements of each of the chains (see Jakab and Világi,
2008, page 19). Reviewing the descriptions of other Bayesian estimated models, it is hard to resist the feeling that some
of the authors of the DSGE models treat convergence in a sort of ritual manner — this is a qualitative difference with
regard to the conduct of researchers specialising in the Bayesian techniques.
90
Assessment of parameters — calibration, optimisation, steady state
WORKING PAPER No. 83 91
6
function) and amounts to about 2.5% of inﬂation and about 4.1% of volume dynamics
7
. The
annual interest rate of about 7.1% corresponds to those values.
In all the DSGE models we are aware of, the steady state level of interest rate has very high
values. Formally, this is so because its level derives from the value dynamics (µ), discount and
the rate of tax on capital in steady state. Within that logic, whilst expecting volume growth
dynamics at the level of 34% and inﬂation at about 2%, it is hard to arrive at interest rate
lower than 56%, even if the discount implies the minimal weight applied to the future and
the tax on capital is quite arbitrarily assumed at low values (or even zero, see Adolfson et al.
(2009, page 29)). There is only one interest rate present in the model, no difference exists
for the rate of the central bank, deposits, loans, short or longterm rate
8
, while observable
equivalents are threemonth rates in the interbank market (Wibor, Euribor). In order to at least
slightly compensate for this inconsistency, in the block of measurement equations of the model
adjusting constants have been introduced. It has been determined by experiment (taking into
account the quality of forecasts and marginal likelihood) that adjustment compliant with data
for the domestic interest rate amounts to 1.35%, 3.4% for the euro area and 2.5% for the USA.
Therefore, by and large, interest rates in Poland shall be by about 2 p.p. higher than the rates in
the euro area in steady state. Corresponding adjustments have also been introduced to the GDP
growth rate and inﬂation in the euro area and in the USA
9
.
Also the share of trade in GDP may raise doubts. In steady state net exports amount to zero
and the share of import in GDP has the level of about 36%, which is below the values that
we have observed (we expected the value of about 40%). In reference to the previous version
of the SOE
PL
model this is, however, a considerable progress (we earlier had less than 29%)
achieved thanks to the explicit consideration of import intensity of exports. Nevertheless, the
model continues to omit import intensity of government consumption, so there are reasons for
underestimated share of trade in GDP.
The values of the most important variables and proportions in the steady state are presented
in Table 6.3. In the Appendix we present an example forecast with a very long horizon, which
(among others) reﬂects the rate of convergence toward the steady state.
Table 6.3. Values of selected variables and the great ratios in steady state
Parameter Value Comments
Annual growth rate of technological progress 0.041181 annualised µ
z
+
Annual growth rate of total investments 0.050825
Annual inﬂation rate 0.025424 annualised π
Annual interest rate 0.070696 R
Share of consumption in GDP 0.626330
Share of investments in GDP 0.183670
Share of government consumption in GDP 0.190000 g
r
Share of import (export) in GDP 0.359007
Source: Prepared by the authors.
7
The carried out experiments, e.g. with higher value dynamics µ, brought about higher inﬂation (instead of growth),
generating at the same time larger ex post forecast errors. Also the received approximation of the marginal likelihood
did not provide arguments for such solution.
8
Based on the structure of the model, all of the transactions have the horizon of one quarter.
9
For inﬂation this was 0.25 p.p. in the euro area and 0.25 in the USA; for the annualised GDP dynamics it was 2.00
p.p. for the euro area and about 1.332 p.p. for the USA.
91
Assessment of parameters — calibration, optimisation, steady state
N a t i o n a l B a n k o f P o l a n d 92
6
6.3.2 Structural changes
The SOE
PL−2009
model allows for structural changes regarding monetary policy, exchange rate
policy and labour market characteristics. We have assumed that due to ofﬁcial introduction of
direct inﬂation targeting in Poland in 1999, interest rate rule parameter values may have changed.
This concerns also characteristics of inﬂation target disturbance. Already before introduction of
targeting, exchange rate regime had gradually been shifting towards free ﬂoat, which — in our
opinion — affected the exchange rate risk premium shock (characteristics of a stochastic process).
Additionally, we assume that introduction of a new pension system and health insurance system
in 1999, with stricter control of the ﬂow of contributions and pension entitlements, affected the
level of wage markup in steady state and the characteristics of the stochastic process governing
this markup. In both cases, we assumed for simplicity the simultaneous occurrence of such regime
changes as from the second quarter 1999. The experiments showed that more reﬁned deﬁning
of moments of change, better reﬂecting the historical events, practically does not affect the
marginal likelihood and the forecasting features of the model, however, leads to a considerable
elongation of the time of computation. Therefore, we have assumed the simpliﬁcation.
Searching for further structural changes, the year of Poland’s accession to the European Union
was also tested, as the possible moment of further structural changes (covering i.e. the labour
market, budget expenditures, effects related to exchange rate). The results of experiments have
not conﬁrmed such assumptions
10
. As a result there are two regimes present in the model,
while change appears at the beginning of the sample, which has an advantage of decreasing
the heterogeneity of the sample, enabling at the same time slightly better identiﬁcation of
parameters in the second regime. A more radical solution — estimation starting from the year
2000 — seemed to us to be too costly. A drawback of the assumed solution is, however, a dozen
or so additional parameters to be estimated or calibrated. Finally, we decided to calibrate the
parameters for the ﬁrst regime, based on the results of experiments made earlier. In the above
presented tables, the changing parameters have been identiﬁed with subscripts, respectively 1
and 2 for the given regimes. In Table 6.4 we present a speciﬁcation of parameters whose values
change in effect of structural changes.
From a formal point of view, the method of representation in the model of the aforesaid structural
changes gives the changes deterministic nature, however, unforeseeable by the agents. This
simpliﬁcation results from the applied technique
11
. We think that the simpliﬁcation may be,
however, rationalised by reference also to subjectmatter arguments. The introduction of the
social security system reform (affecting the labour market) and changes in the strategy of
monetary policy was announced earlier and in the world of rational agents everybody was aware
of that. The information had both a general and qualitative nature. The actual consequences
of such activities were not known even to the authors of the reforms and in that sense the
10
The scope of the experiments made by us was, obviously, limited and several hypotheses remain open. As an
example, the authors of the Hungarian DSGE model, when estimating their model for two regimes of monetary policy,
also arrived at large differences of Calvo probabilities and indexation parameters (see Jakab and Világi, 2008, page 21).
Yet, we have not veriﬁed that issue.
11
Other authors model structural changes with the use of stochastic switching Markov chain, see e.g. (Justiniano
and Primiceri, 2006; Lou et al., 2007; Sims and Zha, 2005). It is also proposed to estimate two models in separate
subsamples (Smets and Wouters, 2007).
92
Assessment of parameters — calibration, optimisation, steady state
WORKING PAPER No. 83 93
6
resultant of all of the changes was unforeseeable, particularly, while speaking about the values
of deep parameters. However, the phenomenon of the growth of general uncertainty of the agent
actions in face of the expected complex structural changes has been omitted. The effect may
be compensated by deﬁning a set of anticipated disturbances — yet, we have left that issue for
separate study.
Table 6.4. Parameters changing values as a result of structural changes
Parameter Regime 1 Regime 2
(calibrated value) (optimised value)
Wage markup
λ
w
1.1366 1.1504
ρ
λ
w
0.6774 0.5839
σ
λ
w
0.7959 0.3802
Risk premium
ρ˜ Φ
0.6431 0.5712
σ˜ Φ
1.7921 2.1293
Interest rate rule and inﬂation target
ρ
R
0.7359 0.8204
r
π
1.4497 1.7995
r
x
0.0100 0.0219
r
y
0.1783 0.1284
σ
ε
R
0.1727 0.1949
σ
¯ π
c 0.3482 0.4658
Source: Prepared by the authors.
From the data provided in the Table 6.4 it appears that the changes in monetary policy discussed
here have been interpreted by agents as the increase in interest rate sensitivity to inﬂation
ﬂuctuations (also in reference to other determinants of the interest rate deﬁned in the rule).
With increase in the variance of interest rate shock, also the persistence of the interest rate
increased. On the other hand, after the year 1999 the inﬂation target (innovations) has also
featured larger variance. At the same time the variance of exchange risk premium innovation
increased with simultaneous decrease in the disturbance’s persistence. More transparent seem
to be the changes in the labour market, where tightening of the insurance system meant the
increase in the level of markup with decrease in the innovation variance and persistence of
the disturbance. The effects of such changes for the impact of the disturbances on observable
variables are illustrated by variances decompositions and reaction functions presented further
herein.
6.3.3 Nominal rigidities
A distinguishing feature of a DSGE model based on the paradigm of a new Keynesian school is the
presence of nominal rigidities — delays in the adjustment of prices. While modelling the rigidities
of prices (wages) in the SOE
PL−2009
model, we have applied the Calvo model supplemented with
the mechanism of dynamic indexation, discussed earlier in more detail (see e.g. equation (4.17)).
In Table 6.5 there are speciﬁed the assessments of parameters characterising Calvo rigidities.
For comparison, there have been provided assessments of analogical parameters (modal value)
received from the earlier version of the SOE
PL
model (see Grabek et al., 2007), the RAMSES
model (case with UIP condition analogical as in the SOE
PL−2009
, (see Adolfson et al., 2007a,
page 25), SOE
Euro
model of the euro area (case with variable utilisation (see Adolfson et al.,
93
Assessment of parameters — calibration, optimisation, steady state
N a t i o n a l B a n k o f P o l a n d 94
6
2005b, page 58), and NAWM ECB (see Christoffel et al., 2007a, page 82), as well as the values
characterising rigidities for the Hungarian economy (see Jakab and Világi, 2008, page 23) for
the case of carrying out the strategy of direct inﬂation target. The cited values comply with the
modal value of posterior distribution or median (RAMSES).
From the Table 6.5 it results, generally, that the SOE
PL−2009
model describes economy with a
more ﬂexible wages and prices than the models of the euro area, Sweden or Hungary — at
least prices are less susceptible to marginal costs’ changes
12
. The difference is particularly visible
in the labour market, however, more persistent indexation mechanism lowers the effects of
shorter delay in reoptimisation of wages. Interesting are also changes in the assessments of
parameters in the subsequent versions of the SOE
PL
, model: the version estimated in 2006
reﬂected relatively stronger adaptative indexation of the prices of domestically manufactured
products; now the κ
d
parameter characterising the effect of inertia is minimal, so with slightly
longer periods between the subsequent reoptimisations indexation takes place mainly based on
the expected (instead of former) prices. Obviously, the change results also from a change in the
speciﬁcation of the model, so we have no grounds to claim that the prices setting mechanism
has changed e.g. after the accession to the European Union.
Table 6.5. Rigidities of prices and wages in the chosen DSGE models
Parameter SOE
PL−2009
SOE
PL−2006
RAMSES SOE
Euro
NAWM (EBC) MNB model
Calvo probability
ξ
w
0.558 0.586 0.752 0.716 0.765 0.711
ξ
d
0.794 0.680 0.838 0.895 0.920 
ξ
mc
0.675 0.585 0.901 0.523  
ξ
mi
0.683 0.617 0.944 0.743  
ξ
mx
0.552     
ξ
x
0.530 0.600 0.883 0.630 0.770 0.827
Dynamic indexation parameter
κ
w
0.417 0.350 0.313 0.453 0.635
κ
d
0.164 0.434  0.173 0.417 
κ
mc
0.441 0.439  0.128  
κ
mi
0.291 0.440  0.192  
κ
mx
0.350     
κ
x
0.202 0.508  0.148 0.489
Source: Prepared by the authors; detailed information in the text of the paper.
12
Christoffel et al. (2007a, page 43) mention that Calvo probability ξ and the average time between price re
optimisation
1
1−ξ
implied by the value does not necessarily inform about the scale of nominal rigidities because in the
Phillips curve built on the basis of the Calvo model, real and nominal rigidities may not be identiﬁed (discriminated)
separately.
94
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 95
7
Chapter 7
Variance decompositions, impulse
response functions and estimation
of disturbances
7.1 Variance decompositions
In the following Tables (7.1–7.4) we have presented variances decompositions (formally —
forecast errors), separately for regime I covering the period up to the ﬁrst quarter of 1999
(Tables 7.1–7.2) and regime II characterising the relationships observed in the other part of the
sample (Tables 7.3–7.2). As we have mentioned before, all of the calculations were made for
parameter values corresponding to the modal value of posterior distribution determined at the
stage of optimisation (we omit the uncertainty of parameters). The horizon has been limited
to shortterm and mediumterm period (maximum 20 quarters), which from the point of view
of the current applications of the model seems to be sufﬁcient. For longer horizons, the results
depend on model speciﬁcation. Real growing variables are determined with nonstationary
disturbances, while the rest of them — with disturbance of the largest persistence. In the earlier
versions of the SOE
PL
model, as well as in the SOE
Euro
model, this was the disturbance of the
inﬂation target (the calibrated autocorrelation coefﬁcient amounted then to 0.975–0.985). In
the current version, the largest persistence is reﬂected by foreign variables which constitute the
SVAR
1
. Certainly, this does not seriously affect the steady state or short and midterm model
1
The aforementioned experiments (sensitivity analyses) showed that autocorrelation of inﬂation target disturbance at
the level of 0.975–0.985 deteriorates the quality of forecasts, therefore, a slightly lower value of the parameter (0.900)
has been assumed. Higher persistence has been achieved by estimation of the characteristics of AR(1) process for an
asymmetric shock and by constructing relations between world’s economy SVAR model and the basic part of the model
(real dollar/euro rate). It is the real dollar/euro rate that currently determines the joint persistence of the observable
external disturbances.
95
7
Variance decompositions, impulse response functions and
estimation of disturbances
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 96
7
dynamics we are the most interested in.
The data gathered in the Tables present the role of 17 single structural disturbances and two
blocks of shocks. The block of external/foreign shocks covers all of the (observable) disturbances
characterising the euro area and the USA, present in the SVAR model. In the block of ﬁscal
disturbances all of the disturbances modelled with the use of SVAR model and additionally
structural disturbances of national insurance contributions and capital tax have been jointly
reﬂected.
The ﬁrst general observation refers to the scale of impact (share) of ﬁscal variables and foreign
variables on the observable variables. Comparing variances decompositions received in the
previous versions of the SOE
PL
model (see Grabek et al., 2007, pages 100–101) with the current
version of the model, the growth of importance of the aforesaid shocks is clearly visible. The
marginal role of impulses originating from the world’s economy was one of the most important
deﬁciencies of previous versions — the effects observed earlier (at the level of 1–3%) were
inconsistent with the concept of a small open economy. Therefore, the currently observed results
are quite interesting, although we think that this time the effects have been overestimated.
The growth of importance of ﬁscal disturbances is also visible, however, it still seems to be
underestimated. Yet, this is — at least partially — caused by the very structure of the DSGE
model (a Ricardian household, no resource effects related to public debt, etc.).
7.1.1 Regime I
In the short run (1–4 quarters), the main inﬂation determinant before 1999:2 was the situation
in the labour market (wage markups), markups in the market of domestic products (GDP deﬂator,
CPI) and markups on the imported investment goods (investment deﬂator). Additionally, the role
of ﬁscal shocks with regard to CPI is visible. Along with the extension of the observation horizon
(8–20 quarters), the impact of disturbances originating from the world’s economy, the inﬂation
target disturbances and technological disturbances (stationary in investments) is growing. In
the same regime, GDP depends (on shortterm basis) on external factors (which quickly abate)
and on domestic shocks related to foreign trade (exchange rate risk premium, export markups,
markups on the import of export components). Further, ﬁscal factors, markups on domestic
intermediate products and wage markups manifest themselves. Along with the longer horizon
of analysis, the impact of technological disturbances (stationary and nonstationary) has been
growing, with the maintained role of shocks originating in the labour market.
According to the data provided in Table 7.1, shortterm consumption is determined by distur
bances in consumption preferences, the role of which abates in the subsequent quarters, as well
as by disturbances in the wage markup (here we observe an inverse trend). Also a moderate
inﬂuence of ﬁscal disturbances may be perceived in midterm horizon (8–12 quarters). Along
with the extension of lag, the impact of stationary technological disturbance and nonstationary
disturbances in investments has been growing. In the case of investments, a dominating role is
played by stationary technological disturbances in investments (nonstationary disturbance in a
longer horizon). Additionally, with a lag of 4–12 quarters, the impact of shocks on wage markup
and consumption preferences may be observed.
96
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 97
7
The disturbance of markups in the labour market — in regime I — is the most important determi
nant of employment and real wages, particularly in shorter horizons. Employment ﬂuctuations
are also triggered by the shocks of consumption preferences and labour supply (2–12 quarters),
ﬁscal shocks (8–20 quarters) and technological shocks (stationary in investments, in the long
term horizon). On shortterm basis (1 quarter) important are also: stationary disturbance in
technology, export markups, exchange rate risk premium and internal disturbances. Real wages
ﬂuctuations result (beside of wage markup shocks) in a shortterm period from ﬂuctuations of
labour supply preferences and domestic product markups. In a longer period, technology is a
dominating factor.
The real exchange rate of US dollar and the interest rate seem to be dominated by external
disturbances along the whole horizon. Interest rate reacts also to ﬂuctuations in labour market
markups and inﬂation target, and in the short period the interest rate shock, exchange rate risk
premium and export markups are important as well. The real exchange rate in the short run (1–4
quarters) reacts to exchange rate risk premium, export markups and markups in consumption
import. “Fundamental” domestic factors affecting the real exchange rate of the dollar on a
midterm basis are clearly missing.
7.1.2 Regime II
In regime II, in the short run inﬂation depends on the inﬂation target, markup on products
manufactured domestically (markup of imported investment goods in the case of investment
deﬂator), ﬁscal disturbances (in the case of CPI) and foreign variables (whose impact grows in
time). In the horizon of 1–4 quarters, also wage markup is important, however, generally the
role of these disturbances (compared to regime I) clearly decreased, particularly in a midterm
horizon. Drop of importance of wage markup is also observable for GDP. Similarly as in regime
I, in a short period of time GDP variance results from disturbances in the markup of products
manufactured domestically and disturbances related to foreign trade (external disturbances,
export markup, risk premium). Impulses coming from the labour market are slightly more visible
with a lag of 412 quarters. Yet, this time technological disturbances (stationary in investments,
nonstationary) and disturbance in labour supply preferences are more important.
The lowering importance of wage markup disturbance is also conﬁrmed in the case of con
sumption and investments. In regime II, in the shortterm horizon (1 quarter), consumption is
clearly impacted only by the shock of consumption preferences and domestic products markups.
Roughly 8–9 shocks (e.g. external shocks, stationary shocks in technology, ﬁscal shocks, inﬂation
target shocks) have the remaining 50% share in the variance (of forecasts errors), while none of
them brings about a stronger impact. In a longer horizon — as in regime I — consumption clearly
depends on ﬁscal and technological shocks. The role of wage markup disturbance has been
taken over, to some extent, by the disturbance in labour supply preferences. For investments,
stationary investment disturbance remains the main determinant on short and midterm basis.
The importance of nonstationary shocks is growing along the horizon.
Conclusions formulated when discussing factors affecting employment and real wages in regime
I remain valid for regime II. We only observe — as in the previous cases — a lower role of wage
97
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 98
7
markup disturbance, with only a partial takeover of the impact by the disturbance in labour
supply preferences.
Although one of the phenomena motivating separate treatment of the two regimes were changes
in the currency market, it is hard to perceive larger changes in the group of the most important
shocks affecting the real exchange rate of the dollar. As before, external disturbances are
dominating. In a very short horizon (1–4 quarters), we may talk about the role of markups in
export, consumption goods import and import of export components as well as risk premium.
The inﬂuence of interest rate disturbance, even in the shortterm horizon, is small (3.6% for
the ﬁrst quarters and dropping) — it is smaller than the inﬂuence of ﬁscal disturbances (about
5.9–5.0% for 1–12 quarters). On the other hand, the consequences of the adoption of the
strategy of direct inﬂation target are bit clearer. The importance of inﬂation target disturbance
grew and the impact of monetary policy shock is stronger (but only in the ﬁrst quarter, in further
quarters a drop may be observed). Clearly, the impact of wage markup disturbance is lower,
while higher is the share of external disturbances.
Thus, investigating the structure of the impact of disturbances in both regimes in a general
way, we see that from the point of view of variance decompositions the essence of changes that
occurred in 1999 was a sort of breakdown in the institutional structure responsible for setting
the price of labour. This phenomenon seems to be the most important, despite hardly spectacular
manifestation (see the respective values of parameters in Table 6.4). In some cases the function
of wage markup disturbance was taken over by the disturbance in the labour supply preferences,
which may suggest that institutional phenomena have been supported by phenomena related to
preferences. Nevertheless, wage markup disturbance is nonidentiﬁable from the point of view
of speciﬁcation of the structural form of wage equation — discrimination requires relatively
extensive expert knowledge (introduced by the speciﬁcation of prior distribution at the stage of
estimation), so it is hard to talk about tangible evidence.
98
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 99
7
T
a
b
l
e
7
.
1
.
V
a
r
i
a
n
c
e
d
e
c
o
m
p
o
s
i
t
i
o
n
s
.
R
e
g
i
m
e
1
,
p
a
r
t
1
D
i
s
t
u
r
b
a
n
c
e
G
D
P
d
e
ﬂ
a
t
o
r
G
D
P
C
o
n
s
u
m
p
t
i
o
n
I
n
v
e
s
t
m
e
n
t
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
S
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
9
.
6
5
.
0
0
.
8
0
.
5
1
.
2
0
.
7
6
.
5
3
.
7
1
.
1
0
.
6
3
.
5
5
.
0
5
.
0
3
.
7
1
.
9
1
.
1
3
.
2
3
.
5
3
.
3
1
.
8
I
n
v
e
s
t
m
e
n
t
s
p
e
c
i
ﬁ
c
t
e
c
h
n
o
l
o
g
y
2
.
2
4
.
8
8
.
8
1
2
.
2
1
3
.
3
2
.
7
8
.
0
1
2
.
9
1
6
.
9
1
2
.
2
5
.
8
5
.
1
1
.
3
7
.
4
2
0
.
7
5
1
.
3
3
9
.
8
2
7
.
5
1
7
.
3
5
.
7
A
s
y
m
m
e
t
r
i
c
t
e
c
h
n
o
l
o
g
y
0
.
0
0
.
2
0
.
3
0
.
5
0
.
7
0
.
6
0
.
0
0
.
0
0
.
1
0
.
3
0
.
1
0
.
1
0
.
2
0
.
3
0
.
4
0
.
3
0
.
3
0
.
5
0
.
6
1
.
0
N
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
8
1
.
3
1
.
6
1
.
8
1
.
7
2
.
8
6
.
9
1
1
.
3
1
9
.
9
3
9
.
1
5
.
6
7
.
2
1
1
.
3
1
7
.
1
2
3
.
7
3
.
0
3
.
4
4
.
4
6
.
4
1
5
.
5
I
n
v
e
s
t
m
e
n
t
n
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
1
.
4
2
.
3
3
.
2
3
.
8
3
.
6
1
.
1
2
.
2
4
.
0
8
.
5
1
9
.
3
1
.
5
1
.
4
1
.
7
2
.
6
5
.
9
0
.
3
2
.
4
5
.
2
8
.
6
2
1
.
6
C
o
n
s
u
m
p
t
i
o
n
p
r
e
f
e
r
e
n
c
e
1
.
1
1
.
8
2
.
9
3
.
9
4
.
4
6
.
5
8
.
4
5
.
8
3
.
4
1
.
3
3
8
.
9
2
9
.
3
1
6
.
7
5
.
8
4
.
1
7
.
8
8
.
9
9
.
6
1
0
.
0
7
.
5
L
a
b
o
u
r
s
u
p
p
l
y
8
.
3
1
0
.
1
5
.
9
2
.
9
2
.
1
1
.
4
7
.
7
7
.
9
4
.
9
1
.
5
2
.
8
5
.
8
7
.
9
7
.
3
3
.
6
1
.
8
3
.
7
4
.
8
5
.
4
4
.
2
D
o
m
e
s
t
i
c
g
o
o
d
s
m
a
r
k
u
p
2
7
.
4
4
.
5
1
.
7
0
.
4
0
.
9
1
2
.
2
5
.
3
1
.
2
1
.
4
0
.
4
7
.
9
4
.
8
2
.
5
1
.
3
1
.
1
6
.
5
3
.
7
2
.
6
1
.
9
0
.
3
I
m
p
o
r
t
e
d
c
o
n
s
u
m
p
t
i
o
n
g
o
o
d
s
m
a
r
k
u
p
1
.
3
2
.
4
1
.
4
0
.
1
1
.
3
0
.
7
1
.
2
1
.
7
2
.
5
1
.
4
0
.
6
0
.
9
0
.
4
2
.
0
2
.
1
0
.
4
2
.
2
3
.
7
4
.
1
2
.
6
I
m
p
o
r
t
e
d
i
n
v
e
s
t
m
e
n
t
g
o
o
d
s
m
a
r
k
u
p
0
.
3
0
.
1
0
.
1
0
.
8
1
.
4
1
.
9
0
.
5
3
.
6
4
.
1
2
.
0
1
.
7
2
.
2
2
.
5
1
.
1
2
.
2
5
.
0
7
.
4
5
.
7
3
.
0
1
.
5
I
m
p
o
r
t
e
d
e
x
p
o
r
t
c
o
m
p
o
n
e
n
t
m
a
r
k
u
p
0
.
3
2
.
4
3
.
1
2
.
6
2
.
1
1
0
.
7
3
.
5
4
.
0
0
.
7
1
.
4
2
.
5
4
.
5
5
.
3
4
.
3
3
.
0
4
.
7
5
.
3
5
.
5
4
.
9
2
.
0
E
x
p
o
r
t
m
a
r
k
u
p
0
.
8
0
.
8
2
.
5
1
.
9
0
.
7
1
4
.
9
6
.
0
2
.
0
1
.
4
0
.
5
4
.
5
3
.
1
0
.
9
0
.
0
0
.
6
0
.
1
0
.
6
1
.
3
1
.
3
0
.
5
W
a
g
e
m
a
r
k
u
p
2
7
.
4
3
5
.
4
2
3
.
7
1
2
.
7
7
.
8
4
.
7
2
5
.
9
2
8
.
5
1
9
.
3
6
.
2
9
.
1
1
9
.
5
2
7
.
8
2
6
.
4
1
3
.
2
5
.
8
1
2
.
2
1
6
.
7
1
9
.
1
1
6
.
1
R
i
s
k
p
r
e
m
i
u
m
1
.
8
1
.
2
1
.
8
2
.
2
1
.
4
1
0
.
2
5
.
6
3
.
2
2
.
8
0
.
2
1
.
7
0
.
0
3
.
1
3
.
9
1
.
8
1
.
2
0
.
4
1
.
8
3
.
3
3
.
7
I
n
t
e
r
e
s
t
r
a
t
e
0
.
7
0
.
9
0
.
3
0
.
2
0
.
3
2
.
4
2
.
2
0
.
1
0
.
1
0
.
2
2
.
5
1
.
5
0
.
8
0
.
5
0
.
4
2
.
1
1
.
3
0
.
9
0
.
6
0
.
1
I
n
ﬂ
a
t
i
o
n
t
a
r
g
e
t
7
.
2
1
0
.
6
1
3
.
2
1
3
.
5
9
.
4
0
.
8
1
.
2
3
.
0
3
.
0
2
.
2
2
.
7
0
.
9
1
.
2
2
.
5
2
.
2
2
.
0
0
.
7
0
.
6
1
.
7
4
.
1
E
n
e
r
g
y
p
r
i
c
e
s
1
.
5
0
.
1
0
.
1
0
.
2
0
.
2
2
.
9
0
.
0
0
.
2
0
.
1
0
.
1
0
.
6
0
.
8
0
.
6
0
.
4
0
.
3
2
.
3
0
.
7
0
.
6
0
.
5
0
.
2
F
i
s
c
a
l
2
.
7
5
.
1
7
.
9
9
.
1
7
.
0
6
.
8
4
.
2
5
.
3
7
.
1
7
.
6
4
.
0
6
.
0
8
.
2
8
.
9
8
.
9
1
.
0
0
.
8
1
.
0
1
.
6
3
.
0
W
o
r
l
d
e
c
o
n
o
m
y
(
S
V
A
R
)
5
.
0
1
0
.
9
2
0
.
6
3
0
.
7
4
0
.
4
1
5
.
9
4
.
6
1
.
6
2
.
8
3
.
4
4
.
0
1
.
7
2
.
5
4
.
5
4
.
0
3
.
3
3
.
1
4
.
1
6
.
2
8
.
6
D
i
s
t
u
r
b
a
n
c
e
E
x
p
o
r
t
s
I
m
p
o
r
t
s
R
e
a
l
e
x
c
h
a
n
g
e
r
a
t
e
I
n
t
e
r
e
s
t
r
a
t
e
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
S
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
3
1
.
3
0
.
2
2
.
1
0
.
2
1
.
9
1
.
8
2
.
7
4
.
1
1
.
7
0
.
0
1
.
0
0
.
2
0
.
4
0
.
0
4
.
2
5
.
7
4
.
4
2
.
8
1
.
2
I
n
v
e
s
t
m
e
n
t
s
p
e
c
i
ﬁ
c
t
e
c
h
n
o
l
o
g
y
0
.
3
2
.
5
1
0
.
6
1
7
.
6
1
2
.
6
3
.
0
6
.
2
1
0
.
5
2
.
6
3
.
3
0
.
1
1
.
6
2
.
8
3
.
0
1
.
5
1
.
2
1
.
2
1
.
5
0
.
7
1
1
.
3
A
s
y
m
m
e
t
r
i
c
t
e
c
h
n
o
l
o
g
y
1
.
2
0
.
9
2
.
0
3
.
2
4
.
2
0
.
8
1
.
3
3
.
6
3
.
3
4
.
0
0
.
5
0
.
4
0
.
4
0
.
4
0
.
3
0
.
4
0
.
3
0
.
3
0
.
4
0
.
5
N
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
1
.
2
2
.
6
8
.
2
1
6
.
0
2
9
.
9
1
.
3
3
.
5
1
2
.
4
1
3
.
2
2
3
.
3
0
.
1
0
.
3
0
.
2
0
.
1
0
.
0
0
.
2
0
.
7
1
.
2
1
.
8
2
.
6
I
n
v
e
s
t
m
e
n
t
n
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
5
0
.
7
2
.
7
6
.
7
1
5
.
2
0
.
3
1
.
4
6
.
3
6
.
9
1
1
.
6
0
.
0
0
.
5
0
.
5
0
.
3
0
.
0
0
.
3
1
.
2
2
.
2
3
.
5
5
.
5
C
o
n
s
u
m
p
t
i
o
n
p
r
e
f
e
r
e
n
c
e
1
.
5
1
.
8
0
.
9
3
.
5
0
.
6
3
.
2
4
.
0
1
.
7
6
.
8
5
.
6
2
.
3
1
.
2
0
.
4
0
.
7
0
.
0
6
.
0
7
.
5
9
.
3
8
.
8
5
.
8
L
a
b
o
u
r
s
u
p
p
l
y
0
.
1
1
.
9
2
.
1
1
.
2
0
.
8
1
.
1
1
.
9
1
.
0
4
.
8
3
.
8
0
.
7
1
.
5
0
.
4
0
.
3
0
.
1
3
.
1
7
.
6
7
.
9
6
.
3
2
.
9
D
o
m
e
s
t
i
c
g
o
o
d
s
m
a
r
k
u
p
0
.
8
0
.
1
3
.
2
2
.
5
0
.
5
8
.
5
0
.
7
6
.
4
3
.
6
0
.
3
1
.
3
0
.
5
0
.
9
0
.
3
0
.
1
2
.
2
3
.
9
1
.
8
0
.
7
0
.
5
I
m
p
o
r
t
e
d
c
o
n
s
u
m
p
t
i
o
n
g
o
o
d
s
m
a
r
k
u
p
5
.
5
8
.
7
5
.
5
1
.
9
0
.
2
6
.
0
8
.
8
5
.
5
1
.
1
1
.
7
1
1
.
1
1
2
.
1
3
.
5
0
.
9
0
.
0
5
.
6
8
.
0
0
.
6
1
.
7
1
.
5
I
m
p
o
r
t
e
d
i
n
v
e
s
t
m
e
n
t
g
o
o
d
s
m
a
r
k
u
p
2
.
1
3
.
8
6
.
0
6
.
0
2
.
7
4
.
9
5
.
7
2
.
8
1
.
0
1
.
7
2
.
9
2
.
5
1
.
4
0
.
9
0
.
3
1
.
3
0
.
7
2
.
1
1
.
7
0
.
8
I
m
p
o
r
t
e
d
e
x
p
o
r
t
c
o
m
p
o
n
e
n
t
m
a
r
k
u
p
1
9
.
4
3
0
.
3
1
8
.
0
5
.
7
0
.
5
3
4
.
0
2
7
.
9
0
.
9
6
.
0
1
.
8
8
.
8
4
.
4
1
.
3
0
.
6
0
.
1
7
.
9
4
.
3
5
.
7
3
.
4
1
.
9
E
x
p
o
r
t
m
a
r
k
u
p
3
0
.
4
2
5
.
8
1
2
.
0
2
.
6
0
.
6
1
7
.
2
2
3
.
0
2
1
.
3
5
.
5
0
.
4
1
1
.
2
6
.
9
1
.
8
0
.
5
0
.
0
1
1
.
8
3
.
3
0
.
3
0
.
3
0
.
4
W
a
g
e
m
a
r
k
u
p
0
.
2
6
.
6
8
.
5
2
.
6
3
.
0
3
.
6
6
.
2
2
.
4
1
5
.
7
1
4
.
1
2
.
6
5
.
1
1
.
6
0
.
9
0
.
3
1
0
.
3
2
5
.
8
2
8
.
3
2
3
.
6
1
1
.
2
R
i
s
k
p
r
e
m
i
u
m
9
.
0
3
.
1
6
.
2
6
.
9
1
.
5
2
.
0
2
.
5
7
.
1
7
.
1
3
.
4
1
8
.
6
0
.
0
1
.
9
1
.
1
0
.
1
1
1
.
4
7
.
4
2
.
8
4
.
1
1
.
7
I
n
t
e
r
e
s
t
r
a
t
e
1
.
4
0
.
7
0
.
5
0
.
5
0
.
1
0
.
1
0
.
1
1
.
5
0
.
9
0
.
1
2
.
8
0
.
2
0
.
2
0
.
1
0
.
0
1
0
.
7
1
.
2
0
.
5
0
.
3
0
.
2
I
n
ﬂ
a
t
i
o
n
t
a
r
g
e
t
0
.
3
1
.
3
3
.
3
2
.
3
0
.
7
0
.
7
0
.
6
0
.
2
2
.
2
4
.
0
1
.
8
0
.
7
0
.
5
0
.
1
0
.
1
1
.
6
6
.
4
8
.
5
9
.
5
8
.
5
E
n
e
r
g
y
p
r
i
c
e
s
0
.
5
0
.
5
0
.
6
0
.
5
0
.
1
0
.
2
0
.
4
1
.
4
0
.
8
0
.
2
0
.
9
0
.
3
0
.
2
0
.
1
0
.
0
1
.
7
0
.
9
0
.
6
0
.
3
0
.
2
F
i
s
c
a
l
0
.
3
0
.
6
1
.
7
1
.
3
0
.
5
0
.
6
0
.
9
0
.
7
1
.
5
3
.
4
5
.
9
5
.
3
5
.
0
3
.
9
3
.
2
5
.
9
3
.
0
5
.
6
7
.
2
6
.
7
W
o
r
l
d
e
c
o
n
o
m
y
(
S
V
A
R
)
2
4
.
9
6
.
7
7
.
7
1
7
.
1
2
6
.
2
1
0
.
6
3
.
1
1
1
.
7
1
3
.
0
1
5
.
4
2
8
.
2
5
5
.
6
7
6
.
7
8
5
.
6
9
3
.
7
1
4
.
3
1
0
.
8
1
6
.
3
2
3
.
0
3
6
.
6
99
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 100
7
T
a
b
l
e
7
.
2
.
V
a
r
i
a
n
c
e
d
e
c
o
m
p
o
s
i
t
i
o
n
s
.
R
e
g
i
m
e
1
,
p
a
r
t
2
D
i
s
t
u
r
b
a
n
c
e
E
m
p
l
o
y
m
e
n
t
R
e
a
l
w
a
g
e
s
I
n
v
e
s
t
m
e
n
t
d
e
ﬂ
a
t
o
r
C
P
I
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
S
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
1
2
.
6
5
.
9
2
.
3
2
.
1
3
.
8
3
.
2
4
.
5
6
.
2
5
.
3
1
.
9
4
.
3
5
.
7
2
.
8
1
.
1
0
.
7
6
.
6
5
.
7
1
.
8
0
.
6
1
.
0
I
n
v
e
s
t
m
e
n
t
s
p
e
c
i
ﬁ
c
t
e
c
h
n
o
l
o
g
y
4
.
1
6
.
3
8
.
1
6
.
1
1
8
.
6
0
.
4
2
.
0
8
.
7
1
8
.
8
1
9
.
0
1
.
1
2
.
7
5
.
2
1
1
.
1
1
6
.
9
1
.
6
4
.
1
7
.
0
1
1
.
6
1
5
.
1
A
s
y
m
m
e
t
r
i
c
t
e
c
h
n
o
l
o
g
y
0
.
3
0
.
1
0
.
0
0
.
0
0
.
1
0
.
1
0
.
1
0
.
1
0
.
3
0
.
3
0
.
3
0
.
3
0
.
2
0
.
2
0
.
4
0
.
2
0
.
3
0
.
2
0
.
4
0
.
6
N
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
1
0
.
2
0
.
7
1
.
7
6
.
2
5
.
7
9
.
4
1
8
.
3
2
9
.
9
3
1
.
4
0
.
4
1
.
0
1
.
3
1
.
8
1
.
9
0
.
6
1
.
2
1
.
4
1
.
8
1
.
8
I
n
v
e
s
t
m
e
n
t
n
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
5
0
.
7
1
.
6
3
.
6
1
2
.
9
2
.
2
3
.
1
5
.
3
9
.
0
1
1
.
5
4
.
1
0
.
3
2
.
1
3
.
7
4
.
1
1
.
0
2
.
2
2
.
8
3
.
7
3
.
9
C
o
n
s
u
m
p
t
i
o
n
p
r
e
f
e
r
e
n
c
e
8
.
0
1
0
.
0
1
0
.
8
1
0
.
5
1
3
.
0
1
.
4
2
.
9
5
.
5
7
.
6
5
.
6
0
.
2
2
.
1
3
.
3
3
.
5
3
.
5
0
.
4
2
.
1
3
.
2
3
.
6
4
.
1
L
a
b
o
u
r
s
u
p
p
l
y
5
.
7
9
.
6
1
1
.
4
8
.
4
0
.
3
1
5
.
0
1
3
.
9
6
.
9
0
.
2
2
.
6
4
.
5
8
.
6
6
.
5
4
.
7
1
.
7
6
.
1
1
0
.
0
6
.
3
3
.
6
1
.
9
D
o
m
e
s
t
i
c
g
o
o
d
s
m
a
r
k
u
p
7
.
9
5
.
0
0
.
0
2
.
3
2
.
8
1
9
.
7
8
.
5
5
.
3
2
.
9
1
.
1
9
.
4
2
.
5
0
.
3
1
.
0
0
.
6
1
7
.
7
1
.
4
1
.
2
0
.
4
0
.
8
I
m
p
o
r
t
e
d
c
o
n
s
u
m
p
t
i
o
n
g
o
o
d
s
m
a
r
k
u
p
0
.
9
0
.
9
0
.
9
1
.
8
0
.
8
1
.
3
1
.
4
1
.
8
2
.
7
2
.
1
2
.
8
2
.
0
5
.
2
2
.
7
0
.
8
8
.
6
0
.
2
4
.
5
2
.
7
1
.
5
I
m
p
o
r
t
e
d
i
n
v
e
s
t
m
e
n
t
g
o
o
d
s
m
a
r
k
u
p
0
.
8
0
.
6
3
.
0
3
.
5
0
.
3
0
.
5
0
.
2
0
.
7
2
.
0
2
.
3
3
0
.
5
1
6
.
9
1
4
.
3
3
.
8
2
.
2
0
.
5
0
.
2
0
.
8
1
.
4
1
.
9
I
m
p
o
r
t
e
d
e
x
p
o
r
t
c
o
m
p
o
n
e
n
t
m
a
r
k
u
p
4
.
5
1
.
8
5
.
4
4
.
4
4
.
5
1
.
6
0
.
9
1
.
5
2
.
7
2
.
6
4
.
0
1
.
5
0
.
7
0
.
2
1
.
6
2
.
0
2
.
3
0
.
9
1
.
3
2
.
0
E
x
p
o
r
t
m
a
r
k
u
p
1
1
.
3
7
.
1
0
.
6
1
.
5
1
.
2
0
.
9
0
.
2
0
.
9
0
.
6
0
.
5
4
.
6
0
.
5
3
.
2
2
.
1
0
.
3
1
.
7
0
.
8
0
.
8
0
.
0
0
.
6
W
a
g
e
m
a
r
k
u
p
1
8
.
8
3
2
.
6
4
0
.
6
3
2
.
2
2
.
0
4
2
.
4
4
4
.
1
2
6
.
1
2
.
2
9
.
0
1
5
.
1
2
9
.
6
2
4
.
0
1
8
.
5
6
.
7
2
0
.
0
3
4
.
8
2
4
.
1
1
4
.
9
7
.
3
R
i
s
k
p
r
e
m
i
u
m
8
.
6
5
.
5
1
.
3
3
.
7
3
.
8
0
.
5
1
.
8
1
.
2
1
.
3
1
.
2
5
.
6
3
.
5
2
.
6
0
.
4
0
.
2
3
.
5
1
.
0
2
.
4
1
.
2
0
.
9
I
n
t
e
r
e
s
t
r
a
t
e
2
.
3
2
.
0
0
.
7
0
.
1
0
.
7
0
.
8
1
.
3
1
.
1
0
.
8
0
.
4
1
.
1
0
.
1
0
.
0
0
.
1
0
.
2
0
.
8
0
.
6
0
.
1
0
.
0
0
.
3
I
n
ﬂ
a
t
i
o
n
t
a
r
g
e
t
0
.
1
1
.
5
3
.
5
4
.
2
4
.
4
0
.
8
1
.
0
3
.
1
4
.
8
2
.
9
5
.
3
9
.
4
9
.
4
1
1
.
7
9
.
2
5
.
8
1
0
.
7
1
1
.
3
1
2
.
4
9
.
5
E
n
e
r
g
y
p
r
i
c
e
s
0
.
5
0
.
4
0
.
4
0
.
4
0
.
5
0
.
9
0
.
6
0
.
5
0
.
4
0
.
3
0
.
9
0
.
4
0
.
0
0
.
1
0
.
2
1
.
2
0
.
2
0
.
0
0
.
1
0
.
2
F
i
s
c
a
l
3
.
7
4
.
2
6
.
8
1
0
.
3
2
0
.
5
1
.
2
2
.
5
3
.
7
3
.
9
1
.
7
1
.
9
4
.
4
5
.
6
8
.
1
7
.
3
1
6
.
7
9
.
7
1
3
.
4
1
2
.
6
6
.
2
W
o
r
l
d
e
c
o
n
o
m
y
(
S
V
A
R
)
9
.
3
5
.
5
2
.
0
3
.
2
3
.
5
1
.
4
1
.
6
2
.
9
4
.
5
3
.
4
3
.
9
8
.
4
1
3
.
4
2
5
.
4
4
1
.
5
5
.
2
1
2
.
6
1
7
.
7
2
7
.
7
4
0
.
5
100
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 101
7
T
a
b
l
e
7
.
3
.
V
a
r
i
a
n
c
e
d
e
c
o
m
p
o
s
i
t
i
o
n
s
.
R
e
g
i
m
e
2
,
p
a
r
t
1
D
i
s
t
u
r
b
a
n
c
e
G
D
P
d
e
ﬂ
a
t
o
r
G
D
P
C
o
n
s
u
m
p
t
i
o
n
I
n
v
e
s
t
m
e
n
t
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
S
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
1
0
.
9
4
.
6
2
.
2
1
.
7
0
.
7
0
.
8
8
.
0
6
.
1
2
.
6
0
.
9
4
.
2
6
.
1
7
.
0
5
.
2
2
.
4
1
.
8
4
.
1
4
.
6
4
.
4
2
.
2
I
n
v
e
s
t
m
e
n
t
s
p
e
c
i
ﬁ
c
t
e
c
h
n
o
l
o
g
y
0
.
7
2
.
0
3
.
9
5
.
4
7
.
2
2
.
8
9
.
1
1
5
.
6
1
9
.
1
1
3
.
7
4
.
7
4
.
6
0
.
9
9
.
3
2
2
.
7
5
0
.
2
4
1
.
1
3
0
.
2
1
9
.
8
6
.
0
A
s
y
m
m
e
t
r
i
c
t
e
c
h
n
o
l
o
g
y
0
.
0
0
.
1
0
.
3
0
.
4
0
.
5
0
.
6
0
.
0
0
.
0
0
.
1
0
.
3
0
.
1
0
.
2
0
.
3
0
.
3
0
.
4
0
.
3
0
.
4
0
.
5
0
.
7
1
.
1
N
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
8
1
.
2
1
.
1
1
.
0
1
.
1
2
.
7
7
.
8
1
3
.
4
2
1
.
1
3
8
.
0
5
.
4
7
.
5
1
3
.
4
2
0
.
2
2
5
.
4
2
.
7
3
.
3
4
.
6
6
.
9
1
7
.
4
I
n
v
e
s
t
m
e
n
t
n
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
1
.
2
2
.
1
2
.
2
2
.
2
2
.
3
1
.
0
2
.
5
4
.
7
8
.
7
1
8
.
4
1
.
3
1
.
3
1
.
8
3
.
0
6
.
2
0
.
6
2
.
2
5
.
3
9
.
3
2
4
.
3
C
o
n
s
u
m
p
t
i
o
n
p
r
e
f
e
r
e
n
c
e
1
.
7
3
.
0
3
.
3
2
.
9
2
.
5
6
.
7
1
0
.
6
7
.
9
4
.
1
1
.
4
3
9
.
1
3
1
.
6
2
0
.
7
7
.
3
4
.
2
6
.
9
8
.
4
9
.
8
1
0
.
7
8
.
5
L
a
b
o
u
r
s
u
p
p
l
y
8
.
4
9
.
7
1
.
7
1
.
4
0
.
6
1
.
5
8
.
8
1
0
.
8
7
.
2
2
.
3
3
.
6
6
.
8
1
0
.
2
9
.
2
4
.
1
2
.
6
4
.
6
6
.
1
6
.
7
4
.
9
D
o
m
e
s
t
i
c
g
o
o
d
s
m
a
r
k
u
p
3
3
.
8
8
.
5
3
.
7
0
.
0
0
.
9
1
2
.
2
7
.
7
0
.
0
1
.
2
0
.
5
8
.
7
6
.
2
4
.
0
2
.
1
1
.
5
7
.
3
4
.
7
3
.
7
2
.
8
0
.
3
I
m
p
o
r
t
e
d
c
o
n
s
u
m
p
t
i
o
n
g
o
o
d
s
m
a
r
k
u
p
1
.
1
2
.
4
1
.
6
1
.
0
0
.
3
0
.
8
2
.
2
0
.
4
2
.
2
1
.
6
1
.
1
1
.
6
0
.
3
1
.
9
2
.
1
0
.
1
1
.
5
3
.
3
4
.
1
3
.
0
I
m
p
o
r
t
e
d
i
n
v
e
s
t
m
e
n
t
g
o
o
d
s
m
a
r
k
u
p
0
.
1
0
.
5
1
.
1
0
.
5
0
.
6
1
.
9
0
.
4
4
.
3
4
.
7
2
.
3
1
.
6
2
.
3
2
.
9
1
.
2
2
.
4
4
.
9
7
.
6
6
.
2
3
.
5
1
.
7
I
m
p
o
r
t
e
d
e
x
p
o
r
t
c
o
m
p
o
n
e
n
t
m
a
r
k
u
p
0
.
2
3
.
2
3
.
7
2
.
2
1
.
3
1
0
.
5
3
.
7
5
.
2
0
.
9
1
.
6
2
.
7
4
.
9
6
.
3
5
.
1
3
.
2
4
.
7
5
.
4
5
.
9
5
.
5
2
.
4
E
x
p
o
r
t
m
a
r
k
u
p
1
.
5
0
.
6
3
.
4
2
.
6
0
.
6
1
5
.
0
8
.
0
2
.
1
1
.
9
0
.
5
3
.
8
2
.
6
0
.
7
0
.
1
0
.
8
0
.
6
1
.
2
1
.
8
1
.
7
0
.
6
W
a
g
e
m
a
r
k
u
p
9
.
5
1
0
.
3
1
.
1
1
.
8
0
.
7
1
.
7
9
.
7
1
1
.
4
7
.
3
2
.
3
4
.
1
7
.
5
1
1
.
0
9
.
7
4
.
3
2
.
9
5
.
1
6
.
6
7
.
2
5
.
0
R
i
s
k
p
r
e
m
i
u
m
2
.
4
1
.
4
2
.
6
2
.
1
0
.
8
1
1
.
1
6
.
1
4
.
5
2
.
8
0
.
4
0
.
7
0
.
9
3
.
7
4
.
0
1
.
7
0
.
2
0
.
4
2
.
3
3
.
5
3
.
3
I
n
t
e
r
e
s
t
r
a
t
e
1
.
9
3
.
0
1
.
6
0
.
2
0
.
4
3
.
3
4
.
4
1
.
0
0
.
1
0
.
3
3
.
8
2
.
9
1
.
9
1
.
1
0
.
8
3
.
3
2
.
4
1
.
8
1
.
3
0
.
1
I
n
ﬂ
a
t
i
o
n
t
a
r
g
e
t
1
3
.
1
2
1
.
7
2
3
.
7
2
0
.
3
1
2
.
9
2
.
1
0
.
7
3
.
6
4
.
2
3
.
0
5
.
1
2
.
8
0
.
7
3
.
3
2
.
7
4
.
0
2
.
2
0
.
1
1
.
9
6
.
1
E
n
e
r
g
y
p
r
i
c
e
s
1
.
8
0
.
1
0
.
1
0
.
1
0
.
2
2
.
8
0
.
1
0
.
1
0
.
1
0
.
2
0
.
7
0
.
9
0
.
8
0
.
6
0
.
4
2
.
3
0
.
8
0
.
7
0
.
6
0
.
2
F
i
s
c
a
l
2
.
2
4
.
4
5
.
5
5
.
2
3
.
3
6
.
8
4
.
6
6
.
6
8
.
5
9
.
0
4
.
4
6
.
7
1
0
.
4
1
1
.
0
1
0
.
0
0
.
6
0
.
7
1
.
3
2
.
2
4
.
0
W
o
r
l
d
e
c
o
n
o
m
y
(
S
V
A
R
)
8
.
7
2
1
.
1
3
7
.
2
4
9
.
0
6
3
.
1
1
5
.
7
5
.
6
2
.
4
3
.
3
3
.
3
4
.
9
2
.
6
3
.
0
5
.
4
4
.
7
3
.
9
3
.
8
5
.
2
7
.
3
9
.
0
D
i
s
t
u
r
b
a
n
c
e
E
x
p
o
r
t
s
I
m
p
o
r
t
s
R
e
a
l
e
x
c
h
a
n
g
e
r
a
t
e
I
n
t
e
r
e
s
t
r
a
t
e
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
S
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
3
1
.
5
0
.
9
1
.
0
0
.
2
1
.
8
1
.
5
3
.
1
4
.
8
2
.
0
0
.
0
1
.
1
0
.
1
0
.
2
0
.
0
3
.
7
5
.
2
4
.
2
2
.
7
0
.
8
I
n
v
e
s
t
m
e
n
t
s
p
e
c
i
ﬁ
c
t
e
c
h
n
o
l
o
g
y
0
.
4
2
.
3
1
0
.
7
1
8
.
9
1
4
.
6
3
.
4
7
.
2
1
1
.
1
3
.
3
4
.
2
0
.
1
1
.
5
2
.
8
3
.
2
1
.
7
2
.
5
5
.
2
7
.
0
5
.
6
6
.
1
A
s
y
m
m
e
t
r
i
c
t
e
c
h
n
o
l
o
g
y
1
.
2
0
.
9
2
.
1
3
.
3
4
.
2
0
.
8
1
.
4
3
.
4
3
.
6
4
.
5
0
.
5
0
.
5
0
.
4
0
.
4
0
.
3
0
.
5
0
.
3
0
.
3
0
.
3
0
.
3
N
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
1
.
2
2
.
7
8
.
3
1
6
.
2
2
9
.
5
1
.
3
3
.
6
1
1
.
4
1
4
.
3
2
6
.
1
0
.
0
0
.
2
0
.
2
0
.
1
0
.
0
0
.
0
0
.
4
0
.
9
1
.
5
2
.
4
I
n
v
e
s
t
m
e
n
t
n
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
5
0
.
9
2
.
7
6
.
5
1
4
.
7
0
.
2
1
.
3
5
.
6
7
.
5
1
3
.
1
0
.
0
0
.
5
0
.
5
0
.
3
0
.
0
0
.
0
0
.
6
1
.
4
2
.
7
4
.
8
C
o
n
s
u
m
p
t
i
o
n
p
r
e
f
e
r
e
n
c
e
1
.
3
1
.
5
1
.
6
3
.
8
0
.
3
3
.
4
4
.
3
1
.
4
7
.
1
6
.
1
2
.
1
0
.
9
0
.
6
0
.
7
0
.
0
6
.
8
9
.
9
1
0
.
3
8
.
9
4
.
6
L
a
b
o
u
r
s
u
p
p
l
y
0
.
1
1
.
8
3
.
2
0
.
9
0
.
2
0
.
8
1
.
4
1
.
8
5
.
3
4
.
0
0
.
6
1
.
4
0
.
7
0
.
0
0
.
0
2
.
2
6
.
0
6
.
2
5
.
0
1
.
9
D
o
m
e
s
t
i
c
g
o
o
d
s
m
a
r
k
u
p
0
.
9
0
.
5
2
.
5
2
.
5
0
.
5
8
.
5
0
.
4
6
.
5
4
.
5
0
.
5
1
.
5
0
.
0
0
.
8
0
.
4
0
.
1
1
.
3
4
.
8
2
.
6
1
.
0
0
.
3
I
m
p
o
r
t
e
d
c
o
n
s
u
m
p
t
i
o
n
g
o
o
d
s
m
a
r
k
u
p
5
.
4
9
.
5
6
.
7
2
.
1
0
.
4
6
.
3
9
.
6
5
.
2
0
.
9
1
.
8
1
0
.
9
1
2
.
9
3
.
8
0
.
9
0
.
0
5
.
5
9
.
7
1
.
7
0
.
9
0
.
8
I
m
p
o
r
t
e
d
i
n
v
e
s
t
m
e
n
t
g
o
o
d
s
m
a
r
k
u
p
2
.
0
3
.
9
6
.
1
6
.
4
3
.
0
5
.
0
6
.
2
2
.
8
1
.
1
2
.
0
2
.
8
2
.
5
1
.
4
1
.
0
0
.
3
1
.
3
1
.
5
3
.
4
3
.
0
0
.
1
I
m
p
o
r
t
e
d
e
x
p
o
r
t
c
o
m
p
o
n
e
n
t
m
a
r
k
u
p
1
8
.
9
3
1
.
8
1
8
.
7
5
.
9
0
.
7
3
4
.
8
2
9
.
6
1
.
1
6
.
6
2
.
0
8
.
4
4
.
6
1
.
4
0
.
6
0
.
1
8
.
6
6
.
2
6
.
5
3
.
2
1
.
2
E
x
p
o
r
t
m
a
r
k
u
p
3
0
.
0
2
7
.
5
1
2
.
2
2
.
2
0
.
6
1
7
.
7
2
4
.
4
2
0
.
2
6
.
4
0
.
5
1
0
.
7
6
.
9
1
.
9
0
.
6
0
.
1
1
3
.
2
3
.
8
1
.
0
0
.
5
0
.
2
W
a
g
e
m
a
r
k
u
p
0
.
1
2
.
0
3
.
3
0
.
6
0
.
2
1
.
0
1
.
6
2
.
2
5
.
9
4
.
1
0
.
6
1
.
5
0
.
7
0
.
0
0
.
0
2
.
5
6
.
6
6
.
7
5
.
2
2
.
0
R
i
s
k
p
r
e
m
i
u
m
9
.
6
2
.
6
6
.
9
6
.
2
0
.
7
2
.
0
2
.
1
7
.
3
7
.
1
2
.
9
2
0
.
4
0
.
7
2
.
0
1
.
0
0
.
1
1
3
.
4
7
.
0
4
.
3
4
.
0
1
.
0
I
n
t
e
r
e
s
t
r
a
t
e
1
.
9
1
.
4
0
.
3
0
.
6
0
.
3
0
.
3
0
.
3
2
.
2
1
.
7
0
.
2
3
.
6
0
.
6
0
.
2
0
.
1
0
.
0
1
1
.
7
2
.
0
0
.
9
0
.
4
0
.
2
I
n
ﬂ
a
t
i
o
n
t
a
r
g
e
t
1
.
0
0
.
9
4
.
1
3
.
5
1
.
1
1
.
1
1
.
0
0
.
5
2
.
5
5
.
7
3
.
2
0
.
3
0
.
7
0
.
2
0
.
1
2
.
9
1
0
.
3
1
2
.
5
1
3
.
9
1
1
.
9
E
n
e
r
g
y
p
r
i
c
e
s
0
.
5
0
.
5
0
.
6
0
.
5
0
.
1
0
.
2
0
.
5
1
.
3
0
.
9
0
.
2
0
.
9
0
.
2
0
.
2
0
.
1
0
.
0
1
.
9
1
.
1
0
.
6
0
.
3
0
.
1
F
i
s
c
a
l
0
.
3
0
.
4
1
.
7
2
.
2
2
.
2
0
.
5
0
.
6
0
.
7
1
.
9
3
.
5
5
.
7
5
.
4
5
.
1
4
.
1
3
.
5
6
.
3
1
.
8
3
.
7
4
.
6
3
.
8
W
o
r
l
d
e
c
o
n
o
m
y
(
S
V
A
R
)
2
4
.
6
7
.
5
7
.
5
1
6
.
9
2
6
.
4
1
0
.
9
3
.
1
1
2
.
1
1
4
.
8
1
6
.
6
2
8
.
0
5
8
.
2
7
6
.
6
8
6
.
2
9
3
.
6
1
5
.
7
1
7
.
5
2
5
.
6
3
6
.
2
5
7
.
4
101
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 102
7
T
a
b
l
e
7
.
4
.
V
a
r
i
a
n
c
e
d
e
c
o
m
p
o
s
i
t
i
o
n
s
.
R
e
g
i
m
e
1
,
p
a
r
t
2
D
i
s
t
u
r
b
a
n
c
e
E
m
p
l
o
y
m
e
n
t
R
e
a
l
w
a
g
e
s
I
n
v
e
s
t
m
e
n
t
d
e
ﬂ
a
t
o
r
C
P
I
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
1
q
4
q
8
q
1
2
q
2
0
q
S
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
1
3
.
0
6
.
3
1
.
0
0
.
6
3
.
4
4
.
1
6
.
7
9
.
4
6
.
4
2
.
4
4
.
1
5
.
2
1
.
5
0
.
3
0
.
2
6
.
9
5
.
0
0
.
4
1
.
2
0
.
5
I
n
v
e
s
t
m
e
n
t
s
p
e
c
i
ﬁ
c
t
e
c
h
n
o
l
o
g
y
4
.
5
7
.
7
1
0
.
7
8
.
7
1
5
.
4
0
.
3
2
.
7
1
1
.
2
1
8
.
1
2
0
.
0
0
.
1
0
.
4
1
.
3
5
.
7
1
1
.
2
0
.
3
0
.
9
2
.
6
5
.
7
8
.
9
A
s
y
m
m
e
t
r
i
c
t
e
c
h
n
o
l
o
g
y
0
.
3
0
.
1
0
.
0
0
.
0
0
.
1
0
.
1
0
.
1
0
.
2
0
.
2
0
.
3
0
.
3
0
.
3
0
.
1
0
.
1
0
.
2
0
.
2
0
.
3
0
.
2
0
.
3
0
.
4
N
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
1
0
.
2
0
.
7
1
.
9
5
.
9
7
.
4
1
3
.
1
2
3
.
3
2
8
.
3
3
2
.
1
0
.
3
0
.
8
0
.
9
1
.
3
1
.
3
0
.
5
1
.
0
1
.
1
1
.
2
1
.
2
I
n
v
e
s
t
m
e
n
t
n
o
n

s
t
a
t
i
o
n
a
r
y
t
e
c
h
n
o
l
o
g
y
0
.
5
0
.
8
1
.
9
4
.
0
1
2
.
2
2
.
8
4
.
3
6
.
7
8
.
4
1
1
.
6
4
.
9
1
.
5
1
.
3
2
.
6
2
.
8
0
.
8
1
.
7
2
.
1
2
.
5
2
.
5
C
o
n
s
u
m
p
t
i
o
n
p
r
e
f
e
r
e
n
c
e
9
.
2
1
3
.
4
1
5
.
3
1
3
.
7
1
2
.
3
1
.
6
3
.
2
5
.
7
6
.
3
5
.
4
0
.
1
3
.
3
3
.
9
2
.
7
1
.
5
0
.
8
3
.
2
3
.
9
2
.
8
2
.
1
L
a
b
o
u
r
s
u
p
p
l
y
6
.
2
1
2
.
0
1
6
.
0
1
2
.
7
1
.
2
1
9
.
2
1
8
.
4
6
.
8
2
.
3
3
.
4
3
.
8
7
.
4
4
.
1
1
.
6
0
.
3
5
.
6
8
.
9
3
.
0
0
.
2
0
.
4
D
o
m
e
s
t
i
c
g
o
o
d
s
m
a
r
k
u
p
9
.
4
7
.
8
1
.
8
2
.
0
3
.
4
2
5
.
9
1
3
.
1
8
.
5
3
.
6
1
.
4
9
.
8
1
.
5
1
.
1
1
.
5
0
.
4
2
0
.
1
3
.
9
2
.
7
0
.
8
0
.
7
I
m
p
o
r
t
e
d
c
o
n
s
u
m
p
t
i
o
n
g
o
o
d
s
m
a
r
k
u
p
1
.
4
2
.
1
0
.
7
1
.
2
0
.
1
2
.
0
1
.
2
0
.
8
1
.
7
1
.
9
3
.
5
1
.
5
6
.
9
4
.
6
0
.
2
9
.
7
1
.
5
6
.
7
2
.
4
0
.
5
I
m
p
o
r
t
e
d
i
n
v
e
s
t
m
e
n
t
g
o
o
d
s
m
a
r
k
u
p
1
.
0
0
.
6
3
.
8
4
.
6
0
.
3
0
.
7
0
.
4
0
.
8
2
.
0
2
.
5
3
4
.
1
2
3
.
5
2
0
.
5
6
.
3
1
.
5
0
.
7
0
.
9
0
.
0
0
.
5
1
.
1
I
m
p
o
r
t
e
d
e
x
p
o
r
t
c
o
m
p
o
n
e
n
t
m
a
r
k
u
p
5
.
0
2
.
2
7
.
3
5
.
7
4
.
1
2
.
2
1
.
4
1
.
6
2
.
4
2
.
6
4
.
3
1
.
9
1
.
4
1
.
3
0
.
6
2
.
2
2
.
8
0
.
8
0
.
6
1
.
0
E
x
p
o
r
t
m
a
r
k
u
p
1
2
.
7
9
.
8
1
.
4
2
.
0
1
.
7
1
.
6
0
.
6
0
.
5
0
.
4
0
.
6
4
.
8
0
.
1
4
.
2
2
.
6
0
.
1
1
.
6
0
.
5
1
.
0
0
.
3
0
.
5
W
a
g
e
m
a
r
k
u
p
6
.
9
1
3
.
2
1
7
.
1
1
3
.
0
0
.
7
2
3
.
8
2
0
.
9
6
.
6
3
.
0
3
.
6
4
.
2
8
.
0
4
.
2
1
.
4
0
.
3
6
.
3
9
.
5
2
.
8
0
.
5
0
.
5
R
i
s
k
p
r
e
m
i
u
m
9
.
7
6
.
6
2
.
3
4
.
7
3
.
2
1
.
1
3
.
1
1
.
5
1
.
0
1
.
2
6
.
6
5
.
6
3
.
0
0
.
3
0
.
3
4
.
4
2
.
0
3
.
2
0
.
9
0
.
4
I
n
t
e
r
e
s
t
r
a
t
e
3
.
9
4
.
4
2
.
3
0
.
3
1
.
2
1
.
7
3
.
3
3
.
0
1
.
6
0
.
8
2
.
2
1
.
6
0
.
6
0
.
5
0
.
2
1
.
9
2
.
4
1
.
0
0
.
3
0
.
3
I
n
ﬂ
a
t
i
o
n
t
a
r
g
e
t
1
.
5
0
.
0
4
.
1
6
.
4
7
.
0
0
.
7
0
.
1
3
.
4
5
.
1
3
.
7
8
.
8
1
8
.
0
1
6
.
6
1
8
.
8
1
2
.
4
9
.
9
2
0
.
5
2
0
.
9
2
0
.
2
1
2
.
6
E
n
e
r
g
y
p
r
i
c
e
s
0
.
4
0
.
3
0
.
4
0
.
5
0
.
5
1
.
2
0
.
9
0
.
8
0
.
5
0
.
3
1
.
0
0
.
3
0
.
2
0
.
2
0
.
1
1
.
3
0
.
1
0
.
2
0
.
1
0
.
1
F
i
s
c
a
l
4
.
1
5
.
4
9
.
6
1
3
.
8
2
4
.
1
1
.
7
3
.
6
4
.
5
3
.
0
1
.
1
1
.
3
3
.
4
3
.
9
4
.
9
3
.
5
1
9
.
1
1
2
.
6
1
4
.
6
1
1
.
9
4
.
7
W
o
r
l
d
e
c
o
n
o
m
y
(
S
V
A
R
)
1
0
.
3
7
.
1
3
.
5
4
.
2
3
.
0
2
.
0
2
.
7
4
.
9
5
.
7
5
.
1
5
.
9
1
5
.
6
2
4
.
4
4
3
.
4
6
3
.
0
7
.
9
2
2
.
1
3
2
.
8
4
7
.
7
6
1
.
6
102
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 103
7
7.2 Impulse response functions
Figures 7.1–7.19 present the responses of selected observable variables to speciﬁed structural
disturbances of the SOE
PL−2009
model, i.e. impulse response function (IRF). Impulse response
functions take into account the existence of all of the simultaneous and lagged interrelations
among model variables, persistence of impulses and (in the case of observable disturbances)
the possible correlation between the disturbances. It is, thus, an illustration of the dynamic
characteristics of the model and the consequences of interrelations present in the model. The
tool is used both for diagnosing the very model (enables to better understand the functioning
of the model, detect the possible incoherence) and — assuming that the model is an adequate
description of a fragment of reality — for the provision of information about the effects of dis
turbances, key reactions to macroeconomic policy variables (direction, strength, lag distribution,
etc.).
As in the case of decomposition of forecast error variances, impulse response functions presented
further herein have been determined for the point estimates of parameters corresponding to
the posterior mode. Thus, the effects of uncertainty of parameters have not been included. The
impulse has the value of one standard deviation and lasts one quarter, taking into account
autocorrelation of the disturbance. So, although each time the impulse is different, it complies
with the values typical for the sample (see Tables 6.2 and 6.1).
For observable variables expressed in annualised form as a percentage (GDP deﬂator, investment
deﬂator, CPI, inﬂation in the euro area, inﬂation in the USA and interest rates), IRF graphs
present deviations from the steady state calculated in percentage points
2
. Impulse response
functions of observable real variables (GDP, consumption, investments, consumption, export,
import, real wages), as well as exchange rates and employment are characterised with percentage
deviations from the steady state. Each of the response functions is presented for both regimes
included in the SOE
PL−2009
model, thus, it must be remembered that differences in responses to
some of the shocks result also from differences in the sizes of impulses and autocorrelations of
impulses.
From the point of view of the purpose for which the SOE
PL−2009
model has been built, interesting
are the impulse response functions enabling the analysis of the characteristics of monetary
transmission mechanism, i.e. response of the economy to monetary impulse (interest rate),
inﬂation target and risk premium. The impulse response functions of the monetary disturbance
(Figure 7.15) have standard characteristics, foreseen by the authors of DSGE models. Increase in
domestic shortterm interest rate following the impulse is transferred to the nominal and real
exchange rate, whose appreciation leads to a drop in exports. The increase of import is, however,
limited due to the income effect — rational forward looking agents reduce investments and
consumption, which is translated into a drop of GDP, employment and wages (with multiplier
effects for consumption and import). Due to the fact that dropping export comprises an import
component, which to some extent may substitute the domestic component, import ﬂuctuations
2
In technical sense, the method of presentation of the response function for growing variables enables their treatment
as the so called cumulative interim multipliers. They inform about the cumulative (in time) effect of the studied impulse
on growing observable variables.
103
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 104
7
occur in the ﬁrst quarters. Domestic currency appreciation, decrease of aggregated demand and
wages result in a drop of inﬂation
3
. Domestic prices, although rigid, include a currently optimised
component, so domestic inﬂation reacts without a lag
4
. Comparing the response functions for
the two regimes we shall notice that a relatively small change in the size of monetary impulse is
accompanied by quite a considerable change of the scale of variables responses — in the second
regime both prices and real categories react more suddenly to the impulse — in the case of CPI,
the maximum response nearly doubled. Characteristic is the fact that this has not signiﬁcantly
affected the time of shock absorption — it is similar in both regimes.
Also responses to exchange rate risk premium impulse seem to be similar to the responses
received in the class of models assuming forwardlooking behaviour of agents and motivating
the exchange rate changes with uncovered interest rate parity (UIP). In such a situation a
growth in risk premium results in immediate depreciation of (nominal and real) exchange
rate, leading to a growth of prices of imported components of consumption, investment and
export goods. This brings about a fast reduction of import and increase in export. Moreover,
imported components are being replaced with domestic components, which despite the drop of
consumption and investments result in an increase in GDP and employment. With increase in
demand for domestic products also the prices of domestically manufactured goods rise, so an
interest rate increase is inevitable. Absorption takes place quite quickly — e.g. the fast response
of interest rate to premium disturbance eliminates the scale of depreciation. A change in regime
that — according to the estimation results — took place in 1999, practically did not change
the responses of most of the observable variables to premium shock (including inﬂation), with
the exception of consumption and investments (with weaker responses in the new regime) and
employment.
Another impulse response function that is interesting for the central bank — the IRF of the
inﬂation target disturbance (see Figure 7.14) — describes the consequences of unexpected
emergence among the manufacturers and consumers of a belief (expectations?) as to the (tem
porary) higher inﬂation (inﬂation target). Due to the fact that the inﬂation target is a basis
for indexation of prices and wages in economy — which is its basic function — inﬂation rises
automatically, while that increase exceeds the scale of the shock. Despite the fact that, in the per
ception of agents, higher inﬂation target considerably alleviates interest rate policy (the central
bank accepts higher inﬂation levels), multiplier effects pushing inﬂation up above the current
target force an increase in interest rate. The real exchange rate strengthens upon relatively
large but shortlasting depreciation (only slightly affecting export) — thus, the consequences of
the expected interest rate increase prevail. Temporary exchange rate depreciation stimulates
export and output, but after the increase in interest rate and exchange rate appreciation, output,
consumption, investment, export and import fall. The impulse response functions of inﬂation
target may, thus, be treated as an illustration of the costs of inﬂation suppression, when such
phenomenon intensiﬁes. Beneﬁts (increase in production, consumption, etc.) abate quickly,
3
In the categories of a theoretical model, whose structural form we have presented in the Appendix, the sequence
of events is initiated with a growth of marginal income utility (a consequence of interest rate growth), which leads to
immediate reduction of consumption.
