STAR模型的滞后阶数选择与稳健性研究

Selection of the Lag Order of STAR Model and Its Robustness Analysis

本文讨论了6种信息准则在STAR模型滞后阶数选择中的适应性及稳健性问题。Monte Carlo模拟结果显示,在多数情况下,数据生成过程中的误差项分布并不影响信息准则正确识别模型最大滞后阶数的能力;对于短STAR模型,ACC准则具有较高的正确识别率,并且对不同平滑转移系数及不同门限值具有很好的稳健性;而对于长STAR模型,SC准则及ACC准则具有更高的正确率及良好的稳健性。

This paper discusses the applicability and robustness of six information criteria for determining the lag order of smooth transition autoregressive （STAR） models via Monte Carlo simulations. Simulation resuhs show that in most situations, the error distribution of the data generation process does not influence the ability of the information criteria to correctly recognize the maximum lag order of a STAR model. For a short STAR model （ a model with a lag order 〈 5） , the ACC criterion can determine the actual maximum lag order with higher accuracy, and exhibits robustness to different smooth transition coefficients and threshold values. The same results are generated by the Schwarz and ACC criteria for a long STAR model （ lag order 〉 5）.