Path-Dependent Stuctural Stability Tests of Markov-Switching Processes
2004/12/14
研討會日期 : 2004-12-14
時間 : 15:00
主講人 : Prof. Tsay, Wen-Jen
地點 : B棟110室
演講者簡介 : Prof. Tsay, Wen-Jen 為Ph.D. in Economics，Michigan State University (1995)。
現為本所副研究員。
其主要領域為Time Seriess及International Finance。
演講摘要 : This paper shows that the one period ahead prediction error generated from Hamilton’s (1993) predictor is likely to be asymmetric in general although the data-generating process is a 2-state AR(1) Markov-switching model and the current state is truly know. Thus, the prediction tests proposed in Box and Tiao (1975) and Barone and Roy (1983) cannot be trivially applied to an N-state Markov-switching model, since they are mainly implemented with symmetric percentiles in detecting the structural stability of a data series. To fully capture the path-dependent nature of a general Markov-switching process, we propose an alternative path-dependent prediction test based on an N-nominal lattice. We also apply this test to the U.S. ex post real interest rate and identify 1973 as a possible shift. This finding is almost identical to that in Garcia and Perron (1996), where they employ filter probabilities to date the change point of the real rate.
