SOME IMPROVEMENTS FOR BOOTSTRAPPING REGRESSION-ESTIMATORS UNDER 1ST-ORDER SERIAL-CORRELATION

Standard t-tests can be severely biased in the presence of trended regressors and serially correlated errors. Veall (Economics Letters, 1986, 21, 41-44) presented simulation evidence that traditional bootstrapping does not substantially reduce this bias. Subsequent research has developed a number of refinements to the bootstrap including accelerated bias correction, prepivoting and iterated bootstrapping. The models considered by Veall were re-examined employing these refinements. It was found that substantial improvements are possible, particularly from prepivoting and iterated bootstrapping.