Download L. Burlon, S. Emiliozzi, A. Notarpietro, M. Pisani Bank of Italy

Medium-term conditional forecasting of
euro-area macroeconomic variables with DSGE
and BVARX models
L. Burlon, S. Emiliozzi, A. Notarpietro, M. Pisani
Bank of Italy
Abstract
The paper assesses the performance of medium-term forecasts of euro-area GDP and inflation obtained with a DSGE model and a BVARX model currently in use at the Bank of Italy. The performance
is compared with that of simple univariate models and with the Eurosystem projections; the same real time assumptions underlying the latter are used to condition the DSGE and the BVARX forecasts.
We find that the performance of both forecasts is similar to that of Eurosystem forecasts and overall more accurate than that of simple autoregressive models. The DSGE model shows a relatively better
performance in forecasting inflation, while the BVARX model fares better in forecasting GDP.
What do we do?
Annual RMSFE
• We focus on quarterly forecasts of euro-area GDP and inflation for
each quarter from 2002Q3 to 2014Q1 obtained with a DSGE and a
BVARX over the same 3 full-years horizon adopted in Eurosystem
projections.
• The DSGE and BVARX forecasts are conditional to the (main) assumptions underlying Eurosystem projections
• Their predictive performance is compared with the Eurosystem
projections and those of simple univariate models
GDP
GDP
Inflation Inflation
1 yr ahead 2 yrs ahead 1 yr ahead 2 yrs ahead
Eurosystem
0.4982
1.9763
0.2716
0.9012
AR(4)
0.4925
2.2831
0.2303
1.0848
0.2373
0.8235
DSGE anticipated
0.4769
2.0080
DSGE unanticipated 0.5052
1.9185
0.2200
0.7555
BVAR
0.4824
1.8367
0.2398
0.9828
BVARX
0.4915
1.8967
0.2577
1.0041
RMSFE
Main results
GDP
1
0.8
• The forecasting performance of both models, using the RMSFE
metric, is comparable to the historical performance of the Eurosystem projections
• The DSGE-based forecasts are relatively more accurate for inflation
• The BVARX fares relatively better in forecasting GDP
0.6
0.4
0.2
1
2
3
4
5
6
7
8
Consumption deflator
0.5
0.4
0.3
DSGE model
0.2
• Two-country New Keynesian model estimated with euro-area and
rest-of-the-world data with nominal and real rigidities
• Estimated recursively from 1995Q1 to 2013Q4
• 12 euro-area and 2 rest-of-the-world observables used in the estimation
0.1
1
AR(4)
2
3
DSGE anticipated
4
5
6
DSGE unanticipated
7
BVAR
8
BVARX
Historical errors
GDP (4 quarters ahead)
– EA: GDP, private and public consumption, investment, exports, imports, total employment, nominal compensation for employee, the consumption and
investment deflators and energy and non-energy components of the HICP.
– RoTW: US 3-month interest rate and World GDP deflator
Consumption deflator (4 quarters ahead)
2
1
1
0.5
0
0
−1
−0.5
−2
−1
−3
−1.5
−4
2002:3 2004:3 2006:3 2008:3 2010:3 2012:3
• In the DSGE, the conditioning over Eurosystem projections’ assumptions differs according to the economic agent’ information set
( anticipated vs unanticipated conditional forecasts )
−2
2002:3 2004:3 2006:3 2008:3 2010:3 2012:3
GDP (8 quarters ahead)
Consumption deflator (8 quarters ahead)
2
1
1
0.5
0
0
−1
BVARX model
−0.5
−2
−1
−3
−4
2002:3 2004:3 2006:3 2008:3 2010:3 2012:3
• The BVARX model contains:
– 6 endogenous variables for the euro area: GDP, imports, exports, pri-
AR(4)
vate consumption deflator, unit labor cost, EA long-run interest rate
– 5 exogenous variables coming from the Eurosystem projections
assuptions
• Estimated recursively from 1985Q1 to 2013Q4
• A Litterman prior is used over the block of endogenous variables
• Priors for the coefficients on exogenous variables are centered on
the elasticities of the Eurosystem models
DSGE anticipated
DSGE unanticipated
Forecast Bias
0.2
• The inflation rate is sistematically underestimated by
both models
0
−0.2
2
3
4
5
6
7
8
Consumption deflator
0.2
0
−0.1
−0.2
1
AR(4)
BVARX
• The DSGE unanticipated
conditional
projections
tend to be more accurate
0.1
• Conditioning set: Euribor 3 month interest rate, euro-area foreign
demand, euro NEER vis-à-vis 20 countries and Brent oil price (in
US dollars)
• We control for high-frequency information coming from surveys
and nowcasts: one step-ahead forecasts are constrained to be
equal to the Eurosystem projection.
BVAR
Some final remarks
GDP
0.4
−0.4
1
Setup of forecast exercise
−1.5
2002:3 2004:3 2006:3 2008:3 2010:3 2012:3
2
3
DSGE anticipated
4
5
6
DSGE unanticipated
7
BVAR
8
BVARX
• Both models fail to predict
2008Q4 fall in GDP
• DSGE forecasts show a
smaller bias than the BVARX
in forecasting the 2008Q4
inflation rate
• Future developments
– Forecasts pooling
– Density forecasts