A recent study conducted by researchers from the University of Chicago Booth School of Business explored the capabilities of OpenAI's GPT-4 in financial statement analysis. The study aimed to determine if the large language model could analyze financial statements purely based on numbers, without any textual context typically found in quarterly earnings reports.
The researchers examined over 150,000 firm-year observations from about 15,000 companies between 1968 and 2021. They found that when given a "simple" prompt, GPT-4 scored slightly worse than human analysts with a 52% accuracy in predicting the direction of future earnings. However, when given a chain-of-thought command prompt, the model achieved a 60% accuracy, outperforming human analysts.
Despite the model's success in financial analysis, the researchers noted that financial analysis and prediction are complex tasks that require judgment, common sense, and intuition, which can stump both humans and machines. The study also pointed out that GPT-4 seemed to perform better at analyzing larger, more mature companies like Apple, as opposed to smaller, more volatile firms.
While the study provides insight into the potential for AI to enhance financial analysis, the researchers emphasized that AI is still complementary to human expertise. They noted that there are areas where human analysts excel, and that AI technology will continue to develop over time.
The question of whether AI could replace human financial analysts remains unanswered. The researchers highlighted the importance of human experts in certain areas where AI may fall short. The study offers a glimpse into the tools that may soon be available to financial analysts to more accurately assess a company's performance. OpenAI and Apple did not provide immediate comment on the study.