Design And Pricing Of Earthquake-related Bonds Based On Neural Network

With the increase in environmental pollution in recent years,the occurrence of large-scale disasters indicates that human society has entered a new era of disasters.As a risk bearer of catastrophic losses such as earthquakes,the insurance industry,due to the limited capital,effective preventive measures must be taken in the face of the strong business risks brought about by catastrophes.Therefore,the issuance of catastrophe bonds to allocate risks to the capital market as a cheaper and more efficient way has begun to gradually develop.Taking earthquake-related bonds as the main research object,on the basis of introducing the successful experience of domestic and foreign catastrophe bonds,explore new ideas for issuance of earthquake-related bonds,and combine China’s disaster data to find a catastrophe bond operation model suitable for the development environment of the domestic financial market.Bond pricing as a key step for the sustainable development of earthquake-related bonds in the market,is the focus of this paper.The empirical process of earthquake-related bonds pricing is divided into three parts: In the first place,collect information about earthquake disasters from1976 to 2019.According to the distribution characteristics of earthquake economic loss data with right-biased,sharp peaks and heavy-tailed,the relevant models of extreme value theory are used for preliminary exploration.Focusing on describing the tail shape of earthquake risk data,and constructing an extended Burr Ⅻ distribution model to fit economic losses.Research shows that compared with generalized extreme value distribution and generalized Pareto distribution.The capture ability is higher for the tail extreme value data,and the fitting accuracy of the overall data is better.Secondly,the generalized regression neural network is selected to provide a new idea for predicting the death toll from the earthquake.By analyzing the historical data of the earthquake,the 7 main factors affecting the death toll from the earthquake are summarized.The comprehensive index processed based on principal component analysis is taken as the input value,train the generalized regression neural network model,reduce the information redundancy,adjust the smooth coefficient to optimize the algorithm.Compared with the traditional BP neural network model,the mean absolute error and root mean square error of generalized regression neural network are smaller.The conclusion is that the generalized regression neural network has high prediction accuracy,short running time,strong generalization ability,and still has good applicability in a small sample environment,which can provide an effective reference for further research on the number of earthquake deaths.Finally,a two-tier trigger mechanism is designed for the principalprotected,part-principally insured,and principal total-loss earthquakerelated bonds,set the number of earthquake deaths as the bond investor’s principal loss ratio when the second trigger value is reached,and the cash flow discount model is used to calculate the issuance price.The LFC empirical model and Kreps pricing model are used to update the yield of earthquake-related bonds.The risk premium spread is measured by considering the expected loss of earthquake risk,unexpected loss,the regional division index of earthquake disaster,and the type of bond,and then the price is updated.The results show that the issue price of earthquake related bonds will rise with the expend of loss amount and fall with the decrease of bond yield.