MarketAxess is leveraging artificial intelligence to accelerate the development and testing of new fixed-income technologies and design more customised execution capabilities for buy-side clients. Speaking at the Fixed Income Leaders’ Summit in Boston, Global Head of Research Julien Alexandre and Adaptive Automation Solutions’ Andrew Cameron discuss how AI has helped the firm cut research and experimentation cycles from weeks to hours and helped them deliver more tailored solutions based on individual buy-side objectives. The next phase of adoption will focus on embedding AI into real-time trading workflows, while improving transparency, verification and user confidence in AI-driven decision-making.
Interview
Josephine Gallagher – Welcome to Trader TV. We’re here at the Fixed Income Leaders’ Summit in Boston, and we are joined by Julien Alexandre, global head of research, and Andrew Cameron, adaptive automation solutions at MarketAxess. Julien, Andrew, welcome.
Julien and Andrew – Thank you, Joe. Thanks for having us.
Josephine Gallagher – So, we’re hearing a lot about how AI is being used internally for the building of technologies and doing so at a fraction of the same time as they used to. So, could you tell us how your research and development team is using AI for fixed-income builds?
Julien Alexandre – Yeah, I mean the same as AI is changing how software engineering works. It’s changing even more, I believe, how core research is being done these days. As we are now able to produce an experiment, you know, in a matter of hours that used to take weeks before to cut them up. So it allows us to, because you know, we do research, so oftentimes we have ideas that don’t pan out and don’t yield anything; it allows us to cut the cost of, like, you know, having an idea that doesn’t work out in, you know, a fraction of the cost. So that has been, you know, a super exciting evolution for us. Things that we wouldn’t be able to do in the past, we’re able to do them now and test them out, and if they work, have something in production relatively rapidly
Josephine Gallagher – Interesting, its that time from idea generation to build has just shortens exponentially.
Julien Alexandre – Exactly.
Josephine Gallagher – Really interesting. So one of the things is that you know buy sides are very different- you know, different sizes of buy sides, different types of buy sides from long only asset managers as well as hedge funds, so how does that factor into your AI builds?
Julien Alexandre – I mean, the first is cutting the cost to a fraction of what it was, is helpful, because allows us to do things that are a lot more customized, because we can build them a lot faster. You know, as well as the fact that those buy sides, like you mentioned, firms, they have very different needs, very different objectives, so to be able to help them out the best we can with the data that we have access to and the knowledge that we have in terms of market structure and data set is to be able to exactly customize the output to their needs. Example, pricing can be dependent on a counterparty, so being able to provide a price product that is counterparty specific is really helpful for our clients.
Josephine Gallagher – Interesting, Andrew, now turning to yourself, so you obviously work on the trade execution side. So tell us, how are these kinds of research that Julien and his team are doing around AI actually being applied to trade execution?
Andrew Cameron – Well, the interesting is we probably have a lot of clients that are using AI without actually realizing that they’re using AI in the form of the CP+ model that Julian’s team has actually built, maintained, and grown over many years. So we actually offer pricing solutions whether you are a liquidity taker, liquidity maker, risk controls – all of those things are kind of customizable or curated based off what that client’s objective might be, right. So different clients, as Julien mentioned, they have different ways of trading, or different theses. We can basically now tailor a actual execution strategy to a particular outcome they’re looking to achieve.
Josephine Gallagher – Okay. Interesting. And now, how do you see the likes of AI applications and these kinds of technologies actually evolving over the next six to 12 months?
Andrew Cameron – From my perspective, it’s a natural progression of where things already go in the market. We already see clients becoming more increasingly interested in terms of how to leverage these concepts, but also how to apply them real time. It’s one thing to have pre-trade analysis, post-trade analysis, but what are you doing at that point of trade, or as that trade is commencing throughout the day. That allows you to kind of adjust based off of what went on the market, or based off of a change in opinion that you may have. So, the big thing is, how do you actually deploying those things in a real-time setting that is again flexible, controllable to a particular client’s needs?
Josephine Gallagher – Okay, understood. And just finally, this is something I’ve definitely been hearing a lot at FILS recently, and it’s the comfort levels with the use of AI and automation across trading functions. So, are you hearing the same, like there has to be more, more done in order to make people more comfortable with the use of these kinds of technologies?
Julien Alexandre – Absolutely. I mean, next 6,12 to 18 months is going to be about the ecosystem of data models and workflows, and the workflow does incorporate the confidence, the verifiability into it to make sure that everybody is super comfortable with using AI. But there are methods that exist around that, including having different types of agents, one to generate the answer, one to verify the answer that are actually being used and being well appreciated. I think the other part that people want to see is the applicability, how to insert into their workflow, how to make it work with their data. So the model providers did an amazing job of of bringing the models to the market and making those models extremely useful, but now integrating that into an ecosystem with clients’ data, with, in their workflow as well is complex, but certainly worth the effort, because that’s going to be yielding a lot of benefits.
Andrew Cameron – I think it’s just general early innings, a lot of how these models are being produced in production, right. We pride ourselves in putting a lot of back testing in to actually ensure these things are producing output that we expect, that a trader or client can anticipate. You want these things to be relatively predictable as, like, an end user of these products. But I think a lot of folks are just really trying to figure out from a diligence perspective, how does this work, right? If I’m ultimately the one that has to count or speak to how the model behaves, they really just need to be kind of armed with that level of information. But ultimately, our main task is to kind of help replicate what a manual decision tree might look like, right? And so now clients can use us on our models to say, okay, what, what should my price be for this particular size, this particular liquidity, or side? We can now curate that for an individual client, which is really exciting. A lot of folks that are having to do that, that type of work manually on a day in, day out basis.
Josephine Gallagher – Really, really interesting. I’d like to thank Andrew and Julien for their insight and, of course, you for watching. This has been Trader TV at FILS.
