Estimation For Automobile Insurance Fraud Rate

As a special type of risk in the insurance industry,insurance fraud seriously erodes the profits of insurance companies,undermines the interests of honest applicants and affects the healthy development of the industry.In the west,where the insurance industry is more developed,insurance companies have regular claims inspection processes and sophisticated anti-fraud policies.In China,due to the lack of practice in insurance fraud management over a long period of time,fraudulent acts in various types of insurance are widespread,with auto insurance being particularly prominent.According to the data released by the Beijing Insurance Regulatory Bureau,the insurance loss caused by auto insurance fraud during the period from 2007 to 2010 was as high as 3.8 billion yuan,accounting for 20% of the total premium income;statistical data on insurance fraud cases show that domestic auto insurance fraud accounts for up to 80% of insurance fraud cases;according to internal estimates of the Chinese insurance industry and China Insurance Regulatory Commission statistics,at least 20% of auto insurance claims are fraudulent,and the national auto insurance fraud claims account for far more than the global average.The fraud rate estimate is one of the classic topics in the study of insurance fraud.This paper estimates the fraud rate of domestic auto insurance based on PRIDIT method.The PRIDIT method is suitable for processing data whose dependent variable is a categorical variable or an ordinal variable,and is suitable for fraud claim identification and fraud rate estimation under unsupervised conditions.However,in the specific application,we found a theoretical defect of PRIDIT method,which may lead to obvious estimation error.In this paper,the principle of mean square error is used to improve this method.It can be verified that under large sample conditions,the improved PRIDIT method is theoretically more complete and the corresponding estimation results are more reliable.By using the improved PRIDIT method,this paper conducts an exploratory study on the fraud rate estimation in the domestic auto insurance market.According to the actual characteristics of the automobile insurance claim data set provided by a large domestic insurance company,we divide it into two subsets(high-risk claim subset and low-risk claim subset)to estimate the fraud rate and compare them.The estimated results are in good agreement with the basic nature of the data set,avoiding the inconsistency of unresolved method estimates.Thesis contribution mainly includes two aspects.One is to improve the fraud rate estimation model based on the PRIDIT method,to overcome the theoretical defects of the model and to avoid possible estimation errors.The second is to use the improved model to study the fraud rate estimation of the domestic auto insurance market.The research methods and results are useful supplements to domestic related research and have important guiding value in anti-fraud practice.