Supervised Dynamic PCA  



报告地点:西南中心516 腾讯会议:482-686-408 密码1108

报告摘要:This paper proposes a novel dynamic forecasting method based on a large number of predictors using a new supervised Principal Component Analysis (PCA). The new supervised PCA provides an effective way to bridge the predictors and the target variables of interest by scaling and combining the predictors and their lagged ones, which is in line with dynamic forecasting. Unlike the traditional diffusion-index forecasting, which does not learn the relationships between the predictors and the target variables before conducting PCA, we first re-scale each predictor according to their significance in forecasting the target variables in a dynamic fashion, and a PCA is then applied to a re-scaled and additive panel, which builds a connection between the predictability of the PCA factors and the target variables. Furthermore, we also use penalized methods, such as the LASSO approach, to select the significant factors that have more predictive power than the others. Theoretically, we show that our estimators are consistent and outperform the forecasts using traditional methods under some mild conditions. We conduct extensive simulations to verify that the proposed method produces satisfactory forecasting results and outperforms most of the existing methods using the traditional PCA. A real example on predictions of U.S. macroeconomic variables using a large number of predictors shows that our method performs better than most of the existing ones in general. Overall, our proposed procedure provides a comprehensive and effective method for dynamic forecasting in large dimensions.

专家简介:高照省,浙江大学“百人计划”研究员,博士生导师。博士毕业于香港科技大学数学系,先后于英国伦敦政治经济学院和美国芝加哥大学从事博士后研究工作。 回国前任职于美国理海大学数学系担任助理教授。 研究方向主要集中在高维和大尺度统计时间序列数据的统计分析,统计机器学习方法和预测,因子学习以及金融资产定价模型。研究成果发表在Journal of the American Statistical Association, Journal of Econometrics、International Journal ofForecasting, Statistica Sinica等统计学和经济学权威期刊上。主持过国家自然科学基金和中央高校科研项目。目前担任国际著名预测学杂志《Journal of Forecasting》的副主编。