Superhuman stockpicking: VanEck on how AI could change the game for investors

As the amount of data grew exponentially, that edge, Gil says, then moved to those best equipped to analyse the surfeit of information available, whether that was through technology or manpower. Now, with the advent of AI, the edge has moved again.

"This was something that was only reserved for those professional grade research firms," says Gil. "I think that's now within the reach of ordinary investors like you and I, and I think that's really exciting."

It means a potential leveling of the playing field where retail and professional investors alike have access to AI tools that go beyond the capabilities of human experts. 

He uses the example of a brilliant research analyst who's read every filing and met the management team of every company they cover. Even they have what Gil calls a "human ceiling", something that AI does not have to contend with, whether it's employed by a retail or professional investor. 

"What AI really enables you to do is it effectively allows you to test that exact same measure with the same level of rigour, but most importantly, without falling in love with a particular stock or maybe even a loser or not getting tired across thousands of different companies," he says.

"The way I like to look at it is you're entering this dark room and rather than shining a torch on something, you're turning the floodlights on."

Inside the GOAT

It goes without saying that many investors may be uncomfortable with handing over the fate of their stock portfolios to AI. Instead, it's VanEck that has chosen to take on the role of pioneer. 

The ETF was developed by VanEck in partnership with Akros Technologies, a South Korean AI and quantitative investing company whose AI engine powers 75 ETFs around the world. It tracks the Akros Enhanced World ex Australia Index, which picks 150 international stocks from an investible universe of around 1,200 using generative reinforcement learning. 

It gets there via a four-step process, says Gil.

"The first step is generating the investment ideas," he says. "For all of the approximately 1,200 or so companies within the investible universe, it's testing tens of thousands of signals across company and industry fundamentals."

This includes everything from return on equity on the fundamentals side, relative strength indices on the technical side, and GDP prints on the macro side. 

These signals are then scored and each company is given what is effectively a report card that tracks its performance and journey.

"In a similar way that a sporting athlete will watch over their tapes for the past couple of games and essentially learn how they performed, each of these companies are effectively measured against decades and decades of data."

The third and fourth steps are where the AI engine's generative learning kicks in, says Gil. It learns from its mistakes and uses that to inform its next decisions, and then validates that as part of an always-on dynamic process.

"Wherever there's new sources of data or new information surfacing in the economy, that is effectively going to get filtered in and that's going to reflect real time. The portfolio then rebalances on a monthly basis to reflect that adaptive nature."

VanEck and Akros have 21 years of simulated data across multiple market cycles to test its index performance, and while past performance is not indicative of future performance, the testing has shown promising results. 

Since the index base date in July 2005, the simulated track record for the Akros Enhanced World ex Australia Index has delivered 12.49% p.a., compared with 9.33% p.a. for the MSCI World ex Australia Index. But the potential returns are only half the story, says Gil.  

"The way I like to look at it is the headline return itself is not as important as how that return premium was achieved," he says.