AI Language Models Outperform Financial Analysts in Predicting Future Earnings

Artificial intelligence (AI) is making significant strides in the financial services sector, with research from the University of Chicago Booth School of Business suggesting that Large Language Models (LLMs), a type of AI trained to understand and generate content, outperform some financial analysts in predicting future earnings. The researchers employed chain-of-thought prompting, a technique that breaks down complex reasoning tasks into smaller steps, to train the language models. In particular, the study found that generative pre-trained transformers (GPTs), a type of LLM, achieved an accuracy of 60.4%, which is 7 percentage points higher than the average analyst prediction.

Interestingly, the researchers did not provide the language model with any narratives or context beyond the balance sheet and income statement. Despite this, the model’s ability to analyze financial statements and predict future earnings was comparable to the consensus forecasts made by analysts during their first month. The researchers noted that the results highlight the importance of “human-like step-by-step analysis” that guides the model in following the typical steps performed by analysts.

The study revealed that the language model’s forecasts were particularly valuable in situations where human biases or inefficiencies, such as disagreements, were present. However, it is important to note that the predictions made by GPTs were not flawless. They were more likely to be inaccurate when dealing with smaller firms, those with higher leverage ratios, firms that record a loss, or those with volatile earnings. This suggests that context plays a significant role when making predictions for smaller or more variable firms. While both GPTs and analysts face challenges in predicting for such companies, analysts tend to navigate complex financial circumstances better, as they factor in soft information and context beyond financial statements.

The researchers concluded that the findings highlight the potential for LLMs to revolutionize financial information processing, not just as tools for investors but also as decision-making aids. However, they cautioned that the performance of AI in real-world scenarios may differ from controlled experiments. The report emphasized that AI and human analysts are complementary rather than substitutes. “Whether AI can substantially improve human decision-making in financial markets in practice is still to be seen,” the authors noted. Nonetheless, the study’s results suggest that AI has the potential to democratize financial information processing and should be of interest to investors and regulators alike.

In a world where technology continues to evolve rapidly, the integration of AI into decision-making processes within the financial services sector has the potential to reshape the landscape. As AI algorithms become more sophisticated and capable, it is essential to strike the right balance between the strengths of human analysts and the capabilities of AI to achieve optimal decision-making outcomes.