Professional Documents
Culture Documents
Albert Rex Bergstrom, Khalid Ben Nowman-A Continuous Time Econometric Model of The United Kingdom With Stochastic Trends (2007)
Albert Rex Bergstrom, Khalid Ben Nowman-A Continuous Time Econometric Model of The United Kingdom With Stochastic Trends (2007)
A Continuous Time
Econometric Model
of the United Kingdom
with Stochastic Trends
Paulo
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao
Cambridge University Press
32 Avenue of the Americas, New York, NY 10013-2473, USA
www.cambridge.org
Information on this title: www.cambridge.org/9780521875493
Estate of Albert Rex Bergstrom and Khalid Ben Nowman 2007
c Cambridge University Press 2007
BN: To Miho
Contents
xi
xiii
xix
page
1.1
1.2
1.3
1
3
1.4
1.5
1.6
1.7
1.8
Introduction
Why Model in Continuous Time
Introduction to General Continuous Time
Models
Continuous Time Models in Finance
Continuous Time Macroeconomic Modelling
Policy Analysis in Continuous Time
Macroeconomic Models
Stochastic Trends in Econometric Models
An Outline of Contents
vii
9
18
31
42
45
47
Contents
Introduction
The Continuous Time Model
The Exact Discrete Model and Its VARMAX
Representation
Estimation and Forecasting
Conclusion
Appendix A: Formulae for the Coefficient Matrices
of Exact Discrete Model
Appendix B: Formulae for the Autocovariance
Matrices
50
50
53
58
67
79
80
85
Model Specification
114
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
3.16
114
116
125
128
130
134
136
138
141
144
145
146
147
147
149
150
Introduction
Equations and General Properties of the Model
Private Consumption
Residential Fixed Capital
Employment
Private Non-Residential Fixed Capital
Output
Price Level
Wage Rate
Interest Rate
Imports
Non-Oil Exports
Transfers Abroad
Real Profits Interest and Dividends from Abroad
Cumulative Net Real Investment Abroad
Exchange Rate
viii
Contents
3.17 Stocks
3.18 Conclusion
Appendix A: Derivation of General Adjustment
Equations
Appendix B: Distributed Lag Relations
151
152
173
4.1
4.2
4.3
4.4
4.5
173
175
180
192
197
197
203
Introduction
The Steady State
Stability Analysis
Stability and Bifurcations
Conclusion
Appendix A: Steady State Level Parameters
Appendix B: Transformed Model
152
164
213
5.1
5.2
5.3
5.4
5.5
5.6
213
214
223
232
240
248
Introduction
Estimation from UK Data
Time Lag Distributions
Steady State and Stability Properties
Post-Sample Forecasting Performance
Conclusion
Appendix A: Linear Approximation
about Sample Means
Appendix B: Data
249
262
269
285
288
References
Author Index
Subject Index
ix
Figures
1.1
1.2
5.1
226
5.2
227
5.3
227
5.4
228
5.5
228
5.6
229
5.7
229
5.8
230
5.9
230
xi
page 22
23
231
231
Tables
1.1
21
5.1
219
5.2
225
5.3
235
5.4
235
5.5
236
5.6
239
242
243
5.7
5.8
xii
Foreword
xiii
Foreword
One unifying force amongst the growing diversity of empirical modelling research in macroeconomics has been
acknowledgement of the importance of long-run behaviour
and the recognition that trending mechanisms in economics
are stochastic. A second area of commonality lies in the use
of nonlinear dynamics, both systems dynamics and volatility
dynamics, the latter being especially important in financial
market applications.
The present volume reports the construction and implementation of a new macroeconomic model of the UK economy that embodies most of these themes. While the model
differs from much of the mainstream of modern macroeconomics in terms of its genesis and form, it shares some
commonality with mainstream work in its detailed attention to nonlinear dynamics, its concern for stochastically
trending data and its use of dynamic optimization principles
in the derivation of adjustment relations. The model developed here is capable of describing diverse patterns of cyclical behaviour in economic aggregates and long-run growth.
It is formulated in continuous time as a system of mixedorder stochastic differential equations. It synthesizes real,
monetary, financial, labour and production sectors of the
economy. It allows for market disequilibria in a systematic way, with parameterized adjustment mechanisms that
xiv
Foreword
measure responsiveness to deviations from partial equilibria whose values are internally consistent and determined
by economic theory. The models long-run properties are
explored analytically, an explicit steady state growth path is
obtained and plausible regions of the parameter space are
determined a priori, thereby providing useful guidelines for
empirical model evaluation. In many of these respects, the
model is very different from prevailing traditions in empirical macroeconomic modelling.
The model is the latest and most sophisticated in a succession of empirical econometric models of the United Kingdom that have been developed under the leadership of the
senior author Rex Bergstrom. The present model therefore
represents the culmination of a long research agenda in
which the cardfile of models has steadily increased in complexity. The incorporation of internally embedded stochastic
trends, the allowance for temporally aggregated flow data
and the introduction of higher order and more complex
lag responses distinguish the present contribution, giving
the new model important elements of realism and features
that are comparable to and in some ways exceed those of
other empirical macro models. Indeed, the reduced form
discrete time version of the present model is a higher order
vector autoregressive (VAR) moving average model with
xv
Foreword
xvi
Foreword
xvii
Foreword
xviii
Preface
Over the last 30 years there has been a growing use of continuous time econometric methods in macroeconomic modelling. Economy-wide continuous time macroeconometric
models have been developed for most of the leading industrial countries of the world. This monograph presents the
first continuous time macroeconometric model of the United
Kingdom incorporating stochastic trends. Its development
represents a major step forward in continuous time macroeconomic modelling. The book describes the new model in
detail and like earlier models it is designed in such a way
as to permit a rigorous mathematical analysis of its steady
state and stability properties, thus providing a valuable check
on the capacity of the model to generate plausible long-run
behaviour. The model is estimated using exact Gaussian estimation methods for continuous time econometric models
xix
Preface
xx
Preface
xxi
CHAPTER ONE
Introduction to Continuous
Time Modelling
1.1 Introduction
1
2
[1950], Phillips [1959], Durbin [1961]) has been well documented in survey papers by Bergstrom [1984a, 1988, 1990,
Ch. 1, 1996]. Our aims in this chapter are to give a brief
introduction to continuous time modelling and econometrics in areas that are important to economists and finance
specialists working in academia, government and financial
markets.
This chapter is organized as follows. Section 1.2 explains
the advantages and problems of continuous time modelling.
Section 1.3 introduces the general idea of a continuous
time model and the estimation of its parameters with available discrete time data. Section 1.4 discusses the application of continuous time models to term structure models in
finance and Section 1.5 introduces continuous time macroeconomic modelling. Section 1.6 discusses policy analysis and
Section 1.7 introduces the idea of stochastic trends in econometrics. Lastly, Section 1.8 outlines the contents of the book.
3
discrete time. There are a number of advantages of modelling in continuous time that we briefly summarize here
(see Bergstrom [1996] for an extensive discussion). First,
since the economy and financial markets are continuously
operating and the underlying decision processes we are trying to model involves millions of decisions by economic
agents within the recorded data observation interval, realistic models will depend on the continuous passage of time.
A continuous system will therefore be sympathetic to the
underlying interactions being modelled and captured in the
data, whereas traditional discrete time models are inherently
less flexible because they restrict the underlying decisionmaking lag structures to match the observation interval in
the data exactly (daily, monthly, quarterly), so that at best
an economys equilibrium and disequilibrium characteristics are being captured at successive points of time. The
potential for misspecification therefore seems greater in a
discrete time system when the underlying phenomenon is
continuous.
Secondly, economic and financial models typically comprise two types of variables. Stock variables observed at
points in time (for example, the money stock, inventories,
fixed capital) and flow variables (for example, consumption,
exports, output and imports) observed as aggregations over
4
5
6
7
8
explosive development of continuous time modelling in economics, and more especially finance over the last 20 years,
together with the enormous increases in computational
capability has made the specification and estimation of continuous time econometric models a much more plausible
practical enterprise.
(1.1)
where x(t) = {x1 (t), . . . , xn (t)} is a n-dimensional continuous time random process, A() is an n n matrix whose elements are functions of a vector = [1 , . . . , p ] of unknown
structural parameters ( p n(n + 1)) and b() is a vector
that is a function of . The error term (dt) is assumed
to be a vector of white noise innovations (see Bergstrom
9
x(t) x(t 1) =
(1.2)
10
(1.3)
The first element of the exogenous vector will typically represent the vector of constants in the system. The approximate simultaneous equations model is now given by
1
A(){x(t) + x(t 1)}
2
1
+ B(){z(t) + z(t 1)} + ut
2
E(ut ) = 0, E ut ut = ,
E z s ut = 0, s, t = 1, 2 . . . .
E us ut = 0, s = t,
x(t) x(t 1) =
(1.4)
11
(1.5)
12
D x(0) = y2 ,
(t 0)
(1.6)
where {x(t), (t > 0)} is a n-dimensional vector of endogenous variables, z(t) is an m-dimensional vector of nonrandom functions (exogenous variables), A1 () and A2 ()
are n n system matrices and B( ) is an n m system matrix
whose elements are known functions of , y1 and y2 are nonrandom n 1 vectors, D is the mean square differential operator and (dt) is a vector of white noise innovations. The first
element of z(t) will normally be unity and so the first column of B will be a vector of constants in the system. Usually
dynamic models include terms for the rate of change, D z(t)
and acceleration, D 2 z(t) of the exogenous variables and this
13
can be done by using the interpolation formulae of Nowman [1991](see also Chambers [1991]). Bergstrom [1986]
provides a precise interpretation of this system.
Associated with these more complex dynamic structures
during the last 20 years, there have been enormous developments in computing technology that have facilitated the
estimation of such models and that use estimation methods, which take account of the exact restrictions on the
distribution of the discrete data implied by the continuous
time model. The most extensively used of these new methods is the exact Gaussian estimation method developed by
Bergstrom [1983, 1985, 1986, 1990] (see also Agbeyegbe
[1984, 1987, 1988], Chambers [1991, 1999], McCrorie
[2000, 2001, 2002], McGarry [2003] and Nowman [1991]).
These methods are applicable to models formulated as systems of higher order and mixed-order stochastic differential equations and under appropriate conditions, yields exact
maximum likelihood estimates of their structural parameters from a sample of mixed stock and flow data.
Typical economic models comprise both stock and flow
variables and we may assume that the stock variables are
observed at specific points in time such as the points 0, 1,
2, . . . , T and the flow variables are observed as integrals over
the unit intervals [0, 1], [1, 2], . . . , [T 1, T ]. We let the
14
x(t) =
y1 =
x s(t)
x f (t)
s
y1
f
y1
,
z(t) =
z s(t)
z f (t)
s
y2
y2 =
f ,
y2
1)}
(t = 1, . . . , T ).
z t = 2 t
f
t1 z (r )dr
(1.7)
(1.8)
15
16
nT
i2 + 2 log mii
(1.10)
i=1
(1.11)
+ 2 (dt)
d[DY(t)] = {3 DY(t) + 4 [DK(t) + C(t) + G(t) Y(t)]} dt
+ 3 (dt),
17
18
19
(1.12)
20
Model
CKLS
Merton
Vasicek
CIRSR
Dothan
GBM
BRSC
CIRVR
CEV
0
0
0
0
0
0
1/
2
1
1
1
1/
2
21
16
UK
US
10
14
8
12
6
10
8
6
2
4
2
0
1992 1994 1996 1998 2000 2002 2004
1992
UK volatility
3.0
2.5
US volatility
2.0
1.6
2.0
1.2
1.5
0.8
1.0
0.4
0.5
0.0
0.0
1992 1994 1996 1998 2000 2002 2004
1992
Japan
Singapore
12
8
10
7
6
5
6
4
3
2
2
1
0
0
1992 1994 1996 1998 2000 2002 2004
1.6
Japan volatility
1992
Singapore volatility
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
1992 1994 1996 1998 2000 2002 2004
1992
(1.13)
24
r (t) = e r (t 1) + (e 1) + t (t = 1, 2, . . . , T )
E(s t ) = 0 (s = t)
E(t2 )
(1.14)
(1.15)
e 2(t ) 2 {r (t 1)} d
t1
2 2
(1.16)
=
(e 1){r (t 1)}2 = m2tt .
2
The Gaussian estimates are then obtained from the Gaussian
r (t) e r (t 1) (e 1)
T
L() =
2 log mtt +
,
2
m
tt
t=1
(1.17)
T
2 log mtt + t2 .
(1.18)
t=1
25
26
27
28
J
X j (t),
(1.19)
j=1
29
(1.20)
30
[2003]
model in the multi-factor cases. Recently, Saltoglu
applied Gaussian estimation to the Brennan and Schwartz
[1979] model. This area of application of continuous time
models continues to be one of the most active research areas
in finance. We now turn to a brief introduction to continuous time macroeconomic modelling.
31
DC
1 Y
= 1 log
(1.21)
C
C
2 e 1 t {Y 4 3 K 4 }1/4
DL
= 2 log
(1.22)
L
L
Dp
Y 1+4
r +
+ 5 k (1.23)
Dk = 3 4 3
K
p
DY = 5 {(1 6 )(C + DK + E) Y}
+ 6 {7 (C + DK + E) S}
32
(1.24)
DI = 7 {6 (C + DK + E) I }
+ 8 {7 (C + DK + E) S}
DE
8 p 9 e 2 t
= 9 log
E
E
Y 4 (1+4 )/4
1 t
1 3 K
we
Dp
10 2
= 10 log
p
p
(1.25)
(1.26)
(1.27)
Dw
2 e 1 t {Y 4 3 K 4 }1/4
= 11 log
(1.28)
w
11 e 3 t
Dr
12 pY 13 r 14
= 12 log
(1.29)
r
M
E
16 e 3 t
Dm = 13 log
+ 14 log
15 I
L
E
16 e 3 t
+ 15 Dlog
+ 16 Dlog
(1.30)
15 I
L
DS = Y + I C DK E
DK
=k
K
DM
=m
M
where
C = real consumption
Y = real net income or output
K = amount of fixed capital
33
(1.31)
(1.32)
(1.33)
E = real exports
I = real expenditure on imports
S = stocks
L = employment
M = volume of money
r = interest rate
p = price level
w = wage rate
k = proportional rate of increase of fixed capital
m = proportional rate of increase in volume of money
t = time
D = d/dt and 0 < 1 < 1, 4 > 1, 0 < 6 < 1, all other
parameters positive (except possibly 3 ).
An important feature of nearly all existing continuous
time macroeconometric models, starting with the model of
Bergstrom and Wymer [1976], is that they are designed in
such a way as to permit a rigorous mathematical analysis
of their steady state and stability properties, thus providing a valuable check on the capacity of the model to generate plausible long-run behaviour. The general methodology used in this analysis was developed by Bergstrom
[1966b, 1967], using smaller models. For the model of Bergstrom and Wymer [1976] and the many continuous time
34
35
36
(1.34)
(1.35)
(1.36)
D 2 logQ = 7 (1 + 2 DlogQ )
{1 9 (qp/pi )10 }(C + Gc + DK + E n + E o )
+ 8 log
Q
(1.37)
D 2 log p = 9 (Dlog(w/p) 1 )
11 4 T2 we 1 t {1 5 (Q /K )6 }(1+6 )/6
+ 10 log
p
(1.38)
37
(1.39)
D 2r = 14 Dr + 15 [13 + rf 14 Dlogq
+ 15 { p(Q + P )/M} r ]
(1.40)
16 Yf 17 ( pf /q p)18
+ 19 log
En
D 2F = 20 DF + 21 {19 (Q + P ) F }
(1.42)
(1.43)
38
(1.45)
26 e3 t
D 2 logM = 26 (3 DlogM) + 27 log
M
En + E o + P F
+ 28 Dlog
( pi /qp)I
E n + E o + P F DKa
+ 29 log
( pi /qp)I
27 pf
D 2 logq = 30 Dlog( pf /qp) + 31 log
qp
En + E o + P F
+ 32 Dlog
( pi /qp)I
E n + E o + P F DKa
+ 33 log
( pi /qp)I
(1.46)
(1.47)
Endogenous variables:
C = real private consumption
E n = real non-oil exports
F = real current transfers abroad
I = volume of imports
K = amount of fixed capital
K a = cumulative net real investment abroad (excluding
change in official reserves)
L = employment
M = money supply
P = real profits, interest and dividends from abroad
39
p = price level
Q = real net output
q = exchange rate (price of sterling in foreign currency)
r = interest rate
w = wage rate
Exogenous Variables:
dx = dummy variable for exchange controls (dx = 1 for
197479, dx = 0 for 1980 onwards)
E o = real oil exports
Gc = real government consumption
pf = price level in leading foreign industrial countries
pi = price of imports (in foreign currency)
rf = foreign interest rate
T1 = total taxation policy variable ((Q + P )/T1 is real private
disposable income)
T2 = indirect taxation policy variable (Q /T2 is real output at
factor cost)
t = time
Yf = real income of leading foreign industrial countries.