4
On the other hand, the cost of working capital bind directly the marginal cost of domestic manufactured products
with interest rate. In this case the interrelation diminishes the antiinﬂation effects of monetary disturbance.
104
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 105
7
while costs (higher inﬂation, drop in employment, output) are perceptible much longer. The new
regime that appeared post 1999 caused much stronger positive reactions of output, consumption
and investments in the ﬁrst quarters. The reactions were strong enough to result in a temporary
increase in employment and real wages (which has not been observed in the ﬁrst regime). In
subsequent quarters the responses of variables are similar in both regimes.
In the group of stationary technological disturbances (TFP, investmentspeciﬁc and asymmetric
disturbances), the ﬁrst of them (see Figure 7.1) represents an increase in total factor productivity
leading to a decrease in the marginal costs of domestic production and demand for production
factors (per unit of product). The effect of such a situation is a drop in the dynamics of prices
and increase in output, consumption, investments and real wages. Lower inﬂation implies a
reduction of interest rate and depreciation of real exchange rate resulting from anticipated
expansionary monetary policy, thus, contributing to increase in the share of domestic products
in the ﬁnal product. The whole leads, of course, to an increase in export and decrease in import.
Also the proportion of productivity factors changes — employment decreases and investments
increase. The fact that the wages are rigid and remuneration for capital services is ﬂexible proves
to be an important element.
105
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 106
7
Figure 7.1. Stationary technology shock
0 12 24 36
1.2
1
0.8
0.6
0.4
0.2
0
GDP deflator
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
GDP
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Consumption
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Investment
0 12 24 36
0.05
0
0.05
0.1
0.15
0.2
0.25
Exports
0 12 24 36
0.4
0.3
0.2
0.1
0
0.1
0.2
Imports
0 12 24 36
0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
Real exchange rate
0 12 24 36
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Interest rate
0 12 24 36
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Employment
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
Real wages
0 12 24 36
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Investment deflator
0 12 24 36
1
0.8
0.6
0.4
0.2
0
CPI
Ver/Regime1
Ver/Regime2
Figure 7.2. Stationary investment shock
0 12 24 36
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
GDP deflator
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
GDP
0 12 24 36
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
Consumption
0 12 24 36
0
1
2
3
4
5
Investment
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Exports
0 12 24 36
0
0.2
0.4
0.6
0.8
1
Imports
0 12 24 36
0
0.05
0.1
0.15
0.2
Real exchange rate
0 12 24 36
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
Interest rate
0 12 24 36
0.2
0.1
0
0.1
0.2
0.3
Employment
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Real wages
0 12 24 36
0.3
0.25
0.2
0.15
0.1
0.05
0
Investment deflator
0 12 24 36
0.3
0.25
0.2
0.15
0.1
0.05
0
CPI
Ver/Regime1
Ver/Regime2
106
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 107
7
Figure 7.3. Asymmetric technology shock
0 12 24 36
0.012
0.01
0.008
0.006
0.004
0.002
0
GDP deflator
0 12 24 36
0
0.01
0.02
0.03
0.04
0.05
GDP
0 12 24 36
0
0.002
0.004
0.006
0.008
0.01
Consumption
0 12 24 36
0
0.02
0.04
0.06
Investment
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
Exports
0 12 24 36
0
0.05
0.1
0.15
0.2
Imports
0 12 24 36
0.06
0.04
0.02
0
Real exchange rate
0 12 24 36
0.01
0
0.01
0.02
Interest rate
0 12 24 36
2
0
2
4
6
8
x 10
3
Employment
0 12 24 36
0
2
4
6
x 10
3
Real wages
0 12 24 36
0.05
0.04
0.03
0.02
0.01
0
Investment deflator
0 12 24 36
0.025
0.02
0.015
0.01
0.005
0
CPI
0 12 24 36
0
0.1
0.2
0.3
Euro zone GDP
0 12 24 36
1
0.5
0
0.5
1
Euro zone inflation
0 12 24 36
1
0.5
0
0.5
1
Euro interest rate
0 12 24 36
0
0.1
0.2
0.3
USA GDP
0 12 24 36
1
0.5
0
0.5
1
USA inflation
0 12 24 36
1
0.5
0
0.5
1
Dollar interest rate
0 12 24 36
1
0.5
0
0.5
1
Dollar/euro exchange rate
Ver/Regime1
Ver/Regime2
Figure 7.4. Nonstationary technology shock
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
GDP deflator
0 12 24 36
0
0.2
0.4
0.6
GDP
0 12 24 36
0
0.2
0.4
0.6
Consumption
0 12 24 36
0
0.2
0.4
0.6
0.8
Investment
0 12 24 36
0
0.2
0.4
0.6
Exports
0 12 24 36
0
0.2
0.4
0.6
Imports
0 12 24 36
0.02
0.01
0
Real exchange rate
0 12 24 36
0
0.01
0.02
0.03
0.04
0.05
Interest rate
0 12 24 36
0
0.01
0.02
0.03
0.04
Employment
0 12 24 36
0
0.2
0.4
0.6
Real wages
0 12 24 36
0
0.02
0.04
0.06
0.08
Investment deflator
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
CPI
0 12 24 36
0
0.2
0.4
0.6
Euro zone GDP
0 12 24 36
1
0.5
0
0.5
1
Euro zone inflation
0 12 24 36
1
0.5
0
0.5
1
Euro interest rate
0 12 24 36
0
0.2
0.4
0.6
USA GDP
0 12 24 36
1
0.5
0
0.5
1
USA inflation
0 12 24 36
1
0.5
0
0.5
1
Dollar interest rate
0 12 24 36
1
0.5
0
0.5
1
Dollar/euro exchange rate
Ver/Regime1
Ver/Regime2
107
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 108
7
Figure 7.5. Investment nonstationary technology shock
0 12 24 36
0
0.05
0.1
0.15
0.2
GDP deflator
0 12 24 36
0
0.1
0.2
0.3
GDP
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
Consumption
0 12 24 36
0
0.2
0.4
0.6
0.8
1
Investment
0 12 24 36
0
0.1
0.2
0.3
Exports
0 12 24 36
0
0.1
0.2
0.3
Imports
0 12 24 36
0.05
0.04
0.03
0.02
0.01
0
Real exchange rate
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
Interest rate
0 12 24 36
0
0.02
0.04
0.06
0.08
Employment
0 12 24 36
0
0.1
0.2
0.3
Real wages
0 12 24 36
0.6
0.4
0.2
0
Investment deflator
0 12 24 36
0
0.05
0.1
0.15
CPI
0 12 24 36
0
0.1
0.2
0.3
Euro zone GDP
0 12 24 36
1
0.5
0
0.5
1
Euro zone inflation
0 12 24 36
1
0.5
0
0.5
1
Euro interest rate
0 12 24 36
0
0.1
0.2
0.3
USA GDP
0 12 24 36
1
0.5
0
0.5
1
USA inflation
0 12 24 36
1
0.5
0
0.5
1
Dollar interest rate
0 12 24 36
1
0.5
0
0.5
1
Dollar/euro exchange rate
Ver/Regime1
Ver/Regime2
Figure 7.6. Consumption preference shock
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
GDP deflator
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
GDP
0 12 24 36
0
0.5
1
1.5
2
Consumption
0 12 24 36
1.5
1
0.5
0
Investment
0 12 24 36
0.5
0.4
0.3
0.2
0.1
0
0.1
Exports
0 12 24 36
0.2
0
0.2
0.4
0.6
0.8
Imports
0 12 24 36
0.25
0.2
0.15
0.1
0.05
0
0.05
Real exchange rate
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
Interest rate
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
Employment
0 12 24 36
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
Real wages
0 12 24 36
0
0.05
0.1
0.15
0.2
Investment deflator
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
CPI
Ver/Regime1
Ver/Regime2
108
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 109
7
Figure 7.7. Labour supply shock
0 12 24 36
0
0.2
0.4
0.6
0.8
1
GDP deflator
0 12 24 36
0.6
0.5
0.4
0.3
0.2
0.1
0
GDP
0 12 24 36
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Consumption
0 12 24 36
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Investment
0 12 24 36
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05
Exports
0 12 24 36
0.2
0.1
0
0.1
0.2
0.3
Imports
0 12 24 36
0.15
0.1
0.05
0
0.05
Real exchange rate
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
Interest rate
0 12 24 36
0.5
0.4
0.3
0.2
0.1
0
Employment
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Real wages
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Investment deflator
0 12 24 36
0
0.2
0.4
0.6
0.8
1
CPI
Ver/Regime1
Ver/Regime2
Figure 7.8. Domestic goods markup shock
0 12 24 36
0
0.5
1
1.5
2
2.5
3
3.5
GDP deflator
0 12 24 36
1
0.8
0.6
0.4
0.2
0
GDP
0 12 24 36
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Consumption
0 12 24 36
0.6
0.5
0.4
0.3
0.2
0.1
0
Investment
0 12 24 36
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
Exports
0 12 24 36
0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Imports
0 12 24 36
0.15
0.1
0.05
0
0.05
0.1
Real exchange rate
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Interest rate
0 12 24 36
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05
Employment
0 12 24 36
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Real wages
0 12 24 36
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Investment deflator
0 12 24 36
0
0.5
1
1.5
2
2.5
CPI
Ver/Regime1
Ver/Regime2
109
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 110
7
Figure 7.9. Imported consumption goods markup shock
0 12 24 36
0
0.05
0.1
0.15
0.2
GDP deflator
0 12 24 36
0.15
0.1
0.05
0
0.05
0.1
GDP
0 12 24 36
0.08
0.06
0.04
0.02
0
0.02
0.04
0.06
Consumption
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
Investment
0 12 24 36
2
1.5
1
0.5
0
Exports
0 12 24 36
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Imports
0 12 24 36
1.5
1
0.5
0
Real exchange rate
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
Interest rate
0 12 24 36
0.08
0.06
0.04
0.02
0
0.02
0.04
Employment
0 12 24 36
0.06
0.04
0.02
0
0.02
0.04
0.06
0.08
Real wages
0 12 24 36
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
Investment deflator
0 12 24 36
0.2
0
0.2
0.4
0.6
0.8
1
1.2
CPI
Ver/Regime1
Ver/Regime2
Figure 7.10. Imported investment goods markup shock
0 12 24 36
0.03
0.02
0.01
0
0.01
0.02
0.03
GDP deflator
0 12 24 36
0.15
0.1
0.05
0
0.05
0.1
0.15
GDP
0 12 24 36
0.05
0
0.05
0.1
Consumption
0 12 24 36
1
0.8
0.6
0.4
0.2
0
Investment
0 12 24 36
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Exports
0 12 24 36
1.2
1
0.8
0.6
0.4
0.2
0
Imports
0 12 24 36
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Real exchange rate
0 12 24 36
0.12
0.1
0.08
0.06
0.04
0.02
0
0.02
0.04
0.06
Interest rate
0 12 24 36
0.1
0.08
0.06
0.04
0.02
0
0.02
Employment
0 12 24 36
0.05
0.04
0.03
0.02
0.01
0
0.01
0.02
0.03
Real wages
0 12 24 36
1
0
1
2
3
4
5
Investment deflator
0 12 24 36
0.1
0.08
0.06
0.04
0.02
0
0.02
CPI
Ver/Regime1
Ver/Regime2
110
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 111
7
Figure 7.11. Imported export component markup shock
0 12 24 36
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
GDP deflator
0 12 24 36
0.4
0.2
0
0.2
0.4
0.6
0.8
GDP
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
0.3
Consumption
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Investment
0 12 24 36
7
6
5
4
3
2
1
0
Exports
0 12 24 36
8
7
6
5
4
3
2
1
0
Imports
0 12 24 36
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Real exchange rate
0 12 24 36
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
Interest rate
0 12 24 36
0.2
0.15
0.1
0.05
0
0.05
0.1
Employment
0 12 24 36
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
Real wages
0 12 24 36
0.6
0.5
0.4
0.3
0.2
0.1
0
Investment deflator
0 12 24 36
0.3
0.25
0.2
0.15
0.1
0.05
0
CPI
Ver/Regime1
Ver/Regime2
Figure 7.12. Export markup shock
0 12 24 36
0.15
0.1
0.05
0
0.05
0.1
GDP deflator
0 12 24 36
1.4
1.2
1
0.8
0.6
0.4
0.2
0
GDP
0 12 24 36
0
0.05
0.1
0.15
0.2
Consumption
0 12 24 36
0.25
0.2
0.15
0.1
0.05
0
Investment
0 12 24 36
8
7
6
5
4
3
2
1
0
Exports
0 12 24 36
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
Imports
0 12 24 36
0
0.2
0.4
0.6
0.8
1
1.2
Real exchange rate
0 12 24 36
0.6
0.5
0.4
0.3
0.2
0.1
0
Interest rate
0 12 24 36
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Employment
0 12 24 36
0.06
0.05
0.04
0.03
0.02
0.01
0
0.01
0.02
0.03
Real wages
0 12 24 36
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Investment deflator
0 12 24 36
0
0.05
0.1
0.15
0.2
0.25
CPI
Ver/Regime1
Ver/Regime2
111
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 112
7
Figure 7.13. Risk premium shock
0 12 24 36
0.05
0
0.05
0.1
0.15
0.2
GDP deflator
0 12 24 36
0.2
0
0.2
0.4
0.6
0.8
1
1.2
GDP
0 12 24 36
0.05
0
0.05
0.1
Consumption
0 12 24 36
0.1
0
0.1
0.2
0.3
0.4
Investment
0 12 24 36
0.5
0
0.5
1
1.5
2
Exports
0 12 24 36
0.6
0.4
0.2
0
0.2
0.4
Imports
0 12 24 36
0
0.5
1
1.5
2
Real exchange rate
0 12 24 36
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Interest rate
0 12 24 36
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Employment
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Real wages
0 12 24 36
0.2
0
0.2
0.4
0.6
0.8
1
Investment deflator
0 12 24 36
0.1
0
0.1
0.2
0.3
0.4
0.5
CPI
Ver/Regime1
Ver/Regime2
Figure 7.14. Inﬂation target shock
0 12 24 36
0
0.2
0.4
0.6
0.8
1
1.2
1.4
GDP deflator
0 12 24 36
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
GDP
0 12 24 36
0.05
0
0.05
0.1
0.15
0.2
Consumption
0 12 24 36
0.3
0.2
0.1
0
0.1
0.2
0.3
Investment
0 12 24 36
0.3
0.2
0.1
0
0.1
0.2
Exports
0 12 24 36
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
Imports
0 12 24 36
0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Real exchange rate
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
Interest rate
0 12 24 36
0.12
0.1
0.08
0.06
0.04
0.02
0
0.02
0.04
Employment
0 12 24 36
0.1
0.08
0.06
0.04
0.02
0
Real wages
0 12 24 36
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Investment deflator
0 12 24 36
0
0.2
0.4
0.6
0.8
1
1.2
1.4
CPI
Ver/Regime1
Ver/Regime2
112
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 113
7
Figure 7.15. Monetary policy (interest rate) shock
0 12 24 36
0.2
0.15
0.1
0.05
0
GDP deflator
0 12 24 36
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
GDP
0 12 24 36
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
Consumption
0 12 24 36
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Investment
0 12 24 36
0.5
0.4
0.3
0.2
0.1
0
Exports
0 12 24 36
0.12
0.1
0.08
0.06
0.04
0.02
0
Imports
0 12 24 36
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Real exchange rate
0 12 24 36
0
0.1
0.2
0.3
0.4
0.5
Interest rate
0 12 24 36
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
Employment
0 12 24 36
0.12
0.1
0.08
0.06
0.04
0.02
0
Real wages
0 12 24 36
0.3
0.25
0.2
0.15
0.1
0.05
0
Investment deflator
0 12 24 36
0.25
0.2
0.15
0.1
0.05
0
CPI
Ver/Regime1
Ver/Regime2
Figure 7.16. Energy prices shock
0 12 24 36
0
0.05
0.1
0.15
0.2
GDP deflator
0 12 24 36
0.25
0.2
0.15
0.1
0.05
0
GDP
0 12 24 36
0.05
0.04
0.03
0.02
0.01
0
Consumption
0 12 24 36
0.12
0.1
0.08
0.06
0.04
0.02
0
Investment
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
0.12
Exports
0 12 24 36
0.08
0.06
0.04
0.02
0
0.02
0.04
Imports
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
Real exchange rate
0 12 24 36
0.08
0.06
0.04
0.02
0
0.02
0.04
0.06
Interest rate
0 12 24 36
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
Employment
0 12 24 36
0.045
0.04
0.035
0.03
0.025
0.02
0.015
0.01
0.005
0
Real wages
0 12 24 36
0
0.05
0.1
0.15
Investment deflator
0 12 24 36
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
CPI
Ver/Regime1
Ver/Regime2
113
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 114
7
Figure 7.17. Capital tax shock
0 12 24 36
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
6
GDP deflator
0 12 24 36
0
2
4
6
8
10
12
14
16
x 10
7
GDP
0 12 24 36
0
0.5
1
1.5
2
2.5
3
3.5
4
x 10
6
Consumption
0 12 24 36
2
0
2
4
6
8
x 10
6
Investment
0 12 24 36
2.5
2
1.5
1
0.5
0
0.5
1
1.5
2
x 10
6
Exports
0 12 24 36
1
0
1
2
3
4
5
6
x 10
6
Imports
0 12 24 36
10
8
6
4
2
0
2
4
6
x 10
7
Real exchange rate
0 12 24 36
0
1
2
3
4
5
x 10
6
Interest rate
0 12 24 36
0
1
2
3
4
5
6
7
x 10
7
Employment
0 12 24 36
0
2
4
6
8
10
12
x 10
7
Real wages
0 12 24 36
0
0.5
1
1.5
2
2.5
3
3.5
4
x 10
6
Investment deflator
0 12 24 36
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
6
CPI
Ver/Regime1
Ver/Regime2
Figure 7.18. Employer social contribution shock
0 12 24 36
0
0.5
1
1.5
2
2.5
3
x 10
3
GDP deflator
0 12 24 36
1.4
1.2
1
0.8
0.6
0.4
0.2
0
x 10
3
GDP
0 12 24 36
1
0.8
0.6
0.4
0.2
0
x 10
3
Consumption
0 12 24 36
2
1.5
1
0.5
0
x 10
3
Investment
0 12 24 36
9
8
7
6
5
4
3
2
1
0
x 10
4
Exports
0 12 24 36
6
4
2
0
2
4
6
8
x 10
4
Imports
0 12 24 36
3.5
3
2.5
2
1.5
1
0.5
0
0.5
1
x 10
4
Real exchange rate
0 12 24 36
0
0.2
0.4
0.6
0.8
1
1.2
x 10
3
Interest rate
0 12 24 36
12
10
8
6
4
2
0
x 10
4
Employment
0 12 24 36
2
1.5
1
0.5
0
x 10
3
Real wages
0 12 24 36
0
0.5
1
1.5
2
x 10
3
Investment deflator
0 12 24 36
0
0.5
1
1.5
2
2.5
x 10
3
CPI
Ver/Regime1
Ver/Regime2
114
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 115
7
Figure 7.19. Wage markup shock
0 12 24 36
0
0.5
1
1.5
2
2.5
3
3.5
GDP deflator
0 12 24 36
2
1.5
1
0.5
0
GDP
0 12 24 36
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Consumption
0 12 24 36
2.5
2
1.5
1
0.5
0
Investment
0 12 24 36
1.2
1
0.8
0.6
0.4
0.2
0
Exports
0 12 24 36
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Imports
0 12 24 36
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
Real exchange rate
0 12 24 36
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Interest rate
0 12 24 36
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Employment
0 12 24 36
0
0.5
1
1.5
2
2.5
Real wages
0 12 24 36
0
0.5
1
1.5
2
2.5
Investment deflator
0 12 24 36
0
0.5
1
1.5
2
2.5
3
CPI
Ver/Regime1
Ver/Regime2
Investmentspeciﬁc technology shock results in an increase in efﬁciency of the transformation of
investment expenditures into ﬁxed assets. Therefore, from a unit of expenditures we receive
more capital which brings rent to households. Thus, the appearance of such shock (see Figure
7.2) results in a large increase in investment expenditures. The increase during the ﬁrst period
leads even to a fall in consumption in the pool of household expenditures. At the same time, a
drop in the price of capital services reduces the marginal cost of domestically produced goods,
which — ceteris paribus — lowers domestic inﬂation and increases export competitiveness. The
total effect is the increase in export but even bigger increase in import (due to high level of
import in investments), and thus appearance of negative net export, which gives an impulse for
real exchange rate depreciation. In the face of exchange rate depreciation and fast GDP growth,
the interest rate is rising.
In a small open economy, the rate of technological progress is identical to the rate in the rest
of the world, there may, however, be present temporary disturbances differentiating growth
dynamics. Such function is fulﬁlled by asymmetric technology disturbance. Appearance of such a
shock increases the dynamics of output in the world, increasing demand for goods exported from
the domestic economy, while the increase in dynamics in the world’s output does not impact the
world’s prices, interest rates and cross rate. The situation is illustrated by the impulse response
functions in Figure 7.3. Faster increase in export than in import leads to (nominal and real)
exchange rate appreciation, which translates into a drop in prices of imported products. The
domestic interest rate reacts stronger to the appearing supply gap than the minimally decreasing
domestic inﬂation. Generally, asymmetric technology shock weakly affects the economy, as in
115
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 116
7
the previous versions of the SOE
PL
model.
Peculiar reactions are observed for nonstationary disturbances (in technology — see Figure 7.4
and in investments — see Figure 7.5). Nonstationary nature of disturbances means a permanent
inﬂuence of shocks on the level of growing observable variables (GDP, consumption, investments,
export, import, real wages, external GDP). The shocks have no permanent impact on inﬂation,
employment, exchange rate dynamics and interest rate.
Thus, real observable variables move to a new path, while price ﬂuctuations are a consequence
of relation between the marginal cost of domestic production and the rate of growth of technical
progress µ
z,t
5
. The graphs of impulse response functions for both aforementioned disturbances
are similar and result from the same mechanism. The only difference refers to investment
deﬂator. In the case of disturbance in technology, responses of the deﬂator are the same as the
responses of other prices. In the case of shock in investments — based on disturbance deﬁnition
— we observe in its wake a drop in investment prices, as faster growth of volume reduces price
dynamics.
The consequences of consumption preference shock and labour supply (leisure preference)
shock are illustrated in Figures 7.6 and 7.7. Momentary increase in interest in consumption
means simultaneous decrease of the relative importance of the other sources of utility speciﬁed
in the function maximised by households. An immediate effect of the shock is an increase in
consumption and a drop in investments. A relative decrease in the importance of leisure leads
to a rise in employment. A combined effect results in an increase in output, employment and
domestic prices, and in real wages decrease. Due to the fact that consumption goods include
imported components, and domestic components become more expensive, import grows (despite
a drop in investments). At the same time — as a result of the increase in prices of domestic
products — the competitiveness of export diminishes, which results in drop of export. With the
rise in inﬂation of the domestically manufactured products, the increase of interest rate becomes
inevitable and leads to nominal exchange rate appreciation — the scale of appreciation is high
enough to translate into the appreciation of the real exchange rate. Changes in exchange rates
increase the depth of response of export and import. Viewing all of the responses to shock in
preferences as boiling down to increase in production and prices, the shock may be treated as a
demand shock, however, the terms “demand/supply shocks” do not ﬁt the logic of the DSGE
models.
Labour supply shock means an increase in the importance of leisure (which automatically lowers
labour supply), and thus a decrease in the relative importance of consumption in the households’
utility function. The effect of such impulse is the reduction of consumption, investments, GDP and
employment. The drop of labour supply is compensated with increased capital utilisation rate,
but due to increase in the price of labour services, the marginal costs of domestic production
rise and thus results in an increase in inﬂation. Increase in domestic prices reduces export
competitiveness and stirs import, which rises despite a drop in domestic demand (the price
effect proves to be stronger than the income effect). Negative net export along with the increase
5
More speciﬁcally, the rate affects the real rent for capital services present in the marginal cost r
t
.
116
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 117
7
in domestic prices results in real exchange rate depreciation, but in the face of inevitable interest
rate increase the exchange rate ultimately appreciates. The total view boils down to increase in
inﬂation and a drop in the level of economic activity.
Taking into account the speciﬁcation of the structural form of the model, the wage markup
shock (markup in the labour market) fulﬁls an analogical function as the labour supply shock
6
,
therefore, the response functions presented in Figure 7.19 replicate Figure 7.7, with accuracy to
the response scale. Nevertheless, the sense assigned to the disturbances is completely different.
In the case of wage markup disturbance, we speak about phenomena of institutional nature,
regulations changing (for example) the position of trade unions in wage negotiations. The IRF
for wage markup shock in both of the regimes explains why the disturbance was so important in
the ﬁrst regime.
A large group of disturbances are markup shocks (for domestically manufactured products,
imported consumer goods, investment goods, as well as reexported products and export
products, see Figures 7.8–7.12). Already a superﬁcial analysis of Phillips curve speciﬁcation for
domestic prices shows that the markup has an identical impact on inﬂation to marginal cost (see
stationary disturbance in technology, however, with a reverse sign). In the case of markup on
import products, where an analogical situation as for domestic products markup is present, there
appear more extensive effects related to substitution of imported components with domestic
production, and the resulting changes in the proportion of the components of aggregated demand.
And consequently, a shock in the markup on imported consumer goods (7.9) increases the prices
of imported components of consumption and automatically the CPI is growing (causing a
response of interest rate). Thus, general consumption is reduced, including consumption import.
Consequently, in the basket of households there will be more relatively cheaper investment goods
but despite the growth of investment import, the import in total decreases, so there appears
a positive net export and exchange rate appreciation. The increase in the share of domestic
production in consumption means also a growth of the prices of components manufactured
domestically, which stirs the growth of inﬂation. As regards disturbance in markup on investment
import, the reactions of observable variables are analogical (see Figure 7.10) — in this case,
however, consumption is growing at the cost of investments.
The shock of markup of imported export component (Figure 7.11) results in replacement of
import with domestic products, while the appearance of positive net export results in exchange
rate appreciation, which immediately impacts export. Growth of demand for domestic products
results in the growth of employment and wages, and the suddenness of this reaction results
in instantaneous growth of the interest rate (the rate depends on the level of gap and its
dynamics). When the effects of appreciation are translated into domestic prices, a similarly
sudden interest rate decrease occurs. With a lower interest rate and higher employment (at least
at the beginning), consumption and investments grow. The shock of export markup (Figure
7.12) reduces export, which immediately leads to a drop in GDP and employment, despite
import reduction. Momentarily there appears a surplus of domestic products and drop in the
6
In the formal statistical sense, the disturbances are undistinguishable. In a DSGE model we identify them only
thanks to the knowledge provided with prior distribution in Bayesian estimation.
117
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 118
7
prices of products. However, depreciation resulting from the appearance of negative net export
makes import more expensive and therefore CPI grows. In the case of drop of inﬂation for the
domestically manufactured products and a drop in GDP, the response of the interest rate is
opposite to the CPI movement — the interest rate decreases. This is caused by a large scale of
GDP drop (calculated in percentage points), with a relatively low increase in CPI inﬂation (a
tenth parts of a percentage point).
7.3 Smoothing — estimation of structural disturbances
Figures 7.20–7.21 present historical decompositions of most important structural disturbances
(and the innovations respective for the disturbances) present in the SOE
PL−2009
model. According
to the earlier suggestions, in DSGE models the disturbances are the most important factor
“driving” the ﬂuctuations of variables. An economy in longterm equilibrium is knocked out
of the equilibrium by the appearance of subsequent shocks, while the decisions made by the
agents (at macro scale) cause the absorption of the disturbances. Was it not for the disturbances,
the economy would develop in compliance with the characteristics of the steady state. In the
world of DSGE models, the very (structural) disturbances are a type of ultimate cause of the
phenomena and events that we may observe in the economy. Therefore, the question about the
sources of observable events shall be answered by searching for disturbances (innovations) that
have occurred. The structural disturbances are, however, a set of artefacts that we construct or
quantify with the use of a speciﬁc DSGE model. Two different DSGE models of the same object
in the same period, usually differently recreate the disturbances that formally fulﬁl the same
function in the models — different series are received. Therefore, the analysis of disturbances
(or broader — historical decompositions) refers to a speciﬁc model (see also Canova and Sala,
2005). Thus, the series presented in the subsequent Figures show an image of a fragment of the
history of Polish economy seen from the perspective of the SOE
PL−2009
model. This is an image
speciﬁc for the version of the model
7
.
In the SOE
PL−2009
model, it has been assumed that structural disturbances are generated by a
stochastic process of AR(1) class and only in a few cases the autocorrelation coefﬁcient takes
zero values (e.g. for interest rate disturbance) — then the graphs of disturbances and innovations
are identical. Different from zero autocorrelation coefﬁcients make nearly all of the disturbances
presented in Figure 7.20 reﬂect a relatively high inertia
8
. The largest inertia is reﬂected by the
inﬂation target disturbance and asymmetric disturbance in technology.
Due to the fact that the “natural” value of disturbances is zero, stronger positive/negative
deviations will be eliminated in a shorter or longer period of time. Therefore, on the basis of
the current estimations of disturbances and the knowledge about the way observable variables
responds, speculations may be formulated regarding the paths of observable variables. Such is
the logic of the forecasting technique with the use of a DSGE model.
7
From the technical side, the procedure of determination/estimation of nonobservable structural disturbances based
on the model having the form of state space model is called smoothing, see Hamilton (1994, Chapter 13).
8
It is worth remembering that some of the autocorrelation coefﬁcients of disturbances have been calibrated.
118
Variance decompositions, impulse response functions and estimation of disturbances
WORKING PAPER No. 83 119
7
Figure 7.20. Estimates of structural disturbances
1998 2000 2002 2004 2006 2008
6
4
2
0
2
4
Stationary technology
1998 2000 2002 2004 2006 2008
1
0
1
2
Investment specific technology
1998 2000 2002 2004 2006 2008
0.5
0
0.5
1
Asymmetric technology
1998 2000 2002 2004 2006 2008
0.4
0.2
0
0.2
Nonstationary technology
1998 2000 2002 2004 2006 2008
0.4
0.3
0.2
0.1
0
Investment nonstationary technology
1998 2000 2002 2004 2006 2008
1
0
1
2
Consumption preference
1998 2000 2002 2004 2006 2008
0.2
0
0.2
0.4
Labour supply
1998 2000 2002 2004 2006 2008
1
0
1
Domestic goods markup
1998 2000 2002 2004 2006 2008
1
0
1
2
Imported consumption goods markup
1998 2000 2002 2004 2006 2008
2
1
0
1
2
3
Imported investment goods markup
1998 2000 2002 2004 2006 2008
10
5
0
5
Imported export component markup
1998 2000 2002 2004 2006 2008
4
2
0
2
4
6
Export markup
1998 2000 2002 2004 2006 2008
0
0.5
1
Wage markup
1998 2000 2002 2004 2006 2008
4
2
0
2
4
6
Risk premium
1998 2000 2002 2004 2006 2008
0.2
0
0.2
0.4
Interest rate
1998 2000 2002 2004 2006 2008
1.5
1
0.5
0
0.5
1
Inflation target
1998 2000 2002 2004 2006 2008
2
0
2
Energy prices
Trying to interpret some of the paths of disturbances presented in the graphs, we see a speciﬁc
role of nonstationary disturbances in technology. In the previous versions of the SOE
PL
model
there was only one nonstationary disturbance in technology present. Its graph represented
approximately the world’s business cycle. Currently, the effect is broken down into two dis
turbances and is no longer so visible, particularly in the view of the fret that deep (negative)
innovations appearing at the end of the sample (related to the crisis in ﬁnancial markets) addi
tionally obscure the proportions (scale). From the beginning of the 21
st
century, nonstationary
disturbance in investments is negative, yet, there appear positive innovations of the stationary
disturbance in investments. Thus, the increase of investment dynamics observed in the years
2005–2007 was — accordingly to the logic of the model — speciﬁc for the Polish economy and
shortlasting. A similar conclusion may be formulated when viewing the stationary disturbance
in technology (of the TFP type), which has been larger than zero since 2004 (with short episode
of negative innovation around the year 2005), while strong positive innovations compensate
the negative shocks resulting from trends common for the whole global economy shaped by
nonstationary disturbances.
The graph, or even midterm trends that may be observed for the disturbance in the inﬂation
target, suggests that after a period of relative stabilisation in the vicinity of steady state in
the years 2003–2005, there appeared a trend to increase the target, i.e. the agents index the
prices of their products based on increasingly growing rates of (future) inﬂation and think
119
Variance decompositions, impulse response functions and estimation of disturbances
N a t i o n a l B a n k o f P o l a n d 120
7
Figure 7.21. Estimates of structural innovations
1998 2000 2002 2004 2006 2008
4
2
0
2
Stationary technology
1998 2000 2002 2004 2006 2008
1.5
1
0.5
0
0.5
1
Investment specific technology
1998 2000 2002 2004 2006 2008
0.6
0.4
0.2
0
0.2
Asymmetric technology
1998 2000 2002 2004 2006 2008
0.4
0.3
0.2
0.1
0
0.1
Nonstationary technology
1998 2000 2002 2004 2006 2008
0.3
0.2
0.1
0
0.1
Investment nonstationary technology
1998 2000 2002 2004 2006 2008
2
1
0
1
Consumption preference
1998 2000 2002 2004 2006 2008
0.3
0.2
0.1
0
0.1
0.2
Labour supply
1998 2000 2002 2004 2006 2008
1
0
1
Domestic goods markup
1998 2000 2002 2004 2006 2008
1
0.5
0
0.5
1
Imported consumption goods markup
1998 2000 2002 2004 2006 2008
2
0
2
4
Imported investment goods markup
1998 2000 2002 2004 2006 2008
10
5
0
5
10
Imported export component markup
1998 2000 2002 2004 2006 2008
5
0
5
Export markup
1998 2000 2002 2004 2006 2008
0
0.5
1
Wage markup
1998 2000 2002 2004 2006 2008
2
0
2
4
Risk premium
1998 2000 2002 2004 2006 2008
0.2
0
0.2
0.4
Interest rate
1998 2000 2002 2004 2006 2008
0.5
0
0.5
Inflation target
1998 2000 2002 2004 2006 2008
2
0
2
Energy prices
that the central bank also adjusts the interest rate by accepting higher inﬂation
9
. A relatively
high positive value of the disturbance implies, however, a change in the trend. In the graph of
innovation, the inﬂation episode related to the accession to the European Union is clearly visible.
The ﬂuctuations from 2000–2001 and 2006 are more difﬁcult to interpret.
Disturbance in foreign exchange risk premium approximately coincides with intuition. Larger
innovations in risk premium are related, among others, to the ﬁnancial crisis of 2008–2009, the
period of zloty liberation (1999–2000) and accession to the European Union. In the graph of the
premium disturbance a phase of strong (speculative?) appreciation in the years of 2006–2008 is
visible, with strong depreciation later on.
Interest rate disturbance (originally identical with innovations) reﬂects more sudden deviations
around the years 1999–2000, which seems to be natural taking into account the introduction
of the ﬂoating exchange rate regime and implementation of the strategy of direct inﬂation
targeting. Explicit consideration of a structural change to interest rate rule resulted in a more
homogenous graph of the disturbance path — earlier versions of the model identiﬁed phases of
sudden ﬂuctuations in the years 1997–1999 and relative stabilisation (clear decrease in variance)
after the year 2001. The ﬂuctuations at the end of the sample suggest, however, the strongest
discretionary negative impulse of interest rate in the last 10 years.
9
Of course, an opposite conclusion may be formulated for the period before the year 2004.
120
Historical decompositions and forecasts
WORKING PAPER No. 83 121
8
Chapter 8
Historical decompositions and
forecasts
8.1 Historical decompositions
The set of calculation procedures that has been built for models in the state space representation
enables to carry out historical decompositions and, thus, to study the impact of shocks (groups
of shocks) on the path of observable variables in the whole historical context of events. The
procedures allow making counterfactual scenarios to answer the question of how the paths
of variables would have been if some of the shocks had not occurred. Therefore, we assess
the inﬂuence and role of the given disturbances for the actual series of events. We are not
referring here a theoretical or potential impact of disturbances on the observable variables
of the model, as in the case of variance decompositions or impulse response functions, but
about the actual inﬂuence on historical events. Variance decompositions and impulse response
functions are determined by the assumption that economy is in steady state, while the studied
disturbance innovation (shock) appears in isolation and its effects are not distorted by other
shocks. Historical decompositions are made in consideration of possible deviations from the
steady state of all of the variables, as well as overlapping/neutralisation of the effects of other
shocks. Using such tools, we may, thus, make an attempt to identify disturbances that had the
largest importance for the given variables or/and historical episodes. As a presentation of the
analytical capacity of the tools, we show an exercise regarding the search for key disturbances
in the graphs of the particular observable variables (the analysis is limited to the period post
2004).
Figures 8.1–8.2 present the paths of observable variables, their historical graphs (green line)
and the graph of their hypothetical development (blue line). In the ﬁrst case we take into
account only the disturbance in consumption preference and in the second case — all of
121
8
Historical decompositions and forecasts
Historical decompositions and forecasts
N a t i o n a l B a n k o f P o l a n d 122
8
the disturbances are present, except the disturbance in consumption preference. In each case
we receive paths of variables resulting from the deviation of economy from the steady state,
the considered disturbances and the relationships between the endogenous variables of the
model. The aforementioned paths show, therefore, that in the studied period consumption
was a category dependent practically on the exogenous shock of preference only, the role of
ﬁscal or technological shocks or work supply preference mentioned as important in variance
decompositions (particularly in a longer horizon, see Chapter 7.1) is hard to perceive. Carrying
out analogical study we shall notice that the path of investment expenditures is determined
by a stationary technological disturbance in investments. It is difﬁcult to observe the role of
nonstationary shocks or preferences. In the case of export a single disturbance cannot be found
to explain the dominating/larger part of the variance of the variable (enabling the recreation of
its historical path). It has been determined by experiment that for the recreation of the graph of
export in the studied fragment of the sample, external disturbances included in the SVAR model
are necessary (except interest rates), as well as disturbances in export markup and markup on
the import of export components. Key factors of import proved to be the disturbances in markup
on imported goods (consumer and investment goods, and export components), as well as export
markups.
The method proposed above enables to state that GDP deﬂator ﬂuctuations result in the studied
period from the realisation of the disturbance in inﬂation target and markup on domestically
manufactured products. On the other hand, the path of the investment deﬂator may be recreated
with the use of disturbance in the markup on imported investment goods and stationary
technological disturbance in investments. Both for the GDP deﬂator and the investment deﬂator
in the last 2–3 years, the exchange rate risk premium is important as well. For the recreation
(arrival at a relatively correct approximation) of the path of real wages in the years 2004–2009,
four disturbances are necessary: wage markup, labour supply preferences, inﬂation target and
markup on domestically manufactured products.
As shown in Figures 8.3 and 8.4, the historical graph of the real dollar/zloty exchange rate may
be recreated with the use of risk premium disturbance and three components of the SVAR model:
change in the nominal dollar/euro rate and inﬂation in the USA and the euro area (the real
cross rate). For the other observable variables (GDP, CPI, domestic interest rate, employment) it
is very difﬁcult to determine a narrow set of shocks to allow the recreation of the paths in the
tested sample.
8.2 Forecasting technique
The SOE
PL−2009
model is an example of a linear DSGE model formulated in a state space
representation. For that class of models the initial stage of making a forecast is the identiﬁcation
of the value of the vector of state variables (with the use of Kalman ﬁlter) at the starting point
of the forecast, i.e. determination in what state, from the perspective of the model, the economy
is. Having the values for the vector of state variables at the starting point, we are able to make a
conditional or unconditional forecast, point forecast or stochastic forecast. Due to the fact that
all of the exogenous variables of the model (disturbances) are described with known stochastic
122
Historical decompositions and forecasts
WORKING PAPER No. 83 123
8
Figure 8.1. Historical decomposition; excluded consumption preference shock
2004 2005 2006 2007 2008 2009
4
2
0
2
4
6
8
10
GDP deflator
2004 2005 2006 2007 2008 2009
1
0.5
0
0.5
1
1.5
2
2.5
GDP
2004 2005 2006 2007 2008 2009
4
3
2
1
0
1
2
3
Consumption
2004 2005 2006 2007 2008 2009
3
2
1
0
1
2
3
4
5
6
Investment
2004 2005 2006 2007 2008 2009
8
6
4
2
0
2
4
6
Exports
2004 2005 2006 2007 2008 2009
10
8
6
4
2
0
2
4
6
8
Imports
2004 2005 2006 2007 2008 2009
20
15
10
5
0
5
10
Real exchange rate
2004 2005 2006 2007 2008 2009
4.5
5
5.5
6
6.5
Interest rate
2004 2005 2006 2007 2008 2009
3
2
1
0
1
2
Employment
2004 2005 2006 2007 2008 2009
0.5
0
0.5
1
1.5
2
Real wages
2004 2005 2006 2007 2008 2009
8
6
4
2
0
2
4
6
8
10
Investment deflator
2004 2005 2006 2007 2008 2009
0
1
2
3
4
5
6
CPI
Yhat
Yobs
processes, no external assumptions regarding exogenous variables are necessary. Unconditional
forecast, assuming that in the future no new shocks shall occur, means that all of the disturbances
identiﬁed at the starting point shall expire and upon an adequately long period of time — due
to the structure of the model — the variables shall come back to steady state. This is illustrated
e.g. in Appendix C. In the simplest case, such prepared forecast may be point forecast, yet, a
more natural environment for the estimated DSGE models (including SOE
PL−2009
) is the world
taking into account the existence of uncertainties. In our case it is possible to quantify the risk
related to:
• uncertainty of parameters — when (full) Bayesian estimation is made, we have the informa
tion about whole distributions, and in the case of classical estimation, the standard errors
may be used;
• state uncertainty — considering the uncertainty related to the forecast of state variables;
• uncertainty of future shocks — information about the variances of shocks enables the
generation of uncertainty related to the realisation of shocks in the future;
• measurement errors — forecasts of observable variables are additionally burdened with
uncertainty related to the measurement (observation) of observable variables.
The above uncertainties are taken into account by conducting stochastic simulations where
we draw parameters (from the posterior distribution), state variables at the starting point, a
sequence of structural shocks and a sequence of measurement errors. We then end up with the
estimation of the predictive distribution which can be depicted in the form of a fan chart or used
123
Historical decompositions and forecasts
N a t i o n a l B a n k o f P o l a n d 124
8
Figure 8.2. Historical decomposition; included only consumption preference shock
2004 2005 2006 2007 2008 2009
4
2
0
2
4
6
8
10
GDP deflator
2004 2005 2006 2007 2008 2009
1
0.5
0
0.5
1
1.5
2
2.5
GDP
2004 2005 2006 2007 2008 2009
4
3
2
1
0
1
2
3
Consumption
2004 2005 2006 2007 2008 2009
2
1
0
1
2
3
4
5
6
Investment
2004 2005 2006 2007 2008 2009
6
4
2
0
2
4
6
Exports
2004 2005 2006 2007 2008 2009
10
8
6
4
2
0
2
4
6
8
Imports
2004 2005 2006 2007 2008 2009
20
15
10
5
0
5
10
Real exchange rate
2004 2005 2006 2007 2008 2009
4
4.5
5
5.5
6
6.5
Interest rate
2004 2005 2006 2007 2008 2009
2
1.5
1
0.5
0
0.5
1
1.5
2
Employment
2004 2005 2006 2007 2008 2009
0.5
0
0.5
1
1.5
2
Real wages
2004 2005 2006 2007 2008 2009
8
6
4
2
0
2
4
6
8
10
Investment deflator
2004 2005 2006 2007 2008 2009
0
1
2
3
4
5
6
CPI
Yhat
Yobs
to calculate probabilities of reaching certain levels by certain variables.
Although the list does not account for all of the potentially important sources of uncertainties
(e.g. broader understood uncertainty of the model, data, uncertainty according to F. Knight, see
e.g. Kłos (2004)), the very fact of the possibility of (selective) quantiﬁcation of major risk factors,
determination of e.g. conﬁdence intervals, shows potential of such class of DSGE models.
With the use of the SOE
PL−2009
model there may also be created conditional forecasts (scenarios),
i.e. with the assumed path of one or more observable variables
1
. Due to the fact that the principle
that shocks explain the behaviour of variables is valid all the time, the user must indicate
what disturbances are to guarantee the fulﬁlment of the assumed path, which enables explicit
determination of the values of shocks
2
. A modest intervention test (modesty metric) by Leeper
and Zha (2003), see also Adolfson et al. (2005a), allows to verify whether the values of such
shocks are not too high. High values of shocks lower the likelihood of conditional forecast, as in
reality they could be treated by economic agents as a change in the prevailing economic regime
or might change the agent’s behaviour (according to Lucas critique).
1
Such forecasts are known in literature as conditional forecasts by WaggonerZha or LeeperZha, see e.g. Waggoner
and Zha (1999).
2
When preparing a scenario, it is possible to start with the shocks, i.e. declare speciﬁc values of the shocks as events
whose effects we study. Although disturbances have economic interpretation, which may help us in determining the
correct value of innovation, the most frequent method of conduct requires monitoring of the response of observable
data, as to which we have intuition and knowledge, and only that gives us basis for ﬁnal determination of the value of
the shocks.
124
Historical decompositions and forecasts
WORKING PAPER No. 83 125
8
Usually, in the above technique of constructing conditional forecasts, unanticipated shocks are
used. In certain situations, the conditional paths may be generally known (e.g. expected tax
changes) and may not/should not be treated as unexpected events. Then, it is possible to use
the anticipated shocks that shall enable agents to react to the very information about a future
event and the realisation of a shock in future shall not be a surprise.
A special case of conditional forecasts, interesting from the point of view of the central bank, are
forecasts using the assumptions with regard to interest rate. According to the historical tradition,
forecasts — or rather projections — are (often) created at central banks with the assumption of
exogenous interest rate (constant or implied by market expectations). In the world of rational
agents (forwardlooking and optimising) it is hard, however, to answer the question what would
their responses (decisions) be, when the central bank stopped to use the rule known to the
agents for several or a dozen or so quarters, i.e. when it stopped to modify the interest rate
accordingly to the occurring events. A break in the functioning of a central bank lasting many
quarters does not comply with the standards, so the knowledge that the central bank shall start
to act on a “nonstandard” basis shall force the agents to “nonstandard” behaviours. If — by
force of assumption — we exclude the possibility of foreseeing by the agents the “nonstandard”
approach of the central bank, we arrive at a contradiction with the initial assumption about the
rationality of the agents. In the ﬁrst case we receive “irrational” forecasts of variables (the effect
of assumption of “nonstandard” policy of the bank — the problem has been discussed by e.g.
Laséen and Svensson (2009)). In the second case, the possible projections shall be internally
Figure 8.3. Historical decomposition; excluded risk premium and real dollar/euro rate shocks
2004 2005 2006 2007 2008 2009
5
0
5
10
GDP deflator
2004 2005 2006 2007 2008 2009
6
5
4
3
2
1
0
1
2
3
GDP
2004 2005 2006 2007 2008 2009
4
3
2
1
0
1
2
3
Consumption
2004 2005 2006 2007 2008 2009
2
1
0
1
2
3
4
5
6
Investment
2004 2005 2006 2007 2008 2009
20
15
10
5
0
5
Exports
2004 2005 2006 2007 2008 2009
10
5
0
5
10
Imports
2004 2005 2006 2007 2008 2009
20
15
10
5
0
5
10
Real exchange rate
2004 2005 2006 2007 2008 2009
2
3
4
5
6
7
8
9
10
Interest rate
2004 2005 2006 2007 2008 2009
4
3
2
1
0
1
2
3
Employment
2004 2005 2006 2007 2008 2009
0.5
0
0.5
1
1.5
2
2.5
Real wages
2004 2005 2006 2007 2008 2009
5
0
5
10
Investment deflator
2004 2005 2006 2007 2008 2009
0
1
2
3
4
5
6
7
8
CPI
Yhat
Yobs
125
Historical decompositions and forecasts
N a t i o n a l B a n k o f P o l a n d 126
8
Figure 8.4. Historical decomposition; included only risk premium and real dollar/euro rate
shock
2004 2005 2006 2007 2008 2009
4
2
0
2
4
6
8
10
GDP deflator
2004 2005 2006 2007 2008 2009
1
0
1
2
3
4
5
6
7
GDP
2004 2005 2006 2007 2008 2009
4
3
2
1
0
1
2
3
Consumption
2004 2005 2006 2007 2008 2009
2
1
0
1
2
3
4
5
6
Investment
2004 2005 2006 2007 2008 2009
5
0
5
10
15
Exports
2004 2005 2006 2007 2008 2009
10
8
6
4
2
0
2
4
6
8
Imports
2004 2005 2006 2007 2008 2009
20
15
10
5
0
5
10
Real exchange rate
2004 2005 2006 2007 2008 2009
2
3
4
5
6
7
8
9
Interest rate
2004 2005 2006 2007 2008 2009
2
1.5
1
0.5
0
0.5
1
1.5
2
Employment
2004 2005 2006 2007 2008 2009
0.5
0
0.5
1
1.5
2
Real wages
2004 2005 2006 2007 2008 2009
8
6
4
2
0
2
4
6
8
10
Investment deflator
2004 2005 2006 2007 2008 2009
0
1
2
3
4
5
6
CPI
Yhat
Yobs
inconsistent and shall be subject to Lucas critique. In each case the projection will not provide a
reliable, methodologically correct answer to the asked question. A solution of the problem is the
preparation of forecasts with the assumption of endogenous interest rate.
8.3 Expost forecasting accuracy of the SOE
PL−2009
model
Using the observations of observable data Y
t
for t = 0, 1, 2, ..., T, the quality of forecasts of the
model in a sample may be veriﬁed by comparison of the theoretical values of the observable
data identiﬁed as
ˆ
Y
t
, with their real values in the sample Y
t
for t = 1, 2, ..., T −1. The observable
variables are forecasted for h = 1, 2, ..., H periods ahead, where H is the maximum forecast
horizon. The forecast of observable variables for period t +h made in period t is denoted as
ˆ
Y
t+h
, and the actual value of the observable variables in period t +h as Y
t+h
. The difference:
γ
t+h
= Y
t+h
−
ˆ
Y
t+h
is called a forecast error in period t for the h forecast horizon. Going through t = 0, 1, 2, ..., T −h,
we may determine various statistics of forecast errors γ
t+h
depending on the determined forecast
horizon h =1, 2, ..., H. Based thereon, the measures of the forecasting quality of the model are
constructed. The analysis will cover onedimensional and multidimensional measures (joint
measures).
126
Historical decompositions and forecasts
WORKING PAPER No. 83 127
8
In the exercises presented below, we assume the maximum forecast horizon of H = 12 quarters,
and the sample covers the period of 2004:4–2009:3, i.e. the period after Poland’s accession
to the European Union. Thus, the test covers the most turbulent period of ﬁnancial crisis
(2008:2–2009:3), which may hardly be deemed typical or representative.
8.3.1 Onedimensional measures of forecasting quality
From the group of onedimensional forecasting quality measures we have used two: root mean
square error (RMSE) and mean percentage error (MPE).
The root mean square error for the forecast horizon h = 1, 2, ..., H for the i
th
observable variable
is calculated as:
RMSE =
T−h
t=0
(γ
t+h
(i))
2
T −h+1
,
where γ
t+h
(i) is the i
th
element of the vector of γ
t+h
. RMSE tells about how much the
ˆ
Y
t+h
forecast is different on the average from the actual value of the observable variable Y
t
. The
mean percentage error for the forecast horizon h = 1, 2, ..., H for the i
th
observable variable is
calculated as:
MPE =
T−h
t=0
γ
t+h
(i)
Y
t+h
T −h+1
.
The value tells by what percentage the
ˆ
Y
t+h
forecast is different on the average from the real
value of the observable variable Y
t
. The basic function of this measure is the reﬂection of the
possible bias. If the actual value of the forecast variable takes values close to zero, the relation
γ
t+h
(i)
Y
t+h
becomes very large even in the case of moderate γ
t+h
(i).
The analysis of RMSE shows several systematic features of the forecasting properties of the
DSGE SOE
PL−2009
model. Firstly, the inﬂation forecast errors (domestic inﬂation, consumption
goods, investment goods and CPI) increase to the horizon of about 3–4 quarters and then fall.
This means that least susceptible to errors is the forecast of inﬂation for 1–2 quarters, and
then around the horizon of 7–11 quarters, depending on the type of inﬂation. The inﬂation in
the perspective of 3–5 quarters is associated with the largest error. The forecasting accuracy
of the model with respect to forecast horizon is different for other variables where the RMSE
increase along the horizon. Secondly, beside single cases, the model’s forecasts always win with
the naive forecasts. The advantage of a DSGE model over naive forecasts increases along the
forecast horizon. Thirdly, the inspection of MPE errors shows that the forecasts of most of the
variables are biased, which means that they are systematically underestimated or overestimated,
regardless of the forecast horizon. The conclusions have been formulated based on the data that
are analysed in details in Appendix D.
127
Historical decompositions and forecasts
N a t i o n a l B a n k o f P o l a n d 128
8
8.3.2 Multidimensional measures of forecasting accuracy
Multidimensional analysis of forecast errors of the model
3
is based on the covariance matrix of
errors Ω
h
, which is determined based on the residuals vector γ
t+h
= Y
t+h
=
ˆ
Y
t+h
for the forecast
horizon h = 1, 2, ..., H:
Ω
h
=
T−H
t=1
γ
t+h
Φγ
T
t+h
T −h
,
where Ψ is the scaling matrix. In the calculations it has been assumed that Ψ is a diagonal
matrix with variances of the subsequent forecasts errors on the main diagonal. For matrix
Ω
h
its eigenvalues λ
h
i
have been determined, as well as the respective v
h
i
eigenvectors. The
determination of the eigenvectors for matrix Ω
h
is equivalent to the determination of the so
called principal components of γ
t+h
forecasts for t = 0, 1, 2, ..., T −H. Principal components are
orthogonal directions along which the speciﬁed value of forecast error variations (variances)
materializes. Each eigenvector of matrix Ω
h
determines one such direction. The larger the
eigenvalue related to the eigenvector, the larger the variability of errors in the corresponding
direction. Should we select eigenvectors or directions — principal components to which the
largest eigenvalues correspond, and check which variables contribute to variability of these
directions to the largest extent, it shall prove which variables generate the largest errors in
the forecasting process. Thus, it is easy to quantify contribution of subsequent variables to the
variance along any given direction, since these contributions are proportional to squares of their
coordinates in the eigenvector deﬁning that direction. Such analysis may be made for every
forecast horizon h =1, 2, ..., H, which gives a clear picture of forecast errors depending on the
forecast horizon. For the presentation purposes, we provide results for horizons h =1, 4, 8, 12
and for each horizon for the directions related to the four largest eigenvalues.
Tables 8.1–8.4 present how the share of variables generating the largest forecast errors in the
total variance of the errors change depending on the forecast horizon. Variables of at least
10% contributions were considered. Table 8.1 shows that the ﬁrst — most important — of the
identiﬁed directions, realizes mainly the errors of export, import and real exchange rate of the
dollar. The contribution of each of the variables is signiﬁcant, regardless of the forecast horizon,
however, the share of exchange rate diminishes for the horizon of over 9–10 quarters.
Table 8.1. The ﬁrst direction with the largest variance of forecast errors
Variable h=1 h=4 h=8 h=12
Export 0.26 0.27 0.20 0.37
Import 0.40 0.33 0.33 0.25
Real USD/PLN exchange rate 0.20 0.30 0.40 0.27
Note: The eigenvalues for the ﬁrst direction amount to 72, 478, 407 and 583 for the forecast horizons of 1, 4, 8 and 12,
respectively.
Table 8.2 shows that in the second of the veriﬁed directions the largest share in the forecast
errors has the real exchange rate of the dollar, with the share for the horizons over 8 quarters,
3
An interesting review of the techniques of DSGE models forecast errors analysis is presented by Adolfson et al.
(2005c).
128
Historical decompositions and forecasts
WORKING PAPER No. 83 129
8
Table 8.2. The second direction with the largest variance of forecast errors
Variable h=1 h=4 h=8 h=12
Investments 0.05 0.07 0.04 0.13
Export 0.04 0.04 0.16 0.07
Import 0.08 0.16 0.20 0.24
Real USD/PLN exchange rate 0.68 0.62 0.41 0.25
Nominal USD/EUR exchange rate 0.04 0.00 0.11 0.16
Note: The eigenvalues for the second direction amount to 26, 110, 104 and 188 for the forecast horizons of 1, 4, 8 and
12, respectively.
amounting even to 80%. For the horizon between 1 and 6 quarters, a major share falls to export
and between 7 and 10 quarters — import. For short horizon a contribution that may not be
neglected is brought by import and EUR/USD exchange rate, while the share of investments is
moderate, yet, realise for most of the horizons.
Table 8.3. The third direction with the largest variance of forecast errors
Variable h=1 h=4 h=8 h=12
Investments 0.00 0.01 0.26 0.45
Export 0.00 0.00 0.01 0.14
Government expenditers 0.19 0.23 0.13 0.00
Nominal USD/EUR exchange rate 0.62 0.65 0.35 0.23
Note: The eigenvalues for the third direction amount to 11, 25, 32 and 45 for the forecast horizons of 1, 4, 8 and 12,
respectively.
As it may be seen from Table 8.3, in the third of the identiﬁed directions, in the case of short
horizons of 1–4 quarters, the dominating contribution to forecast errors is made by investments,
while for horizons longer than 6 quarters — government expenditures and, mainly, the nominal
USD/EUR exchange rate.
In the fourth of the identiﬁed directions — see Table 8.4 — for horizons up to 5–6 quarters,
the largest share in forecast errors falls to investments and nominal USD/EUR exchange rate;
for horizons of 6–8 quarters — export and nominal USD/EUR exchange rate, while for the 12
quarters horizon the forecasts are dominated by investments.
Table 8.4. The fourth direction with the largest variance of forecast errors
Variable h=1 h=4 h=8 h=12
Inﬂation in investments. 0.10 0.46 0.57 0.1
Investments 0.52 0.46 0.57 0.1
Government expenditures 0.05 0.00 0.10 0.21
Real USD/PLN exchange rate 0.28 0.00 0.02 0.02
Nominal USD/EUR exchange rate 0.13 0.20 0.00 0.00
Note: The eigenvalues for the fourth direction amount to 5, 20, 10 and 7 for the forecast horizons of 1, 4, 8 and 12,
respectively.
8.3.3 Rolling forecasts
Figures 8.5–8.6 present the historical graphs of observable variables (thick line) and a series
of forecasts for 12 quarters horizon (made in compliance with the description provided at the
129
Historical decompositions and forecasts
N a t i o n a l B a n k o f P o l a n d 130
8
beginning of paragraph 8.3). In the graphs annual dynamics of variables has been presented
4
.
Using this illustration, the characteristics of forecasting errors of the SOE
PL2009
model, given by
formal errors measures, may be supplemented with the illustration of the ability of the model to
forecast midterm trends, turning points or sensitivity of forecasts to new information.
The ﬁrst observation is that the inaccuracies of forecasts resulting from the earlier presented,
onedimensional measures seem to be slightly exaggerated. In the case of many variables
(e.g. consumption, investments), relatively large root mean square errors result from single
observations (individual large errors during the ﬁnancial crisis). Contrary to the above presented
measures, consumption and inﬂation (CPI, GDP deﬂator) or even investments seem to be
relatively well forecasted. An exception is the period when the economy was exposed to the
shocks of the global ﬁnancial crisis. When such sudden shocks occur, subsequent forecasts
(for example) become instable (subsequent forecasts differ widely). Tendency of the model to
underestimate the dynamics of export and import is conﬁrmed (the effects of which are reﬂected
also in the GDP forecasts), as well as small explanatory power of the model in the case of the
exchange rate. It should be emphasized however, that the forecasts of several turning points of
investments, GDP and inﬂation are accurate.
Figure 8.5. Ex post forecasts for the selected variables of the model (part 1)
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
0
2
4
6
Inflation  GDP delator
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
2
0
2
4
Inflation  investment deflator
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
0
1
2
3
4
5
Inflation  CPI
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
1
2
3
4
5
6
7
GDP
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
4
2
0
2
4
6
8
Consumption
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
0
5
10
15
20
Investments
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
20
15
10
5
0
5
10
Export
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
20
10
0
10
Import
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
0
2
4
6
8
Real wages
4
The exception is employment, which has the form of deviations (absolute values) from the HP trend, and real
exchange rate of dollar expressed in the form of deviations from the linear trend.
130
Historical decompositions and forecasts
WORKING PAPER No. 83 131
8
Figure 8.6. Ex post forecasts for the selected variables of the model (part 2)
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
2
1
0
1
2
3
Employment
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
20
15
10
5
0
5
10
Real exchange rate of dollar
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
4
5
6
7
Interest rate
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
4
2
0
2
Production in the euro area
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
2
0
2
4
Production in the USA
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
0
0.5
1
1.5
2
2.5
Inflation in the euro area
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
1
1.5
2
2.5
3
3.5
Inflation in the USA
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
0
1
2
3
4
5
Interest rate in the euro area
04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4
1
2
3
4
5
6
Interest rate in the USA
131
Final comments
N a t i o n a l B a n k o f P o l a n d 132
Final comments
The presented material documents the results of works over a version of an estimated dynamic
stochastic general equilibrium model SOE
PL−2009
, which in 2010 will be used for building
midterm forecasts and projections of inﬂation processes and business cycle in Poland, i.e.
shall support and supplement the forecasting materials prepared based on the traditional
macroeconometric model and the opinions of experts at the National Bank of Poland.
DSGE models are one of the most important tools of theoretical analyses of modern macroeco
nomics. New theoretical concepts are developed based on DSGE models or the environment of
economy in dynamic general equilibrium. Being an effective tool of theoretical research, DSGE
models are also becoming useful tools for empirical research. More complex estimation of param
eters, particularly with the use of the ideas of Bayesian econometrics, enables matching models
motivated with economic theory to data — referring to the real existing economies, testing the
actual episodes, analysing the reasons of observable events. All of that from the point of view
of explicitly declared economic paradigm (model speciﬁcation). Strict theoretical grounds and
explicitly declared paradigm are features that simplify the interpretation of results but — as it
seemed initially — lower the forecasting potential of models. When we were interested in the
accuracy of forecasts, eclectic models in which more or less random correlations of data enable
the reduction of errors dominated the models with clearer (more explicit) economic contents.
Therefore forecasting experiments with Bayesian estimated DSGE models carried out by F. Smets
and R. Wouters (2004) enjoyed high interest of the institutions managing the macroeconomic
policy. The experiments showed that an estimated DSGE model has not only analytical but also
forecasting potential. The conclusions were conﬁrmed also for other estimated DSGE models
of developed and stable market economies (see e.g. Adolfson et al., 2005c; Christoffel et al.,
2007b).
Forecasting of inﬂation or business cycles for stable developed market economies is different
from the forecasting of the responses of agents subjected for a dozen or so years to institutional
changes, societies that attempt building market institutions from scratch and shape the macro
132
Final comments
Final comments
WORKING PAPER No. 83 133
and microeconomic rationality of ﬁrms, households, social organisations and state institutions.
Unchanging interrelationships assumed within the DSGE model (deep parameters deﬁning the
structure of the model of economy) are more difﬁcult to identify, while the general uncertainty
is much higher. Therefore the task of building a DSGE model for the Polish economy enabling
the construction of midterm forecasts and projections proved to be a harder task than the
construction of a analogical model e.g. for the euro area countries. The experiments with the
family of SOE
PL
models enable, however, to formulate a cautious conclusion that also for the
group of economies in which Poland is included, a DSGE model of signiﬁcant analytical and
forecasting potential may be developed
5
. Using the word “potential” of the model we wish
to emphasise that although the progress achieved in reference to the earlier versions of the
SOE
PL
family of models is signiﬁcant, the current features of the SOE
PL−2009
model, including
the forecasting characteristics, require further work, while the development potential of DSGE
models guarantees the effectiveness of such projects.
5
This is conﬁrmed by the experience of the analysts dealing with DSGE models at the central banks of Hungary and
the Czech Republic (see Benesz et al., 2005; Andrle et al., 2009; Jakab and Világi, 2008).
133
Part IV
Appendix
134
List of equations, list of variables
WORKING PAPER No. 83 135
Appendix A
List of equations, list of variables
Forms of the model
The structural form of the model can be written as:
_
t
_
α
0
˜z
t+1
+α
1
˜z
t
+α
2
˜z
t−1
+β
0
θ
t+1
+β
1
θ
t
_
= 0
θ
t
= ρ θ
t−1
+
t
,
(A.1)
where:
˜
z
t [1x28]
=
_
_
_
_
_
_
_
_
_
_
_
_
_
1 2 3 4 5 6 7 8 9 10
´ π
d
t
´ π
mc
t
´ π
mi
t
´ π
mx
t
´ π
x
t
´
w
t
´
H
t
´c
t
´
i
t
´
ψ
z,t
...
11 12 13 14 15 16 17 18 19 20
...
´p
k
,t
∆
´
S
u
t
´y
t
´
k
t+1
´ u
t
´q
t
´ m
t+1
´ µ
t
´ a
t
´ a
e
t
...
21 22 23 24 25 26 27 28
...
´ γ
mcd
t
´ γ
mid
t
´ γ
mxd
t
´ γ
x,u
t
´x
u
t
´x
x
t
´
R
t
´
E
t
_
_
_
_
_
_
_
_
_
_
_
_
_
, (A.2)
θ
t [1×48]
=
_
θ
s
t [1×26]
, θ
τ
t [1×8]
, θ
t [1×14]
_
, (A.3)
θ
s
t
=
_
_
_
_
_
_
_
_
_
_
_
_
_
1 2 3 4 5 6 7 8 9 10
´ ε
t
´
Υ
t
´
˜ z
t
´ µ
z,t
´ µ
Ψ,t
´
ζ
c
t
´
ζ
h
t
´
ζ
q
t
´
λ
d
t
´
λ
mc
t
...
11 12 13 14 15 16 17 18 19 20
...
´
λ
mi
t
´
λ
mx
t
´
λ
x
t
´
λ
w
t
¯
˜
φ
e
t
¯
˜
φ
e
t−1
¯
˜
φ
u
t
¯
˜
φ
u
t−1
´ ε
R,t
´
π
c
t
...
21 22 23 24 25 26
...
´ ν
w
t
´ ν
k
t
´ τ
k
t
´ τ
w
t
´ τ
s
t
´ π
oil
t
_
_
_
_
_
_
_
_
_
_
_
_
_
,
θ
τ
t
=
_
27 28 29 30 31 32 33 34
´ τ
p
t
´ τ
y
t
´ τ
c
t
´g
t
´ τ
p
t−1
´ τ
y
t−1
´ τ
c
t−1
´g
t−1
_
,
135
Appendix A
List of equations, list of variables
List of equations, list of variables
N a t i o n a l B a n k o f P o l a n d 136
θ
t
=
35 36 37 38 39 40 41
y
e
t
π
e
t
R
e
t
y
u
t
π
u
t
R
u
t
∆
S
x
t
...
42 43 44 45 46 47 48
...
y
e
t−1
π
e
t−1
R
e
t−1
y
u
t−1
π
u
t−1
R
u
t−1
S
x
t−1
.
Upon the solution of the model with AndersonMoore algorithm, anticipated variables are
eliminated (among others) from the structural form. Thus, we can write:
˜ z
t+1
= A˜ z
t
+B θ
t+1
θ
t+1
=ρθ
t
+ε
t+1
or
˜ z
t+1
θ
t+1
ξ
t+1
=
A B ρ
0 ρ
F
ξ
˜ z
t
θ
t
ξ
t
+
B
I
ε
t+1
υ
t+1
.
The state space model has the form of:
ξ
t+1
= F
ξ
ξ
t
+υ
t+1
, (υ
t+1
υ
t+1
) =Q,
Y
t
= A
x
x
t
+H
ξ
t
+u
t
, (u
t
u
t
) = R.
(A.4)
The elements of the matrices of the above formulas are functions of DSGE model parameters.
136
List of equations, list of variables
WORKING PAPER No. 83 137
List of model variables
The list of endogenous variables of the SOE
PL−2009
model, components of vector ˜ z
t
1. π
d
t
— inﬂation of domestic intermediate products [ =
P
d
t
P
d
t−1
]
2. π
x
t
— inﬂation of export products [ =
P
x
t
P
x
t−1
]
3. π
mc
t
— inﬂation of imported consumer goods [ =
P
mc
t
P
mc
t−1
]
4. π
mi
t
— inﬂation of imported investment goods [ =
P
mi
t
P
mi
t−1
]
5. π
mx
t
— inﬂation of imported export components [ =
P
mx
t
P
mx
t−1
]
6. w
t
— real wages
7. H
t
— hours worked
8. c
t
— consumption
9. i
t
— investments
10. ψ
z,t
— marginal utility of income
11. p
k
,t
— relative price of ﬁxed assets
12. ∆S
u
t
— growth of nominal PLN/USD exchange rate
13. y
t
— gross domestic product
14. k
t+1
— ﬁxed assets
15. u
t
— ﬁxed assets’ utilisation rate
16. q
t
— cash
17. m
t+1
— money
18. µ
t
— money growth rate
19. a
t
— total net foreign assets
20. a
e
t
— net foreign assets denominated in euro
21. γ
mcd
t
— ratio of the prices of imported consumption
goods to domestic prices [ =
P
mc
t
P
d
t
]
22. γ
mid
t
— ratio of the prices of imported investment goods
to domestic prices [ =
P
mi
t
P
d
t
]
23. γ
mxd
t
— ratio of the prices of imported components of export
to domestic prices [ =
P
mx
t
P
d
t
]
24. γ
xu
t
— ratio of export prices to the prices in the rest of the world [ =
P
x
t
P
u
t
]
25. x
u
t
— real USD/PLN exchange rate
26. x
x
t
— real USD/EUR exchange rate
27. R
d
t
— domestic shortterm interest rate (gross)
28. E
t
— employment
137
List of equations, list of variables
N a t i o n a l B a n k o f P o l a n d 138
List of exogenous variables of the SOE
PL−2009
model, components of vector
˜
θ
t
, structural
disturbances
1. ε
t
— stationary disturbance in technology (TFP)
2. Υ
t
— stationary disturbance in investments
3. ˜ z
t
— stationary asymmetric disturbance
4. µ
z,t
— nonstationary disturbance in technology
5. µ
Ψ,t
— nonstationary technological disturbance speciﬁc to investments
6. ζ
c
t
— disturbance in consumption preferences
7. ζ
h
t
— disturbance in labour preferences (labour supply)
8. ζ
q
t
— disturbance in preferences of cash holdings
9. λ
d
t
— disturbance in markups on domestic intermediate products
10. λ
mc
t
— disturbance in markups on imported consumption goods
11. λ
mi
t
— disturbance in markups on imported investment goods
12. λ
mx
t
— disturbance in markups on imported export components
13. λ
x
t
— disturbance in markups on exported products
14. λ
w
t
— disturbance in wage markup
15.
˜
φ
e
t
— disturbance in risk premium in the euro market
16.
˜
φ
e
t−1
— as above
17.
˜
φ
u
t
— disturbance in risk premium in the dollar market
18.
˜
φ
u
t−1
— as above
19. ε
R,t
— disturbance in interest rate (monetary policy)
20. π
c
t
— disturbance in inﬂation target
21. ν
w
t
— disturbance in demand for working capital loan
for ﬁnancing labour services
22. ν
k
t
— disturbance in demand for working capital loan
for ﬁnancing capital services
23. τ
k
t
— disturbance in capital tax
24. τ
w
t
— disturbance in national insurance contributions paid by the employees
25. τ
s
t
— disturbance in national insurance contributions paid by the employers
26. π
oil
t
— disturbance in raw materials (energy, oil) prices
List of exogenous variables of the SOE
PL−2009
model, components of vector
˜
θ
t
, observable
disturbances
27. τ
p
t
— disturbance in effective rate of corporate income tax
28. τ
y
t
— disturbance in effective rate of income tax
29. τ
c
t
— disturbance in effective rate of consumption tax
30. g
t
— government consumption
List of exogenous variables of the SOE
PL−2009
model, components of vector
˜
θ
t
, structural
disturbances of the global economy
35. y
e
t
— disturbance in GDP in the euro area
36. π
e
t
— disturbance in inﬂation in the euro area
37. R
e
t
— disturbance in shortterm interest rate in the euro area
38. y
u
t
— disturbance in GDP in the USA
39. π
u
t
— disturbance in inﬂation in the USA
40. R
u
t
— disturbance in shortterm interest rate of the dollar
41. ∆S
x
t
— disturbance in the change of nominal USD/EUR exchange rate
138
List of equations, list of variables
WORKING PAPER No. 83 139
A3. List of equations of the structural form of the model
Below we present the loglinearised equations of the structural form of the model (as given by
A.1). Loglinearised variables are identiﬁed with hats.
