24. Panel Data Analysis and Bootstrapping: Application to China Mutual Funds

Thompson (Journal of Financial Economics 99, 1–10, 2011) argues that double clustering the standard errors of parameter estimators matters the most when the number of firms and time periods are not too different. Using panel data of similar number in firms and time periods on China’s mutual funds, we estimate double- and single-clustered standard errors by wild-cluster bootstrap procedure. To obtain the wild bootstrap samples in each cluster, we reuse the regressors (X) but modify the residuals by transforming the OLS residuals with weights which follow the popular two-point distribution suggested by Mammen (Annals of Statistics 21, 255–285, 1993) and others. We then compare them with other estimates in a set of asset pricing regressions. The comparison indicates that bootstrapped standard errors from double clustering outperform those from single clustering. Our findings support Thompson’s argument. They also suggest that bootstrapped critical values are preferred to standard asymptotic t-test critical values to avoid misleading test results.