Quantopian: Searching for the Vitalik Buterin of quant finance

Hedge funds and high finance are no longer the preserve of a tiny, entitled elite. Quantopian crowdsources raw coding talent onto its open platform where the best trading algorithms are awarded assets, licensed and paid in cash prizes each month.

Thus Quantopian is casting its net wide, searching for new talent in places outside the Ivy League or Wall Street, to attract computer scientists from all over the world. The platform can unearth coders that can do for AI and financial data science what Vitalik Buterin, the 22-year-old who dropped out of a computer science degree in Canada, has done with open source cryptocurrency code.

John Fawcett, CEO and founder of Quantopian, said: "About ten years ago, people realised you could start a web company with your laptop. All you needed was an idea; you had to be capable and you needed just a minimal amount of technology. We are doing the same thing for quants.

"We lowered the activation energy required, so instead of spending years building a rig to find out if your idea works, you can check it in a couple of hours – so that's really transformative.

"I think the analogy with cryptocurrency is that there's this incredible demand in the world for a certain kind of talent, and in our world if you are able to look at data and make predictions with it, there's almost infinite demand for that capability."

The asset management establishment has typically selected their applicants from the same top-tier universities. This tiny pool of applicants are then subjected to gruelling rounds of a very human selection process that includes intensive interviews and tests. This interview process introduces all sorts of biases. Fawcett thinks this is a curious argument to come from a quantitative perspective.

"So you're going to use credentials; you're going to use your own bias. How many people are you going to interview per year – like 20? That's a very small number of people, so you have a tiny little dataset. Basically you're torturing this little sample of data to try to get some very important result. I think really what the major firms rely on is the ability of academic institutions to attract lots of talent and filter it – and that does work, but it's not exhaustive, and it's not scalable."

This traditional model means the largest firms will identify 40 or so candidates at the top schools they want to talk to, and then everybody else identifies the same 40 candidates and then a bidding war ensues, said Fawcett. "There is just so many more people in the world that have this talent that are not in New York or London or Hong Kong, and engaging those people is really important."

Fawcett also points to the fact that prolonging the secretive nature of the industry means there hasn't been an opportunity to unbundle the problem. Firms are doing everything and that's not most efficient way to go; other industries don't operate that way. There should be more of a supply chain for the problem of trading, he says.

While the traditional approach is bounded by the size of the talent pool each firm is fighting over, Quantopian's crowdsoucing model faces its own unique scaling problems – another analogy with the nascent world of open currency chains.

"The core of the problem that's really unique for us as the selector of the individual algorithms is just the scale. Our database has 2.6 million algorithms.

"There's noise and duplication as you'd expect, but after we've winnowed down from the obvious things that don't work at all, duplicates or things that were part of the training process that we are going through, there's still 700,000 original algos to weed through to find the few that are alpha band."

The Quantopian platform is designed to run multiple investment vehicles. Today there is a single vehicle which is pure alpha strategy, and uses a small number of algorithms in the vehicle. Over time the plan is to add other vehicles with different investment criteria and use a much larger share of the database of algorithms.

"Alpha is interesting because it's the toughest problem, but the structure makes it very clear which algo is producing value for the vehicle, and so the relationship we have with those algos is the most clear and simple. That's a really good starting point for crowdsourcing. So it was a trade-off: do you want to tackle the most difficult problem available, but have the best relationship with your community."

As Quantopian evolves and becomes more automated, its selection process, which is also handled by an algorithm, can be subjected to the same level of statistical rigour and stringent testing as the trading algorithms.

"I think what we have realised in the last six months is that if we have a fully automated process for selection, that is itself a kind of algorithm, it can be tested with the same level of rigour. If it's fully automated, we can pick up our own selection process, and do the same kind of simulations and evaluation of that process. That's really powerful because it speeds up the cycle of improvement."

A new and important addition to the platform is a multi-factor risk model. The new contest features a set of constraints participants need to satisfy, such as turnover and also maximum exposures to the risk factors in the model.

"It's hard to overstate how big a deal that is," says Fawcett. "That's one of the things that's most proprietary within asset management or within the vendors. Ours is free. You can analyse any return stream with it. Shortly you'll be able to use the risk factors to optimise your portfolio."

Here is Quantopian's latest free tutorial on how to get started in quant finance/algo trading.