Research On The Application Of Improved RBF Neural Network In The Pricing Of Convertible Bonds

Convertible bonds(also known as convertible bonds,The following all called convertible bonds)can be widely used in the capital market,the root cause is due to their characteristics of conversion.Generally speaking,the value of convertible bonds can be divided into pure bond value and option value and conversion value.Conversion value and net debt value are easy to calculate,but the option value is difficult to determine.So,reasonable and accurate prediction of convertible bond option value,determine the interest rate and volatility is reasonable,the correct development of convertible bond,issued at a reasonable price,can increase the attractiveness to investors,can greatly affect the amount of financing and investment,is of great significance for the development of domestic convertible bond market.The history of the development of convertible bonds in China is relatively short,the use of Black-Scholes pricing model,the hypothesis is very harsh,subjective and fuzzy evaluation is very strong,and thus cannot accurately and efficiently,which makes the model is limited in practical application.Although many scholars with the improvement and optimization of the BP neural network algorithm does improve the accuracy of the pricing model of convertible bond,but because of the defects of BP neural network to model itself is still not up to the best effect.Later,scholars such as Yang Liang Yu introduced the algorithm of RBF neural network to improve the pricing model,and RBF network had better approximation performance in convergence,and obtained more accurate valuation.The number of RBF,center of the hidden layer of network center position,the width of the basis function layer,the network weights will to some extent affect the network performance,this paper studies the basis of previous scholars,in the RBF hidden layer is the width of the basis function selection,an improved algorithm is proposed,the basic idea of the algorithm is the center position and keep the orthogonal least squares algorithm trained RBF network under the same weight vector by the least squares method to calculate the network again,and then adjust the width value of RBF network.Finally this article through convertible bonds in the sample data collected by the Guo Tai An database,Wind database and CITIC Securities trading system,the application of Matlab simulation software,compared with BP neural network of the convertible bond pricing model and RBF neural network simulation results of the convertible bond pricing model RBF neural network bonds the pricing model and the improved.Experimental results showed that in the pricing of convertible bonds,the output error of the improved RBF network with time are significantly lower than the BP network and RBF network results,thus obtains the B-S pricing results of RBF neural network improved by more effective conclusion.It is of great significance for the study of the value evaluation of convertible bonds and the investment decision of the project.