Why Applied Macroeconomists Should Not Use Bayesian Estimation of DSGE Models

This paper examines how Bayesian estimation performs in applied macroeconomic DSGE models when prior beliefs are misspecified. Using controlled Monte Carlo experiments on a standard Real Business Cycle model and a New Keynesian model, the authors show that Bayesian procedures can deliver severely biased and misleading parameter estimates, with posteriors pulled toward the researcher’s prior rather than the true data-generating process. In contrast, a classical simulation-based method, Indirect Inference, remains largely unbiased and robust even under substantial model uncertainty. The results imply that heavy reliance on Bayesian estimation can entrench false conclusions about key structural features, such as the degree of nominal rigidity, and thereby mislead policy analysis. The paper argues for greater use of robust estimation and model-validation techniques, such as Indirect Inference, to ensure that DSGE-based policy advice rests on credible empirical evidence.