基于τ统计量对Markov模型状态数目检验研究初探

Research on the Test of Markov Switching Model According to the τ Statistics

本文根据Markov机制转换模型状态变量预测概率分布特性的分析,提出了基于τ统计量对模型状态数目的检验方法,还运用蒙特卡罗模拟方法模拟了检验所需要的τ统计量的经验分布。检验方法在计算上简单,不存在模型估计过程中多余参量问题的困扰,是一种简便可行的检验方法。最后,运用该检验方法对Hamilton的美国GDP模型进行检验研究,结论支持Hamilton的模型,认为美国GDP确实存在两个不同的增长状态。

For the first time, the paper analyzes the character of the state variable＇ s forecasting probability of the Markov Switching Model and proposes a new method of testing the model according to the statistics. The paper also simulates the empirical distribution which is necessary to the τ test by Mont Carlo method. The method of test has the advantage of simple computation and without nuisance parameter problem. At last, we test the Hamilton＇s America GDP model using our testing method. The paper gets the result that Hamilton＇s model is reasonable and the American GDP growth rate really has two different states.