You are on page 1of 43

Models of Imperfect Information in

Macroeconomics

Jonas Dovern

February 14, 2018

Jonas Dovern CES Lecture 2 February 14, 2018 1 / 38


Brief Review of Previous Lecture

Role of expectations for macroeconomic dynamics already


discussed by Arthur C. Pigou in 1927

Also central to Keynes’ and Friedman’s work

First formal models in 1950/60s:


I Adaptive expectations
I Rational expectations
I Both extremes not very helpful to describe reality

A wide range of modern theories of expectation formation (eg,


Mankiw and Reis, 2018):
I Today: focus on models with imperfect information structures (but
otherwise rational expectations)

Jonas Dovern CES Lecture 2 February 14, 2018 2 / 38


Where Do We Stand Today?

FIRE assumption merely a useful benchmark (although, of course,


many DSGE models used in practice still build on it)

Micro evidence on expectations:


I All kinds of rigidities
I Heterogeneity of expectations

Advances in solution techniques for DSGE models, in MCMC


techniques, and in computing power:
I Easier to work with more complex information structures
I Eg higher order expectations

Jonas Dovern CES Lecture 2 February 14, 2018 3 / 38


Research on Expectations Today

Source: R. Reis at ASSA Meetings 2018.

Jonas Dovern CES Lecture 2 February 14, 2018 4 / 38


Agenda

1 Sticky Information Models

2 Noisy Information Models

3 Rational Inattention Models

4 Policy Implications

Jonas Dovern CES Lecture 2 February 14, 2018 5 / 38


Sticky Information Models

Jonas Dovern CES Lecture 2 February 14, 2018 6 / 38


Sticky Prices vs. Sticky Information
Sticky prices
Info Info Info Info Info Info Info . . .

t
t=0 t=2 t=4 t=6 t=8 t=10

∆P ∆P ∆P ∆P ...

Jonas Dovern CES Lecture 2 February 14, 2018 7 / 38


Sticky Prices vs. Sticky Information
Sticky prices
Info Info Info Info Info Info Info . . .

t
t=0 t=2 t=4 t=6 t=8 t=10

∆P ∆P ∆P ∆P ...

Sticky information
Info Info Info

t
t=0 t=2 t=4 t=6 t=8 t=10

∆P ∆P ∆P ∆P ∆P . . .
Jonas Dovern CES Lecture 2 February 14, 2018 7 / 38
Origins of the Model

Mankiw and Reis (2002) propose to replace the New Keynesian


Phillips Curve by a sticky information version
The NKPC:
pt∗ = pt + αyt

xt = λ ∑ (1 − λ)j Et pt∗+j
h =0

pt = λ ∑ ( 1 − λ ) j xt − j
h =0

πt = αλ2 /(1 − λ) yt + Et πt +1
 

Jonas Dovern CES Lecture 2 February 14, 2018 8 / 38


Origins of the Model
The SIPC:

pt∗ = pt + αyt

xtj = Et −j pt∗


∑ ( 1 − λ ) j xt
j
pt = λ
h =0

πt = [αλ/(1 − λ)] yt + λ ∑ (1 − λ)j Et −1−j [πt + α∆yt ]
j =0

Jonas Dovern CES Lecture 2 February 14, 2018 9 / 38


Origins of the Model
The SIPC:

pt∗ = pt + αyt

xtj = Et −j pt∗


∑ ( 1 − λ ) j xt
j
pt = λ
h =0

πt = [αλ/(1 − λ)] yt + λ ∑ (1 − λ)j Et −1−j [πt + α∆yt ]
j =0

“In the standard sticky-price model, current expectations of future economic


conditions play an important role [. . . ]. In this sticky-information model [. . . ]
the relevant expectations are past expectations of current economic
conditions.” (Mankiw and Reis, 2002, p. 1300)
Jonas Dovern CES Lecture 2 February 14, 2018 9 / 38
Microfoundations

Why do firms set prices based on outdated information?

Observing something is easy . . . figuring out what it means is hard:


I Eg firms observe monetary policy but don’t have resources to
think about the consequences

Also, there might be cost of acquiring updated information ⇒


optimal for agents to do this infrequently (Reis, 2006):
I This can be both in terms of money and time

Jonas Dovern CES Lecture 2 February 14, 2018 10 / 38


Extensions of the Model

Is a constant λ reasonable?
What about:
I Changing cost of acquiring information?
I Changing volatility of shocks?

