Overview
In this latest episode of Ahead of the Curve, our host, Scott Sobolewski, chats with James Mac Hale – Head of Market Data in Post Trade Solutions about the monumental task of creating a centralised market data library of over fifty thousand different types of daily curve inputs as part of our hosted service to support firms in their risk and pricing calculations.
The podcast provides tips and advice for firms that are using or thinking of using the Open Source Risk Engine (ORE) and highlights the importance of putting in the time and effort into planning the inputs into your risk engine to reap automation benefits down the line.
Listen to the podcast
Hello and welcome to Ahead of the Curve, the podcast from Acadia,where we dig deep into the world of derivatives and margin,and all things risk management. My name is Scott Sobolewski,co-head of Acadia's Quant Services division.Today, I'm joined by James Mac Hale, head of Acadia's Market Dataand Risk Services operations. In today's podcast, we wanted to highlightfor the listeners how Acadia has solved the market data questionin its centralized risk services that we offer to clientsfor daily risk and margin calculations. Previous listeners to Ahead of the Curvemay have some previous familiarity with the Open source Risk Engine,or ORE, and our users or interested parties may realizethat the two big inputs into ORE are portfolio data and market data.While we highlight the history of Acadia's market data servicehere in today's podcast, and some of the relevant topicsand challenges that we've solved as Acadia.I'm hopeful that listeners can take away applicationsfor improving or solving similar questions in their local installation of ORE,or their local management, or improvement of their own risk and pricing libraries.James, welcome. -Thanks, Scott.I was hoping maybe we could start with some of the early daysof the uncleared margin rules. Take us back to 2015/2016,and when we first started hearing client interestaround hosted risk and pricing calculations for compliancewith the new uncleared margin rules at the time.Sure. It's been a bit of a journey to get here.Back in 2015, 2016, the company had only been aroundabout six, seven years at that stage. ORE was still really new,but we had a lot of ambition. When ORE came on the horizon,we thought that it was an opportunity for us as a company to grow,as you know, you were there. However, there were a lot of challengesalong the way. When we first tried to get into SIMM,we were trying to restrict ourselves to interest rates and FX,and we were fairly comfortable there, but we immediatelystarted finding things to improve. We realized that we had a big taskahead of us in getting alternative data to the vendorsthat didn't want to work with us. When we finally brought somethingto the market, we started talking to potential clientsand found that the scope wasn't wide enough.It was immediately obvious it was going to be a big task.Around phase four is when it really started heating up, I think.We landed a couple of clients. We'd grown from interest rates and FXto interest rates, FX, and credit with a couple of hacksfor some other asset classes, but nothing formal.We identified Refinitiv as a strong partner for data.I think it was called Refinitiv at that stage.We'd also found limitations in what we could get.Well, what would you like to know? -Well, I would say the companyyou're referring to, the originators or the authors of ORE,originally the predecessor company that you and I were both partof, Quaternion Risk Management. At the time, we were partners with Acadiawell before the LSAG acquisition. I guess this is two acquisitions ago.Two acquisitions ago. -The bulk of the code is to priceand calculate risk sensitivities for those of you who may not be familiar.The SIMM model is the industry standard modelfor the non-clear initial margin calculations globally.It requires risk sensitivities as inputs, deltas, and VEGAs.From a quaternion and Acadia perspective, that code had existed for many yearsfor the majority of the asset classes that were in scope for UMR,the primary investment that Acadia was makingwas to transition this open source code baseinto a production level engine that was to be pairedwith market data and client portfolio data for daily calculations.The consumption of the portfolio data was one challenge,making sure that clients got that data to Acadiain the right formats. However, I would say from my perspective,the much bigger challenge and the one that I'm not sure if you volunteeredor were seconded to take on at the time. -Tempted, I think.However, you grew a dedicated team to manage the sourcingand the automation that it took to feed that market datainto a risk engine to make these on-demand daily calculationsfor hundreds of thousands of trades, and even by now, up to 400,000 tradesa day. By now, millions have been processed dailysince the end of the phase six UMR in the fallof 2022. Maybe from some of those early days,let's say post quaternion acquisition by Acadiaafter the identification of the majority of the clientsthat required us to make those daily calculations,what were some of the asset classes or the marketdata types that were particularly difficult to sourceor feed into