Can GPT Price Options?

Options are financial instruments that grant the holder the right, but not the obligation, to buy or sell an underlying asset at a predetermined price within a specified time frame. Traditional option pricing models, such as the Black-Scholes equation, depend on simplifying assumptions and struggle to effectively capture complex market dynamics. The study investigates the viability of using the Generative Pretrained Transformer (GPT) model to assess the value of stock options. We fine-tune the state-of-the-art GPT-3.5-turbo model by utilizing past option chain data and evaluate its precision in predicting option prices for well-known technology stocks including Apple (AAPL), Google (GOOG), and Microsoft (MSFT). Our model employs many features, including strike price, underlying price, days to expiry, and volatility index (VIX), to precisely predict option prices. Extensive examination undertaken at various degrees of moneyness and time horizons demonstrates that GPT models consistently surpass Black-Scholes in terms of both mean absolute error and root mean squared error. Out of all the models that were assessed, the 5-year data models exhibit the greatest overall accuracy. An error study suggests that pricing for options at the money has improved, but there are issues in appropriately pricing options that are highly in the money or out of the money. The GPT technique shows promise in leveraging transformers for computational finance applications.