Transcript Surach Tanboon, Bank of Thailand

A Panel Discussion on
Recent Developments and Issues on DSGE Modeling
by
Surach Tanboon
Monetary Policy Department
Bank of Thailand
Presented at the SEACEN-CCBS/BOE-BSP Workshop on
Dynamic Stochastic General Equilibrium Modeling and Econometric Techniques
November 23–27, 2009
Manila, Philippines
Recent Workhops on DSGE Modeling
 CCBS/Bank of England (June 2009)
Research Forum: Recent Developments in DSGE Models
– “Incorporating Conjunctural Analysis in Structural Models”*
– Contribution Methodology to incorporate monthly timely
information in estimated quarterly structural DSGE models
 Research Department/IMF (November 2009)
Workshop: Forecasting with Structural Models and Real-Time Indicators
– “Structural Models in Real Time”**
– Contribution Methodology to combine extraneous predictions
in a way consistent with underlying model structure
*Giannoni, Monti, and Reichlin (2009) **Benes, Clinton, Johnson, Laxton, and Matheson (2009)
2
3
Motivation: Why off-model information might be needed?
 When estimate DSGE models, we have assumed all information can be
summarized in small data set
– True only when model is well specified / model variables are observed
 In practice, need potentially informative data/indicators to cope with:
 Unobserved model’s concepts
e.g., total factor productivity
 Imperfectly measured variables
e.g., aggregate price measured by CPI, PCE deflator, GDP deflator
4
Off-model information
Data
Example
External estimates
Macroeconomic Advisers / Blue Chip Economic
of variables
Indicators estimates of GDP
High-frequency observations
Monthly CPI—as indicator for quarterly CPI
on model variables
High-frequency indicators
of model variables
Large data sets
Monthly Private Investment Index—as indicator for
quarterly private investment
Indicators of output, employment, wages, consumption,
investment, interest rates, money, credit, prices, etc.—
to be incorporated in dynamic factor models
5
Boivin and Giannone (2006)
1. Model
3. Model in data-rich environment
F = subset of variables for which large
number of indicators X are observable
Here we impose DSGE model on
transition equations of latent factors
2. Model solution
Measurement equations
with
with
6
Boivin and Giannone (2006)
Case 1 No extraneous information (Smets and Wouters, 2004):
Seven model variables are perfectly observed
Case 2 With extraneous information: Case 1 + seven new indicators of F
7
Benes et al. (2009)
Real-time problem
Data partially available as of
beginning of quarter Q
Options for incorporating
extraneous information
1. Data set truncation
2. Pure model forecast
 Use all available quarterly data
 Solve model for missing obs.
3. “Hard tunes”
 Use estimates obtained outside
model to fill in for missing data
 Leeper’s critique (2003)
X indicates data available
4. “Soft tunes”
– Incorporate off-model predictions
consistently with model structure
through measurement equations
using Kalman filter
8
Soft-tuning
Model solution
Expanded state-space solution
 is extraneous predictions that contain information relating to
 These predictions can be made at any point during the quarter
 Allowing update of estimates each quarter