| 基于广义线性模型与贝叶斯模型的 UBI车险索赔频率预测 |
| Prediction of UBI claim frequency based on integrating generalized linear model with Bayesian model |
| 摘要点击 41 全文点击 0 投稿时间：2026-01-29 修订日期：2026-06-10 |
| 查看/发表评论 下载PDF阅读器 |
| 中文关键词 车险费率厘定；驾驶行为；后验费率厘定；贝叶斯模型 |
| 英文关键词 auto insurance pricing; driving behavior; posterior ratemaking; Bayesian model |
| 基金项目 国家自然科学基金项目（面上项目，重点项目，重大项目） |
| 投稿方向 风险管理与精算 |
| 作者 | 单位 | 邮编 | | 房婷婷 | 新疆财经大学 | 830012 | | 高光远* | 中国人民大学 | 100872 |
|
| 中文摘要 |
| 现有 UBI 车险定价模型常将驾驶行为变量作为固定协变量, 这种建模方法存在两个局限: 一是难以应对新保单缺少车联网数据的问题; 二是无法刻画续保人未来驾驶行为的不确定性. 本文提出了一种联合广义线性模型（GLM）与贝叶斯模型的索赔频率预测框架, 它们分别预测了基于传统风险因子的先验基准索赔频率与基于危险驾驶行为次数的后验索赔频率调整系数. 对于新保单, 只需应用 GLM 预测先验基准索赔频率; 对于续保保单, 还需应用贝叶斯模型得到调整系数, 进而得到后验索赔频率预测. 实证分析表明, 在续保保单的后验索赔频率预测任务上, 本文提出的方法优于将驾驶行为变量视为固定协变量的方法. |
| 英文摘要 |
| Existing UBI pricing models often treat driving behavior variables as fixed covariates. This modeling approach has two limitations: first, it struggles to address the lack of telematics data for new policies; second, it fails to capture the uncertainty in future driving behavior of renewal policyholders. This paper proposes a claim frequency prediction framework that integrates a Generalized Linear Model (GLM) with a Bayesian model. The framework separately predicts the prior baseline claim frequency based on traditional risk factors and the posterior claim frequency adjustment coefficient based on risky driving behavior. For new policies, only the GLM is applied to predict the prior baseline claim frequency. For renewal policies, the Bayesian model is additionally used to obtain the adjustment coefficient, thereby yielding the posterior claim frequency prediction. Empirical analysis demonstrates that the proposed method outperforms the approach treating driving behavior variables as fixed covariates in predicting the posterior claim frequency for renewal policies. |
| 关闭 |
