Robust inference in nonstationary time series models

This paper studies robust inference in unit root and cointegration models. The analysis covers a range of important inference problems including testing stationarity against unit roots, testing for structure change in nonstationary regressions, and testing for cointegration. We analyze these inference problems in a unified regression framework, although separate analysis is given for each specific case when it is needed. The proposed inference procedures are constructed based on residuals of robust M-estimations. The limiting behavior of the proposed tests is investigated, and a Monte Carlo experiment is conducted. The proposed tests are easy to use and have advantages in the presence of non-Gaussian data.