报告人：孔新兵 教授
报告题目：Volatility estimation with continuous time model
报告时间：2014年12月23日下午15:00
报告地点：静远楼1508学术报告厅
主办单位：太阳成集团、科技处
摘 要： Multivariate Bayesian hierachical models with independent components are extensively studied and widely used in practice, while models with certain dependence structure on the mean vector are less well documented. In this paper, we investigate a multivariate heterscedastic Bayesian hierarchical model in which an informative prior with equal correlations for the mean parameters is asserted. This kind of modeling stems from the frequentist's historical commonsense that multivariate observations are variable-wisely correlated. We then estimate the mean vector by the shrinkage estimator based on Stein's unbiased risk estimation (SURE) in determining the hyper-parameters. It is shown that the SURE estimator is superior in terms of the averaged squared loss as the number of means grows ($p\rightarrow \infty$) and close to the performance of the oracle estimator. Compared with the superior SURE estimators under independence (see Xie, Kou and Brown (2012), JASA), SURE estimators under dependence always have lower risk when the means are truly correlated. The simulation studies justify our findings and the results from two real data sets are encouraging
孔新兵个人简介
孔新兵，现为苏州大学数学科学学院教授，研究兴趣为金融统计、高维数据分析、多重检验等。在统计学和计量经济学杂志发表论文多篇，其中包括顶级杂志Journal of the American Statistical Association, Annals of Statistics, Journal of Econometrics，以及权威杂志PLoS ONE, Test, Scandinavion Journal of Statistic。被评为2012年度复旦管院学术新星（唯一获得者）。为SCI杂志Journal of Business and Economic Statistics, Quantitive Finance, Journal of Non-parametric Statistics, Science in China匿名审稿人。