Will Artificial Intelligence Ever Beat the Stock Market?

Society is marching towards artificial general intelligence (AGI), whether we like it or not. The latest milestone towards AGI has been the development of ‘actually useful’ AI chatbots, ChatGPT (Microsoft) and Bard (Google). While in some ways these chatbots may seem trivial, this development is a major milestone that signifies the beginning of an arm’s race that will act as a strong propellant towards hyper-speed, development of new drugs to treat disease, colonization of other planets, and telling your home robot that they brought you the wrong brand of beer. While this new world is certainly scary and fraught with ethical concerns – and brings visions of a potential Bladerunner-style dystopia – it’s also very exciting.

But will AI in and of itself ever be consistently better than humans at picking stocks? The short answer is probably not. Humans are terrible stock pickers, and with the march towards AGI making computers think more like humans, we’re moving in the wrong direction entirely. Tongue in cheek quips aside, quants (including Viewpoint) have been trying to employ AI models as a “core” investment decision-making tool for decades with little success. Even the most successful and legendary quant investor in history, Renaissance Technologies (RenTech), says that they rely on advanced statistics rather than cutting-edge AI methods.

As this WSJ article outlines, “AI Can Write a Song, but It Can’t Beat the Market.” The data sets are simply too small in financial markets relative to other industries, some of which can rely on 175 billion parameters and more. Furthermore, and more importantly, market data is ‘noisier’ than language and other data, making it harder to use to explain or predict market moves. In short, financial markets are three parts chaos and one part sanity; this is not a good recipe on which AI can be trained. The relationship between a company’s future return on their stock and knowable current and historical information is weak (relative to other industries and applications of AI), and it also frequently changes for both obfuscated reasons (we don’t know why) and obvious reasons (some change in fundamentals or business strategy). Does that mean that AI is not a useful tool as part of the investment decision-making process? Absolutely not. The distinction is that AI can be used to spot patterns in relational, complex, or large amounts of data to help enhance a pre-existing edge. It’s this ‘second level’ AI application in portfolio management that has a clear and profitable use case, in sharp contrast to using an AI model to pick stocks or make investment decisions.

Investors will continue to hear about AI models that ‘beat the market,’ almost always referring to picking stocks to outperform an index, which may indeed be true for a given time period. However, just as with any human claiming to be able to ‘beat the market,’ it’s very likely to be mean-reverting over long periods of time.