A multiple DSGE-VAR approach: Priors from a combination of DSGE models and evidence from Japan

I propose a Bayesian VAR model with added priors from a combination of multiple DSGE models. The prior of the combination of multiple DSGE models improves the marginal likelihood of the DSGE-VAR with respect to a single DSGE model. This approach might be useful for model comparison between two or more DSGE models and for measuring the relative degrees of misspecification of DSGE models through comparing impulse responses of DSGE models with those of the multiple DSGE-VAR. From the data of Japanese economy including the “Bubble Boom” and the “Lost Decade”, I demonstrate the multiple DSGE-VAR combined two DSGE models with and without financial frictions, and evaluate misspecification of both DSGE models from their impulse response functions.