Artificial intelligence (AI) can encompass a broad range of technologies which support pattern recognition and analysis, potentially allowing investment firms to develop new signals or triggers for market engagement.
Speaking at TradeTech in Paris, Daniel Leon, global head of trading, treasury management & global solutions at HSBC Asset Management said, “We love to talk about machine learning, but ultimately, it’s about data and advanced analytics. Having the capacity to make decisions which are going to help find liquidity or a provider of liquidity. The big question to discuss with AI is around natural language; how can you use AI in order to assess liquidity through unstructured data?”
George Marootian, head of Technology at Natixis Investment Managers said, “It is a paradigm shift from traditional software engineering or database activities that existed in the past. [We] have seen different software languages come and go. Paradigms that have all contributed to improve activities on the trading desk and other areas of buy side business. AI allows us to really accelerate finding answers to key questions that could drive activities. It isn’t a silver bullet for the tools used everywhere, or things like straight through processing and robotic automation.”
Jesse Greif, COO at OneChronos, a tech company that builds trading venues and matching engines and has a current product that functions as a US equities dark pool or alternative trading system (ATS), outlined how AI has guided his firm’s evolution.
“Here in Europe, we have a unique approach in that we’ve created an environment to compete on the quality of liquidity rather than the speed to access liquidity,” he said. “We do that by running time randomised periodic auctions about 10 times per second, so every 100 milliseconds on average. Each of those auctions has an optimisation process that optimises for best execution outcomes.”
That necessitates using operations research and computer engineering techniques to be able to run auctions quickly and to find solutions for trade matches for its customers.
He continued, “We run a time-constrained optimisation, run it offline for hours using reinforcement learning, it learns more context about the search space of possible candidate solutions which represent potential trade matches, which coach our portfolio of optimisation algorithms to find even better starting points in the future for future auctions that have similar auction dynamics. It’s actually the silver bullet to allow for scale and really couldn’t have been done couldn’t have been done before.”
Some of the limitations in the area are computational, another big issue is explicability, transparently explaining how processes work for customers, stakeholders, and regulators.
“That’s definitely a challenge, skills to progress AI/ML initiatives are not the same as in typical software engineering teams,” said Marootian. “To put together or implement computations and code is one thing, to be able to make it explainable is another, higher order skill set. It requires some specialty knowledge of the domain as well as the algorithmic skill set.”
“It’s definitely going to be a big challenge,” agreed Leon. “And it’s going to be to affect the way we want to use AI. If you have an AI that’s calibrated, meaning you’re doing something based on the past to get the best outcome in the future that’s going to be a challenge. We have axes, we have IOIs, and a lot of unstructured data. I believe AI will help us to pick up the elements of unstructured information, and give you some insights to help us find paths to liquidity.”
©Markets Media Europe 2023
