GPT-4 outperforms human analysts in predicting corporate profits, University of Chicago study finds
Researchers at the University of Chicago have found that large language models (LLMs) can analyze financial statements with accuracy that not only matches but often exceeds that of human analysts. The groundbreaking findings, detailed in a working paper titled “Analyzing Financial Statements with Large Language Models”, have significant implications for the future of financial analysis and decision-making, Venture Beat reports.
The study focused on GPT-4, a large language model developed by OpenAI, and tested its ability to analyze corporate financial statements and predict future earnings growth. It is noteworthy that GPT-4 outperformed human analysts even when presented with only standardized, anonymous balance sheets and income statements without any textual context.
“We find that the prediction accuracy of the LLM is on par with the performance of a narrowly trained state-of-the-art ML model,” the authors write. “LLM prediction does not stem from its training memory. Instead, we find that the LLM generates useful narrative insights about a company’s future performance.”
In the study, GPT-4 outperformed human analysts, who typically predict with an accuracy of 53-57%. The researchers used a novel method that incorporates “thought chain” cues to guide the AI’s reasoning, helping it mimic the analytical process of financial analysts by identifying trends, calculating ratios, and synthesizing information to generate forecasts.
“Taken together, our results suggest that LLMs may take a central role in decision-making,” the researchers conclude.
They attribute GPT-4’s success to its broad knowledge base and ability to recognize patterns and business concepts, which allows it to make intuitive judgments even with incomplete information.