Lei Jin, Texas A&M University, Corpus Christi
Project Description/Abstract
In many econometrics applications, time series are often nonlinear and non-Gaussian. A bootstrap-assisted test is proposed to check the second-order stationarity of nonlinear time series. The test statistic is based on some Walsh ordinates from the Walsh transformation. Under the null hypothesis, the asymptotic normality of the Walsh ordinates is established and their asymptotic covariance matrix is obtained. A blocks-of-blocks bootstrap procedure is applied to estimate the asymptotic covariance matrix. The asymptotic null distribution of the test statistic is derived. In the framework of locally stationary processes, it is shown that the local power of the proposed test tends to one under local alternatives at a certain rate as the length of time series goes to infinity. A simulation study is conducted to examine the finite sample performance of the test with comparisons to some existing competing methods, indicating that the proposed approach works well for nonlinear time series.
Co-author
Suojin Wang, Texas A&M University