Inference Using Simulated Neural Moments

This paper studies method of simulated moments (MSM) estimators that are implemented using Bayesian methods, specifically Markov chain Monte Carlo (MCMC). Motivation and theory for the methods is provided by Chernozhukov Hong (2003). The shows, experimentally, confidence intervals these may have coverage which far from nominal level, a result has parallels in literature overidentified GMM estimators. A neural network be used to reduce dimension an initial set minimum number maintains identification, as Creel (2017). When MSM-MCMC estimation inference based on such moments, continuously updating criteria function, statistically correct all cases studied. illustrated application several test models, including small DSGE model, jump-diffusion model returns S&P 500 index.