基于Lasso时变图模型方法的我国股市网络结构分析

Network Structure Analysis of China’s Stock Market Based on Lasso Time-varying Graphical Model

针对金融资产的动态变化特征,构建动态时变的Lasso图模型理论,通过局部组套索惩戒施加于结构化平稳的模型,以图模型结构时变为前提,进行结构化平稳处理.将无向图模型方法应用于我国2014年1月―2020年3月股市的高维数据建模,利用Lasso方法给出精度矩阵的估计,从而识别得到上证380个股和带有权重的行业指数的图模型结构.实证研究结果表明:不同时点的上证380个股和行业的图模型结构均存在较大差异,股票关联结构呈现动态变化特征.快速上升期(牛市)和快速下跌期(熊市),个股和行业图模型边连接较多,关联结构较强;震荡期,个股和行业图模型边数量较少,关联性相对较弱.

A dynamic time-varying Lasso graphical model is constructed in response to the characteristics of dynamic change of financial assets. Discipline is imposed on the structured smooth model by the local set Lasso, with the premise that the graphic model structure is time-sensitive for structured smoothness processing. The undirected graphical model approach is employed to model the high-dimensional data of China’s stock market from January 2014 to March 2020. Utilizing the Lasso approach, the precision matrix is estimated, with a view to identifying the graphical model structure of the SSE 380 individual stocks and the industry indices with weights. Results of the empirical study reveal that: there are significant discrepancies in the graphical model structure of both SSE 380 individual stocks and industries at different points in time,while the stock correlation structure exhibits dynamic change characteristics. More edge joining and stronger correlation structure are observed between individual stocks and industry graphic models during the fast-rising period(bull market) and the fast-falling period(bear market);during the shock period, there are fewer numbers of edges of individual stocks and industry graphic models, with relatively weak correlations.