Artificial intelligence is not ready to go into live markets, according to one panel at TradeTech in Paris today, who were advising on the better ways to develop algorithms in trading, and took a cautious approach to the veracity of ‘artificial intelligence’ in this discussion.
“In terms of the role AI plays in that, it is important to remember it is often not AI at all, it is just automating a few clicks,” said Daniel Mayston, head of electronic trading and market structure EMEA, at BlackRock. “The FCA published a paper a few years ago on wholesale algos and how to think about them, and I would really recommend it because it discusses the role of the algorithm manufacturer i.e. the sell side and the role of consumer i.e. the buy side.”
He continued. “If you take those terms you can understand what do I as a user need to do to monitor better to get the most of these tools. Understanding the roles that you play is a good set up to support innovating in the right way. As the recipient of sell side innovation, from a buy side point of view it is important to be smart about what you do and what you do not do. It is not my job to second guess a sell side algorithm.”
Where AI is in use, it still needs to be well contained, said Joe Wald, managing director, co-head of electronic trading at BMO Capital Markets.
“At this point there is a separation between machine learning/AI from a quantitative research perspective, and then not letting that loose into the wild in terms of making execution decisions,” he said.
To support development of both AI and more simple algorithms, the starting point is data and the ability to consume it, which is fuelling innovation by market infrastructure providers.
Peter Ho-Spoida, VP of data strategy at Deutsche Börse Market Data + Services said, “We see not only demand for data but demand for the right tools to analyse this data, allowing clients to focus on their core expertise and leave the heavy lifting to other tools which are out there. For example we have create a back testing environment, and algo simulation environment which has embedded in it order book reconstruction and signal generation, so clients can focus directly on that, and back test and run their strategies on our infrastructure. But this raises another point; how much is the client willing to use third party infrastructure?”
For dealers that is one part of the problem, however there are many issues to be considered.
Jas Sandhu, global head of agency electronic solutions at RBC Capital Markets, says, “When I think about algorithms, I think it is multi-faceted. There is more to the development of the algo; that includes its trading behaviour, how you take that infrastructure and make it more multi-asset. We’ve already seen the growth of portfolio trading in credit up 30% year on year, so an asset class that has been electronic becoming more algorithmic. But we would be remiss not to talk about algorithmic documents and rules. I think depending on where we land in that space, whether that be prescribed or principles-based, will dictate how much innovation we see. Because that will create barriers to entry.”
Currently that includes the development and retention of skills which can be applied across the technical and markets areas.
“The war for talent in this space is real,” says Sandhu. “Many years ago, quants or engineers or digital talent would move between banks. Now if you can code in Python or Q you can likely code in Rust and work for crypto firms. So you can get talent that is fungible across industries. What we don’t want is to see key risk takers on the algo side just churning between roles.”
Ho-Spoida added, “You need to bring together markets’ expertise, couple it with the right technology and the data science component which is new. If you are able to bring together these three dimensions, you can get superior performance and improve certainty of execution.”