Lionel Bonaventure | Afp | Getty Images
The proliferation of artificial intelligence programs such as ChatGPT and Alphabet‘s BardAI has already made big waves in financial markets, and a new study suggests that one day those programs may be able to trade in those markets all on their own.
The investment industry has long used algorithms and quantitative trading programs in an effort to boost profits, but the recent breakthroughs in AI opens up new possibilities. While a potential future where AI is running money on its own is likely many years away, a preliminary paper from two academics in South Korea shows that ChatGPT is already a better portfolio manager than just throwing darts.
The paper — called “Can ChatGPT Improve Investment Decision? From a portfolio management perspective” — found that ChatGPT’s ability to choose from a set of assets outperformed random selection on measures of risk-adjusted return and diversification.
“Retail investors, particularly those who may be uneducated or misinformed, can benefit from the democratization of portfolio management,” the paper’s authors, Hyungjin Ko and Jaewook Lee, wrote. “Additionally, professional portfolio managers can improve their productivity by focusing on more important tasks while being assisted by ChatGPT in the selection of diverse assets for a given portfolio.”
The researchers said the early results showed that ChatGPT could be used as a “co-pilot” for investors, but not a “prophet.” That isn’t too far removed from robo-advisors already in use for many retail brokerages. It also mirrors the plan of at least one financial firm, as Morgan Stanley is already testing an OpenAi-powered chatbot with its financial advisors to help with investment decisions.
The experiment, which has not yet been peer-reviewed, went like this: The researchers created a universe of 20 large-cap stocks from different sectors of the US market, and five assets each from the categories of cryptocurrencies, commodities, currencies and bonds.
Then, the researchers ran 10,000 simulations asking ChatGPT to select a various number of the available assets to create a portfolio.
A backtest of the data showed that ChatGPT’s selections created a more diversified portfolio with less correlated assets than random selection.
And then, over the period of Jan. 1, 2022 to Jan. 31, 2023, the ChatGPT portfolios also outperformed the random selection portfolios on a risk-adjusted basis.
For example, in a portfolio of four assets, the ChatGPT portfolio had an average expected return of -13.3% versus -23.3% for the random selection, and a better Sharpe ratio, which is a measure of risk adjusted return.
Stocks and bonds fell sharply in 2022, so the return results being negative is not surprising. The S&P 500 fell more than 14% over this time period, for comparison.
The S&P 500 fell during the time covered in this experiment.
This study does not mean that ChatGPT is ready for prime time and will soon be running a hedge fund. Citi analyst Chris Montagu said in a note to clients that the results were “convincing” but cited the small asset universe, short time frame and “easy-to-beat baseline model” as reasons to be cautious about wider adoption.
But as the language-learning AI models continue to improve and investors steadily migrate to lower-fee products, the AI portfolio manager could one day become a reality.
— CNBC’s Michael Bloom contributed to this report.