Many authors have argued that λ is actually time varying:


I Carroll (2003): Households respond to news about inflation
I Dovern (2013): Professional forecasters more likely to update if far
away from the crowd
I Dräger (2016): Agents switch between costly FIRE and cheaper SI
expectation

Jonas Dovern CES Lecture 2 February 14, 2018 11 / 38


Extensions of the Model

Is a constant λ reasonable?
What about:
I Changing cost of acquiring information?
I Changing volatility of shocks?

Many authors have argued that λ is actually time varying:


I Carroll (2003): Households respond to news about inflation
I Dovern (2013): Professional forecasters more likely to update if far
away from the crowd
I Dräger (2016): Agents switch between costly FIRE and cheaper SI
expectation

Andrade et al. (2016) argue that it might be different for different


variables

Jonas Dovern CES Lecture 2 February 14, 2018 11 / 38


Carroll’s (2003) Model
Households update their expectations infrequently towards the
available professional forecasts
Correlation between frequency of inflation news and gap between
two types of expectations is significantly different from 0

Figure 1: Inflation Versus News Stories, and Michigan Versus SPF Forecasts
21

Jonas Dovern CES Lecture 2 February 14, 2018 12 / 38


Andrade et al.’s (2016) Model
P. Andrade et al. / Journal of Monetary Economic

Fig. This figure shows selected statistics


Jonas1.Dovern CES Lecturefor
2 forecast disagreement from
February 14,the
2018Blue 13
Chip
/ 38Fin
Andrade et al.’s (2016) Model
The model is given by:
 
φ (IK − φ)
Xt = Xt −1 + ε t ,
0 IK

with Xt = [xt1 , . . . , xtK , µ1t , . . . , µK 0


t ]

Jonas Dovern CES Lecture 2 February 14, 2018 14 / 38


Andrade et al.’s (2016) Model
The model is given by:
 
φ (IK − φ)
Xt = Xt −1 + ε t ,
0 IK

with Xt = [xt1 , . . . , xtK , µ1t , . . . , µK 0


t ]

Under FI agents observe yt = [IK 0K ]Xt every period


Here, agents observe the kth element of yt with probability λk :
I Frequency is constant across agents
I Agents do not observe the entire history (tractable . . . but
realistic?)

Model can generate manifold shapes of the “term structure of


disagreement”

Jonas Dovern CES Lecture 2 February 14, 2018 14 / 38


Andrade et al.’s (2016) Model

P. Andrade et al. / Journal of Monetary Economics 83 (2016) 106–128 119

Jonas Dovern CES Lecture 2 February 14, 2018 15 / 38


Alternative: Lahiri and Sheng (2008)
Lahiri and Sheng (2008) propose a different model to generate
long-run disagreement
Initial prior for long-run growth Fi,t,H with a certain precision:
I Heterogeneity of forecasting models
I Judgmental component

Different interpretation of public signals in every period:


Yi,t,h = Lt,h − ε i,t,h with ε i,t,h ∼ N (µi,t,h , 1/bi,t,h )

Bayesian updating of forecasts:

Fi,t,h = λFi,t,h+1 + (1 − λ) (Lt,h − µi,t,h )


ai,t,h+1
with λ = ai,t,h+1 +bi,t,h

Jonas Dovern CES Lecture 2 February 14, 2018 16 / 38


Why Is It Not Used More?
Macro evidence suggest that sticky information models are not
better than conventional NK models (e.g. Coibion, 2010)
Micro evidence about expectations suggest a more frequent
updating than required by SI models to match aggregate features
of the data (e.g. Andrade and Le Bihan, 2013; Dovern, 2013;
Dovern et al., 2015; Binder, 2017)
Technical aspect:
I SI model requires to take into account an infinite number of states
(all past expectations)
I Truncation possible but still nasty when estimating/solving DSGE
models (Trabandt, 2007)

Major aspect in favor of SI model is time-variation of disagreement


. . . we’ll come back to that later

Jonas Dovern CES Lecture 2 February 14, 2018 17 / 38


Noisy Information Models

Jonas Dovern CES Lecture 2 February 14, 2018 18 / 38


The Idea of Noisy Information
Info Info Info Info Info Info

+Noise +Noise +Noise +Noise +Noise +Noise

t
t=0 t=1 t=2 t=3 t=4 t=5

Jonas Dovern CES Lecture 2 February 14, 2018 19 / 38


A General Noisy Information Model
The “island economies” in Phelps (1970) and Lucas (1972, 1973)