The model was estimated using the exact Gaussian estimator for the general case of Bergstrom [1986] on UK
quarterly data over the period 197484 with the predictive
40
performance tested over 198586. As judged by the plausibility of the parameter estimates, the post-sample forecasting
performance and the analysis of its steady state properties,
this model was very successful.
One of the problems in estimating the parameters of continuous time models concerns the issue of the identification
of the structural parameters. We have the aliasing problem
where one cannot distinguish between structures generating cycles whose frequencies (per unit observation period)
differ by integers. For example, the case of half a cycle per
observation period and one and a half cycles per observation period. This phenomenon was originally discussed in
a study by Phillips [1973], where it was shown how identification could be achieved through Cowles Commission
type restrictions on the structural parameters. Hansen and
Sargent [1983] also investigated the aliasing problem. They
showed that there at most a finite number of structures that
are indistinguishable (on the basis of the observations) from
the true structure because of the aliasing phenomenon. One
of the implications of their theorem is that the true structure
will be identifiable provided that the unit observation period
is sufficiently short. In Bergstrom, Nowman and Wymer
[1992] they used an alternative approach to identification,
suggested by the last result of Hansen and Sargent [1983],
41
42
models steady state computed from the estimated parameters was slightly unstable (although, statistically they could
not reject the hypothesis that it is stable), they investigated
the use of simple fiscal policy feedbacks and the use of
control theory considered in Bergstrom [1987] that could
reduce or eliminate the instability. Optimal feedbacks were
derived by minimizing an infinite horizon quadratic cost
function involving deviations of output, the exchange rate
and the policy instruments from their steady state paths and
also the rates of change of these variables. One of the findings was that fiscal policy had an important role to play in
the stabilization of the model than did monetary policy. A
discussion of the important contributions of Barnett and He
[1999, 2002] concerning bifurcation analysis in continuous
time macroeconometric models is given in Chapter 4.
Another important issue in macroeconomic modelling in
discrete and continuous time concerns the famous Lucas
[1976] critique and also the related issues of deep parameters, structural parameters and rational expectations in
macroeconomic modelling. Deep parameters are taken to
be the parameters of Euler equations and are directly
acquired from the parameters of tastes and technology. The
role of rational expectations in macroeconomic modelling
had an important influence on the Bergstrom, Nowman
43
44
45
46
stochastic differential equations, which incorporate unobservable stochastic trends, while the data is a mixture of
stocks (observed at a sequence of equispaced points of time)
and flows (observed as integrals over a sequence of adjoining
intervals of equal length). The algorithm is applicable, therefore, to models which, in addition to incorporating unobservable stochastic trends, involve all the complications in
existing continuous time macroeconometric models.
Such a model has been developed for the UK economy
by the authors of this monograph. Its parameters have been
estimated by the algorithm of Bergstrom [1997], the estimation being the first application of the new algorithm.
The model is the first continuous time macroeconometric
model incorporating stochastic trends, and its development
represents a major step forward in macroeconometric modelling. The purpose of this monograph is to describe the new
model, the estimation of its parameters and its application
to dynamic analysis and forecasting.
47
48
49
CHAPTER TWO
Continuous Time
Econometrics with
Stochastic Trends
2.1 Introduction
his chapter is concerned with the general econometric methodology used throughout the book. The main
objective is to exposit the newly developed exact Gaussian estimation method for continuous time econometric
models that incorporate unobservable stochastic trends. The
approach is based on the article of Bergstrom [1997], and
readers are referred to this article for proofs of the various
results used. Bergstrom [1997] extends the exact Gaussian
estimation method developed in earlier work (Bergstrom
[1983, 1985, 1986, 1990]) in two ways. Firstly, the algorithm is applicable to models incorporating unobservable
stochastic trends. These can provide a more flexible and
50
51
52
(t 0), (2.1)
(t 0),
d(t) = dt + 3 (dt)
x 1 (0), x2 (0), D x2 (0) = y (0),
(2.2)
(t 0),
(0) = 0,
(2.3)
(2.4)
53
54
(2.5)
D x 2 (t) D x 2 (0) =
[A4 x1 (r ) + A5 x 2 (r ) + A6 D x2 (r )
+ B2 z(r ) + C 2 (r ) + b2 ]dr
t
2 (dr ) (t > 0),
+
(2.6)
where
t
0
55
(2.7)
56
model of Harvey and Stock [1988], who are among the few
other econometricians to have dealt with the problem of
estimating structural parameters in continuous time models with unobservable stochastic trends. Harvey and Stock
[1988] assumed that each of the observable (at discrete
intervals) variables is the sum of two unobservable components, the first component being one of a set of completely
unobservable variables generated by an underlying stochastic differential equation system and the second being a linear combination of stochastic trends. In the above model the
stochastic trends are more deeply embedded in the structure
and help to derive the differential equation system, just as,
for example, technical progress helps to derive the economy.
The essential distinction between the two approaches is that
in the above model the stochastic trends play the role of random forcing functions, which help to generate the solution
of the differential equation system, whereas in the model of
Harvey and Stock [1988] the stochastic trends have no effect
on the solution of that system, but are added to the solution
(like observation errors) to obtain the observed variables.
Finally, it should be noted that is introduced as an argument in the vector functions b1 (, ) and b2 (, ) in (2.1) and
(2.2). This allows for the incorporation in the model of the
assumption of long-run expectations, that is, the assumption
57
that the various agents in the economy know the drifts and
take them into account when adjusting the variables that
they control.
58
was more realistic for these variables. There were also some
flow variables, particularly wage and price variables (which
were measured as time averages and had, therefore, to be
treated as flow variables), for which a second-order adjustment equation appeared to be more realistic.
We assume that the stock variables are observed at the
time points 0, 1, 2, . . . , T and the flow variables as integrals
over the intervals [0, 1], [1, 2], . . . , [T 1, T ]. We let the
elements of x2 (t) and z(t) be ordered (without loss of generality) so that
x2 (t) =
x2s (t)
f
x2 (t)
,
z(t) =
z s (t)
z f (t)
,
f
(2.8)
t1
x 2t
= t f
x 2 (r )dr
t1
59
(t = 1, . . . , T ),
(2.9)
x t =
1
2
zt =
x 1t
x 2t
(t = 1, . . . , T ),
{z s (t) + z s (t 1)}
t
z f (r )dr
(2.10)
(t = 1, . . . , T ).
(2.11)
t1
(2.12)
(2.13)
(2.14)
x t = F1
x t1 + F2
x t2 + F0 + E 0
zt + E 1
zt1
+ E 2
zt2 + t
where
(t = 4, . . . , T ),
(2.15)
60
A1
A = 0
A4
A2
0
A5
61
A3
I
A6
62
(2.16)
(2.17)
(2.18)
x t F1
x t1 F2
x t2 F0 E 0
zt E 1
zt1 E 2
zt2
= Mtt t + Mt,t1 t1 + Mt,t2 t2 + Mt,t3 t3
(t = 4, . . . , T ),
(2.19)
(2.20)
(2.21)
(2.22)
and the coefficient matrices M11 , M21 , M22 , M31 , M32 , M33 ,
Mtt , Mt,t1 , Mt,t2 and Mt,t3 (t = 4, . . . , T ) are obtained,
recursively, from (2.23)(2.44).
M11 M11
= 11 ,
63
(2.23)
1
M21 = 21 M11
,
(2.24)
M22 M22
= 22 M21 M21
,
(2.25)
1
M31 = 31 M11
,
(2.26)
1
M32 = 32 M31 M21
,
M22
(2.27)
M33 M33
= 33 M31 M31
M32 M32
,
(2.28)
1
M41 = 41 M11
,
(2.29)
1
M22
M42 = 42 M41 M21
,
(2.30)
1
M42 M32
,
M33
M43 = 43 M41 M31
(2.31)
M44 M44
= 0 M41 M41
M42 M42
M43 M43
,
(2.32)
1
M52 = 52 M22
,
(2.33)
1
M53 = 53 M52 M32
,
M33
(2.34)
1
M54 = 1 M52 M42
M53 M43
,
M44
(2.35)
64
M55 M55
= 0 M52 M52
M53 M53
M54 M54
,
(2.36)
1
,
M63 = 63 M33
(2.37)
1
,
M64 = 2 M63 M43
M44
(2.38)
1
M64 M54
,
M65 = 1 M63 M53
M55
(2.39)
= 0 M63 M63
M64 M64
M65 M65
,
M66 M66
(2.40)
and, for t = 7, . . . , T,
1
Mt,t3 = 3 Mt3,t3
,
(2.41)
1
,
Mt,t2 = 2 Mt,t3 Mt2,t3
Mt2,t2
(2.42)
1
Mt,t2 Mt1,t2
,
Mt,t1 = 1 Mt,t3 Mt1,t3
Mt1,t1
(2.43)
Mtt Mtt = 0 Mt,t3 Mt,t3
Mt,t2 Mt,t2
Mt,t1 Mt,t1
,
(2.44)
65
66
11
= 0
21
0
0
0
12
0 .
22
(2.45)
This matrix and the matrix 33 both occur in formulae for
the i and i j matrices in Appendix B. We shall parametrize
and 33 by writing them as () and 33 (), where
and are vectors of structural parameters. If there are no
restrictions on these matrices, apart from those implied by
Assumption 1, then will have (n1 + n2 )(n1 + n2 + 1)/2 elements and will have k(k + 1)/2 elements.
In addition to the vector [, , , ] of genuine parameters, most of the initial state vector y(0) is unobservable
67
and must be treated as part of the complete vector of parameters to be estimated. The observable part of y(0) is the ns2 1
vector x 2s (0), that is the initial levels of the stock variables.
f
x 1 (0)
yn =
x2 (0) .
D x 2 (0)
(2.46)
68
denotes minus twice the logarithm of the likelihood function (less a constant), then
T
L=
t t + 2log|Mtt | ,
(2.47)
t=1
where 1 , 2 , . . . , T are regarded as functions of the observations and the parameters and can be computed, recursively,
from (2.16)(2.19), that is, from the recursive equations
1
1 = M11
(x 1 G10 G11 y(0) E 11 z1 E 12 z2 E 13 z3 ),
(2.48)
1
(x 2 F21 x 1 G20 G21 y(0)
2 = M22
E 21 z1 E 22 z2 E 23 z3 M21 1 ),
(2.49)
1
3 = M33
(x 3 F32 x 2 F31 x 1 G30 G31 y(0)
E 31 z1 E 32 z2 E 33 z3 M31 1 M32 2 ),
(2.50)
t = Mtt1 (
x t F1
x t1 F2
x t2 F0 E 0
zt E 1
zt1
E 2
zt2 Mt,t1 t1 Mt,t2 t2 Mt,t3 t3 )
(t = 4, . . . , T ).
(2.51)
69
70
71
true values of the parameters (each deviation being multiplied by T 1/2 ) are, in general, asymptotically normally distributed, the deviations of the above estimates from the
true values of the parameters (each deviation being multiplied by an appropriate power of T ) are asymptotically
distributed like very complicated expressions involving integrals of quadratic forms of Brownian motions and integrals
of Brownian motions with respect to Brownian motions.
This is as we should expect in view of the extensive work
of P. C. B. Phillips and his collaborators on the estimation of models with integrated processes (see Phillips and
Durlauf [1986], Phillips [1987, 1991a, 1991b, 1995], Park
and Phillips [1988, 1989]). The computational cost of using
such expressions in assessing the accuracy of the parameter estimates would, however, be very heavy except in very
small models.
In assessing the accuracy of the parameter estimates for
our UK model presented in Chapter 5, we will follow a
simpler Bayesian approach. In this connection, it should
be noted, first, that, if the innovations in the continuous
time model (2.1)(2.4) are Brownian motion and the exogenous variables are quadratic functions of t, then assuming a uniform prior distribution, L(, , , , yn ) is minus
twice the logarithm of the Bayesian posterior probability
72
73
74
(2.52)
(2.54)
x t = F1
x t1 + F2
x t2 + F0 + E 0
zt + E 1
zt1
+ E 2
zt2 + t
(t = 4, . . . , T + R),
(2.55)
1 = M11 1 ,
(2.56)
2 = M21 1 + M22 2 ,
(2.57)
(2.58)
(2.59)
,
,
From this system, together with the estimator [,
75
(r = 1, . . . , R).
(2.60)
(2.62)
T +3 = M T +3,T T ,
(2.63)
T +r = 0
(r = 4, . . . , R),
(2.64)
76
1
3 M t3,t3
,
M t,t3 =
(2.65)
1
2 M t,t3 M t2,t3
M t,t2 =
M t2,t2
,
(2.66)
1
1 M t,t3 M t1,t3
M t1,t1
M t,t2 M t1,t2
,
M t,t1 =
(2.67)
0 M t,t3 M t,t3
M tt M tt =
M t,t2 M t,t2
M t,t1 M t,t1
(2.68)
(t = T + 1, . . . , T + 3).
77
x T +3 = F 1
x T +2 + F 2
x T +1 + F 0 + E 0
z T +3
+ E 1
z T +2 E 2
z T +1 + T +3 ,
(2.71)
x T +r = F 1
x T +r 1 + F 2
x T +r 2 + F 0 + E 0
z T +r
+ E 1
z T +r 1 + E 2
z T +r 2 (r = 4, . . . , R), (2.72)
where F 0 , F 1 , F 2 , E 0 , E 1 and E 2 are obtained by substitut of the parameter vector [, ] into the
ing the estimate [ , ]
formulae for F0 , F1 , F2 , E 0 , E 1 and E 2 given in Appendix A.
Finally, we obtain the forecasts of the vectors x T +1 ,
x T +2 , . . . , x T +R from the accumulation formula
x T + r = x T +
r
x T +s
(r = 1, . . . , R).
(2.73)
s =1
The forecasts given by (2.73) are optimal post-sample forecasts of the discrete observations of the endogenous variables
in the sense mentioned above.
Summarizing the forecasting procedure, the main computational steps are as follows.
1. Compute the matrices M T +1,T , M T +1,T 1 , M T +1,T 2 ,
M T +2,T , M T +2,T 1 and M T +3,T from (2.65)(2.68).
2. Compute the vectors T +1 , T +2 and T +3 from (2.61)
(2.64).
78
2.5 Conclusion
In this chapter, we have formulated an open mixed-order
continuous time dynamic model with mixed stock and flow
variables and unobservable stochastic trends. We have then
described a form of VARMAX model which, under certain
assumptions, is satisfied, exactly, by the discrete observations
generated by the continuous time model, and we have
79
80
0 0 0],
0 I 0
,
I 0 0
0 0 0
,
0 0 I
(2.74)
S4 = [I 0],
S5 = [0 I ],
S1
.
S=
S2
81
82
G20 = S A1 (e A I ) + e A S3 S3 [A2 (e A I ) A1 ]b
+ S A2 (e A I ) + 0.5A1 (e A I ) A1
+ e A S3 S3 [A3 (e A I ) A2 0.5A1 ] C ,
G21 = Se A S3 S3 A1 (e A I ),
G30 = S I + e A S3 S3 A1 (e A I )
+ e A S3 S3 e A S3 S3 [A2 (e A I ) A1 ] b
+ S (A2 + 1.5A1 )(e A I ) A1
+ e A S3 S3 e A S3 S3 [A3 (e A I ) A2 0.5A1 ] C ,
83
where
11
3
S
4
= A1 B 12 + A2 B + A3 B,
2
S5
1
S
4
= A1 B 6 2A2 B 2A3 B,
0
1
1
S
4
= A1 B 12 + A2 B + A3 B,
2
0
L 00
L 01
L 02
1
1
S
4
= A1 B
12 + A2 B + A3 B,
2
0
7
S
4
= A1 B 6 2A3 B,
S5
1
1
S
4
= A1 B 12 A2 B + A3 B,
2
0
L 10
L 11
L 12
L 20
L 21
1
1
S
4
= A1 B
12 A2 B + A3 B,
2
0
1
S
4
= A1 B 6 + 2A2 B 2A3 B,
0
84
11
3
S
4
L 22 = A1 B 12 A2 B + A3 B,
2
S5
7
S4
4
P1 = A1 B
2A2 B + A3 B,
11
S5
6
S4
P2 = A1 B 7 + 3A2 B 2A3 B,
S5
6
1
S4
4
P3 = A1 B
A2 B + A3 B.
