How AI Is Helping to Explain Stock-Price Moves

The earnings surprise is a long-standing tool that compares reported earnings per share against analysts’ forecasts to gauge how much a company beats or misses expectations, though it only explains about 5 percent of same-day stock moves.
Research by Ralph S. J. Koijen and Bradford Levy from Chicago Booth demonstrates that artificial intelligence (AI) models can significantly enhance the explanation of same-day stock movements, explaining about 17 percent compared to just 5 percent.
AI models in this study were able to uncover details from earnings calls that go beyond numbers, including insights like management commentary and forward guidance, as well as nuances in how results were discussed.
AI models provided human-readable explanations of their insights, which could offer financial researchers specific hypotheses about price movements and potentially apply to more affordable models as they advance.
Koijen and Levy acknowledge that their results are preliminary, as most variation in stock movements post-earnings remains unexplained. They have started a competition sponsored by Optiver to further improve predictive models.
The researchers note the challenging nature of predicting market movements, with decades of research explaining only 8 percent of the variation, leaving room for future improvement even with AI.