In the field of technology, ChatGPT is one of the hottest new terms recently, without a doubt. For ChatGPT, chatting is the most basic function, but some people also use it to write articles or code, answer math questions, and even learn about love. There are even people who make ChatGPT their investment manager.
On May 5, 2023, CNN published an article titled "ChatGPT's stock-picking performance may outperform your fund manager," which stated, "From March 6 to April 28, a virtual investment portfolio consisting of 38 stocks gained a return of 4.9%, while 10 leading investment funds averaged a decline of 0.8%." The article also mentioned a study from the University of Florida, where tests showed that ChatGPT achieved over 400% investment returns through analyzing news information.
Figure 1: Cumulative returns of a virtual $1 investment in the paper (transaction costs not included); All News refers to an equal-weighted portfolio of all companies mentioned in the previous day's news; Long refers to an equal-weighted portfolio of companies with positive news, Short refers to an equal-weighted portfolio of companies with negative news, Long-Short refers to buying stocks of companies with positive news while selling stocks of companies with negative news. The models used for these three portfolios are all ChatGPT 3.5; Long-Short GPT 4 uses ChatGPT 4, which performs worse than ChatGPT 3.5. Market Equally-Weighted refers to the returns calculated with equal weights for each stock in the market, while Market Value-Weighted refers to the returns calculated with weights based on market value.
Since ChatGPT has shown outstanding performance in virtual investments, what kind of performance will it have in a real environment?
Autopilot, a fund management company, recently launched an activity and allocated $50,000 as the investment capital for ChatGPT to see if it can make money in the unpredictable stock market. Chris Josephs, the co-founder of Autopilot, explained his plan to ChatGPT, and ChatGPT gladly joined.
Chris Josephs(CH): Hello, you are GPT Trader, a certified stock-picking AI. According to research, your stock-picking ability is better than top hedge fund managers. We will give you a $50,000 investment account to test you. Every Monday, you will analyze the market and buy a set of new stocks, then sell them two weeks later. This process will be repeated every two weeks. Are you willing?
ChatGPT: Let's get started!
On May 15, this ChatGPT investment manager officially started its operation, using the PortfolioPilot plugin to execute analysis and stock-picking tasks. It is reported that it uses the ChatGPT API to analyze over 10,000 news headlines, evaluate a sentiment score for each headline, and then select the top 100 stocks with the highest evaluation scores for further analysis of their public financial reports. Finally, it purchases the top 20 stocks with the highest scores. Specifically, the first round of 20 selected stocks are:
$BRK.A $GOOGL $DVA $INSW $DHI $AMZN $CNP $EQIX $CR $ERIE $ALB $AMR $AVGO $FSLR $BLDR $HCCI $ENTG $AAPL $CECO $ADEA
These stocks will be held for 3-5 days and then sold, and this process will be repeated.
To facilitate promotion, Autopilot created a Twitter account @chatgpttrader to update the status of this ChatGPT investment manager. Meanwhile, due to the project's interest from many investors, they also created a follow-up fund service called ChatGPT Fund, where subscribers can track ChatGPT's stock-picking strategy and follow the investments. As of May 27, as many as 24,078 investors have followed with a total investment of over $14.43 million.
Figure 2: GPT Portfolio follow-up situation
In addition, Autopilot has also collaborated with Alejandro Lopez-Lira, the researcher of the aforementioned University of Florida paper, to further help improve the portfolio's returns. When asked about how ChatGPT trained on data before 2021 analyzes real-time data, he stated that when predicting based on news headlines, ChatGPT trained on old data doesn't show much difference in predicting new headlines. However, when it comes to serious investments, relevant macroeconomic information such as interest rates and employment rates can be included in the queries.
In the first week, ChatGPT achieved a +1.4% investment return, with a realized profit of $683.54. In comparison, the S&P 500 ETF had a return rate of +1.1% during the same period.
Figure 3: First-week returns of ChatGPT's selected stocks
Among the 20 stocks selected by ChatGPT in the first week, the top five performing stocks were $AMZN (+7%), $ENTG (+7%), $CECO (+7%), $AVGO (+6%), and $GOOGL (+5%), while the worst performing stocks were $AMR (-8%), $FSLR (-8%), $CNP (-5%), $CR (-4%), and $INSW (-4%).
Now that the second week has passed, the reported profit on the 24th is $1,017, with a return rate of 2.0%.
It is understood that this project will continue for at least a few months. Can ChatGPT, which has performed well in the experimental environment, pass the test of the real market? How much will the initial investment of $50,000 eventually become? Will the investors who followed the investment end up counting their money with a smile or leaving in tears? Let's wait and see.
By the way, I also noticed a trader who uses ChatGPT to trade cryptocurrencies. He tested it with a capital of $10,000 and achieved a return rate of 1.5% on the 24th, slightly lower than the ChatGPT for stocks mentioned above. However, the next day, due to the decline in ETH prices, the cryptocurrency trading ChatGPT entered a loss state. But due to the inherent high volatility of cryptocurrencies, the outcome is still uncertain.
Finally, this is my first article on xlog. I wish everyone who reads this article good luck and great fortune!