Researchers find GPT-4 can outperform humans in predicting company earnings
2 min. read
Published on
Share this article
Improve this guide
Read our disclosure page to find out how can you help MSPoweruser sustain the editorial team Read more
Key notes
- A study by the University of Chicago found that GPT-4, an AI model, outperformed human analysts in predicting future earnings growth.
- GPT-4 achieved 60% accuracy in predicting earnings direction, exceeding human analysts’ 53-57% range.
- The AI’s success is attributed to “chain-of-thought” prompts that guide its analysis and vast knowledge base for pattern recognition.
A new study by the University of Chicago shows that large language models GPT-4 can outperform human analysts in predicting a company’s future earnings growth. This is impressive because the AI model was only given financial statements with no additional information.
Financial analyst relies on their expertise and experience to analyze a company’s financial health and predict future earnings. However, this study shows that AI models may be able to do this task just as well, or even better!
The model was given anonymized financial data, including balance sheets and income statements, and asked to predict future earnings growth. Even without any additional context, GPT-4 was able to achieve an accuracy rate of 60%, which is higher than the typical range of 53-57% for human analysts.
One of the factors behind GPT-4’s success is the use of “chain-of-thought” prompts. These prompts guide the AI model through the analytical process, which helps it identify trends, calculate ratios, and synthesize information to make a prediction.
The researchers believe that LLMs have a significant advantage over humans due to their vast knowledge base and ability to recognize patterns.
While some experts caution that the study may not have used the most advanced financial analysis models as a benchmark, the overall findings are promising. The fact that a general-purpose AI model can outperform specialized financial models and even human experts suggests that LLMs have the potential to change financial analysis.
If GPT-4 is this good, imagine how good GPT-5 will turn out to be, which apparently is “better at everything.”
More here.
