A resampling method for regression models with serially correlated errors

A method for using bootstrap in regression models with serially correlated errors to estimate the variance of the parameter vector is proposed. This method uses the real Fourier transform to eliminate the serial correlation of the errors. In the resulting heteroscedastic regression model, resampling of (y(i),x'(i)) pairs is performed. The conditional distribution of the bootstrap replications of the ordinary least-squares (OLS) estimator of the parameter vector is shown asymptotically to be close to the sampling distribution of the OLS estimator. The performance of the method is illustrated via simulation. (C) 1997 Elsevier Science B.V.