July 30, 2025Open Access

Research on Option Pricing Prediction Based on Deep Learning

Key Points

The study explores how deep learning algorithms enhance option pricing accuracy compared to traditional methods.
LSTM neural networks outperform conventional pricing models like Black-Scholes in predicting option prices.
Two predictive models were constructed using historical data from SSE 50ETF options and evaluated with MSE and MAE.
Findings suggest that deep learning approaches can better capture market dynamics in option pricing.

Abstract

As a financial derivative with high leverage, options have prominent hedging and arbitrage capabilities. Traditional pricing methods for European options, such as the Black–Scholes pricing model, the Merton model, and the Heston model, rely on strict assumptions and simulate option price trends using stochastic processes. However, due to the discrepancies between these assumptions and actual market conditions, these traditional models often fail to reflect real-world option pricing accurately. Therefore, this paper adopts a data-driven approach using deep learning algorithms to simulate the option pricing process. Based on the classical Black–Scholes pricing theory, we explore the feasibility of using BP neural networks and LSTM neural networks for option pricing prediction. Using historical data of the SSE 50ETF options, we build two predictive models and use MSE, MAE, and R-squared as evaluation metrics to assess prediction accuracy. Experimental results show that the LSTM model significantly outperforms others in predicting the price of SSE 50ETF options.

Read Full Paperexternally

Demander à l'IA

Bookmark

View Full Paper