In recent months, there has been an introduction of various artificial intelligence systems with diverse capabilities and capacities. OpenAI’s ChatGPT 3.5 and GPT-4, Google Bard, Bing, AutoGPT, BabyAGI, and other open-source and proprietary AIs are some of the examples available in the market.
The AI systems mentioned have exhibited remarkable intelligence and capabilities that were once considered exclusive to human cognition. The usage of large language models in stock market analysis and forecast has been in discussions since the beginning of this AI trend.
A finance professor at the University of Florida, Alejandro Lopez-Lira, conducted a recent study revealing that large language models, such as ChatGPT, can be highly valuable in predicting stock prices. In fact, ChatGPT achieved a remarkable 500% return in one investing model, surpassing conventional sentiment analysis models that are commonly utilized by hedge funds.
Lopez-Lira utilized ChatGPT to analyze news headlines to determine whether they were favorable or unfavorable for a stock. The study discovered that ChatGPT’s forecasting ability to predict the direction of the next day’s returns was considerably better than random chance.
The experiment conducted by Alejandro Lopez-Lira highlights the potential of state-of-the-art artificial intelligence. The findings suggest that larger computational resources and improved datasets, such as those used to develop ChatGPT, can give rise to “emergent abilities” or capabilities that were not originally intended during the AI model’s development.
This implies that the potential applications of these AI models could be much broader and more diverse than previously imagined, with possible implications for industries beyond finance.
Lopez-Lira stated that ChatGPT’s ability to comprehend information intended for humans makes it highly likely that there will be return predictability, even if the market does not react in an ideal manner.
Method of Study and Findings
The researchers used a customized prompt to ask ChatGPT to assume the role of a financial expert with experience in stock recommendations and evaluate the impact of a given news headline on a company’s stock price in the short term.
ChatGPT was asked to answer “YES” if the news is good for the stock price, “NO” if it is bad, or “UNKNOWN” if it is uncertain. The researchers then transformed ChatGPT’s recommendation into a score and run linear regressions of the next day’s returns on the ChatGPT score to test its predictive power.
The prompt was specifically designed for financial analysis and asked ChatGPT to evaluate a given news headline and its potential impact on a company’s stock price in the short term.
The researchers assumed that the headline contains sufficient information for an expert in the financial industry to assess its impact on the stock price reasonably. ChatGPT was then asked to explain its answer concisely in one sentence. The authors set a temperature of 0 to maximize the reproducibility of the results.
The study shows that ChatGPT can provide recommendations for news headlines that are predictive of the next day’s stock returns. The ChatGPT score has a statistically significant relationship with the next day’s returns, indicating that the model can generate emergent abilities for financial analysis tasks.
The researchers compared ChatGPT’s results to the sentiment score provided by a news curating company and found out that ChatGPT’s score is more predictive of the next day’s returns.
The study’s conclusion asserts that ChatGPT exhibits superior performance compared to traditional sentiment analysis techniques provided by a prominent vendor.
The research findings highlight the potential of large language models in financial economics and contribute to the expanding collection of literature on the applications of artificial intelligence and natural language processing in this field.
In an interview with CNBC, Lopez-Lira stated that hedge funds had contacted him to gain further insights into his research. He also expressed his belief that ChatGPT’s capacity to anticipate stock price fluctuations might diminish in the future as more institutions begin to incorporate this technology.