The model is given by:


 
φ (IK − φ)
Xt = Xt −1 + ε t
0 IK
Yj,t = [IK 0K ]Xt + ηj,t

with Xt = [xt1 , . . . , xtK , µ1t , . . . , µK 0


t ] and ε t and ηj,t independent

Noise shocks are also independent across agents:


I Agents who track different indicators
I Private information

Jonas Dovern CES Lecture 2 February 14, 2018 20 / 38


The Signal-to-Noise Ratio

Key parameter in the model:


I The ratio of Var [ ε ] to Var [ η ]
t i,j
I σ 2 /σ 2 → ∞ ⇒ FI model
ε η
I σ 2 /σ 2 → 0 ⇒ Ignorant agents always expect µk
ε η t

The ratio determines to what extent agents incorporate news into


their expectations

Here, we take these parameters as exogenous:


I Data uncertainty
I Data interpretation skills

Jonas Dovern CES Lecture 2 February 14, 2018 21 / 38


The Signal-to-Noise Ratio
20

20

20
10

10

10
0

0
-10

-10

-10
-20

-20

-20
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200

Jonas Dovern CES Lecture 2 February 14, 2018 22 / 38


The Signal-to-Noise Ratio
20

20

20
10

10

10
0

0
-10

-10

-10
-20

-20

-20
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200

Jonas Dovern CES Lecture 2 February 14, 2018 22 / 38


Extensions of the Model
Is the signal-to-noise ratio the same for every agent?

Intuition tells otherwise:


I Resources might differ across agents
I Skill of interpreting the data might differ (Lahiri and Sheng, 2008)

In Dovern (2015), I propose to model this ratio as a distribution


across agents:
I Can help matching the persistence of disagreement across agents
I Can account for the positive correlation between disagreement and
forecast performance (Dovern and Hartmann, 2017)

See also Coibion and Gorodnichenko (2012) for extensions of the


model (eg strategic interaction between agents)

Jonas Dovern CES Lecture 2 February 14, 2018 23 / 38


ExtensionsJ. Dovern
of the Model
/ European Economic Review 80 (2015) 16–35

Rank persistence (W) 1

0.75

0.5

0.25

0
-6

-4

-2

25

75

5
0.

0.

1.

2.
1e

1e

1e

1.

1.
2
Variance (σδ )

unction of the signal-to-noise ratio variance. Notes: The rank persistence (coefficient
Source: Dovern of concord
(2015).
odel given by Eqs. (4)–(6) and (7) using the parameterization given in Table 1 of Andrade et al. (2
tion iterations;
Jonas the dashed lines represent CES
Dovern theLecture
corresponding
2 2.5th and February
97.5th14,
percentiles.
2018 24 / 38
Extensions of the Model
Author's personal copy
70 J. Dovern, M. Hartmann

3.5 3.5

3 3

2.5 2.5

MSFE
2 2

1.5 1.5

1 1
0 0.5 1 1.5 2 0 1 2 3 4
E[d]
σε2 / ση2

0 0.5 1 1.5 2
4

3
σε2 / ση2

Fig. 2 MSFEs and average disagreement levels under heterogeneous signal-to-noise ratios
Source: Dovern and Hartmann (2017).
between disagreement and forecast performance. It is evident that the model predicts
a positive and linear relation between those two variables.
Jonas Dovern So far, we have concentrated on one-step-ahead
CES Lecture 2forecasts. But, of course, in reality
February 14, 2018 25 / 38
Rational Inattention Models

Jonas Dovern CES Lecture 2 February 14, 2018 26 / 38


What If We Could Choose the Noise Level?

