Bootstrapping heteroskedasticity consistent covariance matrix estimator (Q1424616)

This paper demonstrates that the wild bootstrap covariance matrix estimator for linear heteroskedasticity models can be calculated directly without simulation, since it is simply a traditional heteroskedasticity consistent matrix estimator (HCCME) proposed by \textit{F. Eicker} [Ann. Math. Stat. 34, 447--456 (1963; Zbl 0111.34003)] and \textit{H. White} [Econometrica 48, 817--838 (1980; Zbl 0459.62051)]. Both theoretical and experimental results make clear that the same forms of the wild bootstrap and of the HCCME estimator have similar finite sample behaviour, and that the wild always would be preferred to the wild bootstrap obtained by simulation (because a finite number of simulations introduces an experimental error).