题目：Adaptive moment method for partially piecewise regression
时间：2018年10月29日(周一)下午14:00-15:00
地点：闵行校区法商南楼135室
报告人：林路 山东大学金融研究院教授、副院长
摘要：Piecewise regression is widely used in some areas such as econometrics. A foundational assumption on such a model is that partial information on the segments is known beforehand. Without the assumption, the difficulty of regression is not merely analytical, but also computational. In this paper, we introduce adaptive moment methods to identify a partially piecewise linear regression, without need of the information on the segments. The new idea is motivated by our findings that the moment conditions of the model contain the information of homogenous parameter and the subgroup-averages of the heterogeneous parameters. Thus we directly use the moment conditions to construct the estimator of the homogenous parameter, and identify the subgroup-averages of the heterogeneous parameters. The resulting estimator for homogeneous parameter has a simple expression, which is similar in form to the common least squares estimator, and is adaptive to various sizes of subgroups of heterogeneous parameters. Based on the subgroup moment estimators, the subgroups of heterogeneous parameters can be identified through mean change-point detections or dimension-reduced informational approaches. The methods are much easier than the existing methods. Our approaches are further illustrated via simulation studies and are applied to non-performing loan model.
报告人简介：林路，山东大学金融研究院教授、博士生导师、副院长；在南开大学获得博士学位后,先在南开大学任教，然后到山东大学任教至今；从事高维统计、非参数和半参数统计以及金融统计等方的研究，在国际统计学、机器学习和相关应用学科顶级期刊Annals of Statistics, Journal of Machine Learning Research, PLoS computational biology和其它重要期刊发表研究论文90余篇；主持过多项国家自然科学基金课题、博士点专项基金课题、山东省自然科学基金重点项目等；获得国家统计局颁发的统计科技进步一二等奖，山东省优秀教学成果一等奖；是国家973项目、国家创新群体和教育部创新团队的核心成员，教育部应用统计专业硕士教育指导委员会成员，山东省政府参事。