改进的RBF神经网络在可转化债券定价中的应用研究

发布时间：2018-01-03 18:44

本文关键词：改进的RBF神经网络在可转化债券定价中的应用研究　出处：《首都经济贸易大学》2017年硕士论文　论文类型：学位论文

【摘要】：可转化债券(又称可转换债券,以下皆称可转换债券)能在资本市场上被广泛的使用,根本原因还是由于其可转换性的特质。一般来说,可转换债券的价值可分为纯债券价值和期权价值以及转换价值。转换价值和纯债价值十分容易计算,而期权价值却是难以确定。所以,合理精准地预测可转换债券期权部分价值,合理确定利率和波动率,正确制定转债条款、以合理价格发行,能增加对投资者的吸引力,能极大的影响融资和投资额度,对国内可转换债券市场的发展具有重要意义。我国可转换债券的发展历史还比较短,使用Black-Scholes定价模型,其前提假设非常严苛,主观性和模糊性很强,因此并不能准确且高效的进行评估,这使得该模型在实际的应用中受到很大的限制。虽然很多学者结合了BP神经网络算法对之进行了改进和优化,也确实提高了可转换债券定价模型的准确性,但是由于BP神经网络自身的缺陷使模型仍然达不到最佳的效果。后来又有杨梁玉等学者引入了RBF神经网络的算法对定价模型进行一定的改进,RBF网络在收敛时具有较为优异的逼近性能,得到了更为精确的估值。由于RBF网络隐含层中心的个数、中心的位置、层基函数的宽度、网络权值都会一定程度地影响其网络性能,本文以前人学者的研究为基础,在RBF网络隐层基函数的宽度选择上,提出一种改进算法,该算法的基本思想是保持正交最小二乘算法训练好的RBF网络的中心位置不变的情况下,用最小二乘方法去再一次计算网络的权值向量,并以此来调整RBF网络的宽度值。最后本文通过国泰安数据库、Wind数据库、雪球财经以及中信证券交易系统中收集的可转换债券的样本数据,应用Matlab软件进行仿真,对比BP神经网络下的可转换债券定价模型与RBF神经网络下的可转换债券定价模型以及改进后的RBF神经网络下的可转换债券定价模型的仿真结果。实验表明:在可转换债券的定价中,改进后的RBF网络下的输出误差与时间都明显低于BP网络和RBF网络下的结果,从而得出改进后的RBF网络下的B-S定价结果更加有效的结论。这对于研究可转换债券的价值评估以及项目投资决策具有十分重要的意义。
[Abstract]:Convertible bonds (also known as convertible bonds, hereinafter referred to as convertible bonds) can be widely used in the capital market, the root cause is due to its convertible characteristics. Generally speaking. The value of convertible bonds can be divided into pure bond value, option value and conversion value. Conversion value and pure bond value are very easy to calculate, but option value is difficult to determine. Reasonable and accurate prediction of convertible bond option partial value, reasonable determination of interest rate and volatility, correct formulation of convertible bond terms and issuance at a reasonable price can increase the attractiveness to investors. Can greatly affect the financing and investment quota, the development of domestic convertible bond market is of great significance. China's convertible bond development history is relatively short. Using Black-Scholes pricing model, its premise assumptions are very strict, subjective and fuzzy, so it can not be accurately and efficiently evaluated. Although many scholars have improved and optimized the model with BP neural network algorithm, the accuracy of convertible bond pricing model has been improved. However, due to the defects of BP neural network itself, the model still can not achieve the best results. Later, some scholars such as Yang Liangyu introduced the algorithm of RBF neural network to improve the pricing model. RBF networks have better approximation performance and obtain more accurate estimates. Because of the number of the center of the hidden layer, the position of the center and the width of the layer basis function in the RBF network. Network weights will affect the network performance to some extent. Based on the previous research by scholars, an improved algorithm is proposed to select the width of hidden layer basis functions in RBF networks. The basic idea of the algorithm is to use the least square method to calculate the weight vector of the RBF network again, while the center position of the RBF network trained by the orthogonal least squares algorithm is not changed. And to adjust the width of the RBF network. Finally, this paper through the Cathay Pacific database data bank wind database, snowball finance and CITIC Securities Exchange system collected in the convertible bond sample data. The simulation is carried out with Matlab software. The simulation results of convertible bond pricing model based on BP neural network and RBF neural network and modified RBF neural network are compared. Indicates that:. In the pricing of convertible bonds. The output error and time of improved RBF network are obviously lower than those of BP network and RBF network. It is concluded that the B-S pricing results based on the improved RBF network are more effective, which is of great significance for the study of the valuation of convertible bonds and the investment decision of projects.
【学位授予单位】：首都经济贸易大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;F832.51

【相似文献】