阈值自回归模型参数估计的小样本性质研究

The Research on Small Sample Properties of Parameters Estimation in the TAR Models

对阈值自回归模型(TAR)的参数进行估计,主要是采用Chan的条件OLS估计法。本文将阈值自回归模型分为不连续的TAR、不连续的冲量阈值自回归(M-TAR)和连续的TAR(C-TAR)三种模型,采用Monte-Carlo模拟技术分别研究其参数估计的小样本性质。结果表明:在小样本中,阈值和自回归参数估计都存在明显的偏差;阈值估计相对于自回归参数估计而言,显得更加不稳定,具有更大的偏差和标准差。进一步研究发现,数据过程均匀率和标准差是影响参数估计小样本性质的主要因素。

The parameters of threshold autoregressive models can be estimated by known conditional least square estimation （Chan, 1993）. Threshold autoregres- sive model （TAR） includes discontinuous TAR, discontinuous momentum thresh- old autoregressive mode[ and continuous TAR mode[. In this paper, we investigate the small sample properties of parameters estimation in the Two-Regime TAR mod- els through a set of Monte-Carlo simulation. The results show there are many obvi- ous bias in the parameters estimation of TAR models; The estimation of the threshold value is more unstable than autoregression parameters＇, and it has large bias and standard errors. Further research shows that even ratio and standard errors of data process play an important role in small sample properties on parame- ter estimation.