The Promise and Peril of Generative AI: Evidence from ChatGPT as Sell-Side Analysts

We examine the accuracy of generative AI models in predicting future earnings. Using the state-of-the-art GPT-4 model, we ask the model to predict future earnings, accompanied by justifications. GPT's forecasts exhibit larger forecast errors compared to analyst consensus. These forecasts are also on average overly optimistic compared to realized earnings. The accuracy of GPT's forecasts improves when firms have a better information environment, and when press releases contain more high-quality content. Overall, our analyses show great promise of generative AI in assisting human effort at information processing and content creation for financial market participants but also highlight the peril of blind reliance on these models when there is a lack of high-quality information.