TradeTech 2023: Artificial intelligence is not a silver bullet, say buy-side experts who see challenges around explainability, resources and transparency

Artificial intelligence and machine learning will play a role in the future of the trading desk, but experts put the brakes on a broad application, citing concerns around potential misuse, explainability to stakeholders and appropriate in-house expertise.

Buy-side organisations are likely to prioritise addressing the mounting concerns over the downside risks of artificial intelligence and machine learning on the trading desk over the comparatively limited benefits of the technologies, experts have opined.

Speaking at TradeTech 2023, representatives from HSBC and Natixis joined head of EMEA equities at BestEx Research Matthew Cousens to discuss how asset managers can leverage the latest applications across their trading desk to access unique liquidity pools, enhance price discovery and access real-time market insights.

While the capabilities of artificial intelligence and machine learning were praised by the experts who admitted they would have a place in the future of the trading desk, it was the application of the technologies among other challenges which were hotly debated.

“It’s a valuable tool because it is a paradigm shift from traditional software engineering activities,” said George Marootian, head of technology, Natixis Investment Managers.

“Anyone that’s been in this industry for a couple of decades has seen different software languages come and go, [they’ve seen] web 1.0 and 2.0, social media and different paradigms that have contributed to improved activities on the trading desk. What AI allows us to do is leapfrog decision signs, and really accelerate finding answers to key questions…but, it’s not a silver bullet, or a tool to use everywhere.

“There’s still a requirement for straight through processing, robotic automation, and there’s a whole other realm of digital technology that’s not AI oriented.”

Daniel Leon, global head of trading, treasury management & global solutions, HSBC Asset Management, had been adamant on another panel in the day that everything should come back to data and finding liquidity, and continued on the same thought path during this panel discussion.

“If you want to do best execution then you need to know what you’re doing so if you’re using AI that’s calibrated based on the past to get the best outcome then it’s not of use that you can trust it to know what you’re doing in the future,” he said. “That’s going to be a challenge and that’s why – for me – it’s not necessarily the way I want to go.

“What I believe AI will help us do is quickly – when you have an illiquid trade – pick up the elements of information – unstructured info – and use it to tell you that last week this traded at this level and especially for the illiquid trade it will help us find this path to liquidity.”

Among the other issues mentioned around the use of AI and ML was explainability, which was a notion repeated throughout the panel.

“We all need to very understandably be able to explain how processes work to our customers, stakeholders and regulators,” said Jesse Greif, COO, OneChronos, a technology outfit which describes itself as being “at the intersection of capital markets, machine learning and mechanism design” providing execution venues to those in the electronic trading world. “There’s this balance of having solutions which are blackbox in nature which potentially inhibits the evolution and application of machine learning techniques. They are very data intensive, you need clean data,” he added.

He described this explainability issue as one of a handful of factors which has “slowed progress in this domain in relation to spaces outside of financial markets where we have seen faster evolution of these type of techniques.”

Also contributing to the slower pace has been resources and expertise, given that some of the skillsets needed to oversee processes and deploy the technologies are specific and outside of the domain of many current team members within buy-side organisations.

Marootian agreed: “The skillset of the resources that are going to be able to help you progress AI/ML initiatives are not typical resources you have in software engineering teams. It’s a much more data intensive, analytics mindset.

“A lot of people trying to move software engineers and data folks into the AI space, or taking visualation folks and training them up and all of a sudden they are data scientists.”

Other concerns include transparency and the potential misuse of AI/ML.

Greif – an expert in the field – said regarding the misuse of the technologies impacting investors that there is a need for transparency, benchmarks and guardrails and urged users of the applications to press their brokers or advisors on how these applications work and who is working on them is important.

“Are you using something offline to coach something that’s deterministic and explainable? Are you using a blackbox solution but you have some benchmark that is a standard explainable benchmark to know if it is performing or underperforming?” he stated. “There’s an emerging field called computation ethics – or machine ethics – which really dives into this topic, specifically talking about a trade-off versus accuracy and fairness.

“[But] you really have to spell out what’s important and therefore the desires have to be sensitive about it,” he concluded.

Ultimately the panel came across very cautious on the use of AI/ML despite conceding it will be a part of the desk of the future. “We want to press ahead, but we need to be careful on how we do that,” said Cousens.

Meanwhile, Marootian added: “It will definitely part of the arsenal, but not the answer to every question.”