AI surpasses financial analysts in predicting corporate earnings, reveals study
Artificial Intelligence (AI) is making significant strides in the financial services sector, according to a new study from the University of Chicago Booth School of Business. The research reveals that Large Language Models (LLMs), a specific type of AI designed to comprehend and generate text, can outperform some financial analysts in "predicting the direction of future earnings." The findings indicate that these models have an accuracy rate of 60.4%, which is 7% higher than the average analyst prediction.
AI's forecasting ability rivals humans
The research team employed a technique known as chain-of-thought prompting, which aids language models in performing intricate reasoning tasks by dividing them into smaller, manageable steps. Remarkably, the AI was given no additional information beyond the balance sheet and income statement data. Yet, it demonstrated an impressive ability to analyze these financial documents and predict future earnings accurately.
Step-by-step analysis outperforms financial analysts
The researchers stated, "our results suggest that GPT can outperform human analysts by performing financial statement analysis even without any specific narrative contexts." They emphasized that their findings underscore the importance of "human-like step-by-step analysis" that helps the model mimic procedures typically undertaken by financial analysts. The study also found that AI predictions were more valuable in situations where human biases or disagreements existed.
Humans and AI struggle with predictions for smaller firms
The study also suggests that both AI models and human analysts face difficulties when making predictions for smaller companies or those reporting losses. However, analysts are better at dealing with complex financial circumstances, likely due to the fact that they take into account "soft information" and context found outside of financial statements, which AI models currently cannot do. The authors concluded that their findings indicate potential for LLMs to democratize financial information processing and should interest investors and regulators.
