500% return on AI stocks? Experts Reveal Mysteries

The tornado of AI has finally reached the financial markets closest to money.

Written by: Kyle

A research report from the University of Florida shocked the financial circle: use ChatGPT to analyze the sentiment of company news, and according to this, you can get more than 500% return on investment by doing long and short in the stock market. While there is some skepticism about the report's astounding return figures, the financial world is being transformed by AI.

Legendary investment banks such as JPMorgan Chase and Goldman Sachs have continuously exposed the news that they are drilling for AI. Regardless of whether the 500% rate of return can withstand scrutiny, it at least shows that the ability of GPT has begun to penetrate into the most front-end link of the financial market-transactions. In the eyes of Rocky, an executive at Web3 Asset Management and Investment Research Institute, the efficient mining and optimization of "alternative factors" by AI has already occurred.

In quantitative institutions and hedge funds, "alternative factors" are the rarest and most precious factors among all strategic factors. It refers to factors other than conventional factors such as company fundamentals, trading volume, and price that affect the market, such as social public opinion and market sentiment. "Top institutions are all looking for alternative factors," Rocky explained. Volume price factors and fundamental factors will inevitably become homogeneous. Alternative factors will play a decisive role and help institutions win by surprise.

As a general-purpose large model, GPT needs to be refined by users if it wants to be directly used in quantitative investment, but a new door has been opened. For ordinary people, with the help of ChatGPT to efficiently verify a large number of strategies and analyze data, they can also find a way to make money that suits them.

The AI tornado has finally reached the financial market closest to money.

"500% ROI" is a sensation in the world

JPMorgan Chase, a top investment bank that has always believed in the power of technology, made a move on AI. On May 26, JPMorgan Chase announced that it is developing a financial service tool called "IndexGPT", which uses cloud computing and artificial intelligence to analyze and select securities, and provides customers with intelligent and personalized investment advice.

This is another sign of JP Morgan Chase to add AI to the trading system.

As early as 2017, JPMorgan Chase began to use the internal artificial intelligence tool code-named LOXM, allowing the machine to summarize experience and lessons from the past billions of real and simulated historical transactions, and then use the fastest and optimal Price executes trading orders, surpassing humans in terms of transaction scale and efficiency.

In 2019, JPMorgan Chase recruited global AI experts to develop a "stock trading robot". The main functions include generating investment reports, automatically searching for investment opportunities, and automatically monitoring "quote requests." At the time, JPMorgan said automated orders had reduced trade execution costs by about 20% over the past few years.

If JPMorgan Chase's early AI investment was intended to "reduce costs", when GPT showed superpowers, the investment bank began to use the most cutting-edge AI technology to enhance its "money capabilities". From the perspective of the layout, the role of AI in JPMorgan Chase has undergone important changes—from an investment assistant to a trader who guides transactions.

JPMorgan Chase's new actions have released a signal of AI's deep involvement in the financial industry, and Goldman Sachs and Morgan Stanley have also been exposed to invest in AI research and development internally.

The news of financial giants engaging in AI was staged on Wall Street, but it did not attract the attention of the public. However, a research report from the Department of Finance of the University of Florida was highlighted, which broke the aesthetic fatigue under the conventional narrative of "AI changes the financial circle".

The university research report titled "Can ChatGPT Predict Stock Price Trends?" was originally released on April 6 this year, and initially received little response. Until May, a tech writer on Reddit recommended the report, arguing that it was a paper ignored by the mainstream media.

After the "500% return on investment" was entered into the question, it instantly detonated inside and outside the financial circle.

According to the paper, researchers from the University of Florida fed GPT-3.5, which is not connected to the Internet, public market data and news from October 2021 to December 2022. These data were obtained through web crawlers, including 67,586 information about 4,138 listed companies. Headlines, and exclude any headlines of stock ups and downs, filtering out meaningless, hot topics, repetitive news, etc. The researchers primarily let ChatGPT evaluate each headline and asked it to decide whether it was positive or negative.

This is classic sentiment analysis and is part of automated trading strategies employed by well-known hedge funds such as DE Shaw, Two Sigma, and others. To give a simple example, when an event happens, the market often disagrees on whether it is good or bad. Accurate sentiment analysis helps to identify the impact of news and make correct investment decisions.

