题目:Predicting the unpaid claims using aBayesian basis expansion model and Hamiltonian Monte Carlo method
报告人:高光远, 澳洲国立大学 商学院
时间:2016年4月27日下午1:30-3:00
地点:博萃楼会议室317
论文:学院内网
【摘要】We propose a Bayesian basis expansion modelwhich uses a natural cubic B-spline basis with knots placed at everydevelopment year to estimate the outstanding unpaid claims liability. Analogousto the smoothing parameter in a smoothing spline, shrinkage priors such asLaplace distribution and Cauchy distribution are assumed in the Bayesianframework. The advantages of a Bayesian basis expansion model include:accommodation of the tail development, quantification of the predictive distributionand incorporation of prior knowledge. For model inference, we use Stan toimplement the no-U-turn sampler (NUTS) which is an automatically tuned HamiltonianMonte Carlo method. We apply the proposed model to a simulated run-off triangleto illustrate the advantages of basis expansion models and to a real run-offtriangle data from WorkSafe Victoria to estimate the outstanding unpaid
claims liability of the doctor benefit.
【报告人简介】高光远, 现为澳洲国立大学统计学博士研究生,预计于2016年6月获博士学位,研究方向为保险精算。