AI Can’t Beat The Market

AI Can’t Beat The Market

Industries across the board are experimenting with AI with varying degrees of success, but one area that it hasn’t conquered yet is beating the market, reports an article in The Wall Street Journal. While ChatGPT may show promising results in bolstering sales, applying AI to investing has had limited progress.

Wall Street had a jump on implementing AI about 40 years ago, when Jim Simons of Renaissance Technologies created algorithms that gave their computers the power to make investing decisions. He, along with other quants, developed models that use past data to identify patterns and make profitable trades with minimal intervention from humans, though they relied more on complex statistics than AI. And not many firms have opted to give all trading control over to their computers. Those that have given over a portion of their operations to machines haven’t seen any kind of significant advancement with self- or reinforcement-learning. And investors generally use datasets that are narrower in scope than those used in AI like ChatGPT, which has 175 billion parameters from centuries of text and data. Also, market data such as earnings, share momentum, and sentiment are only part of the picture, so while AI might identify correlations in the reams of data presented to it, that doesn’t necessarily translate into predicting how the market will move.

In addition, ChatGPT frequently makes obvious mistakes that a skilled investor would not, the article points out. Human behavior is still a vital part of investing, Richard Dewey of fin-tech company Proven tells The Journal. “When it comes to investing, it’s still hard to turn everything over to the machines,” he says. However, investors do seem to be getting more comfortable depending on AI. The hedge fund Voleon is built on the premise of machine-learning, and quant hedge fund Numerai saw gains of 20% in 2022 by using machine-learning methods. Meanwhile, the fund EquiLibre Technologies was founded by three former senior staffers at DeepMind Technologies, the AI subsidiary of Alphabet.

But while AI has the potential to someday make trading more democratic and accessible, there are currently there just aren’t enough firms to bring about the revolution, and those that are implementing AI are still posting very inconsistent results. Still, “once you get them working, these strategies will make more accurate predictions,” says Jon McAuliffe, co-founder of Voleon. And one of EquiLibre’s cofounders, Martin Schmid, told The Journal that he believes “reinforcement learning,” where computers learn through a punishment-and-reward system, could work for stocks and bonds, though his firm has not yet started investing. In the short term, however, AI technology could cause massive change in departments like research and sales. AI technology can be used to create automated and customized data for clients, says  Jens Nordvig of MaretReader, a firm that uses AI to weed through financial news, and “that’s a lot of what salespeople [at investment banks] do.”