Everyone knows fund managers can’t keep up with benchmarks, and the last 12 months have been particularly grim. So when it turns out an exchange-traded fund supposedly driven by artificial intelligence has managed the feat, people take notice.

The AI Powered Equity ETF, ticker AIEQ, is up 11.81 per cent from its debut a year ago through yesterday, just edging out the S&P 500. That’s better than 87 per cent of active managers over the stretch. How’d it do it? Pretty much the only way you could. By buying little-known companies that paid off big.

Running 24/7 on IBM Corp.’s Watson platform, the fund culls data on more than 6,000 U.S. public companies each day before picking about 100 of them to own. Of the top 15 holdings that contributed most to AIEQ’s gain, 10 of them are too small to be in the in the S&P 500 Index, a Bloomberg portfolio analysis shows.

Before delving under the hood, some disclaimers. First, even though the ETF is besting the S&P 500, its track record is far too short to generalize about its superiority. It could be skill, and it could be luck. Second, although it’s fine to look at the decisions made in getting to where it is, trying to discern a machine’s motive is a fool’s errand. That’s part of the point of quant.

“You can’t really conclude that these guys have figured something out until you see their performance through something other than the rather good market conditions that we’ve been enjoying,’’ said Tammer Kamel, CEO of Quandl Inc., an alternative data platform. “If the AI that these guys are using can navigate a correction and still outperform the market, then they’ve really found nirvana.’’

Whatever the case, the ETF has often looked like it knows what it’s doing. Take its interest in a company called Penn Virginia Corp., for example, a US$1.3 billion oil and gas driller based in Houston with fewer than 100 employees. In June, the small cap was the ETF’s largest holding, and over the fund’s lifespan it’s added the most to its return. Penn Virginia is up nearly 100 per cent this year.

Boyd Gaming Corp. is also a top contributor. The Las Vegas company was one of the ETF’s top five holdings when it was launched in October of last year. The stock gained nearly 50 percent before the bottom fell out. But the impact was muted because AIEQ had sold most of its stake.

So AEIQ is taking fliers on tiny companies and riding the wave in small caps to beat the S&P, right? But it outdoes small-cap indexes, too. Since inception, the ETF is beating the Russell 2000 with noticeable help from the “selection effect,’’ meaning picking particularly good securities, Bloomberg data show. AIEQ’s active return over the small-cap gauge is 6.03 per cent, all of which can be explained by individual stock selection.

To true believers, nothing random is happening.

“It’s not serendipity or luck. Rather, it is grunt work analysis of computational analysis of data and looking at your blogs and social media and press releases, and a conglomerate of all that,’’ Rick Roche, managing director at Little Harbor Advisors, a boutique investment firm focused on quantitative strategies, said in an interview at Bloomberg’s New York headquarters. “Machine learning’s power in terms of getting data and uncovering potential alpha is much better in the small cap and the mid cap space than it is large cap.”

It’s easy to be impressed by the way AIEQ’s seems to adapt over time, adding large caps to its portfolio and dialing down smaller companies right as they fell out of favor. In truth, though, other things may be going on. Sure, the machine might’ve sussed out an unwinding of the small-cap trade in the wake of rising rates and less trade bluster. Less amazingly, the ETF may have just needed to buy larger stocks as it got bigger and swelled toward $200 million.

Either way, it helped. Since the start of August, AIEQ’s greater allocation to larger companies has juiced its performance, a Bloomberg analysis shows.

“That’s the benefit of our strategy, right? It is very flexible and dynamic,’’ said Art Amador, COO and co-founder of Equbot, the company behind the ETF’s software. “The idea of, ‘Hey, it’s small caps today,’ doesn’t mean tomorrow we’re going to keep playing in the small-cap universe. In fact, it’s quite the opposite.’’

Fine, but the question of whether the artificial intelligence ETF knows what it’s doing or is simply getting lucky is a long way from solved -- maybe a decade away. To make any sort of judgment, you need so many picks to go right over so long a period that it becomes impossible to credit the gains to mathematical happenstance. That could be eight to 10 years of data spanning a full cycle, according to Sameer Samana, a global quantitative and technical strategist for Wells Fargo Investment Institute.

“In a world where there are so many managers, about the only place where there might be any skill is if you literally occupy the the third percentile and higher,” Samana said by phone, referring to a paper written by Eugene Fama and Kenneth French. “Your top three percent of managers is literally the starting point for anybody who might even have any skill. You can pretty much ignore anybody else.”