Estimation of DSGE models when the data are persistent

An active area of research in macroeconomics is to take DSGE models to the data. These models are often solved and estimated under specific assumptions about how the exogenous variables grow over time. In this paper, we first show that if the trends assumed for the model are incompatible with the observed data, or that the detrended data used in estimation are inconsistent with the stationarity concepts of the model, the estimates can be severely biased even in large samples. Estimates of parameters governing transmission mechanisms can be severely biased. We then consider four estimators that are robust to whether shocks in the model are assumed to be permanent or transitory and do not require the researcher to take a stand on the dynamic properties of the data. Simulations show that when the shocks are not persistent, the proposed estimators are as precise as estimators that correctly impose the stationarity assumption. But when the shocks are highly persistent yet stationary, the proposed estimators are much more precise. These properties hold even when there are multiple persistent shocks. ∗Department of Economics, UC Berkeley, Berkeley, CA 94720 Email: [email protected]. †Department of Economics, Columbia University, 420 West 118 St, MC 3308, New York, NY 10027 Email: [email protected]. This paper was presented at the University of Michigan, the 2007 NBER Summer Institute, and the New York Area Macro Conference. We thank Marc Giannone and Tim Cogley for many helpful discussions and the seminar participants for many helpful comments. The second author acknowledges financial support from the National Science Foundation (SES 0549978).