时变门槛设定下门槛回归模型有限样本性质的模拟分析

Simulation Analysis of Finite Sample Properties of Threshold Model With Time-varying Threshold Value

门槛回归技术被广泛应用于经济管理问题的分析研究,相关文献表明时变的门槛值可能更符合应用背景。然而,当真实模型含有时变门槛时,常数门槛回归模型的性质并未得到考察。鉴于此,文章通过数据生成过程,结合蒙特卡洛模拟对门槛回归模型参数估计、假设检验等问题展开研究,分析参数估计的性质、假设检验的检验功效和检验水平。研究发现:典型的时变门槛设定下,门槛回归模型的参数估计具有严重的系统性偏差,且门槛存在性检验的功效十分低。

Threshold regression has been widely applied in the analysis of economic management issues, and relevant literature indicates that the time-varying threshold value may be more consistent with the application background. However, when the real model contains time-varying threshold, the properties of the constant threshold regression model have not been investigated.In view of this, through the process of data generation and Monte Carlo simulation, this paper studies the threshold regression model parameter estimation and hypothesis testing, and also analyzes the nature of parameter estimation, test efficiency and test level of hypothesis testing. The study finds that under the typical time-varying threshold setting, the parameter estimation of threshold regression model has serious systematic deviation, and the threshold existence test has very low efficacy.