A Study on Unit Root Tests of Linear and Nonlinear Time Serie Models

• Main Authors: Yi-Long Hsu, 許怡隆
• Other Authors: 汪義育
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/32788191460965781247

博士 === 國立政治大學 === 國際貿易學系 === 88 === The nonstationary in time series may influence the empirical results of economic hypothesis, and the construction of econometric empirical models. Since the traditional unit root tests are based on linear and symmetry model, So if the true model is nonlinear or asymmetry, there are model misspecify problems in unit root hypothesis testing, and effect the power of unit root tests. Althought some parametric nonlinear dynamic models have been popularly applied in economic and financial empirical studies, Most research baese on the previous unit root study results, and simply difference the series to be stationay without any pre-testing. The properties of the unit root tests performance under nonlinear model haven''t been pay too much attentions. The main purpose of this essay is to study the effects of linear and nonlinear model specification on unit root hypothesis testing. The topic of first essay is to study the relation between model specification, estimator efficicncy and the size of unit root tests. The second essay study the relation between efficicncy estimator of deterministic trend paramaters and the power of unit root tests. The third essay study the effects of power of unit root tests under nonlinear model specification. The main conclusion from simulation results are : The efficicncy of deterministic trend paramater estimators not only effect the power of unit root tests, but also the size of unit root tests. Under Markov switching unit root model, the traditional ADF test have lower power than linear model, and the power reduction depend on the transition probability. The traditional ADF test is consistent test, but needs more samaple to improve the power than linear model. Using maximum likelihood estimator for Markov switching unit root model to construct unit roo test, may improve the power significant.