Info Info Info Info Info Info

+Noise +Noise +Noise +Noise +Noise +Noise


t
t=0 t=1 t=2 t=3 t=4 t=5

Jonas Dovern CES Lecture 2 February 14, 2018 27 / 38


Rational Inattention
The idea: Agents have a limited capacity to process information

Sims (2003) models this limited attention as a constraint on


information flows:
I Decision makers have to optimize subject to this constraint

Other important references are Woodford (2003) and a number of


papers by Bartosz Maćkowiak and Mirko Wiederholt

The resulting model mechanism is very similar to the NI model

Key differences:
I Signals can also be about past states of the world
I Signals are chosen endogenously
I The signal-to-noise ratio is also chosen endogenously

Jonas Dovern CES Lecture 2 February 14, 2018 28 / 38


The Basic Model
Data follows AR(p) process:

Xt = φ1 Xt −1 + . . . + φp Xt −p + ε t

Agents receive a K-dimensional signal vector:

StK = AXtM + ηtK ,

with XtM = [Xt , . . . , Xt −M +1 ]0

Agents choose K , A, and Ση which denotes the covariance of ηt :


I Interestingly, one can always find an optimal signal with K = 1

Information sets are defined by It = I0 ∪ {S1K , . . . , StK }

Jonas Dovern CES Lecture 2 February 14, 2018 29 / 38


Optimization

min E (Xt − E [Xt |It ])2


 
K ,A,Ση
1
s.t. lim I (X0 , ε 1 , . . . , ε T ; S1K , . . . , STK ) ≤ κ
T →∞ T

Here I (X T ; S T ) equals the difference between unconditional


entropy (“uncertainty”) and conditional entropy:

I (X T , S T ) = H (X T ) − H (X T |S T )
limT →∞ (1/T ) H (X0 , ε t , . . . , ε T ) uncertainty without signals
limT →∞ (1/T ) H (X0 , ε t , . . . , ε T |S1K , . . . , STK ) uncertainty with
signals
Difference measures flow of information to the agent
Jonas Dovern CES Lecture 2 February 14, 2018 30 / 38
Extensions

Also the RI model can be extended

One example is the model in Gorodnichenko (2008) in which there


is strategic interaction across firms:
I Information acquisition for optimal price setting is costly to a firm
I New price is a costless information signal to other firms
I Consequence: Rigidities in price setting without real frictions

Jonas Dovern CES Lecture 2 February 14, 2018 31 / 38


Policy Implications

Jonas Dovern CES Lecture 2 February 14, 2018 32 / 38


Selected Implications
Delayed response of inflation to monetary policy shocks and
persistence of inflation is accounted for (without price indexation)

Complete price stabilization becomes optimal after mark-up shocks


when firms can chose how much attention to pay to aggregate
information in a RI model (Paciello and Wiederholt, 2014):
I Trade-off between reducing inefficient price dispersion and
increasing consumption variance in NK model and NI model
I RI introduces additional effect that dissolves trade-off; reducing
price dispersion → fewer incentives for firms to pay attention to
mark-up shocks → less price increase → less consumption variance

Price level targeting (and no longer inflation targeting) becomes


optimal in a SI model (Ball et al., 2005)

Jonas Dovern CES Lecture 2 February 14, 2018 33 / 38


Signaling Effect of Monetary Policy

Melosi (2017) uses a NI-DSGE model to analyze the effects of


signals that central banks send with their price setting:
I Both firms and central bank act on imperfect information
I Interest rate is observed without error ⇒ non-polluted signal
about central banks view on the economy

Signalling channel:
I Tightening = exogenous MP shock
I Tightening = response to negative supply shock or positive
demand shock

Model identifies signalling channel as major force behind high


inflation (expectations) in the 1970s

Jonas Dovern CES Lecture 2 February 14, 2018 34 / 38


The Signalling Channel
MELOSI SIGNALLING EFFECTS OF MONETARY POLICY 875

Figure 7 Source: Melosi (2017).


Signaling effects of monetary policy on inflation and inflation expectations
Jonas Dovern CES Lecture 2 February 14, 2018 35 / 38
Summary
Contemporary macroeconomic models employ a number of
complex information structures
I Sticky information
I Noisy information/rational inattention
I Not covered: learning models, agent-based models with heuristic
expectations (for more, see Woodford, 2013)

