EXCLUSIVE: “A thoroughly modern Man” – Gary Collier, Man Group in ‘The Fintech Magazine’

Gary Collier, CTO of investment manager Man Group’s Alpha Technology Division, describes how tech augments today’s traders and how it will in the future

Man Group is an active investment management firm with $144.7billion in assets under management. It is the world’s largest listed hedge fund manager. At the heart of the firm, and driving its performance, are two technology divisions. Alpha Technology is responsible for innovation and building the infrastructure that, ultimately, drives all its investment decisions; Trading Platform and Core Technology powers the day-to-day execution of trade.

Last year was a financially strong one for Man Group, with full-year core pre-tax profit coming in 18 per cent higher than 21/22 at $779million – largely due to the firm’s computer-driven macro funds.

We spoke to Gary Collier – who oversees around 230 engineers and data scientists as CTO of Man’s Alpha Technology Division – about strategy,access to data and what traders of the future will need.

THE FINTECH MAGAZINE: How do you build a strong core data set, and how do you then build off that component?

GARY COLLIER: Data isn’t just something that comes in from the outside world and drives our investment decision-making; it also underpins every single part of our value-generation process, across the firm – that’s data ingestion, the quant modelling that we do, the portfolio construction, the risk modelling and trade execution. So, think of data as the foundation, which underpins every single part of that value chain.

In different parts of that chain, there are different challenges that need to be thought about quite deeply. Not just the standard ‘extract, transform, load’, but also mapping data into something we’re going to trade. There are bigger and bigger volumes of data, and we need to build more and more complex models, to handle it because the pursuit of alpha is hard and doing that at scale, and doing it efficiently, is harder still.

TFM: How do you make the most of that data and make sure your traders, or portfolio managers, are able to access it when they need it?

GC: Having investment professionals with market knowledge and access to the right tools and data, so they can self-serve, optimises the end-to-end of data being available and the action being taken off the back of it. So, we can look at things from a purely technical perspective and say ‘how do we minimise latency?’ or ‘how do we make the computer systems faster?’, but the broader way of looking at timeliness of data, is to put it in the hands of people who can make actionable decisions off the back of it, and look to make sure we can optimise that, end to end.

If we want to put data to good use, therefore a number of different pillars come into play. The first consideration is that our investment professionals increasingly need to be well-versed in Python; that’s pretty much the de facto language now for dealing with data and doing data science. So, to make sure our investment professionals are suitably equipped, we have an internal programme providing Python and data science training.

Then we have a powerful, rich technology platform, that provides those professionals with the tools they need, in a very easy-to-access fashion. Literally, anyone in the firm can go to a web browser, type ‘Python’ into the URL bar, and a Python environment materialises for them in which they can start writing code.

Finally, there is the data itself; the fuel to that process, all of which is accessible via an in-house data catalogue. People can search for the data; the data being appropriately permissioned for their job function. The licensing consideration are all taken into account and then the Python code they need to run to access that data is presented to them. By building all of those pillars, the training, access to the technology, and access to the data, it helps optimise that end-to-end process, from data coming into the firm, to actionable insight being taken as a result.

TFM: You’ve just released a new data-focussed product, ArcticDB. Can you tell us a bit more about it?

GC: Regardless of the form data takes when it comes in from the outside – it might be numeric tables, text documents, etc. – it is typically the case that you won’t be able to proceed too far into it before you end up dealing with DataFrames of data.

“People talk about ‘tech’ and ‘the business’ as two separate entities; they’re not… Roughly half of my team are embedded in investment management teams, working side by side with quants and portfolio managers “

Imagine an Excel spreadsheet at an industrial scale, with hundreds of thousands of columns, millions or billions of rows. That’s DataFrames – the sort of complex quant modelling we do might see a model that’s composed of thousands, or tens of thousands of these very large frames of data. Simply put, ArcticDB, allows our investment professionals to interact with DataFrames, the natural unit of what we’re doing here, in as seamless a fashion as possible. It does lots of other clever stuff behind the scenes, too, that are attuned to some of the problems in our industry, such as bitemporal modelling or the ‘point-in-timeness’ of data – not just knowing what the data is now, but what it looked like when it was published, which supports complex back testing.

But I wouldn’t pigeonhole ArcticDB as just a tool for a particular corner of quant finance. Anywhere that you’re doing data science, you’re probably using Python, and wherever you’re using that technology, be it in quant or the discretionary space, there’s a place for it.

TFM: Is it difficult to strike a balance between speed and reliability of data needed to empower your investment managers?

GC: On the face of it, these are two competing concerns, as you imply. There is the need to be agile, but also – when managing billions of dollars’ worth of client funds – the need to be very accurate and reliable.

I think of our technology in terms of two estates. Firstly, there is the research and development estate, the type of technology platform that allows investment professionals to iterate very quickly, in what amounts to a financial laboratory with lots of compute cores, lots of access to data, lots of Python. It’s an environment that facilitates interactive, iterative research and where models are developed and back tested.

Of course, when we take our higher conviction models, and want to trade those in production with real money, different rules apply. Here, there’s a completely different part of our technology estate involved, which I’d refer to as our production estate. And it’s one that’s very carefully locked down.

I’m a CTO, and even I don’t have access to that part of the system; indeed, only a dedicated group of operations specialists do, and they’re the folks who are responsible for making any changes. That isn’t to say we’re not agile there, too; indeed, we’re releasing roughly 5,000 changes per annum to production. To make sure this is done as safely as possible we adopt a very progressive approach to engineering: continuous integration, continuous deployment, a substantial amount of automated testing, automated deployments, etc.

One thing I’d add – in terms of organisational construct – is that I dislike it when people talk about ‘tech’ and ‘the business’ as two separate entities; they’re not. Tech is an integral part of Man Group’s business and that’s reflected in the way i n which a lot of the alpha tech engineers work. Roughly half of my team are embedded in investment management teams, so they’re working side by side with quants and with portfolio managers.

TFM: Earlier this year Man Group ran an experiment with ChatGPT to find out what it ‘thought’ about investment strategy. The results, it would be fair to say, weren’t particularly impressive! So where do you see automation and AI being used in this industry in the next two to five years?

GC: I think there will be an almost relentless focus on general automation efficiency, and using that to help drive innovation across everything we do.

Secondly, there are certainly some pockets readily identifiable now as areas where we will see, or are seeing, increased automation. If we look at the electronification of different asset classes, then credit, for example, is probably following a path like equities took some years ago, in terms of greater electronic market access of liquidity.

When it comes to generative AI, which of course is very topical, absolutely, there’s scope for these technologies to help automate some of the toil that’s inherent in writing software – for example, automation of construction of unit tests as a first pass. It doesn’t take the role of the engineer away, but it certainly gives them a starting point.

TFM: What skills would you advise future traders to be looking at, given an ever-changing ecosystem where suddenly they’ll be able to employ automation and use generative AI tools?

GC: I’d focus on the power of combining deep market knowledge with technical knowledge and with the ability to manipulate data. If you can master those three skills, then it will be an enormously powerful combination to bring to bear on the job, and be enormously impactful upon the organisation you work for. In terms of specific advice for traders, learn Python, learn data science tools, learn about that ecosystem.

It will offer a deeper understanding of how markets work, but it will then also allow you to look at the algorithms your organisation has built and ask how are those algorithms performing; is there scope to fine-tune them? Collectively, I think those skills provide a very powerful arsenal.

This article was published in The Fintech Magazine Issue 28, Page 80-81