Limited Information Estimation and Evaluation of DSGE Models. Working paper #6

We advance the proposal that DSGE models should not just be estimated and evaluated with reference to full information methods. These make strong assumptions and therefore there is uncertainty about their impact upon results. Some limited information analysis which can be used in a complementary way seems important. Because it is sometimes difficult to implement limited information methods when there are unobservable non-stationary variables in the system we present a simple method of overcoming this that involves normalizing the non-stationary variables with their permanent components and then estimating the estimating the resulting Euler equations. We illustrate the interaction between full and limited information methods in the context of a well-known open economy model of Lubik and Schorfheide. The transformation was effective in revealing possible mis-specifications in the equations of LS\'s system and the limited information analysis highlighted the role of priors in having a major influence upon the estimates.