Emily Eimer, e-FX analytics, and John Estrada, global head of e-FX at Credit Suisse, talk about the firm’s new trade analytics system.
Profit & Loss: Talk us through the Real-Time Automated Data Analysis Robot (Radar) you’ve recently launched. What is it and how does it work?
Emily Eimer: Radar is our next generation trade analytics tool. It takes a range of data and uses algos to tell us when to make pricing changes with the objective of achieving optimal client pricing.
The beauty of Radar is that it benefits clients by providing more consistent and tighter pricing whilst simultaneously making us more efficient and effective.
John Estrada: We describe this as taking a “cognitive approach” to client pricing. By automating the process the clients benefit in the near-term by getting more consistent pricing and we benefit in the long-term because it frees up time for a number of our staff to focus on the e-FX business more broadly.
P&L: How does this differ from your previous trade analytics systems?
EE: Version 1.0 was way back when we first started in e-FX and we had a very simple data set. It was a case of watching trades coming in and adjusting the pricing accordingly, which could lead to knee- jerk reactions.
V2.0 involved using tools to run data in a much more sophisticated way, but it was still a pull model where we would have to go through our systems to find out information.
V3.0 – where Radar is headed – is when all this is done for us by a machine. We’re an electronic business: our pricing is done by a machine, our hedging is done by a machine, so why are we not running our trade analytics by a machine?
P&L: How often does Radar run?
EE: At the moment Radar runs on a daily basis but it makes suggestions using historical data, so it’s not just looking at the most recent trades.
P&L: And how customisable is the pricing suggested by Radar?
EE: There is a set of high level rules in the system, which works well for a large number of clients, but we can also do as much customisation as needed in order to provide the most appropriate pricing based on clients’ trading patterns.
So in terms of high level rules for example, Radar can track a client’s average daily volumes with us and then, without us having to do anything ourselves, Radar will know if that client starts doing less business with us and prompt us to improve their pricing.
As an example of where we might customise the pricing, if we go into a meeting and a client tells us that they’re doing a lot more in specific pairs, then we can tell Radar to focus on those. This is really helpful because it means that this information can be entered into Radar, which remembers it and then advises us to make the necessary adjustments.
P&L: Does Radar also serve a compliance and audit function in terms of justifying client pricing?
EE: Yes, because this is a very data-driven approach it provides a great audit trail because we can use this data to explain the rationale for changes.
JE: Audit and compliance may be becoming more onerous and expensive, but this is one of those positive outcomes where we’ve built a tool that helps our client franchise and also has the added benefit of being open and transparent which should make our audit and compliance team happy.
P&L: Are there plans to roll Radar out across other asset classes?
EE: We built Radar with a very flexible framework and so in theory it is scalable and where it makes sense we can roll it out into other asset classes. But there’s still so much within e-FX that we want to do with Radar, for example, we’re working on an optimiser which will look at past performance and past pricing streams and see what other options were available and improving on that. This is getting into machine learning.
JE: Building Radar and setting up the data was a serious piece of work and once you’ve invested so much in something you want to try and fully utilise that investment.
P&L: Does machine learning represent trade analytics 4.0 then?
JE: Right now we’re considering if we want to continue on a subject matter expert approach, which involves supervised learning, or a clustering approach, which is more independent. This is about getting the right outcome for the client and so we want to make sure that the black box doesn’t get too smart too quickly. We don’t want to try to run before we can walk. As a result, we’re leaning towards a subject matter expert approach right now, but if we get good results then we’d be willing to follow another path. It is too early to be making any definitive decisions regarding this, but watch this space…