Stabilizing a GMM Bootstrap for Time Series:A Simulation Study
スポンサーリンク
概要
- 論文の詳細を見る
Inoue and Shintani (2006) demonstrate that in order for their GMM bootstrap to achieveasymptotic refinements for symmetric two-sided confidence intervals and J-statistics of overidentifying restrictions, a kernel of order greater than two must be employed for HAC estimation. A well-known problem of employing a higher-order kernel for HAC estimation is that the resulting covariance estimate does not necessarily become positive semi-definite in finite samples, which often leads to unsatisfactory performance of the bootstrap. This paper proposes to stabilize the bootstrap through employing the nonparametric prewhitened HAC estimator by Xiao and Linton (2002) and Hirukawa (2006), which has the same bias property as a fourth-order kernel but always generates a positive semi-definite estimate in finite samples. Monte Carlo results indicate that the HAC estimator indeed stabilizes the GMM bootstrap.