Understanding Stock Market Reactions to Earnings Results
Unveiling the mystery behind stock market reactions to earnings results, the usual measure employed by investors and market analysts is the earnings surprise. This tool, in existence for several years, compares the reported earnings per share (EPS) against the consensus EPS forecasts from analysts. Despite its widespread use, the metric only accounts for about 5% of same-day stock movements.
Artificial Intelligence Enhancing Earnings Results Analysis
Recent research by Ralph S. J. Koijen and Bradford Levy from Chicago Booth suggests that artificial intelligence (AI) can enhance this percentage significantly, and offer better understanding of stock price movements. The AI models used in their research reportedly explained about 17% of same-day stock moves. The improvement in performance was due to models “learning” how to predict these movements, with the insights being documented into digital notebooks.
AI Models in Action: A Live Test
Instead of relying on simulations using historical data, Koijen and Levy performed a live test, thereby avoiding lookahead bias. The test involved three advanced models and covered nearly 2,000 earnings announcements made across various industries in late 2025.
The Insightful Approach of AI Models
The AI models provided written explanations that were evaluated, revealing that the models picked up on details in the earnings calls, not solely based on numbers. These details incorporated forward guidance from executives, management commentary, and even the language used in the call. Key insights from the models included the fact that good news was often already factored into a stock price, and the models did not overreact to strong results.
Human-readable Explanations and Future Prospects
Koijen and Levy believe that these human-readable explanations can provide researchers with specific hypotheses to test, understanding better what drives price movements. These explanations could potentially be incorporated into smaller, cost-effective models with similar results, suggesting a way forward for scaling these systems at a lower cost.
Invitation to a Competition for Model Submissions
Given the early stages of these results and the fact that most variation in stock moves post earnings releases remains unexplained, a competition has been launched, sponsored by trading firm Optiver. This competition invites individuals to submit their own model predicting price response to earnings-call information.
Despite being seen as a challenging task, Koijen asserts that forecasting market movements can be improved. As it stands, “decades of research can explain just 8 percent of the variation, but we believe this can be increased to 20 percent,” according to him.
Read More US Economic News