Insolvency Prediction Of Life Insurance Company Via SVM

With a broadened business scope,the insurance industry has developed into an important force in the current society.However,the insurance industry is a high-risky industry which includes credit risk,liquidity risk,interest rate risk,underwriting risk and many other risks.In the US A.M.Best rating database,703 insurance companies went bankrupt from 1978 to 2009,accounting for 14.17% of the total number of insurers during this period.In addition to the risks inherent in the insurance industry,there are also risks in the environment.For example,huge natural disasters and man-made disasters.Therefore,supervision of the insurance industry is very necessary.As the largest insurance market.the U.S.established the National Association of Insurance Supervisors which has a database of financial information for all insurance companies across the country.These databases provide the information basis for supervision and management of the United States insurance regulatory agencies.Also,Chinese insurance supervision and management system is constantly undergoing innovation.The relevant regulations were promulgated.On one hand,the country has strengthened supervision of the insurance industry.On the other hand,the academic world has also put a lot of energies in the study of the solvency prediction of insurance companies.Many different methods have been proposed since 1970 s,varying from linear model to non-parametric model.Among those models,support vector machine has attracted many attentions from both industry and academia due to its excellent performance in prediction.It is a new machine learning method based on the optimization and does a quite good job in the binary classification.It has been successfully applied in many real applications.Support vector machine has been introduced into the life insurance company's financial crisis forecasting field in some foreign works,but there are few related domestic researches.I applied this combination method based on the support vector machine to life insurance company's financial crisis forecast to judge the effectiveness.In this thesis,I used three methods including discriminant analysis,logit model and support vector machine to predict the insolvency of insurance companies.Logit model and discriminant analysis are used as comparison methods.Moreover,the variable selection via Information Value(IV)was introduced to improve the prediction accuracy.I also compared SVM with the combination of SVM and IV.It is worth pointing out that the corresponding data in the thesis was from ORBIS Insurance Focus' s insurance company database.After preliminary data compilation,the three-year financial data of 450 U.S.life insurers recorded during the period from 1997 to 2015 were selected.347 were solvency companies and 103 were insolvency companies.Combining the dual considerations of commonly used financial variables in the literature,and data quality,12 financial indicators were selected.When considering kernel functions,Gaussian kernel,polynomial kernel,and sigmoid kernel were used for training.When training the model,a grid search was used to get the parameters.The three-year financial data from American life insurance company was used to in the numerical experiment.The results show that the support vector machine with Information Value achieves the best prediction results,followed by the common support vector machine,logit model,and discriminant analysis.Finally,the paper compared the data of different years and found that the logit model is the most stable one while the support vector machine is the most accurate one.