Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
One of the trickiest challenges for any honest investor is trying to work out whether they are lucky or smart. Is their successful trading strategy the equivalent of a coin toss coming up heads five times in a row? Or is it the result of superior insight or execution? Human nature (and fee structures) being what they are, most investors prefer the latter explanation. In truth, it is often hard to tell.
In an attempt to dial up the smart factor and dial down luck, many investors have resorted to technology. Public market quantitative traders, in particular, have long used mathematical computation and machine-learning systems to spot significant correlations in market data, correct for human bias and execute trades at lightning speed.
This has taken extreme form at Baiont, a Chinese quant fund that hires “nerds and geniuses” with top computer science expertise and zero finance experience. Just as generative artificial intelligence models, such as ChatGPT, are trained to complete the next word in a sentence, they can also predict very short-term price movements, Baiont asserts. “We regard it as a pure AI task,” Feng Ji, Baiont’s founder, told the FT.
That may be a rational, if not necessarily successful, approach in highly liquid, data-rich public markets, where prices are precisely correct. But would that methodology work in private markets, particularly venture capital, where the data is sparse, markets are illiquid and prices are opaque? We are about to find out as a few, pioneering VC funds go all in on quant trading.
One such is QuantumLight, a firm that has just raised $250mn for its latest fund. The business, which tracks 10bn data points from 700,000 VC-backed companies, has already made 17 investments since 2023 driven by its algorithm. Typically, it co-invests $10mn at the series B stage, when a start-up has already acquired a digital footprint. Unlike most other VCs, it never leads a round or takes a board seat.
Traditional VCs still rely on human pattern recognition when deciding where to invest but machines can now perform that task more efficiently and dispassionately, QuantumLight’s chief executive Ilya Kondrashov tells me.
“What do you do in the case where your gut says no, but the machine says yes? We just decided to follow the machine because it’s our mission to prove this can be a good approach,” he says.
Some traditional quant investors are intrigued by how the methodology will play out in the VC field. The most critical determinant of success will be the quality, reliability and usability of the underlying data, says Ewan Kirk, founder of Cantab Capital Partners, a quant investment firm.
And he suggests that the AI technology the quant traders use may itself be disrupting the ways in which start-ups are nowadays built and scaled, confusing pattern recognition algorithms. Start-ups are currently using AI to grow faster than they have before, at lower cost. That can make it difficult to compare start-ups of different vintages.
“It’s all about generalising from historical data,” Kirk tells me. “The problem with VC is how relevant is data about Google’s series B compared with a series B investment you’re making right now?”
To address the data challenge, the quant VC Correlation Ventures has built what it claims is the most complete database of venture deals in the US, drawn from public sources and historical data from 15 VC partners.
It has been co-investing in hundreds of early-stage start-ups since 2011, writing cheques up to $4mn, with mixed results. “When we disagree personally with the model, it turns out, humbly, it’s better to go with the model,” says David Coats, Correlation’s co-founder.
Most mainstream VC firms are not yet ditching human experience and expertise. But the industry’s mythology, which deifies the omniscient investment sage on Silicon Valley’s Sand Hill Road, is being punctured. Almost every VC fund relies on a hybrid approach, using machine-learning tools to scout, select and analyse deals, says Patrick Stakenas, a senior analyst at Gartner.
Stakenas likens the VC quants’ approach to that of Billy Beane, the Oakland Athletics manager profiled in Michael Lewis’s book Moneyball, who used mathematical models to challenge the conventional methods of scouting baseball players to find undervalued talent. “At first, everyone thought they were crazy. Late on, everybody started doing it,” says Stakenas.
Cautious institutional investors, though, will want to see VC quant funds hitting some home runs before they buy into the concept.
john.thornhill@ft.com
