Upload
vuonghanh
View
216
Download
0
Embed Size (px)
Citation preview
1
DSGE modelling at the European
Central Bank
Frank Smets
X Annual Seminar
Banco Central do Brazil
The views expressed are my own and not necessarily those of the ECB.This presentation is based on work by many ECB colleagues, in particular Kai Christoffel,Günter Coenen, Roberto Motto, Massimo Rostagno and Anders Warne.
2
Outline
1. Introduction and motivation
2. Structure of Bayesian DSGE models at the ECB
3. Some empirical applications using the NAWM
4. Conclusions and way forward
3
I. Introduction
4
• Monetary policy makers need a wide range of
macro-econometric models/tools
• for forecasting
• and for policy analysis
• More and more institutions (Fed, ECB, IMF,
Sveriges Riksbank, …) include Bayesian Dynamic
Stochastic General Equilibrium (DSGE) models in
their tool kit for monetary policy analysis and
forecasting.
Introduction
5
• Bayesian DSGE models combine
• a sound micro-founded DSGE structure, characterised by
the derivation of behavioural relationships from the
optimising behaviour of agents subject to technological
and budget constraints and the specification of a well-
defined general equilibrium,
� which makes it suitable for policy analysis;
• with a full-system Bayesian likelihood estimation,
�which provides a good probabilistic description of the
observed data and a good forecasting performance.
Introduction
6
• Advantages of the DSGE approach:
– The general equilibrium structure lends itself to telling
economically coherent stories and structuring forecast-
related discussions accordingly.
– Information about “deep” structural parameters can be
used to calibrate/estimate the model (e.g. breaks, short
time-series) and the model structure (e.g. cross-equation
restrictions) helps to identify parameters and the type of
shocks and to reduce the risk of over-fitting.
– Less subject to the Lucas critique and more suitable for
policy analysis.
– Puts a premium on expectations
– Better feel for which parameters are likely to be policy
invariant and which ones are not.
Introduction
7
• Advantages of Bayesian likelihood approach:
– Formalises the use of prior information and helps
identification, thereby also making the estimation
algorithm of the highly restricted model much more
stable.
– Delivers a full characterisation of the parameter and
shock uncertainty, allowing to construct probability
distributions for unobserved variables (e.g. output gap)
and derived functions (e.g. forecasts)
– Very flexible approach to deal with measurement error,
unobservable variables, different sources of information.
– Provides a framework for empirically evaluating models
and the appropriate input for model averaging and
Bayesian decision making under model uncertainty.
Introduction
8
II. Structure of Bayesian DSGE
models used at the ECB
9
Motivation
• Currently, two Bayesian DSGE models are
routinely used at the ECB:• New Area Wide Model (NAWM) (Christoffel et al, 2008)
• Used for forecasting (in the context of the quarterly ECB Projection
Exercises) and policy analysis;
• There is also a calibrated version, relatively richer in detail and open
for topic-driven extensions.
• Christiano-Motto-Rostagno (CMR) Model• Supports the cross checking through monetary/financial scenarios
• Both models have a similar core, based on Smets
and Wouters (2003, 2007), but • NAWM includes a detailed international block
• CMR includes a detailed financial block
10
Models Overview: Core Structure
Households
- consumption/saving decisions
- labour supply
Production
- Combine labour and capital
Markets: imperfect competition &
price and wage setting as a markup
Monetary Policy
Government
Core structure of both models
Smets-Wouters (2003, 2007)
11
• Important considerations guiding the
development of the estimated version of the
NAWM:
• “to provide a comprehensive set of core projection variables”
• “to allow conditioning the projections on exogenous
assumptions regarding monetary, fiscal and external
developments”
Models overview: NAWM
12
Models Overview: NAWM
Households
- consumption/saving decisions
- labour supply
Production
- Combine labour and capital
Markets: imperfect competition &
price and wage setting as a markup
Monetary Policy
Government
Exports
Imports
Rest of the World
Exchange Rate,
Foreign Demand,
Oil price …
Final Goods
Combining domestic and imported goods
Equations:
• about 50 endogenous
variables
• autoregressive processes
for the model's structural
shocks and the AR/VAR
models for government
consumption and the
foreign variables
• measurement equations
and identities
13
Models Overview: CMR
Households
- consumption/saving decisions
- labour supply
Production
- Combine labour and capital
Markets: imperfect competition &
price and wage setting as a markup
Monetary Policy
Government
Financial Intermediation
Liabilities
- Deposits
(components of M1, M3)
Assets
- Reserves
- Loans
Portfolio Decisions
External Financing
• The CMR expands on the core structure with a monetary and financial block which interacts directly with consumption, investment and price setting.
