Bootstrapping the Breusch-Godfrey autocorrelation test for a single equation dynamic model: Bootstrapping the Restricted vs. Unrestricted model

摘要

We use Monte Carlo methods to study the properties of the bootstrap Breusch-Godfrey test for autocorrelated errors in two versions a) by bootstrapping under the null hypothesis, restricted and b) by bootstrapping under the alternative hypothesis, unrestricted. We use the residual bootstrap for the bootstrap-BG test. Our analysis regarding the size of the test reveals that both bootstrap tests have actual sizes that lie close to the nominal size, with the restricted being better. Regarding the power of the test we find that with bootstrapping under the alternative hypothesis, the unrestricted bootstrap test has the greater power in small samples.