ORE for those daily calculations?Sure. Everything has a problem.That's one thing I've learned over the process.You mentioned IRFX. That's where the majorityof the non-clear trading activity, I'd say 70 or 80 percent.It is. -Of trades are in oneof those two asset classes, but it's always the last.Exactly. It's the 80-20 rule. Even if the majorityof the clients' portfolio is vanilla swaps, vanilla FX options.Some of those swaps might be in a weird currencyreferencing a weird interest rate that we hadn't encountered before.One of the silly ones that I still don't understandhow it hadn't come up was the euro yen TIBOR.We'd been supporting Japanese yen swaps forever sincewe built ORE in the first place. Somehow, that there weretwo different types of TIBOR had never come up.When we really kicked into phase five, we started uncoveringa lot of these things. USD and Euro, we were fine.JPY extra rates. We had some clients coming in with FXforwards referencing the Egyptian pound. We have avast array of Eastern currencies. It says the Singapore dollarwas much more important than we'd expected.Thailand had at the time three different familiesof interest rates. Even in interest rates, we had some trouble.In FX, we were facing client portfoliosthat were usually OTCs, maybe some barriers.From a quantitative point of view, not that hard,but in terms of sourcing market data, a lot more difficult.When you start talking about cross pairs, the data just isn't liquid enough.You can't get the quotes on the market, or at least we couldn'tas a non-participant in the market. We had to develop modelsto proxy cross-pair options,and we did that. Outside of interest rates and effectswhen we started getting into the credit space,we found that the market for credit is huge for single names.Nowadays, we support up to 10,000 different issuersor different issuer curves for single-name swaps.However, at the time we were starting, we were only really expectingto see the CDx investment grade or high yield.Maybe an eye tracks here or there. When we started digging into these phasefive portfolios, that this was a blocker. -That's really where phasesfive and six are. Those types of clients tended to be concentrated primarilyon either large buy-side institutions or smaller regional banks globally.It was, in particular, I'd say, the buy sidethat was driving a lot of the more exotic marketdata requirements speculating in some illiquid geographies.Let's say, perhaps, that has necessitated the build outand even some of the investments that your team has madeinto the automation of sourcing or dynamic configurations,maybe is another way to put it internally. -Exactly. That's where I was going.When we got into the credit market, we started looking at equities.Anybody who's tried to set up ORE locally has encountered problems with scale?Right. How do you set up 20,000 configurations for stocks?How do you support 10,000 configurations for CDx?How do you ensure that your market is built for all of these thingsand that the performance doesn't suffer? That was one of the big problemsthat we had to solve in scaling ORE for our SIMM offering,and we did. -That's, I'd say a big differentiatorbetween the open source version of ORE and let's say,the commercial version that Acadia deploys through its hosted risk services,the existing configurations or sample configurationsthat you'll find on the open source code baseare static, whereas the ones that we use internally, they necessitatea greater flexibility in responding to what clientsdecide to submit each day into our service.We don't know what advance, what those trades are going to be,or which currencies they are going to reference,or which stocks they'll decide to. -Or more importantly,which they're not going to reference. -Right. Maybeyou can just dig a bit deeper into it, and maybeblend this against maybe what you've heard from some clients who have come to usfrom maybe some competing vendors on a hosted basis.What makes Acadia unique from a market data coverageor market data automation standpoint? Are there some areaslike index decomposition? I know the LIBOR transitionis mostly in the rearview mirror by now, but that was a really big projectfor your team at the time to ensure adequate fallbacks.I'm not going to say it was fun. -Probably not completelysolved at this point. We still see legacy CSAsthat reference stale rates and maybe have not been fully negotiatedor transitioned over to the alternative rates.Maybe between those two or three. -A few busy bees.I mean, where would you say Arcadia becomes most differentiated,there from the market data service or the market data coverage?I would say the breadth and depth. The breadth in terms of asset classes.I don't think there's any other vendor out there that's able to serviceinterest rates, including inflation, FX, equities,commodities, fixed income derivatives, and a little bitof asset-backed securities. We cover everything.That was one of the great things about having to build this outwas that we had to be able to support everything.We couldn't just pick and choose what we couldn't because,as we said earlier, the 80-20 rule. If we're able to supportall of our clients' interest rate swaps, all of their effects,and we don't support the