The researchers painstakingly asked ChatGPT to give an answer, and finally they came to a surprising conclusion: ChatGPT, which is good at logical reasoning, outperformed all other sentiment analysis tools. With the help of ChatGPT, the researchers back-tested the return performance of using ChatGPT to guide different investment strategies in the past. In the end, the long-short strategy (buy companies with good news, short-sell companies with bad news) has a return rate of more than 500%, and the short-sell strategy returns The return rate is close to 400%, and the return rate of the long strategy is about 50%.

Strategy performance powered by ChatGPT

In the securities market, any one of the above rates of return is enough to kill 99% of investment managers in the world. The research report noted that buying and holding the S&P 500 ETF returned -12% over the same time period.

Just using ChatGPT for sentiment analysis can bring such a high rate of return? While this report was eye-catching, it also caused netizens to question, "If you found a strategy that can earn a 500% return in less than 2 years, would you make it public?" Others said that even if the report is true, the Once a tactic is widely known, it is no longer effective. "There is no such thing as a free lunch."

AI upgrade alternative factor "excavator"

The layman watched the excitement, while the expert watched the doorway. When the news reached Rocky's ears, he was very excited.

Rocky is an executive of a Web3 asset management and investment research institution. He bluntly said that he was "stunned" by the research report of the University of Florida. He believes that the addition of ChatGPT has made a qualitative leap in the mining and optimization of "alternative factors". , he concluded: “Traders are dead, AI+ investing is the future.”

Rocky explained that before they studied quantification, two points were the most difficult. The first was the data source, and the second was the strategy factor. Common strategic factors include volume-price factors and fundamental factors. In the end, homogeneity is relatively serious. "The ultimate test is the game of alternative factors."

Strategy factors are a common concept in quantitative institutions. Simply understand, after the institution obtains the transaction data, information and public opinion data of the secondary market, it will clean them, and then process the massive data into factors. This is a process of finding important factors affecting the market from massive amounts of information. Integrating these factors into trading strategies can help traders judge the rise and fall of the market.

An effective strategy factor means a "gold mine", and once it is mined, it is not difficult to get a return.

As Rocky said, among the strategy factors, the volume-price factor, fundamental factor and alternative factor account for approximately 60%, 20%, and 20% of the quantitative strategy. Among them, the volume price factor is based on data mining of market trading volume, including asset prices per second, capital flow, technical indicators of various K-lines, etc.; fundamental factors are derived from financial statements, brokerage reports, analysts Expectations and so on; and alternative factors are the "secret weapons" other than the first two. Each institution will use its unique ability to collect factors that affect prices, including social public opinion and store data. The "sentiment analysis" that the University of Florida researchers let ChatGPT do falls into this category.

Common strategic factors (sorted out by Red Bank Research)

Generally speaking, it is difficult to widen the gap between institutions for volume price factors and fundamental factors, because the information is fixed and public, and mining alternative factors will test the skills of institutions. "Now the top hedge funds are investing in alternative factors," Rocky told "Metaverse Daily Explosion". In a duel between masters, conventional moves are difficult to work, and only unique moves can win.

However, the mining cost and difficulty of alternative factors are much higher than volume price factors and fundamental factors.

"It's like picking up shells on a boundless beach. You need to be very patient to pick them up one by one. Usually, a certain type of alternative data can only cover some of the plates, and even if it is excavated, you can only get the benefits on these plates." Li Xiang, general manager of Mengxi Investment, said that the data collection of alternative factors has a certain threshold, because it is not conventional data, either purchased from a third-party data provider, or collected by itself, and even in order to find better data, organizations need to Actively explore valuable data suppliers.

After the data is collected, it is not easy to study alternative data. "How to mine the internal logic of the data, this step also has a high threshold." Li Xiang said that this process is very delicate. It needs to eliminate all kinds of noise, find the internal logic, and then combine factors. After a series of operations are completed, there may be good results. Effect.

Li Xiang likens the process of collecting factors to "mining": at the beginning, some mines at the surface level, which are easier to collect, are collected first, and then dug deeper and deeper.

In terms of mining alternative factors, it is often the most labor-intensive, financially resource-intensive and core work of major investment institutions. They collect information in large quantities, analyze values one by one, backtest the rate of return, trial and error, and may get nothing after a huge and complicated workload. Harvesting effective alternative factors sometimes requires an element of luck.

Now, the emergence of ChatGPT makes the process of mining alternative factors efficient. "Its text-to-text function is very powerful. For example, we can use natural language processing technology to capture netizens' views on a certain type of stock, or even a certain stock." Li Xiang believes that the leapfrog development of GPT can improve some The efficiency of auxiliary work, such as in terms of predictive dimensions, "Its gain for quantitative research is at the data collection end, and ChatGPT can be used to better obtain text-side information."

