7 min read
Outside of talent, technology and data are the biggest issues that need to be addressed.
(This article first appeared on Best Execution, a Markets Media Group publication.)
The early years
A computer science graduate, Fernando started his career with a tech grad scheme at Barclays working in sales trading, experimenting with technical workflows before moving over to a consultancy to hone his skills across a wider variety of clients. From creating trading UIs for an investment bank to building out a price matching engine for a commodities trading house, it gave him management experience across a broad spectrum. “That gave me a lot of insight into the different financial problems you can have across different clients, as well as an understanding of how to build out teams, how to report to stakeholders, and how to deliver in a short timescale,” he said. “The difficulty in a role like that though, is that you don’t really get to choose your career path. You might be working with a client you don’t like, or a project you’re not keen on. It just got to a point where I wanted a bit more control over my own trajectory.”
At the same time, Fernando had started to develop a specific interest in certain areas: particularly around how best to operate in a space which is very data-heavy and fast-moving, and how to incorporate financial data produced by multiple different workflows into effective risk platforms and trading strategies.
That led him to a role with Goldman Sachs Asset Management (GSAM), where he remained for eight years, initially running a core risk management tech function and then latterly focusing more on the front office portfolio and construction elements of the workflow. “I was looking at risk across a plethora of portfolios, both insurance and separately managed accounts, munis, high yield, all within the fixed income space. It was very data-centric, we were looking at how to reduce the workload on traders so that they could optimise these portfolios quickly and produce trades.”
As a discretionary-focused shop, GSAM was keen to reduce costs by automating workflows, and Fernando found that area very exciting. “It led me to start to think about what that could look like if that was the only thing you did and how you could build a business around that.”
From short to fast
That dream became a reality when he joined Man Group in October 2021, and was promoted to head of fast trading technology a few months later. But what does ‘fast trading’ actually mean?
The team sits within AHL, which is one of the systematic engines within Man Group. “As a systematic, the workflow is that we start first and foremost with data, and with data quality. It could be one-day data, it could be streaming data, it could be market data. Then we run signals, we try to think about what this data means in terms of its impact in prices of different assets. We use a variety of mathematical techniques and different models to derive that, based on insights that we find within the data as well as what we see happening in the market. And then from those signals, which are predictions of prices of assets, we can run optimisations to come up with trades.
“The thing that makes it fast trading rather than just a strategy is the frequency with which we’re running those signals and how often we might hold assets off the back of the result of those signals. So in our space, we’re looking at signals which would produce positions with shorter holding periods. That has a lot of interesting repercussions downstream of that signal generation workflow. Because we’re holding positions for shorter periods of time, we often end up with a lot more trades. That means that we really need to be able to automate our trading flow, otherwise we’d be overwhelming the traders by sending them trades far too frequently.
“The team used to be called short-term trading. The change in name to ‘fast’ trading was initially more aspirational, but I think it is something we have achieved over the last few years, and now it’s very fitting.”
From fast to faster
One of the interesting things about Man’s fast trading team is that it is very much cross-asset – and while that can be problematic, it also inspires problem-solving.
“There’s a huge amount of innovation within each asset class, but if you look at FX versus fixed income versus futures and so on, each space is totally different in terms of how you trade, the data you get, the algos and capabilities that are available. So one of the challenges is trying to build out strategies which leverage all of these different workflows. We have a lot of different research experts within each of these asset classes who know them inside out, and then the tech platform supports that whole workflow in terms of onboarding data, providing a trading workflow for all of those things, and then bringing it together as a cohesive platform.”
Another challenge lies in the execution paradigms which are happening as a result. “With numerous fragmented markets that trade using lots of different protocols for each asset class, it’s hard to best optimise the execution. You really need to understand in depth what the best approach is for each situation, and be able to make that look like a cohesive, slick platform for the researchers and traders to use and monitor.”
Although Man Group has a well-established in-house tech stack, the firm is not averse to buying in either. “We look at what’s available in the market and we do use external vendors where it makes sense,” said Fernando. “There’s often some capabilities that we think are best delivered by external vendors such as pricing data and market insight. Tech elements are often built out in-house because we want to make sure that we’re not highly coupled to the solutions. We don’t discount it as an option though – we do use vendor tools where it makes sense.”
The fast trading team is also responsible for optimising the trading flow for all of the other sectors within AHL – and the goal is to try to automate as much as make sense. A key priority is the evolution of its algo usage. “We’re currently looking at how to achieve more cohesion in how our execution algos work cross-asset to our strategies, and making that a pipeline that is seamless, monitorable and visible to the traders,” explained Fernando.
The team views algos from three different perspectives. The first is the broker algos provided to them for individual asset classes, whether a TWAP, VWAP, implementation shortfall and so on, which they evaluate on their own merits. The second is building their own algos to slice their orders, and then leveraging some of those algos to execute directly.
The third, however, is a more macro approach to innovation. “A lot of these algos take a parent order and then just operate on a time window based on the order coming in. But a lot of our strategies are more complex than that. They’re already fully automated. So we think of it as more of a holistic workflow, where we’re already looking at the next order that’s coming in, as well as the in-flight order. That’s where the strategy element comes in. It’s not just an execution algo, it then becomes more of a systematic workflow.
