Finite Sample Properties of Bootstrap GMM for Nonlinear Conditional Moment Models

by Rachida Ouysse.

Abstract: We investigate the finite sample performance of block bootstrap tests for over-identification (J -test) for nonlinear conditional models estimated using Generalized Methods of Moment (GMM). Overall, block bootstrap methods with fixed length blocks outperform the stationary bootstrap which uses a random block length. Randomizing the block length decreases the sensitivity of the distribution of the bootstrap J -test and GMM estimators to the choice of the block size. This sensitivity diminishes with the degree of nonlinearity of the conditional moments. In addition, the accuracy of the block bootstrap approximation diminishes as the dimensionality of the joint test increases, especially in the tails of the distribution of the J-test.

Key Words: Continuously-updated GMM; iterated GMM; non-overlapping and moving blocks bootstrap; stationary bootstrap; rejection probability; QQ-plot; Monte Carlo test

Author:
Rachida Ouysse, rouysse@unsw.edu.au

Editor: Weiming Ke, Weiming.Ke@sdstate.edu

READING THE ARTICLE: You can read the article in portable document (.pdf) format (768776 bytes.)

NOTE: The content of this article is the intellectual property of the authors, who retains all rights to future publication.

This page has been accessed 2067 times since February 28, 2010.


Return to the InterStat Home Page.