A Study On Financial Risk Warning Of Life Insurance Companies Based On Artificial Neural Network Model

China’s insurance industry started late, the regulatory system is still imperfect,while the insurance company without a sound internal control system and risk warningsystem, which led to inadequate risk and solvency problems certain degree of secrecy.The life insurance business conditions directly affect the survival and well-being ofpeople and the socio-economic stability, so the establishment of an effective earlywarning system is extremely necessary.Life insurance company’s financial risk can be divided into asset default risk,pricing risk, interest rate risk and liquidity risk. They run through the company’soperating, financing, investment and earnings for economic activities. These economicactivities in any one part of a problem can lead to cash flow abnormalities, and willultimately be manifested in the form of financial anomalies. Life insurance company’sfinancial risk early warning is utilization of internal and external information, andapply appropriate models and methods, by outputting the results to reflect the financialrisk indicators to alert managers to take appropriate risk management strategies. Thewarning model of early warning indicators and the selection is the key to establish anearly warning system. Indicators directly determine the reasonableness of the modeland warning effect, so select the appropriate indicators is essential. Early warningindicators can be objectively reasonable, comprehensive and dynamically reflect thecompany’s financial position. If you look only at the absolute level of the currentsolvency, it seems to be rather one-sided and hysteresis.This paper selected ninefinancial indicators, while the life insurance company’s capital adequacy, profitability,solvency and ability to grow five aspects of a comprehensive examination of the finalconsolidated a risk to judge the relative value of the life insurance company’s financialrisks. Early warning model is mainly traditional model and the new model. Traditionalearly-warning model includes Univariate Judgment model, Multivariate LinearJudgment model and Multivariate Logistic Regression model. The new model is aneural network model warning, and the two most popular models are Back Propagation(Back Propagation, BP) neural network model and Radial Basis Function (Radial BasisFunction, RBF) neural network model. Compared to traditional model,neural networkmodel has the advantage,which are artificial neural networks have learning adaptive,self-organizing capacity, quickly and efficiently and effectively deal with complexnonlinear problems, better fault tolerance and robustness. In the approximation of the unknown function, to achieve the same accuracy, RBF network parameters than theBP network needs less, and has a faster learning convergence speed. RBF neuralnetwork than the BP neural network can better simulate the biological nervous system.We select36life insurance companies’ data as the study sample, In2007-2009which operated in China.These data are unbalanced time-series cross-sectional data.We use34companies’ data from2007to2009, totaled in765, as the training set andthe36companies’ new data from2007to2010, totaled in1008, as a test set. In thispaper， we use CRITIC (Criteria Importance Through Intercrieria Correlation)weighting method to obtain the desired target vector network. After empirical research,we get the final conclusion. First, useing CRITIC risk weighting method on thecompany’s empowerment and scoring is an objective and reasonable. Second, RBFneural network prediction accuracy is high, relative error of prediction weremaintained at less than5%, while the network fault-tolerant capability and thegeneralization ability in the appropriate spread environment is strong, so the RBFneural network is an ideal risk predictive models. The last chapter is the conclusion ofthis article, in which the inadequacies of this article are these. First, the article does notintroduce qualitative indicators, although it can be integrated to some extent, reflect thecompany’s financial risk, but also a comprehensive place beneath. Second, RBF neuralnetwork model has been required to establish targets vector by quantified dependentvariable, which is the model itself defects. Third, during each of the prediction, wemust also "predict" past data, because of the different number of predicted data is notmeaningful comparisons. This research ideas can be summarized as follows. First,weintroduced life insurance company’s risk, and illustrated the need for early warning.Second, we introduced the commonly used traditional early-warning model, neuralnetwork model and part of the research methods, in order to show the neural networkmodel has the advantage. Specifically, we introduced the CRITIC weighting methodand normalization method. Fourth, the empirical part is that we pretreatment data inputmodel, observe and analyze the results.Finally we draw some conclusions. Fifth, thelast chapter is a summary of this paper.