• The different focus of the two models (international dimension in NAWM and monetary/financial in CMR) make them usable for complementary purposes
14
Data, Model and Shocks
Macroeconomic
VariablesStructural
Shocks
Demand
Technological Forces (e.g TFP)
Exogenous Shifts in Markups
Policy
International (in NAWM)
Financial (in CMR)
National accounts,
Inflation, 3-month interest rate
International (in NAWM)
[exchange rate; deflators; foreign
demand; foreign prices; U.S. FFR;
competitors' export prices]
Financial (in CMR)
[M3; M1; credit; stock-market index;
external finance premium; 10-year
and 3-month interest rate spread]
Model
as an approximation of reality
15
Monetary Policy Transmission
Unanticipated Increase in Interest Rate by 50 bp
Output Inflation Interest Rate
• Response is qualitative same across models
• Considering uncertainty, quantitative differences small
0 5 10 15 20
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
Quarters
Simple Model
NAWM
CMR
0 5 10 15 20
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Quarters
Simple Model
NAWM
CMR
0 5 10 15 20-20
0
20
40
60
Quarters
Simple Model
NAWM
CMR
16
Monetary Policy Transmission: NAWM
0 5 10 15 20-0.9
-0.8
-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
Quarters
OutputConsumptionInvestment
0 5 10 15 20-0.4
-0.3
-0.2
-0.1
0
0.1
Quarters
ExportImports
0 5 10 15 20-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
Quarters
real Ex. Rate
0 5 10 15 20-0.25
-0.2
-0.15
-0.1
-0.05
0
Quarters
Domestic InflationConsumption Inflation
0 5 10 15 20-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Quarters
Rental Rate CapitalWagesMarginal Costs
Additional channels in open economy:
• Appreciation of currency
• Drop in exports amplifies output reaction
• Reduction in import prices implies
stronger inflation response
17
Monetary Policy Transmission: NAWM
Quarters 0 5 10 15 20
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
Quarters
0 5 10 15 20-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
Quarters 0 5 10 15 20
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
NAWMConstant Nominal Ex. Rate
Output Inflation Real Effective Exchange Rate
18
Monetary Policy Transmission: CMR
Deterioration in Economic Activity and Profit Prospects
Balance Sheet Deterioration and Rise in Bankruptcy Rate
Rise in External Finance Premium (over and above risk-free rate). Countercyclical
Price Developments Originally
Unforeseen Can Affect
Financial Position and Real
Economy
2 4 6 8 10 12 14 16 18 20
5
10
15
20
25
30
basi
s po
ints
Marginal
Average
2 4 6 8 10 12 14 16 18 20
0.05
0.1
0.15
0.2
0.25
0.3
Pe
rce
nt
Bankruptcy Rate
Output
5 10 15 20-0.45
-0.35
-0.25
-0.15
-0.05
Perc
ent
CMR baseline
CMR with indexed financial contracts
19
3. Some applications using the
estimated NAWM
(Christoffel, Coenen and Warne,
2008)
20
1. Historical analysis
Model
as a lens to interpret reality
Macroeconomic
VariablesStructural
Shocks
• Model as lens to interpret reality
• Identifying structural shocks
• Decompose variables into contributions from
shocks: Historical decomposition
• Example: what is the contribution of external
developments to euro area GDP growth
21
GDP: Role of international developments
Simple
Model
NAWM
Note: GDP growth, year-on-year % change, in deviation from mean. The sample spans 1999Q1-2006Q4
22
2. Forecasting: Out-of-sample evaluation
23
2. Forecasting: Mean and interval predictions
24
2. Forecasting: Prediction events
2000 2001 2002 2003 2004 20050
0.2
0.4
0.6
0.8
1Real GDP growth (yoy)
Year
Pro
babili
ty
Recession
2000 2001 2002 20030
0.2
0.4
0.6
0.8
1Consumption defl. infla
Year
Pro
babili
ty
0−2% inflationDeflation
25
3. Scenario analysis
Reversing wage moderation
26
3. Scenario analysis
Delay interest rate increase actually started in December
2005 by two quarters and return gradually to actual interest
path
Implication of delayed increase:• Higher GDP growth for 6 quarters
• Inflation path above actual/projected path for 14 quarters
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
2005 2006 2007
2
2.5
3
3.5
4
4.5
5
2005 2006 2007
1
1.5
2
2.5
3
3.5
4
90 confidence70 confidenceactual Output Growth scenario Output Growth
2005 2006 20071.6
1.8
2
2.2
2.4
2.6
2.890 confidence70 confidenceInflation actual pathInflation scenario path
R new scenario pathR actual pathR previous path
Output Inflation Interest Rate
27
Conclusions
28
• Bayesian DSGE models provide an empirically and
theoretically coherent picture useful for monetary policy
analysis.
