 Hello, my I'm Eric Rosenbaum here. I am the chief technologist in the global fsi practice for redhead We're gonna talk today about an open-source approach to modernizing market risk management And I'm joined on stage here with Marius Marius what you choose yourself. Thanks, Eric My name is Marius Bogovici. I'm a chief architect in the North America FSI team at red hat I've been a long time open source first user then contributor and now kind of part of the more like helping our customers modernize their applications and First the reason why we're here today, right the kind of the very reason The reason why we're here today and having this conversation is because of tremendous evolution that opens our software hat over the years and a moving from being a tool to democratize technology to an enabler of innovation and That innovation happens around Like on two coordinates one really getting better and more interesting technologies I think about something like Kubernetes for example, we talk about a lot of other things here But when you think about something like that It came from the open-source community, right and when it did it didn't just do that it enabled new types of business solutions that solve business problems and started creating business value and You know while a lot of the technologies that you see on this map are things that we're going to talk today and Show how we're doing together as part of this architecture You know the real story here is you know, how we can help how we help Creating new innovative business solutions for managing credit market and credit risk, you know with open source So I'm going to hand it over to Eric to talk a little bit about the business domain So before we get started a little show of hands How many people are involved currently in risk management helping either develop develop models supporting the risk management team? Some okay So let's spend a few minutes then I'll just give you a little background so everyone's on the same page with regard to risk management So risk management is the practice of understanding the risk to an organization So that you can then mitigate that risk and not blow up the firm very simple Okay, more detailed explanation on the screen So that being said there's multiple types of risk that a financial institution will come across Today we'll talk about market risk Similarly, there's credit risk when the model we're showing here can also be applied to credit risk We won't talk about other types of risks such as reg risk operational risk strategy risk and so on and so forth But those exist under the umbrella of risk management So if you come across the term risk management a bank, it is a fairly large term. We're gonna focus here really on market risk So has a bank or any FSI understand the risk to them for market risk One approach is something called valued risk var people refer to it as and that's an understanding of based on correlations based on history What is a likely move in the markets and how would that negatively or positively affect you? And it's expressed as something along the lines of in a one-day var in 99% of the cases my risk my loss will not exceed X and And that X is then defined possibly by a risk committee the board of directors It might be a hundred million dollars for a large global bank It may be five million dollars for a smaller hedge fund But that is within 99% of the time in one day the most unlikely to lose is X It doesn't tell you the max loss It's In a probabilistic way what I can likely lose and that's all done in a math math-based way There's Monte Carlo simulations where you can run a thousand five thousand a hundred thousand different simulations to test different things And then you look at the correlations Excuse me the confidence intervals of that risk and see what you're likely to lose Some more interesting approaches are agent-based modeling Where you can take different actors and what what they do would impact others and How that works climate risk is a good way to do agent-based modeling and there's firms that focus on that so to get here today what Marius and I have done is we've talked to a number of our customers a number of our partners and Really surveyed and said what are the challenges you have today? And what we heard is really key three key themes one being an increasing complexity of risk models Whether it's things like FRTB Whether it's things like climate risk the models are getting more complex because life is complex computers are getting faster We want to model what happens in the real world There's a need to bring in other types of data not just market data and not just prices But rather maybe it's a weather model Maybe it's sentiment and other types of things you want to bring into that collectively referred to as alt data We want to be more agile You know as a firm we want to release things more quickly We don't want to spend six months doing something We want to bring things as quickly as possible from the minds of our developers and our quants into production in a safe and efficient way And banks being banks we won't be efficient about this Where do we run this do we run it on-prem do we run in the cloud? Do we run in a serverless? Do we run it in a distributed way? Do we run in VMs? Do we run in Kubernetes? We don't know but we want to be agile about this and we want to be able to be efficient about where we run this So that really that's what's driving. That's what we heard from people So based on that, you know, what we've done is we've gone back and you Envisioned from the ground up what a new architecture would look like and that is event driven in real time It's no longer acceptable to run your risk once a day Markets move during the day positions are put on during the day the idea that okay My risk was you know so-and-so if you had 5 p.m. At London clothes yesterday And now I have no idea where I am. That's not acceptable It's got to be real time and it's got to be based on more than just market data Again, this all data comes in that's gonna you know make me better understand what my risk is. I want to incorporate that We want to incorporate lines of business applications What if a portfolio manager for example wants to say I'm working with the clients I'm working with my desk if I change this and this what does my risk look like? Is it better or is it worse? We want to do that and obviously in real time We want to be able to do