A0867
Title: Posterior inferences on incomplete structural models: The minimal econometric interpretation
Authors: Takashi Kano - Hitotsubashi University (Japan) [presenting]
Abstract: The minimal econometric interpretation (MEI) of DSGE models provides formal model evaluation and comparison of misspecified nonlinear SDE models in concert with atheoretical reference models. The MEI approach recognizes DSGE models as incomplete econometric tools that provide only prior distributions of targeted population moments but have no implications for actual data and sample moments. A Bayesian posterior inference method is developed based on the MEI approach. The prior distributions of targeted population moments simulated by the DSGE model impose restrictions on the hyperparameters of the Dirichlet distributions, which are natural conjugate priors for the multinomial distributions that the corresponding posterior distributions estimated by the reference model follow. The Polya marginal likelihood of the resulting restricted Dirichlet-multinomial model has a tractive approximated log-linear representation of the Jensen-Shannon divergence that the proposed distribution-matching posterior inference uses the limited information likelihood function. Monte Carlo experiments prove that the MEI posterior sampler rightly infers the calibrated structural parameters of an equilibrium nonlinear asset pricing model and detects the correctly specified model with the posterior odds ratios.