报告题目：Cox regression models for bankruptcy prediction with time-varying covariates (具有时变协变量的破产预测的Cox回归模型)
报告时间：2018年5月17日（周四）下午16:00
报告地点：翡翠湖校区科教B楼1101会议室
报 告 人：马骏副教授（澳大利亚 悉尼麦考瑞大学统计学院）
工作单位：澳大利亚悉尼麦考瑞大学统计学院
主办单位：威斯尼斯人5845cc国际官网
报告人简介: 马骏，1983年在安徽大学获得数学理学士学位。1983年曾任合肥工业大学管理工程系教师。分别于1991年和1996年在澳大利亚悉尼麦考瑞大学获得统计学硕士和博士学位。麦考瑞大学统计学院副院长，副教授。澳大利亚统计学会成员，IEEE成员。承担完成多项澳大利亚研究理事会资助的研究项目。出版专著1部，在Statistics in Medicine，Communications in Statistics - Theory and Methods, Computational Statistics & Data Analysis, Australia & New Zealand Journal of Statistics等国际期刊发表论文40多篇，是国际期刊医学统计，多元分析，中国科学--数学等20多种杂志的评审专家。他的研究兴趣包括生存分析，大数据统计计算，医学成像，图像恢复，广义线性模型和半参数回归模型。他指导培养了15名博士生和30多名硕士生。
报告摘要： Recently, Cox regression models have been adopted in risk management for prediction of bankruptcy, where time varying covariates are included. The estimation method, however, are still based on the partial likelihood approach, which has the following shortcomings: (1) the baseline hazard is not estimated, so calculating quantities such as survival probabilities requires a further estimation step; and (2) a covariance matrix for the baseline hazard estimate, needed for example for standard deviation of survival probabilities, is not produced. We address these issues by developing a maximum likelihood method that jointly estimates regression coefficients and the baseline hazard using constrained optimisation. We show in a simulation that our maximum likelihood method gives more accurate regression coefficient estimates than partial likelihood in moderate to small sized samples with heavy censoring and produces a smoother estimate of the baseline hazard than the Breslow (1972) estimator.
