From forecasting the stock market to analyzing the statements of the Federal Reserve; Amazing results of using ChatGPT in the financial field

Judging by the results of the first set of academic studies examining ChatGPT’s performance in finance, much of the hype over the past months about the AI ​​chatbot has been true. For example, ChatGPT can analyze the vague speeches of the governors of the US central bank in some way or to recognize the positive or negative impact of news on the stock prices of different companies.

To Report Bloomberg, this month, two new articles have been published that examine the performance of the ChatGPT artificial intelligence chatbot in financial analysis. One of the articles analyzes the monetary policy of the Federal Reserve based on the statements of its managers, and the second one examines the impact of news related to companies on the prediction of stock market movements.

The results of both studies represent a major step forward in using this technology to turn a range of texts, from news articles to tweets and speeches, into trading signals.

Of course, Quant analysts on Wall Street have long used the language models underlying these chatbots to improve many of their trading strategies. However, these new findings show that the artificial intelligence technology developed by the company OpenAI has reached a new level of framework analysis and detail.

Slavi Marinov, director of machine learning at Man AHL, a platform that has been using a technology called natural language processing for years to read text such as receipts and Reddit posts:

The ChatGPT chatbot was one of those rare cases where the hype is real.

The first article is titled “Can ChatGPT Decode Fedspeak – Vague Statements by Federal Reserve Chairs?” It has been published. The two researchers of this study, who are from the Federal Reserve Office in Richmond, found that ChatGPT’s analysis of whether the central bank’s statements are contractionary or expansionary is very close to human analysis. In this paper, Anne Lundgaard Hansen and Sophia Kazinnik show that ChatGPT beats Google’s BERT model as well as dictionary-based learning models.

According to the article, ChatGPT was even able to explain its classification of the Federal Reserve’s monetary policy statements in a way similar to central bank analysts. Also, in his interpretation, he considered these statements as a human measure.

Consider this sentence from the Federal Reserve’s statement in May 2013: “Labor market conditions have improved in recent months, but the unemployment rate remains high.” Commenting on the line, ChatGPT said that the remarks implied an expansionary monetary policy; Because it shows that the economic conditions have not yet fully recovered. This interpretation was similar to the conclusion of an analyst named Matthew Bryson, whom the article describes as a 24-year-old man known for his intelligence.

The second article is called “Can ChatGPT predict the trend of stock market price movement?” Performance predictability and large language models” has been published. Alejandro Lopez-Lira and Yuehua Tang, researchers at the University of Florida, asked ChatGPT to act as a financial expert and commentator on corporate news headlines. They used the news that was published after November 2021 (Aban 1400); A period whose data was not covered in the chatbot tutorial.

The study found that ChatGPT responses showed a statistical link to future stock market movements; A sign that this technology can correctly analyze the effects of news on the market.

For example, this chatbot explained whether the headline “Rimini Street Firm Fined $630,000 in Oracle Case” was good or bad for Oracle, explaining that it had a positive effect; Because the fine “could potentially boost investor confidence in Oracle’s ability to protect its intellectual property and grow demand for its products and services.”

For most high-level analysts, using natural language processing to gauge how popular a stock is on Twitter or to analyze the impact of the latest news about a company is now almost standard practice. But it looks like recent improvements to ChatGPT could provide a host of new information and make the technology more accessible to a wider range of financial professionals.

Marinov believes that while it’s no surprise that machines can interpret information almost as well as humans, ChatGPT could potentially speed up the entire process.

When IHL was first building its learning models, the hedge fund’s analysts manually determined the positive or negative impact of each sentence on assets to help the machines create a framework for interpreting the language. The London-based company then turned the entire process into a game for all of its employees to rate participants and calculate how much they agreed with each sentence.

These two new papers show that ChatGPT can do similar things without requiring special training. Research by the Federal Reserve showed that the technology of learning without seeing previous examples (Zero-Shot Learning) has made great progress over its previous versions. However, optimizing this chatbot based on a number of specific examples makes the output much better.

Marinov, who previously co-founded a startup based on natural language learning, said:

Previously, you had to label the data yourself. Now you can do it with the right prompt design by ChatGPT.

Bloomberg LP, the parent company of Bloomberg News, also released a large language model-based artificial intelligence chatbot for financial analysis last month.