Inﬂation of domestic prices
_
1 +κ
d
β µ
_
´ π
d
t
=
_
1 −κ
d
__
1 −β µρ
π
_
´
π
c
t
+β µ ´ π
d
t+1
+κ
d
´ π
d
t−1
+
_
1 −ξ
d
_
ξ
d
_
1 −β µξ
d
_
_
´
w
t
+
´
H
t
−
´
k
t
−´ u
t
+
ν
f w
R
R
f w
´
R
t−1
+
ν
f w
(R −1)
R
f w
´ ν
f w
t
+
1 −
´ µ
Ψ,t
+´ µ
z,t
+
τ
s
1 +τ
s
´ τ
s
t
+ν
τ
_
´ π
oil
t
− ´ π
u
t
_
− ´ ε
t
+
´
λ
d
t
_
(A.5)
Inﬂation of prices of imported consumption goods
_
1 +κ
mc
β µ
_
´ π
mc
t
=
_
1 −κ
mc
__
1 −β µρ
π
_
´
π
c
t
+β µ ´ π
mc
t+1
+κ
mc
´ π
mc
t−1
+
_
1 −ξ
mc
_
ξ
mc
_
1 −β µξ
mc
_
_
´x
u
t
+ (ω
mcu
−1) ´x
x
t
−
c
d
c γ
cd
´ γ
mcd
t
+
´
λ
mc
t
_
(A.6)
Inﬂation of prices of imported investment goods
_
1 +κ
mi
β µ
_
´ π
mi
t
=
_
1 −κ
mi
__
1 −β µρ
π
_
´
π
c
t
+β µ ´ π
mi
t+1
+κ
mi
´ π
mi
t−1
+
_
1 −ξ
d
_
ξ
mi
_
1 −β µξ
mi
_
_
´x
u
t
+
_
ω
miu
−1
_
´x
x
t
+
c
m
c γ
cmc
´ γ
mcd
t
−´ γ
mid
t
+
´
λ
mi
t
_
(A.7)
Inﬂation of prices of imported export components
_
1 +κ
mx
β µ
_
´ π
mx
t
=
_
1 −κ
mx
__
1 −β µρ
π
_
´
π
c
t
+β µ ´ π
mx
t+1
+κ
mx
´ π
mx
t−1
+
_
1 −ξ
mx
_
ξ
mx
_
1 −β µξ
mx
_
_
´x
u
t
+ (ω
mxu
−1) ´x
x
t
+
c
m
c γ
cmc
´ γ
mcd
t
−´ γ
mxd
t
+
´
λ
mx
t
_
(A.8)
Inﬂation of prices of export products
_
1 +κ
x
β µ
_
´ π
x
t
=
_
1 −κ
x
__
1 −β µρ
π
_
´
π
c
t
+β µ ´ π
x
t+1
+κ
x
´ π
x
t−1
+
_
1 −ξ
x
_
ξ
x
_
1 −β µξ
x
_
_
−
c
m
c γ
cmc
´ γ
mcd
t
−´ γ
xu
t
− ´x
u
t
+ω
x
´ γ
mxd
t
+
´
λ
x
t
_
(A.9)
Real wages
b
w
β ξ
w
´ π
d
t+1
−b
w
ξ
w
_
1 +κ
w
β
c
d
c γ
cd
_
´ π
d
t
+ b
w
ξ
w
κ
w
c
d
c γ
cd
´ π
d
t−1
−b
w
κ
w
β ξ
w
c
m
c γ
cmc
´ π
mc
t
+ b
w
ξ
w
κ
w
c
m
c γ
cmc
´ π
mc
t−1
+b
w
β ξ
w
´
w
t+1
+
_
λ
w
σ
L
− b
w
_
1 +β ξ
2
w
__
´
w
t
+ b
w
ξ
w
´
w
t−1
−(λ
w
−1)
´
ψ
z
+
,t
+ (λ
w
−1) σ
L
´
H
t
+(λ
w
−1)
´
ζ
h
t
− b
w
β ξ
w
_
1 −κ
w
_
´
π
c
t+1
+ b
w
ξ
w
_
1 −κ
w
_
´
π
c
t
+(λ
w
−1)
´
λ
w
t
+
_
τ
w
(λ
w
−1)
1 −τ
w
_
´ τ
w
t
+
_
τ
y
(λ
w
−1)
1 −τ
y
_
´ τ
y
t
= 0
(A.10)
139
List of equations, list of variables
N a t i o n a l B a n k o f P o l a n d 140
Demand for labour services
´
H
t
=
1
(1 −)
´y
t
−
1
(1 −)
´ ε
t
+
1
(1 −)
1 −
´ µ
Ψ,t
+
(1 −)
´ µ
z,t
−
(1 −)
´
k
t
−
(1 −)
´ u
t
(A.11)
Consumption
_
µ
1−
Ψ
µ
z
_
b´c
t−1
−
_
_
µ
1−
Ψ
µ
z
_
2
+β b
2
_
´c
t
+
_
µ
1−
Ψ
µ
z
_
b β´c
t+1
−
_
β b −
_
µ
1−
Ψ
µ
z
___
b −
_
µ
1−
Ψ
µ
z
__
c
m
c γ
cmc
´ γ
mcd
t
−
_
β b −
_
µ
1−
Ψ
µ
z
___
b −
_
µ
1−
Ψ
µ
z
__
´
ψ
z
+
,t
− b
_
µ
1−
Ψ
µ
z
_
1 −
´ µ
Ψ,t
+β b
_
µ
1−
Ψ
µ
z
_
1 −
´ µ
Ψ,t+1
− b
_
µ
1−
Ψ
µ
z
_
´ µ
z,t
+β b
_
µ
1−
Ψ
µ
z
_
´ µ
z,t+1
−
_
β b −
_
µ
1−
Ψ
µ
z
___
b −
_
µ
1−
Ψ
µ
z
__
τ
c
(1 +τ
c
)
´ τ
c
t
−
_
µ
1−
Ψ
µ
z
_ _
b −
_
µ
1−
Ψ
µ
z
__
´
ζ
c
t
+β b
_
b −
_
µ
1−
Ψ
µ
z
__
´
ζ
c
t+1
= 0
(A.12)
Investments
´p
k
,t
+
´
Υ
t
−ω
i
_
γ
imi
_
η
i
−1
´ γ
mid
t
+
_
µ
z
µ
1
1−
Ψ
_
2
˜
S
´
i
t−1
−
_
1 +β
_
_
µ
z
µ
1
1−
Ψ
_
2
˜
S
´
i
t
+β
_
µ
z
µ
1
1−
Ψ
_
2
˜
S
´
i
t+1
−
_
µ
z
µ
1
1−
Ψ
_
2
˜
S
´ µ
z,t
+
_
µ
z
µ
1
1−
Ψ
_
2
˜
S
β´ µ
z,t+1
−
_
µ
z
µ
1
1−
Ψ
_
2
˜
S
_
1
1 −
´ µ
Ψ,t
_
+
_
µ
z
µ
1
1−
Ψ
_
2
˜
S
β
_
1
1 −
´ µ
Ψ,t+1
_
= 0
(A.13)
Marginal utility of income
´
ψ
z
+
,t
=
´
ψ
z
+
,t+1
−
_
1 −
´ µ
Ψ,t+1
+ ´ µ
z,t+1
_
− ´ π
t+1
−
τ
k
_
1 −τ
k
_
πµ
z
+ −β
µ
z
+ π
´ τ
k
t+1
+
πµ
z
+ −βτ
k
µ
z
+ π
´
R
t
(A.14)
140
List of equations, list of variables
WORKING PAPER No. 83 141
Relative prices of ﬁxed assets
´p
k
,t
= −
´
ψ
z
+
,t
+
´
ψ
z
+
,t+1
+β
1 −δ +δτ
p
_
µ
z
µ
1
1−
Ψ
_ ´p
k
,t+1
−
τ
p
(1 −τ
p
)
_
µ
z
µ
1
1−
Ψ
_
−β
_
µ
z
µ
1
1−
Ψ
_ ´ τ
p
t+1
+
_
µ
z
µ
1
1−
Ψ
_
−β (1 +δτ
p
−δ)
_
µ
z
µ
1
1−
Ψ
_
_
_
ν
w
R
R
f w
−
ν
k
R
R
f k
_
´
R
t
+
ν
w
(R −1)
R
f w
´ ν
w
t+1
−
ν
k
(R −1)
R
f k
´ ν
k
t+1
+
´
w
t+1
+
´
H
t+1
−
_
´ u
t+1
+
´
k
t+1
_
+
τ
s
1 +τ
s
´ τ
s
t+1
_
−
1
1 −
_
_
_
_
β (1 +δτ
p
−δ)
_
µ
z
µ
1
1−
Ψ
_
_
_
_
_
´ µ
Ψ,t+1
−
_
_
_
_
β (1 +δτ
p
−δ)
_
µ
z
µ
1
1−
Ψ
_
_
_
_
_
´ µ
z,t+1
(A.15)
Dynamics of the nominal dollar/zloty exchange rate
_
1 −φ
e
s
_
∆
´
S
u
t+1
=
_
1 −φ
e
s
_
∆
´
S
x
t+1
−φ
e
s
∆
´
S
x
t
+φ
e
s
∆
´
S
u
t
+
_
´
R
t
−
´
R
e
t
_
+
˜
φ
e
a
a
e
´ a
e
t
−
¯
˜
φ
e
t
(A.16)
Real income, balance for the real GDP
´y
t
=
g
y
´g
t
+
c
d
y
´c
t
+
c
d
y
η
c
c
m
c γ
cmc
´ γ
mcd
t
+
i
d
y
´
i
t
+
i
d
y
η
i
i
m
i γ
imi
´ γ
mid
t
+
x
d
x
x
y
_
y
e
y
´y
e
t
+
_
1 −
y
e
y
_
´y
u
t
−
_
η
f u
+ (η
f e
−η
f u
)
y
e
y
_
´ γ
x,u
t
−η
f e
y
e
y
´x
x
t
+
´
˜ z
,+
t
_
+
x
y
x
d
x
η
x x
ω
x
´ γ
mxd
t
+
k
y
r
k
_
µ
z
µ
1
1−
Ψ
_´ u
t
(A.17)
Fixed assets
´
k
t+1
=
1 −δ
_
µ
z
µ
1
1−
Ψ
_
´
k
t
+
_
µ
z
µ
1
1−
Ψ
_
+δ −1
_
µ
z
µ
1
1−
Ψ
_
´
i
t
+
_
µ
z
µ
1
1−
Ψ
_
+δ −1
_
µ
z
µ
1
1−
Ψ
_
´
Υ
t
−
1 −δ
_
µ
z
µ
1
1−
Ψ
_´ µ
z,t
−
1 −δ
_
µ
z
µ
1
1−
Ψ
_
1
1 −
´ µ
Ψ,t
(A.18)
Fixed assets’ utilisation rate
´ u
t
=
1
1 +σ
a
_
_
ν
w
R
R
f w
−
ν
k
R
R
f k
_
´
R
t−1
+
ν
w
(R −1)
R
f w
´ ν
w
t
−
ν
k
(R −1)
R
f k
´ ν
k
t
+
´
w
t
+
´
H
t
−
´
k
t
+
1
1 −
´ µ
Ψ,t
+ ´ µ
z,t
+
τ
s
1 +τ
s
´ τ
s
t
_
−
1
1 +σ
a
ν
aτ
_
∆
´
S
u
t
+ ´ π
oil
t
_
(A.19)
Cash
´q
t
=
1
σ
q
_
´
ζ
q
t
−
´
ψ
z
+
,t
+
τ
k
_
1 −τ
k
_ ´ τ
k
−
R
(R −1)
´
R
t−1
_
(A.20)
141
List of equations, list of variables
N a t i o n a l B a n k o f P o l a n d 142
Money
´ m
t+1
= ´ µ
t
+ ´ m
t
− ´ π
d
t
−
1 −
´ µ
Ψ,t
− ´ µ
z,t
(A.21)
Money growth rate
´ µ
t
=
q
m
´q
t
− ´ m
t
+ ´ π
t
+
ν
k
r
k
k R
m
_
µ
z
µ
1
1−
Ψ
_
_
ν
w
R
f w
−
ν
k
R
f k
_
´
R
t−1
+
ν
k
r
k
k
m
_
µ
z
µ
1
1−
Ψ
_
_
1 −
ν
k
(R −1)
R
f k
_
´ ν
k
t
+
ν
w
m
_
_
_
_
ν
k
r
k
k
_
µ
z
µ
1
1−
Ψ
_
(R −1)
R
f w
+ (1 +τ
s
) w H
_
_
_
_
´ ν
w
t
+
_
_
_
_
ν
k
r
k
k
m
_
µ
z
µ
1
1−
Ψ
_ +
ν
w
m
(1 +τ
s
) w H
_
_
_
_
_
´
w
t
+
´
H
t
+
τ
s
1 +τ
s
´ τ
s
t
_
+
1 −
´ µ
Ψ,t
+ ´ µ
z,t
+
_
1 −
q
m
_
ν
τ
_
´ π
oil
t
− ´ π
u
t
_
(A.22)
Total net foreign assets — classic version
´ a
t
=−c
m
´c
t
−i
m
´
i
t
+ x
_
1 −
_
1 −u
x
_
_
η
f u
+ (η
f e
−η
f u
)
y
e
y
__
´ γ
x,u
t
− x
_
u
c
ω
mcu
+u
i
ω
miu
+u
x
ω
mxu
−1 +
_
1 −u
x
_
η
f e
y
e
y
_
´x
x
t
+
c
m
c
d
c
η
c
γ
cd
´ γ
mcd
t
+
i
m
i
d
i
η
i
γ
id
´ γ
mid
t
+ x
m
η
x x
_
1 −ω
x
_
´ γ
mxd
t
+ x
_
1 −u
x
_
_
y
e
y
´y
e
t
+
_
1 −
y
e
y
_
´y
u
t
+
´
˜ z
,+
t
_
+
−Ra
e
πµ
z
+
_
_
´
R
u
t−1
−
´
R
e
t−1
_
+ a
e
_
˜
φ
u
+
˜
φ
e
a
_
´ a
e
t−1
−
_
1 +
˜
φ
u
a
a
e
_
1
a
e
´ a
t−1
+
_
˜
φ
e
s
−
˜
φ
u
s
_
∆
´
S
u
t
+
_
1 −
˜
φ
e
s
_
∆
´
S
x
t
+
_
˜
φ
e
s
−
˜
φ
u
s
_
∆
´
S
u
t−1
−
˜
φ
e
s
∆
´
S
x
t−1
+
¯
˜
φ
u
t−1
−
¯
˜
φ
e
t−1
_
(A.23)
142
List of equations, list of variables
WORKING PAPER No. 83 143
Total net foreign assets — balance of total incomes and total expenditures
´ a
t
+ g ´g
t
+
_
c
d
+ c
m
_
´c
t
+
_
i
d
+ i
m
_
´
i
t
=
=
−Ra
e
πµ
z
+
_
_
´
R
u
t−1
−
´
R
e
t−1
_
+ a
e
_
˜
φ
u
+
˜
φ
e
a
_
´ a
e
t−1
−
_
1 +
˜
φ
u
a
a
e
_
1
a
e
´ a
t−1
+
_
˜
φ
e
s
−
˜
φ
u
s
_
∆
´
S
u
t
+
_
1 −
˜
φ
e
s
_
∆
´
S
x
t
+
_
˜
φ
e
s
−
˜
φ
u
s
_
∆
´
S
u
t−1
−
˜
φ
e
s
∆
´
S
x
t−1
+
¯
˜
φ
u
t−1
−
¯
˜
φ
e
t−1
_
− x
__
u
c
ω
mcu
+u
i
ω
miu
+u
x
ω
mxu
_
−1
_
´x
x
t
+ x ´ γ
xu
t
+
c
d
c
m
c
η
c
γ
cd
_
1 −γ
mcd
_
´ γ
mcd
t
+
i
d
i
m
i
η
i
γ
id
_
1 −γ
mid
_
´ γ
mid
t
+ y
_
λ
d
−1
λ
d
_
´y
t
+
y (1 −)
λ
d
´
H
t
+
y
λ
d
´
k
t
+
y
λ
d
_
R
f k
−1
R
f k
_
´ u
t
+
y
λ
d
_
(1 −)
R
f w
−1
R
f w
+
R
f k
−1
R
f k
−1
_
ν
τ
_
´ π
oil
t
− ´ π
u
t
_
+
y
λ
d
´ ε
t
−
y
λ
d
1 −
´ µ
Ψ,t
−
y
λ
d
´ µ
z,t
(A.24)
Net foreign assets denominated in euro
´ a
e
t
=
1
a
e ˜
φ
u
a
_
_
´
R
t
−
´
R
u
t
_
−
_
1 −
˜
φ
u
s
_
∆
´
S
u
t+1
+
˜
φ
u
a
´ a
t
+
˜
φ
u
s
∆
´
S
u
t
−
¯
˜
φ
u
t
_
(A.25)
Relation of prices of imported consumption goods
´ γ
mcd
t
= ´ γ
mcd
t−1
+ ´ π
mc
t
− ´ π
d
t
(A.26)
Relation of prices of imported investment goods
´ γ
mid
t
= ´ γ
mid
t−1
+ ´ π
mi
t
− ´ π
d
t
(A.27)
Relation of prices of imported export products
´ γ
mxd
t
= ´ γ
mxd
t−1
+ ´ π
mx
t
− ´ π
d
t
(A.28)
Relation of export prices to the prices in the dollar area
´ γ
xu
t
=
_
´ γ
xu
t−1
+ ´ π
x
t
− ´ π
u
t
_
(A.29)
Real dollar/zloty exchange rate
´x
u
t
= ´x
u
t−1
+∆
´
S
u
t
+ ´ π
u
t
−
_
1 −ω
c
_
_
γ
cd
_
η
c
−1
´ π
d
t
−ω
c
_
γ
cmc
_
η
c
−1
´ π
mc
t
(A.30)
Real dollar/euro exchange rate
´x
x
t
=
_
ρ
x x
_
´x
x
t−1
+ ´ π
u
t
− ´ π
e
t
+∆
´
S
x
t
(A.31)
143
List of equations, list of variables
N a t i o n a l B a n k o f P o l a n d 144
Nominal interest rate
R
t
=ρ
R
R
t−1
+
1 −ρ
R
1 − r
π
π
c
t
+ r
∆π
c
d
c γ
cd
π
d
t
+ r
∆π
c
m
c γ
cmc
π
mc
t
+
1 −ρ
R
r
π
− r
∆π
c
d
c γ
cd
π
d
t−1
+
1 −ρ
R
r
π
− r
∆π
c
m
c γ
cmc
π
mc
t−1
+ r
∆y
y
t
+
1 −ρ
R
r
y
− r
∆y
y
t−1
+
1 −ρ
R
r
x
x
eu
t−1
+ε
R,t
(A.32)
Employment
β ξ
e
˜
E
t+1
+
1 −β ξ
e
1 −ξ
e
H
t
−
1 +β ξ
2
e
˜
E
t
+ξ
e
˜
E
t−1
=0 (A.33)
Steady state solution
Based on theoretical analyses, we assume that in the steady state the following are satisﬁed:
λ
x
= 1, λ
mx
= 1, γ
xu
= 1, γ
xd
= 1, γ
mxd
=1, ℵ
0
y
= x
then the values of other variables are determined in compliance with the following equations:
Technology growth
µ
z
+ = µ
1−
Ψ
µ
z
(A.34)
Inﬂation, inﬂation target
π =
µ
µ
z
+
(A.35)
Nominal interest rate
R =
µ−τ
k
β
1 −τ
k
β
(A.36)
Interest rate of working capital loan for labour services
R
f w
= ν
w
R +1 −ν
w
(A.37)
Interest rate of working capital loan for capital services
R
f k
= ν
k
R +1 −ν
k
(A.38)
Relation of prices of imported consumption goods to domestic prices
γ
mcd
= λ
mcd
γ
xd
gdy γ
xd
= 1 to γ
mcd
=λ
mc
(A.39)
144
List of equations, list of variables
WORKING PAPER No. 83 145
Relation of prices of imported investment goods to domestic prices
γ
mid
= λ
mid
γ
xd
gdy γ
xd
= 1 to γ
mid
= λ
mi
(A.40)
Relation of prices of consumption goods to domestic prices
γ
cd
=
_
_
1 −ω
c
_
+ω
c
(λ
mc
)
1−η
c
_ 1
1−η
c (A.41)
Relation of prices of investment goods to domestic prices
γ
id
=
_
_
1 −ω
i
_
+ω
i
_
λ
mi
_
1−η
i
_ 1
1−η
i
(A.42)
Relation of prices of consumption goods to the prices of imported consumption goods
γ
cmc
=
_
_
1 −ω
c
_
(λ
mc
)
η
c
−1
+ω
c
_ 1
1−η
c
(A.43)
Relation of prices of investment goods to the prices of imported investment goods
γ
imi
=
_
_
1 −ω
i
_
_
λ
mi
_
η
i
−1
+ω
i
_ 1
1−η
i
(A.44)
Rent for lease of ﬁxed assets services
r
k
=
_
β (1 −τ
p
)
µ
z
+ µ
Ψ
−β [1 +δτ
p
−δ]
_
−1
γ
id
(A.45)
Real wages
w =
1
(1 +τ
s
) R
f w
_
λ
d
_
1
_
_
1
1 −
_
1−
_
R
f k
r
_
_ 1
−1
(A.46)
Relation of capital to labour
_
k
H
_
=
1 −
R
f w
R
f k
1
r
k
_
(1 +τ
s
) wµ
Ψ
µ
z
+
_
(A.47)
145
List of equations, list of variables
N a t i o n a l B a n k o f P o l a n d 146
Auxiliary variables
D
1
=
_
_
1 −ω
c
_
_
γ
cd
_
η
c
+ω
c
_
γ
cmc
_
η
c
_
D
2
=
_
_
1 − g
r
_
_
1
µ
Ψ
µ
z
+
_
_
k
H
_
−
_
_
1 −ω
i
_
_
γ
id
_
η
i
+ω
i
_
γ
imi
_
η
i
_
_
µ
z
+ µ
Ψ
+δ −1
µ
z
+ µ
Ψ
_
k
H
_
D
3
=
_
1
A
L
(1 −τ
y
) (1 −τ
w
)
w
λ
w
_ 1
σ
L
D
4
=
1
γ
cd
(1 +τ
c
)
_
µ
z
+ −β b
µ
z
+ − b
_
(A.48)
Labour services
H =
_
D
1
D
4
D
2
D
σ
L
3
_ 1
1+σ
L
(A.49)
Fixed assets services
k = H
k
H
(A.50)
Total consumption
c =
D
2
D
1
H (A.51)
Domestically manufactured consumer goods
c
d
=
_
1 −ω
c
_
_
γ
cd
_
η
c
c (A.52)
Imported consumer goods
c
m
= ω
c
_
γ
cmc
_
η
c
c (A.53)
Investment expenditures
i = k
µ
z
+ µ
Ψ
+δ −1
µ
z
+ µ
Ψ
(A.54)
Domestic component of investment goods
i
d
=
_
1 −ω
i
_
_
γ
id
_
η
i
i (A.55)
146
List of equations, list of variables
WORKING PAPER No. 83 147
Imported component of investment goods
i
m
= ω
i
γ
imi
η
i
i (A.56)
Imported component of export
x
m
= ω
x
x (A.57)
Total export
x =
c
m
+ i
m
1 −ω
x
(A.58)
Production
y =
µ
−
1−
Ψ
µ
−
z
H
k
H
(A.59)
Broad money
m =
ν
k r
k
k
µ
Ψ
µ
z
+
+ν
w
(1 +τ
s
) w H
1 −A
q
(A.60)
Cash
q = A
q
m (A.61)
147
Global economy SVAR model
N a t i o n a l B a n k o f P o l a n d 148
Appendix B
Global economy SVAR model
SVAR model identiﬁcation
In the estimation of the reduced form of the global economy model, the following set of
restrictions has been assumed for the dynamics of a SVAR model
1
:
y
e
t
π
e
t
r
e
t
y
u
t
π
u
t
r
u
t
x
t
=
∗ ∗ 0 ∗ ∗ ∗ ∗
∗ ∗ 0 ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ 0 ∗
∗ ∗ ∗ ∗ ∗ 0 ∗
∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ 0 ∗ ∗ ∗ ∗
y
e
t−1
π
e
t−1
r
e
t−1
y
u
t−1
π
u
t−1
r
u
t−1
x
t−1
+
∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ 0 ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ ∗ 0
∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ ∗ 0
∗ ∗ 0 ∗ ∗ ∗ ∗
y
e
t−2
π
e
t−2
r
e
t−2
y
u
t−2
π
u
t−2
r
u
t−2
x
t−2
+e
t
.
(B.1)
Results of estimation of the SVAR model
Results of estimation of the reduced form of the SVAR model are the following
2
:
y
e
t
π
e
t
r
e
t
y
u
t
π
u
t
r
u
t
x
t
=
1.15 0.15 0.00 0.12 0.11 −0.04 −0.03
0.06 0.28 0.00 0.00 0.10 0.04 0.00
0.05 −0.06 1.10 0.03 0.12 −0.02 0.00
0.39 0.11 −0.35 0.96 0.36 0.00 0.00
0.03 0.10 0.18 0.05 0.34 0.00 0.00
0.06 −0.05 −0.24 0.05 0.25 0.66 0.00
−0.13 −0.51 0.00 0.19 3.21 −0.38 0.22
y
e
t−1
π
e
t−1
r
e
t−1
y
u
t−1
π
u
t−1
r
u
t−1
x
t−1
+
+
−0.31 −0.43 −0.10 −0.13 0.11 0.26 0.02
0.00 0.28 0.00 −0.03 0.19 0.04 0.00
0.01 −0.07 −0.22 −0.05 0.12 0.02 0.00
−0.28 −0.24 −0.12 −0.16 0.14 −0.42 0.02
−0.02 0.10 −0.26 −0.02 0.37 −0.08 0.00
−0.06 −0.06 0.09 −0.01 0.27 0.08 0.00
0.23 −2.47 0.00 −0.85 1.25 2.12 −0.11
y
e
t−2
π
e
t−2
r
e
t−2
y
u
t−2
π
u
t−2
r
u
t−2
x
t−2
+e
t
.
(B.2)
The structuralisation mechanism assumed in the identiﬁcation of structural shocks of the SVAR
model was the following:
1
We omit exogenous variables, as no restriction has been imposed thereon.
2
Exogenous variables have been omitted.
148
Appendix B
Global economy SVAR model
Global economy SVAR model
WORKING PAPER No. 83 149
B =
∗ 0 0 0 0 0 0
∗ ∗ 0 0 0 0 0
∗ ∗ ∗ 0 0 0 0
0 0 0 ∗ 0 0 0
0 0 0 ∗ ∗ 0 0
0 0 0 ∗ ∗ ∗ 0
0 0 0 0 0 0 ∗
. (B.3)
The estimated structural matrix B is the following:
B =
0.43 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.26 0.00 0.00 0.00 0.00 0.00
0.03 0.01 0.09 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.52 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.22 0.00 0.00
0.00 0.00 0.00 0.08 0.03 0.16 0.00
0.00 0.00 0.00 0.00 0.00 0.00 4.33
. (B.4)
149
Convergence to the steady state
N a t i o n a l B a n k o f P o l a n d 150
Appendix C
Convergence to the steady state
The ﬁgure shows the behaviour of observable variables, if the horizon of the example forecast
is extended to 100 quarters. As results from the graph, real categories (GDP, consumption,
investments, etc.) converge to the steady state relatively fast. Deviations of inﬂation (GDP and
investments deﬂator, CPI) from the steady state caused by impulses of the years 2008–2009
abate relatively slowly. Of course, the convergence of observable variables to the steady state is
guaranteed by the structure of the model.
Figure C.1. Example of ex ante forecast with long horizon
Inflation  GDP delator
08:4 12:4 16:4 20:4 24:4 28:4 32:4
1
2
3
4
5
Inflation  investment deflator
08:4 12:4 16:4 20:4 24:4 28:4 32:4
1
0
1
2
3
4
5
Inflation  CPI
08:4 12:4 16:4 20:4 24:4 28:4 32:4
1
2
3
4
5
GDP
08:4 12:4 16:4 20:4 24:4 28:4 32:4
1
2
3
4
5
6
7
Consumption
08:4 12:4 16:4 20:4 24:4 28:4 32:4
2
0
2
4
6
8
Investments
08:4 12:4 16:4 20:4 24:4 28:4 32:4
0
5
10
15
20
Export
08:4 12:4 16:4 20:4 24:4 28:4 32:4
20
15
10
5
0
5
10
15
Import
08:4 12:4 16:4 20:4 24:4 28:4 32:4
20
10
0
10
20
Real wages
08:4 12:4 16:4 20:4 24:4 28:4 32:4
1
2
3
4
5
6
7
8
Employment
08:4 12:4 16:4 20:4 24:4 28:4 32:4
1
0
1
2
Real exchange rate of dollar
08:4 12:4 16:4 20:4 24:4 28:4 32:4
20
15
10
5
0
5
10
Interest rate
08:4 12:4 16:4 20:4 24:4 28:4 32:4
3
4
5
6
7
150
Appendix C
Convergence to the steady state
Analysis of forecasts accuracy
WORKING PAPER No. 83 151
Appendix D
Analysis of forecasts accuracy
Figures D.1–D.5 present the RMSE and MPE errors for observable variables of the model,
depending on the forecast horizon. Tables D.1–D.2 present the mean values and standard
deviations of RMSE and MPE errors calculated for the horizons h = 1, 2, ..., 12. Table D.3
presents the relations of RMSE errors of a DSGE model and naive forecasts depending on the
forecast horizon. The data provided therein shall be analysed further herein in more detail
1
.
Figure D.1. RMSE and MPE for deﬂators: GDP (oPied), consumption (oPiec), investments
(oPiei) and CPI (oPiecpi)
0 5 10 15
0.8
1
1.2
1.4
1.6
0 5 10 15
0
5
10
15
20
oPied
0 5 10 15
0.8
1
1.2
1.4
1.6
0 5 10 15
−60
−40
−20
0
20
oPiec
0 5 10 15
1.4
1.6
1.8
2
2.2
0 5 10 15
−100
−50
0
50
100
oPiei
0 5 10 15
0.8
1
1.2
1.4
0 5 10 15
−40
−30
−20
−10
0
oPiecpi
RMSE — continuous line and left axis; MPE — dotted line and right axis.
The upper left panel in Figure D.1 shows that the root mean square error of the forecast of
domestic inﬂation is the largest, about 1.5 p.p. for the forecast horizon of 3 quarters, afterwards
it drops systematically to about 1 p.p. In relative terms the error ﬂuctuates depending on
the forecast horizon from several percent to nearly 20% for the 5 quarters’ horizon. Positive
1
The set of variables for which the accuracy of projections was tested covered — beside of all of the observable
variables of the model used in the estimation of parameters — also two additional variables: consumption deﬂator
(identiﬁed as oPiec) and expenditures on government consumption (oGov).
151
Appendix D
Analysis of forecasts accuracy
Analysis of forecasts accuracy
N a t i o n a l B a n k o f P o l a n d 152
MPE statistics for each horizon means that inﬂation of domestic products is systematically
underestimated — average MPE error amounts to about 7%, and its mean variability is above
4.5 p.p. Average RMSE error amounts to slightly over 1 p.p., while its mean variability is
nearly 0.2 p.p. The DSGE model wins with the naive forecast in each forecast horizon, while its
advantage grows along the extension of the horizon from about 25% for 1 quarter to 75% for 12
quarters. The mean advantage of the DSGE model over the naive forecast amounts to over 60%.
In the case of inﬂation of consumer goods prices (measured with consumption deﬂator), the
forecast errors are similar, except that the variable is systematically overestimated. The upper
right panel in Figure D.1 shows that root mean square error of the forecast of consumer goods
inﬂation is the largest — about 1.6 p.p. — for the forecast horizon of 4 quarters, afterwards it
drops systematically to about 1–1.2 p.p. In relative terms the error ﬂuctuates depending on the
forecast horizon from several percent to nearly −60% for the 3 quarters horizon. Negative MPE
statistics for most of the horizons means that inﬂation of consumer goods prices is systematically
overestimated — the average MPFE error amounts to −16%, however, its mean variability is
nearly 20 p.p. An average RMSE error amounts to 1.2 p.p., while its mean variability is nearly
0.25 p.p. The DSGE model wins with the naive forecast in each forecast horizon, except 1
quarter horizon when the naive model proves to be better by slightly over 1%. As in the case of
inﬂation of domestic products prices, the advantage of the DSGE model grows along the forecast
horizon — from about 17% for forecasts of 2 quarters to 65% for 12 quarters forecasts. The
mean advantage of the DSGE model over the naive forecast amounts to about 45%.
The bottom left panel of Figure D.1 shows that the root mean square error of the forecast of
inﬂation of investment goods prices reaches the highest value — about 2 p.p. — for 5 quarters
forecast and drops afterwards to about 1.4 p.p. for 8 quarters forecasts, then it grows again to
1.7 p.p. Before the 5th quarter the error oscillates around 1.4–1.8 p.p. MPE statistics show that
inﬂation of investment goods prices is, thus, underestimated for short horizons of the forecast,
overestimated within 3
rd
–8
th
quarters and underestimated from the 9
th
to 12
th
quarter. The
average MPE error amounts to −0.8 p.p., while its mean variability to as much as 35 p.p. The
average RMSE error is 1.7 p.p. and its mean variability is nearly 0.15 p.p. The DSGE model wins
with the naive forecast in each forecast horizon. Identically as in the case of inﬂation of the
prices of domestic products and consumer goods, the advantage of the DSGE model over the
naive forecast amounts to nearly 44%.
Errors of CPI inﬂation forecasts are similar as in the case of consumption deﬂator. The bottom
right panel of Figure D.1 shows that root mean square error of the forecast of CPI inﬂation is
the highest — nearly 1.4 p.p. — for 3 quarters horizon and afterwards it drops to about 0.8
p.p., then, starting from the 9
th
quarter, it grows to 1%. In relative terms the error ﬂuctuates
depending on the forecast horizon from −40% for 3 quarters horizon to −5% for 8 quarters.
MPE for each horizon is negative — CPI inﬂation is, thus, systematically overestimated, and the
average MPE error amounts to −20%, while its mean variability is about 11 p.p. The average
RMSE error amounts to 1 p.p. and its mean variability — to 0.22 p.p. The DSGE model wins
with the naive forecast in each forecast horizon, beside of the ﬁrst three periods when the naive
model proves to be better by 20–50%. The advantage of the DSGE model grows along the
forecast horizon — from 25% for 1 quarter to 65% for 12 quarters. The mean advantage of the
DSGE model over the naive forecast amounts to over 56%.
In RMSE errors of the forecasts of inﬂation of all the four types, a systematic pattern may be
perceived, namely the forecasts are the best in the short horizon (1–2 quarters) and longer
horizon (over 6–7 quarters), while approximately between the 2
nd
and the 6
th
quarter the
forecast errors are signiﬁcantly higher than for short and longer horizons. Additionally, the
advantage of the DSGE model over the naive forecast grows systematically along the forecast
horizon. In the case of other observable variables, the RMSE error has a tendency to grow with
the forecast horizon, which is presented in Figures D.2–D.5.
152
Analysis of forecasts accuracy
WORKING PAPER No. 83 153
Figure D.2. RMSE and MPE or the variables of GDP dynamics (oGdp), consumption (oCons),
investment expenditures (oInv) and export (oExp)
0 5 10 15
0.5
1
1.5
2
2.5
3
0 5 10 15
0
20
40
60
80
100
oGdp
0 5 10 15
1
2
3
4
5
0 5 10 15
−200
−150
−100
−50
0
oCons
0 5 10 15
3
4
5
6
7
0 5 10 15
−800
−600
−400
−200
0
oInv
0 5 10 15
5
10
15
0 5 10 15
−200
−150
−100
−50
0
oExp
RMSE — continuous line and left axis; MPE — dotted line and right axis.
Figure D.3. RMSE and MPE for import dynamics (oImp), real wages (oWage), employment
(oEmp) and real exchange rate of the dollar (oXu).
0 5 10 15
5
10
15
20
0 5 10 15
−600
−400
−200
0
200
oImp
0 5 10 15
0.5
1
1.5
2
2.5
3
0 5 10 15
−200
−100
0
100
200
oWage
0 5 10 15
0.5
1
1.5
2
0 5 10 15
−500
−400
−300
−200
−100
0
oEmp
0 5 10 15
5
10
15
0 5 10 15
−150
−100
−50
0
oXu
RMSE — continuous line and left axis; MPE — dotted line and right axis.
The upper left panel of Figure D.2 shows that root mean square error of the forecast of the
annual growth rate of the GDP grows systematically between the 1
st
and 7
th
quarter from 0.85
to 2.6 p.p., slightly decreasing after the 7
th
quarter, but still remains at the level of above 2 p.p.
In relative terms, the error grows from several percent in the ﬁrst quarters to 60–80% in the last
quarters. Production is, thus, systematically underestimated. The average MPE error amounts to
21%, while its mean variability is 25 p.p. The average RMSE error amounts to 1.9 p.p. and its
mean variability to — 0.65 p.p. The DSGE model wins with the naive forecast in each forecast
horizon. The advantage of the DSGE model grows also with extension of the forecast horizon
— from about 10% for 1 quarter forecast to nearly 50% for 12 quarters forecasts. The mean
advantage of the DSGE model over the naive forecast amounts to 40%.
153
Analysis of forecasts accuracy
N a t i o n a l B a n k o f P o l a n d 154
The upper right panel of Figure D.2 shows that the root mean square error of the forecast of the
annual growth rate of consumption systematically grows in the horizon of the forecast from 1.5
p.p. in the 1
st
quarter to 4.5 p.p. in the 12
th
quarter. The error grows along the whole forecast
horizon also in relative terms — from −24% to −170%. The growth rate of consumption is,
thus, systematically overestimated. The average MPE error amounts to −97%, while its mean
variability is 45 p.p. The average RMSE error amounts to 3.3 p.p. and its mean variability —
to 0.9 p.p. The errors of forecasts of consumption dynamics are, thus, nearly two times higher
than the errors of the forecasts of output growth. The DSGE model wins with the naive forecast
in each forecast horizon. The advantage of the DSGE model grows with the extension of the
forecast horizon — from nearly 20% for 1 quarter to 35% for 12 quarters. The mean advantage
of the DSGE model over the naive forecast is 45%.
In absolute terms, even larger forecast errors are generated by the model for annual growth
rates of investments, export and import, as well as the exchange rate. The bottom left panel
of Figure D.2 shows that the root mean square error of the forecast of annual growth rate of
investments grows in the forecast horizon from about 3 p.p in the 1
st
quarter to nearly 7 p.p. in
the 12
th
quarter. The growing trend is upset only in 5 the quarters’ horizon. In relative terms
the error grows systematically from −60% to nearly −800%. The growth rate of investments
is systematically overestimated. The average MPE error amounts to −100%, while its mean
variable — to 45 p.p. The average RMSE error amounts to 5 p.p. and its mean variability equals
to 0.9 p.p. The DSGE model wins with the naive forecast in each forecast horizon. The advantage
of the DSGE model grows also with the extension of the forecast horizon — from nearly 15%
for 1 quarter to 60% for 12 quarters. The mean advantage of the DSGE model over the naive
forecast amounts to 45%.
From the accuracy point of view, the errors of import and export forecasts are similar. The
bottom right panel of Figure D.2 shows that the root mean square error of the forecast of annual
growth rate of export grows in the forecast horizon from about 6 p.p. in the 1
st
quarter to 14 p.p.
in the 5
th
–6
th
quarter, and then drops to about 10 p.p. to rise again to about 13 p.p. In relative
terms the error grows systematically from minus several or so to −180%, and then drops to
about −90%. The growth rate of import is systematically overestimated. The average MPE error
amounts to −123%, while its mean variability equals 43 p.p. The average RMSE error amounts
to 11.5 p.p. and its mean variability — to 2.3 p.p. The DSGE model wins with the naive forecast
in each time horizon, except the horizon of 1 and 2 quarters when the naive forecast is on the
average by 36% and 6.5% better, respectively. The advantage of the DSGE model also grows
with the forecast horizon, except the 11
th
quarter — from 5% for 3 quarters to 35% for 12
quarters. The mean advantage of the DSGE model over the naive forecast amounts to 17%.
The upper left panel of Figure D.3 shows that the root mean square error of the forecast of
annual growth rate of import grows in the forecast horizon from about 5 p.p. in the 1
st
quarter
to 17 p.p. in the 12
th
quarter, except quarters 5 and 6. In relative terms the error reaches −480%
in the 2
nd
quarter and then drops systematically to (−60)–(−80)%. The import growth rate is
systematically overestimated. The average MPE error amounts to −136% and its mean variability
is 146 p.p. The average RMSE error amounts to 12.5 p.p. and its mean variability equals to about
3 p.p. The DSGE model wins with the naive forecast in each forecast horizon. The advantage
of the DSGE model grows — except quarters 11 and 12 — with the forecast horizon from 3%
for 3 quarters to 37% for 12 quarters. The mean advantage of the DSGE model over the naive
forecast amounts to 32%.
The upper right panel of Figure D.3 shows that the root mean square error of the forecast of
annual growth rate of real wages grows with the forecast horizon from about 0.8 p.p. in the
1
st
quarter to 2.5 p.p. in the 12
th
quarter. In relative terms the error oscillates between −120%
in the 2
nd
quarter and −170% in the 12
th
quarter. The import growth rate is overestimated for
the horizon of 1 and 2 quarters, while it is underestimated for the other horizons. The average
MPE error amounts to −106% and its mean variability equals 94 p.p. The average RMSE error
amounts to 2.2 p.p. and its mean variability is 0.5 p.p. The DSGE model wins with the naive
154
Analysis of forecasts accuracy
WORKING PAPER No. 83 155
forecast in each forecast horizon. The advantage of the DSGE model grows with the forecast
horizon from 15% for 1 quarter to 38% for 12 quarters. The mean advantage of the DSGE model
over the naive forecast amounts to 34%.
The bottom left panel of Figure D.3 shows that the root means square error of the forecast of
annual growth rate of employment grows along the forecast horizon from about 0.5 p.p. in the
1
st
to 1.5 p.p. in the 12
th
quarter, while the trend is upset only in the horizon of 7 quarters. In
relative terms the error reaches the value of −540% in the 6
th
quarter and falls to (−60)–(−90)%
in the 12
th
quarter. The growth rate of employment is moderately overestimated. The average
MPE error amounts to −185% an its mean variability is 125 p.p. The average RMSE error
amounts to 1.15 p.p. and its mean variability — to about 0.3 p.p. The DSGE model wins with
the naive forecast in each forecast horizon. The advantage of the DSGE model grows with the
extension of the forecast horizon — from 13% for 1 quarter to about 43% for 10–12 quarters.
The mean advantage of the DSGE model over the naive forecast amounts to 34%.
Figure D.4. RMSE and MPE for the interest rate (oRd), GDP dynamics in the euro area (oYe)
and in the USA (oYu), plus inﬂation in the euro area (oPiee)
0 5 10 15
0.4
0.6
0.8
1
1.2
1.4
0 5 10 15
2
4
6
8
10
oRd
0 5 10 15
0
1
2
3
4
5
0 5 10 15
−50
−40
−30
−20
−10
0
oYe
0 5 10 15
0.7
0.8
0.9
1
1.1
0 5 10 15
−400
−200
0
200
400
oYu
0 5 10 15
0
0.2
0.4
0.6
0.8
1
0 5 10 15
−20
0
20
40
60
oPiee
RMSE — continuous line and left axis; MPE — dotted line and right axis.
The upper left panel of Figure D.4 shows that the root mean square error of the forecast of
annual nominal interest rate grows together with the forecast horizon, both in absolute and
relative terms. The RMSE error grows from 0.5 p.p. in the horizon of 1 quarter to 1–1.5 p.p. for
quarters 9–12, while the MPE error grows from 3% to 8.5 −9%. The nominal interest rate is
moderately underestimated — the average MPE error amounts to 7% and its mean variability
is about 1 p.p. The average RMSE error amounts to 1 p.p. and its mean variability — to about
0.2 p.p. The DSGE model wins with the naive forecast in each forecast horizon. The advantage
of the DSGE model ranges from 5% in the horizon of 1 quarter, through 40% in the horizon
of 7 quarters, to 17% for 12 quarters. The mean advantage of the DSGE model over the naive
forecast amounts to 28%.
The other three panels of Figure D.4 and the whole Figure D.5 present the errors of forecasts of
foreign variables whose dynamics results from the SVAR model which is exogenous in reference
to the DSGE model and, therefore, we shall not discuss them in detail. For all of the foreign
variables the growing trend of the root mean square error is visible along the forecast horizon.
Production in the euro area is moderately overestimated and production in the dollar area is
underestimated for the horizons of 1 to 9 quarters and afterwards it is overestimated. Inﬂation
155
Analysis of forecasts accuracy
N a t i o n a l B a n k o f P o l a n d 156
Figure D.5. RMSE and MPE for inﬂation in the USA (oPieu), interest rate in the euro area (oRe)
and in the USA (oRu), plus the dynamics of the nominal cross rate (oDsx)
0 5 10 15
1
1.2
1.4
1.6
1.8
0 5 10 15
100
200
300
400
500
oPieu
0 5 10 15
0
0.5
1
1.5
2
0 5 10 15
0
20
40
60
oRe
0 5 10 15
0.5
1
1.5
2
2.5
3
0 5 10 15
0
100
200
300
400
oRu
0 5 10 15
3
4
5
6
0 5 10 15
−120
−100
−80
−60
−40
oDsx
RMSE — continuous line and left axis; MPE — dotted line and right axis.
and nominal interest rates in both areas are usually underestimated, while the real exchange
rate of EUR/USD is overestimated, on the average. In the case of all of the foreign variables the
VAR model generates better forecasts than the naive model for all of the tested forecast horizons,
except inﬂation in the euro area in 11 and 12 quarters forecast horizon and inﬂation in the
dollar area in the horizon of 6 quarters.
Table D.1. Mean and standard deviation of the RMSE statistics
Variable oPied oPiec oPiei oPiecpi oGdp oCons oInv oExp oImp oWage
Mean 1.08 1.20 1.68 1.00 1.90 3.28 5.11 11.51 12.56 2.19
Standard deviation 0.18 0.24 0.16 0.22 0.66 0.91 0.92 2.31 3.08 0.51
Variable oEmp oXu oRd oYe oYu oPiee oPieu oRe oRu oDsx
Mean 1.15 12.45 0.98 2.82 0.92 0.53 1.37 1.26 1.54 4.65
Standard deviation 0.28 2.82 0.17 1.04 0.12 0.18 0.24 0.33 0.67 0.55
Table D.2. Mean and standard deviation of the MPE statistics
Variable oPied oPiec oPiei oPiecpi oGdp oCons oInv oExp oImp oWage
Mean 7.17 16.48 0.83 19.26 21.38 97.27 332.39 123.74 136.14 106.96
Standard deviation 4.64 19.58 35.39 11.17 25.67 45.04 197.47 42.96 146.55 94.09
Variable oEmp oXu oRd oYe oYu oPiee oPieu oRe oRu oDsx
Mean 185.13 81.69 7.55 31.92 168.99 24.96 260.11 33.56 143.20 94.30
Standard deviation 124.96 39.77 1.61 13.09 240.41 15.40 103.26 16.50 89.16 11.58
156
Analysis of forecasts accuracy
WORKING PAPER No. 83 157
Table D.3. Comparison of forecasts of the SOE
PL−2009
model with the naive forecast
Horizon oPied oPiec oPiei oPiecpi oGdp oCons oInv oExp oImp oWage
1 0.76 1.01 0.81 1.53 0.91 0.74 0.85 1.37 0.97 0.84
2 0.62 0.84 0.61 1.36 0.69 0.57 0.69 1.06 0.90 0.81
3 0.56 0.81 0.42 1.19 0.44 0.46 0.59 0.95 0.86 0.69
4 0.40 0.68 0.35 0.86 0.38 0.54 0.47 0.92 0.73 0.64
5 0.33 0.53 0.43 0.62 0.54 0.56 0.40 0.85 0.63 0.60
6 0.30 0.50 0.41 0.54 0.64 0.53 0.35 0.75 0.55 0.63
7 0.26 0.50 0.41 0.55 0.66 0.59 0.35 0.76 0.59 0.62
8 0.25 0.50 0.32 0.50 0.73 0.52 0.32 0.69 0.58 0.62
9 0.28 0.41 0.39 0.37 0.63 0.48 0.35 0.65 0.57 0.56
10 0.25 0.33 0.36 0.31 0.53 0.51 0.36 0.67 0.58 0.60
11 0.31 0.30 0.40 0.30 0.51 0.54 0.36 0.75 0.63 0.64
12 0.25 0.35 0.34 0.37 0.52 0.66 0.40 0.66 0.63 0.62
Horizon oEmp oXu oRd oYe oYu oPiee oPieu oRe oRu oDsx
1 0.87 0.85 0.94 0.83 0.97 0.85 0.82 0.71 0.77 0.79
2 0.78 0.85 0.82 0.78 0.83 0.78 0.82 0.78 0.72 0.72
3 0.79 0.88 0.63 1.05 0.88 0.75 0.94 0.75 0.66 0.60
4 0.73 0.88 0.71 0.98 1.01 0.71 0.79 0.78 0.66 0.67
5 0.69 0.93 0.65 0.88 1.00 0.56 0.98 0.66 0.51 0.81
6 0.68 0.98 0.65 0.81 0.86 0.64 1.09 0.64 0.48 0.93
7 0.56 1.03 0.58 0.84 0.91 0.80 0.83 0.66 0.52 0.90
8 0.56 1.08 0.63 0.81 1.12 0.81 0.94 0.65 0.58 0.83
9 0.55 1.12 0.71 0.80 1.13 0.76 0.90 0.68 0.66 0.76
10 0.57 1.16 0.73 0.78 1.00 0.73 0.81 0.67 0.71 0.73
11 0.56 1.19 0.80 0.78 0.90 1.00 0.97 0.67 0.79 0.88
12 0.58 1.20 0.84 0.80 0.89 1.24 0.85 0.73 0.89 1.25
157
Bibliography
N a t i o n a l B a n k o f P o l a n d 158
Bibliography
ADOLFSON, M., LASÉEN, S., LINDÉ, J., AND SVENSSON, L. E. 2009. Monetary policy tradeoffs in an estimated
openeconomy DSGE model. Sveriges Riksbank, Federal Reserve Board, Manuscript.
ADOLFSON, M., LASEÉN, S., LINDÉ, J., AND VILLANI, M. 2005a. Are constant interest rate forecasts modest
interventions. Evidence from an estimated open economy DSGE model of the Euro Area. Sveriges
Riksbank, Manuscript.
ADOLFSON, M., LASÉEN, S., LINDÉ, J., AND VILLANI, M. 2005b. Bayesian estimation of an open economy DSGE
model with incomplete passthrough. Sveriges Riksbank Working Paper Series 179.
ADOLFSON, M., LASÉEN, S., LINDÉ, J., AND VILLANI, M. 2007a. Evaluating an estimated New Keynesian small
open economy model. Sveriges Riksbank Working Paper Series 203.
ADOLFSON, M., LASÉEN, S., LINDÉ, J., AND VILLANI, M. 2007b. RAMSES a new general equilibrium model for
monetary policy analysis. Riksbank Economic Review 2.
ADOLFSON, M., LINDÉ, J., AND VILLANI, M. 2005c. Forecasting performance of an open economy dynamic
stochastic general equilibrium model. Sveriges Riksbank Working Paper Series 190.
ALTIG, D., CHRISTIANO, L., EICHENBAUM, M., AND LINDÉ, J. 2004a. Firmspeciﬁc capital, nominal rigidities
and the business cycle. Sveriges Riksbank Working Paper Series 176.
ALTIG, D., CHRISTIANO, L., EICHENBAUM, M., AND LINDÉ, J. 2004b. Firmspeciﬁc capital, nominal rigidities and
the business cycle. Manuscript. URL http://www.faculty.econ.northwestern.edu/faculty/christiano/research.
ALTIG, D., CHRISTIANO, L., EICHENBAUM, M., AND LINDÉ, J. 2005. Technical appendix to ﬁrmspeciﬁc capital,
nominal rigidities and the business cycle. Manuscript. URL http://www.faculty.econ.northwestern.edu/
faculty/christiano/research.
ALVAREZLOIS, P., HARRISON, R., PISCITELLY, L., AND SCOTT, A. 2005. Taking DSGE models to the policy
environment. Bank of England, Manuscript. URL http://www12.georgetown.edu/sfs/cges/BOE.pdf.
AN, S. AND SCHORFHEIDE, F. 2005. Bayesian analysis of DSGE models. CEPR Discussion Paper Series 5207.
ANDRLE, M., HLÉDIK, T., KAME
´
NIK, O., AND VLCEK, J. 2009. Implementing the new structural model of the
Czech National Bank. CNB Working Paper Series 2. URL http://www.cnb.cz/.
BENESZ, A., HLEDIK, T., AND VAVRA, D. 2005. An economy in transition and DSGE: What the Czech
National Bank’s new projection model needs. Czech National Bank, Monetary and Statistics Department,
158
Bibliography
Bibliography
WORKING PAPER No. 83 159
Manuscript.
BIRKHOLC, A. 2002. Funkcje wielu zmiennych. PWN.
BOIVIN, J. AND GIANNONI, M. 2005. DSGE models in a datarich environment. Working Paper 12772, NBER.
BROOKS, S. P. 1998. Markov Chain Monte Carlo methods and its applications. The American Statistician
47:69–100.
BROOKS, S. P. AND GELMAN, A. 1996. General methods for monitoring convergence of iterative simulations.
Manuscript. URL http://www.statslab.cam.ac.uk/~steve/mypapers/brog96.ps.
BURRIEL, P., FERNÁNDEZVILLAVERDE, J., AND RUBIORAMIREZ, J. F. 2009. MEDEA — A DSGE Model for the
Spanish Economy. Manuscript.
CALVO, G. 1983. Staggered contracts in a utilitymaximising framework. Journal of Monetary Economics
12:383–398.
CANOVA, F. 2007. Methods for Applied Macroeconomic Research. Princeton University Press.
CANOVA, F. AND SALA, L. 2005. Back to square one: Identiﬁcation issues in DSGE models. Working paper no.
927, Universitat Pompeu Fabra. Manuscript. URL http://www.econ.upf.edu/eng/research/onepaper.php?id=
927.
CHIB, S. AND GREENBERG, E. 1995. Understanding the MetropolisHastings algorithm. The American
Statistician 49:327–335.
CHRISTIANO, L., EICHENBAUM, M., AND EVANS, C. 1998. Monetary policy shocks: What have we learned and
to what end? Working Paper w6400, NBER.
CHRISTIANO, L., EICHENBAUM, M., AND EVANS, C. 2001. Nominal rigidities and the dynamic effects of a shock
to monetary policy. Working Paper 8403, NBER.
CHRISTIANO, L., EICHENBAUM, M., AND EVANS, C. 2003. Nominal rigidities and the dynamic effects of a shock
to monetary policy. Manuscript.
CHRISTIANO, L., EICHENBAUM, M., AND EVANS, C. 2005. Nominal rigidities and the dynamic effects of a shock
to monetary policy. The Journal of Political Economy 113:1–45.
CHRISTIANO, L., MOTTO, R., AND ROSTAGNO, M. 2007a. Financial factors in business cycles. Manuscript. URL
http://faculty.wcas.northwestern.edu/~lchrist/course.
CHRISTIANO, L., MOTTO, R., AND ROSTAGNO, M. 2007b. Technical appendix for Financial Factors in Business
Cycles. Manuscript. URL http://faculty.wcas.northwestern.edu/~lchrist/course.
CHRISTIANO, L., TRABANDT, M., AND WALENTIN, K. 2007c. Introducing ﬁnancial frictions and unemployment
into a small open economy model. Manuscript. URL http://faculty.wcas.northwestern.edu/~lchrist/course/
manuscript.pdf.
CHRISTIANO, L., TRABANDT, M., AND WALENTIN, K. 2007d. Introducing ﬁnancial frictions and unemployment
into a small open economy model. Sveriges Riksbank Working Paper Series 214.
CHRISTOFFEL, K., COENEN, G., AND WARNE, A. 2007a. The New AreaWide Model of the euro area: a
microfounded openeconomy model for forecasting and policy analysis. ECB Working Papers Series 944.
CHRISTOFFEL, K., COENEN, G., AND WARNE, A. 2007b. The New AreaWide Model of the Euro area: Speciﬁca
tion and ﬁrst estimation results. Manuscript.
COENEN, G., MCADAM, P., AND STRAUB, R. 2006. Tax reform and labourmarket performance in the Euro
Area: A simulationbased analysis using New AreaWide Model. EBC, Manuscript.
DIXIT, A. K. AND STIGLITZ, J. E. 1977. Monopolistic competition and optimum product diversity. The
American Economic Review 67:287–308.
159
Bibliography
N a t i o n a l B a n k o f P o l a n d 160
ENGEL, C. 1996. The forward discount anomaly and the risk premium. A survey of recent evidence. Journal
of Empirical Finance 3:123–192.
ERCEG, C. J., HENDERSON, D. W., AND LEVIN, A. T. 2000. Optimal monetary policy with staggered wage and
price contracts. Journal of Monetary Economics 46:281–313.
FAGIOLO, G. AND ROVENTINI, A. 2008. On the scientiﬁc status of economic policy: A tale of alternative
paradigms. Working Paper 2008/03, LEM.
FERNÁNDEZVILLAVERDE, J. 2009. The econometrics of DSGE models. Working Paper 14677, NBER.
FRIEDMAN, M. AND SCHWARTZ, A. J. 1963. Money and business cycles. The Rewiev of Economics and Statistics
45.
GALI, J. 2008. Monetary policy, inﬂation and the Busness Cycle. An Introduction to the New Keynesian
Framework. Princeton University Press.
GEWEKE, J. 1999. Using simulation methods for Bayesian econometric models: Inference, Econometric
Reviews 18:1–126.
GEWEKE, J. 2005. Contemporary Bayesian Econometrics and Statistics. John Wiley & Sons, Inc.
GOODFRIEND, M. AND KING, R. G. 1997. The new neoclassical synthesis and the role of monetary policy.
12th Annual NBER Macroeconomics Conference.
GRABEK, G. AND KŁOS, B. 2009. Wybrane skutki przyst ˛ apienia małej otwartej gospodarki do unii walutowej.
Optyka modeli DSGE SOEPL. In Raport na temat pełnego uczestnictwa Rzeczypospolotej Polskiej w
trzecim etapie Unii Gospodarczej i Walutowej. Aneks. NBP. Cz˛e´s´c 17. URL http://www.nbp.pl.
GRABEK, G., KŁOS, B., AND UTZIGLENARCZYK, G. 2007. SOEPL — Model DSGE małej otwartej gospodarki
estymowany na danych polskich. Metodologia, specyﬁkacja, wyniki estymacji i pierwsze zastosowania.
Materiały i Studia NBP 217. URL http://www.nbp.pl.
GRABEK, G. AND UTZIGLENARCZYK, G. 2009. Gospodarka polska w latach 1997–2006 widziana przez
pryzmat modelu DSGE. Bank i Kredyt 2.
GUERRONQUINTARA, P. A. 2009. What you match does matter: The effects of data on DSGE estimation.
Journal of Applied Econometrics DOI:10.1002/jae.1106. URL http://www.interscience.wiley.com.
HAMILTON, J. D. 1994. Time Series Analysis. Princeton University Press.
HASTINGS, W. 1970. Monte Carlo sampling methods using Markov Chains and their applications. Biometrica
57.
HENDRY, D. F. 1976. The structure of simultaneous equations estimators. Journal of Econometrics 4.
HENDRY, D. F. 1995. Dynamic Econometrics. Oxford University Press.
IRELAND, P. N. 2004. A method for taking models to the data. Journal of Economic Dynamics & Control
28:1205–1226.
JAKAB, Z. AND VILÁGI, B. 2008. An estimated DSGE model of the Hungarian economy. MNB Working Papers
9.
JUSTINIANO, A. AND PRIMICERI, G. E. 2006. The time varying volatility of macroeconomic ﬂuctuations.
Working Paper 12022, National Bureau of Economic Research. URL http://www.nber.org/papers/w12022.
KLEIN, L. R. AND GOLDBERGER, A. 1955. An Econometric Model of the United States, 19291952. North
Holland Publishing Company.
KŁOS, B. 2004. Niepewno´s´c modelu w polityce makroekonomicznej. Zasada odporno´sci. Bank i Kredyt
10:25–40.
KŁOS, B. 2008. SOEPL — Modele DSGE małej otwartej gospodarki estymowane na danych polskich.
160
Bibliography
WORKING PAPER No. 83 161
Dokumentacja techniczna 2008. Instytut Ekonomiczny, Narodowy Bank Polski, mimeo.
KOOP, G. 2003. Bayesian Econometrics. John Wiley & Sons, Ltd.
KYDLAND, F. AND PRESCOTT, E. 1982. Time to build and aggregate ﬂuctuations. Econometrica 50.
LANE, P. 1999. The new open economy macroeconomics: A survey. Journal of International Economics
54:235–266.
LASÉEN, S. AND SVENSSON, L. E. 2009. Anticipated alternative instrumentrate paths in policy simulations.
Manuscript.
LEDUC, S. AND SILL, K. 2001. A quantitative analysis of oilprice shocks, systematic monetary policy,
and economic downturns. Working Paper Research Department. no. 0109, Federal Reserve Bank of
Philadelphia.
LEEPER, E. M. AND ZHA, T. 2003. Modest policy interventions. Journal of Monetary Economics 50:1673–1700.
LOU, Z., WAGGONER, D. F., AND ZHA, T. 2007. Has the Federal Reserve’s inﬂation target changed? Manuscript.
LUCAS, R. E. 1976. Econometric policy evaluation. a critique. In K. B. i A. Meltzer (ed.), The Phillips Curve
and Labour Markets. North Holland.
LÜTKEPOHL, H. 2008. New Introduction to Multiple Time Series Analysis. Springer.
ORPHANIDES, A. 2007. Using DSGE models for forecasting and policy analysis: Where do we go from here?
The limitations. BIS Workshop, Manuscript.
OSIEWALSKI, J. 2001. Ekonometria bayesowska w zastosowaniach. Wydawnictwo Akademii Ekonomiczne
w Krakowie.
PESARAN, H. M. AND SMITH, R. J. 1995. The role of theory in econometrics. Journal of Econometrics 67.
QIN, D. 1993. The formation of econometrics. An historical perspective. Oxford University Press.
SCHMITTGROHÉ, S. AND URIBE, M. 2003. Closing small open economy models. Journal of International
Economics 61:163–185.
SCHMITTGROHÉ, S. AND URIBE, M. 2008. What’s News in business cycles. NBER Working Paper 14215.
SCHORFHEIDE, F. 2000. Loss functionbased evaluation of DSGE models. Journal of Applied Econometrics
15:645–670.
SIMS, C. A. 1980. Macroeconomics and reality. Econometrica 48.
SIMS, C. A. AND ZHA, T. 2005. Were there regime switches in us monetary policy? Manuscript.
SMETS, F. AND WOUTERS, R. 2002. An estimated stochastic dynamic general equilibrium model of the Euro
Area. ECB Working Papers Series 171.
SMETS, F. AND WOUTERS, R. 2004. Forecasting with a Bayesian DSGE model: An application to the Euro
Area. ECB Working Papers Series 389.
SMETS, F. AND WOUTERS, R. 2007. Shocks and Frictions in US Business Cycles: A Bayesian Approach. ECB
Working Papers Series 722.
SPANOS, A. 1990. The simultaneousequations model revisited: Statistical adequacy and identiﬁcation.
Journal of Econometrics 44.
WAGGONER, D. F. AND ZHA, T. 1999. Conditional forecasts in dynamic multivariate models. The Review of
Economics and Statistics 81:639–651.
WARNE, A. 2009. YADA manual — Computational details. EBC, Manuscript. URL http://www.texlips.net/
yada/index.html.
WOODFORD, M. 2003. Interest and Prices. Foundations of a Theory of Monetary Policy. Princeton University
161
Bibliography
N a t i o n a l B a n k o f P o l a n d 162
Press.
˙
ZAK, S. AND CHONG, E. K. P. 2008. Introduction to Optimization. John Wiley and Sons, Inc.
Note: The Internet addresses inform about the place from which the documents have been downloaded.
We cannot guarantee that the aforementioned pages still exist and/or still provide the cited materials.
162
The paper presents the personal opinions of the authors and does not necessarily reﬂect the ofﬁcial position of the institution with the authors cooperate (have cooperated) or to which they are (or have been) afﬁliated. An earlier version , B paper KŁOS GRZEGORZ in polish PL—2009 — An SOE — Model GrzeGorz GGRABEKbohdan kłos, been published KOLOCH, under the title:Estimated Dynamic DSGE GRZEGORZ rabekof, theOHDAN has G,rzeGorz koloch, SOE SOEPL−2009 — An Estimated Dynamic małej otwartej gospodarki estymowany na danych polskich. Specyﬁkacja, oceny parametrów, Applied Stochastic General Equilibrium Model for Policy Analysis and Forecasting, Bureau ofzastosowania, Stochastic General Equilibrium Model for Policy Analysis and Forecasting , Bureau of Applied Materiały Economic ,Institute, National Bank Poland, Warsaw, 2010 Research, iEconomic Institute, National Bank ofof Poland, Warsaw, 2010 Research, Studia NBP 2010. (Document compilation: 860.14.2.2011) (Document compilation: 860.14.2.2011)
PL−2009
Kontakt: B Bohdan.Klos@mail.nbp.pl B Grzegorz.Grabek@mail.nbp.pl B Grzegorz.Koloch@mail.nbp.pl
T (0 48 22) 653 15 87 T (0 48 22) 585 41 08 T (0 48 22) 653 21 79
The paper presents the personal opinions of the authors and does not necessarily reﬂect the ofﬁcial position of the institution with the authors cooperate (have cooperated) or to which they are (or have been) afﬁliated. An earlier version of the paper has been published in polish under the title: SOEPL−2009 — Model DSGE małej otwartej gospodarki estymowany na danych polskich. Specyﬁkacja, oceny parametrów, zastosowania, Materiały i Studia NBP 2010. ,
A Typesetting LTEX (TEX, format: LaTeX2e, ver. 2009/09/24), pdfTEX, ver. 1.40.11 (MikTEX 2.9) A LTEX 2 , pdfTEX, BIBTEX (etc.) implementations, fonts, packages of macros and the utilities available in the MikTEX collection have been used. Detailed information about the licence conditions, software copyrights, legal protection of the stored collection, legally protected trademarks (etc.) are provided on the Internet Web pages of the MikTEX http://www.miktex.org and on the site of TEX system at http://www.ctan.org.
Design: Oliwka s.c. Layout and print: NBP Printshop Published by: National Bank of Poland A Typesetting LTEX (TEX, format: LaTeX2e, ver. 2009/09/24), pdfTEX, ver. 1.40.11 (MikTEX 2.9) Education and Publishing Department A 00919 , pdfTEX, BIBTEX (etc.) implementations, fonts, packages of macros and the utilities available in the LTEX 2 Warszawa, 11/21 Świętokrzyska Street MikTEX collection have 35, fax +48 22 653 13 21 phone: +48 22 653 23 been used. Detailed information about the licence conditions, software copyrights, legal protection of the stored collection, legally protected trademarks (etc.) are provided on the Internet © Copyright by the National Bank of Poland, 2011 and on the site of T X system at http://www.ctan.org. Web pages of the MikTEX http://www.miktex.org E http://www.nbp.pl
2
N a t i o n a l
B a n k
o f
P o l a n d
Contents
Contents
Contents
Introduction
6
I Genesis and anatomy of dynamic stochastic general equilibrium models 10
1 Genesis and anatomy of dynamic stochastic general equilibrium models 1.1 Methods of the Cowles Commission . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Problems with identiﬁcation . . . . . . . . . . . . . . . . . . . . . . . 1.2 LSE and VAR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 LSE methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 VAR methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Methodology of modern macroeconomics . . . . . . . . . . . . . . . . . . . 1.3.1 Real Business Cycle theory . . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Frictionless economy . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.3 Representative household . . . . . . . . . . . . . . . . . . . . . . . . 1.3.4 Representative ﬁrm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.5 General equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.6 Consequences for monetary policy . . . . . . . . . . . . . . . . . . . 1.4 New neoclassical synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Standard newKeynesian model . . . . . . . . . . . . . . . . . . . . . 1.4.2 Representative household . . . . . . . . . . . . . . . . . . . . . . . . 1.4.3 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.4 General equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.5 Consequences for monetary policy . . . . . . . . . . . . . . . . . . . 2 DSGE model — anatomy 2.1 Values of parameters . . . . . . . . . . . . 2.1.1 Calibration . . . . . . . . . . . . . 2.1.2 Maximum likelihood estimation . 2.1.3 Bayesian estimation . . . . . . . . 2.2 Kalman ﬁlter . . . . . . . . . . . . . . . . . 2.3 MetropolisHastings algorithm . . . . . . 2.4 Model selection . . . . . . . . . . . . . . . 2.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 . . 11 . 12 . 13 . 14 . 14 . 16 . 18 . 20 . 20 . 22 . 22 . 23 . 24 . 25 . 25 . 26 . 27 . 28 31 . 35 . 36 . 36 . 38 . 39 . . 41 . 42 . 43
3
WORKING PAPER No. 83
3
6. . .3. SVAR models 5. 4. . .6. .3 Basic features of SOEPL−2009 model . .2 Observable variables. . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Monetary balance . . . . . .2 Income and expenditures of households . . . . . . . . . . . . .6 Macroeconomic balance conditions . . . . . .1 SOEEuro model — prototype of SOEPL family models . . . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . . . . . . . . .1 Growth . . . .5. . . . . . . . . . . Unconditional forecasts . . . . . . . . . .2 Foreign economy . 4. . . . .2 Prior distributions and results of estimation . . . . . . . 6. . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . 4. . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . . .5. . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . . . 43 43 44 45 II Speciﬁcation of DSGE SOEPL−2009 model 47 48 48 50 51 56 56 57 57 58 59 61 61 63 68 69 70 71 72 73 74 74 75 77 3 SOEPL−2009 — general outline 3. . . . . . . . . . . . . optimisation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Structural shocks identiﬁcation Impulse response analysis . . . . . . . .4 Exporters .3. . . .2 2. . 4. . . . . . . . .5. . . .1 Central bank . . 6. . . . . . . . . . .2 Family of SOEPL models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 SVAR models . . . . . . . . .3. . . . . . . . . . . . . . . . . .2 World’s economy SVAR . equilibrium conditions. . . .3 Nominal rigidities . . . . . . .5 Balance of payment . . . . . . . . . . . . . . . . . . . . . . . . . 6. steady state 6. . . . . . .2 Government . . . . . . . . . . . . . . . . . 3.3. . . .6 The aggregate resource constraint . . . . . . . . . . .3 Importers . . 4. . . . . . . . . . . . . . . . . . . .3 Producers . . . . 4. . . . . . . . . . . . . . . . . . . . . .1 2. . . . . . . . . . . . . . . . . . . . .3 2. . . . . macroeconomic balance of the model 4. . . . . . . . . . . . . . . . . . . . . . . . .1 Fiscal SVAR . . . . . .3. . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . 4. . .4 Households . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . .3 State budget . . . . 5. . . . .5. . .1 Forms of model . . . .5 Behaviour of other agents . . . . . . . . . . . . 79 79 80 82 82 83 85 86 87 90 90 92 93 6 Assessment of parameters — calibration. .Contents 2. 4. . . . . .6. . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . III Results of estimation and characteristic features of the DSGE SOEPL−2009 model 78 5 Forms of model. . . 6. . . . . . data . . 4 4 N a t i o n a l B a n k o f P o l a n d . . .1 Calibration of parameters . . . . . . . . . . .2 Domestic intermediate goods ﬁrms . . . . . . . . . . . . . . . . . . . . . . . . Variance decomposition . . .1 Proﬁts in economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Structural changes . . . . . . . . . . . . . . . .1 Aggregators . . . . . . . . . . 4. . . . . . SOEPL−2009 version . .3 Assessment of parameters — conclusions . 4 Decisionmaking problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . data. . . . . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Steady state . . . .6. 4. . . . . 5. . . . . . . . . . . . . . 4. . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 132 134 135 135 136 139 144 A List of equations. . impulse response functions and estimation of disturbances 95 7. . . . . . . . . . . . . . . . . 121 122 126 127 128 129 Final comments IV Appendix . . . . . . . . . . . . . . 8. . . . 95 7. List of model variables . . . . . . . .3 Smoothing — estimation of structural disturbances . . . . . . . . . . . 96 7. . . . . . . . . . . . . . .1. . . . .1. . . . . . . . . . . . . . . . . . . . . 148 Results of estimation of the SVAR model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Multidimensional measures of forecasting accuracy 8. . . . . . . . . . . . . . . . .2 Regime II . . . . . . . . . . . .3. .1 Onedimensional measures of forecasting quality . . . . . . . . . . . . . .1 Variance decompositions . . . . . . . . . . . . . .3 Expost forecasting accuracy of the SOEPL−2009 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 8 Historical decompositions and forecasts 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 5 . . . . . . . . . . . . . . . . . . .Contents 7 Variance decompositions. . . . . . . . . . . . . . . . . . . . . . . . 8. . 103 7. . . . . . . .3. . .2 Forecasting technique . . . . . . . . . . . . . . . . . . . 121 . . . . . . . . . . . . . 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . List of equations of the structural form of the model Steady state solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Impulse response functions . . . . . . . .3. .1 Regime I . . . . . . . . . . . . . . . . . . . . . . . . 8. . . . . . . . . . . . . . . . . . .3 Rolling forecasts . . . . . . . . list of variables Forms of the model . . . . . . . 97 7. . . 148 C Convergence to the steady state D Analysis of forecasts accuracy 150 151 Bibliography Bibliogrphy 158 5 WORKING PAPER No. . B Global economy SVAR model 148 SVAR model identiﬁcation . . . .1 Historical decompositions . . . . . . .
