Vehicle Risk Claim Strength Prediction Based On HGLM And SHGLM Models

After the reform and opening-up,the living standard of our people has been improved,which makes the sales volume of cars increasing day by day.This is a big step forward for the insurance industry under the current situation.In recent years,it has become an indispensable investment and expenditure for car owners.The most important concerns for the purchase of vehicle insurance are:auto insurance price,claim strength and auto insurance rates.In order to determine the price of auto insurance and the strength of claims fairly,it is an essential portion of studying the characteristics of risks and the factors that affect the risk of vehicles.This paper explores the model applicable to auto insurance data for the purpose of predicting the strength of claims.The earliest model used to predict the strength of claims is a generalized linear model,which makes a new breakthrough in predicting the strength of claims for the loose conditions of the dependent variable Y.However,the generalized linear model requires the data to be from independent random variables.It is difficult to satisfy this condition in the actual data.Therefore,the random effect is added to the generalized linear model,and the obtained generalized linear mixed model can be fitted to a certain time,repeated measurement of auto insurance data.In recent years,the research on hierarchical generalized linear models has been gradually developed.It extends the distribution of random effects on the basis of the former,making the model suitable for variable data.On the foundation of the above,considering the regional factors in the auto insurance data,the regional correlation matrix is introduced,and the spatial stratified generalized linear model obtained provides a new possibility for the prediction of claim strength.Firstly,this paper initially deals with the auto insurance business data obtained from the climb,and dissects the all elements affecting the auto insurance claims,and then establishes the generalized linear model,the generalized mixed linear model,the hierarchical generalized linear model and the spatial generalized linear model.The business data is analyzed and tested separately to fit the claim amount.Comparison of the fitted value and the real value.Finally,the fitting results of various models are compared by the sum of residuals,the coefficient of determination,the value of the likelihood function,etc.,and the three models with random effects are better than the generalized linear model,and the applicable data of every model are diverse.The hierarchical generalized linear model and the spatially hierarchical generalized linear model have the best fitting effect.The fitting value of the spatially stratified generalized linear model is close to the actual value.It can be seen that considering the spatial factor of the data is a new idea for predicting the amount of the claim,especially for the data with high regional similarity.The article also presents the features of every mould and the applicable data types to provide a choice for the fit of different data.