1
S5
3
85
t =
K 1 (t r ) (dr ) +
t1
+
t1
K 2 (t 1 r ) (dr )
t2
t2
t3
+
N1 (t r )3 (dr ) +
t1
N2 (t 1 r )3 (dr )
t2
t2
1 =
1
K 11 (1 r ) (dr ) +
N11 (1 r )3 (dr ),
K 21 (1 r ) (dr ) +
K 22 (2 r ) (dr )
K 31 (1 r ) (dr ) +
N21 (1 r )3 (dr ) +
(t = 4, . . . , T )
3 =
N4 (t 3 r )3 (dr )
t4
t3
N3 (t 2 r )3 (dr ) +
t3
2 =
K 4 (t 3 r ) (dr )
t4
t1
t3
K 3 (t 2 r ) (dr ) +
N22 (2 r )3 (dr ),
1
2
K 32 (2 r ) (dr )
+
K 33 (3 r ) (dr ) +
N31 (1 r )3 (dr )
0
2
N32 (2 r )3 (dr ) +
86
N33 (3 r )3 (dr ).
0 =
0
K 1 (r ) K 1 (r )dr +
+
0
+
0
1
+
0
1
0
K 3 (r ) K 3 (r )dr +
N1 (r )33 N1 (r )dr
N3 (r )33 N3 (r )dr
K 2 (r ) K 2 (r )dr
K 4 (r ) K 4 (r )dr
1
+
87
where
1
0
K 1 (r ) K 1 (r )dr
1
0
K 2 (r ) K 2 (r )dr
1
0
K 3 (r ) K 3 (r )dr
0
1
K 4 (r ) K 4 (r )dr
88
e rA e rA dr {WA1 } ,
1
0
0
1
{SA1 }C 33 C {SA2
r e rA dr }
0
1
+ {SA2 } e rA C 33 C e rA dr {SA2 } ,
1
89
1
+ {SA2 (I + e A) SA1 WA2 }
2
C 33 C {SA1 (I + e A) WA1 }
1
+ {SA1 (I + e A) WA1 }
3
C 33 C {SA1 (I + e A) WA1 }
+ {(W 2S)A3 (e A I )}
C 33 C {SA2 (I + e A) SA1 WA2 }
+ {SA2 (I + e A) SA1 WA2 }
+ C 33 C {(W 2S)A3 (e A I )}
1
+ (W 2S)A2
r e rA dr
0
rA
r e dr
+ {(W 2S)A }
0
1
0
= {S(A1 A2 )e A + WA2 (I + e A)
WA1 }C 33 C {S(A1 A2 )e A
+ WA2 (I + e A) WA1 }
1
+ {WA1 (I + e A) SA1 e A}
2
90
0
1
1
0
91
1
{W(A1 A2 )e A}C 33 C {WA1 e A}
2
1
+ {WA1 e A}C 33 C {WA1 e A}
3
+ {WA3 (e A I )}C 33 C {W(A1 A2 )e A}
+ {W(A1 A2 )e A}C 33 C {WA3 (e A I )}
1
WA2
r e rA dr C 33 C {WA1 e A}
0
1 A
{WA e }C 33 C
+ {WA2 }
1
rA
WA
r e dr
0
e rA C 33 C e rA dr {WA2 } .
1 =
K 2 (r ) K 1 (r )dr +
+
+
0
K 4 (r ) K 3 (r )dr +
N3 (r )33 N2 (r )dr
K 3 (r ) K 2 (r )dr
+
0
where
1
K 2 (r ) K 1 (r )dr = {SA1 (I + e A) WA1 }{SA1 }
0
92
K 3 (r ) K 2 (r )dr
0
K 4 (r ) K 3 (r )dr
e rA e rA dr {(S 2W)A1 } ,
0
0
93
(W 2S)A2
r e rA dr C 33 C {SA1 }
0
1
+ {SA (I + e ) WA }C 33 C
A
+ {(W 2S)A2 }
SA
1
rA
r e dr
0
r e rA C 33 C r e rA dr {SA2 } ,
1
0
94
r e rA
0
{WA e }C 33 C
+ {WA2 }
(S 2W)A
1
rA
r e dr
0
95
K 4 (r ) K 2 (r )dr
0
0
1
1
+
N3 (r )33 N1 (r )dr +
N4 (r )33 N2 (r )dr,
2 =
K 3 (r ) K 1 (r )dr
where
1
0
K 3 (r ) K 1 (r )dr
0
K 4 (r ) K 2 (r )dr
1
0
96
1
{S(A1 A2 )e A
2
+ WA2 (I + e A) WA1 }C 33 C {SA1 }
1
{WA1 (I + e A) SA1 e A}C 33 C {SA1 }
3
{(S 2W)A3 (e A I )}C 33 C {SA2 }
C 33 C {SA2 }
1
0
1 A
r e rA C 33 C e rA dr {SA2 } ,
1
0
97
1 A
{WA e }C 33 C
(W 2S)A
1
rA
r e dr
0
+ {WA2 }
3 =
0
where
K 4 (r ) K 1 (r )dr +
1
0
K 4 (r ) K 1 (r )dr
0
0
1
98
WA2
r e rA dr C 33 C {SA1 }
0
1 A
{WA e }C 33 C
2
SA
1
rA
r e dr
0
+ {WA }
e rA C 33 C e rA dr {SA2 } .
11 =
0
K 1 (r ) K 1 (r )dr +
0
where
1
1
0 K 1 (r ) K 1 (r )dr and 0 N1 (r )33 N1 (r )dr are evaluated as
in the formulae for 0 .
1
1
(r )dr + 0 N21 (r )33 N11
(r )dr
21 = 0 K 21 (r ) K 11
where
1
0
K 21 (r ) K 11
(r )dr
= SA1 e A Se A S3 S3 A1 {SA1 }
1
0
N21 (r )33 N11
(r )dr
= SA2 e A SA1 Se A S3 S3 A2 C 33 C {SA2 }
99
1 1 A
SA e Se A S3 S3 A1 C 33 C {SA2 }
2
1 2 A
SA e SA1 Se A S3 S3 A2 C 33 C {SA1 }
2
1 1 A
+ SA1 e A Se A S3 S3 A1 C 33 C SA2
Se A S3 S3 S A2
22 =
r e rA dr
e rA C 33 C e rA dr {SA2 } .
K 22 (r ) K 22
(r )dr
1
1
+
N21 (r )33 N21
(r )dr +
N22 (r )33 N22
(r )dr,
0
K 21 (r ) K 21
(r )dr
where
0
K 21 (r ) K 21
(r )dr
= SA1 e A Se A S3 S3 A1 SA1 e A Se A S3 S3 A1
+ Se A S3 S3 S A2 (e A I ) SA1 e A Se A S3 S3 A1
100
+ SA1 e A Se A S3 S3 A1 Se A S3 S3 S A2 (e A I )
1 rA rA A
+ Se A S3 S3 S A1
e e dr Se S3 S3 S A1 ,
0
1
0
K 22 (r ) K 22
(r )dr =
0
K 1 (r ) K 1 (r )dr ,
101
Se A S3 S3 S A2
r e rA dr
0
C 33 C SA1 e A Se A S3 S3 A1
+ SA1 e A Se A S3 S3 A1
1
r e rA dr
C 33 C Se A S3 S3 S A2
0
1 rA
+ Se A S3 S3 S A2
e
0
C 33 C e rA dr Se A S3 S3 S A2 ,
1
0
N22 (r )33 N22
(r )dr =
K 31 (r ) K 11
(r )dr +
1
0
N31 (r )33 N11
(r )dr ,
where
1
0
K 31 (r ) K 11
(r )dr
A 1 A
= Se S3 S3 A e e A S3 S3 A1 {SA1 }
Se A S3 S3 e A S3 S3 I A2 (e A I ) {SA1 }
+ Se A S3 S3 A1 e A e A S3 S3 A1 {SA2 (e A I )}
1 rA rA
+ Se A S3 S3 e A S3 S3 I A1
e e dr {SA1 } ,
0
102
0
N31 (r )33 N11
(r )dr
= SA1 (e A I ) + Se A S3 S3 A2 e A A1 e A S3 S3 A2
1 A 1 A
Se S3 S3 A e e A S3 S3 A1
2
1 1 A
C 33 C {SA2 }
SA (e I )
2
+ Se A S3 S3 A2 e A A1 e A S3 S3 A2 C 33 C {SA1 }
C 33 C {SA2 }
1 A 1 A
Se S3 S3 A e e A S3 S3 A1 C 33 C {SA1 }
3
Se A S3 S3 e A S3 S3 I A3 (e A I ) C 33 C {SA2 }
+ SA1 (e A I ) + Se A S3 S3 A2 e A A1 e A S3 S3 A2
C 33 C {SA3 (e A I )} Se A S3 S3 (e A S3 S3 I )A2
r e rA dr C 33 C {SA1} + Se A S3 S3 A1 e A e A S3 S3 A1
1
2
rA
C 33 C SA
r e dr + Se A S3 S3 e A S3 S3 I A2
e rA C 33 C e rA dr {SA2 } .
32 =
0
K 31 (r ) K 21
(r )dr +
+
0
N31 (r )33 N21
(r )dr +
103
K 32 (r ) K 22
(r )dr
0
N32 (r )33 N22
(r )dr,
where
1
0
K 31 (r ) K 21
(r )dr
= Se A S3 S3 A1 e A e A S3 S3 A1 SA1 e A Se A S3 S3 A1
+ Se AS3 S3 e AS3 S3 I A2(e AI ) SA1e ASe A S3 S3 A1
+ Se A S3 S3 A1e Ae A S3 S3 A1 Se A S3 S3S A2 (e AI )
1
+ Se A S3 S3 e AS3 S3 I A1 e rA e rA dr Se AS3 S3S A1 ,
0
K 32 (r )33 K 22
(r )dr =
K 21 (r ) K 11
(r )dr ,
104
C 33 C SA2 e A SA1 Se A S3 S3 A2
+ SA1 (e A I ) + Se A S3 S3 A2 e A A1 e A S3 S3 A2
C 33 C Se A S3 S3 S A3 (e A I )
1
A
2
A
rA
+ Se S3 S3 e S3 S3 I A
r e dr
0
C 33 C SA1 e A Se A S3 S3 A1
+ Se A S3 S3 A1 e A e A S3 S3 A1
A
2 1 rA
Se S3 S3 S A
r e dr
C 33 C
0
1 rA
+ Se A S3 S3 e A S3 S3 I A2
e
0
C 33 C e rA dr Se A S3 S3 S A2 ,
1
1
N32 (r )33 N22 (r )dr =
N21 (r )33 N11
(r )dr ,
where
1
0
K 31 (r ) K 31
(r )dr
A 1 A
= Se S3 S3 A e e A S3 S3 A1
105
Se A S3 S3 A1 e A e A S3 S3 A1
+ Se A S3 S3 e A S3 S3 I A2 (e A I )
Se A S3 S3 A1 e A e A S3 S3 A1
+ Se A S3 S3 A1 e A e A S3 S3 A1
Se A S3 S3 e A S3 S3 I A2 (e A I )
1 rA
+ Se A S3 S3 e A S3 S3 I A1
e
0
e rA dr Se A S3 S3 e A S3 S3 I A1 ,
1
0
K 32 (r ) K 32
(r )dr =
1
0
K 21 (r ) K 21
(r )dr ,
106
1 1 A
SA (e I )+ Se A S3 S3 A2 e A A1 e A S3 S3 A2
2
C 33 C Se A S3 S3 A1 e A e A S3 S3 A1
1 A 1 A
+
Se S3 S3 A e e A S3 S3 A1
3
C 33 C Se A S3 S3 A1 e A e A S3 S3 A1
+ Se A S3 S3 e A S3 S3 I A3 (e A I )
C 33 C SA1 (e A I )+ Se AS3 S3 A2 e AA1e A S3 S3 A2
+ SA1 (e A I ) + Se A S3 S3 A2 e A A1 e A S3 S3 A2
C 33 C Se A S3 S3 e A S3 S3 I A3 (e A I )
1
+ Se A S3 S3 e A S3 S3 I A2
r e rA dr
+
C 33 C Se A S3 S3 A1 e A e A S3 S3 A1
+ Se A S3 S3 A1 e A e A S3 S3 A1
1
r e rA dr
C 33 C Se A S3 S3 e A S3 S3 I A2
0
A A
2
+ Se S3 S3 e S3 S3 I A
1
1
0
N32 (r )33 N32
(r )dr =
N21 (r )33 N21
(r )dr ,
107
K 4 (r ) K 21
(r )dr +
K 3 (r ) K 22
(r )dr
0
0
1
1
+
N4 (r )33 N21
(r )dr +
N3 (r )33 N22
(r )dr,
0
where
1
0
K 4 (r ) K 21
(r )dr
= {WA1 e A} SA1 e A Se A S3 S3 A1
+ {WA2 (e A I )} SA1 e A Se A S3 S3 A1
{WA1 e A} Se A S3 S3 S A2 (e A I )
1
1
+ {WA }
e rA e rA dr Se A S3 S3 S A1 ,
0
0
K 3 (r ) K 22
(r )dr =
1
0
K 3 (r ) K 1 (r )dr ,
108
1
{WA1 e A}C 33 C SA1 e A Se A S3 S3 A1
3
+ {WA3 (e A I )}C 33 C SA2 e A SA1 Se A S3 S3 A2
+ {W(A1 A2 )e A}C 33 C Se A S3 S3 S A3 (e A I )
1
+ WA2
r e rA dr C 33 C SA1 e A Se A S3 S3 A1
0
1 A
{WA e }C 33 C
+ {WA2 }
Se
S3 S3
e rA C 33 C e rA dr
N3 (r )33 N22
(r )dr
1
rA
r e dr
0
S A
Se A S3 S3 S A2 ,
N3 (r )33 N1 (r )dr ,
where
K 4 (r ) K 31
(r )dr
0
= {WA1 e A} Se A S3 S3 A1 e A e A S3 S3 A1
+ {WA2 (e A I )} Se A S3 S3 A1 e A e A S3 S3 A1
109
{WA1 e A} Se A S3 S3 e A S3 S3 A1 I A2 (e A I )
1
1
+ {WA }
e rA e rA dr Se A S3 S3 e A S3 S3 A1 I A1 ,
0
K 3 (r ) K 32
(r )dr
= {WA1 (I + e A) SA1 e A} SA1 e A Se A S3 S3 A1
+ {(S 2W)A2 (e A I )} SA1 e A Se A S3 S3 A1
+ {WA1 (I + e A)SA1 e A} Se A S3 S3 S A2 (e A I )
1
+ {(S 2W)A1 }
e rA e rA dr Se A S3 S3 S A1 ,
0
K 2 (r ) K 33
(r )dr =
1
0
K 2 (r ) K 1 (r )dr ,
110
+ {W(A1 A2)e A}C 33 C Se A S3 S3 e AS3 S3 I A3 (e A I )
1
2
rA
+ WA
r e dr C 33 C Se A S3 S3 A1 e Ae A S3 S3 A1
0
1
{WA1 e A}C 33 C Se A S3 S3 e A S3 S3 I A2 r e rA dr
+ {WA2 }
0
0
e rA C 33 C e rA dr Se A S3 S3 e A S3 S3 I A2
N3 (r )33 N32
(r )dr
111
1 A
+ {WA (I + e ) SA e }C 33 C
rA
r e dr
Se A S3 S3 S A2
+ {(S 2W)A2 }
r e rA C 33 C e rA dr
Se A S3 S3 S A2 ,
N2 (r )33 N33
(r )dr =
K 3 (r ) K 33
(r )dr
1
1
+
N4 (r )33 N32
(r )dr +
N3 (r )33 N33
(r )dr,
53 =
K 4 (r ) K 32
(r )dr
where
K 4 (r ) K 32
(r )dr =
1
0
K 4 (r ) K 21
(r )dr ,
112
The integrals
1
0
e rA e rA dr and
1
0
e rA C 33 C e rA dr in the
113
CHAPTER THREE
Model Specification
3.1 Introduction
In this chapter we discuss the new mixed-order continuous time macroeconometric model of the UK economy with
stochastic trends. The model is based on the Bergstrom,
Nowman and Wymer [1992] second-order continuous time
macroeconometric model. The model is the first continuous time macroeconometric model incorporating stochastic
trends, and its development represents a major step forward
in general continuous time macroeconometric modelling
over the last 30 years. The detailed specification of the structural equations and economic theory of previous continuous
time macroeconomic models of various economies of the
world is presented, for example, in Bergstrom and Wymer
[1976], Knight and Wymer [1978], Bergstrom, Nowman
114
Model Specification
115
abroad, cumulative net real investment abroad and the exchange rate. The stock identity is presented in Section 3.17
and conclusions in Section 3.18. This chapter, also, contains
two technical Appendices presenting the formal derivation
of the adjustment equations and the distributed lag relations.
