Financial Derivatives Price Prediction Based On Deep Learning Network Method

The financial derivatives market is an inseparable and important part of the capital market.As one of the most active financial derivatives in the world,options are widely concerned by academics and financial researchers for their pricing and valuation.In recent years,a large number of scholars have proposed the use of nonparametric models to estimate option prices and used them for empirical research.Based on these studies,this paper innovatively proposes a method that combines the parameter partial differential equations of option pricing with deep learning models to realize the estimation of option price.Firstly,the classical BS partial differential equation is transferred into the physical information neural network as a control condition,and the PINN_BS model is cleverly constructed.Secondly,In order to explore the feasibility of its model,according to the assumption of the BS pricing formula,a virtual option is randomly simulated by using the method of data generation,and the PINN BS model is used to predict the pricing of this virtual option,and the comparison analysis with its real value is carried out.Finally,the results confirm that the model is more accurate for the price prediction of virtual options.In the empirical part,The PINN_BS model is also applied to the option observation data of 50ETF stock index in the real market,and the prediction results of the model in the actual observation data are compared with the prediction results of several nonparametric option pricing benchmark models.According to the analysis results of the comprehensive evaluation indicators,it is shown that the PINN_BS model is more accurate in predicting option prices than the MLP benchmark model and the LSTM benchmark model.Based on this research,this paper also explores adding the attention mechanism to the LSTM benchmark model to predict option prices.Empirical evidence shows that compared the current best deep network models in the field of option price forecasting.The PINN_BS model can still achieve better prediction results and model fitting effects.