A BVAR Forecasting Model for the Chilean Economy

Abstract

Doan, Litterman, and Sims have described a method for estimating Bayesian vector autoregressive (BVAR) forecasting models. The method has been successfully applied to the U.S. macroeconomic dataset, which is relatively long and stable. Despite the brevity and volatility of the post-1976 Chilean macroeconomic dataset, this paper shows that a straightforward application of the DLS method to this dataset, with simple modification to allow for delays in the release of data, also appears to satisfy at least one criterion of relative forecasting accuracy suggested by Doan, Litterman, and Sims. However, the forecast errors of the Chilean BVARs are still large in absolute terms. Also, the model's coefficients change sharply in periods marked by policy shifts, such as the floating of the peso in 1982.