| As China’s ageing problem becomes increasingly serious and the proportion of disabled elderly people is increasing year on year,the issue of long-term care for the elderly is gradually becoming a key concern for society.Long-term care insurance,as an important means of solving the care problem,is receiving more and more attention from the industry.However,the development of long-term care insurance in China is still immature,and the commercial long-term care insurance rate setting technology is backward and lacks data accumulation,so it accounts for a very low percentage of health insurance.In addition,the era of big data has stimulated the demand for accurate insurance pricing.As a new technology that has broken through several fields,this paper will explore the research on the pricing of long-term care insurance based on deep neural networks to achieve reasonable pricing for people with different characteristics,which has both theoretical and practical significance.The research content of this paper is as follows:(1)After defining the care status and comparing and analysing four classical long-term care insurance actuarial pricing models,an end-to-end long-term care insurance pricing model based on deep neural networks is proposed.The end-to-end model has the advantages of reducing the complexity of the pricing model,avoiding the accumulation of errors,and meeting the evolving requirements for accurate pricing of LTC insurance.(2)This paper uses data from four surveys of the China Health and Retirement Longitudinal Survey(CHARLS),and expands the number of variables to 22 based on the factors affecting care status selected by previous scholars,including basic information,lifestyle habits and past medical history of individuals..(3)This paper finds that the selected CHARLS data has a high sample imbalance problem and attempts to address this problem using a deep network model.The empirical evidence found that the model with sample imbalance considered was more predictive than the model without sample imbalance considered.The model parameters were then further adjusted to compare the performance of the model with different activation functions,number of hidden layers and number of neurons,resulting in the optimal model structure LTCmodel for predicting individual care status.(4)Based on the LTCmodel,all values of the 22 dimensions were traversed by age and gender groups,and the probability prediction box lines for healthy,mildly disabled,severely disabled and death status were plotted for different genders.The probability prediction box plots for health,mild disability,severe disability and death states were drawn for different genders.(5)The annual premiums for different characteristics of the insured were obtained by combining the LTC insurance product form and LTCmodel.For presentation purposes,the paper controls for the insured’s marital status,number of children,education level and history of smoking and alcohol abuse,and compares the rates for insured persons of different genders,ages,places of residence and the presence of certain types of illnesses.(6)Finally,the paper proposes countermeasures from two perspectives: commercial long term care insurance providers and social long term care insurance providers.The novelties of this paper are:(1)Proposing an end-to-end long-term care insurance pricing model based on deep neural networks.(2)Solving the sample imbalance problem in the database commonly used for long-term care insurance pricing.(3)Combining the LTC insurance product form and the final constructed deep neural network to determine a personalized LTC insurance rate schedule. |
