Recently, a Chicago University paper sparked heated discussions in the industry. The study found that GPT-4The accuracy rate in stock selection is as high as 60%, surpassing most humansStock Analystand professional models. This discovery has led people to question whether humansstockWill analysts face the risk of being laid off? However, some AI experts have questioned this study, pointing out that there may be problems with training data contamination, which affects the accuracy of the results.
The researchers analyzed financial statements and asked GPT-4 to predict future earnings changes. Surprisingly, GPT-4 successfully analyzed the numbers in the financial statements without any narrative context and achieved shockingly good performance. However, some AI experts have suggested that the training data may be contaminated, that is, GPT-4 may have included future stock prices in the training data, which may affect the accuracy of the results.
The researchers also analyzed the narrative insights model generated by GPT-4 and found that this may be one of the reasons for the success of GPT-4. By encoding and analyzing the narrative insights generated by the model, the researchers found that this information is informative for predicting future returns. This discovery gives people a new understanding of the excellent performance of GPT-4.
Although GPT-4 outperformed human stock analysts and professional models in stock selection, researchers pointed out that there may be problems with training data contamination, which affected the accuracy of the results. For the success of GPT-4, more in-depth research and verification may be needed to ensure its reliability and accuracy in the field of stock selection.