Artificial intelligence can certainly make active managers better at their jobs. But will it be enough to stem the tide towards passive?
Given the dreadful run they have been on since the global financial crisis, you can understand why active managers are desperately looking for a glimmer light at the end of the tunnel.
Their hopes have been dashed several times in recent years. We were told, for instance, that a crash would help to revive their fortunes. So too would the end of quantitative easing and the return to more normal interest rates. All of those things have come to pass, and yet active looks more like a busted flush than ever before.
The latest hope that active managers are clinging to is artificial intelligence. Specifically, they’re excited by the potential for large language models (LLMs) to help them pick the right stocks to buy and sell, and possibly time the market too.
It’s a very plausible notion. LLMs like ChatGPT, Google Bard and Claude are, let’s face it, amazing. Their computational power is breathtaking, and it’s no wonder that asset management companies have been using this type of technology for a while. AI allows fund managers to analyse vast amounts of data within seconds. Another huge advantage of AI over flesh-and-blood fund managers is that it avoids the cognitive biases and emotions that human beings are prone to.
Recent studies are encouraging
Two academic studies in recent months have shown that funds driven entirely by AI can produce superior returns to those run by actual fund managers. The first was a study published in May by three academics at the University of Chicago, called Financial Statement Analysis with Large Language Models, which explored whether ChatGPT could forecast stock price movements more accurately than human analysts.
Researchers provided more than 15,000 anonymous financial statements from 1968 to 2021 and tasked ChatGPT-4 with predicting changes in earnings. ChatGPT’s accuracy reached 60%, outperforming human analysts’ 57%. With its chain-of-thought prompting, the model achieved an accuracy close to state-of-the-art neural networks, particularly excelling in predicting earnings for smaller companies. Interestingly, ChatGPT even outperformed the broader, cap-weighted stock market by 37 basis points monthly.
However, this particular model’s focus was on short-term earnings, a space where LLMs excel due to rule-based extrapolation. Longer-term forecasting remains challenging due to AI’s limited grasp of more complex factors, and its tendency to “hallucinate” by generating false content.
Humans and machines are a better combination
A study called From Man vs Machine to Man + Machine: The Art and AI of Stock Analyses, which was published in the October issue of the Journal of Financial Economics, also suggests that, in some areas, AI can have an edge over real-life equity analysts. An AI model trained on publicly available financial data outperformed 54.5% of human analysts in predicting 12-month stock returns. In particular, it was better than the humans at digesting macro-economic variables.
But researchers also found that the human analysts maintained an advantage in situations requiring nuanced understanding and contextual insights, particularly when dealing with intangible assets, financial distress, smaller and less liquid firms, and industries experiencing rapid changes.
In fact the study showed that a combination of man and machine results in superior predictions compared to either alone, reducing extreme errors and leveraging the complementary strengths of both.
What does this mean for indexers?
These are exciting findings for genuinely innovative, tech-savvy fund managers. If human analysts can learn to focus on areas where they retain a competitive advantage over AI, and rely on technology to do what technology does best, it could lead to more robust and reliable forecasting.
But what are the implications for index investors like myself? If AI really can help active managers do their jobs better, does it mean that active funds will start to turn the table on their passive peers? Does AI strengthen the case for using active managers?
Unfortunately for the fund industry, it’s not that simple. Why not? Well, if active managers can use AI models to limit human bias and make their forecasts more accurate, that will only make markets more efficient. That, in turn, will reduce the opportunity to generate alpha through security selection. In other words, AI-augmented funds will be victims of their own success. By getting better at what they do and raising the aggregate skill level, they will make it even harder to outperform not just other active funds, but index funds as well.
As David Booth, co-founder of Dimensional Fund Advisors, explained in an article in the FinancialTimes earlier this year, artificial intelligence is, ultimately, just that, artificial. LLMs, he wrote “struggle to account for unknown factors or real-world changes that come outside their training data.” Markets, on the other hand, are “composed of real, human intelligence and the millions of judgments market participants make.”
Markets are also, he went on, “fantastically complex… so much so that no one knows exactly how much a particular piece of information impacts a price, because there are so many other simultaneous inputs. But the market ensures that a price is the most accurate current representation of the value of a stock or bond. It’s free and available to all. How great is that?”
It all comes down to cost
In conclusion, if you really want to use actively managed funds, choosing funds that harness the power of AI effectively should increase your chances of outperforming other active investors. But you certainly shouldn’t expect to beat the index.
Ultimately, it all comes down to cost. Whichever type of active fund you use, there’s no escaping the arithmetic of active management. The net return is the gross return minus the cost of playing the game. In the words of Nobel laureate William Sharpe, “properly measured, the average actively-managed dollar must underperform the average passively-managed dollar, net of costs.”
So what are you going to rely on: artificial intelligence, on the one hand, or aggregate intelligence, which is far simpler and cheaper, on the other? The choice, for most of us, is surely a no-brainer.
© The Evidence-Based Investor MMXXIV. All rights reserved. Unauthorised use and/ or duplication of this material without express and written permission is strictly prohibited.