Abstract
We consider how to select the auxiliary distribution to implement the wild bootstrap for regressions featuring heteroscedasticity of unknown form. We propose a new class of two-point distributions and suggest using the Kolmogorov–Smirnov statistic as a selection criterion.
