Font Size: a A A

A New Hybrid Neural Network Option Pricing Model Study

Posted on:2022-03-19Degree:MasterType:Thesis
Country:ChinaCandidate:T GaoFull Text:PDF
GTID:2480306482968819Subject:Quantitative Economics
Abstract/Summary:PDF Full Text Request
An option is a contract that gives the holder the right to sell or buy a prescribed asset for a prescribed price at a prescribed time in the future.The earliest trading in options began in the late 18 th century in the American and European markets,and the unified,standardised trading of options contracts began with the opening of the Chicago Board Options Exchange in 1973.The advent of options has greatly enriched the capital markets,meeting the risk control and speculative needs of investors.Since 1973,options have been traded on an increasingly large scale and have played an increasingly important role in the financial markets.In this context,the research significance of option pricing has become more and more prominent,and the current option pricing models can be broadly classified into two categories: parametric models and non-parametric models.In recent years,with rapid advances in computer computing power,data-driven non-parametric methods have been increasingly applied to option pricing research,including neural network methods.Neural network methods are widely used for the following reasons: 1.parametric models determine the relationship between variables before they are fitted,and the resulting model is uniquely invariant,whereas neural network methods fit option pricing models based entirely on market data,and the resulting model is flexible and variable.2.parametric models make many harsh assumptions about the market,whereas neural network methods do not make these harsh assumptions at all.3.parametric models require a complex theoretical foundation to understand and innovate,and are often complex and difficult to understand,whereas neural network methods are often simple to understand and easy to use.The emergence of neural network option pricing models has undoubtedly promoted the development of option pricing theory.Although existing neural network option pricing models are open to a variety of innovations depending on the choice of input variables and output variables,most of them are based on the fully connected neural network framework and few studies have innovated on the neural network framework.In response,based on existing research on neural network option pricing,this paper constructs a new hybrid neural network option pricing model based on LSTM(Long Short-Term Memory)neural networks,fully connected neural networks and the Black-Scholes option pricing formula.The new model has the following characteristics: 1.the Black-Scholes option pricing formula is used as a guide to select key input and output variables.2.for the volatility variables that are difficult to identify in the input variables,instead of simply using historical volatility or prior period implied volatility,an LSTM neural network is introduced to exploit the volatility information embedded in the asset return series by using the LSTM's expertise in handling serial data.3.The new model chooses implied volatility as the output variable of the LSTM hybrid neural network option pricing model,and then uses the Black-Scholes model to calculate option prices on its basis.Compared with choosing option prices as the output variable,this method of using intermediate variables as the output variable can significantly reduce the complexity of the relationship between the input and output variables.To test the performance of the new models,in particular the positive role of LSTM neural networks and implied volatility outputs in option pricing.The paper also develops two benchmark fully connected neural network option pricing models,a benchmark Black-Scholes option pricing model and a benchmark LSTM hybrid neural network option pricing model.After building these models,the paper fitted and predicted each option pricing model with 50 ETF call option data,and further compared the accuracy of each model's option price prediction based on four average accuracy evaluation indicators and the Wilcoxon signed rank test.Empirical tests demonstrate that the new model outperforms both the benchmark fully connected neural network option pricing model and the benchmark Black-Scholes option pricing model in terms of option price prediction accuracy,as reflected by the positive role played by the LSTM neural networks and implied volatility outputs in option pricing.
Keywords/Search Tags:Option pricing, Implied volatility, Historical volatility, Fully connected neural networks, LSTM neural networks, Black-Scholes model
PDF Full Text Request
Related items