4. Computational Experiments and Reality

This study explores three alternative econometric interpretations of dynamic, stochastic general equilibrium (DSGE) models. Under a strong econometric interpretation, these models provide likelihood functions for observed sequences of prices and quantities. Given this interpretation, most DSGE models are rejected using classical econometrics and assigned zero probability in a Bayesian approach. Under a weak econometric interpretation, commonly made in the calibration literature (Kydland and Prescott, 1996), a DSGE model mimics the world only along a carefully specified set of dimensions. Computational experiments provide predictive distributions in these dimensions, which are then compared with the corresponding observed values. The weak econometric interpretation shares the same assumptions as the strong econometric interpretation, however, and it therefore leads again to the conclusion that most DSGE models are not credible. Under a minimal econometric interpretation, introduced by DeJong, Ingram and Whiteman (1996) and further developed in this study, DSGE models provide only population moments for certain functions of prices and quantities. They have empirical content only in the presence of auxiliary models which link population moments with observables. Such models may be entirely atheoretical. This study shows how DSGE models and an atheoretical model may be integrated into a common probability model, and used together to draw conclusions about different DSGE models. The methods and conclusions of the study are all illustrated using models of the equity premium. Iowa were helpful in writing and revising the paper. Responsibility for errors rests with the author, and the views expressed are not necessarily those of the Federal Reserve Bank