Nonparametric minimum-distance estimation of simulation models

We propose a novel nonparametric minimum-distance estimator for the estimation of simulation models. Our approach leverages a nonparametric smoothing step to approximate the distance between real-world observations and data simulated from a model, allowing for the estimation of model parameters without relying on specific auxiliary models or moment selection. By employing sieve estimation techniques, we approximate the objective function using a series of basis functions, ensuring consistency and providing nonparametric rates of convergence. We investigate the asymptotic properties of our estimator and demonstrate its performance through Monte Carlo experiments and an empirical application to financial market data. Our method addresses the limitations of traditional simulation-based estimation techniques, particularly in cases where the stochastic equicontinuity condition is violated, and offers a robust framework for estimating parameters in heterogeneous agents models and other complex systems.