Models vs. managers: who manages money better

Several funds that manage clients' money using algorithms suffered losses this summer. The machines were unprepared for the unpredictability of the US president. But do they always lose to humans?

In June, quant investing pioneer Renaissance Institutional Equities, a fund founded by the late billionaire Jim Simons (now run by his former assistant Peter Brown), lost almost 6%, according to an HSBC report. The summer of 2025 showed how wrong the machines can be. In July, quant funds were synchronously hit by a "junk rally": cheap, shorts-laden stocks rose sharply - just the kind of stocks that models are usually cold to. Goldman Sachs called the beginning and middle of last summer the worst period for this sector in the last five years.

Renaissance Institutional Equities Fund and Renaissance Institutional Diversified Alpha were up 22.7% and 15.6%, respectively, Business Insider reported, citing a source close to the manager. Medallion, also part of Renaissance Technologies, returned 30%, Simons' biographer and Wall Street Journal reporter Gregory Zuckerman wrote on Linkedin.

Generally speaking, in the first half of 2025, many funds where people make decisions have fared noticeably better than most algorithmic strategies. Obviously, the main "culprit" for this is Donald Trump, not the most predictable White House host, whose many decisions are difficult to rationally calculate, and therefore more likely to create volatility in the market.

In their paper"Do Machines Beat Humans? Evidence from Mutual Fund Performance Persistence," economists António Freitas Miguel and Yihao Cheng of Lisbon's Higher Institute of Business and Labor Sciences argue that funds with programmatic decision-makingdo not have a sustainable advantage over funds withhumans at the helm.Evidence from Mutual Fund Performance Persistence" argue that funds with programmatic decision-making do not have a sustainable advantage over funds with humans at the helm.

Algorithms are great at repetition and discipline, but when many models are trained on similar data, and often by the same experts who "roam" from fund to fund, the source of profit quickly dries up: the "secret" ceases to be a secret, similar strategies begin to interfere with each other. In such an environment, the winner is not the one whose model is better trained on historical data, but the one who adapts to new conditions faster. And this is where people on average have better "contextual sensitivity" - the ability to ask themselves in time: "have the rules of the game changed?

Researchers compiled a massive dataset of U.S. equity funds over nearly two decades, from 2000 through 2019. The researchers left 3,915 unique funds, of which 217 are truly"quant" - that is, "quantitative" - "quantitative." In them, the portfolio is built strictly according to the rules of the mathematical model, without the final manual dokrutka. The hypothesis tested by the authors of the study was simple: if algorithms are really smarter than people, then their victories should be more frequent and failures less frequent.

To make the comparison fair, the researchers looked at returns after all fees and also compared "apples to apples": that is, companies in the same row.

Another filter was applied: eliminating the effect of new capital inflows and outflows. If a fund was performing outstandingly, it faced an influx of new clients, causing the strategy to "bloat" and lose its shape. The authors tested this separately and found no significant differences in inflows and outflows between "quant" and "human" funds. So it's not the clients, but the management processes themselves. The main outcome of the study is simple and unpleasant for fans of robo-Buffets. The "quants" have performance persistence closer to the bottom line: funds that lagged often continue to lag. And those that broke into the top often bounce back. Regular active funds behave in the same way, but the "quants" have stronger pullbacks, both on short horizons (quarter) and long horizons (year). This is not to say that people are smarter than machines. The point is that quant funds do not have a noticeable and stable advantage over funds where decisions are made by people.

History also knows purely technical failures: the classic case of AXA Rosenberg, where an error in code and lack of transparency to clients cost hundreds of millions and sanctions from the SEC.

At the same time, we should note the opposite: approaches are evolving. In 2025, AQR Capital Management, one of the world's largest quantitative funds, founded in 1998 by Cliff Asness and colleagues at Goldman Sachs, openly surrendered to the Ma". The company has expanded its use of algorithms across asset classes, and their Apex and Delphi strategies posted double-digit returns for the first half of the year: +11.4% and +11.6%, respectively. But Clif Asness himself admits that it is difficult to explain to clients why a model trades this way and not the other way. Especially in periods of drawdown.

Here's how the findings of Miguel and Cheng's study are interpreted by Larry Swedrow, one of the vocal popularizers of "evidence-based investing."

The lack of meaningful difference between humans and machines in generating market returns once again brings us back to the basic principles of investor safety.

There is no point in overpaying for AI-inside signage, at least not yet. It is much more important to look at what matters with any manager: fund commissions, methods of tax optimization, frequency of reviewing your financial goals and adapting the composition of your portfolio to them.

It is also still not a good idea to succumb to the allure of high performance. Whether a fund is managed by a human or a computer, great performance in the past is still no guarantee of being able to repeat it in the future. Jumping from one "star" fund to another is a sure way to fall behind a simple index.

It is still important to demand from the manager a clear explanation of what exactly his strategy does: what securities it likes, what it avoids, what will happen when market trends change. If there is no such explanation, treat the product as risky, even if it has a beautiful presentation.

Use algorithms where they are strong, and where they are already used by almost all major market players: in discipline and technique, if we are talking about trading. If we are talking about working with information, it is reading large amounts of text, categorizing trends, searching for data and identifying risks. This is how algorithms become a companion of the manager, not a substitute for him. This is the common sense focus of the industry: to optimize the mechanics, not to search for the "eternal grail" of selecting the right stocks.

This article was AI-translated and verified by a human editor