Monte Carlo Test Applied to Models Estimated by Indirect Inference

In this paper, we propose finite-sample inference procedures for parametric econometric models whose likelihood function is intractable and require simulation-based estimation methods, like indirect inference (Gourieroux, Monfort and Renault, 1993) or the efficient method of moments (Gallant and Tauchen, 1996). The procedures proposed are based on extensions of the technique of Monte Carlo tests which applies naturally to any model that can simulated. In particular, we show that the method of maximized Monte Carlo tests allows to control perfectly the level of test procedures for which only asymptotic justifions are typically proposed. The technique is applied to inference on stochastic differential equations.