基于自适应组LASSO估计的高维门限因子模型一致选择

Consistent Model Selection in High Dimensional Threshold Factor Model Based on Adaptive Group-LASSO

门限因子模型设定载荷具有阈值型区制转换结构,可以同时刻画高维时间序列的共变性和区制转换特征。针对高维门限因子模型,本文基于自适应组LASSO技术给出了一种一致模型选择过程。这一模型选择过程将因子个数设定、门限效应推断纳入统一的分析框架,不仅解决了模型选择的一致性问题,还同时实现了模型选择误差的统一控制,这对于高维门限因子模型而言是非常重要的。理论研究和随机模拟结论表明本文给出的一致模型选择过程具有良好的大样本性质和有限样本表现。最后,本文将门限因子模型应用于我国金融市场分析,实证结果进一步验证了本文理论的有效性。

With the threshold-type regime shift in factor loadings,the threshold factor model can delineate the co-movement and regime shift of high-dimensional time series.In this paper,a consistent model selection process of high-dimensional threshold factor model is presented based on an adaptive group-LASSO method,in which the identification of the number of factors and the inference of the threshold effect are integrated into a unified analytical framework.It not only solves the consistency problem of model selection,but also realizes the unified control of model selection error,which is very important for a high-dimensional threshold factor model.The results of theoretical research and random simulation show that the method has a good large sample property and limited sample performance.Finally,the threshold factor model is applied to China’s financial market,and the empirical results further verify the effectiveness of our theory.