Claims Reserving Based On Matrix Completion And Tweedie Network Algorithms

Claims reserving is an important liability of insurance companies.Accurate and reasonable accrual of claims reserving is one of the important preconditions for an insurance company to be able to develop healthily and sustainably.On the one hand,whether the provision of claims reserving is reasonable can reflect the operating performance of an insurance company.Also,it can provide a scientific and effective basis for company's reasonable operation.On the other hand,whether the provision of claims reserves is sufficient can seriously affect the solvency and anti-risk ability of an insurance company is also a basic standard for the overall evaluation of insurance company by regulators.Therefore,the assessment of the reserve for claims is of great significance to insurance companies.In this article,we focus on the prediction of claims reserving for non-life insurance companies.According to its display form,the claim data can be divided into aggregated data and individual data.For aggregated data,we consider the prediction problem from the perspective of machine learning and convert the problem of the prediction lower triangle into the matrix completion,so that we propose the triangle-completion algorithm for aggregated claims based on matrix factorization and the triangle-completion algorithm for aggregated claims based on K-NN and obtain more accurate prediction results than the traditional chain ladder method.For individual data,we consider the distribution information of the claim data and introduce an improved neural network model for claims reserving.A new training process is designed by taking advantage of the widely used Tweedie compound Poisson model,which obtains a more accurate result.