Working Paper: CEPR ID: DP15388
Authors: Boragan Aruoba; Pablo Cubaborda; Kenji Hilgaflores; Frank Schorfheide; Sergio Villalvazo
Abstract: We develop an algorithm to construct approximate decision rules that are piecewise-linear and continuous for DSGE models with an occasionally binding constraint. The functional form of the decision rules allows us to derive a conditionally optimal particle filter (COPF) for the evaluation of the likelihood function that exploits the structure of the solution. We document the accuracy of the likelihood approximation and embed it into a particle Markov chain Monte Carlo algorithm to conduct Bayesian estimation. Compared with a standard bootstrap particle filter, the COPF significantly reduces the persistence of the Markov chain, improves the accuracy of Monte Carlo approximations of posterior moments, and drastically speeds up computations. We use the techniques to estimate a small-scale DSGE model to assess the effects of the government spending portion of the American Recovery and Reinvestment Act in 2009 when interest rates reached the zero lower bound.
Keywords: Bayesian estimation; Effective lower bound on nominal interest rates; Nonlinear filtering; Nonlinear solution methods; Particle MCMC
JEL Codes: C5; E4; E5
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| COPF (F55) | reduced persistence of the Markov chain (C41) |
| COPF (F55) | improved accuracy of Bayesian estimation (C11) |
| COPF (F55) | more stable and reliable likelihood approximations (C62) |
| COPF (F55) | enhanced accuracy of Monte Carlo approximations (C15) |
| COPF (F55) | reduction in computation time (C63) |
| government spending (H59) | larger estimated multipliers during the Great Recession (E19) |