Price Stability in Regression Tree Calibrations
Insurance companies have started to collect more and more information about their individual policyholders.In car insurance this goes so far that GPS location data is recorded second by second(telematics data).This telematics data allows the insurance companies to evaluate the driving habits and driving styles of their individual car drivers.The complexity of this increasingly large data set requires sophisticated methods of analysis.Therefore,classical statistical methods such as generalized linear models are replaced by machine learning methods like regression trees,boosting machines and neural networks.A common drawback of many(data driven)machine learning methods is that they are not very stable under slight changes in the data.The purpose of this paper is to analyze this instability and we present methods to improve on this point.This is of particular interest in insurance pricing because individual premiums should not fluctuate too much in consecutive accounting years.
Insurance pricing Frequency modeling Machine learning Regression tree Boosting machine
Mario V.Wüthrich
Department of Mathematics 8092 Zurich,Switzerland
国内会议
广西桂林
英文
749-762
2017-07-19（万方平台首次上网日期，不代表论文的发表时间）