This type of friction more consistent with micro evidence than


price rigidities coupled with price indexation
Most models imply very similar dynamics of average expectations
⇒ next lecture
Current and future research program:
I More heterogeneity
I More state-dependence (eg Nimark, 2014)
I More micro evidence ⇒ next lecture
Jonas Dovern CES Lecture 2 February 14, 2018 36 / 38
References I
Andrade, P., Crump, R. K., Eusepi, S., and Moench, E. (2016). Fundamental disagreement. Journal of Monetary Economics,
83(C):106–128.
Andrade, P. and Le Bihan, H. (2013). Inattentive professional forecasters. Journal of Monetary Economics, 60(8):967–982.
Ball, L., Gregory Mankiw, N., and Reis, R. (2005). Monetary policy for inattentive economies. Journal of Monetary Economics,
52(4):703–725.
Binder, C. (2017). Consumer forecast revisions: Is information really so sticky? Economics Letters, 161:112–115.
Carroll, C. D. (2003). Macroeconomic expectations of households and professional forecasters. Quarterly Journal of Economics,
118(1):269–298.
Coibion, O. (2010). Testing the sticky information Phillips curve. The Review of Economics and Statistics, 92(1):87–101.
Coibion, O. and Gorodnichenko, Y. (2012). What can survey forecasts tell us about information rigidities? Journal of Political
Economy, 120(1):116 – 159.
Dovern, J. (2013). When are GDP forecasts updated? Evidence from a large international panel. Economics Letters,
120(3):521–524.
Dovern, J. (2015). A multivariate analysis of forecast disagreement: Confronting models of disagreement with survey data.
European Economic Review, 80(C):16–35.
Dovern, J., Fritsche, U., Loungani, P., and Tamirisa, N. T. (2015). Information rigidities: Comparing average and individual
forecasts for a large international panel. International Journal of Forecasting, 31(1):144–154.
Dovern, J. and Hartmann, M. (2017). Forecast performance, disagreement, and heterogeneous signal-to-noise ratios. Empirical
Economics, forthcoming.
Dräger, L. (2016). Recursive inattentiveness with heterogeneous expectations. Macroeconomic Dynamics, 20(4):1073–1100.
Gorodnichenko, Y. (2008). Endogenous information, menu costs and inflation persistence. NBER Working Papers 14184,
National Bureau of Economic Research, Inc.
Lahiri, K. and Sheng, X. (2008). Evolution of forecast disagreement in a Bayesian learning model. Journal of Econometrics,
144(2):325–340.

Jonas Dovern CES Lecture 2 February 14, 2018 37 / 38


References II
Lucas, R. J. (1972). Expectations and the neutrality of money. Journal of Economic Theory, 4(2):103–124.
Lucas, R. J. E. (1973). Some international evidence on inflation-output tradeoffs. American Economic Review, 63:326–334.
Mankiw, N. G. and Reis, R. (2002). Sticky information versus sticky prices: A proposal to replace the new Keynesian Phillips
curve. Quarterly Journal of Economics, 117:1295–1328.
Mankiw, N. G. and Reis, R. (2018). Friedman’s presidential address in the evolution of macroeconomic thought. Journal of
Economic Perspectives, 32(1):xx–xx.
Melosi, L. (2017). Signalling effects of monetary policy. Review of Economic Studies, 84(2):853–884.
Nimark, K. P. (2014). Man-bites-dog business cycles. American Economic Review, 104(8):2320–2367.
Paciello, L. and Wiederholt, M. (2014). Exogenous information, endogenous information, and optimal monetary policy. Review
of Economic Studies, 81(1):356–388.
Phelps, E. S. (1970). Introduction: The new microeconomics in employment and inflation theory. In Phelps, E. S., editor,
Microeconomic Foundations of Employment and Inflation Theory, pages 1–26. Norton, New York.
Reis, R. (2006). Inattentive producers. Review of Economic Studies, 73(3):793–821.
Sims, C. (2003). Implications of rational inattention. Journal of Monetary Economics, 50(3):665–690.
Trabandt, M. (2007). Sticky information vs. sticky prices: A horse race in a dsge framework. Working Paper Series 209,
Sveriges Riksbank.
Woodford, M. (2003). Imperfect common knowledge and the effects of monetary policy. In Aghion, P., Frydman, R., Stiglitz,
J., and Woodford, M., editors, Knowledge, Information and Expectations in Modern Macroeconomics: In Honor of Edmund
S. Phelps, pages 25–58, Princeton. Princeton University Press.
Woodford, M. (2013). Macroeconomic analysis without the rational expectations hypothesis. Annual Review of Economics,
5(1):303–346.

Jonas Dovern CES Lecture 2 February 14, 2018 38 / 38

You might also like