these ad hoc things again not wait for the end of day We want to be agile about this. How do we move models? For the brains of our quants into production as quickly as possible You know we heard from some tier one banks that it takes them six to nine months for that to happen That's really unacceptable, you know markets are moving you look at the oil oil and oil sector things are moving constantly There's you know between Ukraine and the heatwave in Texas and so on and so forth energy markets move You want to be able to build models that? Incorporate what's happening understand your risk and move on to the next problem and obviously you want to be efficient about this reduce the total cost of ownership and Avoid vendor lock-in and the voting vendor lock-in at miss. I don't need to tell everybody here That's a key part of the open source, you know ethos the ability for us to understand what we're doing to have the code there for us to iterate on to improve upon to contribute back So I'm gonna hand it off to Marius next sure. Thanks Eric so kind of coming back to the solution and Coming back to the conversation that we had in the beginning what technologies what advancements what new? characteristics do these systems can these systems incorporate to actually start solving the problems that Eric was talking about right and We have we have new technologies. We have new options Things like you the ability to bring, you know real-time Streaming to be to build event-driven architectures, you know that kind of that collect that connect the different parts Of the system because really these really are connect our complex systems Not only internally but also with with a line of business applications to trigger for example risk calculations in real-time Whenever we need it right on or to You know trigger them for example in response to certain changes in the market I'm gonna talk a bit later about that We have different options to integrate data from a variety of sources using, you know things like APIs for example For for bringing data from from you know from the line of business from external sources from from wherever data is into the application And of course, we have the we have technologies that enable us to be scalable Like the adoption of hybrid cloud architectures where clothes can run on premise on you know in the cloud Can run on the edge, but also, you know have the opportunity to Combine for example existing investment in risk calculation processes and making them more efficient moving them to a different environment With new components are using things like, you know containers virtual machines You know serverless technologies Technology is one thing the other one is concerns the processes and you know the quality of improvements that we make into this and You know applying automation for example You know, how do you solve the problem of moving a model in less than six eggs months, right? Through automation. It's the kind of the software industry has developed practices like DevSecOps For example that are perfectly applicable to to risk model development, right? So incorporating those lessons into the build deployment workflow helps us reduce the time to value also You know starting to incorporate new types of you know new options like Accelerated hardware things like, you know GPUs for example, and when you do that, how do you? figure out how you run applications Like how do you run these hybrid solutions that combine CPU and GPU components, and how do you run them transparently? so Having this like thinking about how to how to run these things is is critically important and Finally, you know, how do you incorporate things like artificial intelligence? How do you how do you add more? You know how you how do you add more intelligence in the way you consume the input data the way you handle this You know the way you manage your processes the way you you you parse your results So all these things are part of this kind of General solution that we envision again. This is not some sort of a grand plan. It's not of a drop-in replacement This is the distillation as as Eric has mentioned of the work that we did With our clients with our partners to understand. What is the journey that? You know that Institutions take to modernize their market and credit risk and can I add one one point just do perfectly clear Red Hat doesn't have a risk management product exactly on a skew that we're selling nothing like that It's all about how do we incorporate open-source technology to solve problems for our clients in the risk management space That's a very that's a very important point And I think we are very emphatic about that. Yeah And it is I think one of the key Things there and where it ties to the open-source ethos is that this is an open architecture It is designed We're going to talk about different components and solutions that fit in there and give examples of technologies But realistically it is designed to to take an a financial institution on a Transformational journey like take what they have right now add new things incorporate what exists You know be able to choose From the different available technologies what works best what was what works best for for them, right? We're gonna because this journey like one of the things that we found out when we talked to When we talked to our clients was Nobody's doing this transformation wholesale Like there is no way to kind of take this and build it and you know take this kind of moonshot project of doing everything At once it's more like incremental Therefore we have a number of themes that we're gonna walk you through which capture, you know the different Portions of the journey or the different paths that organizations take to actually to modernize their market and credit risk And we'll discuss the kind of the appropriate open-source technologies And that's a reflection on the conversations we had with different clients We broke into themes based on their challenges And as Mary steps through it, you'll see that there's specific Fixes, you know solutions to those challenges that we thematically heard right before that before we go that I just want to kind of give you a bit of a lay of the land of this of this solution, right? What you see at the top is your typical risk calculation Process data comes in from different sources is brought into a compute system. It is stored for consumption And then, you know, it is also kind of stored