报告人:胡雪梅（重庆工商大学）

时  间:2018年 11月29日 上午11:00--12:00

地  点:理科楼 LD402

摘  要:Predicting up trends, down trends and sideways trends for stock returns is an important issue in the financial world. Regularized multinomial logit model can not only provide the underlying class probabilities, but also improve the classification accuracy by shrinking the regression coefficients and avoiding multi-collinearity and over-fitting. In this paper we propose regularized multinomial logit model to predict stock return movement directions. To improve predication accuracy, we introduce three regularization criterions, develop the group coordinate descent (GCD) algorithm to identify important group vectors, and show the convergence property of the GCD algorithm. Finally, we apply the Bayes classfier to identify up trends, down trends and sideways trends for stock returns, and propose the three-class receiver operating characteristic (ROC) methodology and a classifier performance index called the volume under surface (VUS) to assess prediction performances for stock return movement directions.

报告人简介:胡雪梅，重庆工商大学数学与统计学院教授，长江上游经济中心(教育部人文社科重点研究基地)博士生导师，中南大学理学博士，中科院数学与系统科学研究院控制论国家重点实验室系统科学博士后，美国《数学评论》评议员，“第五批重庆市高等学校优秀人才支持计划”人选。研究兴趣包括股价趋势预测、高维数据分析、经验似然、半参数统计和随机过程推断；在Journal of Multivariate Analysis、Journal of Nonparametric Statistic、Statistics & Probability Letters和Statistical Papers等刊物上发表论文30多篇， 其中SCI收录22篇；主持完成国家自然科学基金青年项目1项、教育部人文社科青年项目1项、重庆市科委项目3项和市教委科技项目2项；参与完成973和国家社科等项目8项；参与获得重庆市科学技术奖二等奖；2016/12-2017/12获得国家公派高级研究学者、访问学者和博士后项目资助在伦敦政经统计系访问；目前主持重庆市基础研究与前沿探索项目1项，担任Journal of Business & Economic Statistics和Journal of Multivariate Analysis等国际杂志的评审人。

学院联系人:李曼曼 周国立

欢迎广大师生积极参与！