ChatGPT Sentiment Analysis Significantly Outperforms Stock Market Average

A study published by the University of Florida found that using ChatGPT for sentiment analysis of publicly traded companies generated investment returns far in excess of the market average.

ChatGPT, an artificial intelligence chat application developed by OpenAI, is a large language model, which is a type of artificial intelligence algorithm, though it is not one explicitly trained for financial analysis.

The study used headline data for various stocks from October 2021 through December 2022 and tracked actual stock prices for that same time period. The study omitted headlines in which companies were mentioned passively in a story about something else, daily stock movement reports and duplicate headlines. In total, the study drew on 67,586 headlines related to 4,138 companies.

During the study’s time period, the researchers found that a strategy of investing $1 in the market and buying on good news and selling on bad news (as identified by ChatGPT itself) would have turned the $1 into more than $5.50 when not accounting for transaction costs, even though the market average declined over the same time period. This strategy was repeated on a daily basis.

When accounting for transaction costs, investing based on ChatGPT’s positive or negative evaluation outperformed the market average, provided transactions costs are 25 basis points or less per transaction.

The study also found that a strategy of only selling on bad news outperformed the inverse strategy of only buying on good news. Researcher Alejandro Lopez-Lira explained that bad news depresses stock prices more than good news inflates them, and ChatGPT was able to detect this pattern.

Additionally, companies with smaller market capitalization are more sensitive to headline sentiment than larger ones. Lopez-Lira says that, on average, a positive headline can increase the stock price of a small company’s stock by 60 bps, but the effect was only 20 bps for larger stocks.

The study did not set out to explain that gap, but Lopez-Lira suggests one explanation is that because less is known about smaller cap stocks, investors might be more sensitive to headlines than they would be for larger stocks. Lopez-Lira contends that investors can more ably balance the headline against their existing knowledge of the larger company and be less influenced by it.

Lopez-Lira explains that AI digests headline data and performs sentiment analysis much faster than a human can, so the primary advantage of using AI in trading would be the ability to act more quickly on public sentiment than a competitor relying on human expertise could.