管理工程学报2018，Vol.32Issue(3)：142-149,8.DOI:10.13587/j.cnki.jieem.2018.03.017
基于Heston模型和遗传算法优化的混合神经网络期权定价研究
Option pricing model by applying hybrid neural network based on heston model and genetic algorithm
摘要
Abstract
Option is an important kind of derivative securities, and it has an unequivalent property of rights and obligations. Because of this, option is a very favorite investment instrument for investors. For other derivatives, option is also faced with the problem of pricing difficulty. In 1973, financial engineering field entered a golden age because developing the Black-Scholes option pricing formula was a major milestone. The number of studies on derivatives’ pricing is increasing, and financial innovations are made one after another. These two trends have directly led to the rapid development of derivatives market in recent decades. However, how to price efficiently and accurately is still the consistent goal for financial experts. The innovation of financial tools in China has long been inadequate, particularly in the trading of financial derivatives, which directly affects the pace of financial reforms and open market. One side of reasons is the lack of sound market system and environment. The other side of reasons is the lack of an effective theory and pricing tools. In February 9, 2015, the Shanghai stock exchange carried out a pilot trading of stock option and started the transactions of Shanghai 50ETF option. This means that Chinese investors will have a new hedging investment--option. In options trading, the options’ fair value is very important for both sides of transactions. With more and more frequent transactions, the accurate pricing model is very important for the validity of the options market. The traditional method of pricing is divided into parameter models (such as BS model, Cev model, Heston model, etc.) and non-parameter models (such as artificial neural network model). However, each of them has its advantages and disadvantages. Now the frontier theory combines parameter models and non-parameter models to design a hybrid neural network model. The conventional hybrid neural network for option pricing is based on Black-Scholes model. While the Black-Scholes model has very strict assumptions, it does not accord with the actual situations of financial market. In this paper, we use Heston model (a kind of stochastic volatility models) to replace BS model in the hybrid neural network and use BP Artificial Neural Network to fit the difference between actual options data and option prices of Heston model in order to improve pricing accuracy. In addition, BP algorithm has the disadvantage of slow convergence and is easy to fall into local minima. Thus, we use genetic algorithm (GA) to optimize the structure of hybrid neural network. A new option pricing model is established by applying hybrid neural network and genetic algorithm based on Heston model. The pricing accuracy has been demonstrated using the actual Hongkong's Hang Sheng Index (HSI) options and Shanghai 50ETF option. This hybrid approach is shown to provide greater accuracy than either conventional model or the hybrid neural network based on Black-Scholes model. It is a great valuable reference for China option market to trade option pricing. The article is organized as follows: The first chapter introduces the purpose and significance of the topic, studies results of domestic and foreign researchers, and discusses the main research content. The second chapter firstly introduces the Heston model and the derivation process of its closed-form solution, as well as the equivalent martingale measure. The number of model parameters is reduced from six to five. The third chapter describes the calibration theory for Heston model in detail, then compares the calibration--nonlinear least squares method with the adaptive simulated annealing method. Considering the computing time and accuracy, the nonlinear least squares method is finally selected. After combining the calibrated Heston model with BP neural network and genetic algorithm, we establish a BP hybrid neural network based on Heston model and a BP hybrid neural network based on Heston model optimized by genetic algorithm. The fourth chapter is an empirical introduction of Hongkong's Hang Sheng index option and Shanghai 50ETF option., including data description, selection of evaluating indicators, predicted results of the model, performance comparison between different models, and interpretation of results. The fifth chapter includes some analysis and discussion. Future researches are suggested in the end. In conclusion, Heston model is better than BS model. Thus, stochastic volatility assumption is more consistent with the actual market. A hybrid neural network model based on Heson model can receive higher pricing accuracy than other models.关键词
Heston模型/Black-Scholes模型/遗传算法/混合神经网络/期权定价Key words
Heston model/Black-Scholes model/Genetic algorithm/hybrid neural network/Option pricing分类
管理科学引用本文复制引用
张丽娟,张文勇..基于Heston模型和遗传算法优化的混合神经网络期权定价研究[J].管理工程学报,2018,32(3):142-149,8.基金项目
教育部人文社会科学青年基金资助项目(10YJC790380) （10YJC790380）
上海哲学社会科学规划资助青年项目(2011EJB002) （2011EJB002）
上海市教委研创新项目(14YS003) （14YS003）
湖南省国际经济与国际工程管理研究基地招标课题 （）
上海大学经济学院创新课题 （）
