A study on insurance claim risk models

Abstract

Claims constitute a very important issue in the insurance setting since claims can affect the profitability and operational stability of insurers. The insurers need to set aside money in order to pay the claims. This is very similar to the provision of bad debts resulting from delinquency of borrowers. Individual companies develop their own underwriting and monitoring systems in order to evaluate the risk of claims from potential and existing policyholders. Due to the differences in nature of various policies and the legal provisions mandating use of standard financial credit like system for the claim risk setting, many of the risk classification works do not focus on the claim, per se; they address claim fraud directly. These previous works provide the initiation for systemic tools for handling the issue. This thesis develops two insurance claim risk systems based on the concept of the financial credit risk assessment and the preliminary ideas by other scholars. The first system makes use of the basic Least Squares Support Vector Machine (LSSVM) System with variable and parameter setting variations in order to show how the insurance claim risk model can be used for classifying the potential policyholders into high claim risk and low claim risk. This binary classification has been widely used in loan/credit decisions. In the insurance setting, it can help the insurers ease the work in underwriting as well as making one-off decisions. The empirical data analysis found that age and policy variables (age of the first non-general insurance policy purchase, the number of riders/elements involved in the life insurance, and the number of insurance companies to seek non-general insurance) do differentiate low and high claim risk policyholders better. In addition, the simplex and grid search parameter searching methods provide similar classification results but the former seems to have a smaller performance variation in general. The policyholders risk profiles may change from time to time after the insurance has been purchased. Therefore, the second insurance behavioral claim risk model, based on the Fuzzy Markov System, is built. The behaviors of the policyholders through time are captured and used to determine the change in the claim risk status. It can help the insurers identify policyholders with changes in claim risk level through time and is also important for the claim reservation. The thesis develops insurance claim risk systems in both static and dynamic contexts. With the existing dataset for the model which incorporates the characteristics of various insurance policies, it can be extended to different policy settings with different planning horizons.

Date of Award	15 Jul 2015
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Chi Hang Stephen LEUNG (Supervisor) & Kin Keung LAI (Supervisor)