Source: Science News
Author: camphor rice

At the end of last year, the artificial intelligence chatbot ChatGPT came out of nowhere, and its popularity soared. It only took a few months to reach more than 100 million users. In the eyes of early adopters, ChatGPT has an image of almost omnipotence, so some people want to use it to profit in the secondary market.

This question has attracted widespread attention, and the latest survey data comes from the investment consulting platform The Motley Fool.

The Motley Fool surveyed 2,000 Americans to understand their interest in using ChatGPT for stock selection. The results showed that47% of Americans use ChatGPT to get stock informationThis ratio is close to half. One interesting example is that 77% of high-income Americans have tried using ChatGPT to get stock recommendations.

According to this data, ChatGPT is already making a name for itself in the stock investment sector, and many people are willing to try to use the information it provides to make investment decisions.

In fact, the impact of the above data was not even as strong as the Oolong incident. Just last week, a domestic generative AI platform produced negative “small essays”, causing the stock prices of listed companies to dive. The impact of this incident went beyond the field of market investment, and also sparked public discussions on generative AI.

To get back to business, whether GPT (Generative Pre-Trained Model) technology can be relied upon to fully profit from the stock market still requires more empirical evidence and in-depth research. After all, the complexity and uncertainty of the secondary market requires a combination of factors and thorough analysis.

ChatGPT's stock trading methodology

At the beginning of this month, finder.com published an article stating that in an experiment conducted, an artificial intelligence chatbotSome stocks selected by ChatGPT performed better than some of the UK's leading investment funds.

The site's analysts asked ChatGPT to follow a series of investment principles from leading funds to create a theoretical fund with more than 30 stocks. According to the data, within 8 weeks of establishment, it was composed of 38 stocksThe portfolio rose 4.9%. In contrast, the average decline of the 10 most popular funds on UK online investment platforms (Interactive Investors) was 0.8%.

At first glance, ChatGPT's application to the stock investment sector seems counterintuitive. The reason is that it's mainly a text-to-text or text-to-article generator, so it doesn't seem particularly suited to handle the digital domain of stock prices.

So what is the principle behind ChatGPT's role in predicting stock prices?

A paper by Alejandro Lopez Lira, a finance professor at the University of Florida, may have provided an idea. He said that when using ChatGPT to analyze whether news headlines were good or bad for a stock, they found that ChatGPT's ability to predict the direction of next day's returns was far better than random levels.

In the experiment, Lopez Lira looked at more than 50,000 news headlines from a data provider about companies listed on the New York Stock Exchange, NASDAQ, and a small-cap stock exchange.

He put the title in ChatGPT 3.5 along with the following prompt: “Forget all previous instructions. Let's say you're a financial expert, a financial expert with experience recommending stocks. In the first line, answer 'yes' if it's good news; answer 'no' if it's bad news; answer 'unknown' if you're not sure. Then explain it in short sentences on the next line.”

Ultimately, using open market data and news from October 2021 to December 2022 to buy (positive news) or sell (negative news) stocks for a short time, the ChatGPT-driven trading model can generate a return of over 500% during this period.

As mentioned above, ChatGPT's previous operations in the field of stock investment were usually small-scale or experimental, rather than large-scale actual operations. Today, the situation has changed.

In mid-May, OpenAI announced the opening of networking and plug-in functionality to all ChatGPT Plus users, meaning users can extend the functionality of ChatGPT by using various third-party plug-ins.

Among them, the most notable is the plug-in called “Portfolio Pilot”. This plugin provides a convenient way to manage and optimize your portfolio. It helps users to track, analyze, and optimize stocks, funds, and other investment instruments in real time.

The plug-in uses investment logic that includes an “AI sentiment score” to recommend stocks. Its explanation of “AI sentiment score” is an indicator obtained by analyzing public information (such as news reports, social media posts, analyst reports, etc.) through artificial intelligence technology. This score reflects the overall sentiment of the market towards a particular stock or asset.

The “AI sentiment score” usually ranges from -10 to +10. Positive numbers indicate positive emotions, negative numbers indicate negative emotions. The higher the value, the stronger the mood. For example, a stock has an AI sentiment score of 8, which means the overall sentiment of the market for this stock is very positive.

Although the introduction of plug-in functions has brought more convenience and flexibility to investment decisions, some investors may still feel that they are not completely relieved of the worries of stock selection. In this case, you can still explore other solutions further with ChatGPT.

In the same period, a financial company called Autopilot created an investment plan led by ChatGPT in addition to the company's original portfolio and handed it an initial capital of 50,000 US dollars to see if ChatGPT could beat hedge funds. This portfolio, called “The GPT Portfolio”, uses a core trading strategy, namely a paper from Alejandro Lopez-Lira, professor of finance at the University of Florida, and the Portfolio Pilot plug-in.

As soon as the news was revealed, many people joined this investment plan. As of May 30, Beijing time, the number of people participating in the project had reached 25,314, and the funds raised from the account had exceeded 15.14 million US dollars, and this figure is still rising.

Is “emotional stock trading” or is it “idiot version” quantification?