However, GPT is more like a general-purpose large model, and it is not biased towards financial majors, which is destined to not be used out of the box. Rocky said that the data feeding based on the GPT large model is a "universal model", which cannot satisfy the authenticity, validity, and real-time performance of financial cross-sectional and time series data. In the data cleaning process, a professional small model is also needed Do preprocessing and standardization, which shows that ChatGPT is still far from the road of professional quantification.

But Rocky believes that ChatGPT has opened up an obvious door for institutions, and AI can become a high-powered assistant for traders.

Wealth opportunity for ordinary people is here?

The research report of the University of Florida is like an introduction, enough to give JPMorgan Chase a sudden inspiration. AI is likely to become an emotionless "money-making machine" in the trading market, playing money games with real people.

So, can ordinary investors use tools like ChatGPT to participate in quantitative transactions and improve their returns?

In this regard, Rocky feels that it is not realistic. He explained that quantitative trading requires a professional background in financial engineering, advanced mathematics, statistical concepts, financial knowledge, derivatives knowledge, financial regulations and other knowledge reserves. At the same time, GPT, a large model database, does not have real-time performance. You must purchase data sources from Bloomberg and other places. Otherwise, the data is not real-time and you cannot participate in the game. It’s okay to run a profit backtest on GPT, but don’t even think about it in actual combat.”

The financial market is turbulent, and ordinary investors should be particularly cautious in using tools. Once they are seen through by high-end tools, they may become lambs at the mercy of others. However, some people have provided investment ideas that are more suitable for ordinary people. I may not be able to achieve high returns, but there is still a good chance to outperform the mortgage interest rate.

Niu Yifei, the creator of the small program "Aniu Data", has been engaged in low-frequency quantitative trading. Not long ago, he conducted an experiment and asked ChatGPT to write a quantitative strategy and backtest the yield curve.

The strategic logic Niu Yifei provided to ChatGPT is: From the ETFs of the three indexes of SSE 50, ChiNext Index, and 10-year Treasury Bond, select the ETF with the largest increase in the past month (22 trading days) every day. If you have the fund, you will continue to hold the position. If you don’t hold it, you will clear the funds you hold and buy the fund. If the three funds have fallen in the past month, you will clear the position.

Using ChatGPT to write quantitative strategy code process

Soon, ChatGPT gave the corresponding policy code and comments. "The only shortcoming is that the data source is not given. Fortunately, I have a copy of the fund's historical data. After importing the data and running it, I can really see the results of the daily holdings."

Later, Niu Yifei needed to verify the historical performance of the strategy, so he asked ChatGPT to generate a backtest program, and asked the backtest to find out the interval rate of return, annualized rate of return, maximum retracement, etc. of the strategy. After a few seconds, AI gave the program code , and achieved the required indicators. "However, after carefully reviewing the program, I still found some flaws in the details, such as not considering the actual rebalancing time, etc., but the overall completion rate has exceeded 90%."

Niu Yifei said that it optimized the program by guiding the AI, and made simple magic changes manually, and the backtest program was ready. He used this program to backtest the investment performance of the above-mentioned strategy in 2022, and finally got a range rate of return of 9.18%, an annual rate of return of 9.57%, and a maximum drawdown of -12.25%. Compared with professional statistical tools, the net worth curve of the backtest program produced by ChatGPT is almost exactly the same.

In this case, Niu Yifei took the initiative to determine the investment strategy and asked ChatGPT to make automated investment software and backtesting procedures. In fact, he handed over the work of writing code to ChatGPT. Of course, ChatGPT’s code writing efficiency far exceeds that of human engineers, which allows ordinary investors to use this method to efficiently verify the effectiveness of a large number of strategies, and then continuously optimize trading strategies to increase the rate of return.

In Niu Yifei's simple practice, his annualized return on investment has reached 9.57%, which is much higher than that of general bank wealth management products. Niu Yifei revealed to "Metaverse Daily Explosion" that in addition to letting ChatGPT write code, he is also trying to use AI for data analysis, financial report and announcement analysis, etc., so that it can give trading signals.

Li Xiang also believes that ChatGPT will lower the threshold of the quantitative industry to a certain extent. If someone is interested in the quantitative industry, even if they have no experience in the industry and only have some ideas on trading, they can also participate in part with the help of ChatGPT's capabilities. "However, there is still a long way to go from this state to growing into a very professional core researcher of a quantitative institution, with refined strategies and perfect details."