- One study found that ChatGPT reasons similarly to humans when deciphering Fed statements.
- GPT-4, released in mid-March, was more likely to agree with humans than GPT-3.
- A separate study showed the chatbot can also make strong stock predictions based on corporate news.
OpenAI's ChatGPT has shown an ability to decipher the meaning behind Federal Reserve statements and also exhibited predictive stock market capabilities in response to headlines, according to separate studies.
Researchers at the Richmond Fed said in a paper that ChatGPT can reason similarly to humans when decoding Fed speak.
Testing both ChatGPT-3 and GPT-4, the study took 500 sentences from Fed policy statements and asked the AI to sort them based on five categories: dovish, mostly dovish, neutral, mostly hawkish, and hawkish.
Even without prior training, the AI outperformed BERT, another language model used in finance, and was closely aligned to the way four human reviewers categorized the sentences. Moreover, it was able to provide a reasonable justification for its choices:
"The sentence indicates the committee's expectation to start implementing its balance sheet normalization program soon, which is a clear signal of tightening monetary policy, as long as the economy evolves as anticipated," GPT-4 wrote of a statement it deemed hawkish.
GPT-4, released in mid-March, was more likely to agree with humans than GPT-3.
In a second study, University of Florida researchers took financial headlines from after October 2021, which is when ChatGPT's knowledge base ends, and prompted the AI to make investing recommendations based on them.
That way ChatGPT could be tested on how news would move a stock without having prior information on what actually happened. Results showed a strong correlation between ChatGPT's responses and stock market moves.
For instance, the authors asked ChatGPT to decide what the headline "Rimini Street Fined $630,000 in Case Against Oracle" meant for Oracle.
It responded that investor confidence in Oracle would rise as the ruling could improve the company's ability to protect its intellectual property.
"These findings confirm the predictive power of ChatGPT sentiment scores and emphasize the potential benefits of incorporating LLMs into investment decision-making processes," the researchers concluded.