for its archive for for audit and you know for for backtesting but To solve all these other problems of real-time access of of how do you do better around? Processing these results we have an intelligent automation Intelligent orchestration layer that kind of coordinates these things and add some some business process intelligence around it To make these things work. We have an event-driven platform that moves, you know, internal and external events business events and Application relevant events through it and to top it all we have artificial intelligence that adds as I said intelligence in processing data input data orchestrating the processes and of course handling you know and connecting with With the the external applications in line of business kind of giving the right advice to the business stakeholders So let's start going through the things. I think you got it The first one I think and it's common probably the most common one that comes in this conversation is hypercloud and why because a risk calculation is by definition a compute intensive process like Monte Carlo simulation should say though, right, so very You know a lot of the kind of a lot of the existing investment is a high-performance computing now in order to get better results in order to Get results faster, you know, there is a need to acquire more compute capacity and very often institutions what they do They go to the cloud they burst this workload and try to figure out, you know, how they work there When you do that you end up with, you know, different environments you did with different ways of managing So reducing the friction of running in different environments to make it easier, for example to move things from one environment to the other there are, you know components like there are technologies like containerization and Container orchestration platforms things like, you know Linux for example has been an early component for providing a uniform platform for run things Kubernetes added even more abstraction, you know, allowing to run things like Containers even virtual machines to kube-vert serverless to components like k-native and essentially kind of allowing you to pick and choose what's your best technology for running your Your risk calculation processes, right? it You know it enables for example Support for CPUs and GPUs that we were talking about earlier, right? So, you know just you know just enabling this part and enabling the kind of the I can add one thing as well What's important is that in conversations we have a number of clients who are challenged with SLAs There's so much happening in terms of volumes in the market, you know new products They're hitting up against our SLAs and they're looking to you know Do some short-term tactical things that'll give them 5% 10% more capacity Because they're constrained the data center in terms of cooling, you know in power that going to the cloud solves that problem It also reduces their costs Because now instead of running things 24 by 7 they can run them for maybe the 8 hours the 12 hours We're doing compute for the risk sure and you know the other challenge that we've seen and is related to that is Introducing new types of processes side-by-side to existing ones, right? They have a they have an existing platform for example for doing risk calculation that they've been kind of building since you know For backtesting, but how do you do new things like climate risk how you create new models? Do you run on the same platform or do you do some new? And when you do that, how do you do things side-by-side? Want to take intelligent orchestration tells an orchestration. So this is a piece that that's super interesting to make The idea is that how do we orchestrate all these pieces, you know There's a lot to happen to calculate a risk beyond just the mathematical models you have get data from Golden sources the transactional systems, which are typically on-prem How do you transform that data and normalize it your FX system have a different schema than your fixed income than your equities? How do you bring it to where your compute is? Put it into a cache Run those calculations, and then when you're done Archive the stuff your inputs or outputs and the version your models How do you react to? What happened meaning that is there a breach in your var and what do you do there? And that's an important piece because in order to mitigate the risk It's more than just knowing what the risk is But what do you do with it? And there's been a number of failures, you know over the last decades whether it's you know the crypto firms recently whether it is Firms that were hit with Archegos or long-term capital management, you know, you know probably a few decades ago What happens when you do a breach or you get near a breach? So if you're in a warning zone, for example with a client on margin Do you notify the relationship manager so he or she can talk upfront with that customer? Hey, we need to post more collateral. We need to bring down some of your positions if He or she doesn't respond to that alert and it can escalate to his or her manager and Ultimately once they're under margin You can go and clear out that position close the position or hedge the position or do what you need to do Full in a fully automated way. So by comparison, I mean 15 years ago or so. I was an FX firm We calculated our margin risk every second for every position never suffered a debit loss because we were always on top of that customer and Importantly also it helps the customer because if we're doing this in real time Then that customer is not getting a debit balance. They're not going to get a bill from us to say you always a million dollars US 10 million dollars So you're helping the customer in terms of their experience as well. So all of that should be automated All that should be with rules. So when your internal audit comes in you can say very simply here's the rules Here's what we ran here were the inputs here were the outputs here with the versions of the models We ran and it's all very transparent and all very clear And that's a I would argue a very key tenant of a modern risk management system Back to you sir. Thank you I would add here to what Eric was been saying that this is probably one of the places where Artificial intelligence can play a key role because what you've seen for example from you know from talking to our customers It's not just about Getting the data and