credit default swaps,we can't support that client. If we can support their equity,we can support their credit, and if we can't support their commodities,we can't support that client. Because we're dealing primarilywith the buy side, we did see this huge variety.We had to overcome all of these tasksin terms of depth, and the scalabilitythat we achieved, I think, is unmatched among vendors for SIMM.I say a couple of other analytics, too. The way that we built the service outwas so that it would dynamically just pull in what you needed.If you were presenting something to our service for the first time,something that we hadn't seen before. We built automationthat it would get automatically added that the curves that were neededwould be set up, that any new market data that was neededwould get will get pulled in that if there was noif there wasn't market data for something. For example,if you have some esoteric stock out of Finlandthat you would get a proxy service at least so that we could calculate SIMM.It wouldn't interrupt your daily workflows.Wouldn't interrupt your workflows, and it wouldn't needto be manually inspected or configured by somebody from my team.Rather than in some of the very early days of the UMRgo live in 2021, 2022, before some of this automationthat we provide now had fully taken form. If we saw that the finished stockfor the first time, maybe for that particular netting set,and that calculation would have failed. It maybe would have requiredsome manual coverage by either someone on the operations teamor even a client to step in and change the waythat they were submitting data into our service.However, since then, I know for a fact that the number of failed calculationsand the resulting burden on our own client operations teamhave been significantly lowered. I'd say for the last two years now,we're starting to fully realize the benefits.It's remarkable. Honestly, it's remarkable.There was a state before the service launched whereif a client was testing in UAT, and they were it was a novel portfoliothat there would be an angry email, there would be a failed calculation,and it just doesn't happen anymore. Anything that goes wrongwith the service these days? There are obviously some caseswhere something might go wrong, but the vast majority of problemsthat we encounter in the service are trade issueswhere a client is misrepresenting what they've got,and they're unhappy with the result that comes out.I'm pretty proud of what we do. -I think that's a nice lessonfor potential clients or users out there. The investmentis to be made up front in the way that you plan and think about the inputsinto a risk library or production risk engine.With regards to portfolio data and market data,and the various ETL layers or adapters that changeand transform that data into the right formats,and making sure that the teams supporting the daily operationof those either dynamic or batch services. The more time you can put into planningthat up front, realize the significant man years' worth of savings,one or two years down the line if you can get that automation right.With regards to the open source risk engine,definitely get in touch with Arcadia. If you need to get started,we're very happy to share some of our production configurationsthat we use in our hosted service, which may not be made availableon the open source website, but especially for commercial clients of Acadia.Very happy to get them started, and significant time savings,or not able to share the market data directly.No, I wish a big differentiator, but we are able to providesome nice advice based on the five or six yearsof learning that we've harvested here in the Acadia Quant teams.Moving forward, James, you know what? What's ahead of usfrom either a market data or operations standpoint?How are we looking to support clients from a market dataor automation standpoint moving forward into 2025 and beyond?We're at the stage where we've got the coverage that we need.We've scaled everything up that we've needed along the way.It's very rare that a client will come in with an ask that we haven't encounteredor at least that we haven't encountered something very similar.At this stage, we're at the tweaking, right?We're at the point where everything that we've got,we're trying to improve it a little bit. Make it is make it as goodas we can make it, and we're an end-of-day service.We have the flexibility to maybe take a couple of hoursfor automation to run to optimize our curves,optimize our surfaces, and make sure there's no arbitrage hanging around.Nobody likes to see arbitrage in their role,making sure all of our curves are smooth, even commodity forward curves are smooth.Equity forward curves or smooth, maybe bootstrap,and equity options surface. For the moment, that's what we're up to.We're trying to make it as good as it can be.We're working on some machine learning models,in particular in the interest rate space, lifting some inspiration from Bianchi,and so-called to usesome modern techniques that maybe might not have been availableto us at the outset. -That's great.We continue to hear and feel for our clientswho want to bring the same benefits of Acadia's hosted risk servicesto their own local risk and pricing applications throughthe