“But because these strategies are often trading cross-asset, you need to make sure the orders in one asset class are linked to the orders in another asset class – if one fails, you might not want to do the other one. That interaction is tricky to manage and deliver as a tech solution, so that has been a challenging area for us.”
Quants vs traders
Often automation is seen as something which is supposed to help traders, but Fernando has a different approach. “I think automation is something that is supposed to help your clients and your execution. It doesn’t necessarily need to make the traders jobs easier. It’s supposed to help clients and reduce costs.
“We’re trying to get to a place where you can codify the decision-making of a trader. And it’s not always about the trader, sometimes it’s about the decision that a quant would make, if they were given all the information to make that decision. It’s probably not the same decision that a trader would make. A trader is looking at the market: they have a huge amount of market colour, and they have some great relationships. We want to leverage our traders for those relationships and that market insight. But it’s often the job of a quant to understand all of the historical data sets and all the time series data to make the best execution decision. And then after that, then we’ll layer in the trader’s insights to tell us if that makes sense and affirm it. It’s not a decision that gets made on the fly, it’s something that we take time to understand, and then we try to input that into the trading workflow so that it is repeatable and we know how the strategy is going to behave on any given day.
“That can be problematic sometimes because we haven’t coded it to understand every market event. So you do need that trader oversight in order to intervene or to add additional colour to the workflow when those events happen. We’ve recently been looking at hiring quant traders who can do a bit of both, so they can do all of the data analytics but they also have the trading workflow knowledge and relationships with our counterparties.”
Relationships are important, as is underlying trader knowledge, and Fernando doesn’t disagree with that. “It’s just that the way we leverage them can be slightly different. We focus on how we trade, rather than on a specific trade. That’s one of the biggest differences for us.”
That’s why Man Group recently ran a ‘Traders That Code’ programme to help all its traders to understand Python. “We now see them regularly leveraging that data to understand things like liquidity or volumes, and it’s saved a lot of time, because it means that we don’t have to often build out a lot of ad-hoc tools for them, they’re able to self-serve.”
The firm has a number of target areas towards which it is directing tech investment in 2024.
It will come as no surprise that data quality is a key focus. “Being in systematic, we have thousands of different data sets that get delivered every day. It’s really important that we have a way to catalogue those datasets and have checks on them to make sure that they’re relevant for the target workflow.”
The fast trading team uses a data catalogue called Codex, which helps them catalogue all their data items. “That’s been really beneficial in helping our quants search for specific elements. Raising awareness is a challenge, and you don’t want to be onboarding the same dataset twice, so it’s really helpful to have all of our licensing and permissions and visibility in one place. We can then tag those datasets with what are essentially quality assurance workflows, such as ‘alert me if this data isn’t delivered by this time’ or if it has a lack of correlation. Those checks have been fantastic.”
There has also been a notable proliferation in the volume of tick data that’s generated from platforms – and that has led to issues around how to best simulate the trading, especially when you have data that ticks every second or less. “We want to be able to explain a full history of our trading algos. That brings up a huge challenge around how you make sure that the quality of that data is up to scratch. We’ve invested a lot in data quality checks on that type of data,” said Fernando.
Fernando is very excited about the possibilities of GenAI in the research process. “We had to do a lot of due diligence to make sure it was OK for us to leverage that tooling. That was a big hurdle for us, so we created a wrapper called ManGTP which allows us to interact with the OpenAI model – it is very heavily leveraged within the research teams to do things like summarising research or figuring out what might be wrong with a piece of code.”
The other space where Man Group is leveraging AI is within the development environment. “We’re looking at various code interaction workflows through tools like Co-Pilot and Codium to interact with the environment within both our Jupyter notebooks and our Python environments to suggest ways to change our code to be more optimal, or even generate pieces of code for us to be more efficient. There has been huge usage within the tech and research platforms to make us more productive. We’re still thinking of ways we can leverage that technology, but the important thing is that we need to be in control, ensuring that risk is top of mind.”
A core target for 2024 is to improve the back-testing simulation platform within the fast trading space to cope with the amount of data that is being produced.
“Another key area we’re looking at is how we can leverage ArcticDB more heavily within the space to be able to consume those huge datasets and work with that team to more easily run the types of analysis that we do on that platform.”
ArcticDB, a python database developed in-house and run by ex-AHLCTO James Munro, was recently made public, with Bloomberg as the first big client. The firm is also exploring licensing options with other third parties. It is hugely valuable to the team as a means of storing and analysing time series data, and a priority is to build out its capabilities more extensively. “It’s extremely fast, so you can retrieve data very quickly across multiple datasets, and then leverage that within your trading models. We leverage that technology very heavily within our investment process, and we are now looking to collaborate with the open source community to build out functionality and expand its user base.”
So where does the fastest man at Man want to be by this time next year?
“I’m fascinated by the applications of GenAI and new hardware applications in GPUs and fast compute. I’d like us to be leveraging those more in our trading process, that’s really exciting for me,” concluded Fernando. “If I had to choose, those would be my two target areas of progress.”
©Markets Media Europe 2024