CMStatistics 2017: Start Registration
View Submission - CFE
Title: The empirical saddlepoint estimator Authors:  Benjamin Holcblat - University of Luxembourg (Luxembourg)
Fallaw Sowell - Carnegie Mellon University (United States) [presenting]
Abstract: Previous studies have shown that existing moment-based estimation approaches have poor small-sample performance in some applications. We propose an alternative that is based on the ESP (empirical saddlepoint) approximation of the solutions to the empirical moment conditions. Saddlepoint approximations are known to perform well in small sample. The novel estimator proposed, which we call the ESP estimator, is the mode of the ESP approximation. We show that it is consistent and asymptotically normal, and we study its higher-order bias. We propose novel test statistics based on the ESP estimator. Finally, we also investigate the finite-sample properties of the ESP estimator and related test statistics through Monte-Carlo simulations.