Automation, a buzzword of the moment, is drastically changing how trading desks are run. But the way that firms are approaching the technology differs greatly. As traders seek out increasingly marginal gains, the automation game is a competitive one – with numerous approaches.

The primary use case for automation in trading is around small orders. “It saves a lot of time,” says Evan Canwell, equity trader at T Rowe Price. “Automation can take some of those lower value-add jobs that are more repetitive and free up time so we can concentrate more on the higher value-add things, look at market colour.”

At T Rowe, smaller orders have become increasingly handled by machines over the past

eight years. No longer do traders have to take responsibility for tiny trades, selecting which algo to send them off to. “An order comes in, the parameters and characteristics are captured, and it’s sent off to our performance database,” Canwell relays. “We do some statistical analysis, a suggested broker algo is stamped on it, and if it’s a small enough order it will go straight out the door.” Humans still oversee the process, but with a much lighter touch.

It’s a similar story at UBS Asset Management (UBS AM), one that head of European equities trading Stuart Lawrence says aligns with the general approach on the street. “The [small] orders hit our trading desk, automate out to our EMS, execute in the market, come back, and mark themselves,” he recounts. “ The traders use pre-programmed execution

strategies that are driven by the technical factors of an order, which allows us to then automate the entire workflows.”

Lawrence shares that UBS AM has been automating much of its ETF flow over the last year.

“We’re constantly trying to see what else we can automate, freeing our traders to do more thinking and less processing.”

“Without going into specifics, our automation is set up to deal with a host of order types and sizes. We initially set up automation to assist with the trades where traders can add limited value (low ADV, low notional) but we are now expanding our methodology to add new approaches and opportunities to make our workflows more efficient.”

Complex orders

That said, “we don’t just throw everything into automated systems,” Lawrence assures. “We have very strict rules around which orders can be sent through – based on notional, ADV, the market, etc. and we always ensure we have someone on the desk with oversight of the automation at all times.”

For UBS AM, orders requiring more attention include those with a high notional value, super liquid stocks or those that are part of a contingent basket, “That’s where traders can really add value, based on their experience and knowledge of how best to achieve an outcome,” Lawrence adds.

For larger orders, those that traders are expected to focus on once the scraps have been hoovered up, there’s still a place for automation – but as a copilot rather than a controller. “In these cases, an algo suggestion is just that – a suggestion,” Canwell says. “The trader can choose whether to take it or overrule it.”

“There’s a point when the order stops being something that can be automated and needs a trader’s eyes on it,” Lawrence says. “The desk needs to be nimble when it comes to finding that sweet spot. We’ve coded our parameters for automation to allow traders to focus on where they can add the most value.”

Canwell agrees. “Our non-automated flows are those which require some form of human intervention, so they’re typically larger or block-sized, or have specific execution instructions from a portfolio manager.”

Totally random

Broker allocation and algo choice are one of the key uses of automation in trading, but the degree to which these practices should be randomised is up for debate.

At UBS AM, Lawrence explains,“We have full randomisation within our algo wheels, which promotes fairness for measuring performance, so no one broker benefits more from the orders they receive than others. It’s a kind of league, relegation-promotion system; if one broker is consistently underperforming, we either reduce their weighting in the wheel or remove them entirely.”

The wheels are assessed twice a year – Lawrence expects the timeline to be condensed to once a quarter in the near future.

“We need to be constantly asking who’s best for this particular market, this particular strategy,” he says. “It’s important that we’re constantly evolving. We don’t just keep brokers because they’re the incumbents.”

UBS AM uses a mixture of in-house and vendor applications for its automation projects, Lawrence notes. Its algo wheels are built and operated within the firm’s EMS, while OMS to EMS automation is developed internally.

At T Rowe Price, the set-up is very different. There’s a degree of randomness to the allocations – between a 10-20% chance of any broker being selected – to ensure that the top algos are performing consistently well. But for the most part, the order goes to the best performing broker.

“We’ve never really been a fan of randomised algo wheels,” Canwell shares. “We want to replicate the way that a trader would think about an order, rather than replace the trader. If they’re thinking about an order manually, and don’t have an initial feeling for where to send it, they might look at TCA data and see which brokers have performed best in past similar cases. That’s what our service replicates.”

For Canwell, the key is mimicking human behaviour and offering suggestions to traders.

2.0

Lawrence wants to start using ‘waterfall wheels’ at UBS AM, a second generation of algo wheel that goes beyond just routing orders based on set parameters.

“As a simple example, think of it as a trickle-down,” he explains. “First we see if we can get a mid-price fill. If that doesn’t work, we bring it down to the next level and try to get 100% fill on the touch. If that doesn’t work, we might try and rest in dark for a couple of minutes. If that doesn’t work, we finally might say, ‘well, there’s no other liquidity out there that we can use right now’, and route to the algo wheels as we previously did.”

