基于Wasserstein距离的智能体模型估计

摘要

We propose a Simulated Minimum Distance (SMD) estimator based on the Wasserstein distance for estimating parameters in economic models with intractable likelihoods, such as Agent-Based Models (ABM). This approach avoids the need for complex likelihood approximations and is robust even when model and data distributions have non-overlapping support. Simulating the two- and three-type Brock and Hommes (1998) asset pricing model, we show that the estimator is more accurate than a likelihood-based Bayesian benchmark, in particular in the estimation of the behavioral parameters. We conclude with an empirical application of the Brock and Hommes (BH) model to S&P 500 data. Overall, the Simulated Wasserstein Minimum Distance Estimator (SWMDE) offers a flexible and computationally efficient tool for structural estimation in models where likelihood-based inference is infeasible or unreliable.