We consider a set of non-nested and potentially misspecified DSGE models, geometrically combine their likelihood functions, and estimate the parameters using the resulting composite likelihood. We highlight classical asymptotic properties of composite likelihood estimators, describe Bayesian posterior computations via MCMC, and compare the approach to BMA, nite mixture and robustness methods. We propose a criteria to select the models entering the composite likelihood. We robusty inference using the composite posterior distribution of the parameters and the pool of models. We provide estimate the marginal propensity to consume out of transitory income. Joint with Christian Matthes (Federal Reserve Bank of Richmond, United States).

JEL Classification numbers: C13, C51, E17.

Keywords: Model misspecification, composite likelihood, Bayesian model averaging, finite
mixture.