Accuracy Properties of the Statistics from Numerical Simulations

This paper is concerned with the approximation errors involved in the statistics from the sample paths generated by an approximate or computed policy function. Under fairly general assumptions, we establish convergence of these statistics to their true expected values as the approximation error of the policy function converges to zero. Furthermore, under additional regularity conditions we also show that the approximation errors of these statistics are of the same order of magnitude as that of the computed policy function. The constants bounding these orders of convergence are explicitly computed, and are shown to depend on primitive parameters.