Block bootstrap methods for unit root testing.

摘要

For almost three decades, econometricians have been concerned with the problem of testing a time series for a unit root. The problem is significant, since the behavior of such a process differs greatly from that of a stationary process. More recently, some unit root tests have been developed using block bootstrap techniques. In this thesis, two new nonparametric, residual-based bootstrap procedures are proposed for testing for the presence of a unit root in a given time series. One is based on the stationary bootstrap while the other uses the tapered block bootstrap. The procedures give a way of generating a pseudoseries which mimics the original in terms of dependence structure, but that has the property of ensuring the presence of a unit root. The proposed tests have an advantage over many others in the literature in that they are valid for a wide class of weakly dependent processes and are not based on any parametric assumptions on the data-generating process. In addition, the bootstrap methodology is valid for a very general class of test statistics. Large sample theory is developed and the asymptotic validity of the tests is shown via bootstrap functional limit theorems. In addition, simulations are used to investigate the finite sample performance of the procedures.