Shulou(Shulou.com)11/24 Report--

According to news on the morning of April 18, Beijing time, it is reported that the first batch of academic research to apply ChatGPT to the financial industry is coming. Judging from the preliminary results, the high expectations for ChatGPT over the past few months are reasonable.

Two new papers published this month describe how to use ChatGPT for tasks related to financial market analysis. One is to ask ChatGPT to analyze whether the Fed's speech is hawkish or dovish, and the other is to ask ChatGPT to judge whether financial news has a positive or negative impact on a stock.

ChatGPT passed both tests. This suggests that technology may have taken a big step forward in turning textual content, including news, Twitter posts and speech notes, into trading signals.

This transformation process is nothing new on Wall Street, and quantitative analysts have long used the language models that support chatbots to guide strategies. But the latest research shows that the technology developed by OpenAI has reached a whole new level in resolving nuances and contexts.

"this is one of the few cases where hype has become a reality," said Slavi Slavi Marinov, head of machine learning at Man AHL, a quantitative hedge fund. For years, the company has been using natural language processing technology to process text content, including financial statements and Reddit posts.

The first paper is entitled "can ChatGPT decipher the Fed speech". In this paper, two Fed researchers found that ChatGPT is closest to humans in judging whether central bank statements are dovish or hawkish. Research by Anne Lundgard Hansen and Sophia Sophia Kazinnik of the Richmond Fed shows that ChatGPT beats Google's model BERT and dictionary-based taxonomy in this respect.

ChatGPT can even explain its classification of Fed policy speeches in a way similar to that of the Fed's own analysts. The interpretation of Fed analysts is used as a benchmark for the study.

Take a sentence from the Fed's speech in May 2013 as an example: "overall, labour market conditions have improved in recent months, but unemployment remains high." Artificial intelligence explains that this sentence is dovish because it shows that the economy has not fully recovered. This is similar to the conclusion of analyst Bryson, who is described in the paper as "a 24-year-old man, very smart and curious."

The second paper is entitled "can ChatGPT predict stock price movements?" Predictability of returns and large language models. Alessandro Lopez-Lila (Alejandro Lopez-Lira) and Yuehua Tang of the University of Florida asked ChatGPT to pretend to be financial experts and interpret financial news headlines. They used news from the end of 2021, which is not included in ChatGPT's training data.

The study found that there is a statistical correlation between the judgment given by ChatGPT and the subsequent trend of the corresponding stock. This shows that ChatGPT can correctly interpret the meaning of news headlines.

An example in the paper asked ChatGPT to judge whether the news "Rimini Street was fined $630000 in a lawsuit against Oracle" was good or bad for Oracle. ChatGPT believes it is beneficial because the penalty "may enhance investor confidence in Oracle's ability to protect intellectual property rights and increase industry demand for Oracle products and services".

For most senior quantitative analysts, it has become commonplace to use natural language processing technology to measure a stock's popularity on Twitter or to analyze the company's latest news. But the progress demonstrated by ChatGPT seems to open up a whole new world of information and make it easier for a wider range of financial professionals to use such technologies.

For Marinov, it is not surprising that artificial intelligence can read almost as well as humans, but ChatGPT may speed up the use of technology.

When Man AHL first built the model, the quantitative hedge fund needed to manually mark each sentence as positive or negative for an asset, providing a model for interpretive language for artificial intelligence. The company then turned the whole process into a game, ranking participants and calculating their approval of each sentence, so that all employees could participate.

These two new papers show that ChatGPT can even accomplish similar tasks without special training. Fed research shows that this so-called untrained learning has surpassed the previous technology, which will get better after fine-tuning it according to some specific examples.

Marinov was also a former co-founder of a natural language processing startup. "in the past, you had to label the data yourself," he said. "now you can do it by designing the right prompts for ChatGPT."

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.