A Re-Evaluation of the Performance of the Unit Root And Cointegration Tests Under STUR: No More Pitf
2004/10/01
研討會日期 : 2004-10-01
時間 : 15:00
主講人 : Prof. Ga-Won Yoon
地點 : B棟110室
演講者簡介 : Prof. Yoon為Ph.D. in Econometrics，University of California, San Diego（1994）。
現為釜山大學經濟系助理教授。
其主要研究領域為Time-series。
演講摘要 : Previous simulation results reported in Gonzalo and Lee (1998) show that the most widely used cointegration tests would wrongly find that two independent stochastic unit root [STUR] processes are cointegrated. Gonzalo and Lee (1998) call this behavior of the tests the pitfalls in testing for long-run relationships. This paper re-examines the performance of the unit root and cointegration tests with the same data generating processes employed in Gonzalo and Lee (1998), and shows that there are actually no pitfalls under STUR in testing for cointegration, at least for the data generating processes employed herein. The difference occurs because the data generating process implemented in Gonzalo and Lee (1998) is not STUR in fact. When the condition for STUR is properly imposed, the cointegration tests find no spurious cointegrating relationship between two independent STUR processes. STUR is also shown to be very difficult to distinguish from a standard (fixed) unit root process with standard unit root tests, which is already well-known in the literature. These new simulations conducted in this paper also help better understand the previous simulation results in Gonzalo and Lee (1998)..
