Title: Residual bootstrap for VAR models estimated by least absolute deviations
Authors: Hanno Reuvers - Maastricht University (Netherlands) [presenting]
Abstract: The fixed design residual bootstrap method is considered in order to conduct inference in stationary vector autoregressive models. The fixed design residual bootstrap treats the regressor matrix as fixed and adds resampled residuals to construct a bootstrap sample. The asymptotic validity of the method is established. We also show that bootstrapped Wald and LM type of test statistics can be used to test linear hypothesis. Our method does not rely on density estimation and is thus easy to apply. A Monte Carlo study reports on the finite sample performance. The fixed design bootstrap is suggested to have good size properties and better size properties than a recursive bootstrap scheme. The bootstrapped version of the Wald type of test has a higher power than both the bootstrapped version of the LM test and the asymptotically pivotal tests that require density estimation. The bootstrap Wald type of test is therefore recommended to practitioners.