驾驶行为保险的风险预测模型研究

Risk Predicting Models of Usage-Based Insurance

随着车联网技术的发展和我国商业车险费率改革的不断深化,有关驾驶行为保险的研究受到越来越多的关注。驾驶行为保险主要从车联网数据中提取驾驶行为风险因子为汽车保险定价。与传统的风险因子相比,驾驶行为风险因子与保险索赔之间具有更强的因果关系,能够有效改善车险费率厘定结果的准确性和合理性。汽车保险的定价模型包括出险概率模型、索赔次数模型、索赔金额模型和累积损失模型,在实际应用中,需要根据数据的特点选择不同的模型组合来厘定汽车保险的费率。本文主要基于从车联网数据中提取的驾驶行为风险因子建立出险概率的预测模型,具体包括随机森林模型、神经网络模型和XGBoost模型,探讨它们在出险概率预测中的应用效果,并与传统的Logistic回归模型进行比较。实证研究结果表明,与传统的承保风险因子相比,驾驶行为风险因子对于出险概率的预测更加重要和有效;与传统的预测模型相比,XGBoost模型对于出险概率的预测能力更强。基于XGBoost模型的预测结果,可以对驾驶行为风险因子的相对重要性进行排序,从而为驾驶行为保险的承保、定价和理赔提供参考依据。

With the development of telematics technology and the reform ot commercial automobile insurance rate- making in China, the research on UBI is receiving more attention. UBI mainly extracts driving risk factors from telematics data and uses them to determine the automobile insurance premium rate. Compared with traditional risk factors, driving risk factors have astronger causal relationship with insurance claims, thus can effectively improve the accuracy and rationality of automobile insurance premium rating. Automobile insurance pricing models include claim probability models, claim frequency models, claim severity models, and aggregate loss models. In actual prac- tice, different combinations of these models may be applied to determine premium rates. This paper mainly used driving risk factors from telematics data to build claim probability predicting models,which included random forest model, neural network model and XGBoost model. The paper explored the effects of these models in claim probabili- ty prediction and compared them with the traditional Logistic regression model. The empirical study showed that, driving risk factors were more important and effective than traditional underwriting risk factors, and the XGBoost model obtained the best performance in claim probability prediction. Based on the prediction of the XGBoost model, we can arrange driving habit risk factors in an order of importance, and provide reference for the underwriting, ratemaking, and claim settlement of UBI.