关键词: STAR模型, 递归退势, 单位根检验, 有限样本性质

Abstract: This paper focuses on the unit root tests using recursive de-trending procedure against alternative hypothesis where the time series data under investigation follow globally stationary ESTAR processes with alinear deterministic trend. It proposes a unit root test using recursive de-trending method, and derives its asymptotic distribution. The results of comparative study of small sample properties for unit root tests using OLS、GLS and recursive de-trending procedure show that the power of GLS and recursive de-trended unit root tests are better than OLS if the initial condition is negligible. If the initial deviation is sizable, the power of GLS de-trended unit root test declines dramatically, but the recursive de-trended unit root test maintains good power properties, especially for relatively large sample.

Key words: STAR, recursive de-trending, unit root test, finite sample performance