Endogenous Variables
C = real private consumption
E n = real non-oil exports
F = real current transfers abroad
I = volume of imports
K h = residential fixed capital
K = private non-residential fixed capital
116
Model Specification
Exogenous Variables
B = stock of bonds
dx = dummy variable for exchange controls (dx = 1 for
197479, dx = 0 for 1980 onwards)
E o = real oil exports
Gc = real government consumption
K p = public non-residential fixed capital
pf = price level in leading foreign industrial countries
pi = price of imports (in foreign currency)
M = money supply
rf = foreign interest rate
T1 = total taxation policy variable ((Q + P )/T1 is real private disposable income)
117
Structural Equations
1 e {2 (r Dlog p)+3 Dlog p} (Q + P )
DlogC = 1 + 2 + 1 log
T1 C
(3.1)
D 2 logL = 2 (2 DlogL)
4 e 1 {Q 6 5 K 6 }16
+ 3 log
L
118
(3.2)
Model Specification
D 2 logK h = 4 (1 + 2 DlogK h )
7 e {8 (r Dlog p)+9 Dlog p} (Q + P )
+ 5 log
T1 K h
(3.3)
D 2 logK = 6 (1 + 2 DlogK )
5 (Q /K )(1+6 )
+ 7 log
r 10 Dlog p + 11
(3.4)
DlogQ
= 1 + 2 + 8 log
{112 (qp/pi )13 }{1+14 (1 +2 )}
Q
+ 9 log
14 (C + Gc + D K + DKh + DK p + E n + E o )
S
(3.5)
D 2 log p = 10 (Dlog(w/p) 1 )
15 4 T2 we 1 {1 5 (Q /K )6 }(1+6 )/6
+ 11 log
p
(3.6)
D 2 logw = 12 (1 Dlog(w/p)) + 13 Dlog( pi /qp)
4 e 1 {Q 6 5 K 6 }1/6
+ 14 log
16 e 2
119
(3.7)
(3.8)
DlogI
= 1 + 2
12 (qp/pi )13 {1 + 14 (1 + 2 )}
(C + Gc + DK + DKh + DK p + E n + E o )
+ 17 log
( pi /qp)I
14 (C + Gc + DK + DKh + DK p + E n + E o )
+ 18 log
S
(3.9)
DlogE n = 1 + 2 + 19 log
22 Y f 23 ( pf /qp)24
En
(3.10)
DF = 20 {25 (Q + P ) F }
(3.11)
(3.12)
32 pf
+ 25 log
qp
E n + E o + P F DKa
+ 26 (r 33 ) + 27 log
( pi /qp)I
(3.14)
pf
D logq = 24 Dlog
qp
2
(3.13)
120
Model Specification
DS = Q + ( pi /qp)I C DK DKh
DK p E n E o Gc
(3.15)
D1 = 1
(3.16)
D2 = 2
(3.17)
D3 = 3
(3.18)
121
(3.19)
(3.20)
122
Model Specification
123
into account in forming their expectations. The assumption that the agents in the economy know the parameters
is, of course, more realistic than the assumption that they
know the parameters affecting the short-run dynamics of
the model, which would be necessary to justify the assumption of full rational expectations. It is shown in Chapter 4
that the steady state growth rate of most of the real variables
in the model is 1 + 2 . The long-run rational expectations
assumption is incorporated in equations of the form (3.19)
or (3.20) relating to these variables, therefore, by setting the
expectations parameter equal to 1 + 2 .
The reference, in the preceding paragraph, to the steady
state growth rate leads us, naturally, to the discussion of
a third general feature of the model. Like the model of
Bergstrom and Wymer [1976] and the many models for
which that model served as a prototype (including the model
of Bergstrom, Nowman and Wymer [1992]), it is designed
in such a way as to permit a rigorous mathematical analysis of its steady state and stability properties, thus providing
a check on the capacity of the model to generate plausible
long-run behaviour. Because of the smallness of the samples available for the estimation and testing of macroeconometric models, it is very important to ensure, preferably by
the mathematical analysis of their dynamic properties, that
124
Model Specification
125
126
Model Specification
127
form
of
(3.19),
with
128
Model Specification
form
of
(3.20)
with
129
3.5 Employment
In the next few sections, we shall discuss the equations
relating to the production sector of the economy. These
are the adjustment equations for employment, private nonresidential fixed capital, output and the price level. It will be
useful, before discussing these equations in detail, to say
something about the underlying assumptions concerning
firm behaviour and market structure. It is assumed that the
economy is made up of a large number of monopolistic competitors with identical production functions and uniformly
differentiated products. Since the prices charged by the different firms in the economy are equal, it can be treated as
a single product economy, and it tends to a perfectly competitive economy as the degree of differentiation between
the products of the different firms tends to zero, that is as
the price elasticity of the partial demand function for each
firms product tends to minus infinity.
130
Model Specification
It would be possible to derive a set of adjustment equations for employment, output, capital and the price level
by a multidimensional generalization of the analysis in
Appendix A of this chapter, assuming that the agent has
control of a vector of variables, rather than a single variable.
But, to limit the number of parameters to be estimated, we
followed a simpler approach, which takes account of our a
priori knowledge of the relative speeds at which different
factors of production can be adjusted (or the relative costs
of adjusting them at a given speed). We know that it is easier to adjust output by varying the intensity with which the
employed labour force is used (varying the number of hours
per week worked by each employee) than to vary the number of persons employed, and it is easier to vary the number
of persons employed than to vary the stock of fixed capital. We assume, therefore, that, at each point of time, output is adjusting in response to sales, the number of persons
employed is adjusting in response to output and the stock
of capital is adjusting in response to the marginal product of
capital and the real interest rate.
The remainder of this section is devoted to a more detailed
discussion of the employment adjustment equation (3.2).
This is formulated as a second-order differential equation,
131
partly for a priori reasons mentioned in the preceding paragraph, and partly on the basis of the estimated speed of
adjustment parameters in the employment equation in the
model of Bergstrom, Nowman and Wymer [1992]. These
estimates imply that the time lag distribution with which
employment adjusts to output and capital has a mode of
3.48 quarters (the modal time lag). The exponential time
lag distribution generated by a first-order differential equation would, therefore, be a poor approximation to that
distribution.
Equation (3.2) assumes that the acceleration of the logarithm of employment depends on the excess of its steady
state growth rate 2 over its current growth rate and on
the ratio of the partial equilibrium level of employment to
the current level. The function 4 e 1 {Q 6 5 K 6 }1/6
defining the partial equilibrium level of employment is
a constant elasticity of substitution production function,
which incorporates the productivity stochastic trend variable 1 as well as output and capital. This function
plays a central role in the model. Its parameters occur
in several other equations, resulting in a very parsimonious parametrization and cross equation restrictions. The
elasticity of substitution between labour and capital is
1/(1 + 6 ) and can be derived as follows.
132
Model Specification
(3.21)
{5 6 K (1+6 ) }
{Q
5 K 6 } 6
=
K
6
(1+6 ) (1+6 )
= 5 (4 e 1)6 4 e 1{Q 6 5 K 6}1/6
K
(1+6 )
L
= 5 (4 e 1 )6
,
K
and hence
L
=
K
(4 e 1 )6
5
1
1+6
"
(3.22)
133
134
Model Specification
1+6
4 e 1 6
L
6 6
=
5 6 K (1+6 ) ,
{Q
5 K }
K
6
1+6
4 e 1 6
L
=
6 Q (1+6 ) ,
{Q
5 K 6 } 6
Q
6
and hence
{5 6 K (1+6 ) }
Q
=
= 5
K
{6 Q (1+6 ) }
Q
K
(1+6 )
.
135
equals the real interest rate plus the risk premium and capital is growing at its steady state rate, its acceleration is
zero.
3.7 Output
Equation (3.5) assumes that the proportional rate of increase in output depends on the ratio of the partial
equilibrium level of output to the current level of output
and the ratio of the partial equilibrium level of stocks
(inventories) to the current level of stocks. The partial
equilibrium level of stocks is assumed to equal 14 (C +
Gc + DK + DKh + DK p + E n + E 0 ), where the term in brackets equals the total sales for private and public consumption, capital formation and exports, while the parameter 14
is the optimal ratio of stocks to sales. It is assumed that, in
the state of partial equilibrium, the total supply from output
and imports {Q + ( pi /pq)I } is just sufficient to meet current
sales and maintain the constant ratio 14 of stocks to sales,
when sales are growing at their steady state rate 1 + 2 .
The supply required to maintain the constant ratio 14 of
stocks to sales, when sales are growing at the rate 1 + 2 , is
14 (1 + 2 )(C + Gc + DK + DKh + DK p + E n + E 0 ), and the
total supply required to meet current sales and maintain
136
Model Specification
(3.24)
(3.25)
137
138
Model Specification
L i
pi = 15 T2 w
Q i
139
.
L
p = 15 T2 w
Q
.
140
Model Specification
rate of increase of its partial equilibrium level, its acceleration is zero. Finally, it should be mentioned that the speed
of adjustment parameters in (3.6) depends, not only on
adjustment costs (which, for price variables, are relatively
minor), but also on the time lags in the competitive process,
in which each firm is adjusting its price in small steps, taking
account of the current prices of its competitors products (see
Chamberlin [1946, Ch. 5, Section 3]).
141
levels of output and capital) equals 16 e 2 . The labour supply, so defined, is not directly measurable, since it depends,
not only on demographic factors, but also on conditions in
the labour market. It is assumed, therefore, to be an exponential function of the unobservable trend variable 2 with
constant drift parameter 2 and, in the stochastic version of
the model, constant volatility parameter.
The first term on the right-hand side of (3.7) has the same
form as the first term on the right-hand side of (3.6) (discussed above) with the opposite sign. The two equations
together imply that, if the rate of increase in the real wage
rate exceeds the rate of technical progress and all other terms
on the right-hand side of each equation are zero, then there
will be an acceleration of the price level and a deceleration
of the wage rate, while, if the rate of increase in the real
wage rate is less than the rate of technical progress and all
other terms on the right-hand side of each equation are zero,
then there will be an acceleration of the wage rate and a
deceleration of the price level. The term Dlog( pi /q p) allows
for pressure (by trade unions and other organizations) for
higher wages to compensate for the loss of welfare caused
by the fall in the real exchange rate (since, although w/p
is the measure of the real wage that is relevant for firms
employing labour, it is not a measure of welfare).
142
Model Specification
143
144
Model Specification
3.11 Imports
Equation (3.9) assumes that the proportional rate of increase
in the volume of imports depends on the ratio of the partial
equilibrium real value of imports to the current real value
of imports and the ratio of the partial equilibrium level of
stocks to the current level of stocks. It implies that, when
stocks and the real value of imports are at their steady state
levels, the volume of imports will increase at its steady state
rate 1 + 2 . The partial equilibrium real value of imports
has been discussed in Section 3.7, where it was shown that
the price elasticity of demand for imports is (1 + 13 ).
145
146
Model Specification
147
148
Model Specification
149
150
Model Specification
3.17 Stocks
The stock adjustment equation (3.15) is, formally, an identity, although it is not exactly satisfied by the data, because of
errors in measurement. It will be treated as a stochastic differential equation with white noise innovation, therefore,
in the estimation of the parameters of the model.
151
3.18 Conclusion
In this chapter we have specified a new continuous time
econometric model of the United Kingdom. The main innovative feature of the new model is the incorporation of
stochastic trends to represent unobservable variables that,
in earlier models, have been represented by deterministic
trends. This and other general properties of the model have
been discussed in Section 3.2 and the specification of the
individual equations in Sections 3.33.17. Another important general property of the model is that, like the earlier
UK models of Bergstrom and Wymer [1976] and Bergstrom,
Nowman and Wymer [1992], it is designed in such a way as
to permit a rigorous mathematical analysis of its steady state
and stability properties. This analysis will be carried out in
Chapter 4, to which we now turn.
152
Model Specification
153
(3.A1)
154
Model Specification
(3.A2)
(3.A3)
(3.A4)
(3.A5)
155
(r 0)
(3.A6)
(3.A7)
so that y(t + r ) is the deviation of x from its partial equilibrium level at time t + r. Then, from (3.A6) and (3.A7), we
obtain
d 2 y(t + r ) 1
y(t + r ) = 0
dr 2
a
(r 0),
(3.A8)
(r 0),
(3.A9)
(r 0),
(3.A10)
156
Model Specification
r , we must put c 2 = 0 and c 1 = y(t), so that the optimal planned path of y(t + r ) is given by
y(t + r ) = y(t)e r
(r 0).
(3.A11)
(3.A13)
d x(t)
d x(t + r )
,
=
dt
dr
r =0
(3.A14)
157
right-hand side is that of the optimal planned path. It follows from (3.A13) and (3.A14) that the actual path of x
satisfies the differential equation
d x(t)
= + { f (z(t)) x(t)}.
dt
(3.A15)
158
Model Specification
where
C {x(t + r ), d x(t + r )/dr, d 2 x(t + r )/dr 2 }
= {x(t + r ) f (z(t) r )}2 + a{d x(t + r )/dr }2
+ b{d 2 x(t + r )/dr 2 }2
(3.A16)
Cx
(3.A17)
where C x , C x and C x are the partial derivatives of the function C with respect to x(t + r ), d x(t + r )/dr and d 2 x(t + r )/
dr 2 , respectively (see Courant [1936, Vol. II, Ch. 7]). From
(3.A16) we obtain
C x = 2{x(t + r ) f (z(t)) r },
(3.A18)
(3.A19)
(3.A20)
159
dr 4
b
dr 2
1
+ {x(t + r ) f (z(t)) r } = 0 (r 0), (3.A21)
b
which must be satisfied by the optimal planned path of
x(t + r ) (r 0). Then, from (3.A7) and (3.A21) we obtain
d 4 y(t + r ) a d 2 y(t + r )
1
+ y(t + r ) = 0 (r 0),
dr 4
b
dr 2
b
(3.A22)
which is the differential equation that must be satisfied by
the optimal planned path of y(t + r ) the deviation of x(t + r )
from its expected partial equilibrium level.
The solution of (3.A22) depends on the roots of the
equation
u4
a 2 1
u + = 0.
b
b
(3.A23)
160
Model Specification
(r 0), (3.A24)
where D = d/dr.
The general solution of the differential equation (3.A24)
is
y(t + r ) = C 1 e r + C 2 e r + C 3 e r + C 4 e r,
(3.A25)
d
{C 1 e r + C 2 e r }
dr
r =0
dy(t)
,
dt
(3.A26)
(3.A27)
or
C 1 + C 2 = y(t),
C 1 C 2 =
dy(t)
.
dt
(3.A28)
(3.A29)
{y(t) + dy(t)/dt}
,
( )
161
(3.A30)
C2 =
{y(t) + dy(t)/dt}
.
( )
(3.A31)
(r 0),
(3.A32)
or
(r 0),
(3.A33)
d 2 y(t + r )
dy(t + r )
= 1
2 y(t + r ), (3.A34)
dr 2
dr
1
dr 2
dr
+ 2 { f (z(t)) + r x(t + r )}
(r 0),
(3.A35)
162
Model Specification
2
d 2 x(t)
d x(t + r )
=
dt 2
dr 2
r =0,
(3.A36)
d 2 x(t + r )
d x(t + r )
+ 2 { f (z(t)) x(t)}.
= 1
dr 2
dr
(3.A37)
163
(3.B1)
(3.B2)
(3.B3)
164
Model Specification
t
t
e s g(z(s ))ds.
= e
(3.B5)
t
e (ts ) g(z(s ))ds + e t e t g(z(t))
= 2
= {g(z(t)) y(t)},
(3.B6)
165
(3.B7)
We shall assume that the roots of (3.B7) are real and negative, in which case y(t) converges monotonically to g(z(t))
as t . Letting and denote the two roots, we have
+ = 1 and = 2 , so that (3.B6) can be written
dy(t)
d 2 y(t)
+ ( + )
+ y(t) = g(z(t)).
dt 2
dt
(3.B8)
y(t) =
(r )g(z(t r ))dr,
(3.B9)
where
(r ) =
{e r e r }.
