Sequential Monte Carlo Methods in Econometrics with Applications to the Estimation of DSGE Models
Edward Herbst, Principal Economist, Board of Governors
Frank Schorfheide, Professor of Economics, University of Pennsylvania
Location of Workshop: Room 4-02 of the Fordham Law School at 150 West 62nd Street.
Timetable for Tuesday, June 27
1:30 – 2:45p Lecture 1
2:45 – 3:00p Break
3:00 – 4:15p Lecture 2
4:15 – 4:30p Break
4:30 – 5:30p Practical Exercises
- Lecture Notes: Lecture 1 SequentialMonteCarlo, Lecture 2 ParticleFilters. We will not distribute hardcopies of the lecture notes. We recommend that you download / print the pdf prior to the workshop.
- We will conduct some practical exercises in the last hour of the workshop using programs written in MATLAB. Thus, if you would like to run and experiment with sample code during the lab session, please install MATLAB on your laptop. Some of the programs also may run in Octave, which can be downloaded from the internet free of charge.
- Practical Exercises: (i) SMC Exercise Sheet, MATLAB-SMC (ii) Particle Filtering Exercise Sheet, MATLAB-PF
Herbst, E. and F. Schorfheide (2015): Bayesian Estimation of DSGE Models. Princeton University Press. [Companion Website]
Estimated dynamic stochastic general equilibrium (DSGE) models are now widely used by academics to conduct empirical research in macroeconomics as well as by central banks to interpret the current state of the economy, to analyze the impact of changes in monetary or fiscal policy, and to generate predictions for key macroeconomic aggregates. This workshop will introduce participants to sequential Monte Carlo (SMC) methods that can be used to implement Bayesian inference for DSGE models. SMC methods provide an alternative to Metropolis-Hastings algorithms for generating draws from the posterior distribution of DSGE model parameters. They are particularly well suited in applications in which the posterior distribution is non-elliptical and exhibits multiple modes. SMC algorithms can also be used to approximate the likelihood function of nonlinear DSGE models (particle filtering). This likelihood approximation can then be embedded in a posterior sampler for the unknown parameter. The first part of the workshop focuses on the use of SMC algorithms for parameter inference and the second part covers particle filtering. The computational techniques presented in this course are more generally applicable to Bayesian inference for nonlinear state-space models.