Financial Expert: ChatGPT can predict stock prices

AI language models like OpenAI’s GPT-3, on which the AI ​​chatbot ChatGPT is based, could one day accurately predict stock prices. Astonishing prognostic results are already possible, as Alejandro Lopez-Lira, professor at the University of Florida Financial Institute, has discovered.

As part of an experiment, Lopez-Lira and his colleague Yuehua Tang taught ChatGPT to analyze financial news headlines. The AI ​​chatbot should recognize whether this news is good or bad for the future development of the prices of certain stocks.

According to the news situation, ChatGPT should then predict the development of share prices – and according to the scientists, was surprisingly often correct. The AI ​​chatbot did better than conventional stock market barometers. The predictions were also more accurate than random predictions.

However, the experiment also showed the current limits of the language models when it comes to concrete calculations. The Microsoft presentation in February 2023 showed that AI chatbots tend to be at war with mathematics.

ChatGPT did not issue any target corridors for the analyzed stock prices in the stock exchange experiment. Mathematical calculations were not required of him either. Lopez-Lira was only concerned with analyzing and predicting the future mood on the stock market based on headlines.

The financial expert was accordingly surprised by the results. On the one hand, according to Lopez-Lira, the experiment shows that stock market professionals do not yet use the capabilities of AI language models in their trading strategies.

In addition, the potential of AI chatbots is also evident when it comes to analyzes and forecasts in the financial sector. Why, the finance professor asks, should financial firms still pay human analysts when an AI can perform their tasks just as well?

On the other hand, ChatGPT’s ability to predict stock prices would move towards zero to the extent that companies bet on it. Within the next five years, according to Lopez-Lira to CNBChis kind of experiment would then no longer work.

Because the headlines would then be analyzed in real time with AI support and trading adjusted accordingly within seconds – then the news would have been priced into the price development of the coming day long ago.

For their experiment, Lopez-Lira and his colleague fed ChatGPT over 50,000 headlines about companies listed on Wall Street, Nasdaq, and a small-cap stock exchange. The AI ​​chatbot learned what news was good or bad for certain companies and predicted up or down movement for the next day accordingly.

In only one percent of the cases, according to Lopez-Lira, ChatGPT would have predicted the movement by accident – i.e. without prior information from the headlines. The bot was even able to correctly classify headlines that appeared negative at first glance.

In one case, the settlement of a legal dispute with a high fine did not cause bad mood on the stock market, as ChatGPT recognized. Rather, the price of the corresponding share moved upwards. The investors saw the settlement of the legal dispute as positive news and the penalty as the lesser evil.

Lopez-Lira has the results of his experiment in a study published. However, a peer review is still pending.