(3.B10)
e (ts ) g(z(s ))ds .
(3.B11)
166
Model Specification
t
d t (ts )
d
e (ts ) g(z(s ))ds
e
g(z(s ))ds .
dt
dt
(3.B12)
But,
d
dt
t
d
e (ts ) g(z(s ))ds =
e s g(z(s ))ds
e t
dt
t
t
d
e s g(z(s ))ds + e t
e s g(z(s ))ds
= e t
dt
t
=
e (ts ) g(z(s ))ds + e t e t g(z(t))
t
e (ts ) g(z(s ))ds + g(z(t)),
(3.B13)
=
and, similarly,
t
t
d
e (ts ) g(z(s ))ds =
e (ts ) g(z(s ))ds + g(z(t)).
dt
(3.B14)
From (3.B12), (3.B13) and (3.B14), we obtain
t
dy(t)
dt
t
(ts )
e
g(z(s ))ds .
167
(3.B15)
=
e (ts ) g(z(s ))ds
2
dt 2
t
e (ts ) g(z(s ))ds + g(z(t)).
2
(3.B16)
r (r )dr =
=
=
=
=
1
r
r
r e dr
re dr ,
0
0
1
2
2
1
,
+
,
1
1
+ .
(3.B17)
168
Model Specification
log log
.
(3.B18)
(3.B19)
(3.B20)
log r = log r.
(3.B21)
and hence,
Solving (3.B21) for r, we obtain the expression on the righthand side of (3.B18).
We shall deal, finally, with the case in which the roots of
(3.B7) are equal, each being equal to . In this case, (3.B6)
can be written
d 2 y(t)
dy(t)
+ 2
+ 2 y(t) = 2 g(z(t)).
dt 2
dt
169
(3.B22)
where
(r ) = 2 r e r .
(3.B24)
t
te t e s g(z(s ))ds 2
s e t e s g(z(s ))ds
t
t
e s g(z(s ))ds 2 e t
s e s g(z(s ))ds.
= 2 te t
= 2
(3.B25)
t
s e s g(z(s ))ds { 3 e t }
+
+ tg(z(t)) 2 tg(z(t))
t
e s g(z(s ))ds { 2 e t 3 te t }
=
t
s e s g(z(s ))ds { 3 e t }.
(3.B26)
+
170
Model Specification
t
s e s g(z(s ))ds { 4 e t }
+
t
s e s g(z(s ))ds { 4 e t } + 2 g(z(t)).
+
(3.B27)
1 x
x e ,
()
171
where and are parameters and () is the Gamma function defined by
() =
x 1 e x d x.
In the special case where = 2, we obtain the density function defined by (3.B24). The mean of the Gamma distribution is /, and hence the mean of the distributed lag defined
by (3.B24) is 2/. Moreover, it is easily verified that the mode
of the latter distribution is 1/.
172
CHAPTER FOUR
n important feature of the continuous time macroeconometric model specified in this monograph is that,
173
(in some cases zero). Closed formulae from which the exact
steady state paths of the variables can be computed are
derived in Appendix A. In Section 4.3 and Appendix 4.3,
we derive, from the original model, an explicit differential
equation system in the logarithms of the ratios of the variables to their steady state paths and show that the nonlinear
174
(4.1)
dx = 0,
(4.2)
E 0 = E 0 e (1 +2 )t ,
(4.3)
Gc = g (Q + P ),
(4.4)
K p = K p e (1 +2 )t ,
(4.5)
M = M e 4 t ,
(4.6)
pf = pf e 5 t ,
(4.7)
pi = pi e 5 t = pf e 5 t ,
(4.8)
rf = rf ,
(4.9)
T1 = T1 ,
(4.10)
T2 = T2 ,
(4.11)
Yf =
Y f e
1 +2
23
175
"
t
(4.12)
32 =
q p
pf
(4.13)
33 = r ,
(4.14)
1 (0) = 0,
(4.15)
2 (0) = 0,
(4.16)
3 (0) = 0,
(4.17)
(4.18)
E n = E n e (1 +2 )t ,
(4.19)
F = F e (1 +2 )t ,
(4.20)
I = I e (1 +2 )t ,
(4.21)
176
K = K e (1 +2 )t ,
(4.22)
K a = K a e (1 +2 )t ,
(4.23)
K h = K h e (1 +2 )t ,
(4.24)
L = L e 2 t ,
(4.25)
p = p e (3 +4 1 2 )t ,
(4.26)
P = P e (1 +2 )t ,
(4.27)
q = q e (1 +2 +5 3 4 )t ,
(4.28)
Q = Q e (1 +2 )t ,
(4.29)
r = r ,
(4.30)
S = S e (1 +2 )t ,
(4.31)
w = w e (3 +4 2 )t ,
(4.32)
1 = 1 t,
(4.33)
2 = 2 t,
(4.34)
3 = 3 t,
(4.35)
177
where C , E n , F , I , K , K a , K h , L , p, P , q , Q , r , S and
w are constants (hereafter referred to as steady state level
parameters) satisfying the nonlinear simultaneous equation
system (4.36)(4.50).
C = 1 e
Q + P
,
T1
1/6
5 (K )6
,
{(2 3 )(3 +4 1 2 )2 r }
L = 4 (Q )6
(4.36)
(4.37)
Q + P
, (4.38)
= 7 e
T1
1+6
Q
, (4.39)
r = 10 (3 + 4 1 2 ) 11 + 5
K
1 12 q p /pi 13 {1 + 14 (1 + 2 )}
C +G +(1 +2 ) K + K + K + E + E
p
n
o
c
h
8 log
K h
+ 9 log
{(8 9 )(3 +4 1 2 )8 r }
14 C + Gc + (1 + 2 ) K + K h + K p
+E n + E o
= 0,
p = 15 4 T2 w {1 5 (Q /K )6 }(1+6 )/6 ,
6
(Q )
5 (K )
=
178
16
4
(4.40)
(4.41)
6
,
(4.42)
r = 18 rf + 20 { p (Q + P )/M } + 21 (B /M )
+ 17 + 19 (3 + 4 1 2 5 ),
(4.43)
h
17 log
pi /q p I
14 C + Gc + (1 + 2 ) K + K h + K p
+E + E
n
o
+ 18 log
= 0,
(4.44)
E n
25
22 Y f 23
Q + P
F
=
pf
q p
24
,
20 + 1 + 2
,
20
!
" K
21 + 1 + 2
a
=
,
26 + 27 rf 5
P
21
(4.45)
(4.46)
(4.47)
!
"
Q + P
28 + 29 rf r + 30 (3 + 4 1 2 5 )
K a
2
23 + (1 + 2 ) + 22 (1 + 2 )
=
,
(4.48)
23
179
pi I
= E n + E o + P F (1 + 2 )K a ,
q p
(4.49)
p I
(1 + 2 )S = Q + i C + Gc + (1 + 2 )
q p
K + K h + K p + E n + E o .
(4.50)
pi
p f ).
q p
p f
q p
pi
(implicitly, assum-
180
y1 (t) =
y3 (t) =
y5 (t) =
y7 (t) =
y9 (t) =
y11 (t) =
y13 (t) =
y15 (t) =
C (t)
L(t)
log
(t)
=
log
,
y
,
2
C e (1 +2 )t
L e 2 t
K h (t)
K (t)
,
y4 (t) = log
,
log
K h e (1 +2 )t
K e (1 +2 )t
Q (t)
p(t)
,
y
,
log
(t)
=
log
6
Q e (1 +2 )t
p e (3 +4 1 2 )t
w(t)
, y8 (t) = r (t) r ,
log
w e (3 +4 2 )t
I (t)
E n (t)
log ( + )t ,
y10 (t) = log
,
I e 1 2
E n e (1 +2 )t
F (t)
P (t)
,
y
,
log
(t)
=
log
12
F e (1 +2 )t
P e (1 +2 )t
K a (t)
q(t)
,
y
,
log
(t)
=
log
14
K a e (1 +2 )t
q e (1 +2 +5 3 4 )t
S(t)
log ( + )t .
S e 1 2
log{(Q e )
5 (K e ) } y2 , (4.52)
6
181
(4.53)
y6 +y14
(4.54)
13
q p e
pi
13
q p
log 1 12
pi
+ log C e y1 + Gc + K e y4 (D y4 + 1 + 2 )
D y5 = 8 log 1 12
+ K h e y3 (D y3 + y1 + y2 ) +K p (1 + 2 ) + E n e y10 + E o
log C +Gc +(1 +2 ) K + K h + K p + E n + E o y5
+ 9 log C e y1 + Gc + K e y4 (D y4 + 1 + 2 )
+ K h e y3 (D y3 + y1 + y2 )+ K p (1 + 2 )+ E n e y10 + E o
log C +Gc +(1 +2 ) K + K h + K p + E n + E o y15 ,
(4.55)
1
+
6
D 2y6 = 10 (D y7 D y6 ) + 11 y7 y6
6
y5 6
1 + 6
Q e
log 1 5
+
K e y4
6
6
Q
log 1 5
,
(4.56)
K
182
D 2y7 = 12 (D y6 D y7 ) 13 (D y6 + D y14 )
1
6
log (Q )6 5 (K )
+ 14
6
1
y5 6
y4 6
log{(Q e )
5 (K e ) } , (4.57)
6
p e y6 (Q e y5 + P e y12 )
D y8 = 15 D y8 + 16 20
M
p (Q + P )
20
(4.58)
D
y
y
19
14
8 ,
M
2
D y9 = 17 (1 + 13 )(y6 + y14 ) y9
+ log C e y1 + Gc + K e y4 (D y4 + 1 + 2 )
+ K h e y3 (D y3 + y1 + y2 ) + K p (1 + 2 ) + E n e y10 + E o
log C + Gc + (1 + 2 ) K + K h + K p + E n + E o
+ 18 log C e y1 + Gc + K e y4 (D y4 + 1 + 2 )
+ K h e y3 (D y3 + y1 + y2 ) + K p (1 + 2 ) + E n e y10 + E o
log C + Gc + (1 + 2 ) K + K h + K p
(4.59)
+ E n + E o y15 ,
D y10 = 19 [24 (y6 + y14 ) + y10 ],
D y11 = 20 25
Q e y5 + P e y12
F e y11
183
Q + P
F
(4.60)
,
(4.61)
!
" K e y13
K a
a
D y12 = 21 26 + 27 rf 5
,
P e y12
P
(4.62)
y
K a e 13
Q + P
30 (1 + 2 + 5 3 4 )
,
K a
(4.63)
D 2y14 = 24 (D y6 + D y14 ) 25 (y6 + y14 ) + 26 y8
+ 27 log E n e y10 + E o + P e y12 F e y11
K e y13 (D y13 + 1 + 2 )
log E n + E o + P F (1 + 2 )K a
(4.64)
+ y6 + y14 y9 ,
D y15 = Q e y5 + pi /q p I e y9 y6 y14 C e y1
Gc K e y4 (D y4 + y1 + y2 ) K h e y3 (D y3 + y1 + y2 )
K p (1 +2 ) E n e y10 E o /S e y15 Q + pi /q p I
C Gc (1 + 2 ) K + K h + K p
(4.65)
E n E o /S .
184
(i = 1, . . . , 15).
(4.66)
the PoincareLiapounovPerron
conditions f i (y)/|y| 0 as
|y| 0 (i = 1, . . . , 15).
D y1 = 1 y1 1 2 y8 1 (3 2 )D y6 +
1 P
+
y12 + f 1 (y),
Q + P
185
1 Q
Q + P
y5
(4.67)
3 (Q )6
y5
D 2y2 = 2 D y2 3 y2 +
(Q )6 5 (K )6
3 5 (K )6
y4 + f 2 (y),
(4.68)
(Q )6 5 (K )6
D 2y3 = 4 D y3 + 5 (8 9 )D y6 5 y3 5 8 y8
5 Q
5 P
y
y12 + f 3 (y), (4.69)
+
+
5
Q + P
Q + P
D 2y4 = 6 D y4 +
7 10
r 10 (3 + 4 1 2 ) + 11
D y6
7 (1 + 6 )y4 + 7 (1 + 6 )y5
7
y8 + f 4 (y),
r 10 (3 + 4 1 2 ) + 11
(4.70)
8 12 13 (x )13
D y5 = 5 y5 9 y15
(y6 + y14 ) + (8 + 9 )
1 12 (x )13
C y1 + (1 + 2 )K y4 + K D y4 + (1 + 2 )K h y3
+ K h D y3 + E n y10
C + Gc + (1 + 2 ) K + K h + E n + E o
+ (1 + 2 )K p
+ f 5 (y),
(4.71)
D 2y6 = 10 (D y6 D y7 ) 11 (y6 y7 )
! "6
11 5 (1 + 6 ) KQ
+
(y5 y4 ) + f 6 (y),
6
1 5 KQ
186
(4.72)
D 2y7 = 12 (D y6 D y7 ) 13 (D y6 + D y14 )
6
(Q ) y5 5 (K )6 y4
+ 14
+ f 7 (y),
(Q )6 5 (K )6
(4.73)
D 2y8 = 15 D y8 16 19 D y14 16 y8
p (Q + P )
p Q
y6 + 16 20
y5
+ 16 20
M
M
p Q
y12 + f 8 (y),
+ 16 20
(4.74)
M
D y9 = 17 y9 18 y15 + 17 (1 + 13 )(y6 + y14 ) + (17 + 18 )
C y1 + (1 + 2 )K y4 + K D y4 + (1 + 2 )K h y3
+ K h D y3 + E n y10
C + Gc + (1 + 2 ) K + K h + K p + E n + E o
+ f 9 (y),
(4.75)
D y11
Q
P
y5 + 20 25
y12
= 20 25
F
F
Q + P
y11 + f 11 (y),
20 25
F
(4.76)
(4.77)
!
" K
a
(y13 y12 ) + f 12 (y),
D y12 = 21 26 + 27 rf 5
P
(4.78)
187
D y13
Q + P
y8
= {22 + 2(1 + 2 )}D y13 23 29
K a
!
"
Q + P
D
y
r
23 30
+
r
14
23
28
29
f
K a
Q
+ 30 (3 + 4 1 2 5 )}
y5
K a
!
"
+ 23 28 + 29 rf r + 30 (3 + 4 1
!
"
P
r
2 5 )
r
y
12
23
28
29
f
K a
+ 30 (3 + 4 1 2 5 )
Q + P
y13 + f 13 (y),
(4.79)
K a
K D y13
a
+ 27
E n + E o + P F (1 + 2 )K a
+ f 14 (y),
(4.80)
pi I
Q
C
y5 +
(y9 y6 y14 )
y1
=
S
q p S
S
(1 + 2 )K h
(1 + 2 )K
K
y4
D y4
y3
S
S
S
D y15
188
K h
S
D y3
E n
S
y10
p I
Q + q i p C Gc (1 + 2 )
K + K + K E E
p
n
o
h
y15 + f 15 (y).
(4.81)
(4.82)
where C is a 23 23 matrix, whose elements are functions of the parameters of the model (3.1)(3.18) and u =
[u1 , u2 , . . . , u23 ], = [1 , 2 , . . . , 23 ], ui (i = 1, . . . , 23) and
i (i = 1, . . . , 23) are defined as follows.
u1 (t) = y1 (t), u2 (t) = y5 (t), u3 (t) = y9 (t), u4 (t) = y10 (t),
u5 (t) = y11 (t), u6 (t) = y12 (t), u7 (t) = y15 (t), u8 (t) = y2 (t),
u9 (t) = y3 (t), u10 (t) = y4 (t), u11 (t) = y6 (t), u12 (t) = y7 (t),
u13 (t) = y8 (t), u14 (t) = y13 (t), u15 (t) = y14 (t), u16 (t) = D y2 (t),
u17 (t) = D y3 (t), u18 (t) = D y4 (t), u19 (t) = D y6 (t), u20 (t)
= D y7 (t),
u21 (t) = D y8 (t), u22 (t) = D y13 (t), u23 (t) = D y14 (t),
1 (u) = f 1 (y), 2 (u) = f 5 (y), 3 (u) = f 9 (y), 4 (u) = f 10 (y),
189
(i = 8, . . . , 15).