• micro. The very potential of the models of this class seems to be the most important reason for the interest of central banks in that area1 . Merger of the three trends has brought about a class of models — DSGE models — with high analytical and developmental potential. the role of nominal and real rigidities.Introduction Introduction Introduction The paper documents the effects of work on the dynamic stochastic general equilibrium (DSGE) SOEPL model that has been carried out in the recent years at the National Bank of Poland. shift from classical techniques controlled introduction of experts’ knowledge. as well as anticipating and optimising behaviours of agents in an uncertain environment. In 2009. called SOEPL−2009 which in 2010 is to be used to obtain routine midterm forecasts of the inﬂation processes and the economic trends. research that may be directly translated into the practice of monetary policy.and macroeconomic theories (monetary policy issues. strongly motivated with microeconomics). 1 Attention is being drawn to the fact that in the general case the DSGE models do not have to be based on the new Keynesian perspective of economy and do not have to be estimated with the use of Bayesian techniques. In the recent years many researchers have engaged in the work over a class of estimated macroeconomic models (of the business cycle) integrating the effects of at least three important lines of economic and econometric research: • methods of macroeconomic modelling (gradual departure from the traditional structural models towards models resistant to Lucas’ and Sims’ critique. supporting and supplementing the traditional structural macroeconometric model and experts’ forecasts applied so far. improvement of projections accuracy). with emphasis on the consequence of imperfect competition. with a strong shift of point of view towards general equilibrium). to Bayesian techniques with Bayesianspeciﬁc risk quantiﬁcation as well as systematic and 6 6 N a t i o n a l B a n k o f P o l a n d . a team consisting of the authors of this paper developed a new version of the model. • estimation techniques (reduction in parameters calibration. initially at the Bureau of Macroeconomic Research and lately at the Bureau of Applied Research of the Economic Institute.
e. however. 83 7 . at least partially compliant with that which may be achieved with VAR and SVAR models. which eliminates the consequences of the disturbances.e. 2005). econometric methods and the theory of economics. DSGE models give. The model whose details we shall present further herein derives from the structure developed at Riksbank — DSGE model for the euro area4 see Adolfson et al. however. Generally. e. see e. the best we can do is to try to use our model. the user). as well as the analysis of the current policy less inﬂuenced by its rationalisation shall conﬁrm the legitimacy of building and applying models. develop new procedures and thoroughly verify the results. whose forecasting applications are enhanced with experts’ knowledge. veriﬁcation of the directions of economic research and methods of the research. In our opinion. as well as Smets and Wouters (2004). which is likely to be held. DSGE models present a different image of economic processes than classical macroeconometric models — they capture the world from the perspective of structural disturbances. build different versions of DSGE model (a family of SOEPL models) and develop our own procedures of the model application. The issue of applications using the strong sides of the models remains. in which the contribution of experts’ knowledge is considerable but nonformalised or systematic. it is hard to speak about revolution here. methods of estimation and applications received within the technical support of Riksbank enabled us to start several experiments. (2007). We have noticed that the critique refers to a larger extent to the models as such (i. tools) and less to the practice of applying them (i. a number of central banks supplement or even replace the traditional structural macroeconometric models. thus. 2003. particularly DSGE class models.Introduction Along with the development of numerical. Grabek 2 For complete image. In econometric categories. nevertheless. The analytical knowledge and experience gathered in contact with the traditional structural models rather interferes with than helps interpret the results of DSGE models. Following the events of 2008–2009 (global ﬁnancial crisis). (2001. the results of DSGE models are. namely models which attempt to translate the economic processes in a more explicit and systematic manner. In such a situation it is hard to assess whether it is the model that failed or the expert. it may be argued whether the forecasting applications of DSGE models expose the strongest sides of this class of models. restores the economy to equilibrium. a chance of structural (internally consistent and microfounded) explanations of the reasons for the recently observed phenomena and their consequences for the future. The reasons for the global economy problems are searched for in models oversimplifying perception of the world and burdening the decisions regarding economic policy. whereby experts’ knowledge is introduced through Bayesian methods2 . while searching for the reasons for the problems’ occurrence. Grabek et al. Altig et al. we consider that conclusions from a deeper analysis of the sources of 2008–2009 crisis.g. 4 The euro zone model by Riksbank develops the ideas mentioned among others by Christiano et al.e. 3 The issue of correct measuring of accuracy of forecasts. These disturbances set the economy in motion and economic agents respond to them in an optimal way. Orphanides (2007). has been omitted here.g. the usefulness of formalised tools constructed on a uniform. 7 WORKING PAPER No. The euro area DSGE model. It happens although no formal reasons exist for which the ex post veriﬁed accuracy of forecasts within the DSGE models should be higher than that of classical models3 . open. Some of the experiments have been described in separate papers. gather and exchange experience. (2004a). internally coherent (but also restrictive) paradigm for macroeconomic policy tends to be questioned. (2005b). we wish to add that there are also arguments against the engagement of central banks in constructing DSGE models. Therefore. knowhow. with estimated DSGE models. i.
systematic work with the model (preparing forecasts and analyses of their accuracy. The alternative we present in this paper summarizes some of the gathered experience. For the purposes of this paper and the ﬁrst forecast experiments we use only point estimates of the parameters reﬂecting the modal value of posterior distribution. Although it presents only the keynotes. taking into account the effects of the parallel research and the conclusions arrived at during use. A synthetic image of the model characteristics has been presented in Chapters 7–8. illustrating at the same time the correlations with other DSGE models (Chapter 3). in other words our reasoning omits — hopefully temporarily — the issue of uncertainty of the parameters. Identically as in all estimated DSGE models we are aware of. results of estimations and properties of the DSGE SOEPL−2009 model. Although due to the application of the Bayesian techniques. The description of the model speciﬁcation is completed with balance conditions on a macro scale. We present. In the ﬁrst part — relatively independent of the other parts — we have made an attempt to outline the development of the methods of macroeconomic (macroeconometric) modelling and the economic thought related to monetary policy. Additionally. outlines and ideas. The results of the estimation of parameters and assumptions made at the subsequent stages of the work (calibrated values. which describes the responses of observable variables to structural disturbances taken into account in the model (i. the formalisation and precision of presentation required in that case makes the fragment of the paper slightly hermetic — a reader less interested in the techniques may omit that chapter. the Bayesian estimation refers solely to some of the parameters (the rest of the parameters have been calibrated). The second chapter of the ﬁrst part focuses on the technical aspects of construction. variances decompositions (formally — forecast 8 8 N a t i o n a l B a n k o f P o l a n d . characteristics of prior distributions) have been presented in Chapter 6. the number of calibrated parameters has been clearly reduced. estimation and application of DSGE models. drawing attention to mathematical.Introduction and Kłos (2009). pushing aside other classes of models — at least in the academic world. Grabek and UtzigLenarczyk (2009). their equilibrium conditions as well as characteristics of behaviours of the non. Resulting knowledge shall enable the preparation of a more thorough future modiﬁcation of the model.optimising agents. The next chapter deﬁnes decisionmaking problems of the optimising agents. being aware of the consequences of faulty calibration we conducted a sort of sensitivity analysis (examination of the inﬂuence of changes in the calibration of parameters on the characteristics of the model). impulse response functions). therefore.e. The further parts of the paper refer to speciﬁcation. The SOEPL−2009 model has been estimated with the use of Bayesian techniques. simulation experiments and analytical works) may reveal issues and problems that will have to be solved. This paper consists of three basic parts. The considerations are illustrated with simple models of real business cycles (RBC) and DSGE model based on new Keynesian paradigm. a general nontechnical outline of the basic features of the model. We pass the DSGE SOEPL−2009 model for use. statistical and numerical instruments. which brought about the creation of dynamic stochastic general equilibrium models. with a view to considering and analysing other interpretation and understanding of economic processes than that proposed by the traditional models. The presented SOEPL−2009 version takes into account the conclusions we arrived at based on the analysis.
a typical set of information allowing understanding the consequences of the assumptions made at the stage of constructing decisionmaking problems (model speciﬁcation) and choice of parameters. The Appendix presents structural form equations. 83 9 . equations used to determine value at a steady state and a list of variables of the SOEPL−2009 model. examples of historical decompositions (counterfactual experiments) and information about the ex post accuracy of forecasts — this is. thanks to which the structure (relative role) of the impact of shocks on the observable variables may be assessed.Introduction error decomposition). estimation (identiﬁcation) of structural disturbances in the sample. thus. 9 WORKING PAPER No.
Part I Genesis and anatomy of dynamic stochastic general equilibrium models 10 .
see Goodfriend and King (1997).1 Methods of the Cowles Commission In the 1940s and 1950s government institutions of the most important economies started to collect in a systematic way national statistics regarding economic activity.: monopolistic competition. 1. 11 WORKING PAPER No. stochastic model of general equilibrium and.Genesis and anatomy of dynamic stochastic general equilibrium models 1 Genesis and anatomy of dynamic stochastic general equilibrium models Chapter 1 1 Genesis and anatomy of dynamic stochastic general equilibrium models The development of economic models for the purpose of monetary policy analysis has been one of the most exploited research programs within macroeconomics in the last two decades. A lot of effort has been paid to make attempts to understand the correlations between monetary policy. wages. The research is deemed to produce a sort of consensus as to the speciﬁcation of the key elements of a model of economy.g. on which Keynesian elements in the form of real and/or nominal frictions1 are imposed. inﬂation and business cycle. The new Keynesian model is a dynamic. frictions in the process of adjusting prices. frictions in the ﬁnancial market. within which modern macroeconomic analysis is being carried out. Such originating trend of macroeconomic analysis is called a new neoclassical synthesis. thus. a model deriving from neoclassical trend. The basis for its architecture is a model of real business cycle. The speciﬁcation has been named a new Keynesian model. mainly in the aspects important for the monetary policy applied by central banks. 83 11 . Early works over the econometric models of national economies complied with 1 E. The economists gained material which helped them specify quantitative models of national economy and analyze them in empirical tests.
These are problems with structural identiﬁcation and problems with statistical identiﬁcation. as they enabled the consideration of. i. seemingly. Nevertheless. additionally taking into account the interactions occurring among the variables within the different blocks. As mentioned by Hendry (1976) or Qin (1993). the emphasis was put on shortterm analysis of economic aggregates dynamics. The constructed models allowed. A statistical identiﬁcation of model is often being mentioned. The resultant equations were later on combined in a complete model of economy. 1.e. According to the philosophy of the Keynesian trend.1 Problems with identiﬁcation In the terminology derived from the paper of Spanos (1990). 12 12 N a t i o n a l B a n k o f P o l a n d . Their speciﬁcation was based mainly on statistical tests3 . Although model equations were aimed at presenting the dynamics. An example of a model within the discussed class is provided in Klein and Goldberger (1955). macroeconometrics dealt mainly with the analysis of variables dynamics in a short time with the use of partial equilibrium model. according to today’s understanding of structurality in macroeconomics. They were. multiequation econometric models. the reasons for failure of the originators of models maintained in the tradition of the Cowles Commission as regards the application of the models to economic policy analysis may be divided into two groups. sometimes. dynamic. however.e. it is considered that econometrics practiced in the spirit of the Cowles Commission emphasised the structural aspects. The economy was out of (partial) equilibrium for a short period and model simulations answered the question of how to effectively bring the economy to equilibrium. it may be concluded that structurality. forms of equations assumed ad hoc failed to comply with any mechanism of individual choice. among other things. on theories which ignored both the supply side of economy and changes in relative prices. The choice of variables was based mainly on Keynesian ISLM type models.1. The models were called ”structural”. It was also hard to research the mechanism of system response to a change in control of economic policy in a longer time perspective. Groups of variables formed the blocks of a model and each block was researched by a separate group of experts. in the period of works of the Cowles Commission and the later development of multiequation models maintained in this tradition. feedback nature relationships between variables. were the focus of interest of macroeconometrics and the theory of estimation at that time. Dynamics of each of the variables would be modelled with a single equation. which reﬂects the decisions made by economic agents. even several thousand equations.Genesis and anatomy of dynamic stochastic general equilibrium models the paradigms developed during the works carried out by the Cowles Commission. how to stabilise it. important for model estimation techniques. while the role of economic theory was limited to preparing a list of regressors to be taken 1 into account in the particular equations. Such relations and the resulting problem of simultaneity or interrelation. economet2 3 Consisting of several hundred or. Empirical macroeconomic analysis was at that time carried out with often large2 . most often linear. Thus. Therefore. i. or practically its absence primarily accounted for the failure of this trend of modelling. too large to arrive at a general image of the mechanism according to which the propagation of shocks in an economic system took place. correct quantiﬁcation of the consequences of controlling variables that depended on the persons making decisions with regard to economic policy.
The estimations of parameters within a model estimated on data originating from a speciﬁc economic policy regime shall no longer be valid if the policy regime changes. and a smaller emphasis was put on the assessment of the quality of models by virtue of diagnosis of statistical speciﬁcation errors. Although the Cowles Commission was putting emphasis on structural identiﬁcation of the model. and therefore were ineffective for practical purposes of forecasting and policy. the traditional structural macromodels. As a result of endogenous economic policy macroeconomic variables correlate with variables — the controls of policy.e. the models maintained in this convention were not able to sufﬁciently accurately replicate the statistical properties of the processes the dynamics of which they were supposed to represent. interrelation of the modelled phenomena. By assuming erroneously that the policy is exogenous. are worthless for simulation of the effects of changes in economic policy.Genesis and anatomy of dynamic stochastic general equilibrium models rics focused to a large extent on the theory of estimation4 .2 LSE and VAR models Problems with statistical and structural identiﬁcation of traditional multiequation econometric models brought about the development of several trends out of which two had the largest impact 4 Important in the works of the Commission and later works were mainly the issues regarding simultaneity. which were deemed to be structural6 . thus. the approach of the Cowles Commission. and may seem to identify the channel of policy’s impact on economy. in the academic opinion. the endogeneity may be falsely interpreted as a causal relation. the stagﬂation of the 1970s and the related failures of economic policy based on the traditional macroeconometric models disqualiﬁed. 83 13 . may not be applied to analyse the consequences of a change in the regime. 6 Parameters are deemed to be structural or deep if their value does not change under the impact of a change in the economic policy regime. Problems with structural identiﬁcation were more fundamental. Sims (1980) only supports the comments by Lucas claiming that no variable may be deemed exogenous in the world of economic agents that anticipate future events (forwardlooking agents) and whose behaviour is based on intertemporal optimisation. Therefore. Finally. i. 13 WORKING PAPER No. from the perspective of the present day. the parameters of the mode5 . 5 This speciﬁcally refers to the so called reduced form of the model. Pesaran and Smith (1995) conclude that the models represented neither the data. are actually a mixture of structural parameters and parameters related to the expectations of economic agents and. Due to instability of parameters. Structural identiﬁcation was a priority and as a result. in consequence. Therefore. the developed methods are considered. 1 1. among others. often proved ex post. may not be considered to be ﬁxed for various economic policy regimes. according to Lucas. in two papers from the 1970s and 1980s: Lucas (1976) and Sims (1980). Lucas (1976) criticises the status of exogeneity of variables — the controls of economic policy. which is exactly the purpose for which they were created. They are presented. to be unsatisfactory. He points out that the model of structural identiﬁcation proposed by the Cowles Commission does not explicitly take into account the expectations of economic agents. a model estimated within one regime may not be extrapolated outside of the regime and. nor the theory.
Genesis and anatomy of dynamic stochastic general equilibrium models on the practice of macroeconometrics: the socalled LSE method (London School of Economics). loss of information by virtue of reduction must be insigniﬁcant from the point of view of the modelled process. which currently — next to the DSGE model — is the basic tool of macroeconomic analysis. see Lütkepohl (2008). 1 1. For the resulting representation of a process that generates data to be complete. are in fact the generalisation of the LSE approach to vector time series. or y t = Ay t−1 + e t .2 VAR methodology Similar problems are reﬂected in the approach based on the analysis of time series with the use of nonstructural models of vector autoregression (VAR). The errors include. An econometric model is understood as a simpliﬁed representation of an unknown and unobservable stochastic process. (1. The reduction step entails the elimination of a variable or a group of variables from the model and is made with the use of statistical tests. omitting important variables. A completeness of the model is conﬁrmed by the statistical properties of the residuals vector. the LSE approach emphasises the correct statistical identiﬁcation of a model but is not an attempt to solve problems related to structural identiﬁcation. K are autoregressive matrices.. This is the socalled moving average representation of the process ( y t ): 14 14 N a t i o n a l B a n k o f P o l a n d . the model is reduced sequentially to the ﬁnal form. and SCVAR method (Structural Cointegrated Vector Autoregression). It originates from a dynamic model of possibly general speciﬁcation in a given class of models. The e t shocks control the dynamics of endogenous variables such that the variables may be presented as a function of the history of shocks. but are not limited to. 1. so as to cover possibly many various processes. i. Any deviations from the Gaussian white noise certify that speciﬁcation is faulty. It is based on the socalled reduction principle. Further on. The e t process covers shocks controlling the dynamics of endogenous variables.. structural vector autoregression for cointegrated variables. Ak matrices for k = 1. 2. A VAR model of order K ≥ 1 has the following form: K yt = k=1 Ak y t−k + e t . which generated the researched economic observations. Another representation is called a cumulative representation and A matrix is an autoregressive companion matrix.1 LSE methodology There may be numerous possible sources of errors in statistical identiﬁcation of a model.. They are models expressing endogenous variables by their lagged values. see Hendry (1995). or VAR models in reduced form.e.1) where y t is a vector of endogenous variables in t period. erroneous dynamic structure or illegitimate restrictions regarding exogeneity imposed on the variables.2. Thus. Nonstructural VAR models.2. The LSE approach is an attempt to overcome the problems with statistical identiﬁcation. .
or more often — structural vector error correction model. the SVAR methodology makes it possible to provide shocks with structural interpretation. In particular it is assumed that the covariance matrix of ε t shocks is the identity matrix. Next to the advantages of the LSE approach in the form of representation of a process generating the dynamics of endogenous variables through a model of rich dynamic structure. which were identiﬁed in the classical multiequation models. namely VAR. nonintuitive implications in the form of responses to shocks that may hardly be rationalised and 15 WORKING PAPER No. independent ones. SVAR and SVECM models attempt to solve problems regarding statistical identiﬁcation and some of the problems regarding structural identiﬁcation. or — assuming inﬁnite history of the process of endogenous variables: yt = ∞ k=0 1 Ak e t−k . a VAR model lacks a theoretical framework. (1. (1. These advantages explain the popularity of VAR class models in empirical macroeconomics. The e t shocks may not. Such approach is called structural cointegrated vector autoregression method. VAR models speciﬁcation lacks any theoretical basis. Thus. generally happen that e t = Bε t . but may be easily generalised to nonstationary cointegrated processes. The approach is called structural vector autoregression. Despite the aforementioned advantages. i.Genesis and anatomy of dynamic stochastic general equilibrium models t−1 y t = At y0 + k=0 Ak e t−k . shortly SVAR (from structural VAR). shortly SVECM. be structurally interpreted as they do not need to be independent. The e t shocks have the nature of errors or regression residuals (forecast errors) and may form linear combinations of the actual structural shocks which determine the dynamics of endogenous variables. VAR methods are not free of drawbacks. In order to emphasise that there have been proposed at least partial solutions not only for statistical identiﬁcation problems but also for structural identiﬁcation. then.e. where ε t are structural shocks. however. independent shocks with explicit and cohesive economic interpretation. which should be expected from shocks of structural nature. modern macroeconomic analysis is often referred to as structural macroeconometrics. Mainly. and in practice often brings.e. i. and upon assessment it may bring about. The SVAR methodology has been developed not only for stationary processes on which the approach based on the Cowles Commission paradigms must rely. This leads to the approach giving structural interpretation to VAR models.2) where y0 is the initial value of the process ( y t ). shortly SCVAR. The relations between the variables are of clearly statistical nature and even in the case of structural models they may not be referred to any economic mechanism on which the modelled process is founded.3) The basic purpose of the macroeconomic analysis carried out in accordance with the concept of a shock impacting the system of endogenous variables as a source of their dynamics is to identify shocks of structural nature. It may. 83 15 . i.e. (ε t ) = I. and generally are not.
The methodologies of DSGE models are. Whereas the decisionmaking rules of households reﬂect the process of optimisation of welfare. are directly related to VAR class models. i. macroeconomic analysis. It may be considered as a VAR model on the parameters of which. Hence.3 Methodology of modern macroeconomics The views with regard to the method of analysing the dynamics of economic aggregates have signiﬁcantly changed over the last 30 years. on elements of matrices A and B. The decisions of agents are. 1. therefore. i. The restrictions originate in the theory of economics underlying the speciﬁcation of a DSGE model. Speciﬁcally. at least theoretically. nevertheless. as the structural nature of its speciﬁcation imposes strong restrictions on the dynamic structure of the reduced form.e. always optimal a priori. an answer to the problems related to structural identiﬁcation. as the solution of a DSGE model is a VAR class model. In that scope a DSGE model provides.e. namely the form which sets dynamics of endogenous variables around the equilibrium. only partial solutions.Genesis and anatomy of dynamic stochastic general equilibrium models noncohesive forecasts. however not so signiﬁcant because the DSGE models. such as DSGE models. Most often four types of agents are distinguished. a set of restrictions has been imposed. to a major extent. The step is taken at the cost of statistical quality of the model. through VAR class models. They are households. and more speciﬁcally their approximate solutions. Optimisation takes place in a stochastic economic environment and in the case of households the structure of their preferences — the socalled utility function — is also set. The DSGE model speciﬁcation founded on the theory of economics ensures that restrictions imposed on the dynamic structure of its reduced form derive from the optimal decisionmaking rules of rational economic agents operating in an internally cohesive economic reality. ﬁrms as well as government and central bank. as opposed to traditional Keynesian models or multiequation macroeconometric models where the forms of interrelations between economic variables were assumed ad hoc. the best of possible taking into account the available information. a DSGE model expresses a tradeoff between the proper statistical and structural identiﬁcation of macroeconomic processes. to dynamic stochastic general equilibrium models based on the theory of economics. Households decide about 16 16 N a t i o n a l B a n k o f P o l a n d . Thus. thus. They also. work or assets. both physical such as capital and ﬁnancial such as bonds or money. those of ﬁrms result from maximisation of the expected proﬁt. yet ensure its internal cohesion and guarantee structural nature of the identiﬁed shocks. i. reply to the problems related to statistical identiﬁcation. Modern macroeconomics attempts to explain the dynamics of economic aggregates with the use of models based on the socalled microfoundations. particularly at central banks. more and more often refers to models having theoretical foundations. the socalled reduced form of DSGE model. Due to the foregoing disadvantages.e. Usually. DSGE models . a discounted stream of the expected utility. the mechanisms shaping the decisionmaking rules of economic agents are explicitly modelled. The evolution followed from models speciﬁed in the tradition of the Cowles Commission. is a VAR model. They limit the dynamic structure of the model. they refer to three types of economic categories: products and services.through economic theory underlying their speciﬁcation — constitute another 1 step forward to solve the problems with structural identiﬁcation.
In that meaning their decisions are called rational. to set out optimal decisionmaking rules for all agents and apply them. A central issue in the theory of dynamic general equilibrium is the intertemporal dimension of the decisionmaking process. 1 17 WORKING PAPER No. characteristic to the literature of the subject matter. supply of labour and change in the structure of the portfolio of ﬁnancial assets. In that sense the mechanism of building expectations by the economic agents is rational.e. a perfect measurement of the value of all of the variables and shocks impacting the economy in any period. collects taxes. The decisions of agents consist of intertemporal allocation of the available resources. Should it appear ex post that the decisions of the agents are not the best possible to be taken. Beside of the optimisation criterion. however. economic agents make optimal decisions. expends transfers and incurs public debt. the level of investments. Firstly. an unexpected exogenous growth in productivity took place. this is only due to information gap. i.g. It is. the process of making decisions by economic agents takes into account several categories of limiting conditions. based on the model. Longterm equilibrium is also called stationary state or steady state. i. The decision rules of the government and the central bank are usually made ad hoc. e. The individual decisions are coordinated by the market. in a short period it may be out of longterm equilibrium. initial conditions.g. i. equilibrium conditions and constraints regarding the available technology and information structure in the economy8 . see Fagiolo and Roventini (2008). Firms decide about product supply and demand for labour7 . assumed that agents know the complete model of economy. These are most often budget constraints. thus. which brings about decentralised allocation of resources. ex post they do not make mistakes in a systematic manner as to the actions undertaken. due to the fact that after the moment of making the decision an event occurred that could not have been foreseen by the agent. It is. 7 The mentioned division into the decisions of households and ﬁrms is conventional. namely that they know (the real) principles governing the world in which they live and the values of all of its parameters. Secondly. Today’s income may be allocated to future consumption and future income may ﬁnance today’s consumption expenditures. purchase or sale of bonds. etc. The government determines the value of public spending. 8 Ex post there are also imposed the so called transversality conditions of noPonzi game type. That omnipresent transparency and rationality draws critique on the part of alternative paradigms of economic modelling such as multiagent modelling (agentbased computational economics). 83 17 .e. therefore. It reﬂects a shortterm equilibrium in at least two meanings. The central bank controls the nominal interest rate and/or the supply of money. economy is described as a dynamic system. Intertemporal substitution of resources is possible by virtue of participation in the market of ﬁnancial assets. i.Genesis and anatomy of dynamic stochastic general equilibrium models consumption.e. in each point of time economy reﬂects a general equilibrium as understood by Walras. They are also able. e. the level of use of the available capital. Formally. The names are founded on mathematics. The agents build expectations as to the future values of economic variables through the operator of conditional expected value. Even though the economy is — in the above meaning — assumed to remain at all times in shortterm equilibrium. demand for a product.e. assumed that prices always clear the markets. which would require in practice the possibility of making a perfect ﬁltration.
the theory of Real Business Cycle (shortly RBC) has gained the status of a leading macroeconomic theory with regard to economic ﬂuctuations analysis — a model that may be called a standard RBC model from today’s perspective. A model implying the comeback of economy to the steady state upon structural shock occurrence is called a steadystate or stationary model.Genesis and anatomy of dynamic stochastic general equilibrium models Longterm equilibrium is a mathematical concept and refers to the model of economy instead of the economy in itself. DSGE models. which is introduced to DSGE models in an exogenous manner.and mediumterm economic ﬂuctuations has two perspectives — both methodical and conceptual. simulation and validation of RBC models. Both categories are a product of data ﬁltration and their form depends on the chosen method of ﬁltration. It happens 1 because stochastic disturbance. It shall be emphasised that both the trend and the cycle are statistical ﬁction and have no equivalent in the real world. Ad hoc assumptions made with regard to the mechanisms of forming expectations by economic agents gave way to rational expectations. As in reality structural shocks happen at any moment. it shall asymptotically come back to the same equilibrium. structural shocks or structural innovations impacts the economy. There is no direct equivalent in the real world. Economy in a steady state may be knocked out from the state. as ones deriving from the RBC models family. If the effects of shocks abate. 18 18 N a t i o n a l B a n k o f P o l a n d . namely longterm trend and shortterm cyclic ﬂuctuations around the trend. the socalled business cycle. It shall. Ad hoc assumed behavioural equations describing economic aggregates gave way to equations of motion derived from intertemporal solutions of optimisation problems of the economic agents operating in perfectly competitive and frictionless markets. are used in cycle analysis and due to their construction are not able to answer any question regarding the forming of the trend.e. Economy reﬂects longterm equilibrium if all of the variables grow from one period to another according to ﬁxed growth rates9 . 1. the basic tool of business cycle analysis. Examples of structural shocks are technological shocks (increase or drop in total factor productivity). the theory of Real Business Cycle made the dynamic stochastic general equilibrium model a basic tool for macroeconomic analysis. The steady state path may be interpreted as the path taken by the economy would it not be for the shocks. which was reﬂected in the use of the methods of calibration. the real economy shall never come to the same steady state.1 Real Business Cycle theory Since the beginning of 1980s. shock of preferences or contractionary monetary policy. however. This shall happen if the effect of shock is permanent.3. when the ﬁrst papers of the type Kydland and Prescott (1982) appeared. the economy comes back to longterm equilibrium but not necessarily to its preshock state. 9 The deﬁnition covers the case in which some or all variables are unchanged from period to period. The inﬂuence of the RBC revolution on understanding short. The creators of the RBC trend emphasised as well the importance of quantitative aspects of macroeconomic analysis. From the methodological point of view. i. This is the subject of the theory of economic growth. If economy is knocked out of longterm equilibrium by a shock of temporary effect. ﬂuctuate in its environment. i.e. Modern macroeconomic analysis decomposes the time series of economic aggregates into two basic components.
estimated) based on the data that may be generated by nonintuitive statistical artefacts.Genesis and anatomy of dynamic stochastic general equilibrium models Equally fundamental proved to be the conceptual implications of the RBC theory. even if the only shock impacting economy is the shock of productivity. as well as rift between its implications and the practice of economic policy. Firstly. The RBC theory implies. it appears from the RBC theory that a business cycle presents an effective. i.e. even though they are calibrated (today. the dynamics of real variables in economy. RBC models are dynamic stochastic general equilibrium models instead of partial equilibrium models. particularly in the academia. The results contradicted the general opinion that monetary policy impacts the real economy in a short run. They have an internally cohesive theoretical structure. RBC methods could not be considered satisfactory. On the other hand. RBC models try to replicate the ﬂuctuations of products and other economic aggregates.g. The second implication of the RBC theory is the conclusion that the main reason for economic ﬂuctuations are technological shocks. do not depend on monetary policy. derived from microfoundations instead o f being assumed ad hoc. (1998). the results of simulation experiments (e. therefore. role of monetary factors in economy. such as production. RBC models reﬂect. RBC models — due to the results regarding the neutrality of money — were not accepted by central banks. i. RBC methods proposed solutions to problems that discredited the traditional approach based on paradigms developed by the Cowles Commission. cyclic ﬂuctuations — including recessions — do not bring about ineffective allocation of resources — they are fully optimal and do not result from market imperfections. while anticipation of future events inﬂuences their today’s behaviour. The decisionmaking rules of economic agents in an RBC model are structural. optimal path of economic aggregates. Such interpretation of economic ﬂuctuations contradicted the traditional view that changes in productivity contributed to economic growth. the achievements of the revolution of rational expectations. Monetary policy may only affect the nominal values such as the nominal interest rate or nominal supply of money. Thus. which has nothing to do with a business cycle. particularly on vector autoregression methods. mainly technological shocks. This clearly contradicted the standard Keynesian interpretation of recession as a period in which economic resources are used ineffectively and the end of which may be advanced by stimulation of aggregated demand. therefore. Due to the growing evidence of contradictions between the Real Business Cycle theory and empirical research. In other words. RBC models 19 WORKING PAPER No. The expectations of agents are rational. consumption or employment.e. shock response analyses) reﬂect internal cohesion. The central banks continued to focus on classical multiequation models and more and more often on LSE type approach. whose values do not affect the real sphere of economy. The third fundamental conceptual implication of the RBC theory is the limited. This is not the case with VAR class models. Therefore. which temporarily improve or deteriorate the total factor productivity in the economy. both expansion and contraction result from optimal reaction of economic agents to exogenous shocks impacting the real sphere of economy. A standard RBC model implies the neutrality of money even in a short period. that the stabilisation policy of a government may change resources’ allocation but only to less efﬁcient one. see Friedman and Schwartz (1963) and Christiano et al. Hence. or practically nonexistent. 83 1 19 . even the structural ones. forecasts. namely inﬂation. Even though the RBC theory had a considerable inﬂuence on the method of understanding business cycles.
therefore. The assumed simpliﬁcations are nonsigniﬁcant from the point of view of the presented conclusions. which leads to the basic form of a new Keynesian DSGE model. 1. A unit price of bonds in period t amounts to Q t = 1 1+i t divided between consumption C t and savings B t . while consumption increases and labour decreases welfare. One bond purchased in period t at price Q t is worth in period t + 1 one monetary unit. the model corresponds to the standard speciﬁcation of an monetary model which builds on the RBC origins. the reduced form of an RBC 1 model is a VAR type model10 . Subscript d shall stand for demand for labour s reported by a ﬁrm.3. Household’s welfare is measured with a utility function U(C t . The expenditures of a household in period t are.2 Frictionless economy A standard frictionless monetary model is presented below11 . a household receives wage of the worth of H t Wt . 11 The presented version of the model abstracts from investments and government sector.. with a homogenous labour supply12 H t .3 Representative household In exchange for nominal wage Wt . .3. thus. is the income of a household D t = H t Wt + B t−1 . Beside those simpliﬁcations. Pt C t + Q t B t . which is located in bonds with no risk. This feature shows that problems related to statistical identiﬁcation are not ignored in the RBC methods. In the following paragraph the model is extended with Keynesian elements. 20 20 N a t i o n a l B a n k o f P o l a n d . a representative ﬁrm and a central bank. where i t is (products) manufactured by the ﬁrm.. it presents the most important components of a DSGE model. 10 The reduced form of a DSGE model is a process of VARMA class. A standard DSGE economy consists of a representative household. while the unit price of the product is Pt . 12 Subscript s stands for the supply of labour offered by a household. are deep parameters. 1. together with the savings B t−1 in period t − 1. The labour market is perfectly competitive. A household consumes homogenous goods . The introduction of the elements of new Keynesian economics to the classical RBC model proved to be a solution — such derived model is called a new Keynesian model. the models are at least partially resistant to Lucas’ and Sims’ critique. attempts of such modiﬁcation of the Real Business Cycle theory have been made so that the internally cohesive structure of an RBC model be preserved. 2. In other words. H t ). perhaps of inﬁnite order. namely a model covering a wide class of stochastic processes.a household provides a representative ﬁrm in each of the s periods t = 0. Savings are the nominal interest rate determined by the central bank. subscripts s or d shall be omitted when it must be t emphasised that we mean the labour in general equilibrium. The presentation of the standard DSGE model in this paragraph has been founded on the monograph by Gali (2008). It is clear that RBC methods could not be simply rejected. The wage. In general equilibrium H t = hd . All those features indicate that RBC methods well manage the problems of structural identiﬁcation. in practice the process is approximated with a ﬁnite order VAR model. the models allow us to correctly analyse the effects of changes in economic policy regime. Additionally. Thus. their dynamics be related to VAR class models. 1. or at least some of them. Despite a simple structure. We will start with the frictionless monetary model of the economy and then introduce some frictions which render money nonneutral. Thus.Genesis and anatomy of dynamic stochastic general equilibrium models parameters. In exchange. while the role of money be increased in a short time. however.
in t = 0 period a household solves the problem of maximisation of the discounted stream of the expected utility: max s ∞ t=0 s β t U(C t . (1. at the same time. {C t . determines the value of consumption C t depending on the expected consumption C t+1 and inﬂation Π t+1 .8) is recursive. H t ) − ω t (Pt C t + Q t B t − B t−1 − Wt H t )] . (1. B t } for the determined t 1 but also for all of the t = 0.5) where t is the operator of the conditional expected value.} 0 (1. The second equation — labour supply equation — determines. H t ) 1 ∂ s Ht Wt oraz ωt = β Qt t ω t+1 . A household must.4–1. multiperiod nature. σ (1. The choice of {C t . then. Decisions of a household are not. oneperiod problem. Wt Pt s = (H t )σh C t c . decide not only about variables {C t .. 2.1. = (1. the decision rules of a household are the following: Ct = Qt β σc t {(C t+1 ) Π t+1 } 1 σc . consumption’s law of motion. The ﬁrst equation.. then.. t = 0. s the value of Wt H t + B t−1 . ωt = − s ∂ U(C t .7) and for CRRA13 type utility function. as today’s decisions affect the value of resources available tomors row.. a representative household must.e. i.} shall maximise the expected welfare of the household. the supply of labour that needs to be provided by a household in order to generate income s necessary to cover the costs of consumption. i.e.5) may be solved by using Lagrange’s functional in the form: ∞ t=0 s s β t [U(C t . H t . 1.. H t .2. 13 The constant relative risk aversion utility function is deﬁned as: U(C t . 1. Problem (1.. H t ).t=0. The decisionmaking process has a dynamic. such as to maximise utility and by virtue of spending Pt C t + Q t B t not to exceed the budget of their consequences are not important only ”here and now”. 83 21 . however. B t }.H t . make a decision as to variables {C t . 21 WORKING PAPER No. H t . . . H t ) 1 ∂ Ct Pt . In order to know how to do that. 2. H t ) = (C t )1−σc − 1 1 − σc − (H t )1+σh − 1 1 − σh .6) First order conditions are as follows: ωt = s ∂ U(C t . for the assumed real wage value.4) subject to a sequence of budget constraints: s Pt C t + Q t B t = Wt H t + B t−1 .Genesis and anatomy of dynamic stochastic general equilibrium models s In period t..8) System (1.
Generally. (1. the solution of which implies equating real wage with marginal productivity of labour: At = which is exogenous. loglinear form of the model.Genesis and anatomy of dynamic stochastic general equilibrium models 1. In the basic model there are three markets — product market (consumption balancing). market clearing conditions are imposed on all markets. the economy is assumed to remain in general equilibrium state in every period. In order to determine d the H t value. as well as real marginal costs and monopolistic markup which are expressed in logs. Exceptions: real and nominal interest rates. The product market’s clearing condition states that the 14 Small letters denote in this chapter percentage (logarithmic) deviations from the steady state.3.3. Timeless variables denote steady state values. which — for the assumed technology level — determines the value of the product.9) w t − p t = σc c t + σh hst . d Ht subject to the technological constraint: d Yt = A t H t . Therefore.e. The model in its existing form is nonlinear. 1. 1 where A t is an exogenous process of total factor productivity. a ﬁrm solves in each period a static problem of proﬁt maximisation: d max Pt Yt − Wt H t . A decisionmaking variable in the case of a ﬁrm is the demand for labour H t . the equilibrium conditions are superimposed on the simpliﬁed. It manufactures a homogenous real product Yt with the use of production function Yt = A t Nt . w t − pt = at . which are expressed in absolute values. no satisfactory methods exist to solve nonlinear DSGE models. i. 22 22 N a t i o n a l B a n k o f P o l a n d . Wt Pt . which for the presented frictionless model reads as follows14 : ct = t c t+1 − 1 σc (i t − t π t+1 − ρ). y t = a t + hd . labour market and savings (bonds) market.4 Representative ﬁrm d A representative ﬁrm hires labour supplied by a household H t in exchange for nominal wage Wt . t where ρ = − ln β. The product market is perfectly d competitive.5 General equilibrium Upon determination of optimal decision rules of a representative household and a representative ﬁrm.
here — labour productivity.3.6 Consequences for monetary policy Based on the conditions (1. the only structural shock. real wage w t − p t and real interest rate r t = i t − t π t+1 : ct = yt = ht = a. t Whereas households trade bonds only between one another.9) and the market clearing conditions.Genesis and anatomy of dynamic stochastic general equilibrium models whole product supply y t is subject to consumption: yt = ct the labour market clears when: hd = hst = h t . rt = ρ + σc (σh + 1) σh + σ c (ρ − 1)a t . the savings market clears automatically: b t = 0. where ε t ∼ N (0. it results that a change in the nominal interest rate is one to one translated into a change in More precisely. 83 23 . from Fisher equation: r t = i t − 15 16 t π t+1 . the real interest rate equation results from Euler household equation. σh + σ c t σh 1 − σ c σ c σh + σ c σh + 1 at . equal to one (zero for the logarithm): mc t = −λ t = w t − p t − mpn t = 0. the process of total factor productivity (TFP). σ) is a structural shock of labour productivity. where mpn t = a t stands for marginal labour productivity.e. The last of the equations above. Therefore. there may be determined the dynamics of real categories of product y t . i. And in the case of real interest rate — on the expected change in technology t ∆a t+1 = (1 − ρ)a t . 23 WORKING PAPER No. The condition equating real wage with marginal productivity of labour. w t − p t = a t . employment h t . (1. where technology15 is controlled with an exogenous stationary process in the form of: a t = ρa t−1 + ε t . may be read out as the determination of real marginal cost (denoted by mc t ) or — 1 1. after loglinearisation equally — monopolistic markup λ t = −mc t . In a standard RBC model this is rt = it − t π t+1 .10) w t − pt = at . It appears then that the dynamics of real variables in a standard frictionless economy model depends only on the level of technology a t 16 .
On the other hand. which would. A number of papers proposed methods to consider nominal rigidities in the dynamic model of general equilibrium. however.4 New neoclassical synthesis The conclusion regarding the neutrality of money in a standard frictionless monetary model of the economy was a decisive factor that the models of that class could not raise serious interest at institutions such as central banks. In the 1980s and 1990s the analysis covered various microeconomic mechanisms. The equilibrium dynamics of real categories does not depend on the nominal interest rate i t . In the following paragraph. is nonproductive. Monetary policy does not affect the decision rules of ﬁrms and is neutral for the welfare of households. the purpose of which was to bring the general equilibrium models closer to economic reality. Initially — analogically to new Keynesian literature — emphasis was being put on rigidities in the process of adjusting prices. emphasising the reasons for nonneutrality of money. reﬂect the central elements of the trend. This leads to many attempts of developing the originally frictionless monetary models towards speciﬁcations abiding by the methodical achievements of that trend. equivalently.Genesis and anatomy of dynamic stochastic general equilibrium models inﬂation expectations. Consequently. as the interest rate is translated only to nominal category — the expected inﬂation. The proposed models in fact extended the standard frictionless framework to the case of ﬁrms operating in the market of monopolistic competition which are not always able to determine markups or. The choice of distortions in the process of price or markup determination offered the simplest solution guaranteeing that monetary shocks shall have real effects and shall reﬂect adequate persistence. whose practice and understanding of economic categories ﬂuctuations contradicted the implications of Real Business Cycle theory. however. 24 24 N a t i o n a l B a n k o f P o l a n d . implying nonneutrality of money in a short run. we present a standard new Keynesian model based on a standard frictionless engine. In the case of the trend that originated as an effect of merger of the RBC theory with the elements of new Keynesian economics — the trend that is currently called new neoclassical synthesis — the starting point was the model of general equilibrium of frictionless economy (or distortionfree economy). mainly the socalled distortions or noneffectiveness. Therefore. 1. They appeared particularly attractive due to explicit modelling of the decisionmaking process of economic agents and speciﬁcation of the model based on structural parameters — unchanging in the analysis of alternative scenarios of economic policy. including monetary policy applications. requiring high monetary and institutional expenditures. Neutrality of money and the primacy of technological shocks in the course of the cycle contradicted economic practice and the concept that demand factors and monetary shocks should play more than marginal role in forming the cycle. prices at the optimal level. a standard RBC model implies that implementation of monetary 1 policy. the normal course of events seems to be the beginning of construction of a new theoretical trend based on a possibly simple case. microfounded models gained the interest of academic communities because — compared to previous macroeconometric models — they made a signiﬁcant step ahead from the point of view of the method of economic modelling.
e.2 Representative household Whilst consumption of a representative household consists of a continuum of products. The main difference appears in the sector of ﬁrms that possess monopolistic power. e. The central bank implements monetary policy applying the monetary policy rule. (1.4. households face the same decision rules as in the frictionless model. Taylor rule. [0.1 Standard newKeynesian model Economy consists of a representative household.g. In period t.11) w t − pt = σc c t + σh hst .1] (1.12) where η is the elasticity of substitution between any two products in economy.e. They operate in the market of imperfect competition. while total consumption in economy originates as a result of averaging the consumption of the particular products with the use of CES integral aggregator: η−1 η η η−1 Ct = [0. in literature the application of integral aggregators is common. ﬁrms have monopolistic power and may determine the prices of their products. a representative household consumes C t (i) of product i.1] C t (i) di . based on brand naming. It is assumed that there is a continuum of ﬁrms and each of them is represented by a point on [0. Hence. The products of the particular ﬁrms are distinguishable. take the form of: Pt (i)C t (i)d i + Q t B t = B t−1 + Wt H t .4. i. then. Yet.Genesis and anatomy of dynamic stochastic general equilibrium models 1. 1]17 interval. ﬁrms — no longer a representative ﬁrm — and central bank. The economic environment in which a representative household is operating is identical as in the standard frictionless monetary model.13) where Pt (i) stands for the price of product i determined by ﬁrm i in period t. Thus. then. that decisionmaking problem of a representative household. while the elasticity of substitution between products manufactured by various ﬁrms is ﬁnite. so they are no longer price takers. after loglinearisation they take the following form: ct = t c t+1 − 1 σc (i t − t π t+1 − ρ).g. 83 25 . usually monopolistic competition. 1 1. Each ﬁrm manufactures a product different from the products manufactured by other ﬁrms (a variety). as well as its decisionmaking functions are the same as in the standard frictionless monetary model. If we assume that There is no obstacle for the number of products to be ﬁnite. Budget constraints. Integral aggregators are. it faces the same decisionmaking problem as in the frictionless model. 17 25 WORKING PAPER No. In other words. (1. replaced with sumbased aggregators.
With probability The probability ξ does not depend on time that elapsed from the last period in which the ﬁrm had the possibility to reoptimise the price of its product.17) 1. they can set the prices of their products.4. A household decides the share in the total consumption C t of particular products by virtue of maximisation of consumption Ct = ( [0. to the constraint identical as in the frictionless model.1] C t (i) η−1 η di . with probability equal to 1 − ξ. The solution to this problem leads to the function of demand in the form of: C t (i) = Pt (i) Pt −η Ct . η η−1 η for the assumed value of consumption expenditures solving the problem: [0.1] Pt (i)C t (i)di = Z t . 1]. In period t a ﬁrm i determines the price of a product i at the level of Pt (i). In other words.15) i.18) The level of stationary technology A t is common to all ﬁrms. (1.e. 26 26 N a t i o n a l B a n k o f P o l a n d .e.1] Pt (i) 1−η di (1.16) subject to [0. where Z t is an auxiliary variable determining the value of consumption expenditures.3 Firms A ﬁrm i in period t is assumed to manufacture Yt (i) of product i with the use of production function such as that applied by a representative ﬁrm in the frictionless model: d Yt (i) = A t H t (i) (1. a ﬁrm i in period t may determine an optimal price of its product Pt (i). (1.Genesis and anatomy of dynamic stochastic general equilibrium models prices Pt (i) aggregate to the average price Pt in period t according to: 1 1−η Pt = [0. As ﬁrms have monopolistic power.1] max C t (i) Ct = [0. The process of determining prices is subject to frictions. all the ﬁrms which in equal to ξ the ﬁrm is forced to leave the price of its product at the previously determined level.1] C t (i) η−1 η di) η−1 Pt (i)C t (i)di.14) 1 the budget constraint aggregates to: Pt C t + Q t B t = B t−1 + Wt H t (1. which maximises the expected discounted ﬂow of its proﬁts. Price rigidities of the Calvo (1983) type are most frequently applied. Hence. for each i ∈ [0. i.
Firms generate inﬂation by determining an optimal price at the level different from the average price level in the previous period. solve the same optimisation problem. aggregator. in the new Keynesian model the mechanism of inﬂation is modelled explicitly. It may be shown that such aggregation leads to a relation between product. as opposed to the frictionless model. After loglinearisation.Genesis and anatomy of dynamic stochastic general equilibrium models period t may reoptimise prices.1] H d (i)di. Yt+kt = ( P t )−η C t+k stands for the t+k 1 households’ demand in period t + k for the output of a ﬁrm that had the latest opportunity to reoptimise prices in period t. namely determine the same price Pt .t+k stands for the stochastic discount factor. 1.19) where Z t. t The condition of labour market clearing requires that the aggregated demand for labour be equal to supply of labour by a representative household: hd = hst = h t t 18 This dependence requires the elimination of the effects of wage dispersion in economy. P (1. 83 27 . d Demand for labour is aggregated with the use of H t = [0. 27 WORKING PAPER No.4 General equilibrium The condition of clearing the product markets makes demand equal to supply in each market: C t (i) = Yt (i) for every i ∈ [0.t+k (Pt Yt+kt − Ψ t+k (Yt+kt ))}. and Ψ t+k (Yt+kt ) = Wt+k H t+k (Yt+kt ) stands for the nominal cost of such ﬁrm. productivity and employment in the form of18 : y t = a t + hd . 1] and entails the clearing of an aggregated product market: Yt = C t providing that product aggregation is carried out with the use of aggregator in the following form: Yt = [0.1] η−1 η η η−1 Yt (i) di . The price Pt solves the following problem: max Pt ∞ k=0 (βξ)k t {Z t. Thus. the condition takes the form of: yt = ct .4.
the real variables of the new Keynesian model may be Which means that the real cost logarithm mc t shall be zero.21 which — making the real wage dependent only on labour productivity — closed the frictionless model and allowed to determine dynamics of real categories with no need to deﬁne the method of determining the nominal interest rate i t 20 . they force the real marginal cost to be equal to unity19 : mc t = w t − p t − mph t = w t − p t − a t = 0. equivalently.21) Equation which made the real wage dependent on productivity only.22) t π t+1 . Thus. Actually. (1. When ﬁrms operate in monopolistic competition conditions. Assuming the exogeneity of the process generating average markups λ t . 1 1.20) In the frictionless monetary model the decision rules of ﬁrms — making the real wage equal to marginal productivity of labour or. along with the conditions of market clearing. rate and the expected inﬂation was included by virtue of the Fisher rule.Genesis and anatomy of dynamic stochastic general equilibrium models The bonds market clears automatically: b t = 0. closed the frictionless model in the sense that it was possible to determine the dynamics of real variables with no need to refer to the rules of monetary policy. the average real marginal cost in economy mc t does not have to be equal to unity but it is equal to the inverse of the average monopolistic markup in as ﬁrms that may reoptimise the price determine it above the marginal nominal cost. (1. Euler equation determined only the dynamics of real interest rate r t = i t − economy λ t : mc t = w t − p t − mph t = w t − p t − a t = −λ t = 0 (1. 20 19 28 28 N a t i o n a l B a n k o f P o l a n d .4. Equation 1. y t = a t + hd . Equation 1. t where ρ = − ln β. Equation 1.22 relates the real wage with productivity and average monopolistic markup. making the nominal marginal cost of production equal to market price — made the real wage dependent on employment and productivity. w t − p t = σc c t + σh hst .5 Consequences for monetary policy The decision rules of households and aggregated production function of a standard newKeynesian model corresponds to that in the presented frictionless monetary economy: ct = t c t+1 − 1 σc (i t − t π t+1 − ρ).22 as ﬁrms that may reoptimise the price determine it above the marginal nominal cost.
equivalently. then.Genesis and anatomy of dynamic stochastic general equilibrium models expressed by RBC model variables and by average markups: yt = yt − ht = ht − 1 σ c + σh 1 σ c + σh λt . Due to disturbances in the prices formation process changes of the nominal interest rate are not one to one translated into the expected inﬂation but affect the real interest rate. More precisely. Therefore. In the case of price difference equation for the optimal price. equivalently. In reply to the monetary policy shock not all ﬁrms shall be able to determine markup on the level of the optimal response. 29 WORKING PAPER No. the process of determining optimal prices must be disturbed because if we assume perfect ﬂexibility of prices. Average markups depend on the expected increment of the optimal level of prices. σh σ c + σh t (λ t+1 (1. The markup shall.e. be ﬁxed and equal to λ t = ln η−1 . the sector of ﬁrms shall be reduced to a representative monopolist determining the optimal price in each period at the level of: Pt = η η−1 ψt where ψ t stands for the nominal marginal cost of a representative monopolist. the monetary policy shall inﬂuence real categories. the average markup may be determined as the function of the expected optimal price in the next period: λt = λ + βξ (1 − βξ) t (p t+1 − p t ). w t − p t and r t variables are that resulting from the frictionless model. Consequently. (1. as long as the monetary shock has not expired. by the process of determining optimal prices. i. which within the model are determined endogenously. which upon linearisation has the form of: p t = p t−1 + βξ t {p t+1 η rigidities of Calvo type.23) λt . Based on that equation. w t − p t = (w t − p t ) − rt = rt − 1 σ c + σh 1 where the values of y t . 83 29 . − λ t ).24) ˆ where mc t = mc t − mc = −(λ t − λ) stands for the deviation from the average real marginal cost mc t in economy from its value in steady state mc or. it may be seen that nonneutrality of monetary policy in the presented standard model takes place by virtue of the process of determining optimal margins or. the solution of the decisionmaking problem of ﬁrms leads to a stochastic ˆ − p t } + (1 − βξ)mc t + π t . the minus deviation of the average markup λ t in economy from its value in steady state λ = −mc. h t . it may be seen that the model with price rigidities shall not be closed such as the frictionless model or model with a representative monopolist did. Thus. λt . which would close the considered model so that the dynamics of real categories would be independent of monetary policy. exclude price rigidities. namely in the case of absence of price rigidities and in perfect competition.
the level of which is determined by the central bank. (1−ξ)(1−βξ)(σc +σh ) . i. meaning one that may be expressed by the level of technology a t . They are ad hoc assumed equations deﬁning the response of interest rate to the changes in macroeconomic aggregates. ξ and ˜t = y t − y tn stands for the output gap. i. the difference y where: r tn = ρ + σc t (1 − ρ)a t is the natural interest rate. r tn and a t . If the interest rate reacts to inﬂation changes and output gap. ˜t . an equation for i t is assumed in the form of the socalled monetary policy rules.shall be introduced. y Nevertheless.e. as may be expected in the light of the presented motivation. The equation is known as the newKeynesian Philips curve.e. an equation for i t . to the socalled dynamic IS equation: ˜t = y y t { ˜ t+1 } − 1 σc (i t − 1 + σh σc (1 − α) + σc + α n t {π t+1 } − r t ). Most commonly. Euler equation brings about the dependence of output gap on the path of differences between the real interest rate and its natural level. 30 30 N a t i o n a l B a n k o f P o l a n d . namely the product that would be materialised in the case of perfectly ﬂexible prices. 1 where κ = between product y t and the socalled natural product y tn . The last three equations and the process deﬁning the technology a t = ρa a t−1 + η t deﬁne the form of the new Keynesian model with regard to π t . it shall be decided how the central bank is to control the nominal interest rate. Such type of rule is known as Taylor rule. the form of the rule is following: y i t = ρ + φ y ˜ t + φπ π t . In order to be able to solve the model. i. equations includes the i t variable — the nominal interest rate.e.Genesis and anatomy of dynamic stochastic general equilibrium models Optimal price policy of ﬁrms implies the inﬂation equation in the following form: πt = β y t {π t+1 } + κ ˜ t .
The y t vector is called a vector of endogenous variables of the model. 83 31 .e. (2. the solution of model (2. the so called information set. the DSGE model shall be a stochastic difference equation in the following form1 : t f ( y t+1 . y t . shocks that have materialised today. The mechanism represented by the structural form of the model includes the encoded decision rules of economic agents. i. it is necessary to decode the rules and represent them in operational form. vector and structural shocks in vector ε t . Knowing the solution of the model. be gathered in y t . i. disturbances are endogenous variables. except structural shocks.e. In particular. conditions constraining their decisions and equilibrium conditions.e.1) where {ε t I t−k } = 0 for k ≥ 1 and I t represents the information about the state of the world possessed by the economic agents in t period. where t is a structural shock.2) 1 Vector y t includes all of the variables of the model except structural shocks. i. form enabling their direct implementation. Should all the variables of the model.1) is any g function in the form of: y t = g( y t−1 . Both former states and today’s shocks come within the information set I t .e. 31 WORKING PAPER No. The operationality of the model solution from the perspective of t period is understood to be such property that today’s y t state may be determined based on the knowledge of its previous value y t−1 and the value of ε t .1) is called a structural form of a DSGE model. ε t ). (2. Formally. states. it represents the economic mechanism determining the dynamic properties of economy. In order to solve the model. y t−1 . The form (2. Variables in the form of θ t = ρ θ t−1 + t are called model disturbances. economic agents are able to implement their decisionmaking rules.DSGE model — anatomy 2 DSGE model – anatomy Chapter 2 DSGE model — anatomy 2 Formally. possessed by the economic agents in t period. i. From the point of view of model representation in the so called state space. it may include autoregressive variables in the form of: θ t = ρ θ t−1 + t . a DSGE model is a system of ﬁrstorder conditions for economic agents. ε t ) = 0. state vector or simply state.
0). i. instead of considering the nonlinear model (2. the law of motion for the state or policy function.3) The condition implies that the decision rules represented by the g function are actually possible and optimal. model (2. (2. ε t ). the state vector y t . Therefore.1) into ﬁrstorder Taylor 2 Formally — the values of the differentials. Such approximation is called a logarithmiclinear (loglinear) approximation. the values of their increments2 d y t have the interpretation of percentage deviations from the point in the neighbourhood of which the model is approximated. a decision rule.1): t f g g( y t−1 . i. Equations of motion determine the values of all variables of the model in t period.1) into the Taylor series takes place in the neighbourhood of a chosen point. The latter name originates from the optimal control theory. It has become customary that the point is the point of longterm economy equilibrium. g( y t−1 . ¯ . Moreover. ε t . ε t ). ¯ .e. expansion of (2. the model is most often expressed in logarithms of variables such that after linearisation. shall satisfy the equation (2. if no structural shock occur — economic agents in the steady state have no motivation to change their decisions. based on the previous state of economy y t−1 and structural shocks ε t that occurred in the current period. ε t+1 ) and f ( ¯ . Thus.DSGE model — anatomy which solves the structural model. This considerably simpliﬁes the procedure of model solving.e. (eg. The g function represents the decision rules of economic agents decoded 2 from the structural form of the model. assuming zero values of structural shocks ε t = 0: ¯ = g( ¯ . y y (2. ε t+1 ). Solving a DSGE model is a difﬁcult task in general. The g function is called a reduced form of a DSGE model. the values of the increments of variables d y t are interpreted as the percentage deviations from the value in the steady state. steady state is also a ﬁxed point of g representation.5) y y y Since y t+1 = g(g( y t−1 . This is the socalled expansion into the ﬁrstorder Taylor series. and thus. The deterministic steady state is deﬁned as each ¯ point satisfying: f ( ¯ .e. 32 32 N a t i o n a l B a n k o f P o l a n d . however.4) In other words. ¯ . steady state is a state for which selfrepeating shall be optimal. does not lead to determination of the g function but a linear approximation. determines the dynamics of the model. 2002). ε t = 0. its linear approximation is being considered. If the model is loglinearised in the neighbourhood of a steady state. 0) = 0. The expansion of model (2. 0) = 0. due to the nonlinear form of the f function and absence of adequately general analytical or numerical methods.e. from the values consistent with the longterm equilibrium. y t−1 . in this chapter identiﬁed as ¯ . Determining the g function is equivalent to solving the DSGE model. describes the laws of motion for all of the variables of the model.1) is expanded into ﬁrstorder Taylor series in the neighbourhood of a deterministic y y steady state.1). i. its solution. i. y y y (2. Birkholc. ¯ .
the solution of the model boils down to the determination of matrix A from equation 2. Determination of A and B matrices satisfying equation (2.9. A. (2. The last equation indicates that the reduced form of a DSGE model has the representation of a VAR model. with accuracy to linear approximation.6) dε t = 0.DSGE model — anatomy series in the neighbourhood of steady state ¯ has the form of: y ∂f t ∂ y t+1 ∂f ∂ y t+1 ( ¯ )[ y ( ¯) y ∂g ∂y ( ¯ )]2 + y ∂g ∂ε ∂f ∂ yt ( ¯) y ∂f ∂ yt ∂g ∂y ( ¯) + y ∂g ∂ε ∂f ∂ y t−1 ( ¯) y ( ¯) y d y t−1 + (2.9) for every t. A2 . ∂ y ( ¯ ). equivalently: d y t ≈ Ad y t−1 + Bdε t . i. The solution of linear approximation of a DSGE model or.7).e.8) for every t and: t (A0 AB + A1 B + S)du t = 0 (2.7) System 2. e. ε t ) ≈ ¯ + Ad y t−1 + Bdε t y or.7 is a stochastic matrix difference equation with unknown matrices A and B. As equation (2. B and (A0 A + A1 A + A2 )d y t−1 + (A0 AB + A1 B + S)dε t = 0. ∂ y y y ∂ y t+1 t ∂f ∂g ( ¯ ). equivalently.7). A1 . both addends of a sum under the sign of the expected value must be equal to zero. the deviation is interpreted as percentage deviation. And the differential d y t represents the deviation of state y t from the steady state and. it is required that: t (A0 A2 + A1 A + A2 )d y t−1 = 0 (2. Special numerical algorithms are applied for that purpose. the linear approximation 33 WORKING PAPER No.10) Thus. the linear approximation of the model takes the form of: t ∂f ∂g ∂f ( ¯ ). y ( ¯) y ( ¯) + y ∂f ∂ε Denoting matrices S respectively. i.7) must be satisﬁed for every d y t and dε t . The former equation is a stochastic matrix quadratic equation with respect to A. and the latter — if only A has been determined — shall enable to calculate B from the condition: B = −(A0 A + A0 )−1 S. ∂g ∂y ( ¯) y ( ¯) + y ( ¯ ). if the variables of a model are expressed in the form of logarithms. 2 2 (2. 83 33 . because up to the ﬁrstorder Taylor y series expansion around ¯ deterministic steady state reads y t = g( y t−1 . ∂ ε ( ¯ ) y y ∂ y t−1 and ∂f ( ¯) y ∂ε by A0 . enables the determination of linear approximation of policy function g.e. solution of equation (2.g the basic BlanchardKahn method or the AndersonMoore algorithm.
model may be considered a family of From the statistical — classical — point of view. 34 34 N a t i o n a l B a n k o f P o l a n d .DSGE model — anatomy of its reduced form3 : y t = Ay t−1 + Bε t is represented in the socalled statespace: Yt = H y t + u t . shocks and disturbances from the distribution p(Y. the representation of model (2. and in particular they may include trends. we assume that the variables of the model are expressed in the form of percentage deviations from the steady state. y t = Ay t−1 + Bε t . y.12) and conditional distributions of observables with regard to nonobservable data. ε. the states. the already known linear approximation of the model solution.e. uθ ) 3 From now on. (2. upon integration of states. Alternatively. The former equation is called the measurement equation and presents the relation between observations. (2.13) The form. The elements of matrices A. and endogenous variables y t of the model.e. ε. which shall be identiﬁed with θ . shocks. i. Random variable u t is called a measurement error or disturbance. Ω is the space of measurement errors u. which in the context of model representation in the statespace are called state variables or simply states. but also from a classical point of view. u. B and H are nonlinear functions of model parameters. ε. θ ∈ Θ . Representation of the reduced form of a DSGE model in the state space (2. They may also depend on the values of states in steady state4 . the model represents the family of conditional probability distributions of its variables conditional upon the θ parameters: p(Y. namely observable variables Yt . R) oraz ε t ∼ N (0. to the end of this chapter. i. measurement errors and parameters: p(Y  y. 4 For the simplicity of notation. functions of exogenous variables. I). (2.11) 2 and: u t ∼ N (0. Υ is the state space y. ε ∈ Ξ. θ ).e. with regard to where Ξ is the space of shocks ε. y ∈ Υ . Formally. Hereinafter we shall call it a solution or reduced form. θ ∈ Θ . and Θ is the parameters space. i. uθ ). θ ∈ Θ ⊂ Rn for some n. The latter equation is called transition equation and describes the dynamics of state variables y t .11) makes the observations dependent only on states. This equation is a linear approximation of the reduced form. Observations may be functions of endogenous variables y t .11) together with the assumptions of ε t shocks distribution and u t observation errors shall be abbreviated with the symbol called the model. of the model. solution. u ∈ Ω. y.
This may be done at least in two ways — the values of parameters may be calibrated or estimated. the most often used is the maximum likelihood method. i.15) 2 Upon integration of states.14) which is the central object for the estimation of the model parameters. i. the parameters of the model shall be assigned numerical values.1 Values of parameters In order to be able to solve a model and then carry out analyses based thereon. shocks and disturbances with regard to parameters. The kernel of posterior distribution of parameters is the central object of Bayesian estimation of parameters. states. i. A1 .e.DSGE model — anatomy leads to the likelihood function L(θ Y ): L(θ Y ) = p(Y θ ) = p(Y. disturbances and parameters. estimate the values of shocks in a sample. Among the methods of estimation. y. may next serve various simulation experiments. ε. 2. ε. which gives rise to factorisation into the conditional distribution of observations.e. prior assignment of numerical values to its parameters. uθ )d y dεdu. we receive the socalled kernel of posterior distribution (posterior kernel) (θ Y ) = (θ Y ): (2. ε. This initially resulted from the fact that the estimation of DSGE models with the maximum likelihood method with the use of deterministic optimisation algorithms results in numerical problems that prevent the satisfactory use of the method. uθ )p(θ )d y dεdu = p(Y θ )p(θ ) being the product of the likelihood function of parameters and density of the probability of their prior distribution. 83 35 . A2 i S. forecast endogenous variables and their functions — the socalled observable variables.16) Ω Ξ Υ p(Y. θ ) = p(Y. calibration and estimation are applied in a complementary manner. states. y. u. Currently. both positive and counterfactual. which deﬁne the linear approximation of its structural form. the elements of the matrices — parameters of the model — shall be assigned numerical values. shocks and disturbances from the joint distribution. the Bayesian approach becomes more and more popular. In order to solve the model. method of moments and Bayesian estimation. A solved DSGE model. shocks. its reduced form. uθ )p(θ ). The process of solving a DSGE model requires calculations made on matrices A0 . Nevertheless. ε. From the statistical — Bayesian — point of view the model may be considered joint distribu tion of observations. This may be done in many ways. (2. y. the Bayesian approach to DSGE models estimation becomes more common 35 WORKING PAPER No.e. Ω Ξ Υ (2. and unconditional distribution — the socalled prior distribution — of parameters: p(Y. Currently. some of the parameters of the model are calibrated and some of them are estimated. In practice. all the experiments require the solution of the model. It is also possible to quantify the response of economy to structural shocks. y.
Y ) = p(Y θ ) originates as model. where From the operational point of view. In practice. ε shocks and u disturbances from the conditional density p(Y. ( )) = p(Y θ . such values of DSGE model parameters may be assumed as to make its dynamics (response of economy to structural shocks) and longterm equilibrium correspond with economic intuition. Yt } denotes the information set in t period.1. It entails the determination of the values of θ . From the point of view of Bayesian estimation. 2.. (θ Y. . such method of reasoning enables the formal reﬂection of that what is common in the classical approach but happens ad hoc and in an implicit manner. The elasticities speciﬁed therein may be assumed to be the one chosen for the DSGE model. In order to ( ) stands (θ Y ) function or development of a numerical method of its approximation. marginalisation of states. in the calibration process. ( )). the likelihood function is denoted by for ﬁltration made with the use of the model. the likelihood function observables Y conditional with respect to θ parameters:: a result of integration of y states. Thus. )) is measured by using an iterative T )) = p(Y0 ) t=1 p(Yt I t−1 ). where I t = {Y0 . such as VAR class models.DSGE model — anatomy than only the methods of solving technical problems. calibration boils down to the assignment to the calibrated parameters a possibly informative prior distribution. The use of such formula 36 36 N a t i o n a l B a n k o f P o l a n d . uθ ). This results mainly from the fact that the approach enables the consideration of knowledge a priori. calibration is a mixture of both techniques. in an explicit and formal manner. The method requires the determination of the (θ Y ) is a density function of conditional distribution of (θ Y ) = p(Y θ ). Technically. Formally. ε. ( ( (θ Y. the results of other studies or indication of models of more empirical purpose. Filtration is conditioned with regard to the emphasise this fact. The likelihood function may be considered a function measuring the likelihood of θ parameters in the sense that it takes higher value for such conﬁguration of parameters for which the density of probability — determined by p(Y θ ) that data Y was generated by the parameters have θ value — is larger. The likelihood function model whose (θ . y. for which the likelihood function analytical form of (θ Y ) obtains the highest value. As an example.. shocks and disturbances takes place in the process of their ﬁltration. there are empirical studies dedicated to the estimation of elasticity of demand for domestic and foreign products in various sectors of many 2 economies.1 Calibration Calibration may entail the assignment of values to the model parameters in an arbitrary manner or can be based on other research. likelihood formula: (θ Y. Y1 .. 2.1. parameters. On the other hand.2 Maximum likelihood estimation Among the classical methods of estimation the most often used is the maximum likelihood method.
