Macroeconomic Forecasting in Chile: a Structural Bayesian Approach

We evaluate the performance of different models projecting the inflation rate, the output gap, the real exchange rate, and the interest rate related to monetary policy. We consider linear models widely used by central banks: a BVAR, a standard reduced-form New Keynesian model, and a DSGE model—all of them estimated by Bayesian econometrics. Our benchmarks are univariate time series models (AR(1) and random walk), but estimated by ordinary least squares. The results indicate that: i) introducing microfoundation (DSGE model) produces reasonable projections within a one-year horizon for inflation, the output gap and the monetary policy rate (MPR), ii) the priors for the parameter estimates are particularly useful for inflation and MPR forecasting, iii) those priors are helpful only if they come from well-founded models, iv) the reduced-form Keynesian model obtains the worst results, v) forecasting the real exchange rate gap, the univariate models remain better than the multivariate versions that we considered (Meese-Rogoff puzzle).