190
conditions
f i (y)/|y| 0
as
|y|
191
of the original model (3.1)(3.18) as asymptotically stable. For, convergence of the variables ui (t) (i = 1, . . . , 23) to
zero implies that the proportional deviations of the variables
C(t), E n (t), . . . , w(t) from their steady state paths converge
to zero. The steady state level parameters are of interest in
themselves, and, assuming they have, already, been computed, the additional cost of computing the unknown elements of C and the eigenvalues of this matrix is comparatively light. Finally, it should be noted that the eigenvalues
of C provide information not only about the stability of the
steady state solution but also about the cyclical properties of
a solution of the model in the neighbourhood of the steady
state.
192
Q
Q e (1 +2 )t
log
T1
T1
.
193
the limiting case of in which case the feedback relation becomes instantaneous. Bergstrom, Nowman and Wandasiewicz [1994] found that this simple fiscal policy feedback
in cases of either instantaneously or with an exponential
time lag had no definite stabilizing influence on the dynamic
behaviour of the model.
In Bergstrom, Nowman and Wandasiewicz [1994] they
also investigated stability analysis, using the more advanced
approach of optimal control theory. They proved the existence of stabilizing feedbacks involving the fiscal policy variable and the money supply as combined policy instruments.
They also derived optimal feedbacks by minimizing an infinite horizon quadratic cost function involving deviations of
output, the exchange rate and the monetary and fiscal policy instruments from their steady state paths (and also, the
rates of change of the policy instruments). One of the findings was that fiscal policy had an important role to play
in the stabilization of the model than does monetary policy (see Bergstrom, Nowman and Wandasiewicz [1994] for
full details). In a more recent study, Bergstrom [2001] also
investigated the stability of the Bergstrom, Nowman and
Wymer [1992] model, using the estimates of the model
in Nowman [1996] by looking at the wage-acceleration
parameter that determines the effect of labour demand
194
195
196
4.5 Conclusion
We have shown, in this chapter, that, under certain assumptions concerning the paths of the exogenous variables,
the continuous time macroeconometric model specified in
Chapter 3 has a steady state solution in which all variables
grow at constant proportional rates (in some cases, zero). We
have, also, derived sufficient conditions for the asymptotic
stability of the steady state solution in terms of the coefficients of the linear part of a transformed model. We now
turn to some empirical results from UK data in the next
chapter.
197
h(r ) =
!
"
!
"
21 23 26 + 27 rf 5 {28 + 29 rf r
+ 30 (3 + 4 1 2 5 )}
(21 + 1 + 2 ){23 + (1 + 2 )2 + 22 (1 + 2 )}
Therefore,
Q + P
1
=
.
Q
1 h(r )
(4.A1)
(4.A2)
198
(4.A3)
(4.A4)
(4.A5)
13
p
q
{1 + 14 (1 + 2 )} C + Gc
Q = 1 12
pi
+ (1 + 2 ) K + K h + K p + E n + E o ,
(4.A6)
13
{1 + 14 (1 + 2 )} C + Gc + (1 + 2 )
(4.A7)
(K + K h + K p ) + E n + E o ,
S = 14 C + Gc + (1 + 2 ) K + K h + K p + E n + E o .
pi I
= 12
q p
q p
pi
(4.A8)
Q =
16
4
1 5
199
K
Q
6 1/6
(4.A9)
6 /(1+6 ) 16
(
+
)
r
10
3
4
1
2
+ 11
16
,
1 5
Q =
4
5
(4.A10)
C + Gc + (1 + 2 ) K + K h + K p + E n + E o
Q
1
=
.
! "13
1 12 q pp
{1 + 14 (1 + 2 )}
(4.A11)
(4.A12)
where
x =
and
q p
q p
=
,
pf
pi
1 e (2 3 )(3 +4 1 2 ) e 2 r
(r , x ) =
g +
T1
(1 + 2 )7 e (8 9 )(3 +4 1 2 ) e 8 r
+
T1
1
1 h(r )
200
1/1+6
r 10 (3 + 4 1 2 )
+
11
+ (1 + 2 )
! "23
(1 + 2 )K p + 22 Y f
(x )24 + E o
+
6 /(1+6 ) 16
r 10 (3 +4 1 2 )+11
16
1 5
4
5
1
.
{112 (x )13 }{1 + 14 (1 + 2 )}
(4.A13)
P
h(r )
=
,
Q
1 h(r )
(4.A14)
from (4.A1),
h(r )
21 + 1 + 2
K a
!
"
=
,
Q
1 h(r )
21 26 + 27 r 5
(4.A15)
from (4.46),
F
25 20
=
,
Q +P
20 + 1 + 2
201
(4.A16)
(4.A17)
= 0.
Q Q q p Q
Q
Q
Q
(4.A18)
(4.A19)
where
! "23
24
22 Y f
(x )
+ Eo
(r , x ) =
16
4
1 5
r 10 (3 +4 1 2 )+11
5
6 /(1+6 ) 16
12 (x)13
1
25 20
112 (x )13
1 h(r )
20 + 1 + 2
(1 + 2 )(21 + 1 + 2 )
h(r )
!
" .
1
+
1 h(r )
+
r
21
26
27
202
p =
r 18rf 21 (B /M ) 17 19 (3 + 4 1
2 5 )
20
M
,
P + Q
(4.A20)
{1 h(r )}{r 18 rf 21 (B /M ) 17
19 (3 + 4 1 2 5 )}
p =
20
M
.
(4.A21)
203
From (3.1),
DlogC = 1 + 2 + 1 log1 2 (r Dlog p)
3 Dlog p + log(Q + P ) logT1 logC ,
we obtain
D y1 = 1 log1 2 (y8 + r ) + (2 3 )(D y6 + 3 + 4
1 2 )+log(Q e y5 + P e y12 ) logT1 log(C e y1 ) .
(4.B1)
From (4.36),
logC = log1 2r + (2 3 )(3 + 4 1 2 )
+ log(Q + P ) logT1 ,
204
(4.B2)
1
log{(Q )6 5 (K )6 },
6
(4.B4)
+ log(Q e y5 + P e y12 ) logT1 log(K h e y3 ) ,
(4.B5)
From (4.38),
logK h = log7 + (8 9 )(3 + 4 1 2 ) 8 r
+ log(Q + P ) logT1 ,
205
(4.B6)
From (3.4),
D y4 = 6 D y4 + 7 log5 + (1 + 6 ) log
2
Q e y5
K e y4
log{y8 + r 10 (D y6 + 3 + 4 1 2 ) + 11 }
= 6 D y4 +7 [log5 +(1+6 )(logQ logK + y5 y4 )
log{y8 + r 10 (D y6 + 3 + 4 1 2 ) + 11 }],
(4.B7)
from (4.39),
log5 + (1 + 6 )(logQ logK )
= log{r 10 (3 + 4 1 2 ) + 11 },
(4.B8)
206
(4.B9)
from (4.40),
8 log 1 12
q p
pi
13
+ log{1 + 14 (1 + 2 )}
+ log C + Gc + (1 + 2 ) K + K h + K p +E n + E o
+ logw + y7
1 + 6
6
log 1 5
Q e y5
K e y4
6
log p y6 ,
(4.B11)
from (4.41),
log p = log15 + log4 + logT2 + logw
6
Q
1 + 6
, (4.B12)
log 1 5
6
K
207
From (3.7),
D 2y7 = 12 (D y6 D y7 ) 13 (D y6 + D y14 )
1
log{(Q e y5 )6
+ 14 log4 log16
6
y4 6
5 (K e ) } ,
(4.B13)
from (4.42),
log4 log16 =
1
log{(Q )6 5 (K )6 },
6
(4.B14)
D 2y8 = 15 D y8 + 16 17 + 18 rf
19 (D y14 + 1 + 2 + 5 3 4 )
y6 y5
p e (Q e + P e y12 )
B
+12
y8 r ,
+ 20
M
M
(4.B15)
from (4.43),
r = 17 19 (1 + 2 + 5 3 4 ) + 8rf
p (Q + P )
B
+ 20
+ 21
,
(4.B16)
M
M
208
From (3.9),
q p
D y9 = 17 log12 + (1 + 13 ) log
pi
+ (1 + 13 )(y6 + y14 ) + log{1 + 14 (1 + 2 )}
+ log C e y1 + Gc + K e y4 (D y4 + 1 + 2 )
+ K e y3 (D y3 + 1 + 2 ) + K p (1 + 2 ) + E n e y10 + E o
logI y9 + 18 log14 + log C e y1 + Gc + K e y4
(D y4 + 1 + 2 ) + K h e y3 (D y3 + 1 + 2 )
+ K p (1 + 2 ) + E n e y10 + E o logS y15 , (4.B17)
from (4.44),
q p
+ log{1 + 14 (1 + 2 )}
17 log12 + (1 + 13 ) log
pi
+ log C + Gc + (1 + 2 ) K + K h + K p + E n + E o
logI +18 log14 + log C + Gc + (1 + 2 )
(4.B18)
K + K h + K p + E n + E o logS = 0,
and, from (4.B17) and (4.B18), we obtain (4.59).
From (3.10),
D y10 = 19 log22 +
23 logY f
q p
24 log
pf
209
(4.B19)
from (4.45),
q p
logE n = log22 + 23 logY f 24 log
pf
+
,
(4.B20)
1
,
D y11 = (1 + 2 ) + 20 25
F e y11
(4.B22)
from (4.46),
Q + P
= 20 + 1 + 2 ,
F
and, from (4.B22) and (4.B23), we obtain (4.61).
20 25
(4.B23)
From (3.12),
!
" K e y13
a
1 ,
Dlog p = 21 26 + 27 rf 5
P e y12
(4.B24)
from (4.B24),
D y12 = (1 + 2 ) + 21
!
" K e y13
a
26 + 27 rf 5
1
,
P e y12
(4.B25)
from (4.47),
!
" K
a
21 26 + 27 rf 5
= 21 + 1 + 2 , (4.B26)
P
210
from (4.B27),
D 2y13 = 22 (D y13 + 1 + 2 ) (D y13 + 1 + 2 )2
!
"
+ 23 28 + 29 rf r y8
30 (D y14 + 1 + 2 + 5 3 4 )}
y5
Q e + P e y12
1 ,
(4.B28)
K a e y13
from (4.48),
!
"
22 (1 + 2 ) + (1 + 2 )2 + 23 = 23 28 + 29 rf r
Q + P
+ 30 (3 + 4 1 2 5 )
,
(4.B29)
K a
211
y y y
y
Q e + ( pi /q p )I e 9 6 14 C e 1
K e y4 (D y4 + 1 + 2 ) K h e y3 (D y3 +1 +2 )
K p (1 + 2 ) E n e y10 E o Gc
,
+
S e y15
(4.B32)
212
(4.B33)
CHAPTER FIVE
213
214
215
x1 (t) = [logC(t), logE n (t), F (t), logI (t), P(t), logQ (t),
logS(t)] ,
x2 (t) = [logK(t), K a (t), logKh (t), logL(t), log p(t),
logq(t), r (t), logw(t)] ,
z(t) = [logB(t), logK p (t), logM(t), DlogK p (t), dx , logE 0 (t),
logGc (t), log pf (t), log pi (t), r f (t), logT1 (t), logT2 (t),
logY f (t), Dlog pf (t), Dlog pi (t)] ,
(t) = [1 (t), 2 (t), 3 (t)] .
(t 0),
(5.1)
(t 0),
(5.2)
(5.3)
216
217
218
Parameter
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Prior Lower
Bound
Prior Upper
Bound
0.8000
0.0000
0.0000
0.1000
0.0100
0.0100
1.0000
0.0000
0.0000
0.0000
0.0000
0.1000
0.5000
0.0000
1.0000
20000
1.0000
0.5000
0.0000
0.0050
0.0050
10000
0.1000
0.5000
1.0000
4.0000
4.0000
10.000
1.0000
1.0000
10.000
4.0000
4.0000
1.0000
0.0500
0.3000
0.5000
1.0000
1.1000
40000
0.0100
1.0000
1.0000
0.0500
0.0500
35000
1.0000
1.5000
Estimate
0.9351
0.2014
1.6176
0.1494
0.2630
0.2965
3.5000
0.1815
3.0574
0.1535
0.0077
0.2169
0.4033
0.7556
1.0000
23150
0.0832
0.9312
0.1593
0.0073
0.0060
29500
0.5000
1.0000
Standard
Error
0.0073
0.0138
0.0203
0.0298
0.0001
0.0043
0.1401
0.0042
0.0945
0.0365
0.0005
0.0102
0.0598
0.0258
44.2813
0.1087
1.6399
0.0224
0.0007
0.0142
451.7023
0.3562
0.0614
(continued)
219
Parameter
25
26
27
28
29
30
31
32
33
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Prior Lower
Bound
0.0000
0.0000
0.0000
2.0000
0.0000
0.0000
0.0000
0.5000
0.0000
0.0010
0.1000
0.0010
0.1000
0.0010
0.1000
0.0001
0.0010
0.0010
0.1000
0.0010
0.1000
0.0010
0.0010
0.1000
0.0010
0.0010
Prior Upper
Bound
0.0100
0.0200
1.0000
2.0000
100.00
10.000
0.1000
1.5000
10.000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
220
Estimate
0.0083
0.0081
0.2520
0.1303
57.0548
0.1577
0.0007
0.7059
0.0089
0.5294
0.2200
0.0100
4.9999
0.2760
0.0976
0.0009
0.1029
0.0944
0.2384
0.0015
4.3549
0.6494
0.0220
1.4070
0.0292
0.1039
Standard
Error
0.0021
0.0003
0.0315
0.0057
12.9646
0.0305
0.0001
0.1997
0.4913
0.0735
0.1650
0.0145
0.0336
0.0001
0.0040
0.00001
0.0028
0.0015
0.0195
0.000001
0.0749
0.0056
0.0016
0.5086
0.0039
0.0062
Parameter
18
19
20
21
22
23
24
25
26
27
1
2
3
Prior Lower
Bound
Prior Upper
Bound
Estimate
0.0010
0.0010
0.0010
0.0010
0.1000
0.0010
0.1000
0.0010
0.0010
0.0000
0.0000
0.0100
0.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
5.0000
0.0100
0.0100
0.0500
0.0917
0.1019
2.8565
0.7942
0.2033
0.0075
3.9345
0.1808
0.1244
0.0009
0.0028
0.0061
0.0098
Standard
Error
0.0010
0.0002
0.4912
0.2061
0.0811
0.0014
0.1045
0.0010
0.0047
0.00003
0.0019
0.0025
0.0017
221
222
speed of adjustment parameters 1 , 2 . . . , 27 is fairly obvious. Less obvious and, perhaps, of more interest are the
time lag distributions with which the variables adjust, each
in response to changes in the other variables on which
it is directly dependent. These time lag distributions are
presented and discussed in Section 5.3, to which we now
turn.
223
of the form (3.A15), the time lag distribution is an exponential distribution given by (3.B4). The density is a decreasing
function of the time lag, and the mean of the distribution is
the reciprocal of the speed of adjustment parameter .
When the adjustment equation is a second-order differential equation of the form (3.A37), the time lag distribution
is of a more complicated form, depending on the roots of
the quadratic equation (3.B7). In the normal case, where
these roots are real and negative, it will be a unimodal distribution with a positive mode. But, its exact form depends
on whether the roots are equal or unequal. When they are
unequal, the density function of the distribution is given
by (3.B10), its mean by (3.B17) and its mode by (3.B18),
while when they are equal, the density function is given
by (3.B24), the mean is 2/ and the mode is 1/. Estimates
of the means and modes of the lag distributions have been
derived from the estimated values of the parameters, for all
variables in the model that adjust through differential equations of the form (3.A15) or (3.A37) and are presented in
Table 5.2.
As can be seen, the private non-residential fixed capital
stock variable adjusts with the longest time lag and the financial flow variables with the shortest time lags. In particular,
the private non-residential fixed capital variable logK
224
Variable
logC
logEn
F
logI
logKh
logK
Ka
logL
P
logQ
r
0.9172
27.7487
10.9215
9.6795
3.0672
225
0.6
Density
0.5
0.4
0.3
0.2
0.1
0
0
10
Quarters
Turning to the first-order equations, we find that the nonoil exports variable logE n , output variable logQ and the
import variable logI adjust with mean time lags of approximately 21/2 years; the consumption variable logC and profits,
interest and dividends from abroad variable P adjust with
mean time lags of approximately 1/2 a year, and finally the
international transfers variable F adjusts with a mean time
lag of approximately 1 month. The graphs of the density
functions of the lag distributions are presented in Figures 5.1
to 5.11.
