Working Paper, Federal Reserve Bank of Atlanta 2003-26
Cogley and Sargent provide us with a very useful tool for empirical macroeconomics: a Gibbs sampler for the estimation of VARs with drifting coefficients and volatilities. The authors apply the tool to a VAR with three variables-inflation, unemployment, and the nominal interest rate-and two lags. This tool is a serious competitor to the identified-VAR-cum-Markov-switching technology recently developed by Sims (1999) and Sims and Zha (2002) for the study of economies that are subject to regime changes. However, the Gibbs sampler suffers from a curse of dimensionality: as more variables or more lags are added to the system, the computational burden of the estimation quickly grows out of proportion. My suggestions here are mainly aimed at making the tool more flexible, and hence more widely applicable.