基于半连续两部模型的保险损失预测

Prediction of insurance loss based on semicontinuous two-part model

【目的】提高保险领域中保单累积损失预测的准确率。传统的Tweedie回归模型只能对非零均值建立回归模型,却不能对零概率建立回归模型,从而导致该模型的拟合效果并不理想。【方法】考虑到保单损失数据中往往包含着大量的零索赔,此时可视其为一种半连续型数据。因此,基于半连续两部模型,并考虑到累积损失中非零连续部分的分布类型,提出3种不同的累积损失预测模型,并结合一组实际损失数据进行模型对比分析。【结果】与Tweedie回归模型相比,本研究所提出的半连续两部回归模型的赤池信息准则值(Akaike information criterion,AIC)和贝叶斯信息量准则值(Bayesian information criterion,BIC)更小,具有较好的拟合效果。【结论】本研究结果可为保险领域中的保单累积损失预测提供参考。

[Objective]It is imperative to improve the prediction accuracy of policy accumulated loss in the insurance field.The traditional Tweedie regression model can only establish a regression model for non-zero mean value,but not for zero probability,so the fitting effect of the model is not ideal.[Method]Considering that the policy loss data often contains a large number of zero claims,it can be regarded as semicontinuous data.Therefore,based on the semicontinuous two-part model and considering the distribution type of non-zero continuous part of accumulated loss,three different accumulated loss prediction models were proposed,and a set of actual loss data was combined to make a comparative analysis of the models.[Result]The results show that,compared with the Tweedie regression model the AIC(Akaike information criterion)and BIC(Bayesian information criterion)values of the two semicontinuous regression models are smaller,which have better fitling effects.[Conclusion]The results can provide a reference for the prediction of policy accumulated loss in the insurance field.