Real-Time DSGE Model Density Forecasts During the Great Recession - A Post Mortem

Transcription

1 The views expressed in this talk are those of the authors and do not necessarily reflect the views of the Federal Reserve Bank of New York or the Federal Reserve System. Real-Time DSGE Model Density Forecasts During the Great Recession - Marco Del Negro Federal Reserve Bank of New York Frank Schorfheide University of Pennsylvania, CEPR, NBER September, MFM Meeting

2 How to think about DSGE models...

3 One of the events: point forecasting under quadratic loss (RMSEs) RMSE ratios: DSGE / AR().

4 DSGE Model Starting point is Smets and Wouters (7) model, modified to incorporate information from inflation expectation data. We add financial frictions along the lines of Bernanke, Gertler, and Gilchrist (999).

5 From SW to SWπ Model: Incorporating Inflation Expectations Why? High-inflation rates from lead to fairly large estimate of steady-state inflation rate (4 % annualized). = Upward bias in current inflation forecasts. Information about stance of monetary policy at forecast origin. How? Anchor target inflation rate using long-run inflation expectations. Augment measurement equations: [ ] πt O,4 4 = π + E t π t+k. 4 Modify policy rule: k= R t = ρ R R t + ( ρ R ) ( ψ (π t π t ) +... Time-varying inflation target evolves according to: π t = ρ π π t + σ π ɛ π,t.

6 From SWπ to SWπ-FF: Incorporating Financial Frictions In addition to SW model, we consider a model with financial frictions along the lines of Bernanke, Gertler, Gilchrist (999). Gross nominal return on capital: R k t = λr k t + ( λ)q k t q k t + π t SW model: arbitrage condition between return on capital and return on nominal bond: E t [ R k t+] = R t + b t, where R k t is treated as latent and b t is a shock. SW-FF Model: arbitrage condition is E t [ R k t+] = R t + b t +ζ sp,b ( q k t + k t n t ) + σω,t where R k t R t is treated as observed, σ ω,t is an additional shock, and n t is an additional endogenous variable.

7 Generating Forecasts with a DSGE Model DSGE Model = State Space Model Measurement Eq: y t = Ψ (θ) + Ψ (θ)t + Ψ (θ)s t State Transition Eq: s t = Φ (θ)s t + Φ ɛ(θ)ɛ t Posterior distribution of DSGE model parameters: p(θ Y :T ) = p(y :T θ)p(θ), p(y :T ) = p(y :T θ)p(θ)dθ. p(y :T ) Objective of interest is predictive distribution: p(y T +:T +H Y :T ) = p(y T +:T +H θ, Y :T )p(θ Y :T )dθ. Use numerical methods to generate draws from predictive distribution.

8 Using Real-Time Data Observables: output, consumption, investment, real wage growth, hours worked, inflation, Federal Funds rate, (Baa vs. -year treasury spread). Recursive out-of-sample forecasting; all estimation samples start in 964. Real time data, following Edge and Gürkaynak (): forecast horizons and data vintages are aligned with Blue Chip survey publication dates: we consider January, April, July, and October, ending April Forecast End of Est. Optional Spread & Forecast Origin Sample T FFR from T + h = h = Apr 8 7:Q4 8:Q 8:Q 8:Q Jul 8 8:Q 8:Q 8:Q 8:Q Oct 8 8:Q 8:Q 8:Q 8:Q4 Jan 9 8:Q 8:Q4 8:Q4 9:Q

9 Forecasting the Crisis: Model Versions SWπ: Smets-Wouters model with time-varying inflation target anchored by long-run inflation expectations. SWπ-FF: Smets-Wouters model with time-varying inflation target anchored by long-run inflation expectations and financial frictions. Utilizes data on spreads until period T. SWπ-FF-Current: Smets-Wouters model with time-varying inflation target anchored by long-run inflation expectations and financial frictions. Also use FFR and spread from current quarter T +. Spreads: based on Baa bonds versus -year treasury rate.

10 Forecasting the crisis: Oct, 7 (7:Q data) SWπ SWπ-FF SWπ-FF-Current

11 Forecasting the crisis: July, 8 (8:Q data) SWπ SWπ-FF SWπ-FF-Current

12 Forecasting the crisis: Jan, 9 (8:Q data) SWπ SWπ-FF SWπ-FF-Current

13 Forecasting the Crisis: Inflation with SWπ-FF-Current October, 7 July, 8 January,

14 What about GDP Growth Forecasts from an -variable VAR? Mixed frequency (monthly and quarterly) VAR

15 Summary No evidence that DSGE models have forecast worse during the Great recession than professional forecasters who may or may not have used them and had access to a wealth of additional information. Empirical DSGE models with financial frictions have been around at least since Christiano, Motto, and Rostagno (). What about the story? While the BGG mechanism generates amplification of shocks, the explanation for the Great Recession is mainly one of large adverse shocks, some of them originating on the financial side. Let s take a look...

16 Output Historical Decomp: Jan, 9 (8:Q data) g financial z markup r m π *.6

17 Food for Thought Is the DSGE community to blame for not having incorporated financial frictions into their standard model prior to the crisis?

18 Difference in Forecasting Accuracy Over Time: SWπ and SWπFF-Current Difference in 4-quarter-ahead Rolling RMSEs Output Inflation

19 In Closing... Early warning? Compelling, parsimonious economic story? Good policy prescriptions?

20 Inflation Historical Decomp: Jan, 9 (8:Q data) g financial r m π g financial z markup r * m π *

21 Inflation Historical Decomp: Jan, 9 (8:Q data) g financial z markup r m π *