( )) in one run of the Kalman ﬁlter. NewtonRaphson method requires the speciﬁcation of initial values of θ0 parameters and is a local method — i. it is vital to know expected values (Yt I t−1 ) of which are — in the case of normal distribution — identical as the analytically derived formulas for the values. mainly for (θ Y. Yet. the gradient (θk Y. Deterministic iterative optimisation methods are then ineffective and usually the obtained θK estimation is not much different from the speciﬁed initial values of θ0 .. with the use of NewtonRaphson method is time consuming. 83 37 .e. T .e. (θ Y. stochastic optimisation methods have recently become more popular. 4. For k = 0 the initial value of parameters must be assumed. i. ( )) ˙ iterative methods based on NewtonRaphson method are most often used (see Zak and Chong.. T .. Numerical determination of the ﬁrst two differentials of function i.Therefore. 2008). i.. ( )) and hessian H (θk Y. i. 2.. 5. . determination of (θk Y. . 6.e. ( )) (H (θk Y. ( )) for θ = θk . particularly simulated annealing.DSGE model — anatomy requires the determination of probability density p(Yt I t−1 ) for t = 1.e. In order to determine values of the θ parameters maximising the likelihood function (θ Y. Determination of the value of the likelihood function for θ = θk . 2. Tk 1 In order to calculate the likelihood of parameters. the equations 2 model for θ = θk .. Finally. Making an iterative step of the NewtonRaphson method. Assuming the normality of shocks and disturbances. DSGE model estimation with the maximum likelihood method. which is a numerically instable task. the density is the density of normal distribution: p(Yt I t−1 ) =(2π)− 2 ( (Yt I t−1 )− 2 × × exp − and variances 1 2 Yt − (Yt I t−1 ) ( (Yt I t−1 ))−1 Yt − (Yt I t−1 ) . 2. ( )) ) −1 . This is because NewtonRaphson method is iterative and each iteration requires the solution of a model (point 2) and running the Kalman ﬁlter in order to determine the value of the likelihood function requires numerical determination of a gradient and Hessian of the likelihood function. For that purpose Kalman ﬁlter is used. K. assumption that θk+1 = θk − (θk Y.... it 37 WORKING PAPER No. Veriﬁcation of algorithm convergence criteria. The general sequence of estimation of DSGE model parameters with the maximum likelihood method is the following: 1. the largest problem related to the application of the maximum likelihood method to estimate DSGE model parameters is insufﬁcient curvature of the likelihood function. Additionally. ( )) (point 3). namely the presence of the socalled plateau. Solution of the B= ∂g  ( ¯ ). k = 0. Hence. 2. determination of matrices A = ∂g y  ( ¯) ∂ y θ =θk and 3. . ( )) . Parameters assume the value θk . it does not guarantee that the obtained estimation θK maximises globally the likelihood function. y ∂ ε θ =θk (Yt I t−1 ) for t = 1.e. 1.
For point estimates of θ parameters one assumes. Bayesian inference fundamentally differs from the classical one. A priori assessment merged with classical likelihood of parameters. i. . 5 ). ( )) likelihood. Hence. p(Y  ( ( )) )) . for example. Formally. The analytical determination of posterior distribution is usually impossible.1. ( )). e. Maximum likelihood method poses also other difﬁculties. MetropolisHastings algorithm. Such )) is a normalising factor — a density function of data subject to ﬁltration )) by calculation is usually. where p(Y  ( ( ( )) = p(θ )p(Y θ . the maximum of posterior density function in order to which is determined in the process of ﬁltration conditionally with regard to the model procedure leads to posterior distribution: p(θ Y. it is also possible to formally include a priori knowledge. ( )) The hypothesis states that the . in the Bayesian interference the probability of the formulated hypotheses — and the statement that θ parameters of the 5 DSGE model take a speciﬁc value illustrates such hypothesis — is assessed not only on the basis of how strongly the Y data conﬁrm the hypothesis. Such subjective assessment is called a priori assessment. Therefore. while determining the maximum of posterior distribution. the Bayesian estimation task entails the determination of the posterior distribution: p(θ Y ) = p(θ )p(Y θ ) p(Y ) (θ Y.DSGE model — anatomy that reason.3 Bayesian estimation Although the initial reasons for applying Bayesian estimation to DSGE models were strictly technical. Unlike as in the classical case. so simulation approach. generated the data Y . consider uncertainty with parameters in the form of posterior distribution and make formal smallsample inference that does not refer to asymptotic properties. but also based on subjective assessment of the probability that is not related to the Y data. with the likelihood function a posteriori assessment of probability of the truth of the postulated hypothesis. Analytical determination of marginal distribution p(Y  ( ( )) = p(θ )p(Y θ . DSGE models are often estimated with Bayesian methods. results in the of parameters θ . The application of the Bayesian approach to estimate parameters equals — from the numerical point of view — the use of the maximum likelihood method with full information. 2. (2. the Bayesian approach makes the nature of statistical analysis of DSGE models other 2 than classical. which allow to increase the likelihood function curvature by virtue of a change of its shape with the socalled prior distribution. the normalising constant tends to be ignored and maximisation is applied only to the product of posterior distribution and likelihood function: (θ Y. Density p(Y θ ) is obtained based on the (θ Y. whose parameters have the θ value.model.e. is applied.17) approximate its modal value. 38 38 N a t i o n a l B a n k o f P o l a n d .g.
T . 2. 83 39 . but constitutes an exact result. 2. the approximation is a linear ﬁlter. If prior distribution is highly informative.. This may be both.. The informative nature of prior distribution increases along with the decrease of its variance. Let us assume that we have observable a mechanism processing Y data in order to determine the expected values and variances of y t endogenous variables of the model. informative distribution and uninformative distribution. 2. ( )). Y = {YT . ﬁltration and prediction may also be derived analytically from the probabilistic viewpoint. which is a sum of prior distribution logarithm ln p(θ ) and a loglikelihood function ln p(Y θ .e. as instead of maximisation of the kernel maximise its logarithm: ln (θ Y. 1. . it is considered to have a "strong prior". or under the condition of an information set I T . T − t is called a ﬁltration. ε t shocks and u t disturbances in t period. Firstly — the determination of the 39 data. Rationalisation of such conduct results from the fact that posterior distribution is proportional to its kernel: p(θ Y.. 1.. ( )). then we speak of a ﬁlter. under the condition of an informative set I t . in contrast to uninformative distribution.. then then we speak about a smoother. There are many methods of ﬁltration and prediction. while the task of determining the expected values and variances under the condition of set I t−k for k = 1. subjective assessment draws a relatively growing attention along with a drop in variance. most often used is the socalled Kalman ﬁlter. Since we operate on a linear approximation of a DSGE model.DSGE model — anatomy i. .. . Hence the ﬁlter has statistical interpretation and the likelihood function based thereon is not the approximation which would occur in the case of more general assumptions with regard to the distribution of shocks and measurement errors. 2. 2 )) = ln p(θ ) + ln p(Y θ .2 Kalman ﬁlter A DSGE model may be perceived from many perspectives. The value of likelihood function may be determined with the use of Kalman ﬁlter equations. t is called a prediction. it is possible to equivalently (θ Y. Prior distribution is assumed arbitrarily. Assuming the normality of shocks and measurement errors of the model. It may be perceived as a system of conditions deﬁning the optimal behaviour of economic agents or as a stochastic matrix difference equation. The task of determination of the expected values and variances under the condition of set I t+k for k = 0. In the context of DSGE models... YT −1 . It may also be viewed as the socalled ﬁlter. i. ( )) ∝ (θ Y. ( )). Determination of posterior kernel requires — as in the case of the maximum likelihood method — the determination of the likelihood function logarithm.. namely observations from periods t = 0.. . The effect of the ﬁlter application has several dimensions. The ﬁlter is WORKING PAPER No. Yt observations.. ( ( )).e. ( (θ Y. the socalled posterior kernel. Y0 }. Obtained derivations coincide with Kalman ﬁlter equations.. The equations of the Kalman ﬁlter may be derived from the optimisation calculus by minimisation of oneperiod forecasterrors of observable variables of the model. and strictly speaking — its )).
so its analytical form is known. the Kalman ﬁlter is optimal in the class of linear ﬁlters.21) and (ε t I t ) (2.20) of structural shocks. Thirdly — determination of the enables one to determine the value of the likelihood function method and Bayesian estimation.. 40 (2. In consequence of the assumptions regarding the normality of structural shocks. provided that it is actually fulﬁlled.DSGE model — anatomy expected values and variances: ( y t I t ) or: ( y t I T ) and ( y t I T ) (2. an analytical form of the iterative likelihood function of Y observations is known. ( y t I t ) of states for t = 1. is a density function of a multidimensional normal distribution. conditional with regard to information set of t − 1 period.e. which is equivalent to their identiﬁcation. This assumption. i. shocks and observations.19) and ( y t I t ) (2.23) 40 N a t i o n a l B a n k o f P o l a n d . however. there are more effective nonlinear ﬁlters in the aforementioned sense.18) 2 of endogenous variables. Since it is a multidimensional normal distribution. and (Yt I t−1 ) (θ Y. If the assumption of normality is not fulﬁlled. has some advantages. . the determination of the distribution of Yt observations in each t period. therefore... do not come within set Y . (2. Secondly — determination of the expected values and variances: (ε t I t ) or: (ε t I T ) expected values and variances: (Yt I t−1 ) the value of (θ Y. the Kalman ﬁlter includes the following equations. 2.is an optimal ﬁlter with regard to minimisation of variances of the estimations of states. Kalman ﬁlter enables. Determination of and (ε t I T ) (2. ( y t I t ) = A ( y t−1 I t ) A + B (ε t I t ) B . the density function of Yt — conditional upon I t−1 . which enables one to know the estimated values of those among them which are not perfectly observable. It appears that Kalman ﬁlter — if shocks reﬂect normal distribution . For the initial values of ( y0 ) and ( y0 ).22) ) for θ . T are determined The expected values ( y t I t ) and variances recurrently based on the following relations: ( y t I t ) = A ( y t−1 I t ) + B (ε t I t ). ) is an element of estimation both according to the maximum likelihood Deriving the Kalman ﬁlter that has statistical interpretation requires an assumption that structural shocks of the model have normal distribution.
The distribution is not normal with regard to θ parameters (it is normal with regard to nonlinear functions of θ parameters). M . 83 41 .. the subsequence of which the p(θ Y. ( rameters. p(θk−1 Y. the shape of the )) function is approximated with simulation methods. ( )) .g.24) determined recursively. Yt I t−1 ). Yt I t−1 ) (Yt I t−1 )−1 (Yt − (Yt I t−1 )). r). (2. (θ Y. Draw θ from the normal distribution N (θk−1 . Yt I t−1 ). the modal value of p(θ Y. . Yt I t−1 ) (Yt I t−1 )−1 cov( y t−1 . .. (θ Y. assuming the normality of structural shocks. ( )) where acceptance threshold r is speciﬁed by: r = 3. ( ( (θ Y.. ( )) kernel of )) posterior distribution. (2. also the moments of parameter estimates may be determined. Formally. Based on the members of the Θ set. Thus. 2. Yt I t−1 ) (Yt I t−1 ) cov(ε t . distribution. The approximation of posterior distribution p(θ Y. or. k = 1. θ1 . otherwise. (2.. Let θ0 be an initial value of parameters. θk−1 .. (Yt I t−1 ) = H A ( y t−1 I t−1 ) + A x x t ... ( )).3 MetropolisHastings algorithm ( )) of pa Bayesian estimation entails the determination of posterior distribution p(θ Y. The kth. Yt I t−1 ) (Yt I t−1 )−1 (Yt − (Yt I t−1 )).. The expected values (ε t I t ) and variances relationships: (ε t I t ) = cov(ε t−1 . 2. T are (ε t I t ) = Ψ − cov(ε t .. eral case.DSGE model — anatomy while assuming the normality of structural shocks.. γΣ) with the expected value θk−1 and variance γΣ. and let us assume that Θ = . Assume: Θ = Θ ∪ {θk }..25) 2 The expected values (Yt I t−1 ) and variances determined recursively based on the following relationships: (Yt I t−1 ) of observations for t = 1.. the Θ sequence is a Markov chain. θ M originates — for an adequately large m — from unknown posterior distribution )). The shape of the density function p(θ Y.. θ M }.26) 2. 2. For that purpose sampling methods are used. equivalently. The result of the operation of the algorithm is a sequence of parameters Θ = {θ0 . θm . step of MetropolisHastings algorithm runs as follows: 1. 2. ( )) WORKING PAPER No. based on the following (ε t I t ) of structural shocks for t = 1.. T are (Yt I t−1 ) = H A ( y t−1 I t−1 )A H + R. p(θ Y. 41 = p(θ Y. . let Σ be a numerically determined second differential of )) in θ0 .. whose stationary distribution is ( ( )) involves the determination of a histogram of the elements of the Θ set. θm+1 . (θ Y. e. most often the MetropolisHastings algorithm. Assume θk = θ. its analytical form is not known in the gen( )). −1 ( y t−1 I t ) = ( y t−1 I t−1 ) − cov( y t−1 . and let γ be a positive constant. the following applies: ( y t−1 I t ) = ( y t−1 I t−1 ) + cov( y t−1 . with probability min(1. ( )) (θk−1 Y. . .
At the end of the day a single speciﬁcation is used6 . ( )) (θ Y. which enable the determination of Θ chain convergence. One way to approximate (2. where W and B are estimates of within and between chain variances of this parameter when length of each chain equals n.1 – 1. thus. as it requires that chain Θ be convergent to stationary distribution. There are analytical tools. i. However. The Gelman and Rubin diagnostic suggests that for.28) p(Y  ( θ ∈Θ θ ∈Θ is a marginal density of data Y provided that model is used.e. The Geweke test simply applies the difference of means test to two overlapping parts of the chain to check if the two parts of the chain come from the distribution with the same mean. is not credible. the posterior mode) and its 42 42 N a t i o n a l B a n k o f P o l a n d .29) )) = (2π) 2 n ˆ (Σθ )p(θ Y. ( ))dθ (2.27) for 1. the ﬁlter is run in each iteration of the MetropolisHastings algorithm.4 Model selection In the process of model building one usually works with alternative speciﬁcations or alternative calibrations of the same model. the shrink or scale reduction factor: r= (1 − 1 )W + 1 B n n W should not be too hign (perhaps lower than 1. one could in principle work with many models using methods of bayesian averaging for policy experiments like stochastic simulations or forecasting. In case of normal distribution this method is known as Laplace approximation and the estimate of p(Y  ( p(Y  ( ˆ where θ and Σθ ˆ covariance. approximation of posterior distribution p(θ Y.28) is to assume a functional form of the posterior kernel. This integral in most cases is a difﬁcult one and has to approximated. each parameters.2). in order to reduce the modele selection risk. The MetropolisHastings method is highly timeconsuming. 6 )) is: ( ˆ ))p(θ  ( )) (2. Y  ( ))dθ = p(θ )p(Y θ . convergence diagnostics.DSGE model — anatomy Calculation of the value of ﬁlter. ˆ denote estimate of θ using model (e. 2. which happens for very large M values from the practical point of view. To select a single model from a family of models we may want to compare them in a pairwise manner by ratios of their posterior distributions (the so called posterior odds ratio): p( 1 )Y 2 Y = p( p( 1 )p(Y  ( 1 )) 2 )) 2 )p(Y  ( (2.g. ( )) (point 2) requires making one run of the Kalman 2 To verify convergence of Θ we may apply Gelman and Rubin diagnostic or the Geweke diagnostic. Without convergence veriﬁcation. 2 ∈ where: )) = p(θ .
has a structure of a VAR model in reduced form: y t = Ay t−1 + Bε t Based on Y data all the simulation exercises that have been developed for VAR class models may be done for a DSGE model. while in the periods t + 1.5 Applications With a reduced form of a DSGE model. values of y t .5. T .... Let us shortly discuss them in the following paragraphs. t + 2. In this context. ∂g ∂ε ( ¯ ) oraz B = y ( ¯) y Presentation of the response of endogenous variables to a shock which takes place in period t depending on index k ≥ 1 is called an impulse response function (shortly IRF). 43 WORKING PAPER No.e. 83 43 . 2. All the exercises are based on the property that the reduced form of a DSGE model. t + 1. it is possible to do several standard exercises: to identify structural shocks. It should be supplemented by veriﬁcation of models’ in sample and... i. . i. the analysis aims to quantify the responses of endogenous variables in periods t.. at the same time.e. (ε t I t ). to an impulse from a structural shock ε t in period t. 2.2 Impulse response analysis Analysis of response functions (or impulse response analysis) replies to the question of what happens in the economy after the occurrence of a structural shock. More precisely..... this may be 2. carry out historical decompositions and forecasts. occurs. perform variance decomposition of endogenous variables.1 Structural shocks identiﬁcation The basic exercise that may be done on basis of the reduced form DSGE model is identiﬁcation of structural disturbances ε t in a sample... if possible.. in y t period a structural shock ε t . for periods 0. with the use of the following relationship: ∂ y t+k ∂ εt where: A= ∂g ∂y = Ak Bε t . ε t+1 = ε t+2 = . y t+1 . i.e.DSGE model — anatomy Comparison of posterior odds ratios is a technical procedure. t + k no structural shocks occur. analyse responses of endogenous (observable) variables to structural shocks. . = ε t+k = 0. its solution.e. 1. i. 2 2. . The dynamic reaction of the economy may be determined on the basis of the reduced form of a DSGE model. As it has been said.. It is assumed. out of sample forecasting power as well as by qualitative inspection of models’ theoretical structure. solution or the reduced form of a DSGE model may be considered not only as a VAR model but also as a linear ﬁlter. Shocks identiﬁcation consists in derivation of expected values (ε t I t ) and variances done in the course of one run of the Kalman ﬁlter. that before t period the economy stays in the longterm equilibrium ¯ .5..
it may be seen that the necessary and sufﬁcient condition for stability of the longterm equilibrium is convergence of matrix Ak along with k to a zero matrix: lim Ak = 0. Thus.. since d y t is known (estimated). so it is permitted that ˜ = ¯ .e. 2. the recursive y forecast error may be calculated in period t + h from: h−1 y ∆h = y t+h − ˆt+h = Ak Bε t+h k=0 44 44 N a t i o n a l B a n k o f P o l a n d . Thus. 2. Shocks with exploding effects are excluded for lack of economic interpretation. it y y returns to a steady state ˜ . one assumes d ˆt = d y t . y Nevertheless. it shall asymptotically y 2 return to the same steady state ¯ ..e. If a DSGE model speciﬁcation includes such shocks. all the shocks of the model are of transitory or stationary nature. If this is the case. the condition of stability is not fulﬁlled. As the reduced form of a DSGE model has the form of an autoregressive equation: y t = Ay t−1 + Bε t the iterative or mechanical point forecasts of the values of endogenous variables y t+h for h = 1. i. different than the one in which it was before the occurrence of the y y shock.5. This condition ensures that the response of the model to any type of structural shock — with stationary or nonstationary effects — shall not be exploding. read: ˆt+h = Aˆt+h−1 y y and in period t. or longterm. that  ¯ − ˜  < ∞. The condition is satisﬁed if and only if: y y k→∞ ˜ lim Ak = A ˜ for a some matrix A with ﬁnite elements.3 Variance decomposition Variance decomposition of endogenous variables answers the question which structural shocks have the largest importance for the dynamics of given variables. k→∞ The condition is satisﬁed if and only if all the eigenvalues of matrix A are smaller in modulus than 1. it is required that upon the occurrence of a shock y with longterm effect. i.DSGE model — anatomy Based on the above relationship. nonstationary effects. . Therefore. some of the structural shocks do not need to be transitory — they may have permanent. It is replaced with a more general condition. Economy being in a steady state ¯ . H. should it be subjected to shock ε t .. the model variables move away from the initial steady state ¯ by a ﬁnite value at the most. if and only if the all the eigenvalues of matrix A are not larger than 1 in modulus. Should the economy in a steady state ¯ be subjected to a shock ε t with permanent effects. it is possible to determine which shocks and to what extent determine the dynamics of the particular variables.
YT +2 . usually. i.. distribution is given by: p T (YT +1 . T + h. YT +h observations. If the forecast variables are observable variables. By identifying the i j element of matrix Ak B(B) (Ak ) as pikj ..30) where: p T (YT +1 . while the determination of uncertainty related to the forecast based on the central path entails the determination of the distribution of observable variables in the forecast horizon. the determination of the modal value of predictive posterior distribution. we may be interested in quantiﬁcation of uncertainty related to the determined central path. .... we are interested in calculay T +1 . i. The distribution is called a predictive distribution. . YT +h θ )p T (θ )dθ .e. .. T + 2. Below we present an algorithm simulating the predictive distribution of observable variables and endogenous variables of the . simulation methods are applied. and p T (θ ) = p(θ I T ) is a poste involves.. where I T = Y . y T +2 . . θ ∈ 1. YT +2 .. Draw states y T from normal distribution with mean ( y T I T ) and variance ( y T I T ).. .. Determination of the central path of observable variables ). and endogenous variables. y T ∼ N ( y T I T )... The density is a multidimensional integral with a large support and may not be calculated analytically. YT +2 . We shall identify the elements of matrix The diagonal elements h dii (∆h ) by dihj .. Let us assume that y+ = and Y+ = . YT +h ).e. Draw parameters θ parameters from the posterior distribution of the model p(θ Y ).5. (∆h ) = [dihj ]. ( y T I T ) . YT +1 . . we may be interested in determination of the socalled central forecast path.. we get that the contribution of the jth structural shock to the variance of an ith endogenous variable of the model in the horizon of the forecast h amounts to: σh ( j) i = h−1 k 2 k=0 (pi j ) h dii (∆h ) are variances of forecast errors of endogenous 2 . The forecasting process may refer both to observable variables. we shall assume that of matrix variables in the forecast horizon h. Firstly. YT +2 . y T +h states.e. Therefore.. distribution with the density p T (YT +1 . 45 WORKING PAPER No.e. YT +h I T . 2.. 83 45 .. i. ie.... rior distribution of θ . i.. i. which usually corresponds to the expected value of forecast variables or the modal value of their predictive distribution. (2. YT +h ) = Θ p(YT +1 . . . YT +2 . YT +2 . namely the determination of the distribution of forecast variables in the forecast horizon. ie. 2. .e.DSGE model — anatomy as well as the covariance matrix: h−1 (∆h ) = k=0 Ak BB (Ak ) .4 Unconditional forecasts The forecasting process may be of at least dual nature. Additionally.... Let us assume that the forecast horizon is h ≥ 1 periods. their predictive tion of forecasts for periods T + 1. YT +h ) = p(YT +1 .
. u T +2 .DSGE model — anatomy 3.. ie.. . Using the sequence (ε) and equation of motion of states generate a respective has a normal distribution with expected value equal to zero and Ψ. for a large number of times leads to sets y+ and Y+ .g. The set y+ includes realizations of forecast paths of unknown values of endogenous variables in periods t + 1.. ( y T I T ) have been determined with the use 2 Repeating steps 15. ε T +2 . current state y T may be eliminated by assuming y T = ( y T I T ) in step two. distributions of states and observations entails the generation of histograms from the obtained The uncertainty resulting from the variance of measurement errors may be eliminated by assuming (u) = (0. Draw a sequence of observation errors (u) = (u T +1 . . by calculating the variance of forecasts). 0) in step three. namely: 1.. 3. Using (u) and measurement equation generate a respective sequence of has a normal distribution with expected value equal to zero and variance R. 2.... Ψ). ε T +h ). R).. ... These paths come from the predictive distribution of endogenous variables. h.. and: 4...... if parameters have been calibrated. . in which every element i = 1. The approximation of predictive samples of y+ and Y+ respectively.. observable variable (Y ) = (YT +1 . 4. by determining the modal values of sets y+ and Y+ ) and quantify the uncertainties related to them (e. The paths come from the predictive distribution of observable forecasts (e. YT +h ).t + h. 0. variance of the estimator of parameters θ 7 .. The set Y+ includes the realizations of forecast paths of unknown values of observable variables variables.. t + 2. y T +h ). u T +h ). By inspecting the statistical values of sets y+ and Y+ we may learn the central paths of in periods t + 1. ε T +i ∼ N(0. variance of the estimator of the current state y T .t + h. 46 46 N a t i o n a l B a n k o f P o l a n d . . Uncertainty of forecast can therefore be modelled by virtue of consideration of the four sources of uncertainty. ie. h. in which every element i = 1. Distribution p(θ Y ) in point 1 has been determined with the use of MetropolisHastings algorithm.g.. 5.. YT +2 . y T +2 ... The expected value ( y T I T ) and variance of the Kalman ﬁlter.. sequence of state variables ( y) = ( y T +1 . .. . The uncertainty 7 The uncertainty shall not be present. Assume y+ = y+ ∪ ( y) and Y+ = Y+ ∪ (Y ). 2... 2. u T +i ∼ N(0.. structural shocks variance. measurement errors variance. t + 2. The uncertainty resulting from the estimation of the resulting from the estimation of parameters θ may be eliminated by assuming in step one that θ is always the mode of posterior distribution p(θ Y ). Draw a sequence of structural shocks (ε) = (ε T +1 .
Part II Speciﬁcation of DSGE SOEPL−2009 model 47 .
Eichenbaum and Ch.). 2004) may be noticed.g. 2007a) — hereinafter the model shall be referred to as SOEEuro . see Warne (2009). and also (partially) the authors of the Dynare package (M. The essence of the methods applied by the ECB may be reconstructed by virtue of analysis of the construction of the YADA package (a collection of scripts of the Matlab software). as well as Bayesian estimation1 . which enables its application to the description of the Polish economy..d) and Christiano et al. Adolfson et al. S. Evans (Christiano et al. Juillard. They used the SOEEuro elements to construct a model describing the economy of Sweden — this is how the RAMSES DSGE model of the Riksbank (see Adolfson et al.. 2007b) was created. see (por. Smets and R. 2005b. More complete references are provided at the end of the paper. Christiano et al. the experience of Riksbank has been used by the analysts of the European Central Bank building the NAWM model for the purposes of the ECB (see. They also use the broader understood methods of constructing and applying DSGE models. 2007a). Schorfheide (2000). The software we use is a modiﬁcation (usually very farreaching) of the scripts prepared at Riksbank for the purposes of SOEEuro . The family of SOE models refer to the economic ideas of the SOEEuro model by Riksbank. 2001. (2004a). Wouters. from which the family of DSGE SOEPL models derives. and in particular Altig et al. 2003.b). Christiano. Adjemian et al.. the aforementioned F..SOEPL2009 — general outline SOE Chapter 3 PL–2009 3 – general outline SOEPL−2009 — general outline 3 3. Christiano. Christoffel et al. (2007a. 2005).1 SOEEuro model — prototype of SOEPL family models The family of SOEPL models originates directly from the estimated DSGE model of the euro area developed by the analysts of the Central Bank of Sweden (Sveriges Riksbank). An important source of ideas we used when modifying the initial speciﬁcation of the models were the subsequent works by L. e. The Riksbank’s DSGE model for the euro area uses the pattern of a small open economy. Also the inﬂuence of the model by Smets and Wouters (2002. M. 48 48 N a t i o n a l B a n k o f P o l a n d . The Swedish analysts acted similarly. During the several years’ work 1 The authors of the SOEEuro model used the work and experience of other researchers. On the other hand. (2007c. The three models deﬁne the line or school of constructing DSGE models. The SOEEuro model was based on the ideas included in the model by L.
and capital adjustment costs. GDP foreign trade turnover.g. nonstationary disturbance (a stochastic trend) interpreted as the trend of technical progress (see also Altig et al. Hence. consumption. The pricesetting mechanism is related to the appearance of nominal rigidities (delays in adjustment of prices to market conditions) — a phenomenon with which the new Keynesian school explains the effectiveness of macroeconomic (monetary) policy in a short run. expressed in the state space representation. variable capital utilisation. following the hints included in the literature of the socalled new open economy macroeconomics. at least partially. Another characteristic of the new Keynesian school are real rigidities (an idea derived from the Real Business Cycle school (RBC)).SOEPL2009 — general outline and repeated reconstruction of the model and of the computation procedures. Lindé and his team have supplemented the CEE model speciﬁcation with issues related to international exchange. In our opinion. then. a model was created in which the optimising (rational) households maximise the utility originating (among others) from the consumption of products manufactured from domestic and imported components. 1983)) cause also that the exchange rate passthrough is incomplete. 2008. we enable the growth of all of the variables indicating the trend (here e. investments. Lane (1999). As we have mentioned before.g. In the SOEEuro model it has been assumed that there exists consumption habit persistence. which enables — by the application of the Kalman ﬁlter — to determine the value of the likelihood function and further on to apply the formalised (classical or Bayesian) techniques of parameters estimation. (to which we shall hereinafter refer as CEE model) — a model representing the newKeynesian point of view with regard to the economic processes (see Gali. The characteristic feature of the SOEEuro model was also the use of a single.e. 2003). of the growth (i. 2004b. Woodford. 2005). It must be emphasised in that context that it is vital to construct a block of measurement equations in the state space model (equation approximating the relations of the variables of a theoretical model with observable variables). the emphasis put on that aspect distinguishes the methods applied by Riksbank (and further by the ECB and by us) from the techniques applied e.g. An inﬁnite number of specialised agents manufacture domestic products and import consumption and investment goods. 1985) and. the logic of work with a DSGE model has remained unchanged — upon loglinearisation of equations. The rigidities of the prices of imported and exported products (in the SOEEuro approximated with Calvo model (Calvo.. which along with the stochastic nature of the technical progress explain the business cycle. As a result. the starting point for the formation of the SOEEuro model by Riksbank was the DSGE model of a closed economy by Christiano et al. J. 83 49 . characteristics of nonstationary disturbance) are elements of the model speciﬁcation. while the characteristics . to exceed the shortterm analysis horizon characteristic of business cycle models — the SOEEuro model had also the potential to explain 3 49 WORKING PAPER No. it is possible. the SOEPL model is solved numerically (brought to the reduced form) with the AndersonMoore algorithm (Anderson and Moore. see e. by a large number of Dynare package users. real wages). The specialisation of manufacturers enables them to set prices in a manner characteristic of imperfect competition. By including nonstationary disturbance in the speciﬁcation of a general equilibrium model.
imported consumption goods. In 2008 wider studies were conducted of the problems related to Poland’s accession to the euro area. Experiments with stochastic wage markup were the consequence of problems with interpretation of the labour supply disturbance identiﬁed in the sample. There were attempts to estimate the model (with Bayesian techniques) on the Polish data. extended risk premium — RAMSES). Whereas some of the experiments implemented ideas present in other models (e. the implemented alternative of working capital (in stochastic version. the construction of premium for foreign exchange risk and the construction of the tax system (more precisely the role of taxes and national insurance contributions in the process of generation of the manufacturing costs). Another distinguishing feature of the SOEEuro model was a large number of shocks (disturbances) — larger than in other then constructed models — including structural shocks: several technological shocks (stationary. labour supply.2 Family of SOEPL models.g. In the group of solutions to which the authors of the SOEEuro model attached greater significance was the socalled working capital channel. nonstationary. as well as observable disturbances (ﬁscal and originating from the world’s economy — derived from separately estimated SVAR models). when the share of payments made in advance is subject to stochastic disturbance) proved to be hardly useful in most of the applications. and the assessment is more reliable as the parameters of the model are (largely) estimated. that technical progress has an exogenous nature. The authors of the model have emphasised that thanks to the above. contributions). ﬁnanced from a loan. as well as stochastic wage markup. what translates into direct. the solution determining the demand of manufacturing ﬁrms for money (partial payment of wages in advance. stationary investmentspeciﬁc). the demand for money by households results — traditionally — from the utility provided by cash to the households. Structural changes were taken into account. A special version of a DSGE SOEPL created for the purposes of that study (see Grabek and Kłos.SOEPL2009 — general outline midterm trends2 . the interest rate rule. markup shocks (domestic products. cash demand). 2009) allowed (among others) to compare the method of absorption of disturbances in a We shall remember. The reason was also a relatively small role of monetary aggregates in the policy of central banks. etc. therefore. the original version of the SOEEuro model was subject to a series of experiments conducted at the National Bank of Poland (NBP). the solution was reduced to deterministic version or even marginalised. that the addition of the ﬁnancial sector to DSGE models would restore importance of this solution. preference shocks (consumption. exported products). positive inﬂuence of the interest rate (the cost of working capital loan) on the marginal cost of production of intermediate products and inﬂation). 2 50 50 N a t i o n a l B a n k o f P o l a n d . so the conclusions regarding a severalyear’s horizon shall be very cautious. It seems to us now. other attempted to better adjust the model to institutional framework of the Polish economy (taxes. the assessment of a relative role of the disturbance in shaping business cycles is possible. However. imported investment goods. We made experiments with various collections of observable variables. the ﬁscal rule. 3 3. however. SOEPL−2009 version In the recent years.
when the quality of forecasts would force the resignation from logic or coherence — the economic contents of the model — priority was given to the economic contents. the analyses showed that forecasts of the dynamics of investment expenditures and investment deﬂator were highly imprecise. it did not justify such enormous errors. due to limited resources (time. however. the criterion of quality (accuracy) of forecasts has not dominated our choices. An example is the labour market in which (as in SOEPL . At least some of the problems shall be the subject of our work in future but it is worth mentioning in advance that the growing size of the model is a natural barrier and it may not be expected that one model shall answer all the questions.g. The above experience. In the current economic situation these could increase the chances for the model to better explain the ongoing processes and. 83 51 . a dynamic stochastic general equilibrium model to be used for forecasting purposes and not a ”forecasting model”. according to the logic of DixitStiglitz aggregator (Dixit and Stiglitz. perfect competition in the markets of ﬁnal products and capital services. An important role in designing the changes in the speciﬁcation was also played by the analyses of accuracy of the forecasts received from the earlier. 1977). The preference for future applications is. therefore. 3 3. probably improve the accuracy of forecasts. Although it is very hard to model (and forecast) investment expenditures. In some cases the applied simpliﬁcations are disputable. and SOEPLEuro ) unemployment may not occur. therefore. Although the SOEPL−2009 version was created in consideration of forecasting applications. trends (appreciation) of foreign exchange rates. computational capacity of computers.3 Basic features of SOEPL−2009 model The speciﬁcation of SOEPL−2009 is based on a framework typical in the class of DSGE models derived from CEE model: a representative. Finally two types of goods for domestic use (consumption 51 WORKING PAPER No. On the other hand. or the issue of indebtedness of households and governments. The SOEPL−2009 model is. reﬂected in omission of interesting threads and deeper research of problems that are not directly related to forecasting — here an example may be the hypotheses related to structural changes that have probably occurred at the end of the 20th century and beginning of the 21st century in the Polish economy. imperfect competition in intermediate products and labour markets.) we could not implement in the discussed version the solutions that were the object of our former works (e. Merging of domestic and imported components into ﬁnal products is made with the use of CES function. In that version of the model. extension of the ﬁnancial sector). the world economy consisted of two areas: the monetary union and the rest of the world. experimental versions. In any case. etc. In the hereinafter presented version of the model there are also many problems that have not been solved — they have only been outlined. Final products are assembled from domestic and imported intermediate products. forwardlooking and optimising consumer. nominal and real rigidities. As an example.SOEPL2009 — general outline small open economy functioning within a monetary union and outside of it. as well as additional ideas resulting from separate research fed into the construction of the newest version of the DSGE model named SOEPL−2009 . the issue of catching up and the changes in the share of foreign trade in the GDP midterm .
Additional sources of income are proﬁts of intermediate goods producers as well as transfers from the budget. The labour and capital services are purchased in competitive market and set to minimise the cost of production. leisure and cash holdings under budget constraint. given the speciﬁc nature of manufactured goods. the dynamics of prices (and wages) is described by Phillips curves. Interest on each type of currency deposit is calculated taking into account risk premium. which is ensured by a special type of insurance levelling the income. The mechanism of prices rigidity is based on a slightly modiﬁed Calvo model.. Households are also the owners of ﬁxed capital assets and they receive income on account of lease of the assets. The technique enables merging of the pattern of a representative consumer with differentiation of qualiﬁcation of the labour force supply by households (households are different but their consumption standards. Households maximise utility attained from the consumption (with some habit persistence). as well as products to be exported. All of the income and also consumption expenditures are burdened with a set of taxes. Households are the sole administrator of labour force. 2000). It is assumed that there are inﬁnitely many rational (optimising and forwardlooking) households in the economy. different for each of the currencies. Generally. are the same). in which inﬂation expectations appear explicitly (the anticipated rate of inﬂation). Pragmatic reasons (avoidance of the further extension of the model and complication of equations) caused that the goods consumed by the government do not include imported components. 3 52 52 N a t i o n a l B a n k o f P o l a n d . the manufacturers solve a dynamic (intertemporal) decisionmaking problem. the costs of working credit and national insurance contributions paid be the employers (an additional charge on the labour costs). in which the Calvo type rigidity mechanism is assumed. The possibility of generating income from leasing the capital services causes that households are interested in increasing this resource — investments. Another source of income is interest from domestic deposits and deposits in foreign currencies. while the unique qualiﬁcations of each of the households give them a monopolistic position in the processes of wage negotiations (see Erceg et al. an inﬁnite number of domestic ﬁrms produce heterogeneous intermediate goods by using the CobbDouglas technology with homogenous capital and labour inputs. In the SOEPL−2009 model we have assumed that households may deposit their savings in domestic currency and in euro or dollar. The marginal costs of domestic production of intermediate products depend on the costs of capital and labour. A departure from the aforesaid principle refers to the goods consumed by the government — the goods consist only of the domestic component3 . Yet. A side effect of such simpliﬁcation is the understating of the steady state share of foreign trade in GDP .SOEPL2009 — general outline and investment goods) are created. 3 SOEEuro only consumer and investment goods cover the imported component. or more generally — the structure of expenditures. The prices of intermediate products (and also wages) are set under imperfect competition with rigidities in the adjustment processes. Part of the capital lease rent (use of capital services) and wage bill must be paid in advance and is ﬁnanced with a working capital loan. all the households reﬂect the same consumption pattern. while the applied markup has a stochastic nature. manufacturers may set prices of their products in a manner characteristic of monopolies. Upon the solution of the respective decisionmaking problems. Maximisation of utility takes place in the perspective of inﬁnite horizon with classical time discount. In order to set the prices of intermediate products. Next to households.
see also Leduc and Sill (2001). transforming them into heterogeneous products (e. instead of the actual decisionmaking process observed from inside of the bank. This is. We deﬁne in the model two speciﬁc agents who do not have clear objective functions and their behaviour is described with ad hoc rules. The rule is an effect of the manner in which the rational and forwardlooking agents — households and ﬁrms — perceive the behaviour of the central bank. thus.SOEPL2009 — general outline A special solution applied in SOEPL−2009 is the inclusion in the model of a disturbance representing the effects of ﬂuctuations in the prices of energy (e. The market in which the exporters sell their products is fully competitive. while the decisions regarding the prices (wages) lead to maximisation of 4 3 We have also carried out experiments with a version assuming the observable nature of such shock. The government assigns its expenditures for public (collective) consumption and lumpsum transfers to households. also on the prices of all of the ﬁnal products. The equilibrium at a micro scale results from ﬁrst order conditions for each group of optimising agents. The marginal cost of imports is the function of the price of goods in the world’s markets weighted with the geographic structure. The interest rate of the central bank follows the interest rate rule. optimising and forwardlooking households may. set by maximising proﬁts under monopolistic competition. In other words.g. in spite of that. which would mean the appearance of budget deﬁcit in periods and repayment of the debt in others. The second one is based on the direct inﬂuence of that disturbance on the marginal costs of domestically manufactured intermediate products and. investments. Importing ﬁrms purchase homogenous goods in the world’s market (the euro area and the USA). These are the government dealing with the collection of taxes and expending income and the central bank that controls the interest rate. The ﬁrst one works through the costs related to capital utilisation rate. consumers in both parts of the world pay identical prices. the prices of which are. In other words. another segment of economy with imperfect competition. maintain the level of expenditures. The disturbance has a structural nature4 and affects the economic processes through two channels. Final products (consumption. by branding). savings (and their currency structure).g.g.) bring households to the maximum of their expected utility. the level of capital utilisation (etc. investment and export goods) consist of domestic intermediate products and imported components. Decisions regarding consumption. 53 WORKING PAPER No. if they decide to avail of foreign deposits. such as it is proposed e. (2007a). the transfers may be considered negative. Theoretically. hence. Exporting ﬁrms purchase domestic and imported components and produce heterogeneous export goods. thus. The budget deﬁcit in such situation immediately reduces the disposable income but the rational. oil). by Christiano et al. set their prices and charge a markup (it is assumed that markups have a stochastic nature). the model characterises the point of view of economic agents and their perception of monetary policy. No public debt category appears here. distributes the income from taxes without creating the budget deﬁcit. The government plays only a passive role. the geographical structure of exports has no importance for the relations described in the model. The whole reﬂects the Ricardian behaviour. 83 53 . Decisions regarding the production level and proportion of production factors bring ﬁrms to minimise the costs.
(2009). (2004a. disturbances (technology. An exception are investment expenditures.) have stationary nature — it has been assumed that they shall have the nature of ﬁrstorder autoregressive process (AR(1)). we assume that the trend is a resultant of two processes (nonstationary disturbances). etc. the world’s economy is heterogeneous in SOEPL−2009 and consists of two areas: the euro area and the dollar area (the rest of the world identiﬁed with the USA). which are sort of a standard.b). a similar version is also proposed by Christiano et al. Equilibrium conditions on every market depend. Thus. in a long run real investments do not have to grow at the same rate as consumption or GDP Other structural . it is also required that all markets are clear.d). the total value of expenditures is equal to the total income and the state budget revenues is equal to expenditures. The key characteristics of the world’s economy and their relations are approximated with a structural vector autoregression (SVAR) model. Additionally. which guarantees that demand for services of production factors in each period are equal to the total supply of these factors. are structural unanticipated shocks. Macroscale balances are satisﬁed. in SOEPL−2009 the possibility of occurrence of disturbances anticipated by the agents has been permitted. More precisely. 54 54 N a t i o n a l B a n k o f P o l a n d . Economic growth is described with a stochastic trend — nonstationary disturbance characterising technical progress. Formally. however. The total impact of both types of technical progress gives. Burriel et al. whose dynamics of value is the same as other variables but the division into the growth of volume and price is different — accordingly to the characteristics of the second of nonstationary disturbances and the applied technology. The anticipated disturbances are. In the SOEPL−2009 model there are several groups of stochastic shocks (disturbances). thanks to which the time series reﬂecting the trend related to technical progress are modelled in consideration of that process. risk premium. however. In similar manner the fragments of the ﬁscal block have been treated — here also the SVAR model has been used for approximation of the interdependencies between budget expenditures and (a part of) budget revenues. the structure of anticipated disturbances coincides with the supplementation of the disturbance structure with MA class component — we have used here the convention proposed by SchmittGrohé and Uribe (2008). basic group. preferences. Identical for all of the growing variables is the dynamics of values and the division of the dynamics into the growth of volume and prices. (2007a. an option of which we have not availed in the current version (and. The second process impacts the prices of investment expenditures and ﬁxed assets as proposed by Christiano et al. beside of the unanticipated structural and observable disturbances. The ﬁrst directly impacts labour5 such as in the original version of SOEEuro by Riksbank. markups. Another group of disturbances are observable disturbances (also of unanticipated nature). therefore. 2005). see also Altig et al. As we have mentioned before. among others on the type of competition and the speciﬁc values of elasticity of substitution — generally. The ﬁrst. identical dynamics to all of the variables growing in steady state. 5 3 It is also proposed here to extend the interpretation of the disturbance for demographic effects. omit it in the further description). (2007c.SOEPL2009 — general outline expected proﬁts. The SVAR model is estimated separately and is used to describe the mechanisms generating the shocks (here observable disturbances) from the foreign environment.
Also the stimulation of economic activity with monetary policy instruments may be effective in a short run however. ﬁscal stimulation of the economy is effective only in a short run. and the possible deviations from the longterm trends are absorbed already in medium term. so we shall return to the issue of the role of anticipated disturbances at further stages of the research. For example. 83 55 . 3 We use interchangeably the terms of steady state and longterm equilibrium. The function of institutions managing the macroeconomic policy has been adjusted to the above logic (i. with explicitly and clearly deﬁned steady state6 . In consequence. model dynamics. while the costs of such adjustments may exceed the previous gains. agents decide to possess cash resources. The characteristics of adjustment processes. the rules of behaviours derived by the rational agents based on the observation of the activities of institutions). i. Additionally.SOEPL2009 — general outline Nevertheless. accordingly as in SOEEuro they have been excluded. while the shortterm effects — thanks to inclusion in the model of real and nominal rigidities — shall have a more Keynesian nature. the analytical potential of this solution seems to be considerable. the monetary variables do not appear in the set of observable variables. in the long run the only growth determinant is the (exogenous) technical progress.e. Coenen et al. the SOEPL−2009 speciﬁcation brings the model to neoclassical responses in the long run. In effect of optimisation of such costs. Their policy excludes any forms of game with the agents and is timeconsisted. (2006). demand for money is also reported by ﬁrms that need to pay in advance for some share of labour services and of capital7 .e. A competitive method of deriving cash models is the deﬁnition of the costs of transaction. therefore. see e. the model may be used solely for making analyses in which the condition shall be satisﬁed at least approximately. The existence and functions of money result from the households’ objective function (holding some cash resources is useful for households). As a result. Stochastic monetary effects (disturbances in cash demand among households and demand among ﬁrms) — as a result of several experiments made before the year 2009 — proved to be unsuccessful as the disturbances of this class have not increased the potential of the model and. Looking from another angle. 7 6 55 WORKING PAPER No.g. is a consequence of optimising (forwardlooking) behaviours of agents (the decisionmaking problems of agents formally derived from intertemporal optimisation problems).
µz σµz . The common technological trend (z t ) for all growing 56 56 N a t i o n a l B a n k o f P o l a n d . we assume the existence of a technological trend speciﬁc for capital/investment goods Ψ t . macroeconomic balance of the model 4 Decisionmaking problems.t = 1 − ρµΨ µΨ + ρµΨ µΨ. µz. stochastic growth which is driven by changes in the level of technology (z t ). z t−1 4 The SOEPL−2009 is a model of exogenous.t Ψt . macroeconomic balance of the model Decisionmaking problems. Ψ t−1 are governed by the process: µΨ.t = µz . whose changes.3. µΨ σµΨ .t−1 + µz . is translated into other macroeconomic categories and + extends the neutral technological trend.t ≡ µΨ.t = µΨ .1 Growth zt . equilibrium conditions. is governed by the stochastic µz .t .Chapter 4 Decisionmaking problems. and µz is a longterm growth rate of technology. where ρµz is the persistence coefﬁcient.t = 1 − ρµz µz + ρµz µz. The technological trend has a neutral nature — it refers to all of the macroeconomic categories characterised by growth. model macroeconomic balance of the 4.t µΨ .2). The growth rate of technology.t ≡ process: µz. µz. by use of capital as a factor of production (see Chapter 4. ∼ N 0. equilibrium conditions.t ∼ N 0.t−1 + µΨ . Beside of that. The presence of an additional technological trend speciﬁc for capital goods. equilibrium conditions. µΨ.
Pte is the level of prices in the euro area.t . We assume that the currency of the rest of the world is the dollar and in the euro area — the euro. e xt ≡ e S t Pte Ptc .2) Therefore. nominal wages are translated into stationary real wages: wt ≡ Wt + Ptd z t . may be presented as: + z t = z t Ψ t1− .5) 4. e u denoted respectively by: S t . S tx . as well as export goods. t (4.4) where Ptu is the level of prices in the rest of the world. therefore.2 Foreign economy Domestic economy functions in the environment of two foreign economies: the euro area and the rest of the world. (4.e. To real exchange rates. S t . xu ≡ t e S t S tx Ptu Ptc .g. 83 57 . analogically as for the nominal rates.1) Additionally. 1− µz + . e. at the ﬁnal stage the whole model may be presented with the use of stationary variables and explicitly determined steady state. (4. macroeconomic balance of the model variables.t = µz. i. where + is the share of capital in production. Interactions with those economies entail exchange of goods and ﬁnancial ﬂows. euro/zloty and dollar/euro. The level of technology for capital goods z t Ψ t + and all the other categories z t allow to express the growing variables in a stationary form (usually denoted with small letters). (4. except capital goods. it ≡ It + zt Ψt .Decisionmaking problems. investment and export goods.3) Additionally. nominal variables are stationarized with the use of the level of prices Ptd . imported consumption.3 Producers There are ﬁve markets of intermediate goods: domestic goods.1] interval) manufacturing heterogeneous intermediate 57 WORKING PAPER No.: yt ≡ Yt + zt . 4 4. we deﬁne real exchange rates: x tx ≡ S tx Ptu Pte . etc.t µΨ. (4. equilibrium conditions. three nominal exchange rates: dollar/zloty. We use. the following applies: e x u = x tx x t . and Ptc is the level of domestic consumer prices. The exchange rates satisfy: e u S t = S tx S t . In each of the markets there are inﬁnitely many agents (continuum determined in the [0.
mx. X m . equilibrium conditions.t di . macroeconomic balance of the model products of a given type that are aggregated to a homogenous ﬁnal product representing the production of the given market.1 Aggregators Heterogeneous intermediate products must be aggregated1 . imported investment goods (mi). mc. Oi. t imported consumption goods (mc). (4. we may apply economic interpretation thereto.t . They purchase heterogeneous intermediate products and transform them into a homogenous ﬁnal product (taking the prices of intermediate products and the price of the ﬁnal product as given).6) where λo is the value of markup in steady state. o ∈ {d. ∼ N 0.3. λo = λo . λ o t 1 ≤ λo < ∞.t Oi. For each market we assume the existence of inﬁnitely many ﬁrms (the agents do not consume resources or generate added value).9) 58 58 N a t i o n a l B a n k o f P o l a n d . λo is the markup in the market o. export products (x). C m . di 1−λo t .7) Proﬁt maximisation by the aggregator leads to the demand function for intermediate products of the ith producer: o Pi. t (4.8) we obtain the equation for the price of the homogenous ﬁnal product in the given market: Pto = 1 0 o Pi. Markups speciﬁc for each of the markets are described with stochastic processes: λ o = 1 − ρλ o λ o + ρ λ o λ o + t t−1 λo . and o identiﬁes market: domestic products (d).8) o where Pto is the price of the homogenous ﬁnal product in market o. λo σλo .t 1 1−λo t 1 Aggregation of heterogeneous products into a homogenous product is a technical operation necessary from the point of view of the model operability. x}.t is the price of the intermediate product of the ith producer. mi. mx. which operate under perfect competition and use the same production function. X }) takes the form of the CES function: Ot = 1 0 1 λo t 4 where Ot is the production of the ﬁnal good. (4.t = Pto Ot .Decisionmaking problems. t o ∈ {d. x}.t is the production by the ith intermediate goods producer. 4. Using equations (4. mc.t t − 1−λo t λo Oi. imported goods intended for export (mx). however. I m . (4. mi.6) and (4. The production function of the ﬁnal good in each of the markets O (O ∈ {Y. Pi. λo .
1− 1− Yi. speciﬁed explicitly only at the level of loglinearised form.t = ε t z t H i.Decisionmaking problems. 83 . τs is the rate of national insurance contribution paid by the t employer. whose costs are represented by F tτ function.t = z t ε t Ki.t −1 H i.t .t H i. fk (4.12) with respect to H i.t H i.t are the inputs of labour (hours) and capital services determined by the i th producer. (4.t H i.t . σ ε .t z t ε t Ki.g. is a linear function of structural shock representing the dynamics of the prices of energy (e.t .t and Ki. We assume also that the use of labour and capital services involves the use of “energy”.t .t Yi.t min R t F tτ Wt H i.H i.t .t . ε t = 1.14) First order conditions of the decisionmaking problem (4. we arrive at the equation of real marginal cost of the domestic intermediate goods producers: 59 59 WORKING PAPER No. must be ﬁnanced with a working capital loan hence the presence of effective gross nominal interest rates. ·) function.t 1− Ki. ν w and ν k . We assume that in each period a fraction of the wage and capital fund. ε.t . oil): πoil = 1 − ρπoil πoil + ρπoil πoil + t t−1 are: R t F tτ Wt 1 + τs = (1 − t R t F tτ Rk = t fk fw 1− − ) λ t Pi. πoil .12) 4 where Wt is the nominal wage.t and λ t 1− 1− λ t Pi.10) where H i. (4.t ∼ N 0. σ πoil . the CobbDouglas production function. given by: R t ≡ ν w R t−1 + 1 − ν w . The total factor productivity ε t is described with a stochastic process: ε t = 1 − ρε + ρε ε t−1 + costs minimisation: 1− 1− ε t Ki. Ki.t z t ε t Ki.2 Domestic intermediate goods ﬁrms The producers of domestic intermediate goods are the only actual generators of the GDP Using . (4. they use individually determined labour and capital inputs to produce: Yi.t 1 + τs + R t F tτ Rk Ki. R t fw and R t .t − z t t t fw fk ε. macroeconomic balance of the model 4. λ t is the Lagrange t multiplier.13) where R t−1 is the gross nominal interest rate.3.t .t ∼ N 0. Rk is the gross nominal rental rate per unit of capital services. Based on the ﬁrst order conditions of the problem of costs minimisation. equilibrium conditions. (4.t − λ t Pi. fw fk R t ≡ ν k R t−1 + 1 − ν k . with production technology identical for all of the producers.11) The optimal values for inputs of capital and labour are determined based on the problem of Ki.t . The F tτ (·. in the cost function. πoil .
macroeconomic balance of the model d mc t ≡ λ t = 1 εt 1 1 1− 1− r kRt t fk wt Rt fw 1− F tτ . given by (4. it sets its price to maximise the ﬂow of future proﬁts.16) κo 1 + κo βµ 1 − ξo πo − π t t−1 c 1 + κo βµρπ 1 − ξo βµ (4. already in a loglinearised but relatively legible form: πo = t + βµ 1 + κo βµ 1 + κo βµ πo − π t+1 + t+1 πt + c c ∼ N 0. Also the impact of price rigidities can be seen — the smaller the price rigidity (smaller ξo ).t In steady state inﬂation target is equal to the steady state level of inﬂation π ≡ πd . where υ t+s is the marginal utility of the households’ nominal income3 . In every period any of the manufacturers. the producer takes into account the demand for their ouput given by equation (4. assuming that it will not be allowed to reoptimise the price in the future. there are some limitations in the spirit of Calvo price setting (Calvo (1983)). the larger is the importance of marginal costs and markup in the market for the current inﬂation. The last problem the manufacturers have to cope with is the determination of the optimal level of employment (number of full time employees). described with the t exogenous process (4. . . may set the optimal price of its output Pt 1 − κd ): d Pt+1 = πd t κd d. Thus. the proﬁt in the particular period is weighted with the marginal utility of the households’ nominal income. When solving the proﬁt maximisation problem above. 60 60 N a t i o n a l B a n k o f P o l a n d .t+s . With probability ξd the price cannot be set in the optimal way and it is then indexed to previous inﬂation (with weight κd ) and the current inﬂation target2 (with weight πc t+1 1−κd Ptd . πd t t+s−1 κd πc . in which the main inﬂation determinants d become the real marginal costs mc t . which means that manufacturers produce heterogeneous products and may set their prices. equilibrium conditions.t+s − M Ci. and markup λd .17) If a producer is allowed to reoptimise its price.8). At the same time. The process of adjusting employment involves Calvotype 2 The inﬂation target πc has a stochastic nature and is given by exogenous process: t π c = 1 − ρπ c π c + ρπ c π c + t t−1 πc .new d Yi.new . (4. (4. t t Current inﬂation depends on the difference between the inﬂation target and past/expected inﬂation.t . and β is a discount factor. directly on the very inﬂation target and the current standing of producers (their markup and marginal costs). the 4 decisionmaking problem takes the form: max d. πc t+1 t+s 1−κd Pt d.new Pt ∞ t s=0 υ t+s βξd s πd .t+s Yi. 3 Due to the fact that in each period the proﬁt generated by the ﬁrm is transferred to households.18) ξo 1 + κo βµ λo + mc o .7)4 . The solution of the problem takes the form of the Phillips curve for domestic intermediate goods. πc σ c πc . . 4 Below we present an example of a Phillips curve (binding for each of the markets of intermediate products).15). πc .15) The market of domestic intermediate products is characterised with monopolistic competition. with probability 1 − ξd . based on the number of hours worked determined in the process of costs minimisation.Decisionmaking problems. (4. .
mi.4). take the form: mc mc t ≡ mi mc t ≡ mx mc t ≡ u S t Ptu 4 the importers’ proﬁt.3 Importers The imported consumption.u ) . Heterogeneous products are. under monopolistic competition. The solution of the decisionmaking problem describes the level of employment in the economy. with ﬁxed geographic structure of import.u and ωmx.u determine the share of the rest of the world in the basket of imported consumption. 1 − ωmi. then.Decisionmaking problems. and differentiate them. The total demand for imported consumption and investment goods depends on the decisions of households (see Chapter 4.19) where n the number of hours per employee. macroeconomic balance of the model rigidities — with probability 1 − ξe the producer is allowed to set the level of employment in an takes the form: min new Ei. we arrive at three Phillips curves in the form compliant with the formula Ptmc u S t Ptu Ptmi u S t Ptu Ptmx ωmc.20) where ωmc.t+s 2 . In each of the markets the importers purchase foreign goods (from the euro area and the rest of the world — we assume the stability of the geographic structure of the import). The marginal costs. equilibrium conditions. When producer is allowed to reoptimise the level of employment. while the demand for import of goods intended for export is determined by the exporters.u + e S t Pte Ptmc e S t Pte Ptmi e S t Pte Ptmx (1 − ωmc. 4. heterogeneous export goods X i.u + ωmx. described with exogenous processes (4. (4. As we assume free ﬂow of products between the euro area 61 WORKING PAPER No. the decisionmaking problem ∞ s=0 βξe s new ni Ei. while the process runs similarly to the case of domestic intermediate goods producers (with speciﬁc ξo . The monopolistic competition implies that importers may set prices of their products. 83 61 . for which they may set prices x Pi. 4. ωmi.u .3. (1 − ωmx.t .7). investment and export goods make three separate markets of imported products. Solving the problem of maximisation of presented in the footnote — full versions are provided in the Appendix. (o ∈ {mc. investment goods and goods intended for export.t optimal way.3.u ) .4 Exporters Similarly to the domestic intermediate goods producers and importers.t . while with the probability ξe the producer cannot change the level of employment.t − H i. exporters produce. κo and λo . (4. investment and export goods. Due to the fact that production is intended for the world market. purchased by the aggregators and transformed into homogenous ﬁnal products. the prices set by exporters are expressed in dollars. determine inﬂation for imported consumption goods. The real marginal costs and markups in the markets. mx}) parameters for each market of imported products).u + ωmi.u .
u + It x.e = −η f .e + It x. the demand for domestic export on the part of both economies and both types of products is expressed with the following equations: Ct x. κ x i λ x speciﬁc for the export market. and ω x is the share of the imported component. Assuming that the income of foreign economies is entirely divided between consumption and investments. the price for the euro area market is the same price converted into euro x based on the USD/EUR exchange rate Pi. u where C te C tu and I te I t are consumption and investments in the euro area (rest of the world). d m X i.t + 1 − ωx d X i.u = Ptx Ptu −η f .u C tu . The marginal cost together with the markup described with the exogenous process (4.21). we may express the demand for domestic export as a function of foreign income: X t ≡ Ct x. Each exporter must solve the problem of costs minimisation: 4 subject to (4.t = ω x m X i.7) is reﬂected in the Phillips curve for the export market. Ct x. After solving the problem of proﬁt maximisation we obtain the Phillips curve for the export market.e Yte . (4. equilibrium conditions.u Ytu + S tx Ptx Pte −η f . The production function takes the form: 1 ηx x η x x −1 ηx x 1 ηx x η x x −1 ηx x ηx x η x x −1 X i.e η f .e + Ct x. Ptd is the price of the domestic component.23) 62 62 N a t i o n a l B a n k o f P o l a n d . while η f . It x.X i. Assuming that consumption and investments in the euro area and in the rest of the world are determined based on CES functions with domestic export being one of the inputs.u u It . It x.t .t .t d X i. (4. (4. macroeconomic balance of the model and the rest of the world.u = Ptx Ptu S tx Ptx Pte −η f .e I te .Decisionmaking problems. The process of setting the price runs similarly as in the case of domestic goods producers and importers with parameters ξ x .22) where Ptmx is the price of imported component. The solution of the problem leads to the marginal costs of exporters: mc tx = 1 e S t S x Ptx ω x Ptmx 1−η x x + 1 − ωx Ptd 1−η x x 1 1−η x x .t and .t + Ptd X i.t S tx .21) where η x x is the elasticity of substitution between domestic and imported products. Export good of the i th exporter is produced with the use of domestic products imported products m X i.u is the elasticity of substitution between domestic export products and the euro area (the rest of the world) products.u = Ptx Ptu −η f .t m d min Ptmx X i.e = S tx Ptx Pte −η f .e C te .t .
ωc ωi is the share of import in consumption (investments) and ηc ηi is the elasticity of 63 63 WORKING PAPER No. consumption. (4. ζc .t is the supply of labour (hours). The preferences regarding consumption. Households provide labour and capital services to the producers of domestic intermediate products. This enables households to acquire state contingent securities making them homogenous with regard to the possessed resources and incurred expenditures. In each period households divide their income between domestic and foreign deposits. 83 . (4. Q j.4 Households The households maximise utility consisting of consumption. ζlt = 1. σq is the elasticity of demand for cash with respect to interest rate. The utility function of a jth household takes the form: 4 ∞ t s=0 where C j. q}. leisure and cash. σ L is the inverse of the elasticity of labour supply with respect to wage. σζl . ζh and ζ t . the demand for domestic export depends on the relation of export prices and world prices and income (output) abroad. All of the income is adequately taxed. where C td I td and C tm I tm are domestic and import components of consumption (investments). 4. investments and purchase/sale of new.24) ζlt = 1 − ρζl + ρζl ζlt−1 + products: Ct = It = ζl .26) 1 ηi ηi −1 ηi + ωi ηi −1 ηi ηi −1 . The income of households consists of domestic and foreign deposits plus interest.t is the consumption in period t.25) Consumption and investment goods purchased by households consist of domestic and imported 1 − ωc 1 − ωi 1 ηc C td I td ηc −1 ηc + ωc 1 ηi 1 ηc C tm I tm ηi ηc −1 ηc ηc ηc −1 .t is cash holdings. l ∈ {c. as well as proﬁts transferred in the form of dividend. remuneration for the labour and capital services.t+s − bC j. macroeconomic balance of the model Thus. leisure and cash holdings. installed capital. which means that utility is derived not so much from the absolute current level of consumption as from a change in the level of consumption in reference to the previous period. equilibrium conditions. and additionally direct transfers from the state budget are allowed.Decisionmaking problems.t L h t+s c β ζ t+s ln C j. h.t z t Pt 1−σq 1 − σq . (4. thanks to which the model may be made operational. Moreover. as well as cover the cost of maintenance of capital that has not been lent to producers. ζl ∼ N 0. Households are characterised with the so called internal habit persistence.t+s−1 − ζ t+s A L + Aq ζ h t+s 1+σ L 1+σ Q j. are given by exogenous processes: t t q h j. we assume that ﬁnancial markets are complete. h j.
t+1 = (1 − δ) K k.t is the purchase/sale of new. In other words. we arrive at the equation of demand for domestic and imported components of consumption and investment: C td = 1 − ωc Ptc Ptd Pti Ψ t Ptd ηi ηc Ct . x = µ+ µΨ . z (4.Decisionmaking problems. we assume that: ˜ ˜ S (x) = S (x) = 0 and ˜ ˜ S (x) ≡ S > 0. I j.t . C tm = ωc Ptc Ptmc Pti Ψ t Ptmi ηc Ct . ﬂuctuations in investment expenditures generate costs.30) ˜ Function S is not explicitly speciﬁed. (4. ∆ j.t .t . equilibrium conditions. . which creates the mechanism of smoothening of investment expenditures. The prices of ﬁnal consumption goods Ptc and investment goods Pti are then expressed with Ptc = Pti Ψ t = 1 − ωc 1 − ωi Ptd 1−ηc + ωc (P mc )1−ηc + ωi 1−ηi Ptmi 1 1−ηc 1 1−ηi .t I j.t . Solving the problems of consumption and investment maximisation.t+1 evolves according to: K j.28) The households’ physical capital stock K j. called the investmentspeciﬁc technology shock or 64 64 N a t i o n a l B a n k o f P o l a n d .31) This means that full transformation of investments into physical capital takes place when investment expenditures grow at the steady state level. (4. We assume that aggregation is made in such a way as to maximise the values C t and I t subject to budget constraints: Ptd C td + Ptmc C tm = Ptc C t . (4. macroeconomic balance of the model substitution between domestic consumption (investment) goods and imported goods.t + Υ t F I j. Pt I td Ψt + Ptmi I tm = Pti I t .t−1 I j. 1−ηi Ptd (4. I tm = ωi ηi . Ptmc and Ptmi are the prices of the domestic components and imported consumption and investment components. It Ψt 4 I td = 1 − ωi the following equations: It .27) where Ptd .t−1 = ˜ 1−S I j.t−1 + ∆ j. An additional factor affecting the effectiveness of transformation of investments into capital goods is the exogenous process Υ t .29) where δ is the capital depreciation rate. installed capital. I j. Function F is the function of transformation of investment expenditures into physical capital: F I j.
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
effectiveness of transformation of investment into capital: Υ t = 1 − ρΥ + ρΥ Υ t−1 +
Υ ,t , Υ ,t
∼ N 0, σΥ ,
Υ t = 1.
(4.32)
The physical capital stock is fully or partially leased to the intermediate goods producers in the form of capital services K j,t . With u j,t , u j,t ≡
K j,t
steady state u = 1). We assume that incomplete use of the capital resource generates cost for
K j,t Ψt τ . Function Fa,t represents a part
K j,t
, we denote utilization rate of capital (in
τ households, depending on the utilitzation rate — Fa,t a(u j,t )
of the cost depending on the changes in the prices of energy and — similarly to the function F tτ — is a function of energy price shock πoil (the solution is based on the work by Christiano t et al. (2007a)). Function a(u j,t ) is not explicitly speciﬁed, we assume only that a(1) = 0 and The budget constraint of households takes the form of:
e e x u c c i M j,t+1 + S t B e j,t+1 + S t S t B j,t+1 + Pt C j,t 1 + τ t + Pt I t
a ≥ 0.
+ Ptd +
τ Fa,t
a(u j,t ) Ψt Ae t−1
+ z t−1
K j,t + Pk ,t ∆ t
y
= R t−1 M j,t − Q j,t + Q j,t + 1 − τk Π t t 1 − τw W j,t h j,t t
e St B e j,t e S t S tx B u + T R t + D j,t j,t
4
p 1 − τt
Rk u j,t K j,t + 1 − τ t t
e e ˜e , s t s t−1 , φ t−1 t
+ Re Φ t−1 + Ru Φ t−1 − τk t +
Au t−1
+ z t−1
˜u , su su , φ t−1 t t−1
t
Rb − 1 t−1 Ru Φ t−1
M j,t − Q j,t + Au t−1
+ z t−1 t
Re Φ t−1
Ae t−1
+ z t−1
(4.33)
e − 1 St B e j,t
e e ˜e , s t s t−1 , φ t−1 t
˜u , su su , φ t−1 t t−1
e − 1 S t S tx B u j,t
e e e e x + B e S t − S t−1 + B u S t S tx − S t−1 S t−1 j,t j,t
+ τ t Ptd F tτ
p
a(u j,t ) Ψt
K j,t + τ t Pt Pk ,t δK j,t ,
p
where M j,t are domestic ﬁnancial assets, B e and B u are assets denominated in euro and dollar, j,t j,t Pk ,t is the relative price of capital goods, τc is the consumption tax rate, τk is the capital tax t t
p
rate (on interest from deposits and dividends), τ t is the corporate income tax rate, Π t are the proﬁts of intermediate goods producers (domestic, exporters and importers), T R t are lumpsum transfers from state budget, D j,t is the income from state contingent securities, τ t and τw are the t personal income tax rate and the rate of national insurance contribution paid by an employee. Foreign assets, B e and B u , bear interest according to the interest rates for the euro area, Re , t j,t j,t and the rest of the world, Ru , adjusted for risk premium, see e.g. (Adolfson et al., 2007a, page 8) t
y
65
WORKING PAPER No. 83
65
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
and (SchmittGrohé and Uribe, 2003; Engel, 1996): Φ Φ where: Ae ≡ t Ae t−1
+ z t−1 e e ˜e , s t s t−1 , φ t−1 t
for assets denominated in euro, (4.34) for assets denominated in dollar,
Au t−1
+ z t−1
,
˜u su su , φ t−1 t t t−1
e e S t B t+1
Ptd
,
Au ≡ t
u St Bu t+1
Pt
,
e st ≡
, e S t−1
e St
su ≡ t
u St u S t−1
,
˜u ˜e while φ t and φ t are the risk premium shocks described with stochastic processes: ˜e ˜e φ t = ρφ e φ t−1 + ˜ ˜u ˜u φ t = ρφ u φ t−1 + ˜
˜ φ e ,t , ˜ φ u ,t , ˜ φ e ,t ˜ φ u ,t
∼ N 0, σφ e , ˜
˜e φ t = 0, ˜u φ t = 0.
∼ N 0, σφ u , ˜
(4.35)
4
Risk premium for assets in the given currency depends on the position in those asses at the scale of the whole economy, while function Φ is strictly decreasing in Ae Au . For total foreign t t assets, a t ≡ denominated in euro are positive (then au = −a e ).
At + zt
=
Ae t + zt
t + z + , we assume that in steady state they are equal to 0, while foreign assets t
Au
Based on the utility function (4.24), budget constraint (4.33) and the law of motion of capital (4.29) we may formulate an optimisation problem and the Lagrange functional: max
∞
u 0 ˜ c j,t ,i j,t ,u j,t ,k j,t+1 ,m j,t+1 ,q j,t ,∆ j,t ,b e j,t+1 ,b j,t+1 t=0
βt
t,
t
= ζc ln C j,t+s − bC j,t+s−1 − ζh A L t+s t+s
h j,t
1+σ L q + Aq ζ t+s
Q j,t z t Pt
1−σq
1 + σL
1 − σq
p
+ υ t R t−1 M j,t − Q j,t + Q j,t + T R t + D j,t + 1 − τk Π t + 1 − τ t Rk u j,t K j,t t t + 1 − τt + Ru Φ t−1 − τk t +
y
1 − τw W j,t h j,t + Re Φ t t−1 Au t−1
+ z t−1
Ae t−1
+ z t−1
e e ˜e , s t s t−1 , φ t−1 t
e St B e j,t p
˜u , su su , φ t−1 t t−1
t
τ e S t S tx B u + τ t Ptd Fa,t j,t
p
a(u j,t ) Ψt
K j,t + τ t Pt Pk ,t δK j,t
e − 1 St B e j,t
R t−1 − 1 Ru Φ t−1
M j,t − Q j,t + Au t−1
+ z t−1 t
Re Φ t−1
Ae t−1
+ z t−1
e e ˜e , s t s t−1 , φ t−1 t
˜u , su su , φ t−1 t t−1 −
e e e − 1 S t S tx B u + B e S t − S t−1 j,t j,t
e e x + B u S t S tx − S t−1 S t−1 j,t
e e x u M j,t+1 + S t B e j,t+1 + S t S t B j,t+1
+ Ptc C j,t 1 + τc + Pti I t + Ptd t
τ Fa,t
a(u j,t ) Ψt
K j,t + Pk ,t ∆ t
+ ω t (1 − δ) K k,t + Υ t F I j,t , I j,t−1 + ∆ j,t − K j,t+1
66
66
N a t i o n a l B a n k o f P o l a n d
Decisionmaking problems, equilibrium conditions, macroeconomic balance of the model
where υ t and ω t are Lagrange multipliers. First order conditions of the above decisionmaking problem create a system of equations determining consumption, investments, physical capital stock and its utilization rate, cash holdings, foreign assets denominated in euro, marginal utility of nominal income and nominal exchange rate (USD/PLN). After stationarization, for which we additionally deﬁne:
+ υ t z t Ptd ≡ ψz+,t
(4.36)
and upon the consideration of the solution symmetry, the conditions may be presented as follows: ζc t c t − bc t−1 µ 1 +
z ,t
−βb
ζc t+1
t
c t+1 µz + ,t+1 − bc t υ t Ptd pk ,t Ψt
− ψz + ,t γcd 1 + τc = 0, t t
= ωt ,
−ψz + ,t γid + ψz + ,t pk ,t Υ t F1 i t , i t−1 , µz + ,t µΨ,t t +β
t
4
ψz + ,t+1 pk ,t+1 Υ t+1 F2 i t+1 , i t , µz + ,t+1 µΨ,t+1 = 0, ψz + ,t+1 1 − τk t+1 R t − 1 + 1 = 0,
−ψz + ,t + β
t
π t+1 µz + ,t+1
−ψz + ,t pk ,t + β
p
ψz + ,t+1
t
µz + ,t+1 µΨ,t+1
1 − τ t+1
p
τ ¯t+1 u t+1 − F t+1 a(u t+1 ) rk
+ τ t+1 pk ,t+1 δ + pk ,t+1 (1 − δ)
τ ¯tk = Fa,t a (u j,t ), r
= 0,
ζ t Aq q t
q
−σq
− ψz + ,t 1 − τk t
R b − 1 = 0, t−1
−ψz + ,t + β
ψz + ,t+1
t
µz + ,t+1 π t+1
e e e e ˜e s t+1 Re Φ a t , s t+1 s t , φ t t
k 1 − τk t+1 + τ t+1
= 0,
−ψz + ,t + β
ψz + ,t+1
t
µz + ,t+1 π t+1
e x ˜u s t+1 s t+1 Ru Φ au , su su , φ t t t t+1 t t
k 1 − τk t+1 + τ t+1
= 0.
