Incentivised to explore the state-of-the-art
Our ML researchers have huge amounts of (clean) data and massive compute at their fingertips, with which they predict the future of financial markets. Because this is a very mature prediction problem, finding the 1% of difference, working at the very cutting edge of developments, is the place where success happens. We are, in effect, incentivised to explore the state-of-the-art. We do not work on cool problems for the sake of it, we work on them because we know one small discovery tilts the balance in our favour. While we start with the standard ML toolkit, to make a model work our researchers really need to understand what’s going on, not just throw an out-of-the-box solution at a dataset.
Unlike pure problems, our researchers get near-instantaneous feedback on their work in the form of absolute numbers. There is no dotting i’s and crossing t’s (unless that improves your work): you build a model, back-test it extensively, then put it into production. It either works or it doesn’t – there is no debate or politics, just quantitative outcomes.
Collaborate on a common code base
A typical project lifecycle starts with the pencil and paper stage, and instead of selling your work, we run back tests. If it’s profitable, that’s enough. Our equivalent of sending off to peer review in academia is going to out-of-sample stage. Rather than publish we put research into production. As well as thousands of GPUs and tens of thousands of CPUs, our researchers share a common code base, tools, and techniques, collaborating on these to the benefit of everyone. We are also actively exploring next-generation forms of computing to maintain our edge.
While the UK COVID-19 lockdown continues, G-Research is happy to confirm that we will continue to interview, hire, and onboard new staff remotely. Please do not hesitate to send in your application for a role.