226
0.04
0.035
Density
0.03
0.025
0.02
0.015
0.01
0.005
0
0
10
20
30
40
Quarters
Density
0.05
0.04
0.03
0.02
0.01
0
0
10
15
Quarters
227
20
0.009
0.008
0.007
Density
0.006
0.005
0.004
0.003
0.002
0.001
0
0
20
40
60
80
100
120
140
Quarters
Density
0.08
0.06
0.04
0.02
0
0
10
15
Quarters
228
20
0.025
Density
0.02
0.015
0.01
0.005
0
0
10
20
30
40
Quarters
Density
0.08
0.06
0.04
0.02
0
0
10
15
Quarters
229
20
0.12
0.1
Density
0.08
0.06
0.04
0.02
0
0
10
15
20
Quarters
Density
2.5
2
1.5
1
0.5
0
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
Quarters
230
1.8
0.9
0.8
0.7
Density
0.6
0.5
0.4
0.3
0.2
0.1
0
0
Quarters
Density
0.005
0.004
0.003
0.002
0.001
0
0
10
20
30
40
50
60
Quarters
231
70
232
233
234
Assumed Value
B
E 0
g
Kp
M
pf
pi
r f
T1
T2
Y f
4
5
178786
1580
0.2400
172434
17143
1.0000
1.0000
0.0098
1.3103
1.1526
0.5514
0.0048
0.0000
Variable
C , En , F , I , K , Kh , K a , P , Q, S
L
p
w
r
q
1.92
0.80
0.00
1.12
0.00
0.00
235
Variable
C
En
F
I
Kh
K
Ka
L
P
p
Q
q
r
S
w
Steady State
Steady State Level in Actual Level in Last
Level Parameter Last Quarter of 1996 Quarter of 1996
65,787
26,456
765
20,218
246,283
741,387
11,365
23,151
120
0.5457
92,488
1.5174
0.0100
70,553
1.6230
100,365
40,362
1,168
30,844
375,733
1,131,074
17,339
27,606
182
0.5457
141,102
1.5174
0.0100
107,637
2.0764
100,619
42,999
1,096
47,103
579,636
713,621
18,154
26,244
1,570
1.2449
142,808
0.9139
0.0184
116,833
4.5106
236
237
238
Imaginary Part
4.9633
2.8613
1.4068
0.7990
0.5419
0.1597
0.1558
0.1033
0.1016
0.0642
0.0533
0.0469
0.0295
0.0017
0.0066
4.2409
0.0699
0.0012
0.1695
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.1792
0.0094
0.0202
0.0389
239
The model also generates a cycle with a period of approximately 9 years associated with the pair of eigenvalues
4.2409 0.1792. This could be the well-known economic
cycle commonly referred to as the trade cycle or business
cycle and generally recognized as having a period of 711
years. Although the pair of eigenvalues with which this cycle
is associated has a negative real part, the cycle will be prevented from dying out by the stochastic innovations in the
model, and its average amplitude will depend on the variances of these innovations. We may conclude, therefore,
that the model (3.1)(3.18) generates plausible long-run
behaviour, with fluctuations about the steady state neoclassical growth path including both a trade cycle with a period
of about 9 years and a longer (Kondratieff) cycle with a
period of about 40 years.
240
241
Table 5.7. Root Mean Square Errors of Post-Sample MultiPeriod Forecasts 1995Q11996Q4
Variable
logC
logEn
logI
logKh
logK
logL
log p
logQ
logq
r
logS
logw
RMSE of Forecasts
from VARX Model
0.0477
0.0797
0.0450
0.0350
0.0024
0.0137
0.0121
0.0395
0.0472
0.0012
0.1396
0.0130
0.0148
0.1131
0.1329
0.0024
0.0077
0.0057
0.0218
0.0181
0.0453
0.0007
0.0096
0.0227
242
Quarter
Actual Value
Forecast
logC
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
11.4364
11.4306
11.4545
11.4424
11.4579
11.4630
11.4862
11.4761
11.4436
11.4609
11.4739
11.4881
11.4990
11.5165
11.5424
11.5599
logEn
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
10.5943
10.5545
10.6041
10.6157
10.6551
10.6283
10.6432
10.6472
10.5315
10.5325
10.5348
10.5379
10.5418
10.5466
10.5523
10.5586
l ogI
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
10.6232
10.6450
10.6737
10.6913
10.7258
10.7203
10.7384
10.7746
10.6627
10.6566
10.6562
10.6598
10.6671
10.6761
10.6888
10.7010
(continued)
243
Quarter
Actual Value
Forecast
logKh
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
13.2508
13.2540
13.2564
13.2593
13.2616
13.2644
13.2667
13.2700
13.2566
13.2659
13.2753
13.2849
13.2941
13.3053
13.3166
13.3280
logK
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
13.4279
13.4345
13.4415
13.4476
13.4558
13.4627
13.4697
13.4771
13.4273
13.4337
13.4407
13.4481
13.4561
13.4644
13.4732
13.4824
logL
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
10.1576
10.1594
10.1558
10.1639
10.1593
10.1611
10.1680
10.1721
10.1629
10.1675
10.1713
10.1744
10.1772
10.1798
10.1824
10.1852
244
Quarter
Actual Value
Forecast
log p
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
0.1714
0.1817
0.1810
0.1870
0.2023
0.2058
0.2109
0.2199
0.1650
0.1710
0.1766
0.1819
0.1868
0.1917
0.1966
0.2017
logQ
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
11.7785
11.7749
11.7855
11.7857
11.7927
11.8043
11.8117
11.8270
11.7669
11.7767
11.7907
11.8079
11.8275
11.8486
11.8730
11.8967
logq
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
0.1332
0.1735
0.1769
0.1759
0.1761
0.1678
0.1624
0.0854
0.1064
0.1172
0.1284
0.1391
0.1497
0.1613
0.1722
0.1824
(continued)
245
Quarter
Actual Value
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
0.0207
0.0195
0.0195
0.0185
0.0203
0.0199
0.0196
0.0184
0.0192
0.0189
0.0187
0.0186
0.0184
0.0183
0.0182
0.0181
logS
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
11.6071
11.6243
11.6358
11.6506
11.6474
11.6565
11.6566
11.6709
11.6085
11.5754
11.5404
11.5114
11.4908
11.4813
11.4712
11.4822
logw
1995 Q1
Q2
Q3
Q4
1996 Q1
Q2
Q3
Q4
1.4552
1.4489
1.4535
1.4613
1.4983
1.4884
1.4856
1.5049
1.4382
1.4464
1.4542
1.4615
1.4686
1.4776
1.4862
1.4968
246
Forecast
247
5.6 Conclusion
In this final chapter, we have presented and discussed the
estimates of the parameters of the model in Chapter 3, which
were obtained from UK data by the exact Gaussian estimation procedure described in Chapter 2. Since the latter procedure is applicable only to a model that is linear in the
variables, although, generally, nonlinear in the parameters,
it was necessary, for the purpose of estimation, to use a linear (in the variables) approximation to the model specified
in Chapter 3. The parameter estimates are, on the whole,
very plausible, and all of the time lag distributions derived
from the speed of adjustment parameters are very plausible.
We have examined the steady state and stability properties of
the estimated model, using the results obtained in Chapter 4,
and shown that it generates plausible long-run behaviour.
More specifically, it generates fluctuations about the steady
state neoclassical growth path, including both a trade cycle
with a period of about 9 years and a longer (Kondratieff)
248
cycle with a period of about 40 years. Finally, we have examined the post-sample forecasting performance of the estimated model and compared it with that of a second-order
VARX model (vector autoregressive model with exogenous
variables). We conclude that the forecasting performance is
satisfactory, particularly considering the very parsimonious
specification of the model.
[A1 ]11 = 1 ,
1
[A1 ]15 = 1
,
e logQ + P
249
e logQ
[A1 ]16 = 1
[A1 ]22
e logQ + P
= 19 ,
,
[A1 ]33 = 20 ,
[A1 ]35 = 20 25 ,
[A1 ]36 = 20 25 e logQ ,
[A1 ]41 =
(17 + 18 )e logC
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
(17 + 18 )e logEn
e logC + e logGc + e logK DlogK + e logK h DlogK h
+ e logK p DlogK p + e logEn + e logEo
,
[A1 ]44 = 17 ,
[A1 ]47 = 18 ,
[A1 ]55 = 21 ,
[A1 ]61 =
(8 + 9 )e logC
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
[A1 ]66
(8 + 9 )e logEn
e logC + e logGc + e logK DlogK + e logK h DlogK h
[A1 ]67 = 9 ,
250
,
[A2 ]43 =
,
,
[A2 ]45 = 17 (1 + 13 ),
[A2 ]46 = 17 (1 + 13 ),
[A2 ]52 = 21 {26 + 27 (r f Dlog pf )},
[A2 ]61 =
(8 + 9 )e logK DlogK
e logC + e logGc + e logK DlogK + e logK h DlogK h
+ e logK p DlogK p + e logEn + e logEo
251
,
(8 + 9 )e logK h DlogK h
[A2 ]63 =
[A2 ]65
[A2 ]66
,
(17 + 18 )e logK
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
(17 + 18 )e logK h
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
(8 + 9 )e logK
e logC + e logGc + e logK DlogK + e logK h DlogK h
+ e logK p DlogK p + e logEn + e logEo
252
,
[A3 ]63 =
(8 + 9 )e logK h
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
3 e 6 logQ
e 6 logQ 5 e 6 logK
11 5 (1 + 6 )e 6 (logQ logK )
1 5 e 6 (logQ logK )
27 e
logE n
,
e logEn + e logEo + P F DKa
27
=
,
logE
logE
n + e
o + P F DK
e
a
= 27 ,
27
=
,
logE
logE
n
o
e
+e
+ P F DKa
253
where 3 =
3 T
2
14 e 6 logQ
,
e 6 logQ 5 e 6 logK
7 (1 + 6 ),
7
,
r 10 Dlog p + 11
23 ,
!
"
23 29 e logQ + P ,
[A5 ]33 = 5 ,
[A5 ]37 = 5 8 ,
[A5 ]41 =
[A5 ]44
[A5 ]51
[A5 ]55
3 5 e 6 logK
,
e 6 logQ 5 e 6 logK
= 3 ,
11 5 (1 + 6 )e 6 (logQ logK )
,
=
1 5 e 6 (logQ logK )
= 11 ,
[A5 ]58 = 11 ,
[A5 ]65 = 27 25 ,
[A5 ]66 = 27 25 ,
[A5 ]67 = 26 ,
[A5 ]75 = 16 20 e (log p+logQ logM3 ) + P e (log plogM3 ) ,
254
[A5 ]77 = 16 ,
[A5 ]81 =
14 5 e 6 logK
e 6 logQ 5 e 6 logK
,
[A6 ]11 = 6 ,
[A6 ]15 =
[A6 ]22
[A6 ]26
7 10
,
r 10 Dlog p + 11
= 22 ,
!
"
= 23 30 e logQ + P ,
[A6 ]33 = 4 ,
[A6 ]35 = 5 (8 9 ),
[A6 ]44 = 2 ,
[A6 ]55 = 10 ,
[A6 ]58 = 10 ,
[A6 ]62 =
[A6 ]65
27
[A6 ]66 = 24 ,
[A6 ]76 = 16 19 ,
[A6 ]77 = 15 ,
[A6 ]85 = 12 13 ,
[A6 ]86 = 13 ,
[A6 ]88 = 12 ,
[B1 ]1,11 = 1 ,
[B1 ]1,28 = 19 24 ,
255
[B1 ]2,13 = 19 23 ,
[B1 ]42 =
,
(17 + 18 )e logK p
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
(17 + 18 )e logE o
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
[B1 ]49
(17 + 18 )e logGc
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
[B1 ]5,10 = 21 27 K a ,
[B1 ]5,14 = 21 27 K a ,
[B1 ]62 =
(8 + 9 )e logK p DlogK p
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
(8 + 9 )e logK p
e logC + e logGc + e logK DlogK + e logK h DlogK h
+ e logK p DlogK p + e logE n + e logEo
256
,
[B1 ]66 =
(8 + 9 )e logEo
e logC + e logGc + e logK DlogK + e logK h DlogK h
,
(8 + 9 )e logGc
e logC + e logGc + e logK DlogK + e logK h DlogK h
+ e logK p DlogK p + e logE n + e logEo
[B1 ]69 =
8 12 13 e 13 (logq+log plog pi )
1 12 e 13 (logq+log plog pi )
27 e logEo
[B2 ]69 = 27 ,
[B2 ]6,14 = 24 ,
257
,
[B2 ]71 = 16 21 e (logBlogM) ,
[B2 ]73 = 16 20 e (log p+logQ logM3 ) + 20 P e (log plogM3 )
+ 21 e (logBlogM) ,
where 3 = 32T ,
[B2 ]7,10 = 16 18 ,
[B2 ]8,15 = 13 ,
[C 2 ]41 = 3 ,
[C 2 ]51 = 11 ,
[C 2 ]73 = 16 20 e (log p+logQ logM3 ) + P e (log plogM3 ) ,
[C 2 ]81 = 14 ,
[C 2 ]82 = 14 ,
!
"
[b1 ]1 = 1 + 2 + 1 log1 + 1 log e logQ + P
logQ e logQ
P
1
,
1
e logQ + P
e logQ + P
[b1 ]2 = 1 + 2 + 19 log22 ,
[b1 ]3 = 20 25 e logQ {1 logQ },
[b1 ]4 = 1 + 2 + 17 log12 + 18 log14
+ 17 log{1 + 14 (1 + 2 )} + (17 + 18 )
log e logC +e logGc +e logK DlogK +e logK h DlogK h
+ e logK p DlogK p +e logEn +e logEo (17 + 18 )
258
logE n
logE o
+e
logE n +e
logE o
,
e logC +e logGc +e logK DlogK +e logK h DlogK h
+ e logK p DlogK p + e logEn + e logE o
259
260
[b2 ]4 = 2 2 + 3 log4
+
3
log e 6 logQ 5 e 6 logK
6
3 e 6 logQ
e 6 logQ 5 e 6 logK
3 5 e 6 logK
logQ
e 6 logQ 5 e 6 logK
logK ,
1 5 e 6 (logQ logK )
[b2 ]6 = 25 log32 26 33
+ 27 log e logEn + e logE o + P F DKa
27 e logEn logE n + e logE o logE 0 + P F DKa
,
261
logQ
e 6 logQ 5 e 6 logK
14 5 e 6 logK
logK .
+
e 6 logQ 5 e 6 logK
Appendix B: Data
Sources of data:
BLUE United Kingdom National Accounts: The Blue
Book, 1997, ONS
BOE
Bank of England
ET
Economic Trends
ETAS
FS
Financial Statistics
IFS
MD
262
The series used in this study consisted of quarterly observation for the period 197496 and were defined as follows:
(C ) Real Private Consumption
Consumers expenditure at current prices deflated by gross
domestic product implicit price deflator, p, as defined below
(source: ETAS).
(E n ) Real Non-Oil Exports
Exports of goods and services less oil exports at current prices
deflated by p (sources: ETAS and ET).
(F ) Real Current Transfers Abroad
Net government and private transfers abroad deflated by p
(source: ETAS).
(I ) Volume of Imports
Imports of goods and services at 1990 market prices (source:
ETAS).
(Kh ) Residential Fixed Capital
Cumulative net real residential fixed capital formulation
since the beginning of 1975 plus a constant derived from
the official estimate of the capital stock for dwellings at the
end of 1996 adjusted to 1990 prices. Net real residential fixed
capital formulation equals gross residential fixed capital formulation at current prices deflated by p minus depreciation
at current replacement for residential dwellings deflated by
p (sources: ETAS and BLUE).
263
264
( p) Price Level
Gross domestic product at market prices divided by
gross domestic product at 1990 market prices (source:
ETAS).
(Q ) Real Net Output
Net domestic product at current prices deflated by p (source:
ETAS).
(q) Exchange Rate (price of sterling in foreign
currency)
Sterling effective exchange rate average (1990 = 1.0000)
(source: BOE).
(r ) Interest Rate
Yield on long-dated UK government securities adjusted to
instantaneous rate per quarter (source: ETAS).
(S) Inventories
Value of real physical increase in stocks at current prices
since the beginning of 1975 plus a constant derived from
the official estimate of the book value of stocks and work
in progress held at the end of 1996 adjusted to 1990 prices.
(sources: ETAS and BLUE).
(w) Wage Rate
Income from employment (including pay of forces) divided
by sum of employees in employment and HM forces
(sources: ETAS and MD).
