A research paper by the University of Chicago suggests that GPT-4, a generalized AI model, could be more effective in predicting a company’s future earnings than human experts. The researchers found that the AI model achieved an accuracy of 60% in its predictions, significantly higher than the 53% accuracy achieved by human analysts.
The experiment involved providing GPT-4 Turbo with a company’s balance sheet and income statement in a standardized format. The AI model was then asked to analyze these documents like a financial analyst. A technique called Chain-of-Thought (CoT) prompting was used to train the AI model to mimic the analysts. In order to ensure that the AI model relied solely on its analysis of the financial data, the researchers did not provide it with any additional information such as the company’s name, discussions about management, or other corporate data. They also kept the year the financial data came from hidden from GPT-4.
The researchers found the AI model performed about the same or even slightly better than specialized machine learning (ML) models frequently used by financial analysts to analyze numbers and make predictions. GPT-4 seems to benefit from its unbiased and dispassionate analysis, which is free of human sentiment and emotions that often lead to biased predictions.
The study also alludes to the broader issue of AI replacing traditional professions, including those that require high levels of expertise such as financial analysis. According to a 2020 study, actively managed investment funds were outperformed by the S&P 500 benchmark index, with only 11% of financial analysts in investment firms managing to beat the index. The researchers of the current study suggest GPT-4’s superior performance could potentially democratize financial information processing, making expert human financial analysts less relevant.
This research raises the potential of AI platforms, such as ChatGPT, to take the place of fund managers and other financial professionals, providing advice on buying or selling stocks. While it remains uncertain how GPT-4 manages to outperform human specialists in financial prediction, the study’s findings suggest replacing finance professionals with AI models may become reality sooner than expected.
