A Study Of Option Price Prediction Based On LSTM Neural Network

Options have the functions of venture capital,value discovery,hedging and so on.They are important tools for investors to manage assets.However,due to unrealistic assumptions and omitted variables,traditional option pricing models will generate pricing systematic error and arbitrage error in the process of practical application.Therefore,the establishment of an effective option prediction model will play a crucial role in the steady development of financial market,and help investors in understanding options.Based on Black Scholes pricing model,LSTM neural network and investor sentiment,this thesis focusing on 50 ETF option establishes an option price prediction model.First of all,considering Black Scholes pricing model and option trading status,this thesis selects underlying security close price,option strike price,option remaining term,riskless rate,underlying securitiy historical volatility,option volume,option position,option settlement price and option opening price as input variables.At the same time,considering that domestic financial market is an emerging market and the supervision system is not sound,the irrational behavior of investors exists.This thesis constructs a composite index of investor sentiment to discuss whether the option price is affected by investor sentiment.Secondly,considering LSTM with multiple hidden layers has strong nonlinear approximation ability,and with memory unit can learn time series data,therefore this thesis uses LSTM to build option price prediction model.In order to explore whether LSTM neural network can get better forecast in the option price prediction,this thesis uses SVM,XGBoost,BP neural network to build constrast models,and choses MAE,MSE,MRE as accuracy evaluation criterion.Finally,the prediction model is optimized from the aspects of network structure and model input,for instance the number of hidden layers is increased and the price of BS theory is introduced as the input,so as to discuss whether the prediction accuracy of the model is improved.The conclusions of this thesis are as follow:Firstly,investor sentiment can have a significant impact on option price prediction,indicating that option investors will be influenced by irrational factors when making investment decisions.Secondly,the prediction accuracy of LSTM neural network is always better than that of BP neural network,SVM and XGBoost prediction model.Because the LSTM neural network model has a higher level of nonlinear combination operation capability,and the existence of gate mechanism helps the LSTM neural network model to learn the historical information of a long period,the LSTM neural network is more suitable for fitting the time series data of high dimension and local clustering.Thirdly,the model prediction error will be further reduced after considering the option theoretical price BS and increasing the number of hidden layers appropriately or changing the training algorithm,indicating that option pricing theory has guiding significance to option market price,and model structure is related to prediction accuracy.This thesis put forward an option price prediction model based on LSTM neural network with higher prediction accuracy,which is of guiding significance to help investors realize the irrational fluctuation of option price and make better use of information to predict option price.