2025-03-24-IMF-使用DSGE模型中的冲击分解评估历史事件（英）_41页_2mb

Evaluating Historical Episodes Using Shock Decompositions in the DSGE Model

This working paper analyzes the impact of different shock decomposition methods on interpreting historical macroeconomic episodes within a Dynamic Stochastic General Equilibrium (DSGE) model.

Key Findings

1. Problem & Methodology

   • DSGE models use state space representations to decompose economic variables into contributions from exogenous shocks over time.
   • The paper compares three shock decomposition methods:
     • DC1: Isolates shocks occurring after the start of the historical period.
     • DC2: Uses subsamples of full-sample decompositions (common but conceptually flawed).
     • DC3: Adjusts the full-sample decomposition through differencing (proposed as the best method but rarely used).
   • Key Insight: Different methods yield profoundly different interpretations of historical events due to how they handle initial conditions and prior shocks.

2. Model & Data

   • An extended DSGE model based on Smets-Wouters (2007) including fiscal policy, financial frictions, and preference shocks.
   • Uses U.S. data (1948-2020) to estimate model parameters and perform shock decompositions.

3. Empirical Analysis

   • Applies all three decomposition methods to analyze key historical periods: 1964-1966, 1979-1987, 2006-2009, 2016-2020, and 2020-2023.
   • 1964-1966: Shows DC1 attributes economic growth to prior shocks, while DC2 highlights fiscal role.
   • 2006-2009: DC3 clearly shows spread shocks and monetary policy's role, while DC2 obscures the crisis’s origins.
   • 2020 (COVID): All methods agree preference and private investment shocks drove the contraction, but DC1 highlights initial pre-pandemic deviations from the balanced growth path.
   • 1979-1987 (Volcker): All decompositions capture Volcker's contractionary policies, but DC1 avoids contamination from prior shocks.

4. Takeaways & Conclusion

   • DC1 (preferred) synthetically isolates the impact of shocks within the chosen period and should be used to avoid conflating historical context with the event under analysis.
   • DC2 (subsample) isolates period shocks but retains legacy effects.
   • DC3 (differencing) focuses on net changes but ignores initial conditions' influence.
   • Policy Implications: Improved historical analysis informs future policy design and model calibration.