• Like most policy models, they are still very much work in
progress:
– Need for richer specification of labour market, asset
markets, fiscal policy, sectoral and input/output
structure, international interactions, …
– Rational expectations? Allow for learning and imperfect
information.
– Strategic interactions between different agents and
different actions are often minimised in order to
maintain tractability.
Conclusions
29
– But there is a risk of creating new monsters:
• A single large model may be more difficult to
understand, more difficult to handle and lead to a
lack of robustness.
– General trend:
• One model in forecasting process
�Ensures consistency across projection rounds
• Suite of models for policy/scenario analysis
�Ensures robustness and flexibility in addressing the
multitude of questions.
– Question: How large should the benchmark model be?
Conclusions
30
Background Slides
31
Comparison of model characteristics
YesYesNoYesYes50NoNAWM
Yes
Limited
Limited
Forward-
looking
NoYesYesNo29NoCMR
No
No
Financial
sector
No
No
Micro-
founded
OpenRegular
forecast
SizeCountry
Details
YesYes~ 500YesMCM
YesYes119NoAWM
32
Data Coverage
Nominal interest
rate
Compensation
per employee
GDP deflator,
Consumption
Deflator,
Extra Euro
Area Import
Deflator
Employment,
Extra Euro
Area Exports
and
Imports
Output,
Investment,
Consumption,
Government
Consumption,
NAWM
3-month Interest
RateCompensation
GDP deflator,
Investment
deflator
Hours
workedCMR
Prices, Wages and Interest RatesNational Accounts
Common Block
• CMR and NAWM share common core set of
variables
• Additional variables according to model focus
33
Data Coverage (continued)
• Model Specific Variables
M1, M3, Credit, Stock Market Index
External Finance Premium, 10-year and 3-month
interest rate spread
CMR
Nominal effective exchange rate,
Foreign demand,
Foreign Prices (weighted GDP deflator)
U.S. federal funds rate
Competitors’ export prices
Oil Prices
NAWM
Financial BlockInternational Block
Specific Blocks
34
Key Parameter Estimates
0.890.250.230.100.27CMR
10.630.240.410.08NAWM
10.300.280.170.08Simple
Model
Indexation to
Productivity
Indexation in
Wage
Wage
Adjustment
Probability
Indexation
in
Prices
Price
Adjustment
Probability
Wage-SettingPrice-Setting
Price and Wage Setting
Note: Figures in the first and third columns indicate the estimated price/wage adjustment probability. The indexation parameter in
price and wage setting (second and fourth columns), is a measure of second-round effects and corresponds to the estimated weight
placed on past inflation in the estimated price-inflation and wage-inflation equations, respectively. The complements to 1 of those
coefficients provide some indication of the “anchoring” of inflation expectations.
• Price and Wage Nominal Rigidities Are Important:• Price Adjustment: Higher Rigidity in NAWM
• Price Indexation: Higher in NAWM
• Wage Adjustment: Similar Across Models
• Wage Indexation: Higher in NAMW
35
Key Parameter Estimates (cont.)
0.1 [credit development]0.860.30 [speed of economic activity]1.86 [expected inflation]
0.26 [inflation acceleration]
CMR
0.810.08 [speed of economic activity]1.75 [lagged inflation]
0.40 [inflation acceleration]
Simple
Model
0.870.15 [speed of economic activity]1.90 [lagged inflation]
0.19 [inflation acceleration]
NAWM
Additional Terms“Gradualism”Reaction to Economic
Activity
Reaction to Inflation
Estimated Monetary Policy Reaction Functions
Note: The coefficient on inflation, output and credit indicate the increase in the policy instrument in response to an increase in those indicators.
Main Features of Monetary Policy Reaction:
• Response to Inflation Developments
• Restrain from Reacting to Each Twist and Turn in Data
• In CMR, Attempt to Capture Monetary Pillar Features
36
Model Overview
Households
- consumption/saving decisions
- labour supply
Production
- Combine labour and capital
Markets: imperfect competition &
price and wage setting as a markup
Monetary Policy
Government
Exports
Imports
Rest of the World
Exchange Rate,
Foreign Demand,
Oil price …
Final Goods
Combining domestic and imported goods
Portfolio decisions
Liabilities
- Deposits
Assets
- Reserves
- Loans
Financial Intermediaries
External financing