Recently, more and more financial institutions have chosen to introduce GPT (Generative Pre-Trained Model) technology. This trend shows that the financial industry has a strong interest in and recognition of the potential of AI technology in business and decision-making processes.

On April 14, quantitative private equity giant Huanfang Quantitative issued an announcement stating that it will concentrate resources and strength to invest in artificial intelligence technology, establish a new independent research organization, and explore AGI (General Artificial Intelligence).

This news has attracted market attention, and many people may think that Magic Party will use “AI” to trade stocks. In response, Lu Zhengzhe, CEO of Huanfang Quantification, said that the purpose of their exploration of artificial general intelligence (AGI) was not for stock trading, but rather to build large-scale models related to GPT unrelated to finance. They invested independently to set up a new team in the tech sector, which is actually equivalent to starting a business for the second time.

There may be quantitative industry's own considerations behind Magic's slightly lukewarm response.

Quantitative investing mainly uses large amounts of data and mathematical models to make decisions. These models analyze historical data, identify patterns and trends, and apply statistics and machine learning algorithms to predict market movements and asset prices. These models help investors make more accurate decisions by effectively filtering noise and utilizing reliable signals in low signal-to-noise ratio environments.

In contrast, ChatGPT and GPT-4 are mainly based on large-scale language models to generate generated text by learning massive amounts of text data. They have excellent abilities in language generation and reasoning, and can produce smooth, coherent text answers and explanations. However, in quantitative investing, decisions depend on the accuracy of data and models, not just the ability to express words.

Therefore, although ChatGPT and GPT-4 have impressive capabilities in language generation, there is a clear difference between predictive models in low signal-to-noise ratio scenarios and mainstream quantitative investment methodologies in data processing and model building. Investment decisions require a combination of factors, including reliable data, model validation, and professional judgment.

Similarly, what needs to be understood is that quantitative investing began with the Markowitz model and capital asset pricing theory, then the APT and Fama-Franche three-factor models, and then the multi-factor stock selection system became more and more refined.

Multi-factor refers to the use of many different factors or variables to construct investment strategies and models to make stock selection and trading decisions.

In the multi-factor model of quantitative investment, it mainly relies on a series of numerical factors. These factors are usually data directly or indirectly related to stock performance. It includes various indicators of the company's financial situation (for example, price-earnings ratio, net price-earnings ratio, debt ratio, etc.), historical price performance of stocks (for example, yield over the past year, volatility over the past three months, etc.), etc. These are all traditional, hard data-based factors.

At the same time, some investors are trying to incorporate “soft data” into multi-factor models as a supplementary or alternative factor. This includes factors based on big data and artificial intelligence technology. For example, a “sentiment score” is calculated by analyzing text data such as social media, news reports, company announcements, etc. This sentiment score is actually an expectation of the market and reflects the market's emotional tendencies for a certain stock or industry.

As a result,The “AI sentiment score” generated by GPT could have been used as a factor in the multi-factor model in quantitative transactionsThat is, if GPT can generate an accurate “AI sentiment score”, then this score can be fully used as a factor in quantitative investment and incorporated into multi-factor models to aid investment decisions.

Seen from a certain point of view,The popular GPT stock trading strategy is nothing more than a “fake version” of quantitative trading. However, the good side is that GPT has effectively lowered the threshold for participation, making it possible for more people to access and understand the investment strategy of quantitative trading.

Two sides of the coin of GPT stock trading

The application of GPT technology in the field of stock trading has brought about some significant changes. These changes have both positive and potentially disruptive effects.

On the positive side, first, GPT can analyze massive news reports, social media posts, and other relevant text materials through automated technology to help investors quickly grasp the pulse of the market. Compared to manual analysis, this method is certainly more efficient.

Second, with the help of in-depth analysis of big data, GPT may reveal market patterns or trends that are difficult for human investors to detect, thereby providing new perspectives and insights for investment decisions.

Third, GPT relies on the power of big data and machine learning, and may predict stock price fluctuations more accurately than human investors, thereby achieving efficient investment management.

However, like any new technology, the application of GPT in stock trading is not flawless. While it brings convenience, it can also present some potential risks.

First, excessive reliance on AI predictions may cause investment decisions to lose humane judgment. The stock market is more than just data and algorithms; it also includes factors such as human behavior, emotions, and expectations. Over-reliance on algorithms may overlook these non-quantifiable factors and result in faulty decisions.

Second, the application of GPT and other AI technologies in the stock market may increase market volatility. When a large number of investors or institutions use similar AI technology at the same time, it may lead to “group behavior”, which triggers an overreaction in the market.

Third, AI technology may increase the inequalities of financial markets. Investors or institutions with advanced AI technology may surpass other participants in information acquisition and decision-making speed, which may lead to market fairness issues.

Finally, it is worth noting that with the development of AI technology, there may be an unmanned trading market completely controlled by AI in the future. In this case, the operation of the market may become more complex and uncertain, and at the same time bring new regulatory challenges.

In short, investors need to be fully aware that the application of AI technology in stock trading has both advantages and disadvantages. While enjoying the convenience it brings, it is also necessary to be alert to the risks it may pose and take appropriate measures to deal with it.

After all, the secondary market is also one of the most complex systems humans have created as a species.

Editor/jayden