The labour market is characterised by monopolistic competition — households provide heterogeneous labour services h j,t and set wages W j,t . In a similar manner to creating homogenous
67
WORKING PAPER No. 83
67
t+1 = πc t κw πc t+1 1−κw µz + . with probability 1 − ξw . . λw = λw . πc t+1 t+s 1−κw y 1 − τw × t+s × πc . the current value of the inﬂation target (with weight 1 − κw ) W j. t (4. ﬁrms).t+1 W j. 68 68 N a t i o n a l B a n k o f P o l a n d . macroeconomic balance of the model ﬁnal products by the aggregators.t . . it may only be indexed to previous inﬂation of model) — in each period. . also the heterogeneous labour services are aggregated into a homogenous labour services H t . .e. 4. t ∼ N 0. equilibrium conditions. The agents have not been assigned any formal. which may be then used by the domestic intermediate goods producers: Ht = 1 0 1 λw t h j. with and the current technology growth: 4 consumer prices (with weight κw ).t d j λ w t . a household may set optimal wage. λw σλw . it maximises the difference between the utility of income on account of wage and disutility of leisure reduction: ∞ t s=0 s max new Wt βξw − ζh A L t+s × h j. .t .37) The process of wage setting runs similarly to the process of price setting by the producers (Calvo probability ξw wage cannot be reoptimised.t+s 1+σ L 1 + σL κw + υ t+s 1 − τ t+s πc . λw . i. autonomous object functions.t+s . charges taxes from which expenditures are ﬁnanced. The ﬁrst order condition of the above decisionmaking problem leads to the equation of real wage in economy.5 Behaviour of other agents Apart from optimising agents (households. the SOEPL model explicitly considers the existence of two additional agents — the central bank and the government. . It is only assumed that the purpose of the central bank is to control price dynamics.t+s Wtnew h j.Decisionmaking problems. and the only instrument the bank has at its disposal is the interest rate. πc t t+s−1 × µz + . The other agent — the government — fulﬁls a passive function of managing budget funds.t+1 . µz + . 1 ≤ λw < ∞.t where wage markup is described with an exogenous process: λ w = 1 − ρλ w λ w + ρλ w λ w + t t−1 λw . When a household is allowed to set the wage in an optimal way.
On the other hand.t . sensitivity of the interest rate to GDP deviations from its level in a steady state. equilibrium conditions. The standard version + r∆π ∆πc + r∆ y ∆ y t + εR. We assume that the real exchange rate present in the rule is the effective rate deﬁned as: x ue = x u + 1 − γ xux x tx . The stationary value of the target (in the second meaning) may but does not have to be compliant with the target declared by the institution responsible for the monetary policy. In each case when we speak about the inﬂation target. i.1 Central bank The reactions of the central bank5 are characterised from an external point of view. therefore. without prior reference to the decisionmaking problem. change in the value of some of the parameters of the rule.e. The disturbance of the interest rate (monetary policy. the agents derive characteristics of the rules of behaviour of the central bank: interest rate persistence. etc.t . From the actually applied monetary policy.39) In the current version of the model we assume that rπc ≡ 1. Such interpretation eliminates the automatic assignment of any interest rate changes to the decisions of the central bank. The interpretation of the component 5 Depending on the type of interest rate to which the model refers.5. — contrary to other shocks appearing in the model — is deﬁned as innovation. therefore. c 69 WORKING PAPER No. In microeconomic decisionmaking problems of households and ﬁrms there appear values related to the interest rate policy. the interest rate rule describing the activities of the central bank is deﬁned directly in a loglinearised form. when interest rate is derived from the interbank market.e. The interpretation of activities of the central bank and independent conclusions regarding the perspectives of the inﬂation processes. In a typical situation. During estimation we allow the possibility of a structural change occurrence. macroeconomic balance of the model 4. rational and anticipating active participants of the economic processes — ﬁrms and households. In other words. sensitivity of the interest rate to inﬂation deviations from the reference point (inﬂation target6 ).Decisionmaking problems. it is silently assumed that the central bank is not trying to carry out any form of game intended for exploitation of the perception error created by the policy. monetary disturbance) εR. 83 69 . an extending interpretation of the agent or group of agents responsible for interest rate change is possible. This model does not refer to the ofﬁcially declared targets of the monetary policy. t (4. The form of the rule is sufﬁciently general for covering a possibly broad spectrum of interest rate policies. optimal decisions of setting prices and wages depend on the values. we mean the second concept. i. As a consequence of the above mentioned assumptions. π t shall be interpreted as the perception of the policy of the central bank (inﬂation target) by the agents. we assume that monetary policy is reliable. as well as further policy of the central bank must be reliable. The second concept allows for a stochastic nature of the target oscillating around the stationary value (steady state). temporary incoherence or inconsistency is absorbed by the ﬂuctuations of the inﬂation target. Possibly. there are grounds for claiming that interest rate ﬂuctuations result at the same time from the activities of the central bank and the responses to the current events in the interbank market. t t of the interest rate rule has then the form: R t = ρR R t−1 + 1 − ρR eu rπc π t + rπ πc − π t + r y y t−1 + r x x t−1 t−1 c c 4 (4.38) where γ xux determines the currency structure of international settlements. 6 We differentiate the inﬂation target declared by the central bank from the inﬂation target being the result of perception of the inﬂation processes and monetary policy by the optimising agents.
transfers (T R) received by households may be negative. The Ricardian logic of optimising and forwardlooking households rationalises this solution (today’s budget deﬁcit is balanced with the growth of taxes tomorrow. equilibrium conditions. Thus. one of the functions included in the ﬁscal policy is.. Therefore. If households are not inclined to reduce their expenditures. thus. therefore. In the SOEPL−2009 model there are explicitly present national insurance contributions paid by the employers (τs ) and employees (τw ). Taxes are not an instrument of ﬁscal policy — their level is relatively constant. the income tax (τ y ) and the corporate income tax (τ p ). such as it is made in DSGE models with exogenous variables. the function of government is reduced to administration of the state income redistribution. disputable. We do not deﬁne separate institutions dealing with management of such class of funds — in the applied pattern of a representative agent there is no place for households living only of the pension payments. so households shall immediately and directly reduce disposable income. the tax on consumption (τc ). the existence of budget deﬁcit and public debt has not been explicitly provided. Within the above outlined redistribution system. therefore. the ﬁscal policy impact is represented on the micro level. in the households budget constraint the category of special state contingent insurance (D) appears. however. Emphasis should be placed on the fact that the aforesaid approximation may be considered to be sufﬁcient for an economy that only sporadically experiences a deﬁcit or surplus in the budget (public ﬁnance). such as in the known models built for the purposes of central banks. 1998). a set of the basic ﬁscal charges imposed on agents. Technically. such redistribution of the state revenues that the differentiated households reﬂect a uniform pattern of consumption behaviour. macroeconomic balance of the model suggesting itself — discretionary component of the monetary policy — is. it requires a reduction in spending already today). The model is based on the pattern of a representative consumer. In the SOEPL−2009 model the disturbances are approximated with the AR(1) process or a SVAR model. the reasoning here is cohesive. so it must be ﬁnanced by the budget. Therefore. therefore. An element of the system is also the pension and disabilitypension block. 4. If the budget deﬁcit has a structural 70 70 N a t i o n a l B a n k o f P o l a n d .Decisionmaking problems. they may ﬁnance them with negative foreign deposits. it should be assumed that the information set of agents (model) and the information set of the central bank are different. Formally. Christiano et al. there are only observed temporary deviations from the longterm level (steady state). their behaviour.2 Government In the class of models deriving from the research of L. the system is not characterised in detail. namely reduction of disposable income. The deviations have the nature of disturbances.5. however. thus. the speciﬁcation of decisionmaking problems of ﬁrms and households shows the role of national insurance contributions and taxes in determining the budget of the agents and. it has been assumed that monetary disturbance is not correlated with variables present in the information set (see e. national insurance contributions are treated as the revenues of the government (budget) that are spent on a current basis. Christiano.g. Although the ﬂuctuations of tax rates have been 4 treated in a slightly simpliﬁed manner. the capital tax (τk ). Macroeconomic consequences of ﬁscal policy are treated in a more simpliﬁed manner. This is. Traditionally.
. which means that the changes e. macroeconomic balance of the model nature and the public debt reﬂects no trend of stabilisation. The model has the following form: Γ0 τ t = Γ0 Γ (L) τ t−1 + Γ0 ετ. ετ.. the role of the government sector may be larger than it appears from the proposed approximation. in government expenditures impact e. equilibrium is identiﬁed with simultaneous balancing of income and expenditures of households and the government. Joint reﬂection of budget expenditures and income leads to the following speciﬁcation: Pt G t + T R t + D t = R t−1 M t+1 − M t + τc Ptc C t + τw + τs + τ t t t t + τk t Π t + R t−1 − 1 Mt − Q t y 1 − τw t Wt H t e e e e e e ˜e + Re Φe (a t−1 .t ∼ N 0. G t t p y ετ. the stream of services of productivity factors is sufﬁcient for manufacturing 71 WORKING PAPER No. a separate model has been built trying to explain jointly the ﬂuctuations of government expenditures and the rates of income tax. . The issue is discussed in more detail in the following paragraphs. similarly as the tax rates.g. φ t−1 ) − 1 S t B t + B t S t − S t−1 t−1 (4.τ u t . τc . (4...t = B0.t . collective consumption) around a steady state are approximated with a (separately estimated) SVAR model. a stochastic nature.40) .τ ) enable the determination of the matrix of disturbance covariance (Στ ). 4. ετG .t . τ t .t δ K t .41) 4 The estimated matrices of parameters (Γ0 ) i (B0. τ t ≡ τ t .t . the aggregated supply and demand (factor resources).t a u t K t − Pt Pk . φ t−1 ) − 1 t−1 t−1 1 Ψt u St Bu t + Bu t u St u − S t−1 Rk u t K t − t τ Pt Fa.t Structural decomposition of disturbances is given by: Γ0 ετ. A new solution introduced to the discussed version of the SOEPL model is the use of the above speciﬁcation in building a consolidated (integrated) balance of total expenditures and income. More precisely.42) + + τt p ˜u Ru Φu (au . ετc .t ≡ ετp . The ﬂuctuations of government expenditures (government consumption. net foreign assets. (4. we obtain a system guaranteeing that in each period the total of expenditures is equal to the total of income.6 Macroeconomic balance conditions At macro scale. Στ . . 83 71 .g. as well as the balance of the banking sector and international transactions (payment balance). u · u = I.. As a result. The ﬂuctuations (deviations from the steady state) of the tax on capital and national insurance rates are treated as structural disturbances and are approximated in the DSGE model with the form of AR(1) process.Decisionmaking problems.. ετ y . corporate income tax and tax on consumption. . equilibrium conditions.t . The amount of government expenditures (G t ) has.
equilibrium conditions.6. instead proﬁts appear in steady state.1 Proﬁts in economy 4 We take into account ﬁrms manufacturing intermediate products. proﬁts generated by ﬁrms may be estimated at macro scale with the use of marginal costs assessment. Contrary to the previous versions of the model. (2007b.) ≡ f td . A new solution also consists in the explicit derivation of the balance of expenditures and income of the government (equivalent to state budget). proﬁts result from a difference between the marginal cost and the actual price. among others. macroeconomic balance of the model domestic intermediate products. Domestic products manufacturers In the case of intermediate products manufacturers we have: 1 d Πt = 0 d P jdt Y j t d j − Ptd mc t 1 0 Yj t d j The value the function of markup f (λd . Taking into account the above. More precisely. the derivation of ﬁrms’ proﬁt accounts (being the income of households).. calculated at current prices. export and import). According to the suggestion of Christiano et al.. i. The balance enables the determination of net foreign assets. We assume that total proﬁt in economy is the sum of the proﬁts generated in particular sectors of economy (domestic production.Decisionmaking problems. 4. In particular. proﬁts at macro scale may be estimated as: d d Π t = Ptd Yt − Ptd mc t ¯ Ptd Ptd λd t 1−λd t d = Ptd Yt − Ptd mc t Yt ¯ Ptd Ptd 1−λd t λd t . The system of macroeconomic balance is considerably much more extended in comparison with the previous versions of SOEPL and SOEEuro . is the allocation inefﬁciency. for the ﬁrst time macroeconomic balance of income and expenditures of households has been introduced. however. page 26–28). Net export is balanced with foreign net assets. in the case of loglinearisation it is also justiﬁed t to treat the price relations as equal one. (4.: d m Π t = Π t + Π tx + Π t . which upon integration with the balance of income and expenditures of households covers all of the expenditures at the scale of the whole economy and the methods of their ﬁnancing (sources of income). . which in combination with the carried out import and export ensures the levelling of demand and supply at macro scale. Such expression may be approximated with Yt f td . we assume that ﬁxed costs are absent in domestic production. which requires. importers and exporters.43) 72 72 N a t i o n a l B a n k o f P o l a n d . so it replaces a classical balance of payments. taking into account the ineffectiveness of allocation resulting from price setting with the use of the Calvo model. who transfer their proﬁts to households where the proﬁts are taxed with the tax on dividend (tax on capital).e.
d m xt . omitting the existence of ﬁxed costs.45) where f tx is deﬁned as f td .d m it + γmx d t − γt .15).2 Income and expenditures of households We assume that the total of net income and expenditures of households is balanced. Merging the income side with the side of expenditures.Decisionmaking problems. (4.44) where the marginal cost mc d is expressed with equation (4. Proﬁts in export Assuming. (4. further.d m m m ct + it + x t = (4. we arrive at the aggregated version of the budget condition of households: 73 WORKING PAPER No. 83 73 . we have: e Π tx = Ptx S t S tx C tx + I tx 1 − mc tx f tx . equilibrium conditions. macroeconomic balance of the model Making it stationary and then bringing down to ﬁxed prices. Bringing it to the stationary version. After making it stationary.d m ct + γmid t − γt . 4. we arrive at: m t = m Πt + zt P d t m m m = γmcd c t + γmid i t + γmx d x t − γ t t t t . that proﬁts in export (calculated in domestic currency) are subject to domestic tax. we obtain: d t ≡ d Πt + Ptd z t d d = y t − mc t y t f td = y t 1 − mc t f td .47) = γmcd t − γt .6.46) 4 Proﬁts in import Out of the two possible methods of deﬁning proﬁts (on the micro level with further aggregation or with direct reference to the macro scale). (4. we obtain: x t = e Ptx S t S tx Ptd 1 − mc tx f tx x c tx + i tx = γ t d 1 − mc tx f tx c tx + i tx . we have used the macroeconomic convention: e m Π t = Ptmc C tm + Ptmi I tm + Ptmx X tm − S t S tx Pt C tm + I tm + X tm .
s tx . we arrive at (see equation (4.6. s t .Decisionmaking problems.4 Monetary balance As in the original version of the SOE model. approximately. s t−1 .42)): Pt G t + T R t + D t = R t−1 M t+1 − M t + τc Ptc C t + τw + τs + τ t t t t + τk Π t + R t−1 − 1 t + + τt p y e e e e e e ˜e + Re Φe (a t−1 .6.48) e e e − M t+1 − S t B t+1 − S t S tx B u − 1 + τc Ptc C t − Pti I t − Pt Pk . φ t−1 S t B t + τk S t−1 B t t t t−1 1 − τk t R t−1 − 1 Mt − Q t e x e e e x ˜u + 1 − τk Ru Φ au . the sector of public ﬁnance and a fragment of the ﬁnancial sector specialising in pension insurance..t ∆ t = 0. φ t−1 ) − 1 t−1 t−1 1 Ψt u St Bu t + Bu t u St u − S t−1 Rk u t K t − t τ Pt Fa. 74 74 N a t i o n a l B a n k o f P o l a n d . s t−1 . s t−1 .t a u t K t + τ t Pt Pk . We notice that in the steady state the following applies: πd = µ µz + .t δ K t p (4. Upon adaptation to the domestic tax system and the possibility of depositing savings in two currency markets (next to domestic deposits). φ t−1 ) − 1 S t B t + B t S t − S t−1 t−1 Mt − Q t 1 − τw t Wt H t (4.. . equilibrium conditions..t a u t K t − Pt Pk . φ t−1 S t S tx B u + τk S t−1 S t−1 B u t t t t t−1 t−1 + 1 − τk Π t + 1 − τ t t + 1 − τt y p Rk u t K t − t 1 Ψt τ Pt Fa. The standard form of the balance of revenues and expenditures of a government stems from the 4 original version of the SOEEuro . .. t (4. macroeconomic balance of the model e e e e e e e ˜e + 1 − τk Re Φ a t−1 . In the above formula we assume that national insurance contributions paid by the employees and the employer represent the revenues of the government budget..50) The equation (after loglinearisation) shall be used for explaining the resource of money.49) ˜u Ru Φu (au . s t . 4.t δ K t . we deﬁne the broad money dynamics as: µt ≡ M t+1 Mt = m t+1 mt µz + .t πd ..3 State budget The term state budget or ”government budget” means here. t t+1 1 − τw Wt H t + M t + T R t + D t t 4.
Therefore. . estimation of parameters. Yet. 83 75 . s t−1 .e. s t . 2008). expenditures and income in economy.53) e ˜u + S t S tx B u Ru Φu au .6. it was the one ﬁnally chosen. .t µz + .5 Balance of payment The basic function of the payment balance in the model is the determination of the value of net foreign assets compliant with the level of activity in foreign trade. prices at which the products were purchased.d m m m ct + it + x t e + s t s tx πd µz + . Such approach complies with the approach assumed in the previous versions of the SOE model. equilibrium conditions. t−1 75 WORKING PAPER No.Decisionmaking problems.t + ν tw Fτ 1 + τs w t H t = t r µt mt πd µz + . t−1 The classical form of the payment balance (after bringing it to stationary state) has the form: e au + a t = γ t t x. e e e e e S t S tx B u + S t B t+1 = S t S tx Ptx C tx + I tx − S t S tx Pt t+1 4 C tm + I tm + X tm (4.t t e e x ˜u Ru Φu au . s t−1 t−1 t−1 (4. we add import to the speciﬁcation. φ t−1 . the geographic structure of net foreign assets is presented explicitly (see Kłos.54) e a t−1 e + st d π t µz + . The selection was to be done by virtue of experiments. as in the SOEPL model. s t . t t−1 t−1 e e e ˜e + S t B t Re Φe a t−1 . . and it becomes a component of export. φ t−1 . Classical version In the payment balance. when we consider net export.d x t − γt au t−1 .51) 4. analyses of the model characteristics (impulse response function) and analyses of forecasting accuracy. However.t t − qt .52) (4. s t−1 . . φ t−1 . and from the integrated balance of expenditures and incomes. macroeconomic balance of the model The banking system must provide ﬁnancing of ﬁrms with working capital loans and must provide cash for households. . . Making it stationary and expressing in ﬁxed prices. t t t r where M t+1 = µ t M t . Below we present both alternatives. we have the dependence: ν tk Fτ Rk K t + ν tw Fτ 1 + τs Wt H t = M t+1 − Q t .t e e e ˜e Re Φe a t−1 . Here we assume that in the balance of payment import is calculated at global prices. φ t−1 . . the experiments showed that the alternative of the income balance increases the potential possibility of the model. we obtain the following form: ν tk Fτ t r k kt t µΨ. In SOEPL−2009 both solutions were implemented. therefore. Doubts appear at the moment of choosing the price index to be used for expressing import at current prices. Foreign assets may be determined at least in two ways: from the classical version of balance. (4. i. s tx .
φ t−1 S t S tx B u . φ t−1 M t+1 − Q t e e St Bt e ˜u + Ru Φ au . investment expenditures. t t−1 t−1 Taking into account balance of the banking sector.. s t ... equilibrium conditions..d e S t S tx Pt Ptd .. φ t−1 t−1 e e s t a t−1 πd µz + . the payment balance obtained by the integration of income and expenditures of households. the income and expenditures of the budget and the balance of the banking sector (after making it stationary and bringing down to real categories) takes the form: e a t + au + g t + γid i t + γcd c t + t t t Fτ a u t k t a. and capital adjustment are ﬁnanced from the proﬁts. t t−1 t−1 t−1 Finally. φ t−1 S t S tx B u . .. x}.56) fw Pt Fτ a u t a.. φ t ) = exp −φa a t − a i − φsi i i ˜i s t+1 s t − 1 + φ t .t e e e e ˜u ˜e + Re Φ a t−1 . we obtain: e e e Pt G t + Pti I t + Ptc C t +S t B t+1 + S t S tx B u = Π t + 1 + τs t t+1 g + Fτ R t − 1 + 1 Rk u t − t t fk 1 Ψt Fτ R t − 1 + 1 Wt H t t Kt (4.48) and the state budget (4. The above equation shows that expenditures calculated at the macro level of economy (consumer expenditures. e S t S tx Ptx t i ∈ {u.t + 1 + τs t r k kt t µΨ. . i ∈ {u.. (4.. e} .t µz + .Decisionmaking problems. which upon simpliﬁcation takes the following form: e e e Pt G t + Pti I t + Ptc C t + S t B t+1 + S t S tx B u = Π t + 1 + τs Wt H t t t+1 g 4 + Rk u t − t 1 Ψt τ Pt Fa a u t Kt + Re Φ t−1 + R t−1 − 1 e ˜e a t−1 .. e at = e e S t B t+1 + Ptd z t i st = i St i S t−1 . .42).t µz + . φ t−1 S t B t + Ru Φ au .. . γt x..t t ˜u + Ru Φ au . Version with the income balance Merging the income and expenditure balance of households (4.t = Fτ R t − 1 + 1 w t H t t fw = d t + m t fk + x t (4. s t+1 . . e. φ t−1 t−1 t−1 πd µz + . .. we arrive at the formula...55) and au = t e S t S tx B u t+1 + Ptd z t . new foreign deposits (net foreign assets).t µΨ..57) e s t s tx au t−1 + Fτ R t − 1 + 1 t e ˜e + Re Φ a t−1 .d = Ptd γt = ..t t . government expenditures.. income from labour. income 76 76 N a t i o n a l B a n k o f P o l a n d . macroeconomic balance of the model where: i i i ˜i i ˜ ˜i Φi (a t .
t a ut k t . More generally.6 The aggregate resource constraint The starting point for further considerations is the formula in which all of the components are expressed at ﬁxed prices. starting with the identity of GDP calculated in real terms: d d d y g t + c t + i t + x t = ˜t (4. plus bringing it to the stationary form.t µΨ. The obtained inequality has the form: G t + C td + or assuming equality: Gt + zt 1 Ψt τ I td + Fa.t µΨ. Merging the above and assuming equality of resources and demand for the resources. macroeconomic balance of the model from capital and revenues from (mature) foreign deposits. we obtain: d d g t + ct + it + 1 − ω x f tx ω x γ t mx.d 1−η x x γt + 1 − ωx ηx x 1−η x x xt (4. we may determine: d xt = X td + zt 4 = 1 − ωx f tx ω x mx.58) + C td + zt + 1 + Ψt zt τ I td + Fa.t a u t K t + X td ≤ t 1− 1− zt Kt H t (4.6.59) and using the solution of the decision problem of exporters (4.t a u t k t − φ.61) µz + . 77 WORKING PAPER No. 83 77 .Decisionmaking problems. equilibrium conditions.22).60) we arrive at an equation that upon loglinearisation will explain the real income: yt = g t + + d ct + d it τ Fa + 1 − ωx f tx ωx mx. Net foreign assets amount in total to e a t = a t + au . t 4.d 1−η x x γt + 1 − ωx 1−η x x ηx x Xt + zt .d 1−η x x τ Fa = + Ψ t−1 z t−1 1− zt + 1 − ωx ηx x 1−η x x xt + zt t 1− Ht kt − µz + .t a u t K t + X td + zt = yt (4. In resources constraint we omit the factor characterising the ineffectiveness of allocation (the effect of Calvo price settings).
Part III Results of estimation and characteristic features of the DSGE SOEPL−2009 model 78 .
Forms of model, data, SVAR models
Chapter 5
5 Forms of model, data, SVAR models
Forms of model, data, SVAR models
5.1
Forms of model
Upon the determination of the ﬁrst order conditions from the decisionmaking problems presented earlier and their loglinearisation, we arrive at the linear structural form of the model which may be presented in matrix form in the following manner: ˜ ˜ ˜ t α0 z t+1 + α1 z t + α2 z t−1 + β0 θ t+1 + β1 θ t = 0, θ =ρθ t t−1 + t
(5.1)
5
where the vector of exogenous variables (θ t disturbances) consists of structural disturbances θ ts and observable ﬁscal disturbances θ tτ and observable foreign disturbances θ t . The approximation of the processes controlling observable disturbances is received by separate estimation of two SVAR models: ﬁscal model and world’s economy model. Fragments of matrix ρ and the matrix of observable disturbances covariance are equivalent to the respective matrices of the SVAR model, i.e. they remain ﬁxed during (Bayesian) estimation of the whole DSGE model. In the Appendix we shall present a list of variables of the structural form and a list of loglinearised equations which constitute the structural form of the model. Transition to the reduced form of the model — which eliminates the forwardlooking variables — is made with the use of AndersonMoore algorithm. Thus, we obtain: ˜ ˜ z t+1 = A z t + B θ t+1 , θ t+1 = ρ θ t + ε t+1 .
79
WORKING PAPER No. 83
79
Forms of model, data, SVAR models
The system may be transformed in the state space form of the model. Then the transition equation shall be: ˜ z t+1 θ t+1
ξ t+1
=
A 0
Bρ ρ
Fξ
˜ zt θt
ξt
+
B I
υ t+1
ε
t+1
and the state space model takes the form of: ξ t+1 = Fξ ξ t + υ t+1 , (υ t+1 υ t+1 ) = Q, (u t u t ) = R,
where Y t is a vector of observed variables and u t is a vector of measurement errors. During the construction of the state space model, the state vector is supplemented with lagged variables. Variables that are not necessary in the measurement equations may be eliminated. In ˜ order to avoid introduction of further symbols, we shall omit the fact that the z t vector present in ˜ the structural form and the z t vector in the general case are different. As results from the notation
Y = A x + H ξ + u , x t t t t
(5.2)
5
of equations, on a standard basis we assume the appearance of measurement errors in the model (state space representation). The elements of the variancecovariance matrix of measurement errors R increase the pool of the model parameters. For simplicity, we shall assume that R is diagonal. The appearance of measurement errors in the equation is explained by inaccuracies in statistics, approximation errors appearing in measurement equations (loglinearisation of nonlinear dependencies) and the possibility of errors in model speciﬁcation1 , therefore, arrival at (additional) statistical identiﬁcation is a side effect2 . Formally, in the presented version of the model there are in total 30 shocks, including 11 observable disturbances (processes estimated with SVAR models). A list of the model variables and a list of equations of the structural form presented in the Appendix characterise a more general version of the SOEPL−2009 model. The version discussed further on originated as an effect of elimination of the disturbance of national insurance contributions paid by the employer and the assumption that there is only one interest rate risk premium for the tested economy, ˜ ˜ ˜ common for both currency markets (φ u = φ e = φ ). In both cases we deal with a simpliﬁcation
t t t
of the model motivated by the problems with shocks identiﬁcation.
5.2
Observable variables, data
The data used in a DSGE model estimation and SVAR models come from the ofﬁcial publications of the National Bank of Poland (NBP), the Central Statistical Ofﬁce (GUS), the ECB and
1 The subject of the selection of observable variables and the role of measurement errors speciﬁed in DSGE models were analysed by GuerronQuintara (2009). According to his conclusions, consideration of measurement errors during a DSGE model estimation allows to increase the model resistance to speciﬁcation errors and incorrect selection of observable variables; see also Boivin and Giannoni (2005). 2 The issue of statistical identiﬁcation was discussed in more detail by e.g. Ireland (2004), AlvarezLois et al. (2005), Canova and Sala (2005).
80
80
N a t i o n a l
B a n k
o f
P o l a n d
Forms of model, data, SVAR models
OECD available till November 2009. The observable variables in the Y t vector in equation (5.2) are: GDP deﬂator, investment deﬂator, consumer price index (CPI), GDP consumption , (calculated together with changes in inventories), investment expenditures, export, import, total employment, real wages, domestic shortterm interest rate (Wibor 3M), real USD/PLN exchange rate, GDP deﬂator in the euro area, GDP deﬂator in the USA, shortterm interest rate for dollar (OECD assessments of threemonth’s dollar rate), shortterm interest rate for euro (Euribor 3M), GDP in the USA, GDP in the euro area, the nominal cross rate (dollar/euro)3 — 19 series in total. The whole sample for the DSGE SOEPL−2009 model covers the period of 1996:2–2009:3. For the estimation of the SVAR model of the world’s economy, we have used the longest available series. The data referring to domestic national accounts stem from the system of measurement at ﬁxed prices of the year 2000. At the initial stages of preparing the data, the consistency of the calculations (volume and deﬂators) was veriﬁed, and then the series were subjected to transformations that referred to the elimination of their seasonality with the X12 method (only variables that may reﬂect quarterly seasonality), elimination of means (exchange rate), ﬁnding the logarithm, transformation of variables into quarterly growth rates. Exceptions are the dynamics of prices, which have an annualised form, interest rate (remaining in the natural form) and employment. In the case of the last variable, the data regarding the level have been ﬁltrated (with HodrickPrescott ﬁlter) and deviations from the HP trend act as an observable variable. Bearing in mind the solutions assumed by the authors of the SOEEuro model, an additional adjustment of the export and import series was made. Whereas the model does not attempt to explain why the share of foreign trade in the GDP changes (grows), an adjustment ensuring constant share of import and export was made. Another arbitrary correction refers to the trend appearing in the observable exchange rate (real exchange rate of dollar/zloty). The phenomenon of nominal exchange rate appreciation (in consequence also the real exchange rate) is speciﬁc for economies catchingup with the development gap and similarly as in the case of trade share in the GDP the SOEPL−2009 model does not undertake to explain the process, therefore, at the , ﬁnal phase of transformation (linear) trend is eliminated from the exchange rate. Although the practice of elimination of series of various arbitrarily deﬁned components (generally — the ﬁltration practice) used in estimation of DSGE or (S)VAR models is widespread, it is worth to remember that the informative content of the series and their possible interrelations are distorted. As an example, by eliminating a trend from employment or exchange rate, we leave the consequences of the existence of the trends in other series (e.g. prices, GDP) and such set of data is further used for searching relations between series. We wish to point out that foreign variables are present in the set of observable variables, i.e. the variables fulﬁl a double function in the model: they are disturbances and observable variables. Thus, we receive a better identiﬁcation of the other disturbances referring to the world (export
3 We present here a set of variables that was used in the estimation of the discussed version of the model. The version is an effect of many experiments, out of which a part referred to the selection of observable variables. We have experimented (among others) with monetary aggregates, export deﬂators, import deﬂators, consumption deﬂators (including changes in inventories), ﬁscal series (government expenditures, revenues from VAT tax), competitive versions of employment series and change in the real exchange rate of dollar.
5
81
WORKING PAPER No. 83
81
40).Forms of model.5 1 1998 2000 2002 2004 2006 2008 2 1 4 3 US inflation 6 5 4 3 2 1 1998 2000 2002 2004 2006 2008 USD interest rate Change of dollar/euro exchange rate 10 5 0 5 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 5. asymmetry in the level of technology.5 1 0. Data GDP deflator 25 20 15 10 5 0 1998 2000 2002 2004 2006 2008 6 4 2 0 2 4 1998 2000 2002 2004 2006 2008 2 4 1998 2000 2002 2004 2006 2008 GDP 4 2 0 Consumption 15 10 5 0 5 10 Investment 20 10 0 10 20 30 1998 2000 2002 2004 2006 2008 Exports 1998 2000 2002 2004 2006 2008 Imports 15 10 5 0 5 10 15 1998 2000 2002 2004 2006 2008 10 5 0 5 10 15 20 Real exchange rate 25 20 15 10 5 1998 2000 2002 2004 2006 2008 Interest rate 2 1 0 1 2 1998 2000 2002 2004 2006 2008 Employment 3 2 1 0 1 2 1998 2000 2002 2004 2006 2008 Real wages 1998 2000 2002 2004 2006 2008 Investment deflator 60 40 20 15 10 5 0 CPI 1 0 1 2 1998 2000 2002 2004 2006 2008 Euro zone GDP 3 2 1 0 1 1998 2000 2002 2004 2006 2008 Euro zone inflation 5 4 3 2 1 1998 2000 2002 2004 2006 2008 Euro interest rate 5 20 0 20 40 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 US GDP 1.5 0 0. interest rate and economic growth in domestic and world economy). The above described raw time series are subject to transformations. The purpose of ﬁscal SVAR model and world’s economy SVAR model is to receive a description of a stochastic process approximating the disturbances represented by the variables of the SVAR models. etc. consumption tax (state budget revenues from intermediate taxes referred to individual consumption) and government expenditures (public consumption). formally represented by equation (4. one of the fundamental assumptions of the model of a small open economy is veriﬁed: the common steady state (identical inﬂation. additionally. data. Tax rates are seasonally 82 82 N a t i o n a l B a n k o f P o l a n d . Figure 5.1 Fiscal SVAR Fiscal SVAR. effective corporate income tax rate (the revenues of state budget from corporate income tax by the GDP decreased for households income). SVAR models and import markups.3.) and.1. Any possible departures from that principle are compensated with constants in the block of measurement equations. 5. risk premium.3 SVAR models SVAR models are a separate component of the model. Figure 5.1 illustrates the graphs of observable variables of the model after all of the transformations. consists of four variables: effective personal income tax rate (calculated as the revenues of state budget from personal income tax by the estimates of households income. increased for the health insurance contribution).
πe . Inﬂation and the nominal exchange rate are included in the model as quarterly logarithmic growth rates of the series of GDP deﬂators and nominal exchange rate of USD/EUR. The model consists. while the data from before 1995 have been reconstructed based on the database of a Area Wide Model (AWM) of the European Central Bank. inﬂation (πe and πu ) and nominal interest rate (Re and Ru ). The dollar area is approximated by the USA economy. An exchange rate shock enters the 83 WORKING PAPER No. For the propagation of structural disturbances between the euro area and the dollar area one quarter delay has been assumed. Beside of the variables t t t t characteristic of each of the areas. as well as the cross rate USD/EUR. For modelling each of the areas we have used three variables: real product ( y te and y tu ). Variables of the euro area shall be identiﬁed with superscript e. 5. thus. The data for the euro area starting from the ﬁrst quarter 1995 come from the OECD and Eurostat databases. πu . i. starting from the ﬁrst quarter of 1980. Vector autoregression system in reduced form is given by: y t = A1 y t−1 + A2 y t−2 + d + ωt + e t where: y t = ( y te . A series of the nominal USD/EUR exchange rate comes from the Eurostat database. For the purpose of estimation of the structural form of the VAR model within each of the areas. Cholesky type structuralisation has been applied in the following sequence: production. y tu . A logarithm was found for the series of government expenditures at ﬁxed prices and then the series was seasonally adjusted with X12 method and converted into percentage deviations from the trend determined with the use of HodrickPrescott ﬁlter. vector d consists of constants and vector ω consists of the trend parameters. Re .3. the VAR model includes the dynamics of the nominal exchange rate of USD/EUR (∆x t ). Interest rate is expressed in percentage points at quarterly values. ∆x t ). the model includes a deterministic trend and a constant. data. Output is included in the model as a cyclic component of a quarterly HP ﬁlter applied to the logarithm of an output expressed at market prices deﬂated with the GDP deﬂator. Ru . The lag order of the system is 2. of three variables for the euro area and three variables for the dollar area. Such constructed SVAR model has been estimated on a sample 1996:1–2009:3. 83 5 83 . e t = B u t . Foreign variables of the model are exogenous and consist of two areas: the euro area and the dollar area. t t t t Matrices A1 and A2 are autoregressive matrices. inﬂation and interest rate. The sample based on which the model parameters have been estimated covers 166 observations for each of the variables.Forms of model.e.2 World’s economy SVAR The task of the VAR model is to determine the dynamics of foreign variables and of dollar/euro nominal exchange rate. while the vector of error terms e t depends on the vector of structural shocks through matrix B. the reduced form covers 144 parameters.e. The variables have been seasonally adjusted with the ARIMA X12 method. i. SVAR models adjusted with X12 method and then converted into percentage deviations from the mean in the sample (according to the logic of loglinearised variables of the model). while the variables of the dollar area with subscript u. The data for the dollar area come from the OECD and Eurostat databases.
as well as the elements of vectors d and ω. SVAR models structuralisation orthogonally to the other shocks. inﬂation (πe and πu ) and the interest rate (Re and Ru ). An effect of contractionary monetary policy of the value of 1 percentage point in the dollar area is similar in terms of quality. thus. have been estimated on quarterly data with estimated general least squares method (EGLS). the elements of matrices A1 and A2 .05%. Matrix B making the forecasts errors e t dependent on structural shocks u t has been estimated with the maximum likelihood method.Forms of model. On the parameters of the reduced form dynamic restrictions have been imposed (the matrix structure. output declines by about 1%. y u ). insigniﬁcant along the whole length of the impulse response function. on the product ( y e . display larger inertia than output. The parameters of the reduced form. i. while the effect is signiﬁcant between the 5th and 15th quarter and the maximum effect materializes in the 12th quarter. Output responds signiﬁcantly between the 2nd and 5th quarter and drops by about 0. the imposed restrictions and the results of the estimation have been presented in the Annexe).2 presents the effect of positive structural shock to the interest rate (Re for the euro area and Ru for the dollar area) of the value of 1 p.2. After monetary policy contraction in the euro area. the response of inﬂation is much weaker — drop by about 0. Figure 5. data. Contractionary monetary policy shock in a SVAR model 5 84 84 N a t i o n a l B a n k o f P o l a n d .25%. Figure 5.65% in the 6th quarter.e. Impulse responses seem to be sensible. however. Inﬂation drops by about 0. Prices.p. while the maximum drop occurs 9 quarters after the shock and the effect becomes statistically insigniﬁcant after about 3 years.
Assessment of parameters — calibration, optimisation, steady state
Chapter 6
6 Assessment of parameters – calibration, optimisation, steady state
Assessment of parameters — state
calibration, optimisation, steady
The procedure of estimation of large DSGE models, motivated with Bayesian ideas1 , consists of two basic stages. At the ﬁrst stage we determine with numerical optimisation methods the approximation of modal value of posterior distribution, i.e. point estimates of the parameters. At the second stage we estimate the shape of parameters distributions in the neighbourhood of the posterior mode, by applying MCMC techniques (e.g. Metropolis algorithm versions). Thus we generate a sample from model parameters posterior distribution2 . Practice has shown that already at the stage of optimisation, problems with parameters identiﬁcation appear, while convergence of numerical procedures is achieved only if prior distributions shall explicitly determine at least a subset of parameters. In such a situation most of the authors of the DSGE models decide to divide parameters into two groups: calibrated and estimated parameters. The estimation procedure (e.g. with Bayesian techniques) is then applied only to some of the parameters. In principle, one would have to be absolutely certain about the values of the calibrated parameters and only those parameters to which our knowledge (which we represent by prior distribution) is uncertain shall be (Bayesian) estimated. Unfortunately, in our case (or probably not only in our case, see Adolfson et al. (2007a, page 12)), this was not the motivation for applying calibration to at least some of the parameters — we calibrated parameters which cannot be estimated because of technical constraints.
1 We refer here to practice, the formulated generalisation refers to the users of the DYNARE package (academic communities), YADA package (ECB) and Riksbank. Generally, DSGE models do not have to be estimated with Bayesian techniques and the manner of the Bayesian techniques application may be different than we suggest. 2 We omit here (and further on in the presentation) the fact that some of the parameters of the model are estimated separately with the use of SVAR models.
6
85
WORKING PAPER No. 83
85
Assessment of parameters — calibration, optimisation, steady state
6.1
Calibration of parameters
In the SOEPL−2009 model, which as for the standards of estimated DSGE models may be deemed to be large, about 8090 parameters from a set of over 150 parameters are calibrated. The calibration refers, among others, to 19 variances of the measurement errors of observable data, 27 parameters directly or indirectly characterising the steady state, 22 parameters characterising the stochastic disturbance processes, 8 parameters centring the variables included in the SVAR model, as well as 12 other parameters which from economic point of view may be treated as important. Above enumeration omits zero covariance of measurement errors and a collection of parameters related to the stochastic processes of observable disturbances (SVAR), yet, it includes the values of parameters that are subject to structural changes (11) and those that are determined by the speciﬁcation (economic content) of the model. The percentage of calibrated parameters is relatively higher, in reference to the estimated DSGE models developed in other countries, but our model is larger and data sample remains short and nonhomogenous. Thus we face greater challenges than the authors of models in e.g. the euro area countries or the USA. Due to the diversity of the parameters excluded from the estimation, no uniform procedure has been applied to determine their values. In case of characteristics of steady state variables that have their equivalents among observable data, the average values of the sample were the main indicator. In a number of cases, the ﬁrst approximations were the values assumed by other authors of DSGE models, and in particular the assumptions made for the models of SOEEuro , MEDEA (Burriel et al., 2009), NAWM and RAMSES (Adolfson et al., 2009), however, we always
6
made attempts to verify such assumptions by experimenting with competitive values, i.e. we tested the consequences of changes in the values of such parameters, making a sort of sensitivity analysis, such as e.g. the authors of NAWM (Christoffel et al., 2007a, pages 44–45). At the initial stages of the works over the model, we used Laplace approximation of marginal likelihood as the criterion of assessment of the model, later we additionally applied the measures of accuracy of ex post forecasts for the last 45 years of the sample, all the time controlling the dynamic features of the model (impulse response functions). The experiments proved that restrictions imposed on a DSGE model by virtue of calibration of some of the parameters were very strong but in many cases failed confrontation with empirical material. Seemingly small adjustments of calibrated parameters signiﬁcantly inﬂuenced the accuracy of forecasts, yet, they are not recognisable for other criteria (e.g. marginal likelihood, impulse response functions). Small departure from the “typical” calibrated values of parameters enabled us to improve the quality of the SOEPL−2009 model, or at least its forecasting features, without a major breach of its economic content. Despite the efforts, we are aware that there is large spare capacity in that area and potential consequences of assuming wrong values are serious. Speciﬁcation of the most important calibrated parameters of the model has been provided in Table 6.1. The speciﬁcation omits the parameters obtained from estimation of SVAR models.
86
86
N a t i o n a l B a n k o f P o l a n d
Assessment of parameters — calibration, optimisation, steady state
Table 6.1. Values of the most important calibrated parameters in the SOEPL−2009 model
Parameter δ β AL Aq σL σq νfk νfw µ gr τk τw τs τc τp τy ωc ωi ωx ωmcu ωmiu ωmxu ηf u ηf e u y e, ua e m b ˜ S σa λx λmx ρε ρλ d ρλ mx ρε,R ρπ c ¯ ρτ k ρτ s ρπoil cRd R kr a j Value 0.250 0.025 0.9995 7.500 0.135 1.000 10.620 0.400 0.600 1.0165 0.190 0.0125 0.180 0.180 0.1772 0.0125 0.1301 0.450 0.650 0.330 0.460 0.370 0.333 4.500 3.200 0.6810 0.0095 0.6401 6.9682 1.1276 1.0000 1.0000 0.700 0.100 0.500 0.000 0.900 0.900 0.900 0.900 1.3500 0.0400 0.0100 0.6774 0.6431 0.7959 1.7921 0.1727 Comments Source based on literature, veriﬁed by experiment based on literature and sample means of the great ratios based on literature, veriﬁed by experiment based on literature based on data based on literature based on literature based on literature, veriﬁed by experiment based on literature, veriﬁed by experiment determined by experiment mean from the sample of share of consumption in GDP empirically veriﬁed expert assessments average national insurance contributions average national insurance contributions mean from the sample of effective VAT rate mean relation of revenues from CIT to GDP mean from the sample of effective PIT rate determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment conclusion from model assumptions conclusion from model assumptions determined by experiment based on literature, veriﬁed by experiment based on literature based on literature based on literature, veriﬁed by experiment determined by experiment determined by experiment determined by experiment determined by experiment expert assessments expert assessments determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment determined by experiment
steady state steady state steady state steady state steady state steady state steady state steady state
6
R ot oczenie ρλ ,1 w ρΦ,1 ˜ σλ ,1 w σΦ,1 ˜ σε
R,1
observable variable adjustment variance of measurement error variance of measurement error change of regime change of regime change of regime change of regime change of regime
σπc ,1 0.3482 change of regime ¯ λw,1 1.1366 change of regime ρR,1 0.7359 change of regime rπ,1 1.4497 change of regime r x,1 0.0100 change of regime r y,1 0.1783 change of regime Source: Prepared by the authors.
6.2
Prior distributions and results of estimation
As it has been mentioned before, the DSGE model estimation is a twostage process. At the ﬁrst stage we search for the values of parameters that maximise the value of the function, which is given by the sum of the logarithms of the likelihood function and prior distribution. These parameters represent the mode of posterior distribution and the stage is called optimisation. At the second stage an attempt is made to generate a sample from (posterior) distribution of 87
WORKING PAPER No. 83
87
Assessment of parameters — calibration, optimisation, steady state
parameters by using e.g. an iterative algorithm in which the starting point is the distribution modal value, and consecutive values depend on the Hessian obtained during optimisation. Generation of such sample allows — among others — to better understand the scale of uncertainty related to the estimated parameters. That piece of information may be used to asses forecasts uncertainty. The formal aspects of Bayesian estimation, MCMC techniques and speciﬁcs of Bayesian estimation of DSGE models have been shortly outlined in the ﬁrst part of the paper. Other recommended sources include i.e.: Osiewalski (2001), Koop (2003), Geweke (1999), Geweke (2005), Hastings (1970), Chib and Greenberg (1995), Brooks (1998), Brooks and Gelman (1996) and Schorfheide (2000), An and Schorfheide (2005), Canova (2007), Warne (2009), FernándezVillaverde (2009), providing the details of the use of Bayesian techniques in DSGE models estimation.
Table 6.2. Results of estimation — characteristics of prior and posterior distributions
Parameters ξw ξd ξmc ξmi ξmx ξx ξe κw κd κmc κmi κmx κx ντ ˜u φa ˜e φa ˜u φs ˜ φe
s
Type Beta Beta Beta Beta Beta Beta Beta Beta Beta Beta Beta Beta Beta TNor InvG InvG Beta Beta
Prior distribution CVal Std/DF 0.600 0.600 0.600 0.600 0.600 0.600 0.600 0.500 0.500 0.500 0.500 0.500 0.500 0.125 0.125 0.125 0.250 0.250 0.800 1.700 0.300 0.000 0.125 0.075 1.100 1.200 1.200 1.200 5.000 5.000 5.000 1.008 1.008 0.700 0.900 0.600 0.600 0.750 0.750 0.750 0.100 0.100 0.100 0.100 0.100 0.100 0.100 0.140 0.140 0.140 0.140 0.140 0.140 0.075 2 2 0.150 0.150 0.085 0.150 0.065 0.065 0.065 0.065 0.075 0.075 0.075 0.075 2 2 2 0.002 0.002 0.100 0.100 0.100 0.100 0.100 0.100 0.100
Optimisation results Mode InvH
Mean
Sample form posterior distribution Mode Median 5% perc. 95% perc. 0.557 0.795 0.675 0.685 0.553 0.530 0.704 0.419 0.164 0.440 0.297 0.347 0.204 0.121 0.429 0.186 0.193 0.180 0.827 1.795 0.234 0.020 0.128 0.127 1.137 1.203 1.290 1.139 3.792 4.097 4.622 1.008 1.007 0.777 0.971 0.596 0.585 0.782 0.614 0.685 0.427 0.749 0.573 0.576 0.416 0.431 0.641 0.231 0.079 0.228 0.138 0.172 0.095 0.021 0.184 0.109 0.051 0.053 0.768 1.596 0.184 0.048 0.039 0.072 1.055 1.098 1.199 1.033 3.146 3.271 3.540 1.007 1.005 0.683 0.848 0.430 0.412 0.683 0.436 0.498 0.690 0.837 0.774 0.783 0.687 0.631 0.755 0.628 0.286 0.682 0.529 0.577 0.375 0.240 1.483 0.318 0.421 0.316 0.877 2.010 0.286 0.005 0.225 0.187 1.241 1.311 1.383 1.283 4.813 5.880 6.637 1.009 1.009 0.855 0.999 0.748 0.741 0.862 0.782 0.833
6
ρR,2 rπ,2 r∆π r x,2 r y,2 r∆ y
w λ2 λd λmc λmi ηc ηi ηx x µz µΨ
Beta TNor TNor Norm Norm TNor TNor TNor TNor TNor InvG InvG InvG TNor TNor Beta Beta Beta Beta Beta Beta Beta
ρΥ ρz ˜ ρµ z ρµ Ψ ρζ c ρζh ρλmc
Price rigidities 0.558 0.079 0.558 0.547 0.790 0.026 0.794 0.790 0.677 0.057 0.675 0.672 0.687 0.062 0.683 0.690 0.554 0.081 0.552 0.562 0.527 0.060 0.530 0.530 0.698 0.034 0.701 0.704 0.417 0.124 0.423 0.385 0.164 0.062 0.171 0.146 0.441 0.144 0.446 0.402 0.291 0.113 0.310 0.278 0.350 0.123 0.357 0.333 0.202 0.081 0.215 0.175 Foreign exchange rates and energy prices effects 0.150 0.068 0.125 0.122 0.382 0.206 0.610 0.295 0.190 0.063 0.196 0.164 0.217 0.114 0.208 0.133 0.196 0.082 0.182 0.202 Interest rate rule 0.820 0.035 0.825 0.829 1.800 0.124 1.798 1.782 0.237 0.031 0.234 0.231 0.022 0.017 0.020 0.015 0.128 0.057 0.130 0.120 0.128 0.033 0.128 0.124 Markets organisation, technological progress 1.150 0.060 1.142 1.121 1.213 0.063 1.204 1.209 1.299 0.052 1.290 1.286 1.160 0.071 1.146 1.119 3.740 0.454 3.861 3.723 3.964 0.607 4.454 3.777 4.544 0.861 4.800 4.362 1.008 0.001 1.008 1.008 1.007 0.001 1.007 1.007 Characteristics of disturbances 0.773 0.052 0.774 0.785 0.957 0.044 0.953 0.999 0.592 0.097 0.593 0.607 0.586 0.099 0.583 0.573 0.778 0.053 0.778 0.774 0.625 0.102 0.612 0.629 0.685 0.101 0.678 0.698
Continued on next page
88
88
N a t i o n a l B a n k o f P o l a n d
466 0.570 0.500 0. 4 It is worth to remember that reaching for Bayesian techniques is not always the consequence of the Bayesian point of view or Bayesian beliefs of the authors.250 ¯ Source: Prepared by the authors.623 0.2 Type Beta Beta Beta Beta InvG InvG InvG InvG InvG InvG InvG InvG InvG InvG InvG InvG InvG InvG Prior distribution CVal Std/DF 0.639 0.453 0.220 0.021 0.007 0. Instead of additionally extending the set of calibrated parameters.392 2.287 0. selection of an effective algorithm for optimisation3 .504 0.610 0.584 0.092 0.010 0. HInv — approximation of standard deviation from the assessment of parameter based on Hessian inverse.633 0. Variances (degrees of freedom) of prior distributions have been deﬁned such as to prefer information included in the sample (unless we had a stronger belief regarding the value of the parameters and the sample proved to be sufﬁciently informative)4 .062 1.384 2.081 0.425 0.277 0. 83 89 . for the inverse Gamma distribution the modal value and the number of the degrees of freedom have been speciﬁed.114 0.558 0.500 0.205 0.467 Sample form posterior distribution Mode Median 5% perc.396 0.2 ˜ σε σΥ σz ˜ σµ z σµ Ψ σζ c σ ζh σλ d σλmc σλmi σλmx σλ x σλw .436 0. In the speciﬁcation provided there.750 0. The central values of the distributions.385 2.443 0.423 0.282 0. 95% perc.2 InvG 0.000 1.582 0.379 0.582 2.557 0.000 1.846 0.563 0. an inverse Gamma distribution for variances of disturbance innovation. The type of prior distribution assumed for particular parameters results from knowledge about the allowed interval of values of the parameters.425 2.707 0. however.750 1.100 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Optimisation results Mode InvH 0.750 0.571 2.104 3.025 σν oil InvG 2.372 0.155 0.205 0.791 0.047 0.214 0.396 0.688 0.537 1. As in the case of calibrated parameters.121 0.631 0.304 2.682 0.208 0.192 0.107 0.361 0.g. csminwel algorithm by Ch.524 0.664 0.000 1.322 2.707 0.442 0.359 0.296 1.134 5. as in this case.856 0.565 0. optimisation. Sometimes. Note: Nrm — normal distribution.371 0.264 0.101 9. the Bayesian procedures are applied instead of the classical procedures.739 0.547 6. of which we have.000 0.180 0.418 1.100 0.102 0.051 0.675 0.617 0.544 0.773 1.417 5.697 0.555 0. CVal — central value.971 0. at the cost of considerable longer time of computation.216 0.012 0. completely undistinguishable for an 3 Often applied by researchers.000 σR. After several experiments it appeared that the unconditional optimisation algorithm implemented in the optimisation toolbox of Matlab deals much more effectively with the extreme’s determination. (2007.043 0.289 0.869 0.515 3.250 σπc .484 5.761 0.079 0.843 0. 0.037 2. We also continue to stick to conditional estimation — optimisation takes place in the presence of the expert knowledge introduced formally with prior distribution.2 InvG 0. proved to be debatable.306 0.844 0.810 1.666 0.440 0. Calibration of parameters does not solve all the problems which appear during the optimisation. prior distributions of models built before 2006 are available.010 σ τs InvG 0.100 0.531 0.100 0. on a standard basis there is assumed Beta distribution for e.445 0.500 0. etc.750 0.513 6.018 1.700 0.289 0.060 0.205 Mean 0.500 0.025 0. InvG — inverse Gamma distribution.449 1.711 2. e.961 2.602 1. of course. extensive literature describes the experience of the authors of Bayesian estimated DSGE models5 .553 3.370 0.757 1. Also in this case it appears that relatively small changes in the central values of distributions.129 0.031 2.687 0.006 0.583 2.500 0.055 2.102 0.856 1.031 0.546 0. Sims (provided by the author on the Internet site in the form of Matlab scripts. normal distribution for parameters that may assume any values.274 0. we avail of techniques that slightly reduce the arbitrariness of the study. steady state Parameters ρλmi ρλ x ρλw .380 2.714 0.g. 6 89 WORKING PAPER No. however. Therefore.263 0.2 ρφ e .489 0.000 0.954 2.220 0. due to the fact that the available sample does not include a sufﬁcient amount of information to carry out the traditional estimation (ﬂat likelihood function).794 0.139 0.365 0. implemented in Dynare and YADA packages.750 1. therefore.382 0.025 2.561 0. pragmatic considerations are decisive.483 0.700 0.095 0.295 0.058 0.532 0. 5 Grabek et al.261 1.053 2.015 1.2 σ e˜ φ.Assessment of parameters — calibration.285 0.202 0.666 0.663 0.055 0. TNor — truncated normal distribution.015 0.115 0. Calvo probabilities or correlation coefﬁcient (AR processes of the shocks).195 0.994 0.802 στ k InvG 0.860 0.088 0.855 0. pages 69–70) present several examples drawn from the DSGE models with structure similar to SOEPL−2009 .298 0. also available in the form of C++ function) proved to be ineffective in our case. availed while verifying our own analyses and beliefs.064 0.
steady state expert. as well as assessments of parameters and approximations of standard errors arrived at during optimisation have been provided in Table 6. On the other hand.1 Assessment of parameters — conclusions Steady state When all the disturbances (shocks) described in the model expire. it is hard to resist the feeling that some of the authors of the DSGE models treat convergence in a sort of ritual manner — this is a qualitative difference with regard to the conduct of researchers specialising in the Bayesian techniques. Already a superﬁcial analysis of the values of parameters for the modal value received from optimisation and determined from the Markov chain reﬂects a considerable divergence between the corresponding values. The annualised value of the parameter implies the growth rate at the level of about 6.000 elements has been generated. except investments.000 elements each (see Christoffel et al.Assessment of parameters — calibration.000 elements was used.000.000 elements rejected). Therefore. 90 90 N a t i o n a l B a n k o f P o l a n d . page 16). 2008. the Metropolis procedure was used for generation of two Markov chains of the lengths of 300.000 elements.3 6. Generally. convergence was tested with the use of four parallel chains of the length of 500. In the DSGE model built for the Central Bank of Hungary.000 end elements for inference see (see Adolfson et al. The division of the value dynamics into the dynamics of volume and prices is identical for all the growing categories. test statistics based on cumulated sums. The characteristics of posterior distribution of the NAWM model (ECB) have been determined based on a chain of the length of 500. Due to limited resources at our disposal we could not further generate the chain6 . during further works we used only the point estimates of parameters originating from the ﬁrst stage of estimation — optimisation.000 elements using 500.000 elements for a signiﬁcantly smaller DSGE model (see SchmittGrohé and Uribe. while the data presented in the Table have been taken from a sample of 500. Then.2 provides also selected characteristics of the distribution received as a result of the application of Metropolis algorithm (in the random walk version). a signiﬁcant issue of the uncertainty of parameters at the current stage of the works has not been completed — we shall come back to the issue after the existing technical limitations has been reduced. S..000800. SchmittGrohé and M. 6 6. The ﬁnal result of our research — a complete set of assumptions regarding prior distributions.000. originating from the merger of the last 250. which generates a sample of distribution convergent toward posterior distribution. More thorough convergence analyses (logarithms of posterior probability for the elements of the chain. the economy reaches a steady state. Markov chain of the length of 750. Uribe used the end 4. page 42).3. optimisation.000 elements of each of the chains (see Jakab and Világi..) provided additional doubts as to whether the convergence is actually present. however. Table 6. 2007a.6%.2. page 28). as well as the calibrated share of capital in a product ( — parameter of CobbDouglas production 6 As an example. During the second stage of the estimation procedure. page 19). 2008. 2005b. The division depends on the estimated parameters characterising the growth rate of technical progress µz and µΨ .000 elements. all the nominal categories grow at an identical rate determined by the calibrated parameter µ. Geweke tests for the particular parameters. and for reasoning a sample of 500.000 elements (with ﬁrst 50. etc.000 elements remaining upon the rejection of the ﬁrst 250.000 elements from a chain of 10. suggesting lack of convergence. Reviewing the descriptions of other Bayesian estimated models. are important — they impact the characteristics of the model and its forecasting features. the authors of the SOEEuro model generated a chain of the length of 700.
see Adolfson et al.5% for the USA.025424 0. Also the share of trade in GDP may raise doubts. for the USA. discount and the rate of tax on capital in steady state. by and large. all of the transactions have the horizon of one quarter. Value 0. In reference to the previous version of the SOEPL model this is. 9 For inﬂation this was 0. generating at the same time larger ex post forecast errors. with higher value dynamics µ. whilst expecting volume growth dynamics at the level of 34% and inﬂation at about 2%.070696 0.041181 0. which is below the values that we have observed (we expected the value of about 40%).25 in the USA. steady state function) and amounts to about 2. so there are reasons for underestimated share of trade in GDP . optimisation.5% of inﬂation and about 4. In the Appendix we present an example forecast with a very long horizon. In order to at least slightly compensate for this inconsistency.3. Euribor). loans. even if the discount implies the minimal weight applied to the future and the tax on capital is quite arbitrarily assumed at low values (or even zero. which (among others) reﬂects the rate of convergence toward the steady state. The values of the most important variables and proportions in the steady state are presented in Table 6.626330 0. a considerable progress (we earlier had less than 29%) achieved thanks to the explicit consideration of import intensity of exports. In all the DSGE models we are aware of.1% corresponds to those values.183670 0. the model continues to omit import intensity of government consumption.25 p. in the euro area and 0. deposits.050825 0. 91 WORKING PAPER No. brought about higher inﬂation (instead of growth). Formally. 6 Table 6. for the euro area and about 1.4% for the euro area and 2. Within that logic.190000 0. 83 91 . 8 Based on the structure of the model. (2009.p. in the block of measurement equations of the model adjusting constants have been introduced.1% of volume dynamics7 . the steady state level of interest rate has very high values. 3. no difference exists for the rate of the central bank.p. There is only one interest rate present in the model. Corresponding adjustments have also been introduced to the GDP growth rate and inﬂation in the euro area and in the USA9 . page 29)). short. e. interest rates in Poland shall be by about 2 p. higher than the rates in the euro area in steady state. In steady state net exports amount to zero and the share of import in GDP has the level of about 36%.359007 Comments annualised µz + annualised π R gr 7 The carried out experiments.332 p. for the annualised GDP dynamics it was 2. while observable equivalents are threemonth rates in the interbank market (Wibor. Also the received approximation of the marginal likelihood did not provide arguments for such solution. It has been determined by experiment (taking into account the quality of forecasts and marginal likelihood) that adjustment compliant with data for the domestic interest rate amounts to 1.p. Nevertheless. Values of selected variables and the great ratios in steady state Parameter Annual growth rate of technological progress Annual growth rate of total investments Annual inﬂation rate Annual interest rate Share of consumption in GDP Share of investments in GDP Share of government consumption in GDP Share of import (export) in GDP Source: Prepared by the authors.00 p. it is hard to arrive at interest rate lower than 56%. Therefore. this is so because its level derives from the value dynamics (µ). The annual interest rate of about 7.35%.p. however.3.or longterm rate8 .g.Assessment of parameters — calibration.
The results of experiments have not conﬁrmed such assumptions10 . also arrived at large differences of Calvo probabilities and indexation parameters (see Jakab and Világi. while change appears at the beginning of the sample. we decided to calibrate the parameters for the ﬁrst regime. obviously. limited and several hypotheses remain open. rationalised by reference also to subjectmatter arguments. We think that the simpliﬁcation may be. affected the level of wage markup in steady state and the characteristics of the stochastic process governing this markup. 2007).Assessment of parameters — calibration. see e. It is also proposed to estimate two models in separate subsamples (Smets and Wouters. which — in our opinion — affected the exchange rate risk premium shock (characteristics of a stochastic process). interest rate rule parameter values may have changed. Additionally. In both cases. We have assumed that due to ofﬁcial introduction of direct inﬂation targeting in Poland in 1999.2 Structural changes The SOEPL−2009 model allows for structural changes regarding monetary policy. we assumed for simplicity the simultaneous occurrence of such regime changes as from the second quarter 1999. enabling at the same time slightly better identiﬁcation of parameters in the second regime. with stricter control of the ﬂow of contributions and pension entitlements. Lou et al. which has an advantage of decreasing the heterogeneity of the sample. Searching for further structural changes. From a formal point of view. This concerns also characteristics of inﬂation target disturbance. Finally. The introduction of the social security system reform (affecting the labour market) and changes in the strategy of monetary policy was announced earlier and in the world of rational agents everybody was aware of that.. Therefore. As a result there are two regimes present in the model.e. The actual consequences of such activities were not known even to the authors of the reforms and in that sense the 10 The scope of the experiments made by us was. the authors of the Hungarian DSGE model. In the abovepresented tables. when estimating their model for two regimes of monetary policy. we have assumed the simpliﬁcation. respectively 1 and 2 for the given regimes. As an example. 2006. better reﬂecting the historical events.g. exchange rate regime had gradually been shifting towards free ﬂoat. 2007. A drawback of the assumed solution is. 6 92 92 N a t i o n a l B a n k o f P o l a n d . leads to a considerable elongation of the time of computation. 2005). as the possible moment of further structural changes (covering i. optimisation. Already before introduction of targeting. based on the results of experiments made earlier. (Justiniano and Primiceri. budget expenditures. the changing parameters have been identiﬁed with subscripts. The experiments showed that more reﬁned deﬁning of moments of change. Sims and Zha.3. the labour market. A more radical solution — estimation starting from the year 2000 — seemed to us to be too costly. however. 11 Other authors model structural changes with the use of stochastic switching Markov chain. 2008. Yet. however. This simpliﬁcation results from the applied technique11 . however. In Table 6. steady state 6. we assume that introduction of a new pension system and health insurance system in 1999. unforeseeable by the agents. the method of representation in the model of the aforesaid structural changes gives the changes deterministic nature. a dozen or so additional parameters to be estimated or calibrated. The information had both a general and qualitative nature. practically does not affect the marginal likelihood and the forecasting features of the model. exchange rate policy and labour market characteristics. page 21). we have not veriﬁed that issue.4 we present a speciﬁcation of parameters whose values change in effect of structural changes. effects related to exchange rate). however. the year of Poland’s accession to the European Union was also tested.
Assessment of parameters — calibration. the RAMSES model (case with UIP condition analogical as in the SOEPL−2009 . λw ρλ w σλ From the data provided in the Table 6. In Table 6. steady state resultant of all of the changes was unforeseeable.7921 2.4 it appears that the changes in monetary policy discussed here have been interpreted by agents as the increase in interest rate sensitivity to inﬂation ﬂuctuations (also in reference to other determinants of the interest rate deﬁned in the rule). SOEEuro model of the euro area (case with variable utilisation (see Adolfson et al. there have been provided assessments of analogical parameters (modal value) received from the earlier version of the SOEPL model (see Grabek et al.8204 rπ 1.7959 0. (see Adolfson et al. particularly.1783 0. discussed earlier in more detail (see e. For comparison.5839 0.4658 ¯ Source: Prepared by the authors.3482 0. At the same time the variance of exchange risk premium innovation increased with simultaneous decrease in the disturbance’s persistence. 6 6. On the other hand. Parameters changing values as a result of structural changes Parameter Regime 1 (calibrated value) Regime 2 (optimised value) Wage markup 1. after the year 1999 the inﬂation target (innovations) has also featured larger variance.5712 ˜ σΦ 1. However.1949 R σπ c 0.4.1504 0. page 25). we have applied the Calvo model supplemented with the mechanism of dynamic indexation.. 2007).0219 ry 0.3802 w Risk premium ρΦ 0.1284 σε 0.0100 0..1366 1.3.6431 0. where tightening of the insurance system meant the increase in the level of markup with decrease in the innovation variance and persistence of the disturbance.3 Nominal rigidities A distinguishing feature of a DSGE model based on the paradigm of a new Keynesian school is the presence of nominal rigidities — delays in the adjustment of prices. equation (4. The effects of such changes for the impact of the disturbances on observable variables are illustrated by variances decompositions and reaction functions presented further herein. 2007a.17)).7995 rx 0. while speaking about the values of deep parameters. Table 6.5 there are speciﬁed the assessments of parameters characterising Calvo rigidities.7359 0. The effect may be compensated by deﬁning a set of anticipated disturbances — yet. we have left that issue for separate study. While modelling the rigidities of prices (wages) in the SOEPL−2009 model.4497 1. optimisation.6774 0.1727 0. More transparent seem to be the changes in the labour market.1293 ˜ Interest rate rule and inﬂation target ρR 0. 93 WORKING PAPER No. 83 93 . the phenomenon of the growth of general uncertainty of the agent actions in face of the expected complex structural changes has been omitted. With increase in the variance of interest rate shock. also the persistence of the interest rate increased.g..
Assessment of parameters — calibration.920 0..883 0. page 43) mention that Calvo probability ξ and the average time between price re1 optimisation 1−ξ implied by the value does not necessarily inform about the scale of nominal rigidities because in the Phillips curve built on the basis of the Calvo model.765 0.417 0. 2007a.711 0.944 0. steady state 2005b.453 κd 0.164 0. page 58).148 Source: Prepared by the authors. Rigidities of prices and wages in the chosen DSGE models Parameter SOEPL−2009 SOEPL−2006 RAMSES SOEEuro NAWM (EBC) 0. as well as the values characterising rigidities for the Hungarian economy (see Jakab and Világi. that the SOEPL−2009 model describes economy with a more ﬂexible wages and prices than the models of the euro area.291 0. Obviously.617 0.600 0.683 0. and NAWM ECB (see Christoffel et al. The difference is particularly visible in the labour market. so with slightly longer periods between the subsequent reoptimisations indexation takes place mainly based on the expected (instead of former) prices. The cited values comply with the modal value of posterior distribution or median (RAMSES). page 23) for the case of carrying out the strategy of direct inﬂation target.313 0.439 0.585 0.173 κmc 0. detailed information in the text of the paper. page 82).530 0.630 Dynamic indexation parameter κw 0. 2008.675 0. From the Table 6. (2007a. so we have no grounds to claim that the prices setting mechanism has changed e. 94 94 N a t i o n a l B a n k o f P o l a n d .558 0.523 ξmi 0.128 κmi 0.417 0.489 MNB model 0.752 0.5.350 κx 0.g. generally.441 0.838 0. optimisation. the change results also from a change in the speciﬁcation of the model. model: the version estimated in 2006 reﬂected relatively stronger adaptative indexation of the prices of domestically manufactured products. Interesting are also changes in the assessments of parameters in the subsequent versions of the SOEPL .552 ξx 0.350 0. more persistent indexation mechanism lowers the effects of shorter delay in reoptimisation of wages.635 0. however.192 κmx 0.5 it results.716 ξd 0.794 0.586 0. Sweden or Hungary — at least prices are less susceptible to marginal costs’ changes12 .202 0.770 0.434 0. after the accession to the European Union.508 0.440 0. now the κd parameter characterising the effect of inertia is minimal.895 ξmc 0.680 0.  12 Christoffel et al.901 0.743 ξmx 0. Table 6.827 6 Calvo probability ξw 0. real and nominal rigidities may not be identiﬁed (discriminated) separately.
Variance decompositions.2) and regime II characterising the relationships observed in the other part of the sample (Tables 7. a slightly lower value of the parameter (0. The horizon has been limited to shortterm and mediumterm period (maximum 20 quarters). 83 95 . the results depend on model speciﬁcation.1–7. all of the calculations were made for parameter values corresponding to the modal value of posterior distribution determined at the stage of optimisation (we omit the uncertainty of parameters). impulse response functions and estimation of disturbances Chapter 7 Variance decompositions.985 deteriorates the quality of forecasts. which from the point of view of the current applications of the model seems to be sufﬁcient.4) we have presented variances decompositions (formally — forecast errors). impulse response functions and estimation of disturbances 7.985). impulse response functions and estimation of disturbances 7 Variance decompositions.975–0. while the rest of them — with disturbance of the largest persistence. It is the real dollar/euro rate that currently determines the joint persistence of the observable external disturbances. the largest persistence is reﬂected by foreign variables which constitute the SVAR1 . this does not seriously affect the steady state or short.1–7. separately for regime I covering the period up to the ﬁrst quarter of 1999 (Tables 7.975–0. As we have mentioned before.3–7.and midterm model 1 The aforementioned experiments (sensitivity analyses) showed that autocorrelation of inﬂation target disturbance at the level of 0.900) has been assumed. Certainly. Higher persistence has been achieved by estimation of the characteristics of AR(1) process for an asymmetric shock and by constructing relations between world’s economy SVAR model and the basic part of the model (real dollar/euro rate). 95 WORKING PAPER No. For longer horizons. as well as in the SOE Euro 7 model.1 Variance decompositions In the following Tables (7. In the earlier versions of the SOE PL model. this was the disturbance of the inﬂation target (the calibrated autocorrelation coefﬁcient amounted then to 0.2). therefore. Real growing variables are determined with nonstationary disturbances. In the current version.
