We propose a generalized conditional Monte Carlo technique for computing
densities in economic models. Global consistency and functional asymptotic
normality are established under ergodicity assumptions on the simulated
process. The asymptotic normality result allows us to characterize the
asymptotic distribution of the error in density space, and implies faster convergence
than nonparametric kernel density estimators. We show that our
results nest several other well-known density estimators, and illustrate potential
applications. |