265
(B) Bonds
Sterling debt (British government stock) held by the
domestic private sector, million, stock outstanding (source:
BOE).
(E o ) Real Oil Exports
Oil exports at current prices deflated by p (source: ET).
(Gc ) Real Government Consumption
Central government consumption deflated by p plus local
authorities consumption deflated by p (source: ETAS).
(K p ) Public Non-Residential Fixed Capital
Cumulative net real public non-residential fixed capital formulation since the beginning of 1975 plus a constant derived
from the official estimate of the capital stock for public nonresidential fixed capital at the end of 1996 adjusted to 1990
prices. Net real public non-residential fixed capital formulation equals gross public non-residential fixed capital formulation at current prices deflated by p minus depreciation at
current replacement for public non-residential investment
deflated by p (sources: ETAS and BLUE).
( pf ) Price Level in Leading Foreign Industrial
Countries
Average of implicit price deflator of US GDP and implicit
price deflator of German GDP (source: IFS).
266
267
References
Adkins, L. C. and Krehbiel, T. [1999]: Mean Reversion and Volatility of Short-Term London Interbank Offer Rates. International
Review of Economics and Finance 8, 4554.
Agbeyegbe, T. D. [1984]: The Exact Discrete Analog to a Closed
Linear Mixed-Order System. Journal of Economics Dynamics and
Control 7, 363375.
Agbeyegbe, T. D. [1987]: The Exact Discrete Analog to a Closed
Linear First-Order System with Mixed Sample. Econometric
Theory 3, 142149.
Agbeyegbe, T. D. [1988]: An Exact Discrete Analog of an Open
Order Linear Non-Stationary First-Order Continuous-Time
System with Mixed Sample. Journal of Econometrics 39, 237
250.
At-Sahalia, Y. [1996]: Testing Continuous-Time Models of the
Spot Interest Rate. Review of Financial Studies 9, 385342.
Andersen, T. G. and Lund, J. [1997]: Estimating Continuous-Time
Stochastic Volatility Models of the Short-Term Interest Rate.
Journal of Econometrics 77, 343377.
269
References
Babbs, S. H. and Nowman, K. B. [1999]: Kalman Filtering of Generalized Vasicek Term Structure Models. Journal of Financial and
Quantitative Analysis 34, 115130.
270
References
271
References
272
References
273
References
Chambers, M. J. [1999]: Discrete Time Representation of Stationary and Non-Stationary Continuous Time Systems. Journal of
Economic Dynamics and Control 23, 619639.
Chambers, M. J. [2001]: Temporal Aggregation and the Finite
Sample Performance of Spectral Regression Estimators in Cointegrated Systems: A Simulation Study. Econometric Theory 17,
591607.
Chambers, M. J. [2003]: The Asymptotic Efficiency of Cointegration Estimators under Temporal Aggregation. Econometric Theory
19, 4977.
Chen, L. [1996]: Interest Rate Dynamics, Derivatives Pricing, and Risk
Management. Springer, Lecture Notes in Economics and Mathematical Systems.
Chen, R. R. and Scott, L. [1992]: Pricing Interest Rate Options in
a Two-Factor CoxIngersollRoss Model of the Term Structure.
Review of Financial Studies 5, 613636.
Coddington, E. A. and Levinson, N. [1955]: Theory of Ordinary Differential Equations. New York: McGraw Hill.
Corradi, V. [1997]: Comovements between Diffusion Processes:
Characterization, Estimation, and Testing. Econometric Theory 13,
646666.
Courant, R. [1936]: Differential and Integral Calculus. London:
Blackie and Sons Ltd., vol. II.
Cox, J. C. [1975]: Notes on Option Pricing 1: Constant Elasticity of
Variance Diffusion. Working Paper, Stanford University.
Cox, J. C. and Ross, S. [1976]: The Valuation of Options for Alternative Stochastic Processes. Journal of Financial Economics 3, 145
166.
Cox, J. C., Ingersoll, J. E. and Ross, S. A. [1985]: A Theory
of the Term Structure of Interest Rates. Econometrica 53, 385
407.
274
References
Cox, J. C., Ingersoll, J. E. and Ross, S. [1980]: An Analysis of Variable Rate Loan Contracts. Journal of Finance 35, 389403.
Dahlquist, M. [1996]: On Alternative Interest Rate Processes. Journal of Banking and Finance 20, 10931119.
Dothan, U. L. [1978]: On the Term Structure of Interest Rates.
Journal of Financial Economics 6, 5969.
Donaghy, K. P. [1993]: A Continuous-Time Model of the United
States Economy. In G. Gandolfo (Ed.), Continuous Time Econometrics. London: Chapman and Hall.
Duffie, D. and Kan, R. [1996]: A Yield-Factor Model of Interest
Rates. Mathematical Finance 6, 379406.
Durbin, J. [1961]: Efficient Fitting of Linear Models for Continuous
Time Stationary Time Series from Discrete Data. Bulletin of the
International Statistical Institute 38, 273282.
Durbin, J. [1963]: Trend Elimination for the Purpose of Estimating Seasonal and Periodic Components of Time Series. In M.
Rosenblatt (Ed.), Time Series Analysis. New York: John Wiley.
Einstein, A. [1906]: Zur Theorie Der Brownschen Bewegung.
Annalen Der Physik 19, 371381.
Episcopos, A. [2000]: Further Evidence on Alternative Continuous
Time Models of the Short-Term Interest Rate. Journal of International Financial Markets, Institutions and Money 10, 199212.
Engle, R. F. and Granger, C. W. J. [1987]: Cointegration and Error
Correction: Representation Estimation and Testing. Econometrica
55, 251276.
Franses, P. H. [1996]: Periodicity and Stochastic Trends in Economic
Time Series. Oxford: Oxford University Press.
Gandolfo, G. [1981]: Quantitative Analysis and Econometric Estimation
of Continuous Time Dynamic Models. Amsterdam: North-Holland.
Gandolfo, G. and Padoan, P. C. [1982]: Policy Simulations
with a Continuous Time Macrodynamic Model of the Italian
275
References
276
References
277
References
Kawai, K. and Maekawa, K. [2003]: A Note on Yu-Phillips Estimation for a Continuous Time Model of a Diffusion Process. Unpublished
Paper, Hiroshima University.
Kirkpatrick, G. [1987]: Employment, Growth and Economic Policy: An
278
References
Lucas, R. E. [1976]: Econometric Policy Evaluation: A Critique. Carnegie-Rochester Conference Series on Public Policy 1,
1946.
Malinvaud, E. [1966]: Statistical Methods of Econometrics. NorthHolland, Amsterdam.
Melino, A. [1994]: Estimation of Continuous-Time Models in
Finance. In C. A. Sims (Ed.), Advances in Econometrics: Sixth World
Congress. Cambridge: Cambridge University Press, vol. II.
Merton, R. C. [1969]: Lifetime Portfolio Selection under Uncertainty: The Continuous Time Case. Review of Economics and
Statistics 51, 247257.
Merton, R. C. [1971]: Optimum Consumption and Portfolio Rules
in a Continuous Time Model. Journal of Economic Theory 3, 373
413.
Merton, R. C. [1973a]: Theory of Rational Option Pricing. Bell
Journal of Economics 4, 141183.
Merton, R. C. [1973b]: An Intertemporal Capital Asset Pricing
Model. Econometrica 41, 867887.
Merton, R. C. [1990]: Continuous Time Finance. New York: Oxford
University Press.
McCrorie, J. R. [2000]: Deriving the Exact Discrete Analog of a
Continuous Time System. Econometric Theory 16, 9981015.
McCrorie, J. R. [2001]: Interpolating Exogenous Variables in Open
Continuous Time Dynamic Models. Journal of Economic Dynamics
and Control 25, 13991427.
McCrorie, J. R. [2002]: The Likelihood of the Parameters of a Continuous Time Vector Autoregressive Model. Statistical Inference for
Stochastic Processes 5, 273286.
McGarry J. S. [2003]: The Exact Discrete Time Representation of
a System of Fourth-Order Differential Equation. Computers and
Mathematics with Applications 46, 213230.
279
References
Niewenhuis, H. J. [1995]: Continuous Time Macroeconometric Modelling with An Application to the Dutch Economy. Den Haag:
Labyrinth.
Nowman, K. B. [1991]: Open Higher-Order Continuous Time
Dynamic Model with Mixed Stock and Flow Data and Derivatives of Exogenous Variables. Econometric Theory 7, 404
408.
Nowman, K. B. [1993]: Finite-Sample Properties of the Gaussian
Estimation of an Open Higher-Order Continuous-Time Dynamic
Model with Mixed Stock and Flow Data. In G. Gandolfo (Ed.),
Continuous Time Econometrics: Theory and Applications. London:
Chapman-Hall.
Nowman, K. B. [1996]: A Note on Continuous Time Dynamic
Disequilibrium Macroeconometric Modelling of The United
Kingdom. In W. A. Barnett, G. Gandolfo and C. Hillinger
(Eds.), Dynamic Disequilibrium Modelling. Cambridge: Cambridge
University Press.
Nowman, K. B. [1997]: Gaussian Estimation of Single-Factor Continuous Time Models of the Term Structure of Interest Rates.
Journal of Finance 52, 16951706.
Nowman, K. B. [1998a]: Continuous-Time Short Term Interest
Rate Models. Applied Financial Economics 8, 401407.
Nowman, K. B. [1998b]: Econometric Estimation of a Continuous Time Macroeconomic Model of the United Kingdom with
Segmented Trends. Computational Economics 12, 243254.
Nowman, K. B. and Sorwar, G. [1999]. An Evaluation of Contingent Claims Using the CKLS Interest Rate Model: An Analysis
of Australia, Japan, and United Kingdom. Asia Pacific Financial
Markets 6, 205219.
Nowman, K. B. [2001]: Gaussian Estimation and Forecasting of
Multi-Factor Term Structure Models with an Application to
280
References
281
References
Saltoglu,
B. [2003]: Comparing Forecasting Ability of Parametric and Nonparametric Methods: Application with Canadian
Monthly Interest Rates. Applied Financial Economics 13, 169176.
Sargan, J. D. [1974]: Some Discrete Approximations to Continuous Time Stochastic Models. Journal of the Royal Statistical Society,
Series B, 36, 7490.
282
References
Sargan, J. D. [1976]: Some Discrete Approximations to Continuous Time Stochastic Models. In A. R. Bergstrom (Ed.), Statistical
Inference in Continuous Time Economic Models. Amsterdam: NorthHolland.
Sassanpour, C. and Sheen, J. [1984]: An Empirical Analysis of the
Effect of Monetary Disequilibrium in Open Economies. Journal
of Monetary Economics 13, 127163.
Simos, T. [1996]: Gaussian Estimation of a Continuous Time
Dynamic Model with Common Stochastic Trends. Econometric
Theory 12, 361373.
B. [1993]: CONTIMOS A Continuous Time Econometric
Sjo o,
Model for Sweden Based on Monthly Data. In G. Gandolfo (Ed.),
Continuous Time Econometrics. London: Chapman and Hall.
Stanton, R. [1997]: A Nonparametric Model of Term Structure
Dynamics and the Market Prices of Interest Rate Risk. Journal of
Finance 52, 19732002.
Stavrev, E. [2001]: A Small Continuous Time Macro-Econometric
Model of the Czech Republic. Empirical Economics 26, 673705.
Stefansson, S. B. [1981]: Inflation and Economic Policy in a Small Open
Economy: Iceland in the Post-War Period. Ph.D. Thesis, University
of Essex.
Sundaresan, S. M. [2000]: Continuous-Time Methods in Finance:
A Review and an Assessment. Journal of Finance 55, 15691622.
Tse, Y. K. [1995]: Some International Evidence on the Stochastic Behavior of Interest Rates. Journal of International Money and
Finance 14, 721738.
Tullio, G. [1981]: Demand Management and Exchange Rate Policy:
The Italian Experience. IMF Staff Papers 28, 80117.
Van Loan C. F. [1978]: Computng Integrals Involving the Matrix
Exponential. IEEE Transactions on Automatic Control AC-23, 396
404.
283
References
284
Author Index
Adkins, L. C., 21
Agbeyegbe, T. D., 14, 51
At-Sahalia, Y., 28
Andersen, T. G., 29
285
Author Index
Henry, S. G. B., 56
He, Y., 43, 195, 196
Hiraki, T., 21
Houthakker, H. S., 10
Hull, J., 29
Ingersoll, J. E., 19
Ito, K., 2
Jiang, G., 28
Johansen, S., 46, 66
Jonson, P. D., 31, 42, 115
Kan, R., 29
Karolyi, G. A., 19
Kawai, K., 27
Kirkpatrick, G., 31, 42, 115
Knight, M. D., 31, 114, 115
Knight, J., 28
Kondratieff, N. D., 239
Koopmans, T. C., 3
Krehbiel, T., 21
Kydland, F. E., 196
Langetieg, T. C., 29
Levinson, N., 35, 174, 191
Longstaff, F. A., 19, 29
Lo, K. M., 27
Lucas, R. E., 43
Lund, J., 29
Malinvaud, E., 29
Maekawa, K., 27
Mathieson, D. J., 31, 115
McCrorie, J. R., 14
McGarry J. S., 14
Mckibbin, W. J., 42
Melino, A., 20
Merton, R. C., 2, 7, 18, 19
Moses, E. R., 31, 115
Niewenhuis, H. J., 31, 115
guez, T. M., 27
N
Padoan, P.C., 31, 42, 115
Park, J. Y., 46, 66, 72
Petit, M. L., 42
Peters, S., 56
Phillips, A. W., 3, 17, 42
Phillips, P. C. B., 11, 12, 16, 17,
24, 26, 27, 28, 29, 31, 35, 41,
46, 66, 72, 115
Prescott, E. C., 196
Revuz, D., 27
Ross, S., 19
Saltoglu,
B., 31
Sanders, A. R., 19
Sargan, J. D., 10
Sargent, T. J., 41
Sassanpour, C., 42
Scholes, M. 2, 7, 18
Schwartz, E. S., 19, 29, 31
Scott, L., 29
Sheen, J., 42
Simos, T., 46
B., 31, 115
Sjo o,
Sorwar, G., 28
Stanton, R., 28
Stavrev, E., 32, 115
Stefansson, S. B., 42
Stock, J. H., 18, 46, 56, 57
Sundaresan, S. M., 7, 18, 19
286
Author Index
Takezawa, N., 21
Taylor, L. D., 10
Tenney, M. S., 29
Trevor, R. G., 42
Tse, Y. K., 21
Tullio, G., 42
Wiener, N., 2
Wren-Lewis, S., 56
Wymer, C. R., 10, 31, 32, 34,
35, 36, 41, 42, 44, 45, 48,
58, 114, 115, 121, 123, 124,
125, 126, 132, 152, 164, 173,
174, 192, 194, 195, 196, 218,
223
Yor, M., 27
Yu, J., 26, 27, 28
Zadrozny, P. A., 18
287
Subject Index
aliasing, 41
approximate discrete model, 25,
27
asymptotic stability, 35, 174,
197, 238
autocovariance, 48, 85
Bank of England, 30, 262
Bank of Japan, 30
bifurcation, 43, 174, 192, 195,
196
bond, 2, 19, 2829, 117, 127,
135, 144145, 148149,
266267
Brownian motion, 19, 27, 30,
56, 69, 72, 74, 76, 80, 217,
221
(BRSC), 26, 28
capital, 4, 18, 3334, 39, 58,
115118, 128136, 142,
149150, 222, 224225,
236237, 263264, 266
288
Subject Index
identification, 4142, 44
imports, 4, 34, 3940, 58,
115117, 136137,
145146, 150, 222, 232,
263, 267
income, 1719, 28, 33, 40,
117118, 126127,
144149, 248, 265, 267
indirect taxation, 40, 118, 138,
267
inflation, 37, 45, 48, 118, 127,
134135, 141, 143,
148149, 222
interest rate, 1834, 40, 115,
117, 127, 131, 135136,
144145, 148151, 180,
197, 225, 242
investment, 39, 116117,
135, 147148, 225, 264,
266
kalman filter, 18, 30
labour, 32, 37, 4445, 118,
123, 131134, 139143,
194, 222, 232233, 236,
264
Lag distributions, 13, 58, 214,
223224, 226, 248
Lebesgue measure, 54
Lucas critique, 196
macroeconometric, 31, 3436,
43, 4647, 51, 71, 114,
124, 173, 196197, 213,
247
mean reversion, 1920, 27
289
Subject Index
Poincare-Liapounov-Perron,
35,
174, 185, 191
290