主讲人：Prof. Chenlei Leng (新加坡国立大学)

时 间： 11月8日（周二）16：00-17：00

地  点：数学科学学院 312 教室

备  注：摘要When a parametric likelihood function is not specified  for a model, estimating equations (EEs)  provide an instrument for statistical inference. Qin and Lawless (1994) illustrated that empirical likelihood (EL)  makes optimal use of the EEs in inferences for fixed (low) dimensional unknown parameters. In this paper, we study  EL  for general EEs with growing (high) dimensionality and propose an EE-based penalized empirical likelihood (PEL) approach for parameter estimation and variable selection. Theoretically, we quantify the asymptotic properties of EL and PEL. We further show that the PEL  has the Oracle property. Namely, with probability tending to one, PEL identifies the true sparse model. In addition, the efficiency of the estimated nonzero coefficients is optimal. The performance of the proposed PEL  is illustrated via four simulated applications and a data analysis. This is joint with Chengyong Tang.