Screening And Auditing Strategies For Fraudulent Claims Under CSV Framework

This thesis sets up a two-phase detecting process for fraudulent claims under CSV(costly state verification)framework at the background of true-life insurance company.Based on this,the optimal detecting model as well as a suspicious claim characteristic index model are built.The optimal detecting model shows that insurance company's ordinary using SD(Selective Detection)scheme is not always the best choice.Under certain circumstances,WD(Without Detection)scheme or CD(Complete Detection)scheme may be the better alternatives.In addition,the optimal detecting strategies about the average investigation cost DC when CD scheme is operated and the optimal detecting matrixD(p,C)about the auditing rate and the average investigation cost when SD scheme is put into effect are given in the thesis.By using MATLAB analogue simulation,it also shows more clearly about different detecting schemes and strategies on the influence of model parameters such as fraudulent rate,average sum of claim,screening result and auditing ability.The other important thing should be auditing rate,which is one of the decision variables in the SD scheme,controlling the feasibility of detecting schemes.If insurance company could find out the most suspicious claims and send them to be audited again successfully,then we can see that insurance company will save cost and time as well as improve their auditing efficiency.Therefore,for the screening process,the thesis builds up a suspicious claim characteristic index model to improve the effectiveness of screening.For better using of the screening model,empirical research method to calibrate and analyze the model with automobile insurance claims data from real life is tried in this thesis.All the data originates from the insurance company,all the claim characteristics comes from the company's internal system which is aiming to find out fraudulent claims.The results say that appropriate characteristics index can help a lot to distinguish suspicious claims from normal ones.One important point should be remaindered that integrity of the data from insurance company influence model's predicting capacity.The thesis contributes on the following three points.Firstly,it builds the SD scheme for fraudulent claims detection by combining the classical CSV framework with true-life insurance company management process.Secondly,by using screening model and detecting model to describe screening phase and auditing phase respectively,we gain the optimal detecting strategies with controlled conditions.Thirdly,the theory results could provide insurance company with necessary theoretical foundation as well as methodological support when they are making anti-fraud management strategies or in the need of realizing total cost minimalistic target.