Comments on “bootstrapping time series models”
质疑Li和Maddala关于自举残差总是首选方法的观点，指出残差的自协方差结构可能与真实误差不同，并提出半参数方法作为替代。
We take issue with the main suggestion in Li and Maddala (LiMa) that bootstrapping residuals is always the preferred approach and question some of their guidelines. We show that it can be potentially misleading to mimic the autocovariance structure of residuals, since it can be very different from that of true errors. We emphasize that the residuals are sensitive to model misspecification and generally not a part of the information set. We make constructive suggestions and propose a semiparametric method.