Working Paper

American Option Pricing using Self-Attention GRU and Shapley Value Interpretation

• Construct four different machine learning models, including multilayer perceptron (MLP), long short-term memory (LSTM), self-attention LSTM, and self-attention gated recurrent unit (GRU) in comparison to the traditional binomial model.
• The results show that the self-attention GRU model with historical data outperforms other models due to its ability to capture complex temporal dependencies embedded in the historical data.
• Employ the SHapley Additive exPlanations (SHAP) method to interpret and analyze the prediction results of the self-attention GRU model with historical data, which provides insights that the current spot price, strike, and moneyness have the greatest impacts on predicting the current SPY call price.

Are AI Risks Priced in the U.S. Stock Market? Evidence from Financial News Factors

• Construct four domain-specific AI-risk news factors from 7,787 Wall Street Journal articles published between 2016 and 2025, combining latent Dirichlet allocation (LDA) with the Domain Taxonomy in the MIT AI Risk Repository.
• Find that only the taxonomy-mapped Misinformation factor (D3) is robustly priced in U.S. stock returns. Its high-minus-low beta portfolio earns monthly risk-adjusted alphas of 0.49%–0.57%, with a positive price of risk across a broad range of specifications.
• Narrative analysis links prominent D3 peaks to AI governance and policy developments. Together with higher investment rates among high-D3-beta firms, this evidence suggests that the AI-policy environment is a plausible channel underlying the pricing results.