Additionally.). Along with the extension of the observation horizon (8–20 quarters). the currently observed results are quite interesting. Also a moderate inﬂuence of ﬁscal disturbances may be perceived in midterm horizon (8–12 quarters).Variance decompositions. According to the data provided in Table 7. the role 7 of ﬁscal shocks with regard to CPI is visible. it still seems to be underestimated. The block of external/foreign shocks covers all of the (observable) disturbances characterising the euro area and the USA. shortterm consumption is determined by disturbances in consumption preferences. markups on the import of export components). pages 100–101) with the current version of the model.1. Along with the extension of lag. however. the inﬂation target disturbances and technological disturbances (stationary in investments) is growing.1.1 Regime I In the short run (1–4 quarters). a dominating role is played by stationary technological disturbances in investments (nonstationary disturbance in a longer horizon). The data gathered in the Tables present the role of 17 single structural disturbances and two blocks of shocks. this is — at least partially — caused by the very structure of the DSGE model (a Ricardian household. present in the SVAR model.. The growth of importance of ﬁscal disturbances is also visible. impulse response functions and estimation of disturbances dynamics we are the most interested in. with a lag of 4–12 quarters. Yet. Further. markups in the market of domestic products (GDP deﬂator. In the same regime. the growth of importance of the aforesaid shocks is clearly visible. as well as by disturbances in the wage markup (here we observe an inverse trend). with the maintained role of shocks originating in the labour market. Therefore. Along with the longer horizon of analysis. the impact of technological disturbances (stationary and nonstationary) has been growing. markups on domestic intermediate products and wage markups manifest themselves. In the case of investments. GDP depends (on shortterm basis) on external factors (which quickly abate) and on domestic shocks related to foreign trade (exchange rate risk premium. In the block of ﬁscal disturbances all of the disturbances modelled with the use of SVAR model and additionally structural disturbances of national insurance contributions and capital tax have been jointly reﬂected. etc. Comparing variances decompositions received in the previous versions of the SOEPL model (see Grabek et al. 7. although we think that this time the effects have been overestimated. the main inﬂation determinant before 1999:2 was the situation in the labour market (wage markups). The ﬁrst general observation refers to the scale of impact (share) of ﬁscal variables and foreign variables on the observable variables. Additionally. The marginal role of impulses originating from the world’s economy was one of the most important deﬁciencies of previous versions — the effects observed earlier (at the level of 1–3%) were inconsistent with the concept of a small open economy. 96 96 N a t i o n a l B a n k o f P o l a n d . ﬁscal factors. no resource effects related to public debt. the impact of stationary technological disturbance and nonstationary disturbances in investments has been growing. the role of which abates in the subsequent quarters. CPI) and markups on the imported investment goods (investment deﬂator). the impact of shocks on wage markup and consumption preferences may be observed. the impact of disturbances originating from the world’s economy. 2007. export markups.
generally the role of these disturbances (compared to regime I) clearly decreased. impulse response functions and estimation of disturbances The disturbance of markups in the labour market — in regime I — is the most important determinant of employment and real wages. On shortterm basis (1 quarter) important are also: stationary disturbance in technology. risk premium). markup on products manufactured domestically (markup of imported investment goods in the case of investment deﬂator). while none of them brings about a stronger impact. Employment ﬂuctuations are also triggered by the shocks of consumption preferences and labour supply (2–12 quarters).g. The lowering importance of wage markup disturbance is also conﬁrmed in the case of consumption and investments. export markups and markups in consumption import. 83 7 97 . export markup. 7. to some extent.1. We only observe — as in the previous cases — a lower role of wage 97 WORKING PAPER No. nonstationary) and disturbance in labour supply preferences are more important. In a longer horizon — as in regime I — consumption clearly depends on ﬁscal and technological shocks. stationary investment disturbance remains the main determinant on short. in the short run inﬂation depends on the inﬂation target. particularly in shorter horizons. Roughly 8–9 shocks (e. exchange rate risk premium and internal disturbances. this time technological disturbances (stationary in investments. The real exchange rate of US dollar and the interest rate seem to be dominated by external disturbances along the whole horizon. exchange rate risk premium and export markups are important as well. by the disturbance in labour supply preferences. inﬂation target shocks) have the remaining 50% share in the variance (of forecasts errors). I. “Fundamental” domestic factors affecting the real exchange rate of the dollar on a midterm basis are clearly missing. Impulses coming from the labour market are slightly more visible with a lag of 412 quarters. ﬁscal shocks. The importance of nonstationary shocks is growing along the horizon. and in the short period the interest rate shock. In regime II. Yet. consumption is clearly impacted only by the shock of consumption preferences and domestic products markups.2 Regime II In regime II.and midterm basis. technology is a dominating factor. The role of wage markup disturbance has been taken over. also wage markup is important.Variance decompositions. export markups. however. external shocks. Conclusions formulated when discussing factors affecting employment and real wages in regime I remain valid for regime II. For investments. In the horizon of 1–4 quarters. stationary shocks in technology. in a short period of time GDP variance results from disturbances in the markup of products manufactured domestically and disturbances related to foreign trade (external disturbances. The real exchange rate in the short run (1–4 quarters) reacts to exchange rate risk premium. particularly in a midterm horizon. Interest rate reacts also to ﬂuctuations in labour market markups and inﬂation target. Real wages ﬂuctuations result (beside of wage markup shocks) in a shortterm period from ﬂuctuations of labour supply preferences and domestic product markups. in the longterm horizon). Drop of importance of wage markup is also observable for GDP Similarly as in regime . ﬁscal shocks (8–20 quarters) and technological shocks (stationary in investments. ﬁscal disturbances (in the case of CPI) and foreign variables (whose impact grows in time). in the shortterm horizon (1 quarter). In a longer period.
wage markup disturbance is nonidentiﬁable from the point of view of speciﬁcation of the structural form of wage equation — discrimination requires relatively extensive expert knowledge (introduced by the speciﬁcation of prior distribution at the stage of 7 estimation). is small (3.0% for 1–12 quarters). we may talk about the role of markups in export. investigating the structure of the impact of disturbances in both regimes in a general way. Thus. in further quarters a drop may be observed). so it is hard to talk about tangible evidence. the consequences of the adoption of the strategy of direct inﬂation target are bit clearer. external disturbances are dominating. despite hardly spectacular manifestation (see the respective values of parameters in Table 6.4). As before. In some cases the function of wage markup disturbance was taken over by the disturbance in the labour supply preferences. impulse response functions and estimation of disturbances markup disturbance. which may suggest that institutional phenomena have been supported by phenomena related to preferences. the impact of wage markup disturbance is lower. even in the shortterm horizon.6% for the ﬁrst quarters and dropping) — it is smaller than the inﬂuence of ﬁscal disturbances (about 5. while higher is the share of external disturbances. consumption goods import and import of export components as well as risk premium. Clearly. In a very short horizon (1–4 quarters). The inﬂuence of interest rate disturbance. Nevertheless. 98 98 N a t i o n a l B a n k o f P o l a n d . with only a partial takeover of the impact by the disturbance in labour supply preferences. it is hard to perceive larger changes in the group of the most important shocks affecting the real exchange rate of the dollar. The importance of inﬂation target disturbance grew and the impact of monetary policy shock is stronger (but only in the ﬁrst quarter.Variance decompositions.9–5. we see that from the point of view of variance decompositions the essence of changes that occurred in 1999 was a sort of breakdown in the institutional structure responsible for setting the price of labour. This phenomenon seems to be the most important. Although one of the phenomena motivating separate treatment of the two regimes were changes in the currency market. On the other hand.
7 12.8 0.1 1.4 0.2 13.8 0.1 0.0 0.4 2.6 2.9 15.3 6.2 1.8 0.8 3.1 5.3 7.7 2.1 5.1 0.0 5.8 0.1 0.1 0.5 1.4 0.6 13.7 5.3 0.1 2.9 35.5 13.6 10.8 15.9 6.4 0.6 13.3 0.3 0.8 0.9 20.6 7.7 1.4 1.9 0.7 0.5 2.6 4.3 17.5 5.9 0.0 0.5 0.5 0.6 0.2 0.2 13.4 0.3 25.4 1.1 0.8 7.0 1.8 3.1 3.0 2.6 0.4 10.2 2.9 1.5 1.3 0.9 7.7 5.4 19.2 0.7 7.8 0. impulse response functions and estimation of disturbances 99 Disturbance Stationary technology Investment speciﬁc technology Asymmetric technology Nonstationary technology Investment nonstationary technology Consumption preference Labour supply Domestic goods markup Imported consumption goods markup Imported investment goods markup Imported export component markup Export markup Wage markup Risk premium Interest rate Inﬂation target Energy prices Fiscal World economy (SVAR) 1q 0.2 0.5 0.3 11.7 1.3 0.9 6.9 6.6 0.4 11.6 1.3 1.9 1.4 0.4 4.5 1.6 2.8 6.1 0.1 0.9 1.2 0.9 1.4 0.9 0.1 6.4 12.4 1.6 3.3 0.3 17.1.2 3.5 38.9 4.3 17.0 0.3 12.2 2.9 8.5 Investment 8q 3.7 0.9 0.1 0.7 0.0 8.5 0.0 1.7 1.0 15.1 0.4 1.1 10.6 2.9 6.0 1.3 2.5 3.1 7.7 5.0 4.6 GDP 8q 3.2 0.6 76.6 5.5 0.4 5. Variance decompositions.6 18.1 0.0 0.9 20.0 10.1 2.7 1.3 1.5 1.8 0.1 0.5 4.6 6.5 0.5 1.3 0.3 0.2 39.9 3.5 0.1 4.9 2.7 4.2 3.7 7.7 1.7 3.6 12.1 12q 4.2 0.0 2.5 1.9 4.0 17.3 4.9 4.1 0.1 4.1 1.7 3.8 1.0 4.3 3.0 3.9 34.1 0.2 2.1 8.2 1.4 30.7 27.7 Exports 8q 0.5 26.9 4.6 1.0 1.1 1.6 3.3 19.5 0.5 9.1 7.8 1.5 0.6 12.4 0.5 6.3 0.6 6.5 7.7 36.9 3.2 2.4 8.4 23.6 1.5 0.7 0.1 1.9 11.3 8.6 Consumption 4q 8q 12q 5.2 10.7 1.5 27.3 1.0 4.1 0.5 0.7 3.9 1.7 7.5 0.2 0.3 16.5 1.7 2.2 Imports 8q 2.6 30.6 6.1 3.5 1.6 20q 0.3 1.2 8.9 0.7 1.2 2.6 2.0 40.3 3.5 6.9 55.8 0.2 3.7 1.5 0.4 15.4 4.6 0.3 0.4 1.7 1.3 3.5 5.7 3.7 0.5 3.1 2.3 3.5 2.1 2.8 5.1 3.3 0.7 11.8 30.0 0.3 7.0 Variance decompositions.1 20q 0.6 3.4 0.6 0.1 5.3 0.6 3.6 1.0 0.6 29.0 0.5 4.1 0.5 5.1 2.2 4.7 0.2 1.3 0.0 18.8 3.7 0.0 3.4 1.2 3.1 2.3 1.7 1.3 3.2 0.0 6.8 10.5 2.2 7.0 0.6 1.9 4.8 2.9 0.7 14.7 1.7 0.2 0.2 8.3 1.3 4.7 85.2 0.8 0.5 0.7 20q 1.7 3.4 0.5 0.5 4.3 8.8 2.3 23.8 5.9 0.7 2.9 8.3 4q 5.5 0.1 0.5 1.2 5.6 12q 1.0 3.8 2.7 2.5 4.4 0.3 4q 3.1 4q 1.7 12q 2.8 5.4 1.5 9.2 1.8 12.5 0.2 0.0 1q 1.1 2.3 0.4 5.3 7.0 3.3 9.0 3.3 0.4 1q 3.2 2.6 7.8 7.1 12q 3.8 1.5 0.4 0.2 7.3 3.5 1.9 0.3 0.7 1.5 0.0 0.8 7.0 5.2 0.6 2.2 0.9 2.2 0.1 0.0 0.0 23.8 2.5 15.1 0.3 0.9 4q 6.3 13.8 1. part 1 WORKING PAPER No.4 8.7 5.9 99 7 .6 5.8 4.7 0.1 3.2 0.4 2.8 5.8 0.5 2.2 1.3 1.5 9.2 1.3 11.2 8.5 21.5 16.4 2.4 1.7 0.8 2.5 0.2 1.8 4.7 Real exchange rate 4q 8q 12q 1.9 3.0 4.7 8.4 5.4 14.2 29.3 0.5 6.4 2.1 51.2 5.7 3.5 1.6 1.3 2.2 1.7 1q 4.5 1.3 11.2 2.7 5.7 0.0 12.9 1.9 5.3 25.3 16.0 20q 1.8 Interest rate 8q 12q 4.1 0.6 0.4 0.7 2.8 1.0 1.1 4.2 0.1 17.2 20q 1.8 2.6 0.4 0.5 0.7 5.8 2.1 5.7 0.2 3.0 0.5 2.5 2.4 6.6 4.2 0.2 1.3 2.7 1.2 6.6 6.9 2.8 1.2 9.5 0.7 0.3 23.3 28.8 7.0 8.3 0.0 3.1 0.8 0.1 0.9 3.7 5.2 1q 1.2 12.6 1.1 2.1 0.1 8.6 4.0 20q 1.9 23.6 6.6 0.1 7.6 12.0 4.9 2.8 0.9 9.8 1.7 2.3 0.0 6.6 1.6 1.5 1.4 1q 0.1 3.0 19.3 0.4 1.4 0.8 1.0 5.1 1.8 8.8 1.2 0.5 0.4 0.1 3.1 8.4 0.4 0.4 7.3 0.3 4.8 0.4 0.2 1.7 10.2 0.2 7.3 5.8 6.8 0.2 0.1 19.3 0.6 1.6 4.2 5.5 27.9 10.9 14.9 1.7 0.6 4q 1.1 0.5 0.5 2.0 8.7 0.4 0.8 2.0 0.2 11.8 11.9 0.5 0.7 5.5 1.2 0.0 0.5 6.8 27.1 19. Regime 1.5 8.3 27.0 0.4 2.2 2.2 10.8 26.0 25.0 5.9 3.4 2.3 0.9 0.2 0.2 5.0 0.1 3.2 93.0 2.3 0.4 2.0 28.3 0.5 0.1 3.4 0.0 5.4 7.9 0.2 6. 83 GDP deﬂator 4q 8q 12q 5.7 5.9 0.5 4.0 8.3 0.0 0.7 1.5 3.2 1.2 0.4 5.2 0.6 1.1 0.7 7.0 20q 1.3 5.7 2.8 7.0 4.8 20q 0.1 1.3 7.4 3.2 16.7 10.2 0.2 1.3 0.6 3.3 1.1 0.9 1.4 6.1 1.3 0.1 0.3 0.5 6.1 0.9 2.2 0.0 0.0 0.1 3.6 3.2 3.1 0.4 23.1 0.8 2.8 6.2 0.0 6.7 3.6 1.4 1.0 0.0 11.5 3.7 2.6 3.4 2.5 0.9 21.2 11.4 1.7 12.9 28.0 6.1 0.8 1.5 1.2 2.0 2.4 1.0 6.4 2.2 0.4 1q 0.9 5.9 4.8 2.2 2.6 2.1 19.0 0.3 1.8 0.8 0.9 10.4 0.7 1.3 4.6 3.0 0.0 1.Table 7.5 0.6 2.6 2.9 8.3 1.7 0.2 0.7 1.3 3.8 1.3 39.6 10.9 2.1 16.2 9.1 3.9 2.8 1.2 6.5 0.5 3.9 1.2 16.5 5.7 0.3 8.3 1.3 24.3 1.6 4.6 Disturbance Stationary technology Investment speciﬁc technology Asymmetric technology Nonstationary technology Investment nonstationary technology Consumption preference Labour supply Domestic goods markup Imported consumption goods markup Imported investment goods markup Imported export component markup Export markup Wage markup Risk premium Interest rate Inﬂation target Energy prices Fiscal World economy (SVAR) 1q 9.9 0.3 2.2 0.9 0.5 4.6 2.
6 0.5 1.1 5.4 4.0 3.9 4.6 6.0 1.8 0.3 0.2 0.9 1.5 9.9 2.5 0.4 1.1 5.0 2.6 1.9 13.5 9.1 12.8 0.7 0.6 0.1 17.7 0.3 29.1 0.9 0.9 0.0 13.2 12.9 5.1 2.4 3.7 1.1 1.4 9.1 0.4 0.7 0.8 18.6 24.8 30.7 0.4 0.3 2.7 1.1 2.5 0.0 1.6 0.8 5.1 2.7 20.7 16.7 12.2 6.6 3.7 4.6 11.0 0.3 4.3 5.3 8.1 3.7 2.0 0.3 31.1 6.7 5.5 8.8 0.4 4.8 7.7 3.3 5.1 2.5 9.6 10.4 7.2 44.6 2.3 2.1 0.2 1.6 7.4 15.9 1.1 3.6 32.7 1.4 4.9 14.5 1.3 3.6 12q 5.8 0.9 4.8 0.1 0.7 100 Employment 4q 8q 12q 5.5 0.6 0.2 0.0 0.4 0.3 6.0 0.8 3.7 12q 0.5 0.5 1.5 3.6 4.0 2.3 3.8 0.9 1.7 16.5 0.5 1.7 3.6 1.1 0.9 2.3 0.7 4.1 3.0 0.4 2.9 0.5 2.3 100 N a t i o n a l B a n k o f P o l a n d .1 0.3 1.6 Real wages 8q 6.7 0.3 1.4 2.2.3 18.1 11.6 0.0 18.1 0.9 Investment deﬂator 4q 8q 12q 5.9 19.3 0.6 1.4 4q 4.5 4.5 1.7 2.1 4.3 0.0 3.4 13.9 0.0 0.1 29.2 0.1 18. impulse response functions and estimation of disturbances Table 7.1 0.1 6.9 4.7 8.6 2.2 7.4 2.6 0.2 16.3 5.2 9.0 12.1 5.2 2.2 1.6 3.8 5.4 0.8 1.1 0.4 2.1 2.9 0.0 3.5 32.3 0.2 0.6 1.0 0.0 5.2 8.5 3.1 0.4 0.9 13.4 5.2 1.9 3.5 3.8 10.6 27.7 5.2 0.2 0.8 8.2 2.3 0.4 5.4 8.2 11. Regime 1.5 Variance decompositions.0 14.2 2.4 11.1 0.5 6.4 0.2 1.0 0.8 1.7 2.8 4.5 1q 3.0 2.6 0.5 0.1 3.6 2.5 1.2 0.3 1.8 3.2 0.0 0.5 4.3 1.6 0.4 6.6 10.1 6.1 8.0 0.6 11.2 0.8 0.7 9.2 0.5 4.4 0.5 2.2 4.3 2.6 1.3 0.7 1.0 0.2 1.5 3.4 17.7 2.3 41.3 0.4 0.3 0.0 3.2 2.2 0.0 1.3 1.4 25.9 1.5 20q 1.3 1.9 8.4 11. part 2 Disturbance Stationary technology Investment speciﬁc technology Asymmetric technology Nonstationary technology Investment nonstationary technology Consumption preference Labour supply Domestic goods markup Imported consumption goods markup Imported investment goods markup Imported export component markup Export markup Wage markup Risk premium Interest rate Inﬂation target Energy prices Fiscal World economy (SVAR) 1q 12.0 7.0 19.4 0.0 0.8 1.2 9.8 3.4 0.6 2.0 10.4 0.0 1. Variance decompositions.1 9.1 1.7 0.1 0.8 1.2 0.9 0.2 4.3 0.3 2.2 2.2 6.8 1.7 0.8 4.2 5.1 0.4 1q 4.2 2.2 1.5 1.3 1.5 3.7 2.7 0.9 20q 0.3 9.9 0.9 0.6 0.4 CPI 8q 1.2 1.8 0.8 24.8 0.6 8.9 2.5 0.9 42.4 1.2 2.0 4.9 0.1 1.9 9.0 15.3 0.0 0.6 0.3 3.3 0.8 34.4 1.1 10.7 20q 1.5 1.5 1q 6.7 1.6 1.7 7.1 9.8 10.2 40.5 3.8 0.7 2.5 11.9 26.8 1.5 8.8 0.7 0.0 5.5 2.6 0.6 15.6 40.8 2.7 2.0 0.2 6.3 18.4 3.2 20q 3.6 3.3 1.5 5.2 4q 5.5 20.5 2.1 1.
4 1.8 0.5 15.4 5.7 0.0 0.2 4.9 1.4 10.2 0.3 0.7 2.2 0.7 2.3 5.2 1.5 6.0 0.7 8. 83 GDP deﬂator 4q 8q 12q 4.8 57.0 6.6 20.1 0.9 1.6 5.1 0.9 6.2 0.7 7.4 11.8 6.5 1.2 29.2 20q 0.1 6.4 3.9 1.0 0.4 2.5 2.8 1.0 9.4 3.8 0.0 2.0 34.0 0.2 3.0 2.9 0.0 2.5 6.7 2.1 1.0 Imports 8q 3.4 1.4 1.8 6.2 3.5 1.9 0.0 2.2 0.2 8.7 11.6 13.1 2.6 2.2 2.6 0.3 2.0 18.6 1.6 1.4 25.3 0.5 1.1 7.7 1.0 0.8 20q 2.9 10.4 3.4 1.6 1.7 1.5 Exports 8q 0.2 5.7 0.7 1q 1.4 19.8 3.7 0.5 3.8 8.1 0.4 5.2 2.2 3.1 15.4 1q 1.2 14.5 3.7 Variance decompositions.3 1.0 1.2 1.6 0.0 0.3 8.8 0.8 6.4 Investment 8q 4.3 2.6 1.0 9.9 2.8 0.1 0.1 Real exchange rate 4q 8q 12q 1.2 49.2 1.7 2.7 2.9 2.Table 7.0 4. part 1 WORKING PAPER No.3 0.3 0.6 2.0 0.5 4.6 1.5 0.8 2.7 4.9 0.0 2.8 5.3 4.3 0.4 0.5 13.1 2.5 5.0 0.7 6.3 0.8 50.2 76.2 0.3 6.5 1.1 0.6 20.6 3.1 0.4 1.5 6.2 2.2 0.8 2.2 0.0 1.0 GDP 8q 6.1 7.2 5.6 86.4 3.1 11.1 0.2 0.1 0.5 0.5 11.0 4.0 20q 0.3 1.2 4q 1.2 0.2 5.9 0.8 0.4 0.3 1.2 0.4 20.4 0.2 6.5 12q 1.9 7.3 0.6 7.2 1.7 0.2 0.2 0.5 3.4 0.6 0.9 1.1 1.7 0.3 0.0 0.6 0.8 4.4 5.3 2.5 9.3 0.4 8.7 2.6 14.7 0.3 0.8 0.1 2.8 1.3 4.8 1.3 2.4 12.2 21.3 0.5 3.4 1.6 8.0 9.2 21.4 0.4 0.6 6.9 1.9 5.6 0.1 0.1 8.7 4.8 0.6 3.1 3.2 4.8 0.5 1.3 38.8 12q 4.7 0.1 4.2 4.5 0.6 5.9 0.7 2.4 12.6 19.2 0.8 2.4 4.7 7.5 2.6 25.9 0.1 1.7 7.5 4.9 0.2 6.3 5.8 2.9 3.9 3.8 6.6 0.7 1.6 0.3 0.4 7.3.5 0.6 2.0 0.1 21.2 7.7 3.8 1.2 0.9 9.0 0.5 0.0 0.3 1.0 0.9 0.0 0.4 1.0 0.9 7.4 0.2 0. Variance decompositions.4 4.1 2.1 2.5 0.4 5.3 5.3 3.1 4.3 5.0 0.4 33.8 4.2 8.7 0.6 12q 2.6 4.6 2.6 3.5 1.1 8.5 1.7 7.6 1.1 5.3 63.4 2.0 13.0 0.7 5.3 16.6 101 7 .7 1.7 23.6 6.5 14.3 8.2 0.1 17.3 2.3 0.0 0.3 0.3 1.4 2.5 12.3 24.2 1.3 3.0 1.7 0.2 11.9 4.5 0.1 4.1 0.1 6.7 2.2 0.2 1.9 2.1 0.1 3.6 1.5 2.7 0.2 2.3 10.5 5.3 1.7 0.8 4.9 20q 0.4 0.0 3.7 4q 8.6 6.9 0.3 15.0 0.6 9.6 2.8 0.2 0.4 6.2 2.5 10.2 0.6 1.4 0.2 1.3 3.6 0.7 0.5 6.1 0.7 3.1 0.1 0.0 5.4 Consumption 4q 8q 12q 6.9 0.6 1q 3.8 6.0 0.0 18.3 0.1 0.1 0.9 4q 1.8 9.9 1.6 1.1 3.0 31.0 6.0 1.5 13.2 2.9 10.7 10.0 10.5 1.2 4.1 0.4 9.4 2.1 1.2 4.9 30.6 2.5 2.2 4.7 1.2 4.1 0.2 16.7 3.6 0.7 0.0 0.8 1.7 7.3 1.6 1.8 0.0 5.5 26.2 2.2 4.3 1q 4.5 0.4 1.8 17.2 9.0 1.6 0.7 0.1 4.9 13.9 2.6 1q 0.2 0.9 0.7 1.7 4q 5.2 1.8 3.0 2.1 1.1 0.0 6.0 3.8 1.3 4.6 4.3 1.1 8.9 0.4 4.4 12.3 20q 0.5 93.1 0.3 0.7 1.2 26.0 0.1 13.9 0.5 4.4 3.1 4.9 10.3 0.7 6.6 0.3 2.0 1.2 20q 0.6 2.9 9.6 0.2 0.2 29.3 0.7 12.1 1.7 0.9 14.9 5.7 6.3 7.9 5.7 3.8 7.5 3.0 4.9 6.0 2.9 3.1 5.3 9.7 28.4 10.1 0.0 2.9 20q 2.0 2.1 20.9 1.5 13.0 2.5 3.0 18.7 2.4 0.3 20q 2.9 3.5 Interest rate 8q 12q 4.1 4.7 4.7 0.5 0.2 0.3 0.7 3.3 8.9 1.9 31.7 0.6 36.0 1.0 3.2 0.3 4.4 1.8 10.0 7.3 1.7 6.2 2. impulse response functions and estimation of disturbances 101 Disturbance Stationary technology Investment speciﬁc technology Asymmetric technology Nonstationary technology Investment nonstationary technology Consumption preference Labour supply Domestic goods markup Imported consumption goods markup Imported investment goods markup Imported export component markup Export markup Wage markup Risk premium Interest rate Inﬂation target Energy prices Fiscal World economy (SVAR) 1q 0.5 7.9 0.2 1.8 4.3 0.0 2.0 0.7 0.5 1.0 4.7 0.9 2.5 5.9 6.6 2.4 4.1 2.2 2.2 3.1 0.3 0.1 3.5 7.1 5.9 6.5 2.1 6.4 0.6 3.7 0.5 2.8 6.6 1.3 2.8 15.3 0.3 0.0 0.5 2.2 7.7 1.7 1.1 9.7 20.4 11.5 0.6 2.6 10.2 3.3 2.1 0.7 8.7 1.8 2.8 0.8 8.5 10.6 6.0 3.4 0.0 0.2 3.1 1q 0.4 3.6 3.3 1.5 10.3 2.0 0.8 27.2 0.5 9.7 0.2 1.1 6.8 0.9 4q 4.6 5.5 1.1 0.0 5.6 3.7 4.2 3.5 7.7 12.4 4.1 58.5 1.1 0.0 3.4 0.7 1.0 0.9 2.5 0.1 1.5 2.3 1.7 0.6 6.8 2.9 1.3 2.4 1.6 8.9 9.8 17.3 3.2 1.1 0.0 9.4 4.3 39.1 37.2 0.6 5.4 2.1 18.3 0.1 3.6 24.2 5. Regime 2.4 11.6 0.7 10.8 0.4 5.5 0.6 7.1 0.4 3.1 6.9 0.6 1.3 6.1 0.2 5.6 2.1 0.1 1.2 0.6 0.9 0.4 22.1 0.4 Disturbance Stationary technology Investment speciﬁc technology Asymmetric technology Nonstationary technology Investment nonstationary technology Consumption preference Labour supply Domestic goods markup Imported consumption goods markup Imported investment goods markup Imported export component markup Export markup Wage markup Risk premium Interest rate Inﬂation target Energy prices Fiscal World economy (SVAR) 1q 10.1 41.1 12q 4.3 0.6 3.2 3.2 0.0 2.2 4.1 0.7 0.3 0.9 13.0 0.1 1.9 2.1 3.6 6.1 3.8 0.1 2.2 3.8 3.0 1.5 0.9 1.9 1.3 7.5 1.4 2.7 0.4 24.1 2.2 9.5 0.5 16.3 1.4 2.9 1.3 5.2 6.5 0.8 3.2 4.0 3.6 30.3 6.3 4.7 2.7 0.5 1.1 4.7 7.1 11.1 7.
9 3.0 1q 6.7 4.1 9.9 0.9 3.5 0.4 0.6 2.9 20q 0.3 0.9 1.6 32.4 4.1 1.1 4.9 0.3 4.3 1.7 10.0 3.8 4.4 0.5 3.7 20q 2.5 2.3 32.0 16.7 1.2 1.5 0.7 9.1 3.6 3.9 4.0 0.0 13.1 1.9 1.0 12.8 3.8 1.5 20.7 0.5 6.7 0.7 0.1 3.2 1.4 6.0 5.4 15.4 0.4 1.5 5.9 4.3 0.1 0.5 0.3 4.4 2.3 2.1 0.8 2.5 8.7 20q 0.1 0.6 20.5 34.6 20.6 23.2 2.1 13.1 0.3 6.4 0.4 2.1 5.3 0.7 0.1 2.4 1.2 3.8 1.0 0.0 1q 4.2 18.6 0.4 1.9 0.0 20.3 20.2 8.8 5.9 1.3 3.2 1.7 7.4 13.9 3.9 0.7 7.7 102 Employment 4q 8q 12q 6.3 13.3 0.4 4.1 4.1 5.6 2.2 6.1 3.5 0.8 1.3 4.2 11.9 4q 5.2 9.8 1.7 0.8 0.4 3.5 1.0 1.9 0.3 5.4 0.3 0.5 2.2 25.5 63.6 5.1 1.3 2.1 0.2 0.3 4.5 0.3 0.7 6.7 2.2 23.1 4.2 28.8 1.6 23.1 1.6 0.8 0.3 0.3 1.6 2.0 1.6 0.1 1.0 2.6 1.1 Real wages 8q 9.0 12.1 0.2 1.7 3.3 0.8 9.5 4.4 15.6 1.0 0.4 0.7 2.8 3.1 0.3 1.1 0. part 2 Disturbance Stationary technology Investment speciﬁc technology Asymmetric technology Nonstationary technology Investment nonstationary technology Consumption preference Labour supply Domestic goods markup Imported consumption goods markup Imported investment goods markup Imported export component markup Export markup Wage markup Risk premium Interest rate Inﬂation target Energy prices Fiscal World economy (SVAR) 1q 13.8 4.3 12.5 0.8 1.7 12q 6.2 0.9 15.6 3.4 43.7 2.1 0.2 0.1 11.5 0.0 0.1 0.4 0.7 6.7 5.6 13.7 12.7 3.1 0.7 6.4 2.9 9.5 0.3 1.7 3.6 24.5 18.8 1.0 13.4 3.6 3.4 0.1 1q 4.7 2.2 2.8 0.4 20.5 0.1 13.4 1.4 5.8 0.1 1.2 0. Variance decompositions.0 4.6 2.1 11.4 0.0 2.3 5.2 0.4 1.5 1.7 0.1 4.5 0.0 2.4 1.1 10.9 9. impulse response functions and estimation of disturbances Table 7.4 3.6 0.2 0.1 0.3 0.1 7.7 0.0 1.4 11.1 0.6 0.3 4.5 24.0 0.6 Variance decompositions.9 3.7 1.3 0.2 12.9 1.2 5.6 0.9 1.8 8.9 47.4 9.9 2.0 4q 6.2 17.8 12q 1.4 20.7 0.3 8.9 Investment deﬂator 4q 8q 12q 5.5 6.2 0.8 1.2 1.2 7.3 19.5 6.5 2.2 2.7 8.0 4.7 0.3 1.3 2.0 0.2 3.0 5.9 2.3 1.2 1.5 1.5 9.6 3.6 22.4.5 0.5 1.6 7.7 2.0 0.9 0.0 0.2 20q 3.6 6.8 2.2 7.9 4.4 1.4 0.0 6.2 1.8 7.3 0.2 1.5 0.1 0.7 0.0 2.8 0.2 6.1 12.2 8.1 7.4 CPI 8q 0.3 0.9 2.5 2. Regime 1.6 8.7 61.9 2.6 0.3 5.4 1.4 0.6 0.3 3.8 0.5 3.1 3.4 18.3 3.2 1.4 0.6 19.6 5.7 1.6 1.8 1.2 0.4 1.3 0.3 0.9 12.0 0.0 2.8 4.5 0.2 0.2 12.0 4.5 4.6 18.3 0.1 6.5 9.3 1.3 102 N a t i o n a l B a n k o f P o l a n d .6 1.3 0.8 3.9 0.2 14.1 3.0 16.2 0.3 1.1 3.2 1.4 2.0 0.0 1.
etc. The impulse has the value of one standard deviation and lasts one quarter. interesting are the impulse response functions enabling the analysis of the characteristics of monetary transmission mechanism. taking into account autocorrelation of the disturbance. Impulse response functions take into account the existence of all of the simultaneous and lagged interrelations among model variables. whose appreciation leads to a drop in exports. .19 present the responses of selected observable variables to speciﬁed structural disturbances of the SOEPL−2009 model. it must be remembered that differences in responses to some of the shocks result also from differences in the sizes of impulses and autocorrelations of impulses. key reactions to macroeconomic policy variables (direction. The increase of import is. 83 103 .e.2 Impulse response functions Figures 7. Due to the fact that dropping export comprises an import component.e. as well as exchange rates and employment are characterised with percentage deviations from the steady state. response of the economy to monetary impulse (interest rate). which to some extent may substitute the domestic component. inﬂation in the euro area. CPI. From the point of view of the purpose for which the SOEPL−2009 model has been built.1–7. import. import ﬂuctuations 2 In technical sense. effects for consumption and import).1).). consumption. the effects of uncertainty of parameters have not been included. investment deﬂator. it complies with the values typical for the sample (see Tables 6. however. inﬂation in the USA and interest rates). impulse response functions presented further herein have been determined for the point estimates of parameters corresponding to the posterior mode. real wages). the method of presentation of the response function for growing variables enables their treatment as the so called cumulative interim multipliers.15) have standard characteristics. thus. persistence of impulses and (in the case of observable disturbances) the possible correlation between the disturbances. impulse response functions and estimation of disturbances 7. IRF graphs present deviations from the steady state calculated in percentage points2 . i. investments. For observable variables expressed in annualised form as a percentage (GDP deﬂator. It is. The impulse response functions of the monetary disturbance (Figure 7. limited due to the income effect — rational forward looking agents reduce investments and consumption. Thus.Variance decompositions. an illustration of the dynamic characteristics of the model and the consequences of interrelations present in the model. although each time the impulse is different. strength. detect the possible incoherence) and — assuming that the model is an adequate description of a fragment of reality — for the provision of information about the effects of disturbances. Each of the response functions is presented for both regimes included in the SOEPL−2009 model. lag distribution. inﬂation target and risk premium. thus. which is translated into a drop of GDP employment and wages (with multiplier . They inform about the cumulative (in time) effect of the studied impulse on growing observable variables. As in the case of decomposition of forecast error variances. i. foreseen by the authors of DSGE models. So. impulse response function (IRF). Impulse response functions of observable real variables (GDP consumption. The tool is used both for diagnosing the very model (enables to better understand the functioning of the model. export. Increase in domestic shortterm interest rate following the impulse is transferred to the nominal and real exchange rate.2 and 6. 7 103 WORKING PAPER No.
so an interest rate increase is inevitable. investment. although rigid. In the categories of a theoretical model.) abate quickly. which despite the drop of consumption and investments result in an increase in GDP and employment. investment and export goods. Moreover. 3 104 104 N a t i o n a l B a n k o f P o l a n d . so domestic inﬂation reacts without a lag4 . the sequence of events is initiated with a growth of marginal income utility (a consequence of interest rate growth). consumption. Comparing the response functions for the two regimes we shall notice that a relatively small change in the size of monetary impulse is accompanied by quite a considerable change of the scale of variables responses — in the second regime both prices and real categories react more suddenly to the impulse — in the case of CPI. export and import fall. the consequences of the expected interest rate increase prevail. 4 On the other hand. Also responses to exchange rate risk premium impulse seem to be similar to the responses received in the class of models assuming forwardlooking behaviour of agents and motivating the exchange rate changes with uncovered interest rate parity (UIP). Due to the fact that the inﬂation target is a basis for indexation of prices and wages in economy — which is its basic function — inﬂation rises automatically. Beneﬁts (increase in production. Temporary exchange rate depreciation stimulates export and output. Domestic currency appreciation.g. imported components are being replaced with domestic components. with the exception of consumption and investments (with weaker responses in the new regime) and employment. but after the increase in interest rate and exchange rate appreciation. The real exchange rate strengthens upon relatively large but shortlasting depreciation (only slightly affecting export) — thus. Absorption takes place quite quickly — e. In such a situation a growth in risk premium results in immediate depreciation of (nominal and real) exchange rate. when such phenomenon intensiﬁes. the cost of working capital bind directly the marginal cost of domestic manufactured products with interest rate. higher inﬂation target considerably alleviates interest rate policy (the central bank accepts higher inﬂation levels). which leads to immediate reduction of consumption. impulse response functions and estimation of disturbances occur in the ﬁrst quarters. In this case the interrelation diminishes the antiinﬂation effects of monetary disturbance. A change in regime that — according to the estimation results — took place in 1999.14) — describes the consequences of unexpected emergence among the manufacturers and consumers of a belief (expectations?) as to the (temporary) higher inﬂation (inﬂation target). This brings about a fast reduction of import and increase in export. Another impulse response function that is interesting for the central bank — the IRF of the 7 inﬂation target disturbance (see Figure 7. the maximum response nearly doubled. practically did not change the responses of most of the observable variables to premium shock (including inﬂation).Variance decompositions. whose structural form we have presented in the Appendix. The impulse response functions of inﬂation target may. the fast response of interest rate to premium disturbance eliminates the scale of depreciation. Despite the fact that. output. be treated as an illustration of the costs of inﬂation suppression. consumption. include a currently optimised component. decrease of aggregated demand and wages result in a drop of inﬂation3 . thus. etc. multiplier effects pushing inﬂation up above the current target force an increase in interest rate. while that increase exceeds the scale of the shock. With increase in demand for domestic products also the prices of domestically manufactured goods rise. leading to a growth of prices of imported components of consumption. Domestic prices. Characteristic is the fact that this has not signiﬁcantly affected the time of shock absorption — it is similar in both regimes. in the perception of agents.
consumption. The reactions were strong enough to result in a temporary increase in employment and real wages (which has not been observed in the ﬁrst regime). Lower inﬂation implies a reduction of interest rate and depreciation of real exchange rate resulting from anticipated expansionary monetary policy. to an increase in export and decrease in import. The new regime that appeared post 1999 caused much stronger positive reactions of output. In subsequent quarters the responses of variables are similar in both regimes. The whole leads. The effect of such a situation is a drop in the dynamics of prices and increase in output. The fact that the wages are rigid and remuneration for capital services is ﬂexible proves to be an important element. of course. output) are perceptible much longer. impulse response functions and estimation of disturbances while costs (higher inﬂation. In the group of stationary technological disturbances (TFP investmentspeciﬁc and asymmetric . 7 105 WORKING PAPER No. thus. investments and real wages. contributing to increase in the share of domestic products in the ﬁnal product. 83 105 .1) represents an increase in total factor productivity leading to a decrease in the marginal costs of domestic production and demand for production factors (per unit of product). the ﬁrst of them (see Figure 7. disturbances). consumption and investments in the ﬁrst quarters.Variance decompositions. Also the proportion of productivity factors changes — employment decreases and investments increase. drop in employment.
5 0.06 0.02 Real exchange rate 0 0.4 0 12 Imports 0.5 0.7 0.2 0 12 24 36 0.6 0. Stationary technology shock GDP deflator 0 0.05 0 Real wages 0 0.3 0.2 0.6 0.1 0.25 0.35 0.3 0.6 0.Variance decompositions.7 Investment deflator 0 0.05 0.35 0 GDP deflator 0.2.3 0.05 0 24 36 0 Real exchange rate 0.6 0.15 0.3 0.15 0.15 0.35 0.1 0 12 24 36 0 Investment 12 24 36 Exports 0.3 0.1 0 0 12 24 36 0 0 12 0.1 0 0.5 0.2 0.2 0.1 0 0.05 0.2 0.25 0.5 0.2 0.4 0.4 0.2 0.05 0.2 0.05 0 0.05 0 12 24 36 0 0 Consumption 0.3 CPI 12 24 36 0 12 24 36 106 106 N a t i o n a l B a n k o f P o l a n d .3 0.15 0.1 0.25 0.2 0.2 0.4 0.15 0.25 0.1 0.1 0.15 0.1 0.02 0 0.1 0.7 0.4 0 Employment 0.4 0.2 0.05 0 0. Stationary investment shock 7 0 0.8 0.2 0.2 0.6 0.05 0 0 Real wages 0 0.7 0.4 0. impulse response functions and estimation of disturbances Figure 7.1 0.2 0.2 0.6 0.12 0.15 0.1 0.15 0.15 0.4 0.4 GDP 0.2 0 12 24 36 0.4 0.08 0.2 0.3 0.3 0.1 0.2 0.2 0.8 1 0 12 24 36 0 12 CPI Ver/Regime1 Ver/Regime2 12 24 36 0 12 24 36 24 36 Figure 7.4 0.6 0.2 0.1 0.4 Interest rate 24 36 0 12 24 36 0 12 24 36 0 0.3 0.15 0.3 0 12 24 36 0 Consumption 5 4 3 2 1 0 12 24 36 0 Investment Ver/Regime1 Ver/Regime2 12 24 36 12 24 36 Exports 0.3 0.1 0.2 1 Imports 0.1 0.25 0.2 0.2 0.3 0.1 0 GDP 0.3 0.1 0.1 0.05 0 12 24 36 0.25 0.1 0.1 0.25 0.5 0.1.1 0 0.25 0.2 12 24 36 0 Interest rate 12 24 36 Employment 0.15 0.25 0.05 0.8 1 1.04 0.5 0.3 0.3 12 24 36 0 Investment deflator 0 0.2 0.35 0.35 0.4 0.05 0.25 0.2 0.1 0.05 0.2 0.1 0 0.15 0.
2 0.6 0.1 0.2 0.004 0.4 0.5 0 0.02 0.6 0.04 0.5 1 USA inflation 1 0.5 0 0.05 0.01 0 0 12 24 36 0.15 0.2 0.025 0 12 24 36 Real wages 0 0.4 0.06 0.02 0.3 0.25 0.6 0.04 0.5 0 0. impulse response functions and estimation of disturbances Figure 7.02 0 0 12 24 36 0 0 12 24 36 0 0 12 24 36 0.8 0.01 0.2 0 0 Investment 7 36 12 24 Exports 0.2 0 Imports 0 0.5 1 USA inflation 1 0.3.006 0.02 12 24 36 0 0 12 CPI 24 36 Euro zone GDP 0.2 Consumption 0.06 0.008 0.15 0.01 0.04 0.5 1 Dollar interest rate 1 0.03 0.004 0.04 0.012 0 12 24 36 GDP deflator 0.03 0.002 24 36 0 0 Consumption 0.6 0.05 Investment deflator CPI 0 12 24 36 1 0.01 0.5 0 0.1 0.5 0 0.04 0.06 Real exchange rate 0.04 0.008 0.04 0.02 0.1 0.01 0 0 12 GDP 0.006 0.02 0 12 24 36 0 Real exchange rate 0.02 0 Investment 12 24 36 0 12 24 36 Exports 0.5 1 Euro zone inflation 1 0.Variance decompositions.6 0.05 0 0 x 10 8 6 4 2 0 2 0 12 24 36 2 0 6 4 12 24 36 0.5 0 0.02 0.05 0.1 0.08 0.08 0.3 0.6 0.4 0.2 0 Real wages 0.01 0.02 0.01 12 24 36 0 0 Interest rate 0 12 24 36 12 24 36 Employment 0.02 0 12 24 36 0 0 Investment deflator 0.02 0.02 0.04 0.005 0.06 0. Asymmetric technology shock 0 0.03 0.2 0.002 0.01 0.1 0 USA GDP 0 12 24 36 0 12 24 36 0 12 24 36 0 12 24 36 1 0.03 0.5 0 0.01 Interest rate 3 Employment 3 24 36 0 12 24 36 0 0.5 1 Dollar/euro exchange rate Ver/Regime1 Ver/Regime2 0 12 24 36 0 12 24 36 0 12 24 36 WORKING PAPER No.05 0 0 x 10 12 Imports 0 0.4 0.4 0.08 0.5 1 Euro zone inflation Euro interest rate 0.5 0 0.4 0.06 0.5 0 0.4 0.2 GDP 0.04 0.5 1 Dollar interest rate 1 0. Nonstationary technology shock GDP deflator 0.5 1 Dollar/euro exchange rate Ver/Regime1 Ver/Regime2 0 12 24 36 0 12 24 36 0 12 24 36 Figure 7.4.2 0 USA GDP 0 12 24 36 0 12 24 36 0 12 24 36 0 12 24 36 1 0.5 0 0.2 0 1 0.015 0.1 0 1 0.6 0.6 0.5 1 0 12 24 36 0 12 24 36 Euro zone GDP 0.2 0 0. 83 107 107 .5 1 Euro interest rate 0.01 0 0.4 0.
1 0 0 0.05 0 12 24 36 0 Real exchange rate 0.12 0.14 0.04 0.18 0.2 0.4 0.2 0.4 0.1 0.02 0.3 0.2 0.06 0.5.25 0.7 0. impulse response functions and estimation of disturbances Figure 7.3 1 0.1 0 12 24 36 0 12 24 36 24 36 108 N a t i o n a l B a n k o f P o l a n d 108 .3 0.2 0 12 24 36 0 Investment 12 24 36 Exports 0.3 GDP 0.5 0 0.25 24 36 0 Real exchange rate 0.05 0 0.3 0.06 0.5 0 0.3 0.1 0 Real wages 0 0.2 0.1 0.02 12 24 36 0 0 Interest rate 0 12 24 36 12 24 36 Employment 0.4 0.2 0.5 0.5 1 1 0.2 0 12 24 36 0 12 Imports 0.1 0 0 0.1 0 0.06 0.5 1 USA inflation 1 0.02 0 12 24 36 0 0 12 CPI 0.6 0.1 0 0 12 GDP 2 Consumption 0 Investment 1.16 0.02 0 0 12 24 36 0.1 0 USA GDP 0.5 0 0.6 0.1 0 0.1 0.5 1 0 Investment deflator 0.1 0 Imports 0 0.2 0.03 0.16 0.05 0 0 12 24 36 0.1 0 Ver/Regime2 12 24 36 0 12 24 36 Employment 0.2 0.5 Interest rate Ver/Regime1 0.04 0 0.2 0.1 0 0.04 0.4 0.1 0.5 Euro zone inflation Euro interest rate 0.2 0 0.4 0.6.3 0.05 0. Investment nonstationary technology shock GDP deflator 0.02 0.2 0.5 0 0.05 0.12 0.02 0 0 12 24 36 0.05 0 12 24 36 0 0 Consumption 1 0.01 0.1 0.5 0 12 24 36 Exports 0.04 0.15 0.6 0 12 24 36 1 0.3 0.14 0.5 1 0 12 24 36 0 12 24 36 0 12 24 36 0 12 24 36 1 0.1 0.5 1 Dollar interest rate 1 0.5 0.05 0 CPI 12 24 36 0 12 24 36 Euro zone GDP 0.08 0.1 0.4 0.04 0.08 0.15 0.6 0.08 0.8 0.2 0.15 0.12 0.1 0.3 0.2 0.4 0.06 0.2 Investment deflator 0.5 0 24 36 0 12 24 36 1.06 0.5 0.Variance decompositions.08 0.8 0.5 1 Dollar/euro exchange rate Ver/Regime1 Ver/Regime2 0 12 24 36 0 12 24 36 0 12 24 36 Figure 7.15 0.2 0.14 0.2 0. Consumption preference shock 7 GDP deflator 0.1 0.3 0.2 0.08 0.18 Real wages 0.16 0.04 0.15 0.5 0.
4 0.2 0.8 0.5 0.2 0.25 0 12 24 36 1.15 0.4 0 0.2 0.2 0.2 0.2 0 12 Imports 0.7 0.5 0.05 0.2 0.05 0.7 0.05 0.5 1 0. impulse response functions and estimation of disturbances Figure 7.1 0.4 0 Consumption 0 0.9 0.4 0.5 0 12 24 36 0.35 0.3 0.5 1 0.7 0.6 0.1 0.1 0.3 0.05 0 0.05 0.2 0.2 0.Variance decompositions.8 0.15 Real exchange rate 0.35 0 Employment 0 0.2 0.1 0.1 0 0 Real wages 0.7 0.15 0.4 0.15 0.4 0.8 0.2 Imports 0.1 0.5 0 12 24 36 0 12 CPI Ver/Regime1 Ver/Regime2 12 24 36 24 36 109 WORKING PAPER No.3 0.2 0.35 0 12 24 36 0.5 0. Labour supply shock GDP deflator 1 0.4 0.6 0 GDP 0 0.15 0.2 1 0.2 0 0.15 0 12 24 36 0 12 24 36 Employment 0 0. 83 109 .8 0.3 0.25 0.2 0.8 0.2 0.5 0.3 0.9 12 24 36 0 Investment 12 24 36 Exports 0.8 0.05 Real exchange rate 0.5 0.35 0.3 0.2 0.15 0.1 0.3 0.4 0 Consumption 0 0.4 0.5 0 0 12 24 36 0.2 0 CPI Ver/Regime1 Ver/Regime2 12 24 36 0 12 24 36 Figure 7.1 0.5 2 1.25 0.4 0.2 0.15 0.35 0.1 0.4 1.2 0.15 0.9 Real wages 1.1 0.1 0.5 2 1.1 0.1 12 24 36 0 0 Investment deflator 1 0.45 0.3 0.3 0.4 1.2 0.05 Interest rate 24 36 0 12 24 36 0 0 12 24 36 0.6 0.6 0.05 0 0.8 0.4 0.2 0 12 24 36 0. Domestic goods markup shock GDP deflator 3.8 0.1 0 Interest rate 0.3 0.1 0.3 0.6 0.6 0.25 0.4 0.5 0.4 0.3 0.6 0.05 0.3 0.4 0.2 1 0.25 0.4 0.05 0 0.5 3 2.8.6 12 24 36 0 Investment 7 12 24 36 Exports 0.6 0.05 0.1 0.05 0 0.8 1 0 12 24 36 0.6 0.1 0.7.05 0.25 0.5 0.1 0.4 0.1 0 0.6 0 12 24 36 GDP 0 0.3 0.2 0 0 12 24 36 0 0.1 0.2 0 0 12 24 36 0 Investment deflator 2.6 1.
3 0.1 0 12 24 36 0.15 0 12 24 36 0 12 24 36 0 12 24 36 GDP 0.08 0.05 0.04 0.5 0.4 0.02 0 0.2 1 0.8 1 1.6 0.02 0.4 Investment 7 0.6 0.2 Imports 0 Real exchange rate 0.5 0.06 0.02 0.06 0 Real wages 0.02 0 0.02 0 0.05 0.02 0 0.06 0.08 0.5 1.6 0 12 24 36 1.04 0.04 0.02 0 3 2 1 0 1 0.02 0.03 0. Imported consumption goods markup shock GDP deflator 0.8 1 1 1.1 0.06 0.2 0.05 0.7 0.1 0.04 0.08 0 12 24 36 0.05 0.5 12 24 36 0 Investment deflator 1.2 0.06 0.02 0.1 0 0.2 0.6 0.4 0.1 0.2 0.4 0.05 0 12 24 36 0 12 24 36 0.3 0.04 0.6 0.6 0.04 0.02 0 0.8 0.15 0 0.3 0.3 0.2 0.02 0.4 0.8 0 12 Exports 0 0.Variance decompositions.06 0.04 0.03 0.01 0.05 0 Real wages 5 4 Investment deflator 0.2 0.05 0 0.2 24 36 0 12 Imports 0 0.2 0.5 0.08 0.06 0.1 Consumption 0 0.9.02 0.6 Interest rate 0.04 0.15 0.35 Real exchange rate 0.1 0.4 0.12 0 12 24 36 0 Interest rate 24 36 12 24 36 Employment 0.4 0.03 0 0.06 0.01 0 0.1 0 12 24 36 0 12 CPI Ver/Regime1 Ver/Regime2 12 24 36 24 36 110 110 N a t i o n a l B a n k o f P o l a n d .6 0.2 1.1 0 0.1 0.03 0.02 0.04 0.02 0.8 1 0.15 0.4 0.1 2 0 12 24 36 1.3 1 0.08 Consumption 0. Imported investment goods markup shock GDP deflator GDP 0.25 0.05 0.5 0.1 0.3 0.01 0 0.2 0.2 0.10.1 0.4 0.5 0.15 0 12 24 36 0 12 24 36 0 0.02 0.1 0.04 0. impulse response functions and estimation of disturbances Figure 7.05 0.06 0.01 0.5 0 12 24 36 0 0 12 24 36 Employment 0.2 0 0.08 0.02 0.2 12 24 36 0 12 CPI Ver/Regime1 Ver/Regime2 24 36 Figure 7.1 0 0 Investment 12 24 36 0 Exports 0 0.04 0.
2 0 12 24 36 0.8 0.3 CPI Ver/Regime1 Ver/Regime2 12 24 36 0 12 24 36 Figure 7.5 0.1 0.35 0.1 0.15 0.03 0.03 0.01 0 0.07 0.4 0.3 0.1 0.11.6 0.12.3 0.3 0.15 0.05 0.4 0.2 0.15 0.05 0.06 0 Real wages 0.2 0 0.02 0.4 0 12 24 36 0 0 12 24 36 0.06 0.15 0.15 0.1 0.8 0.14 0.5 0. Export markup shock GDP deflator 0.15 0 12 24 36 0 0.4 GDP deflator 0.2 0.45 0 12 24 36 0.5 0.1 0.05 0.3 0.6 0.6 0.3 0.25 0.6 12 24 36 0 Investment deflator 0 0.4 0.2 0.2 1 0.7 0.1 0.2 0.1 0.25 0.4 0.18 0 12 24 36 0.1 0.03 0.7 0.8 0. impulse response functions and estimation of disturbances Figure 7.8 0.5 3 3.6 0.1 0.05 0.8 0.6 Interest rate 24 36 0 12 24 36 0 12 24 36 0 12 24 36 Employment 0 0.2 0.5 0.1 0 Investment 0 12 24 36 0 12 24 36 0 12 24 36 0 1 2 3 4 5 6 7 0 12 Exports 0 1 2 3 4 5 6 7 8 24 36 0 12 Imports 0 0. Imported export component markup shock 0 0.5 2 2.08 0.15 0.4 0.2 0.5 1 1.01 0.4 0.2 0.2 0.3 0 12 24 36 0 Interest rate 24 36 12 24 36 Employment 0.1 0 0.3 0.25 0.08 0.2 0.05 0.9 Real exchange rate 0.4 0.5 4 4.05 0.02 0.4 0.2 0.Variance decompositions.4 0.1 0.2 0.25 0.16 0.04 0.1 0.5 Imports 1.6 0.5 0.8 1 1.3 0.05 0 0.3 0.02 0.1 0.25 0 12 24 36 0.2 1.6 GDP 0.2 0 Real exchange rate 0 0.05 0.01 0 0 Real wages 0 0.2 0.1 0.1 0.4 0.02 0.04 0.15 0.04 0.06 0.05 0.2 0.12 0.9 0. 83 111 .1 0.05 0 0.05 0 Consumption 0.7 0.1 Investment deflator 0.2 GDP Consumption 0 Investment 7 0 1 2 3 4 5 6 7 8 0 12 Exports 0 0.05 0 0 12 24 36 0 12 CPI Ver/Regime1 Ver/Regime2 12 24 36 24 36 111 WORKING PAPER No.1 0 0.2 0.09 0.
2 12 24 36 0 0 Investment deflator 1.05 GDP Consumption 0.02 0.1 0 Interest rate 12 24 36 0 12 24 36 Employment 0.06 0.6 0.5 0.1 0 0 0.4 1.02 0 0 Real wages 1 0.2 0.2 1.05 24 36 0 Consumption 0.1 0.2 0.1 0.15 0.4 0.05 0 0.4 0.02 0.1 0.4 0.3 12 24 36 0 Investment 7 1.8 0.12 0 12 24 36 0.3 0.2 0.4 0.05 0 0.1 0.08 0.5 1 0.2 1 0.8 0.2 0 0.5 0.2 0 12 24 36 Exports 0.04 0.1 12 24 36 0 Interest rate 12 24 36 Employment 0.15 0.1 0.15 0.1 12 24 36 0 12 CPI Ver/Regime1 Ver/Regime2 24 36 Figure 7.1 0.5 0 0.1 0 0.05 0 0.25 0.4 0.1 0 Investment 12 24 36 Exports 0.3 0 12 24 36 0.1 0 12 24 36 0.04 Real wages 1.2 0.05 0 0.2 0.5 0.05 0 12 24 36 0.1 0.35 0.1 0.3 0.1 0.6 0.2 0.05 0.6 0.4 1.08 0.2 0 0.06 0.5 0 1 0.15 0.2 0.3 0.12 0.1 0 0.4 0.14.8 0.Variance decompositions.2 12 24 36 0 0 12 CPI Ver/Regime1 Ver/Regime2 24 36 112 112 N a t i o n a l B a n k o f P o l a n d .5 0.4 0.05 0.05 24 36 0 Real exchange rate 0.02 0 0.6 0.2 12 24 36 0 Investment deflator 0.2 1 0.05 0 0.25 0. Risk premium shock GDP deflator 0.2 0.1 0. impulse response functions and estimation of disturbances Figure 7.3 0.2 0.2 0 12 24 36 0 0.4 GDP deflator 0.8 0.04 0.6 0.1 0 0.08 0.35 0.2 0.06 0.4 0.8 0.15 0.1 0.05 0 0.4 0.14 0.15 0 12 24 36 0 12 GDP 0. Inﬂation target shock 1.5 0 24 36 0 Real exchange rate 0.15 0.4 2 0.05 0.13.15 0 12 Imports 0.2 0.4 0.3 0.2 1 0.8 0.04 0.1 0 0.6 0.2 1 0.2 0.1 0.05 0 12 24 36 1.2 0.6 0 12 24 36 0 12 Imports 2 1.2 0.3 0.2 0.3 0.7 0.1 0 0.1 0.
012 0.18 0.08 0.04 0.15 0.15 0.05 0.1 0. Energy prices shock 0.02 0 Real exchange rate 0.2 GDP deflator 0 0.04 0.1 0.2 0.3 0.02 0.05 Consumption 0 0.25 0.016 0.16 0.2 0 12 24 36 0. 83 113 .002 0 0 12 24 36 0 0.Variance decompositions.01 0.5 0.04 0.06 0.18 0 Consumption 0 0.4 24 36 0 Real exchange rate 0.06 0.3 0. impulse response functions and estimation of disturbances Figure 7.16.06 0.25 0 12 24 36 0 12 24 36 0 Exports 0.02 0 12 24 36 0 12 24 36 0 0 12 24 36 113 WORKING PAPER No.05 0.15 0.05 0.12 0.15 0.06 0.05 0.4 0 12 24 36 GDP 0 0.4 0.04 0.04 0.1 0.06 0.018 0.004 0.045 0 Real wages 0.35 0.1 0.01 0.08 Interest rate 24 36 0 12 24 36 0 12 24 36 Employment 0.02 0.005 0.08 0 12 Imports 0.2 0.2 0.08 0.04 0.04 0.2 0.08 0.1 0.16 0.15.15 0.1 0.08 0.25 CPI Ver/Regime1 Ver/Regime2 12 24 36 0 12 24 36 0 12 24 36 Figure 7.02 0.15 Investment deflator 0.04 0.01 0.04 0.2 0.1 0.08 0.1 0.3 0.06 0.18 0 12 24 36 0 0.08 0.12 0 Real wages 0 0.02 0.1 0.02 0.1 0.1 0 Interest rate 12 24 36 0 12 24 36 Employment 0 0.35 0.14 0.14 CPI Ver/Regime1 Ver/Regime2 0.3 0.25 0.008 0.02 0 0 12 24 36 0.08 0.25 0.1 0.16 0.1 0.05 0.06 0.25 0.035 0.08 0.1 0.5 0 12 24 36 0 0.02 0 0.02 0.04 0.12 0.06 0. Monetary policy (interest rate) shock GDP deflator 0 0 0.025 0.03 0.1 0.04 0.12 0 12 24 36 0 Investment 0.02 0.14 0.12 0.02 0.2 0.06 0.1 0.04 0.12 0 12 Imports 0 0.05 0.02 0.12 0.1 0.35 12 24 36 0 Investment 12 24 36 Exports 0 0.2 0.15 0.15 0.1 7 12 24 36 0.04 0.15 GDP 0 0.05 0.4 0.02 0 0.014 0.03 0.05 0.1 0.06 0.015 0.2 0.04 0.06 0.05 0.3 Investment deflator 0 0.3 0.006 0.
5 0 0.4 0 12 24 36 0 12 24 36 1 0.5 3 2.5 0 6 Exports 6 5 4 3 2 1 0 1 12 24 36 x 10 6 Imports 6 4 2 0 2 4 6 8 10 x 10 7 Real exchange rate 5 4 3 2 1 0 x 10 6 Interest rate 0 12 24 36 0 12 24 36 0 12 24 36 x 10 7 6 5 4 3 2 1 0 7 Employment 12 10 8 6 4 2 0 x 10 7 Real wages 4 3.4 x 10 3 Consumption 0 x 10 3 Investment 7 2.5 2 2.2 1.5 1 0.5 0.5 3 2.5 0 0.5 0 0 12 24 36 6 GDP deflator 16 14 12 10 8 6 4 2 0 x 10 7 GDP 4 3. Employer social contribution shock x 10 3 3 GDP deflator 0 0.5 1.5 4 3.5 2 1.8 1 1.6 0.5 2 1.5 2 2.5 0.2 1 0.Variance decompositions.5 4 3.5 x 10 6 Investment deflator x 10 4.5 3 2. Capital tax shock x 10 4.5 3 2.5 1 0.5 2 0 12 24 36 x 10 0 1 2 3 4 5 6 7 8 9 0 4 Exports 8 6 4 2 0 2 4 12 24 36 6 x 10 4 Imports 1 0.5 0 x 10 6 Consumption 8 6 4 2 0 2 x 10 6 Investment 0 12 24 36 0 12 24 36 0 12 24 36 x 10 2 1.5 1 0.4 x 10 3 GDP 0 0.5 0 0.5 1 0.2 0.5 2 1.5 1 1 Ver/Regime1 Ver/Regime2 1 1.5 0.4 0.6 0. impulse response functions and estimation of disturbances Figure 7.8 1 0 12 24 36 1.5 3 3.17.5 6 CPI Ver/Regime1 Ver/Regime2 0 12 24 36 0 12 24 36 0 0 12 24 36 0 0 12 24 36 Figure 7.2 0.5 1 0.18.6 0.5 x 10 4 Real exchange rate 1.5 2 1.5 1 1.5 0 2 12 24 36 0 12 24 36 0 0 12 24 36 0 12 24 36 114 114 N a t i o n a l B a n k o f P o l a n d .2 0 x 10 3 Interest rate 0 12 24 36 0 12 24 36 0 12 24 36 x 10 0 2 4 6 8 10 12 0 4 Employment 0 x 10 3 Real wages 2 x 10 3 Investment deflator 2.5 1 1.5 1.8 0.5 1 0.5 2 x 10 3 CPI 0.5 2 1.
In the face of exchange rate depreciation and fast GDP growth.5 0 Investment deflator 3 2. Such function is fulﬁlled by asymmetric technology disturbance.6 0.4 0.4 1.4 Consumption 0 0. interest rates and cross rate. The domestic interest rate reacts stronger to the appearing supply gap than the minimally decreasing domestic inﬂation.6 0.5 0 2 0 12 24 36 0 12 24 36 1. The increase during the ﬁrst period leads even to a fall in consumption in the pool of household expenditures.6 0.5 2 1. Appearance of such a shock increases the dynamics of output in the world.8 0. however.4 0. The total effect is the increase in export but even bigger increase in import (due to high level of import in investments). 83 115 .3.2) results in a large increase in investment expenditures.8 1. impulse response functions and estimation of disturbances Figure 7.5 3 2.2 1 0.5 1 1.5 0 0 1. there may. increasing demand for goods exported from the domestic economy.4 0.6 0. which gives an impulse for real exchange rate depreciation.5 2 1. a drop in the price of capital services reduces the marginal cost of domestically produced goods. Faster increase in export than in import leads to (nominal and real) exchange rate appreciation. asymmetric technology shock weakly affects the economy.19. while the increase in dynamics in the world’s output does not impact the world’s prices.8 0 12 Imports 0. the appearance of such shock (see Figure 7.8 0. as in 115 7 WORKING PAPER No.Variance decompositions.3 0.5 2 2.5 0 12 24 36 0 Investment 12 24 36 Exports 0 0. the interest rate is rising.2 0. Thus.6 0 12 24 36 0. Generally.4 0. which — ceteris paribus — lowers domestic inﬂation and increases export competitiveness.5 0 CPI Ver/Regime1 Ver/Regime2 12 24 36 0 12 24 36 0 12 24 36 Investmentspeciﬁc technology shock results in an increase in efﬁciency of the transformation of investment expenditures into ﬁxed assets.2 0.5 1 0.2 1.4 0.8 1 1.2 0. the rate of technological progress is identical to the rate in the rest of the world.2 0 12 24 36 1 0.8 1 1.4 1.5 0 GDP 0 0. In a small open economy. from a unit of expenditures we receive more capital which brings rent to households. At the same time.5 2 Real wages 2.8 1 1. Wage markup shock GDP deflator 3. The situation is illustrated by the impulse response functions in Figure 7. and thus appearance of negative net export.5 1 0.2 Interest rate 12 24 36 0 0 12 24 36 Employment 0 0.2 0.6 1.5 1 0.5 1 0.5 2 1.4 0. which translates into a drop in prices of imported products.5 1 2.5 24 36 0 Real exchange rate 1.1 0 0.4 0. be present temporary disturbances differentiating growth dynamics. Therefore.2 1.2 0 0.6 0.2 0.6 0.2 0.1 0.
the marginal costs of domestic production rise and thus results in an increase in inﬂation. A relative decrease in the importance of leisure leads to a rise in employment. as faster growth of volume reduces price dynamics.4 and in investments — see Figure 7.5).6 and 7. An immediate effect of the shock is an increase in consumption and a drop in investments. responses of the deﬂator are the same as the responses of other prices. which results in drop of export. import. investments. GDP and employment. employment. the rate affects the real rent for capital services present in the marginal cost r t . In the case of shock in investments — based on disturbance deﬁnition — we observe in its wake a drop in investment prices. The graphs of impulse response functions for both aforementioned disturbances are similar and result from the same mechanism. but due to increase in the price of labour services.Variance decompositions. The effect of such impulse is the reduction of consumption. Changes in exchange rates increase the depth of response of export and import. A combined effect results in an increase in output. the shock may be treated as a demand shock. however. export. The only difference refers to investment deﬂator. In the case of disturbance in technology. and in real wages decrease. Thus. Negative net export along with the increase 5 7 More speciﬁcally. Due to the fact that consumption goods include imported components. The drop of labour supply is compensated with increased capital utilisation rate. Momentary increase in interest in consumption means simultaneous decrease of the relative importance of the other sources of utility speciﬁed in the function maximised by households.7. The shocks have no permanent impact on inﬂation. Increase in domestic prices reduces export competitiveness and stirs import. and thus a decrease in the relative importance of consumption in the households’ utility function. Nonstationary nature of disturbances means a permanent inﬂuence of shocks on the level of growing observable variables (GDP consumption. and domestic components become more expensive. The consequences of consumption preference shock and labour supply (leisure preference) shock are illustrated in Figures 7. 116 116 N a t i o n a l B a n k o f P o l a n d . import grows (despite a drop in investments). With the rise in inﬂation of the domestically manufactured products.t 5 . Peculiar reactions are observed for nonstationary disturbances (in technology — see Figure 7. Labour supply shock means an increase in the importance of leisure (which automatically lowers labour supply). the terms “demand/supply shocks” do not ﬁt the logic of the DSGE models. real observable variables move to a new path. the increase of interest rate becomes inevitable and leads to nominal exchange rate appreciation — the scale of appreciation is high enough to translate into the appreciation of the real exchange rate. Viewing all of the responses to shock in preferences as boiling down to increase in production and prices. impulse response functions and estimation of disturbances the previous versions of the SOEPL model. investments. At the same time — as a result of the increase in prices of domestic products — the competitiveness of export diminishes. employment and domestic prices. real wages. which rises despite a drop in domestic demand (the price effect proves to be stronger than the income effect). . exchange rate dynamics and interest rate. while price ﬂuctuations are a consequence of relation between the marginal cost of domestic production and the rate of growth of technical progress µz. external GDP).
impulse response functions and estimation of disturbances in domestic prices results in real exchange rate depreciation. the disturbances are undistinguishable. A large group of disturbances are markup shocks (for domestically manufactured products.12). the reactions of observable variables are analogical (see Figure 7. In the case of markup on import products. in the basket of households there will be more relatively cheaper investment goods but despite the growth of investment import. which stirs the growth of inﬂation. despite import reduction. there appear more extensive effects related to substitution of imported components with domestic production. regulations changing (for example) the position of trade unions in wage negotiations. The IRF for wage markup shock in both of the regimes explains why the disturbance was so important in the ﬁrst regime. As regards disturbance in markup on investment import. while the appearance of positive net export results in exchange rate appreciation. and the resulting changes in the proportion of the components of aggregated demand. which immediately impacts export. In a DSGE model we identify them only thanks to the knowledge provided with prior distribution in Bayesian estimation. 7 117 WORKING PAPER No. with accuracy to the response scale. And consequently. The increase in the share of domestic production in consumption means also a growth of the prices of components manufactured domestically. the sense assigned to the disturbances is completely different. With a lower interest rate and higher employment (at least at the beginning). Already a superﬁcial analysis of Phillips curve speciﬁcation for domestic prices shows that the markup has an identical impact on inﬂation to marginal cost (see stationary disturbance in technology. Growth of demand for domestic products results in the growth of employment and wages.Variance decompositions. Consequently. the import in total decreases.10) — in this case. the response functions presented in Figure 7. Momentarily there appears a surplus of domestic products and drop in the 6 In the formal statistical sense. When the effects of appreciation are translated into domestic prices. The shock of export markup (Figure 7. The total view boils down to increase in inﬂation and a drop in the level of economic activity. as well as reexported products and export products.7. general consumption is reduced. 83 117 . investment goods. which immediately leads to a drop in GDP and employment.8–7. Taking into account the speciﬁcation of the structural form of the model. however. Thus. the wage markup shock (markup in the labour market) fulﬁls an analogical function as the labour supply shock6 .12) reduces export.11) results in replacement of import with domestic products.9) increases the prices of imported components of consumption and automatically the CPI is growing (causing a response of interest rate). where an analogical situation as for domestic products markup is present. but in the face of inevitable interest rate increase the exchange rate ultimately appreciates. In the case of wage markup disturbance. The shock of markup of imported export component (Figure 7. we speak about phenomena of institutional nature. including consumption import. with a reverse sign). a shock in the markup on imported consumer goods (7. and the suddenness of this reaction results in instantaneous growth of the interest rate (the rate depends on the level of gap and its dynamics). consumption is growing at the cost of investments. so there appears a positive net export and exchange rate appreciation. see Figures 7. however. a similarly sudden interest rate decrease occurs. consumption and investments grow. therefore.19 replicate Figure 7. imported consumer goods. Nevertheless.
the analysis of disturbances (or broader — historical decompositions) refers to a speciﬁc model (see also Canova and Sala. while the decisions made by the agents (at macro scale) cause the absorption of the disturbances.20 reﬂect a relatively high inertia8 . it has been assumed that structural disturbances are generated by a stochastic process of AR(1) class and only in a few cases the autocorrelation coefﬁcient takes zero values (e. 8 It is worth remembering that some of the autocorrelation coefﬁcients of disturbances have been calibrated. the economy would develop in compliance with the characteristics of the steady state. Was it not for the disturbances. In the case of drop of inﬂation for the domestically manufactured products and a drop in GDP the response of the interest rate is . the series presented in the subsequent Figures show an image of a fragment of the history of Polish economy seen from the perspective of the SOEPL−2009 model. on the basis of the current estimations of disturbances and the knowledge about the way observable variables responds. The largest inertia is reﬂected by the inﬂation target disturbance and asymmetric disturbance in technology. The structural disturbances are.20–7. Different from zero autocorrelation coefﬁcients make nearly all of the disturbances presented in Figure 7. in DSGE models the disturbances are the most important factor “driving” the ﬂuctuations of variables. the question about the sources of observable events shall be answered by searching for disturbances (innovations) that have occurred. see Hamilton (1994. however. Therefore. This is caused by a large scale of GDP drop (calculated in percentage points). the very (structural) disturbances are a type of ultimate cause of the phenomena and events that we may observe in the economy. the procedure of determination/estimation of nonobservable structural disturbances based on the model having the form of state space model is called smoothing. 7. stronger positive/negative deviations will be eliminated in a shorter or longer period of time.g. a set of artefacts that we construct or quantify with the use of a speciﬁc DSGE model. However. Such is the logic of the forecasting technique with the use of a DSGE model. According to the earlier suggestions. Therefore. opposite to the CPI movement — the interest rate decreases. This is an image speciﬁc for the version of the model7 . Thus. Chapter 13). with a relatively low increase in CPI inﬂation (a tenth parts of a percentage point). 118 118 N a t i o n a l B a n k o f P o l a n d . 7 7 From the technical side. Two different DSGE models of the same object in the same period. In the SOEPL−2009 model. impulse response functions and estimation of disturbances prices of products. usually differently recreate the disturbances that formally fulﬁl the same function in the models — different series are received. In the world of DSGE models. for interest rate disturbance) — then the graphs of disturbances and innovations are identical. Due to the fact that the “natural” value of disturbances is zero.21 present historical decompositions of most important structural disturbances (and the innovations respective for the disturbances) present in the SOEPL−2009 model. 2005).Variance decompositions. Therefore.3 Smoothing — estimation of structural disturbances Figures 7. speculations may be formulated regarding the paths of observable variables. depreciation resulting from the appearance of negative net export makes import more expensive and therefore CPI grows. An economy in longterm equilibrium is knocked out of the equilibrium by the appearance of subsequent shocks.
suggests that after a period of relative stabilisation in the vicinity of steady state in the years 2003–2005. we see a speciﬁc role of nonstationary disturbances in technology. the agents index the prices of their products based on increasingly growing rates of (future) inﬂation and think 7 119 WORKING PAPER No. Currently.2 0 0.5 1 1. From the beginning of the 21st century.2 0 0. Its graph represented approximately the world’s business cycle. the increase of investment dynamics observed in the years 2005–2007 was — accordingly to the logic of the model — speciﬁc for the Polish economy and shortlasting. impulse response functions and estimation of disturbances Figure 7. In the previous versions of the SOEPL model there was only one nonstationary disturbance in technology present. nonstationary disturbance in investments is negative.1 0. particularly in the view of the fret that deep (negative) innovations appearing at the end of the sample (related to the crisis in ﬁnancial markets) additionally obscure the proportions (scale). which has been larger than zero since 2004 (with short episode of negative innovation around the year 2005).20. 83 119 .e. there appear positive innovations of the stationary disturbance in investments.5 Inflation target 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 Energy prices 2 0 2 1998 2000 2002 2004 2006 2008 Trying to interpret some of the paths of disturbances presented in the graphs.4 0.3 0. there appeared a trend to increase the target.4 Nonstationary technology 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 Investment nonstationary technology 0 0.2 0.5 0 Risk premium 0.Variance decompositions. or even midterm trends that may be observed for the disturbance in the inﬂation target. yet.2 0.5 0 Asymmetric technology 0. A similar conclusion may be formulated when viewing the stationary disturbance in technology (of the TFP type).4 0. The graph. Estimates of structural disturbances Stationary technology 4 2 0 2 4 6 1998 2000 2002 2004 2006 2008 Investment specific technology 2 1 0 1 0. Thus. while strong positive innovations compensate the negative shocks resulting from trends common for the whole global economy shaped by nonstationary disturbances.2 0 0.5 0 0. the effect is broken down into two disturbances and is no longer so visible.5 1998 2000 2002 2004 2006 2008 1 0.2 Interest rate 1 0.2 Labour supply 1 0 1 Domestic goods markup 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 6 4 2 0 2 4 1998 2000 2002 2004 2006 2008 Imported consumption goods markup 2 1 0 1 1998 2000 2002 2004 2006 2008 Imported investment goods markup 3 2 1 0 1 2 1998 2000 2002 2004 2006 2008 6 4 2 0 2 4 Imported export component markup 5 0 5 10 1998 2000 2002 2004 2006 2008 Export markup Wage markup 1 0. i.4 1998 2000 2002 2004 2006 2008 2 1 0 1 Consumption preference 0.