open source risk engine at a very cost-efficient price, i.e., free.It is open source, but we continue to hear that getting the portfolio datainto the XML format that ORE expects is, like I said at the outset, one thing.The much bigger challenge continues to be the sourcingand formatting of market data into ORE, especially for use casesthat extend beyond the uncleared margin rules.For every other type of analytics that already does very well,Acadia does not host a daily calculation service for thingslike potential future exposure xVAs value at risk.What does the future hold there for Acadia,trying to make ORE users' lives easier with regardsto either getting familiar with or testing calculations,and ORE without having to go through the months of effortthat it takes to plug ORE into a local market database?We've become known as the SIMM vendor, as the SIMM calculating company.However, a lot of people don't know that ORE did start offas an xVA calculator. The source code was initially builtfor a bank that wanted to mitigate counterparty credit riskfrom a dynamic initial margin point of view.All of that architecture is still there. We feel we're at a pointnow where we can have this great calculatorthat can do a variety of calculations. xVA, maybe market risk calcs,SIMM calcs, SACCR calcs, PFA especially, are coming up moreand more at the moment? Not entirely sure why,but I can maybe guess. -Some of the guidance that came outat the beginning of the year was on counterparty credit risk.I'd say, along with Basel three, endgame in the US and Europe, and the UK.Runway risk as well. -Yes, there is definitely nowthat the uncleared margin rule problem has been mostly solved by the industry,it seems like there will be a renewed focuson counterparty credit risk in particular. -However, I think we'rein a great position with this calculator and the market data servicethat we've built, with the breadth and depthto offer calculations for other analytics. One of the thingsthat we're hoping to develop over the coming monthsand maybe release into the new year is a hosted notebook servicehosted calculation service using Jupyter Notebooks.It's something I'm very excited about. I've been an evangelistfor using Python for ORE calcs for many years.The idea there is that a client will be ableto log in to Arcadia's website, to Arcadia Hub,see a pop-up Jupyter notebook that is tied in to our calculating serversand tied into our market data and tied into the configurationsthat we're using in production for some calculations.With all of these pieces, if they're a SIMM client,they already draw in their portfolio and use that portfolioand all the rest to perform PFE calcs, to perform generic market risk calcs,to perform SACC calcs, just to really take advantageof all of the power that what we've got can offer.Extending the full functionality of ORE. -The full functionality.The two or three percent is utilized to produce market risk sensitivitiesor SIMM calculations. Extending that full functionality of OREpaired with the existing market data service that we have centralized,and the dynamic configurations that we mentioned earlierthat were built out to support a daily hosted service.With the guarantees that we offer for the SIMM calculation servicewith the market data is held to a good standardis reconciled every day against counterparty calculations.It's strictly back to the beginning of 2008.We often hear from clients that their market data historyneeds to go back at least that far. We know from a regulatory standpointin the SIMM world, it does as well, especially for historical VaR calculationsor back testing. -Yes. We have to access the SIMM model.That is the service that we offer. We have a few clients.I don't know if I should say how many backtesting their same with usat every quarter. -Over 30.Over 30 backtesting same with us every quarter.For those back tests, we need data be it real or proxied.From today, all the way back to 1st January, 2008.That's been one of the big challenges that we've had to overcome over the years.However, now hopefully a broader client basethat extends beyond the world can also utilizethat centralized market data service and market data historyfor any sort of risk or pricing calculation for pretty much any tradedinstrument or structured product that already supports the hundredsby now across all asset classes. I think that's all offeredwithin the same Acadia UI that our clients are used to.Alongside all the client support and the guaranteesthat Acadia has offered and performed well for the marketfor the last decade, and now as part of the LSEG post-trade solutions family.Even I'd say greater synergies into 2025 and beyond, potentially.I think we're coming up on time. James, is there anything elsethat you'd like to share? -I'd make one joke.If anybody wants to try and build a market data servicefor daily calcs for 400,000 trades. Good luck. If anybody everwants to do that at the same time are building a backtesting servicethat goes back to 2008. Please don't talk to me ever again.All right. -Once is enough.All right. With that, that concludes our time for today.Again, thank you for joining Acadia's, Ahead of the Curve.We hope to see you in a future episode. Thank you again.Thank you.