Peter Fuleky
University of Washington
Abstract
Duffee and Stanton (2008) argued for the superiority of Indirect Inference (II) over the Efficient Method of Moments (EMM) for estimating term structure models. I arrive at a different conclusion by using an alternative formulation of EMM (herafter EMM-2). The EMM-2 estimator is asymptotically equivalent to EMM, but it has additional features in finite samples. I conduct a thorough comparison of EMM, EMM-2, and II in the context of continuous time models of the interest rate. Such models have a natural auxiliary counterpart given by their crude Euler discretization. While the EMM estimator corrects the discretization bias, the EMM-2 estimator also corrects the finite sample bias, which dominates the former one in parameter estimates of persistent processes. As a result, the EMM-2 mean reversion estimates have a 50% smaller root mean squared error for realistic parameter values in sample sizes of 1000 weekly observations. Consequently, the distributional properties of bond prices or mildly non-linear transformations of the EMM-2 estimates are also improved. In addition to its finite sample bias correction properties, the EMM-2 estimator performs better than the original one in coefficient tests, and overall its behavior closely matches that of II.