Edgeworth expansions for the Wald and GMM statistics for nonlinear regressions Export

@inbook{hansen:2006, abstract = {The Edgeworth expansion is derived for the GMM distance statistic for a real-valued nonlinear restriction on a normal linear regression. A refinement of the Edgeworth expansion for the Wald statistic (Park and Phillips 1988) is provided. The leading coefficients are shown to be the same in these two expansions. This establishes that, to the order of approximation of the Edgeworth expansion, the GMM distance statistic has a better approximation to the chi-square distribution than does the Wald statistic. The Monte Carlo simulation of Gregory and Veall (1985) is updated to include both heteroskedasticity-robust covariance matrix estimation and the GMM distance statistic. If the robust covariance matrix is calculated under the null, the GMM has near perfect finite sample type I error even in sample sizes as small as 20.}, address = {New York}, author = {Hansen, Bruce}, booktitle = {Econometric Theory and Practice}, chapter = {1}, citeulike-article-id = {1459390}, editor = {Corbade, Dean and Durlauf, Steven N. and Hancen, Bruce E.}, keywords = {asymptotics, econometric\_theory, econometrics, edgeworth\_expansion, higher\_order\_asymptotics}, posted-at = {2007-07-16 15:58:35}, priority = {2}, publisher = {Cambridge University Press}, title = {Edgeworth expansions for the Wald and GMM statistics for nonlinear regressions}, year = {2006} }

TY - CHAP ID - hansen:2006 L3 - citeulike-article-id:1459390 N2 - The Edgeworth expansion is derived for the GMM distance statistic for a real-valued nonlinear restriction on a normal linear regression. A refinement of the Edgeworth expansion for the Wald statistic (Park and Phillips 1988) is provided. The leading coefficients are shown to be the same in these two expansions. This establishes that, to the order of approximation of the Edgeworth expansion, the GMM distance statistic has a better approximation to the chi-square distribution than does the Wald statistic. The Monte Carlo simulation of Gregory and Veall (1985) is updated to include both heteroskedasticity-robust covariance matrix estimation and the GMM distance statistic. If the robust covariance matrix is calculated under the null, the GMM has near perfect finite sample type I error even in sample sizes as small as 20. AD - New York TI - Edgeworth expansions for the Wald and GMM statistics for nonlinear regressions PB - Cambridge University Press T2 - Econometric Theory and Practice KW - asymptotics KW - econometric_theory KW - econometrics KW - edgeworth_expansion KW - higher_order_asymptotics AU - Hansen, Bruce A2 - Corbade, Dean A2 - Durlauf, Steven A2 - Hancen, Bruce PY - 2006/// ER -