producing a ton of results is the way you interpret them the way you handle You know a large volume of data you have like we were talking about with banks that have like 7,000 dashboards You're producing more results. What do you want? Probably not more dashboards You want someone to kind of go to and sift through that data separate signal from noise and make you understand You know what what's going on in there and that is part of the intelligent orchestration to and that Introducing that intelligence into the process is a key piece here now of course We have data like we have processes running in in the cloud and locally we have you know this intelligent orchestration process How do we how do we manage? How do we bring the data to the computation? And this is the place where you know Storage and performance or in modern storage and caching solutions play a key role caching brings essentially data close to the computation like these are very very intensive Required computations require low latency access. They do a lot of churning. They do a lot of work. You want to have the data there You want to have Essentially also you want to have data stored in your file systems in such a way that if you're moving for example the workload from You know from local to the hybrid cloud you still have the data accessible in these different places When you do that you have to do it with consideration to security, right? This is like, you know, this is very confidential data. Maybe not PII, but it still contains for example proprietary information on how the bank does its calculations, so Enabling that you know in late enabling this You know access across different clouds enabling You know a seamless integration between the kind of the file system the cash layer is is very very important And that this is an area and that and that's an important piece also We've talked about you talked before about the open-source aspect of this that really what we're looking to help banks modernize You know, so I mean I see for example our friends from Hazelcast You know that's a that's a place where place where you can plug in the cash is there I mean really it's agnostic about what do we do that's best for that customer where their investment lies How they leverage technology such as Hazelcast or an Finnish fan or something else so Yes Sure, what I would argue for the risk models you you need to understand What is it the inputs you need at all that should be in the cash? So you're probably talking about a larger cash, so we have a conversation with one One firm this week they keep 74 days for their their look back for var to do correlation So 74 maybe 180 or something like that But once you start going to disk off that cash the var models are going to be run much more slowly If you're happy, I mean after we can have a deeper conversation Happy to take that it's a great point though Like it's part of the we didn't enter in all the details, but happy to have that conversation afterward Yeah, and it depends on asset class as well, you know options for 30 years would be enormous, you know equities be less so Now as we solve the problems of Where the computation happens, you know, how we orchestrate things how we move the data where we store it Comes how do we put all these things together, you know, as I said, this is an open architecture. It's modular It's composed in multiple parts. So getting the different pieces to talk to each other Requires a decomp like requires a decomposed. I said I said it again modular architecture and event-driven Architecture are probably the best way to to connect things together event-driven architecture does not just give you the kind of The composability and you know the option to replace and evolve different parts independently also solves problems of real time Real time come to how do you trigger for example a computation in real time? When we talk about events, we don't we don't just talk about technical events or messages, right? We're talking about business events We're talking about Internal kind of compute internal events like a computation has ended. What do I do now? We're talking about events that are produced for example by the intelligent orchestration layer. I detected a breach What do I do now? We talk about events for example, synthetically produced from the alternative data Market moves in a certain way. This means for example, like there's you know, they kind of the you know the Twitter for example is is is suing Elon Musk right now. How does that what does that mean for my market and for my risk calculation process like? Interpreting and incorporating that and translating that into into triggers for the risk calculations the key piece here and This is also again a place for artificial intelligence like applying, you know technologies like sentiment analysis or you know NLP for example for parsing things like ESG reports or parsing things like you know parsing things like speech like analyzing, you know the voice patterns in certain earnings calls for example produce valuable data that is That is that is introduced in the system, right? so I Think the the last but not least we have another kind of slide here Which presents you a different view of this entire process not just from the kind of not just from the Kind of how the system is structured, but how the different personas can work together to to solve this problem of acceleration of Delivering the model time to value You know the key idea is that you know, we want to enable a quick Turnaround from getting data from the line of business getting in the hands of of quants to analyze it Enabling like taking those models quickly in the hands of developers to build applications Deploying them for production and basically kind of enabling this continuous virtuous cycle of moving data through the system, right? With that said I think We're at the end of our presentation, but I would love to start and get some questions for me Yeah questions and if anybody would like to copy the slides I'm happy to share them Just you know giving your contact info and we can share them And if you'd like we're here right now, we're gonna be here for the rest of the day So don't be shy And but if you were want to talk to us personally, you can find us in the in the attendance So thank you very much. I'm sorry. There was a question. Oh, hey thumbs up Okay, thank you all I appreciate that I'll take that as a question. Thank you