With the liquidity landscape changing, and electronic providers increasingly prominent, there are a lot more flows to tap into. Desks want to take advantage of that.

“It treats smaller orders in a similar way to bigger ones, so we always see if there’s an opportunity to trade in the spread through liquidity providers who offer mid-spread liquidity,” Lawrence continues. “We can eke out small gains from that. If you’re constantly getting an extra 30% of your fills at the mid rather than on the touch, you’ll see a meaningful difference to performance.”

READ MORE: Fighting information leakage with innovation

This kind of tool is already used at rival asset managers including Groupama, former head of trading Eric Heleine told Global Trading last year.

Some market players have suggested that while this kind of waterfall logic is nothing new, it has been automated in recent years thanks to increased bilateral trading and the democratisation of more advanced execution management system technology.

Pre-trade

Beyond allocation, automation also has a significant role to play in pre- and post-trade analysis to enhance execution.

T Rowe’s Canwell detailed one way that the firm is thinking of bringing automation into its data processing.

“We put together a daily trading note manually, bringing together Bloomberg messages, broker emails, things like that. In the future, we could compile all those into a nice format – probably using an AI tool like an LLM – and then have a trader decide whether that output is good enough or needs changing.”

It’s something that’s front-of-mind for investors across the board. “Automate the noise, focus on execution,” is a focus of many.

There is some question, however, as to how much data will be accessible for such systems – big data providers may be less keen for their outputs to be scraped. Equally, trading firms may not be ready to hand over their data. One trader observed that although there is a lot of talk about the use of AI on the desk, the reality is that few are actually implementing it in meaningful ways.

T Rowe’s automated trading services have all been developed in-house, with just one external vendor used. “That’s purely for supplying data points, such as pre-trade cost estimates, to help choose an execution algorithm,” Canwell affirms.

Already, according to Global Trading’s recent buy-side survey, just 14% of traders are using in-house tools for pre-trade automation and complex execution. The majority outsource these tasks to a vendor tool (59%) or use a broker tool via website access (27%). Automation can be both a helpful tool for firms with limited resources and another budgeting headache – especially for smaller companies.

READ MORE: Trading under the microscope

Earlier this year, Mathias Eriksson, senior trader equities and credits at Swedish pension fund AP2, shared how the firm uses vendor AI technology to wrangle data and lead its broker allocation decisions.

READ MORE: Trading at the human-machine interface

From a vendor perspective, Ovidiu Campean, product director for LSEG Execution

Solutions, sees the benefit in giving clients the opportunity to customise their solutions. “The more flexibility we can provide to clients to incorporate their own logic and criteria into the wheel and into the automation, the more we see clients implementing our own automation and wheels,” he affirms.

“We’ve seen shifts towards Alternative Trading Systems (ATS) and conditional block cross platforms, which are meant to help clients find bigger blocks of liquidity while minimising slippage and market impact, and reducing adverse selection. There are also many that employ diverse AI models to improve execution quality.

Giving insight into LSEG’s offerings, he added, “Our order and execution management systems operate as normaliser layers across both lit and dark type venues, so we can provide venue agnostic access to liquidity, improving clients’ chance for best execution. We’re continually adding venues to our platform, so we’re riding on this trend, giving clients optionality and flexibility across global asset classes.”

How low can you go?

It’s easy to get swept away by the thought of algorithms and automation taking over the trading floor – but even enthusiastic market participants are clear that this is some way off.

One trader explained that automation could, for equities, have reached its ceiling. Aside from small tweaks here and there, and advancements made possible through enhanced data visibility (which the consolidated tapes intend to bring), there is not that much room to grow in the asset class.

Others disagree, believing that automation can go further – with caveats.

From a governance perspective, regulation of automation needs to include internal policy frameworks for fiduciary duty and trading ethics, ensuring that brokers or routes are not favoured in a way that harms clients, Campean notes.

“Another thing to monitor is data quality and input controls, where you see how much data quality contributes to the decision an automated system is making. You have to have high-integrity, latency-aligned, real-time data, or the model will perform on stale sources.”

On a practical level, he adds, “For a while you’ll need human oversight and controls, especially kill switch and escalation path capabilities. If the model goes haywire due to high volatility or the system connectivity to markets breaks, you need to have controls in place, otherwise the model could flood the system or the exchanges.”

“Reasoning artificial intelligence may still be a decade away, and until such robust systems and models emerge, the idea of eliminating human oversight remains highly improbable.”

That said, Campean does expect to see AI being increasingly used in trading automation, acting as a co-pilot to traders and providing increasingly complex insights into growing reams of data.

Canwell sits between the two perspectives.

“There’s already some level of machine learning in algos, used to fine-tune parameters over time,” he says. “I can see that increasing, with more brokers pulling in more data to improve their offerings. The world is only getting more quantitative, and I think traders want to be able to benefit from that.”

But, he concludes, “the near-term future won’t be too dissimilar from where we are now.”