According to a study by the University of Chicago Booth School of Business,OpenAI of GPT-4existfinanceThe study, published last week, found that GPT-4, a large language model, performed well in analyzing financial statements and predicting future earnings. "Even without any narrative or industry-specific information, the language model outperformed financial analysts in predicting changes in earnings," the study said. "The language model showed relative strength in situations where financial analysts often struggle."
Source: The image is generated by AI, and the image is authorized by Midjourney
According to the study, the model was able to predict future earnings more accurately than human analysts by identifying financial statement trends and calculating various financial ratios through “thought chaining” prompts. “When we used thought chaining prompts to simulate human reasoning, we found that GPT achieved an accuracy of 60%, which is much higher than the accuracy of analysts.” The study said that the prediction accuracy of human analysts was closer to the low 50% range.
The study also found that trading strategies based on GPT-4's predictions are more profitable than the stock market, with higher share ratios and alphas. "We found that long-short strategies based on GPT predictions outperform the market and generate significant alphas and Sharpe ratios."
The study’s findings could have important implications for the financial services industry, which is increasingly adopting generative AI techniques. The development comes as OpenAI released a new version of GPT-4o, which is a machine learning algorithm that can generate and predict data.up to dateAfter version 1, GPT-4o is the successor of its ChatGPT bot.up to dateversion that can reason across audio, vision, and text.
Meanwhile, GPT-4 was released in March 2023. At the time of its release, OpenAI described GPT-4 as a large multi-model model (accepting image and text inputs and generating text outputs) that, while not as good as humans in many real-world scenarios, showed human-level performance on a variety of professional and academic benchmarks.
OpenAI's ChatGPT model was introduced to the public in November 2022. In January of the following year, ChatGPT reached 100 million registered users, achieving this milestone just 64 days after its launch.