贝叶斯模型平均下的时变系数预测回归模型  

Time-varying Coefficient Predictive Model Combined With Bayesian Model Average

作　　者：杨光艺[1] Yang Guangyi(Guanghua School of Management,Peking University,Beijing 100871,China)

机构地区：[1]北京大学光华管理学院

出　　处：《统计与决策》2019年第19期20-24,共5页Statistics & Decision

基　　金：中国博士后科学基金会面上一等资助项目(2017M610001)

摘　　要：文章采用时变系数模型对中国股票市场的超额收益率进行预测,并结合贝叶斯模型平均的方法对各模型进行模型平均,得出稳健有效的预测结果。实证比较表明:(1)时变系数模型相比常系数模型具有更好的预测效果;(2)时变系数模型能帮助投资者在市场剧烈变化时迅速做出反应;(3)单变量情形下,宏观经济类变量具有更好的预测效果。This paper employs the time-varying coefficients model to predict the excess return of Chinese stock market,and uses the method of Bayesian model average to average each model,obtaining the robust and effective prediction results.Empirical comparison shows that 1)time-varying coefficient model has better prediction effect than constant coefficient model;2)The time-varying coefficient model can help investors to react quickly when the market changes violently;3)In the case of single variable,macroeconomic variables have better predictability.

关 键 词：时变模型 模型平均 预测回归 

分 类 号：F832.5[经济管理—金融学]