Bayesian Estimation of Economic Simulation Models using Neural Networks

摘要: Recent advances in computing power and the potential to make more realistic assumptions due increased flexibility have led prevalence of simulation models economics. While this class, particularly agent-based models, are able replicate a number empirically-observed stylised facts not easily recovered by traditional alternatives, such remain notoriously difficult estimate their lack tractable likelihood functions. estimation literature continues grow, existing attempts approached problem primarily from frequentist perspective, with Bayesian remaining comparatively less developed. For reason, we introduce protocol that makes use deep neural networks construct an approximation likelihood, which then benchmark against prominent alternative literature. Overall, find our proposed methodology consistently results accurate estimates variety settings, including financial heterogeneous agent identification changes dynamics occurring incorporating structural breaks.