Bootstrapping regression estimators under first-order serial correlation

It is well known that with highly trended time series data and strongly autocorrelated disturbances, there will be a marked tendency for standard GLS techniques to over-reject true null hypothesis in finite samples. There is also a potential problem because most applications of GLS are in conjunction with a pretest such as a Durbin-Watson test. An application of bootstrapping to these problems is considered here using a small Monte Carlo experiment; the results provide no evidence that standard bootstrapping provides an improvement.