AI can enable trade flow automation through triage of high- and low-touch trades

Published on 1 October 2020

Triaging between bonds by liquidity profile can be more easily achieved using machine-learning algorithms. The level of automation this provides to the sell-side desk dramatically increases responsiveness to request-for-quote orders, while buy-side desks can reduce workload allowing traders to focus on more complex trades.

Vuk Magdalenic, CEO of Overbond, explains how his firm is offering these tools via cloud-based visualisation, API integration and integration on the Refinitiv Eikon terminal to support execution via Tradeweb.

Dan Barnes Welcome to Trader TV, I’m Dan Barnes. Joining me today is Vuc Magdalenic, CEO of Overbond, an innovator in the application of algorithms and A.I. in the fixed income markets. We’re going to be talking about how the application of technology is overcoming some of the challenges buy-side traders face today. Vuc, welcome to Trader TV.

Vuc Magdalenic Thank you for having me.

Dan Barnes So tell us, what do you think are the most pressing challenges that you see on fixed income trading desks today?

Vuc Magdalenic I think it is about having a true view of price and liquidity. That doesn’t necessarily only mean liquidity that I see on one venue or in my own trading floor operations, but liquidity across venues or across multiple data sources. And there very quickly, we run into a situation where it becomes apparent that aggregating data across different sources becomes really difficult or at least difficult to do in real time.

Dan Barnes Is there a need for traders to triage between different types of trades as they try to find prices in the liquidity on the desk?

Vuc Magdalenic Absolutely. I think the notion of automating the entire trading workflow and the whole volume on the desk is unrealistic given all of the market conditions. Automating a part of the trading workflow where the liquidity profile of a particular ISIN is stable, the market supports and triaging trades that qualify and don’t qualify, I think is the innovation that would bring the desk the most benefit.

Dan Barnes And that requires having the right data in the first place.

Vuc Magdalenic Absolutely. And that’s where novel aggregation that can happen in real time has been the challenge, both from just the sheer volume of data in the last couple of years, but also technologically. Reconciling multiple reference data streams in real time is pretty challenging, and that’s where the application of A.I. and server-less cloud technologies can enable the trade flow automation we’re talking about.

Dan Barnes How are you supporting clients and overcoming these challenges?

Vuc Magdalenic A lot of talk in the industry is around solving for OTC nature or disparate data nature of the fixed income market through the regulatory kind of post trade repositories. We realized that there’s some way until that is able to be consumed at the desk. So what we focused on in terms of the approach is aggregating data client-side with pre-structured algorithms that can map transactions data to the reference master, and enrich it with sellerman-layer data and corporate fundamentals. So from the data aggregation layer, we’re now going into the modeling layer where the risk algorithms that monitor liquidity or liquidity stability per ISIN, do the trade tiering, and then for tiers that are not as stable or don’t have as much trading liquidity, we have models that do deeper AI benchmarking of similar bonds and similar issuers and curve-fitting to optimize for the best executable price.

Dan Barnes What do you expect the impact of that to be on buy-side and sell-side trading workflows?

Vuc Magdalenic Through all of our BAC tests, there is about 30 percent of the entire flow on a regular kind of diversified corporate coverage desk that can be automated, as well as about 30 percent of situations when there is unstable liquidity profile per ISIN. Now the model of this nature can actually price with 75% precision.

Dan Barnes Which instruments and markets is that covering?

Vuc Magdalenic Investment grade corporates and obviously benchmark rates. From the ratings perspective, there are ISIN’s and issuers in kind of cuspy high-yield category that models can support. But obviously, as we go into a deeper, high-yield data availability and the number of bonds outstanding for issuers drops off radically.

Dan Barnes What sort of performance changes might you expect in trading on RFQ platforms, for example?

Vuc Magdalenic Let’s say a sell-side trading desk that has two traders and on average fields a volume of five hundred RFQs per day; from a manual standpoint they can respond to about 100 a day. A model like this can support doubling or tripling that response volume, adhering to like a two minute or sub two minute response time and preserving the hit ratio. Now, obviously, it’s not just about the speed, it’s about precision. So we do monitor direct PNL impacts through two factors. First, one is obviously a fair value price that the model is able to minimize to kind of historical levels. If the trader, let’s say, sets a five percent precision acceptance threshold, as well as the margin optimization and I think margin optimization level or sell-side desks is very important, given that each ISIN, each counterparty in each market condition would merit differently.

Dan Barnes So how will Overbond support clients in accessing this and the use of it going forward?

Vuc Magdalenic We’ve been focusing on various types of integrations. So the first and foremost is a fully cloud-enabled visualization that is very easy to access. The second is API deployment, where data can feed OMS or any internal system. And the third one, which we’re really excited about, is the side-by-side integration with Refinitive and their Eikon terminal, where we can have a 1-click-review of the model price and then obviously through Eikon terminal, 1-click trade on Tradeweb.

Dan Barnes That sounds fantastic. I’m sure you’re going to have great success with this.

Vuc Magdalenic Thanks so much, Dan.

Dan Barnes I’d like to thank Vuc for his insights and, of course, you for watching. To catch up on our other shows or to subscribe to our newsletter, go to TraderTV.NET or ETFTV.NET.