Estimation of Quasi Rational DSGE Models

This paper investigates the forecasting performance of the class of small-scale New Keynesian Dynamic Stochastic General Equilibrium (DSGE) business cycle monetary policy models under the assumption that the agents are `boundedly rational' and have quasi-rational expectations (QRE). The QRE hypothesis maintains that the agents compute their conditional forecasts by using their `best fitting' statistical model for the variables. The agents' forecasting model is a VAR system when all variables are observed and is a state-space system when some of the variables are unobserved. In particular a QRE-DSGE model is obtained from the baseline DSGE model by suitable augmenting its dynamic structure such that the reduced form solution of the system has the same time-series representation of the agents' forecasting model. This approach provides a `natural' remedy to the typical difficulties DSGE models based on rational expectations have to account for the rich contemporaneous and dynamic correlation structure of the data. The proposed approach suggests a way to connect DSGE and VAR modeling. An application based on U.S data illustrates the advantages of using QRE in terms of forecasting performance.