Larger innovations in risk premium are related. the period of zloty liberation (1999–2000) and accession to the European Union. Explicit consideration of a structural change to interest rate rule resulted in a more homogenous graph of the disturbance path — earlier versions of the model identiﬁed phases of sudden ﬂuctuations in the years 1997–1999 and relative stabilisation (clear decrease in variance) after the year 2001.3 1998 2000 2002 2004 2006 2008 1 0 1 2 Consumption preference 0. Interest rate disturbance (originally identical with innovations) reﬂects more sudden deviations around the years 1999–2000. to the ﬁnancial crisis of 2008–2009.4 Nonstationary technology 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 Investment nonstationary technology 0. however. impulse response functions and estimation of disturbances Figure 7.6 Asymmetric technology 0.5 0 4 2 0 2 1998 2000 2002 2004 2006 2008 Risk premium 0.1 0.5 1998 2000 2002 2004 2006 2008 Energy prices 2 0 2 1998 2000 2002 2004 2006 2008 that the central bank also adjusts the interest rate by accepting higher inﬂation9 . Estimates of structural innovations Stationary technology 2 0 2 4 1998 2000 2002 2004 2006 2008 Investment specific technology 1 0. among others.5 Inflation target 0 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 0.2 1998 2000 2002 2004 2006 2008 Labour supply 1 0 1 Domestic goods markup 0.4 0. 120 120 N a t i o n a l B a n k o f P o l a n d .1 0.2 Interest rate 0. A relatively high positive value of the disturbance implies. with strong depreciation later on.21.3 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 Imported consumption goods markup 1 0.5 1998 2000 2002 2004 2006 2008 0.5 1 1998 2000 2002 2004 2006 2008 Imported investment goods markup 4 2 0 2 1998 2000 2002 2004 2006 2008 Imported export component markup 10 5 0 5 10 1998 2000 2002 2004 2006 2008 5 0 5 Export markup 1998 2000 2002 2004 2006 2008 Wage markup 1 0.3 0. Disturbance in foreign exchange risk premium approximately coincides with intuition. 7 The ﬂuctuations from 2000–2001 and 2006 are more difﬁcult to interpret. In the graph of innovation. the inﬂation episode related to the accession to the European Union is clearly visible.1 0 0. The ﬂuctuations at the end of the sample suggest.2 0.1 0 0. which seems to be natural taking into account the introduction of the ﬂoating exchange rate regime and implementation of the strategy of direct inﬂation targeting. an opposite conclusion may be formulated for the period before the year 2004. the strongest discretionary negative impulse of interest rate in the last 10 years. a change in the trend.4 0.2 0.2 0.2 0 0.2 0.1 0 0. 9 Of course.5 0 0.5 1 1.5 0 0.1 0. however.Variance decompositions. In the graph of the premium disturbance a phase of strong (speculative?) appreciation in the years of 2006–2008 is visible.2 0 0.
Figures 8. thus. The procedures allow making counterfactual scenarios to answer the question of how the paths of variables would have been if some of the shocks had not occurred.1 Historical decompositions The set of calculation procedures that has been built for models in the state space representation enables to carry out historical decompositions and. thus. Using such tools. we may. 83 121 . Historical decompositions are made in consideration of possible deviations from the steady state of all of the variables. Variance decompositions and impulse response functions are determined by the assumption that economy is in steady state.Historical decompositions and forecasts 8 Historical decompositions and forecasts Chapter 8 Historical decompositions and forecasts 8. Therefore. as in the case of variance decompositions or impulse response functions. make an attempt to identify disturbances that had the largest importance for the given variables or/and historical episodes. We are not referring here a theoretical or potential impact of disturbances on the observable variables of the model. In the ﬁrst case we take into account only the disturbance in consumption preference and in the second case — all of 8 121 WORKING PAPER No. but about the actual inﬂuence on historical events. their historical graphs (green line) and the graph of their hypothetical development (blue line).1–8. As a presentation of the analytical capacity of the tools. as well as overlapping/neutralisation of the effects of other shocks.2 present the paths of observable variables. we assess the inﬂuence and role of the given disturbances for the actual series of events. to study the impact of shocks (groups of shocks) on the path of observable variables in the whole historical context of events. we show an exercise regarding the search for key disturbances in the graphs of the particular observable variables (the analysis is limited to the period post 2004). while the studied disturbance innovation (shock) appears in isolation and its effects are not distorted by other shocks.
labour supply preferences. In each case we receive paths of variables resulting from the deviation of economy from the steady state. the historical graph of the real dollar/zloty exchange rate may be recreated with the use of risk premium disturbance and three components of the SVAR model: change in the nominal dollar/euro rate and inﬂation in the USA and the euro area (the real cross rate).4. external disturbances included in the SVAR model are necessary (except interest rates). the role of ﬁscal or technological shocks or work supply preference mentioned as important in variance decompositions (particularly in a longer horizon. As shown in Figures 8. the path of the investment deﬂator may be recreated with the use of disturbance in the markup on imported investment goods and stationary technological disturbance in investments. and export components). It is difﬁcult to observe the role of nonstationary shocks or preferences. The method proposed above enables to state that GDP deﬂator ﬂuctuations result in the studied period from the realisation of the disturbance in inﬂation target and markup on domestically manufactured products.e. Key factors of import proved to be the disturbances in markup on imported goods (consumer and investment goods.1) is hard to perceive. is very difﬁcult to determine a narrow set of shocks to allow the recreation of the paths in the tested sample.Historical decompositions and forecasts the disturbances are present. determination in what state.2 Forecasting technique The SOEPL−2009 model is an example of a linear DSGE model formulated in a state space representation. we are able to make a conditional or unconditional forecast. the considered disturbances and the relationships between the endogenous variables of the model. It has been determined by experiment that for the recreation of the graph of export in the studied fragment of the sample. For the recreation (arrival at a relatively correct approximation) of the path of real wages in the years 2004–2009. Due to the fact that all of the exogenous variables of the model (disturbances) are described with known stochastic 122 122 N a t i o n a l B a n k o f P o l a n d . the exchange rate risk premium is important as well. employment) it . as well as disturbances in export markup and markup on the import of export components. four disturbances are necessary: wage markup. see Chapter 7. Carrying out analogical study we shall notice that the path of investment expenditures is determined by a stationary technological disturbance in investments. except the disturbance in consumption preference. For the other observable variables (GDP CPI. as well as export markups. the economy is. therefore. from the perspective of the model. In the case of export a single disturbance cannot be found to explain the dominating/larger part of the variance of the variable (enabling the recreation of its historical path). that in the studied period consumption was a category dependent practically on the exogenous shock of preference only. Having the values for the vector of state variables at the starting point. domestic interest rate.3 and 8. i. For that class of models the initial stage of making a forecast is the identiﬁcation of the value of the vector of state variables (with the use of Kalman ﬁlter) at the starting point of the forecast. On the other hand. Both for the GDP deﬂator and the investment deﬂator in the last 2–3 years. The aforementioned paths show. point forecast or stochastic forecast. inﬂation target and markup on domestically manufactured products. 8 8.
in Appendix C. a more natural environment for the estimated DSGE models (including SOEPL−2009 ) is the world taking into account the existence of uncertainties. This is illustrated e. generation of uncertainty related to the realisation of shocks in the future. such prepared forecast may be point forecast. state variables at the starting point. the standard errors may be used. We then end up with the estimation of the predictive distribution which can be depicted in the form of a fan chart or used 8 123 WORKING PAPER No.Historical decompositions and forecasts Figure 8. excluded consumption preference shock GDP deflator 10 8 6 4 2 0 2 4 2004 2005 2006 2007 2008 2009 2.5 Real wages 10 8 6 4 2 0 2 4 6 8 Investment deflator 6 5 4 3 2 1 0 2005 2006 2007 2008 2009 2004 2005 2006 CPI Yhat Yobs 2007 2008 2009 2005 2006 2007 2008 2009 2004 2005 2006 2007 2008 2009 2004 processes. Unconditional forecast. yet. • state uncertainty — considering the uncertainty related to the forecast of state variables. and in the case of classical estimation.5 1 2004 2005 2006 2007 2008 2009 GDP 3 2 1 0 1 2 3 4 2004 2005 Consumption 6 5 4 3 2 1 0 1 2 2006 2007 2008 2009 3 2004 2005 Investment 2006 2007 2008 2009 Exports 6 4 2 0 2 4 6 8 2004 8 6 4 2 0 2 4 6 8 10 2005 2006 2007 2008 2009 2004 2005 Imports 10 5 0 5 10 15 20 2006 2007 2008 2009 2004 Real exchange rate 6. no external assumptions regarding exogenous variables are necessary. we have the information about whole distributions.1. Historical decomposition.5 6 5.5 0 0.5 2 1. • uncertainty of future shocks — information about the variances of shocks enables the • measurement errors — forecasts of observable variables are additionally burdened with uncertainty related to the measurement (observation) of observable variables.5 1 0.5 5 4. In our case it is possible to quantify the risk related to: • uncertainty of parameters — when (full) Bayesian estimation is made.5 2005 2006 2007 2008 2009 2004 2005 Interest rate 2006 2007 2008 2009 Employment 2 1 0 1 2 3 2004 2 1.5 1 0. a sequence of structural shocks and a sequence of measurement errors.5 0 0.g. 83 123 . In the simplest case. means that all of the disturbances identiﬁed at the starting point shall expire and upon an adequately long period of time — due to the structure of the model — the variables shall come back to steady state. The above uncertainties are taken into account by conducting stochastic simulations where we draw parameters (from the posterior distribution). assuming that in the future no new shocks shall occur.
uncertainty according to F. Although disturbances have economic interpretation. the very fact of the possibility of (selective) quantiﬁcation of major risk factors. as to which we have intuition and knowledge.5 2 2004 2005 2006 2007 2008 2009 2 1. data. A modest intervention test (modesty metric) by Leeper and Zha (2003). Due to the fact that the principle that shocks explain the behaviour of variables is valid all the time.5 4 2005 2006 2007 2008 2009 2004 2005 Interest rate 2006 2007 2008 2009 Employment 2 1. with the assumed path of one or more observable variables1 . allows to verify whether the values of such shocks are not too high. 1 8 124 124 N a t i o n a l B a n k o f P o l a n d .e.5 0 0. High values of shocks lower the likelihood of conditional forecast.g.5 5 4. Historical decomposition. the user must indicate what disturbances are to guarantee the fulﬁlment of the assumed path. declare speciﬁc values of the shocks as events whose effects we study. With the use of the SOEPL−2009 model there may also be created conditional forecasts (scenarios). Such forecasts are known in literature as conditional forecasts by WaggonerZha or LeeperZha. the most frequent method of conduct requires monitoring of the response of observable data. Waggoner and Zha (1999).5 2 1. conﬁdence intervals.5 0 0.2. 2 When preparing a scenario. as in reality they could be treated by economic agents as a change in the prevailing economic regime or might change the agent’s behaviour (according to Lucas critique). Knight. (2005a). see e. determination of e. Kłos (2004)). i. shows potential of such class of DSGE models. Although the list does not account for all of the potentially important sources of uncertainties (e. included only consumption preference shock GDP deflator 10 8 6 4 2 0 2 4 2004 2005 2006 2007 2008 2009 2.5 1 0.5 1 0. which may help us in determining the correct value of innovation.5 1 0. broader understood uncertainty of the model. which enables explicit determination of the values of shocks2 . i.5 1 1.e.5 1 2004 2005 2006 2007 2008 2009 GDP 3 2 1 0 1 2 3 4 2004 2005 Consumption 6 5 4 3 2 1 0 1 2 2006 2007 2008 2009 2004 2005 Investment 2006 2007 2008 2009 Exports 6 4 2 0 2 4 6 2004 2005 2006 2007 2008 2009 8 6 4 2 0 2 4 6 8 10 2004 2005 Imports 10 5 0 5 10 15 20 2006 2007 2008 2009 2004 Real exchange rate 6.5 2004 2005 Real wages 10 8 6 4 2 0 2 4 6 8 2006 2007 2008 2009 2004 Investment deflator 6 5 4 3 2 1 0 2005 2006 2007 2008 2009 2004 2005 2006 CPI Yhat Yobs 2007 2008 2009 to calculate probabilities of reaching certain levels by certain variables.Historical decompositions and forecasts Figure 8. and only that gives us basis for ﬁnal determination of the value of the shocks.g. see e. it is possible to start with the shocks.g. see also Adolfson et al.5 0 0.g.5 6 5.
g. are forecasts using the assumptions with regard to interest rate. Then. we arrive at a contradiction with the initial assumption about the rationality of the agents. i. unanticipated shocks are used. so the knowledge that the central bank shall start to act on a “nonstandard” basis shall force the agents to “nonstandard” behaviours. 83 125 . in the above technique of constructing conditional forecasts. If — by force of assumption — we exclude the possibility of foreseeing by the agents the “nonstandard” approach of the central bank. to answer the question what would their responses (decisions) be.3.5 1 0. A break in the functioning of a central bank lasting many quarters does not comply with the standards. the possible projections shall be internally Figure 8.Historical decompositions and forecasts Usually. when it stopped to modify the interest rate accordingly to the occurring events. the conditional paths may be generally known (e.e. Historical decomposition.5 2004 2005 Real wages 10 Investment deflator 8 7 6 5 5 4 0 3 2 5 1 0 CPI Yhat 2005 2006 2007 2008 2009 2006 2007 2008 2009 2004 2005 2006 2007 2008 2009 2004 Yobs 125 WORKING PAPER No. it is possible to use the anticipated shocks that shall enable agents to react to the very information about a future event and the realisation of a shock in future shall not be a surprise.5 0 0. expected tax changes) and may not/should not be treated as unexpected events. Laséen and Svensson (2009)). however. In certain situations. In the ﬁrst case we receive “irrational” forecasts of variables (the effect of assumption of “nonstandard” policy of the bank — the problem has been discussed by e. In the second case. excluded risk premium and real dollar/euro rate shocks GDP deflator 10 3 2 1 0 1 2 3 GDP 3 2 1 0 1 2 3 4 2005 2006 2007 2008 2009 2004 2005 Consumption 6 5 4 3 2 1 0 1 2 2006 2007 2008 2009 2004 2005 Investment 5 0 4 5 6 5 2004 2005 2006 2007 2008 2009 2004 2006 2007 2008 2009 Exports 5 0 5 10 15 20 2004 2005 2006 2007 2008 2009 10 Imports 10 5 0 Real exchange rate 10 9 8 7 6 5 Interest rate 8 5 0 5 10 15 5 4 3 2 2005 2006 2007 2008 2009 2004 2005 2006 2007 2008 2009 10 2004 2005 2006 2007 2008 2009 20 2004 Employment 3 2 1 0 1 2 3 4 2004 2005 2006 2007 2008 2009 2. forecasts — or rather projections — are (often) created at central banks with the assumption of exogenous interest rate (constant or implied by market expectations).g.5 2 1. when the central bank stopped to use the rule known to the agents for several or a dozen or so quarters. interesting from the point of view of the central bank. A special case of conditional forecasts. In the world of rational agents (forwardlooking and optimising) it is hard. According to the historical tradition.
the measures of the forecasting quality of the model are constructed..5 2 2004 2005 2006 2007 2008 2009 1.. 2. variables are forecasted for h = 1..5 1 1. T .4....5 0 0.5 0 0. In each case the projection will not provide a reliable. The observable horizon.3 Expost forecasting accuracy of the SOEPL−2009 model Using the observations of observable data Yt for t = 0. is called a forecast error in period t for the h forecast horizon.Historical decompositions and forecasts Figure 8. . 2.5 2004 2005 2 Real wages 10 8 6 4 2 0 2 4 6 8 2006 2007 2008 2009 2004 Investment deflator 6 5 4 3 2 1 0 2005 2006 2007 2008 2009 2004 2005 2006 CPI 2007 2008 2009 inconsistent and shall be subject to Lucas critique.. 2.5 1 0. where H is the maximum forecast 126 126 N a t i o n a l B a n k o f P o l a n d . 1... with their real values in the sample Yt for t = 1.. The analysis will cover onedimensional and multidimensional measures (joint measures). The forecast of observable variables for period t + h made in period t is denoted as ˆ Yt+h . the quality of forecasts of the model in a sample may be veriﬁed by comparison of the theoretical values of the observable ˆ data identiﬁed as Yt . and the actual value of the observable variables in period t + h as Yt+h .5 1 0. T − h.. H. T − 1. H periods ahead. . The difference: ˆ γ t+h = Yt+h − Yt+h we may determine various statistics of forecast errors γ t+h depending on the determined forecast horizon h = 1. included only risk premium and real dollar/euro rate shock GDP deflator 10 8 6 4 2 0 2 4 2004 2005 2006 2007 2008 2009 7 6 5 4 3 2 1 0 1 2004 2005 2006 2007 2008 2009 GDP 3 2 1 0 1 2 3 4 2004 2005 Consumption 6 5 4 3 2 1 0 1 2 2006 2007 2008 2009 2004 2005 Investment 2006 2007 2008 2009 Exports 15 8 6 4 2 0 2 4 6 8 10 Imports 10 5 0 5 10 15 20 2005 2006 2007 2008 2009 2004 Real exchange rate 9 8 7 6 5 4 3 2 2005 2006 2007 2008 2009 2004 2005 Interest rate Yhat Yobs 10 5 0 5 2004 2005 2006 2007 2008 2009 2004 2006 2007 2008 2009 Employment 2 1.. Based thereon.. A solution of the problem is the preparation of forecasts with the assumption of endogenous interest rate. .. 1. Going through t = 0. 2. Historical decomposition. 8 8. 2. methodologically correct answer to the asked question. .. .
2. Firstly. which means that they are systematically underestimated or overestimated. and then around the horizon of 7–11 quarters. The inﬂation in the perspective of 3–5 quarters is associated with the largest error. The mean percentage error for the forecast horizon h = 1.. Secondly. 83 127 . the relation γ t+h (i) Yt+h T −h+1 . The analysis of RMSE shows several systematic features of the forecasting properties of the DSGE SOEPL−2009 model. The root mean square error for the forecast horizon h = 1. H for the i th observable variable is calculated as: T −h γ t+h (i) t=0 Yt+h T −h+1 .3.e. i. If the actual value of the forecast variable takes values close to zero. . regardless of the forecast horizon.. investment goods and CPI) increase to the horizon of about 3–4 quarters and then fall. the inspection of MPE errors shows that the forecasts of most of the variables are biased. we assume the maximum forecast horizon of H = 12 quarters. the model’s forecasts always win with the naive forecasts.. beside single cases. Thirdly. RMSE tells about how much the Yt+h forecast is different on the average from the actual value of the observable variable Yt . MPE = ˆ The value tells by what percentage the Yt+h forecast is different on the average from the real value of the observable variable Yt . which may hardly be deemed typical or representative.. Thus. 8 127 WORKING PAPER No. The forecasting accuracy of the model with respect to forecast horizon is different for other variables where the RMSE increase along the horizon. the period after Poland’s accession to the European Union.. and the sample covers the period of 2004:4–2009:3. the test covers the most turbulent period of ﬁnancial crisis (2008:2–2009:3). consumption goods.. This means that least susceptible to errors is the forecast of inﬂation for 1–2 quarters. 2. depending on the type of inﬂation. the inﬂation forecast errors (domestic inﬂation.Historical decompositions and forecasts In the exercises presented below. becomes very large even in the case of moderate γ t+h (i). 8. .1 Onedimensional measures of forecasting quality From the group of onedimensional forecasting quality measures we have used two: root mean square error (RMSE) and mean percentage error (MPE). H for the i th observable variable is calculated as: T −h 2 t=0 (γ t+h (i)) RM SE = ˆ where γ t+h (i) is the i th element of the vector of γ t+h . The advantage of a DSGE model over naive forecasts increases along the forecast horizon. The basic function of this measure is the reﬂection of the possible bias. The conclusions have been formulated based on the data that are analysed in details in Appendix D.
T − H. Such analysis may be made for every forecast horizon h = 1.. since these contributions are proportional to squares of their coordinates in the eigenvector deﬁning that direction. The ﬁrst direction with the largest variance of forecast errors Variable Export Import Real USD/PLN exchange rate h=1 0. The contribution of each of the variables is signiﬁcant. the share of exchange rate diminishes for the horizon of over 9–10 quarters. the larger the variability of errors in the corresponding direction. 478... 8. respectively. 407 and 583 for the forecast horizons of 1.2 Multidimensional measures of forecasting accuracy Multidimensional analysis of forecast errors of the model3 is based on the covariance matrix of ˆ errors Ωh . For the presentation purposes. which gives a clear picture of forecast errors depending on the forecast horizon.. and check which variables contribute to variability of these directions to the largest extent. H. Each eigenvector of matrix Ωh determines one such direction. with the share for the horizons over 8 quarters.. (2005c)..33 0. 4. 3 An interesting review of the techniques of DSGE models forecast errors analysis is presented by Adolfson et al..40 0. For matrix Ωh its eigenvalues λh have been determined. Table 8. we provide results for horizons h = 1. In the calculations it has been assumed that Ψ is a diagonal matrix with variances of the subsequent forecasts errors on the main diagonal. 2.1–8.20 0. 12 and for each horizon for the directions related to the four largest eigenvalues. as well as the respective vih eigenvectors. realizes mainly the errors of export.26 0. Variables of at least 10% contributions were considered.27 0.4 present how the share of variables generating the largest forecast errors in the total variance of the errors change depending on the forecast horizon. H: Ωh = T −H t=1 γ t+h Φγ T t+h where Ψ is the scaling matrix. Should we select eigenvectors or directions — principal components to which the largest eigenvalues correspond.33 0. . .20 h=4 0. which is determined based on the residuals vector γ t+h = Yt+h = Yt+h for the forecast horizon h = 1. 1. 128 128 N a t i o n a l B a n k o f P o l a n d . regardless of the forecast horizon. Thus.2 shows that in the second of the veriﬁed directions the largest share in the forecast errors has the real exchange rate of the dollar. 4.27 Note: The eigenvalues for the ﬁrst direction amount to 72.Historical decompositions and forecasts 8. . it shall prove which variables generate the largest errors in the forecasting process. 8 dollar.3. 8 and 12. Tables 8. it is easy to quantify contribution of subsequent variables to the variance along any given direction. 2. 2..40 h=12 0. Principal components are eigenvalue related to the eigenvector.37 0..30 h=8 0. Table 8. Table 8. The i determination of the eigenvectors for matrix Ωh is equivalent to the determination of the so called principal components of γ t+h forecasts for t = 0.1 shows that the ﬁrst — most important — of the identiﬁed directions. import and real exchange rate of the orthogonal directions along which the speciﬁed value of forecast error variations (variances) materializes.25 0.1. The larger the T −h . however.
2. 10 and 7 for the forecast horizons of 1. respectively.13 0. 4.4. respectively. 8 and 12.21 0.23 0.00 0.13 h=4 0.00 0.Historical decompositions and forecasts Table 8. the dominating contribution to forecast errors is made by investments.10 0. in the third of the identiﬁed directions. while for horizons longer than 6 quarters — government expenditures and. 110.4 — for horizons up to 5–6 quarters. in the case of short horizons of 1–4 quarters. realise for most of the horizons.19 0. respectively.07 0. for horizons of 6–8 quarters — export and nominal USD/EUR exchange rate.00 h=12 0.01 0. the nominal USD/EUR exchange rate. Investments Government expenditures Real USD/PLN exchange rate Nominal USD/EUR exchange rate h=1 0. For short horizon a contribution that may not be neglected is brought by import and EUR/USD exchange rate.28 0. As it may be seen from Table 8.65 h=8 0.07 0.16 0.26 0.3. In the fourth of the identiﬁed directions — see Table 8. the largest share in forecast errors falls to investments and nominal USD/EUR exchange rate.5–8.01 0.1 0.00 0.16 Note: The eigenvalues for the second direction amount to 26. a major share falls to export and between 7 and 10 quarters — import.04 0.02 0.3. 4. Table 8.00 h=8 0.14 0. while for the 12 quarters horizon the forecasts are dominated by investments.1 0.05 0.41 0. The second direction with the largest variance of forecast errors Variable Investments Export Import Real USD/PLN exchange rate Nominal USD/EUR exchange rate h=1 0.10 0. yet.00 0.04 0.57 0.52 0.62 0.20 h=8 0.3 Rolling forecasts Figures 8.24 0.04 h=4 0. 8 and 12.16 0.6 present the historical graphs of observable variables (thick line) and a series of forecasts for 12 quarters horizon (made in compliance with the description provided at the 129 WORKING PAPER No.25 0.02 0. 83 129 .68 0. 8 and 12.05 0.13 0.57 0. 4. amounting even to 80%.62 h=4 0.20 0. Table 8.35 h=12 0.45 0.08 0. The fourth direction with the largest variance of forecast errors Variable Inﬂation in investments. 32 and 45 for the forecast horizons of 1.00 0.46 0.11 h=12 0.00 0. For the horizon between 1 and 6 quarters. 8. 104 and 188 for the forecast horizons of 1. The third direction with the largest variance of forecast errors Variable Investments Export Government expenditers Nominal USD/EUR exchange rate h=1 0.3.23 Note: The eigenvalues for the third direction amount to 11.00 8 Note: The eigenvalues for the fourth direction amount to 5.04 0. 20.46 0. 25. mainly. while the share of investments is moderate.
CPI 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 GDP 7 6 5 4 8 6 4 2 0 2 4 Consumption 20 15 10 5 0 Investments 8 3 2 1 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 Export 10 5 0 5 10 15 20 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 10 10 0 Import 8 6 4 2 20 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 Real wages 0 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 4 The exception is employment. and real exchange rate of dollar expressed in the form of deviations from the linear trend. consumption and inﬂation (CPI. that the forecasts of several turning points of investments. turning points or sensitivity of forecasts to new information. In the graphs annual dynamics of variables has been presented4 .investment deflator 5 4 3 2 1 0 Inflation .g.Historical decompositions and forecasts beginning of paragraph 8. consumption. It should be emphasized however. 130 130 N a t i o n a l B a n k o f P o l a n d . subsequent forecasts (for example) become instable (subsequent forecasts differ widely). Figure 8. may be supplemented with the illustration of the ability of the model to forecast midterm trends. Ex post forecasts for the selected variables of the model (part 1) Inflation . as well as small explanatory power of the model in the case of the exchange rate. onedimensional measures seem to be slightly exaggerated. investments).5. GDP and inﬂation are accurate. the characteristics of forecasting errors of the SOEPL2009 model. Using this illustration. The ﬁrst observation is that the inaccuracies of forecasts resulting from the earlier presented.GDP delator 6 4 2 0 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 4 2 0 2 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 Inflation . which has the form of deviations (absolute values) from the HP trend. GDP deﬂator) or even investments seem to be relatively well forecasted. Contrary to the above presented measures. When such sudden shocks occur. In the case of many variables (e. An exception is the period when the economy was exposed to the shocks of the global ﬁnancial crisis.3). relatively large root mean square errors result from single observations (individual large errors during the ﬁnancial crisis). Tendency of the model to underestimate the dynamics of export and import is conﬁrmed (the effects of which are reﬂected also in the GDP forecasts). given by formal errors measures.
5 1 0.5 Inflation in the euro area 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 0 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 Inflation in the USA 5 3. 83 131 .5 1 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 4 3 2 1 Interest rate in the euro area 6 5 4 3 2 1 0 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 Interest rate in the USA 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 8 131 WORKING PAPER No. Ex post forecasts for the selected variables of the model (part 2) Employment 3 2 1 0 1 2 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 10 5 0 5 10 15 20 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 4 6 5 Real exchange rate of dollar 7 Interest rate 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 Production in the euro area 2 0 2 4 04:4 05:4 06:4 07:4 08:4 09:4 10:4 11:4 4 2 0 2 Production in the USA 2.5 2 1.Historical decompositions and forecasts Figure 8.5 2 1.5 3 2.6.
Adolfson et al. shall support and supplement the forecasting materials prepared based on the traditional macroeconometric model and the opinions of experts at the National Bank of Poland. DSGE models are also becoming useful tools for empirical research.g. Smets and R.. When we were interested in the accuracy of forecasts.e. i. Therefore forecasting experiments with Bayesian estimated DSGE models carried out by F. The conclusions were conﬁrmed also for other estimated DSGE models of developed and stable market economies (see e.. Forecasting of inﬂation or business cycles for stable developed market economies is different from the forecasting of the responses of agents subjected for a dozen or so years to institutional changes. More complex estimation of parameters. Christoffel et al. eclectic models in which more or less random correlations of data enable the reduction of errors dominated the models with clearer (more explicit) economic contents. 2007b). DSGE models are one of the most important tools of theoretical analyses of modern macroeconomics. Strict theoretical grounds and explicitly declared paradigm are features that simplify the interpretation of results but — as it seemed initially — lower the forecasting potential of models. societies that attempt building market institutions from scratch and shape the macro 132 132 N a t i o n a l B a n k o f P o l a n d . New theoretical concepts are developed based on DSGE models or the environment of economy in dynamic general equilibrium. Being an effective tool of theoretical research. particularly with the use of the ideas of Bayesian econometrics. enables matching models motivated with economic theory to data — referring to the real existing economies.Final comments Final comments Final comments The presented material documents the results of works over a version of an estimated dynamic stochastic general equilibrium model SOEPL−2009 . 2005c. analysing the reasons of observable events. Wouters (2004) enjoyed high interest of the institutions managing the macroeconomic policy. testing the actual episodes. All of that from the point of view of explicitly declared economic paradigm (model speciﬁcation). which in 2010 will be used for building midterm forecasts and projections of inﬂation processes and business cycle in Poland. The experiments showed that an estimated DSGE model has not only analytical but also forecasting potential.
the current features of the SOEPL−2009 model. 2008). require further work. while the general uncertainty is much higher.Final comments and microeconomic rationality of ﬁrms. for the euro area countries. 133 WORKING PAPER No.. however. to formulate a cautious conclusion that also for the group of economies in which Poland is included. while the development potential of DSGE models guarantees the effectiveness of such projects. Andrle et al. Unchanging interrelationships assumed within the DSGE model (deep parameters deﬁning the structure of the model of economy) are more difﬁcult to identify. 2005. 83 133 . Using the word “potential” of the model we wish to emphasise that although the progress achieved in reference to the earlier versions of the SOEPL family of models is signiﬁcant. including the forecasting characteristics. The experiments with the family of SOEPL models enable. a DSGE model of signiﬁcant analytical and forecasting potential may be developed5 . social organisations and state institutions. Therefore the task of building a DSGE model for the Polish economy enabling the construction of midterm forecasts and projections proved to be a harder task than the construction of a analogical model e. 2009. households.g. 5 This is conﬁrmed by the experience of the analysts dealing with DSGE models at the central banks of Hungary and the Czech Republic (see Benesz et al. Jakab and Világi..
Part IV Appendix 134 .
83 135 .. .t 20 e at ..3) θst = .. (A.List of equations.. ν tw θτt = ν tk 27 p τk t 28 y τw t 29 30 τs t 31 p πoil t 32 33 34 y . list of variables Appendix A Appendix variables List of equations. list of A List of equations.1) πd t 11 πmc t 12 u ∆S t 22 πmi t 13 πmx t 14 wt 16 Ht 17 ct 18 it 19 ψz...t 14 µΨ..u γt ut 25 qt 26 m t+1 27 µt 28 at .. pk . where: ˜ z t [1x28] = . τt τt τc t gt τ t−1 τ t−1 τc t−1 g t−1 ..t πt .. . 135 WORKING PAPER No. 8 εt 11 Υt 12 ˜ zt 13 x λt 23 µz. θ τ t [1×8] ... list of variables Forms of the model The structural form of the model can be written as: z z z t α0 ˜ t+1 + α1 ˜ t + α2 ˜ t−1 + β0 θ t+1 + β1 θ t = 0 θ t = ρ θ t−1 + t . γmcd t γmid t γmx d t xu t x tx Rt Et . 1 2 3 4 5 x πt 15 6 7 8 9 10 (A.2) t [1×14] (A..t 15 ζc t 16 ζh t 17 ζt q λd t 19 λmc t 20 c . 18 λmi t 21 λmx t 22 λw t 24 ˜e φt 25 ˜e φ t−1 26 ˜u φt ˜u φ t−1 εR.t 21 yt 23 k t+1 24 x... θ 1 2 3 4 5 6 7 . 9 10 θ t [1×48] = θ s t [1×26] ..
we can write: ˜ ˜ z t+1 = A z t + B θ t+1 θ t+1 = ρ θ t + ε t+1 or ˜ z t+1 θ t+1 ξ t+1 = .. anticipated variables are eliminated (among others) from the structural form. 35 36 37 38 39 40 41 y te 42 e y t−1 πe t 43 Re t 44 y tu 45 u y t−1 πu t 46 Ru t 47 ∆S tx 48 x S t−1 .. (υ t+1 υ t+1 ) = Q. list of variables θ t Upon the solution of the model with AndersonMoore algorithm. 136 136 N a t i o n a l B a n k o f P o l a n d . Y t = Ax x t + H ξt + ut ..4) The elements of the matrices of the above formulas are functions of DSGE model parameters. = A 0 Bρ ρ Fξ ˜ zt θt ξt + B I υ t+1 ε t+1 . πe t−1 Re t−1 πu t−1 Ru t−1 . Thus.List of equations. The state space model has the form of: ξ t+1 = Fξ ξ t + υ t+1 .. (A. (u t u t ) = R.
17. 27.t pk .t u ∆S t yt k t+1 ut qt m t+1 µt at e at mcd γt γmid t γmx d t xu γt inﬂation of imported consumer goods [ = real wages hours worked consumption investments marginal utility of income relative price of ﬁxed assets growth of nominal PLN/USD exchange rate gross domestic product ﬁxed assets ﬁxed assets’ utilisation rate cash money money growth rate total net foreign assets net foreign assets denominated in euro ratio of the prices of imported consumption P mc goods to domestic prices [ = Pt d ] t Ptmc mc ] Pt−1 Ptmi investment goods [ = P mi ] t−1 P mx export components [ = P tmx t−1 ] ] ratio of the prices of imported investment goods to domestic prices [ = ratio of the prices of imported components of export Ptmx to domestic prices [ = P d ] t Ptmi Ptd ] xu t x tx Rd t Et ratio of export prices to the prices in the rest of the world [ = real USD/PLN exchange rate real USD/EUR exchange rate domestic shortterm interest rate (gross) employment Ptx Ptu ] 137 WORKING PAPER No. 16. πd t x πt — — — — — — — — — — — — — — — — — — — — — — — — — — — — inﬂation of domestic intermediate products [ = inﬂation of export products [ = inﬂation of imported inﬂation of imported Ptx x Pt−1 Ptd d Pt−1 ] πmc t πmi t πmx t wt Ht ct it ψz. 10. 11. components of vector z t 1. 9. 28. 24. 18. 20. 5. 2. 6. 25. 22. 12.List of equations. 8. 26. list of variables List of model variables ˜ The list of endogenous variables of the SOEPL−2009 model. 21. 15. 7. 3. 23. 13. 4. 83 137 . 14. 19.
g t ˜ List of exogenous variables of the SOEPL−2009 model.t 5. φ t ˜u 18. list of variables List of exogenous disturbances — 1. Ru t 41. y tu 39. observable disturbances p — disturbance in effective rate of corporate income tax 27. ν tk τk t τw t τs t πoil t — — — — — ˜ variables of the SOEPL−2009 model. λmx t x — 13. 25. Re t — disturbance in GDP in the USA 38. 24. φ t ˜e 16.List of equations. ∆S tx — disturbance in the change of nominal USD/EUR exchange rate 138 138 N a t i o n a l B a n k o f P o l a n d . λmi t — 12. τ t — disturbance in effective rate of income tax 29. εR. components of vector θ t . structural disturbances of the global economy — disturbance in GDP in the euro area 35. Υ t ˜ 3. ζc — t — 7. µΨ. φ t−1 — — 19. structural stationary disturbance in technology (TFP) stationary disturbance in investments stationary asymmetric disturbance nonstationary disturbance in technology nonstationary technological disturbance speciﬁc to investments disturbance in consumption preferences disturbance in labour preferences (labour supply) disturbance in preferences of cash holdings disturbance in markups on domestic intermediate products disturbance in markups on imported consumption goods disturbance in markups on imported investment goods disturbance in markups on imported export components disturbance in markups on exported products disturbance in wage markup disturbance in risk premium in the euro market as above disturbance in risk premium in the dollar market as above disturbance in interest rate (monetary policy) disturbance in inﬂation target disturbance in demand for working capital loan for ﬁnancing labour services disturbance in demand for working capital loan for ﬁnancing capital services disturbance in capital tax disturbance in national insurance contributions paid by the employees disturbance in national insurance contributions paid by the employers disturbance in raw materials (energy. τc — disturbance in effective rate of consumption tax t — government consumption 30. y te — disturbance in inﬂation in the euro area 36. πe t — disturbance in shortterm interest rate in the euro area 37. φ t−1 — ˜u — 17. πu — disturbance in inﬂation in the USA t — disturbance in shortterm interest rate of the dollar 40. ε t — 2. oil) prices ˜ List of exogenous variables of the SOEPL−2009 model. λmc t — 11. λw t ˜e — 15. z t — — 4. components of vector θ t . τ t y 28. λd t — 10. 23. µz. ζh t q 8. λ t — 14. components of vector θ t . ζ t — — 9.t — 20. πc t — 21. 26. ν tw 22.t — 6.
10) + (λw − 1) ζh − bw β ξw 1 − κw π t+1 + bw ξw 1 − κw π t t τw + t τ y (λw − 1) 1 − τy y τt = 0 139 WORKING PAPER No. Inﬂation of domestic prices 1 + κd β µ πd = 1 − κd t + 1 − ξd ξd 1 − β µ ξd + 1 − β µ ρπ π t + β µ π d + κ d π d t+1 t−1 wt + 1− Ht − µΨ. list of variables A3.1). List of equations of the structural form of the model Below we present the loglinearised equations of the structural form of the model (as given by A.8) Inﬂation of prices of export products x 1 + κx β µ πt = 1 − κx x x 1 − β µ ρπ π t + β µ π t+1 + κ x π t−1 c + Real wages 1 − ξx ξx 1 − β µ ξx − cm c γcmc x xu γmcd − γ t − x u + ω x γmx d + λ t t t t (A.t + ut + τ s c νfwR Rf w R t−1 + ν f w (R − 1) Rf w νt fw (A.t + (λ − 1) σ L H t + (λw − 1) λw + t τw (λw − 1) 1 − τw w c (A. 83 139 .5) 1 + τs τs + ν τ πoil − πu − ε t + λd t t t t Inﬂation of prices of imported consumption goods 1 + κmc β µ πmc = 1 − κmc t + ξmc 1 − β µ ρπ π t + β µ πmc + κmc πmc t+1 t−1 1 − β µ ξmc x u + (ωmcu − 1) x tx − t c 1 − ξmc cd cγ γmcd + λmc t cd t (A.6) Inﬂation of prices of imported investment goods 1 + κmi β µ πmi = 1 − κmi t + 1 − ξd ξmi 1 − β µ ξmi 1 − β µ ρπ π t + β µ πmi + κmi πmi t+1 t−1 cm c γcmc γmcd − γmid + λmi t t t c x u + ωmiu − 1 x tx + t (A. Loglinearised variables are identiﬁed with hats.9) bw β ξw πd −bw ξw t+1 1 + κw β cm c γcmc cd c γcd πd + bw ξw κw t cm c γcmc cd c γcd πd t−1 −bw κw β ξw w πmc + bw ξw κw t πmc t−1 w t + bw ξw w t−1 c +bw β ξw w t+1 + λw σ L − bw 1 + β ξ2 w − (λ − 1) ψz + .t + kt − µz.7) Inﬂation of prices of imported export components 1 + κmx β µ πmx = 1 − κmx t + 1 − ξmx ξmx 1 − β µ ξmx 1 − β µ ρπ π t + β µ πmx + κmx πmx t+1 t−1 cm c γcmc γmcd − γmx d + λmx t t t c x u + (ωmxu − 1) x tx + t (A.List of equations.
11) bc t−1 − 1− µΨ µz 2 1− + β b 2 c t + µΨ µz b β c t+1 1− − β b − µΨ µz 1− − β b − µΨ µz 1− b − µΨ µz 1− b − µΨ µz cm c γcmc ψz + .t+1 − − τk 1 − τk µΨ.t (A.13) 1− ˜ S µz.t 1− + µz µΨ ˜ S β 1 1− µΨ. list of variables Demand for labour services Ht = − Consumption 1− µΨ µz 1 (1 − (1 − ) ) yt − kt − 1 (1 − (1 − ) ) εt + 1 (1 − ) 1− µΨ.t + β b 1− µΨ µz 1− µΨ µz µz.t+1 1− τc µΨ.t + β b 1− µΨ.t+1 − π t+1 1− π µz + − β k π µz + − βτk τ t+1 + Rt µz + π µz + π (A.14) 140 140 N a t i o n a l B a n k o f P o l a n d .List of equations.t+1 =0 Marginal utility of income ψz + .t + ut (1 − ) µz.t + µz µΨ 2 ˜ S β µz.t+1 (A.t + Υ t − ωi γimi 1 ηi −1 γmid t 1 1− µz µΨ 1 2 1− + µz µΨ 1 1− − µz µΨ 1− − µz µΨ 2 ˜ S i t−1 − 1 + β 1 1− 2 ˜ S it + β 1 1− µz µΨ 2 ˜ S i t+1 (A.t+1 1 1 2 ˜ S 2 µΨ.t = ψz + .t γmcd t −b −b 1− µΨ µz 1− µΨ µz µz.t+1 + µz.12) 1− − β b − µΨ µz 1− − µΨ µz 1− b − µΨ µz (1 + τc ) τc t ζc = 0 t+1 1− b − µΨ µz ζc + β b t 1− b − µΨ µz Investments pk .
u xd x x y x x d y te + 1 − y tu − η f u + (η f e − η f u ) r k 1 1− µz µΨ ye y ˜ x tx + z t .18) + µz µΨ 1 1− +δ−1 Υt − 1−δ µz µΨ 1 1− µz.List of equations.t + µz.+ (A.t + τs s t 1 + σa + πoil t τk 1 − τk 141 τk − R (R − 1) R t−1 (A.t+1 1− 1− µz µΨ µz µΨ τ k ν t+1 + w t+1 + H t+1 − u t+1 + k t+1 + τs 1 + τs t+1 e u u x ˜e ˜e 1 − φse ∆S t+1 = 1 − φse ∆S t+1 − φse ∆S tx + φse ∆S t + R t − Re + φa a e a t − φ t t (A.19) µΨ. balance for the real GDP yt = + + g y gt + cd y ye y ct + cd y ηc cm cγ ye y y γmcd + cmc t id y it + id y ηi im i γimi γmid t ye y γt − η f e x. 83 141 .t − 1−δ µz µΨ 1 1− 1 1− µΨ.20) WORKING PAPER No.t+1 + β 1 1− µz µΨ 1−δ+δτ µz µΨ 1 1− p pk .17) y x η x x ω x γmx d + t k ut Fixed assets 1 k t+1 = 1− µz µΨ 1 1− µz µΨ 1−δ 1 kt + 1− µz µΨ 1 1− µz µΨ +δ−1 it (A. list of variables Relative prices of ﬁxed assets 1 pk .15) − 1 1− − ν (R − 1) Rf k Dynamics of the nominal dollar/zloty exchange rate β (1 + δ τ p − δ) β (1 + δ τ p − δ) µ µ 1 1 Ψ.t + q νw R Rf w 1 1− − νk R Rf k R t−1 + τs 1+τ ν w (R − 1) Rf w − 1 ν tw − ν aτ ν k (R − 1) Rf k u ∆S t ν tk + w t + H t − k t (A.t = − ψz + .t+1 − νw R Rf w τ p 1− µz µΨ 1 1− (1 − τ p ) νk R Rf k Rt + −β τ t+1 p µz µΨ + − β (1 + δ τ p − δ) µz µΨ k 1 1− − ν w (R − 1) Rf w s w ν t+1 (A.t+1 − z.t + ψz + .t Fixed assets’ utilisation rate ut = 1 1 + σa + Cash qt = 1 σq ζ t − ψz + .16) Real income.
23) e ˜ ˜e e Ru − Re φ u + φa a t−1 t−1 t−1 + a ˜u − 1 + φa a e 1 ae u ˜ ˜u a t−1 + φse − φs ∆S t u x ˜ ˜u ˜u ˜e ˜ ˜ + 1 − φse ∆S tx + φse − φs ∆S t−1 − φse ∆S t−1 + φ t−1 − φ t−1 142 142 N a t i o n a l B a n k o f P o l a n d .t + µz.t (A.+ + x 1 − ux + −R a e π µz + y te + 1 − (A.21) νk rk k R m µz µΨ Rf k 1 1− Rf w ν tk − νk Rf k R t−1 + + + + m m µz µΨ k 1 1− 1− k ν k (R − 1) ν r k 1 1− k (R − 1) Rf w w µz µΨ k ν r k µz µΨ 1 1− + ν m 1− µΨ. list of variables Money m t+1 = µ t + m t − πd − t Money growth rate µt = q m qt − mt + πt + νk rk k m ν w 1− νw µΨ.22) τs 1 + τs τs t Total net foreign assets — classic version a t = −c m c t − i m i t + x 1 − 1 − u x −x + η f u + (η f e − η f u ) ye y x tx γt x.t − µz.u c m c d ηc c uc ωmcu + ui ωmiu + u x ωmxu − 1 + 1 − u x η f e γcd γmcd + t ye y i m i d ηi i γid ye y γmid + x m η x x 1 − ω x γmx d t t ye y ˜ y tu + z t .List of equations.t + 1 − (1 + τs ) w H w t + H t + q m ν τ πoil − πu t t + (1 + τs ) w H ν tw (A.
24) γid λd 1 − γmid γmid t kt + y λd +y + + y λd y λd Rf k − 1 Rf k ut Rf w − 1 Rf w Rf k − 1 Rf k µz.30) Real dollar/euro exchange rate x x tx = ρ x x x t−1 + πu − πe + ∆S tx t t (A. list of variables Total net foreign assets — balance of total incomes and total expenditures at + g g t + c d + c m ct + i d + i m it = = −R a e π µz + e ˜ ˜e e Ru − R e φ u + φa a t−1 t−1 t−1 + a ˜u − 1 + φa a e 1 ae u ˜ ˜u a t−1 + φse − φs ∆S t −x + c d c m ηc c xu uc ωmcu + ui ωmiu + u x ωmxu − 1 x tx + x γ t u x ˜ ˜u ˜u ˜e ˜ ˜ + 1 − φse ∆S tx + φse − φs ∆S t−1 − φse ∆S t−1 + φ t−1 − φ t−1 γcd λd − 1 λd (1 − εt − 1 − γmcd γmcd + t yt + ) y (1 − λd + y λd ) i d i m ηi i Ht + y (A.List of equations.28) (A.27) (A.31) 143 WORKING PAPER No.25) Relation of prices of imported consumption goods γmcd = γmcd + πmc − πd t t t t−1 Relation of prices of imported investment goods γmid = γmid + πmi − πd t t t t−1 Relation of prices of imported export products γmx d = γmx d + πmx − πd t t t t−1 Relation of export prices to the prices in the dollar area xu xu x γ t = γ t−1 + π t − πu t (A.t − Net foreign assets denominated in euro e at = 1 ˜u a e φa u u ˜u ˜u ˜u ˜u R t − Ru − 1 − φs ∆S t+1 + φa a t + φs ∆S t − φ t t (A. 83 143 .29) Real dollar/zloty exchange rate u x u = x u + ∆S t + πu − 1 − ω c t t t−1 γcd ηc −1 πd − ωc γcmc t ηc −1 πmc t (A.26) (A.t y 1− − 1 ν τ πoil − πu t t λd µΨ.
λmx = 1. list of variables Nominal interest rate R t = ρR R t−1 + 1 − ρR + 1 − ρR rπ − r∆π 1 − rπ π t + r∆π cd c γcd πd + t−1 c cd c γcd πd + r∆π t cm c γcmc πmc t πmc t−1 (A.38) (A. γmx d = 1.35) Nominal interest rate R= 1 − τk β µ − τk β (A.37) 144 144 N a t i o n a l B a n k o f P o l a n d .34) Inﬂation.39) (A.List of equations.36) Interest rate of working capital loan for labour services Rf w = νw R + 1 − νw Interest rate of working capital loan for capital services Rf k = νk R + 1 − νk Relation of prices of imported consumption goods to domestic prices γmcd = λmcd γ x d gdy γx d = 1 to γmcd = λmc (A. ℵ0 y = x then the values of other variables are determined in compliance with the following equations: Technology growth 1− µ z + = µΨ µ z (A. we assume that in the steady state the following are satisﬁed: λ x = 1. γ x d = 1.t 1 − ρR rπ − r∆π cm c γcmc ˜ β ξe E t+1 + 1 − β ξe ˜ ˜ 1 − ξe H t − 1 + β ξ2 E t + ξe E t−1 = 0 e (A. inﬂation target π= µ µz + (A.33) Steady state solution Based on theoretical analyses.32) + r∆ y y t + Employment eu 1 − ρR r y − r∆ y y t−1 + 1 − ρR r x x t−1 + εR. γ xu = 1.
45) Real wages w= 1 (1 + τs ) R f w λ d 1 1 1− 1− R fk r 1 −1 (A.43) Relation of prices of investment goods to the prices of imported investment goods γimi = 1 − ωi λmi ηi −1 + ωi 1 1−ηi (A.44) Rent for lease of ﬁxed assets services r = k µz + µΨ − β [1 + δτ p − δ] β (1 − τ p ) −1 γid (A.List of equations.41) Relation of prices of investment goods to domestic prices γid = 1 − ωi + ωi λmi 1−ηi 1 1−ηi (A. list of variables Relation of prices of imported investment goods to domestic prices γmid = λmid γ x d gdy γx d = 1 to γmid = λmi (A.47) 145 WORKING PAPER No.42) Relation of prices of consumption goods to the prices of imported consumption goods γcmc = 1 − ωc (λmc )ηc −1 + ωc 1 1−ηc (A.46) Relation of capital to labour k H = Rf w 1 1− Rf k rk (1 + τs ) w µΨ µz + (A. 83 145 .40) Relation of prices of consumption goods to domestic prices γcd = 1 − ωc + ωc (λmc )1−ηc 1 1−ηc (A.
48) (1 − τ y ) (1 − τw ) 1 µz + − b (1 + τc ) µz + − β b Labour services H= D1 D4 D2 σ D3 L 1 1+σ L (A.49) Fixed assets services k=H k H (A. list of variables Auxiliary variables D1 = D2 = − D3 = D4 = 1 AL γcd 1 − ωc 1 − gr 1 − ωi γcd ηc + ωc γcmc k H + ωi γimi w λw 1 σL ηc 1 µΨ µz + γid ηi ηi µz + µ Ψ + δ − 1 µz + µΨ k H (A.54) Domestic component of investment goods i d = 1 − ωi γid ηi i (A.52) c m = ωc γcmc ηc c (A.51) Domestically manufactured consumer goods c d = 1 − ωc Imported consumer goods γcd ηc c (A.53) Investment expenditures i=k µz + µ Ψ + δ − 1 µz + µΨ (A.50) Total consumption c= D2 D1 H (A.55) 146 146 N a t i o n a l B a n k o f P o l a n d .List of equations.
83 147 . list of variables Imported component of investment goods i m = ωi γimi ηi i (A.56) Imported component of export x m = ωx x cm + im 1 − ωx k H (A.61) 147 WORKING PAPER No.58) 1− y = µΨ µ− z − H (A.57) Total export x= Production (A.60) Cash q = Aq m (A.List of equations.59) Broad money m= νk rk k µΨ µz + + ν w (1 + τs ) w H 1 − Aq (A.
27 1.13 0.10 −0.08 0.00 0.02 −0.1) Results of estimation of the SVAR model Results of estimation of the reduced form of the SVAR model are the following2 : y te πe t e rt u yt πu t u rt xt = 1.03 −0.47 0.06 −0.10 0. the following set of restrictions has been assumed for the dynamics of a SVAR model1 : y te πe t e rt u yt πu t u rt xt = ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 0 ∗ ∗ ∗ ∗ 0 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 0 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ e y t−1 πe t−1 e r t−1 u y t−1 πu t−1 u r t−1 x t−1 + et .19 0.25 −0.28 −0.96 0.10 −0.2) We omit exogenous variables.42 −0.28 −0.66 −0.00 0.02 −0.00 −0.Global economy SVAR model Appendix B Appendix model Global economy SVAR B Global economy SVAR model SVAR model identiﬁcation In the estimation of the reduced form of the global economy model.28 −0.00 0.00 r t−2 u 0.00 0.12 0.15 0.04 0.43 0.11 0.26 0.05 0.10 −0.22 e y t−1 e π t−1 e r t−1 u y t−1 + u π t−1 u r t−1 x t−1 e y t−2 0. Exogenous variables have been omitted.10 0.06 0.13 −0.00 r t−2 −0.23 −0.14 0.16 −0. u π 0.39 0.00 t−2 e 0.00 1.24 0.06 0.00 0.26 0.11 0.03 0.04 −0.19 −0.02 −0.03 0. + ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 ∗ ∗ ∗ ∗ 0 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 ∗ ∗ 0 ∗ e y t−2 πe t−2 e r t−2 u y t−2 πu t−2 u r t−2 x t−2 (B. 148 148 N a t i o n a l B a n k o f P o l a n d .15 0.02 πe 0.25 3.05 0.11 x t−2 The structuralisation mechanism assumed in the identiﬁcation of structural shocks of the SVAR model was the following: 1 2 −0.04 0.00 0.07 −0.00 0.06 0.05 −0.38 −0. as no restriction has been imposed thereon.12 0.18 −0.00 0.01 + −0.37 0.00 0.05 −0.12 (B.00 0.06 −2.08 2.02 0.36 0.12 0.02 y t−2 + e t .51 0.21 0.00 0.03 0.12 −0.85 0.09 0.11 0.35 0.34 0.00 t−2 u 0.05 0.22 −0.24 0.00 −0.31 0.01 −0.
(B.33 .00 0.00 0.01 0.00 0.00 0.00 0.52 0.00 0.00 0.00 0.03 0.00 0.00 0.00 0.00 4.00 0.00 0.03 0.00 0.09 0.00 0.00 0.00 B= ∗ ∗ ∗ 0 0 0 0 0 ∗ ∗ 0 0 0 0 0 0 ∗ 0 0 0 0 0 0 0 ∗ ∗ ∗ 0 0 0 0 0 ∗ ∗ 0 0 0 0 0 0 ∗ 0 0 0 0 0 0 0 ∗ .00 0.4) 149 149 WORKING PAPER No.08 0.00 0.00 0.00 0. 83 .00 0.00 0.22 0.00 0.00 0. (B.16 0.43 0.26 0.00 0.00 0.3) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.Global economy SVAR model The estimated structural matrix B is the following: B= 0.00 0.
.Convergence to the steady state Appendix C Appendix C state Convergence to the steady Convergence to the steady state The ﬁgure shows the behaviour of observable variables. Figure C. As results from the graph. real categories (GDP consumption.GDP delator 5 4 3 2 1 08:4 12:4 16:4 20:4 24:4 28:4 32:4 Inflation . Deviations of inﬂation (GDP and investments deﬂator. etc. if the horizon of the example forecast is extended to 100 quarters. Of course.) converge to the steady state relatively fast. CPI) from the steady state caused by impulses of the years 2008–2009 abate relatively slowly. investments.investment deflator 5 4 3 2 1 2 0 1 08:4 12:4 16:4 20:4 24:4 28:4 32:4 1 4 5 Inflation .CPI 7 6 5 3 4 3 2 1 08:4 12:4 16:4 20:4 24:4 28:4 32:4 GDP 08:4 12:4 16:4 20:4 24:4 28:4 32:4 Consumption 8 6 15 4 2 0 2 08:4 12:4 16:4 20:4 24:4 28:4 32:4 10 5 0 20 Investments 15 10 5 0 5 10 15 20 08:4 12:4 16:4 20:4 24:4 28:4 32:4 Export 20 10 0 10 20 Import 08:4 12:4 16:4 20:4 24:4 28:4 32:4 08:4 12:4 16:4 20:4 24:4 28:4 32:4 Real wages 8 7 6 5 4 3 2 1 08:4 12:4 16:4 20:4 24:4 28:4 32:4 1 1 2 Employment Real exchange rate of dollar 10 5 0 5 6 5 4 3 08:4 12:4 16:4 20:4 24:4 28:4 32:4 7 Interest rate 0 10 15 20 08:4 12:4 16:4 20:4 24:4 28:4 32:4 08:4 12:4 16:4 20:4 24:4 28:4 32:4 150 150 N a t i o n a l B a n k o f P o l a n d . Example of ex ante forecast with long horizon Inflation . the convergence of observable variables to the steady state is guaranteed by the structure of the model.1.
about 1. Table D. MPE — dotted line and right axis. . consumption (oPiec).p. investments (oPiei) and CPI (oPiecpi) 1. The upper left panel in Figure D.4 1. RMSE and MPE for deﬂators: GDP (oPied). The data provided therein shall be analysed further herein in more detail1 .2 1 0.8 −100 15 oPiei oPiecpi RMSE — continuous line and left axis.1–D.5 p. Figure D.p.1.8 0 5 10 20 0 −20 −40 −60 15 0 −10 −20 −30 0 5 10 −40 15 oPied oPiec 2.6 1.8 1. Positive The set of variables for which the accuracy of projections was tested covered — beside of all of the observable variables of the model used in the estimation of parameters — also two additional variables: consumption deﬂator (identiﬁed as oP iec) and expenditures on government consumption (oGov).2 present the mean values and standard deviations of RMSE and MPE errors calculated for the horizons h = 1. depending on the forecast horizon.8 0 5 10 20 15 10 5 0 15 100 1.6 1.1–D.3 presents the relations of RMSE errors of a DSGE model and naive forecasts depending on the forecast horizon.Analysis of forecasts accuracy Appendix D Appendixaccuracy D Analysis of forecasts Analysis of forecasts accuracy Figures D. 2.2 2 1. for the forecast horizon of 3 quarters. afterwards it drops systematically to about 1 p..4 0 5 10 −50 0. Tables D. 12. In relative terms the error ﬂuctuates depending on the forecast horizon from several percent to nearly 20% for the 5 quarters’ horizon.2 1 0.4 50 0 1.. 1 151 WORKING PAPER No.2 1 1. 83 151 .6 1.5 present the RMSE and MPE errors for observable variables of the model.1 shows that the root mean square error of the forecast of domestic inﬂation is the largest.4 1..
The advantage of the DSGE model grows along the forecast horizon — from 25% for 1 quarter to 65% for 12 quarters. while its mean variability to as much as 35 p. a systematic pattern may be perceived. In the case of inﬂation of consumer goods prices (measured with consumption deﬂator). while its advantage grows along the extension of the horizon from about 25% for 1 quarter to 75% for 12 quarters.22 p. its mean variability is nearly 20 p. the advantage of the DSGE model grows along the forecast horizon — from about 17% for forecasts of 2 quarters to 65% for 12 quarters forecasts.p. Additionally. beside of the ﬁrst three periods when the naive model proves to be better by 20–50%.7 p. the advantage of the DSGE model over the naive forecast amounts to nearly 44%.p. As in the case of inﬂation of domestic products prices. The mean advantage of the DSGE model over the naive forecast amounts to over 56%. which is presented in Figures D.8 p.8 p. then it grows again to 1..2 p. Identically as in the case of inﬂation of the prices of domestic products and consumer goods.4 p. underestimated for short horizons of the forecast.15 p. MPE statistics show that inﬂation of investment goods prices is.4–1.p. MPE for each horizon is negative — CPI inﬂation is. Negative MPE statistics for most of the horizons means that inﬂation of consumer goods prices is systematically overestimated — the average MPFE error amounts to −16%. The mean advantage of the DSGE model over the naive forecast amounts to about 45%.p.2 p. and its mean variability — to 0. Errors of CPI inﬂation forecasts are similar as in the case of consumption deﬂator.p. the advantage of the DSGE model over the naive forecast grows systematically along the forecast horizon. systematically overestimated. — for the forecast horizon of 4 quarters.2 p.p.4 p. The bottom left panel of Figure D. Before the 5th quarter the error oscillates around 1.p.p. and its mean variability is above 4.2–D. overestimated within 3rd –8th quarters and underestimated from the 9th to 12th quarter.p. In relative terms the error ﬂuctuates depending on the forecast horizon from −40% for 3 quarters horizon to −5% for 8 quarters. thus.. thus. — for 3 quarters horizon and afterwards it drops to about 0. for 8 quarters forecasts.p.5.p. and its mean variability is nearly 0. starting from the 9th quarter. The DSGE model wins with the naive forecast in each forecast horizon. it grows to 1%. except that the variable is systematically overestimated. — for 5 quarters forecast and drops afterwards to about 1. In relative terms the error ﬂuctuates depending on the forecast horizon from several percent to nearly −60% for the 3 quarters horizon. however.1 shows that the root mean square error of the forecast of inﬂation of investment goods prices reaches the highest value — about 2 p.p.p. afterwards it drops systematically to about 1–1.1 shows that root mean square error of the forecast of CPI inﬂation is the highest — nearly 1.p. while its mean variability is about 11 p. and the average MPE error amounts to −20%. An average RMSE error amounts to 1. the forecast errors are similar. The average RMSE error is 1. The mean advantage of the DSGE model over the naive forecast amounts to over 60%..7 p. The DSGE model wins with the naive forecast in each forecast horizon.1 shows that root mean square error of the forecast of consumer goods inﬂation is the largest — about 1.p..p. except 1 quarter horizon when the naive model proves to be better by slightly over 1%.Analysis of forecasts accuracy MPE statistics for each horizon means that inﬂation of domestic products is systematically underestimated — average MPE error amounts to about 7%. The upper right panel in Figure D. Average RMSE error amounts to slightly over 1 p.p.25 p. The average RMSE error amounts to 1 p. then. In the case of other observable variables. The DSGE model wins with the naive forecast in each forecast horizon.p. The DSGE model wins with the naive forecast in each forecast horizon.p. namely the forecasts are the best in the short horizon (1–2 quarters) and longer horizon (over 6–7 quarters). 152 152 N a t i o n a l B a n k o f P o l a n d . while its mean variability is nearly 0.p. while approximately between the 2nd and the 6th quarter the forecast errors are signiﬁcantly higher than for short and longer horizons.6 p.5 p. The average MPE error amounts to −0.p. The bottom right panel of Figure D. while its mean variability is nearly 0. the RMSE error has a tendency to grow with the forecast horizon. In RMSE errors of the forecasts of inﬂation of all the four types.8 p.
Figure D. MPE — dotted line and right axis.5 15 0 0 −100 −200 −300 −400 15 oWage 2 1. but still remains at the level of above 2 p.p. real wages (oWage). In relative terms. The advantage of the DSGE model grows also with extension of the forecast horizon — from about 10% for 1 quarter forecast to nearly 50% for 12 quarters forecasts. MPE — dotted line and right axis.5 2 1.2.p.5 1 0. 153 WORKING PAPER No. The upper left panel of Figure D. The average MPE error amounts to 21%. The DSGE model wins with the naive forecast in each forecast horizon.85 to 2.5 1 5 10 200 100 0 −100 −200 15 0 −50 10 −100 5 −150 15 oImp −600 0. The mean advantage of the DSGE model over the naive forecast amounts to 40%. employment (oEmp) and real exchange rate of the dollar (oXu). slightly decreasing after the 7th quarter.p. Production is. RMSE and MPE for import dynamics (oImp).p.p. 83 153 .3. investment expenditures (oInv) and export (oExp) 3 2.5 2 1. the error grows from several percent in the ﬁrst quarters to 60–80% in the last quarters.65 p.9 p. The average RMSE error amounts to 1.6 p.Analysis of forecasts accuracy Figure D.. and its mean variability to — 0.5 0 5 oEmp 10 −500 15 0 5 oXu 10 RMSE — continuous line and left axis. while its mean variability is 25 p.5 0 5 10 100 80 60 40 20 0 15 0 −200 −400 10 −600 0 5 10 −800 15 5 0 5 10 5 4 3 2 1 0 5 10 0 −50 −100 −150 −200 15 0 −50 −100 −150 −200 15 oGdp oCons 7 6 5 4 3 15 oInv oExp RMSE — continuous line and left axis.5 1 0. 20 15 10 5 200 0 −200 −400 0 5 10 3 2. thus. RMSE and MPE or the variables of GDP dynamics (oGdp). consumption (oCons). systematically underestimated.2 shows that root mean square error of the forecast of the annual growth rate of the GDP grows systematically between the 1st and 7th quarter from 0.
p. while its mean variability equals 43 p. in the 12th quarter.5 p. In relative terms the error grows systematically from minus several or so to −180%. The advantage of the DSGE model grows also with the extension of the forecast horizon — from nearly 15% for 1 quarter to 60% for 12 quarters.p. and its mean variability — to 0. The DSGE model wins with the naive forecast in each forecast horizon.9 p.p. and its mean variability equals to about 3 p. In relative terms the error reaches −480% in the 2nd quarter and then drops systematically to (−60)–(−80)%. The DSGE model wins with the naive forecast in each forecast horizon. The import growth rate is systematically overestimated. The average RMSE error amounts to 2. The average RMSE error amounts to 12. The average MPE error amounts to −97%.p. export and import.p. systematically overestimated. From the accuracy point of view. except the horizon of 1 and 2 quarters when the naive forecast is on the average by 36% and 6. The growing trend is upset only in 5 the quarters’ horizon.p. The mean advantage of the DSGE model over the naive forecast amounts to 17%.3 p.Analysis of forecasts accuracy The upper right panel of Figure D. the errors of import and export forecasts are similar.p. The average RMSE error amounts to 11. to rise again to about 13 p.p.p. and its mean variability — to 2. The growth rate of consumption is. The advantage of the DSGE model also grows with the forecast horizon. In relative terms the error oscillates between −120% in the 2nd quarter and −170% in the 12th quarter.p. The upper right panel of Figure D. The average MPE error amounts to −136% and its mean variability is 146 p.p. except quarters 5 and 6.p. except the 11th quarter — from 5% for 3 quarters to 35% for 12 quarters. The mean advantage of the DSGE model over the naive forecast amounts to 45%.5 p.p. thus. The DSGE model wins with the naive 154 154 N a t i o n a l B a n k o f P o l a n d .p.p in the 1st quarter to nearly 7 p.2 shows that the root mean square error of the forecast of annual growth rate of investments