保险公司异常赔款数据对准备金估计的影响研究

发布时间：2019-04-03 21:16

【摘要】：未决赔款准备金作为非寿险公司最大的负债项目,其计提水平将直接影响保险公司的盈利能力、偿付能力和产品定价。在所有准备金估计方法中,链梯法思路明确,方法简单,是最常用的确定性方法之一,但这种方法对原始数据有非常严格的假设条件且无法完成统计检验。因此,研究人员考虑在随机模型的范围内建立准备金估计模型,这其中发展最为完善的就是广义线性模型。广义线性模型因其具有适宜于保险精算的良好特性,为精算学的发展提供了有力的工具。而以此为基础的两阶段广义线性模型通过引入理赔次数变量,又实现了对广义线性模型的拓展。在末决赔款准备金的估计过程中,由于突发事件、信用欺诈等问题所造成的异常赔款数据时有发生,这些数据往往会使准备金的估计结果产生较大的偏差,并最终影响非寿险公司的运营决策。因此,文章通过对多种常用准备金估计方法的比较,以研究不同模型下异常赔款数据对准备金估计结果的影响程度,并尝试建立更加合理、稳健的准备金估计模型,为保险公司准确计提准备金提供参考。基于对链梯法和的广义线性模型的改进,本文引入两种更加稳健的模型：稳健链梯法和稳健广义线性模型,并在两阶段广义线性模型和稳健广义线性模型的基础上提出了两阶段稳健广义线性模型。同时,本文对模型的估计结果利用Bootstrap方法进行了统计检验。通过实证分析发现,稳健估计方法对原始数据中的异常值起到了很好的规避作用,大大降低了异常值对估计结果的影响,保险公司将可以得到与原始数据中不存在异常值时类似数量的准备金。并且,通过链梯法与稳健链梯法的比较,可以迅速的定位异常值出现的位置,方便保险公司精算人员据此探究异常数据出现的原因并对原始数据做出修正。
[Abstract]:As the largest liability item of non-life insurance companies, the amount of outstanding indemnity reserve will directly affect the profitability, solvency and product pricing of insurance companies. Among all reserve estimation methods, chain ladder method is one of the most commonly used deterministic methods because of its clear train of thought and simple method, but this method has very strict assumptions on the original data and can not complete statistical test. Therefore, researchers consider establishing a reserve estimation model within the scope of stochastic models, among which the most perfect development is the generalized linear model. The generalized linear model provides a powerful tool for the development of actuary because of its good characteristics suitable for actuarial insurance. The two-stage generalized linear model based on this model extends the generalized linear model by introducing the variable of claim number. In the process of estimating the final compensation reserve, the abnormal compensation data caused by unexpected events, credit fraud and other problems often occur, and these data often cause a large deviation in the estimation result of the reserve. And ultimately affect the operation of non-life insurance companies decision-making. Therefore, through the comparison of many common reserve estimation methods, this paper studies the influence degree of abnormal compensation data on reserve estimation results under different models, and tries to establish a more reasonable and robust reserve estimation model. Provide reference for the insurance company to accurately calculate the reserve. Based on the improvement of the generalized linear model of the chain ladder method, this paper introduces two more robust models: the robust chain ladder method and the robust generalized linear model. Based on the two-stage generalized linear model and the robust generalized linear model, a two-stage robust generalized linear model is proposed. At the same time, the Bootstrap method is used to test the estimated results of the model. Through the empirical analysis, it is found that the robust estimation method has a good evading effect on the abnormal values in the original data, and greatly reduces the influence of the outliers on the estimation results. Insurers will be able to obtain reserves similar to those in the original data when there are no exceptions. By comparing the chain ladder method with the robust chain ladder method, the position of the abnormal value can be located quickly, and the actuarial staff of the insurance company can explore the reason of the abnormal data and revise the original data according to the comparison between the chain ladder method and the robust chain ladder method.
【学位授予单位】：天津财经大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F840.3;F224

【参考文献】