Option Pricing Based On Deep Learning And Application Research

As one of the most frequently used financial derivatives,options play an irreplaceable role in risk management,hedging,speculation,and arbitrage.In option trading,the price of the option has a direct impact on the specific income of both parties to the transaction.Therefore,how to reasonably use mathematical models to determine the price of options is the most important core issue here.The most classic pricing model is the Black-Scholes model,but the classic model also has shortcomings,for example,it cannot perfectly fit the changes in the price of bid assets in the actual market.At the same time,option investors generally rely on past investment experience when investing,and are more likely to be affected by their own emotions in the process of choosing investment strategies.Therefore,it is of great significance to establish an option pricing model that is more in line with the changes in the underlying asset price in the financial market and an investment strategy model that can learn independently and make automatic decisions.This paper studies the option pricing model and option investment strategy model,the main work is as follows:Firstly,the existing data-driven models for pricing European options are mainly based on two methods: deep learning and ensemble learning.The method based on deep learning will generate a large number of hyperparameters;The effect of the method based on ensemble learning on data feature extraction is not ideal.This paper combines the advantages of the two types of methods,and proposes a European option pricing model based on deep ensemble learning,adopts the framework of embedding modular ensemble learning into the network learning structure,uses the ensemble learning method to reduce hyperparameters,and uses the network learning structure to retain the effect of data feature extraction.The results show that our proposed model has a smaller error in pricing European options.Secondly,in view of the uncertain exercise time of American options,deep learning methods cannot be well applied to the problem of American option pricing.This paper based on the idea of BAW method,a hybrid American option pricing model based on deep learning and parameter method is proposed.Consider the pricing of American options as the sum of the European option value and premium,use Attention-LSTM to price the European option part,use the BAW method to approximate the premium,and combine the advantages of deep learning methods and parameter methods to improve the overall prediction accuracy.Thirdly,for the problem of option investment strategy selection,this paper introduces the idea of deep reinforcement learning into the field of option investment strategy research,and proposes an option investment strategy model based on deep reinforcement learning.Two different neural network methods are used to extract market data features,and then the features are fused;based on the reinforcement learning decision module,the Double DQN and Dueling DQN methods are combined to solve the problem of Q value overfitting and improve the convergence speed of the model.