Mar 1, 2013

Hybrid Models & Optimization Techniques for Real-Time Counterparty Credit Risk Exposures

Dr. Serguei Issakov, Global Head of Quantitative Research at Numerix , joins host and CMO Jim Jockle to discuss how Hybrid models can be leveraged in the real-time calculation of CVA, FVA and DVA computations and in terms of implementation – the types of optimizations financial institutions must consider in order to achieve real-time objectives, including the use of Factor models and Shared models.

Lastly, Dr. Issakov discusses the benefits of computing exposures for CVA by the exact same models used for pricing – resulting in greater mitigation of model risk and a very high level of accuracy for CVA computation.

Weigh in and continue the conversation on Twitter @nxanalytics, LinkedIn, or in the comments section below.

Video Transcript: Hybrid Models & Optimization Techniques for Real-Time Counterparty Credit Risk Exposures

Jim Jockle (Host): Hi and welcome to Numerix Video Blog, I’m your host Jim Jockle. Joining me again, Dr. Serguei Issakov, Global Head of Quantitative Research at Numerix. Welcome Serguei.

Serguei Issakov (Guest): Thank you Jim.

Jockle: Wanted to continue our conversation regarding counterparty credit risk around a paper that was introduced at Risk USA, back in November of 2012, and in our last video blog we talked a little bit about the adoption of the modern approach for counterparty credit risk as well as calculating risk exposures for vanillas via super swap as well as structured products through the newly introduced algorithmic exposure so please, if you haven’t seen that post look back online and you can get the background on that.

But one of the things we talked Serguei, is hybrid modeling and the use of hybrid modeling. Could you for our viewers go into that a little bit more? What actually is a hybrid model? What is the importance of hybrid modeling and how does that effect calculations such as CVA, DVA, FVA etc.?

Issakov: Well for hybrid models, across worldwide hybrid models are important for counterparty risk. The requirements are that you have to account calculation between all asset classes. If you have equity, interest rates, FX, inflation, commodities in your portfolio, you have to account for calculations between all these (risks in your portfolio). 

Hybrid models do exactly that. The construction of hybrid models is as follows: first you start from component model that belong to particular asset class such as: interest rate, equity or FX. Then when each model is constructed/calibrated you build the hybrid model as a container of all those models with some modifications that are necessary to do like drift adjustments, and then the hybrid model, then you correlate all the models in your portfolio by applying external correlation metrics, and this is where correlations come into play. And then this model is ready to basically compute exposures for any deals in your portfolio.

Jockle: In terms of a practical implementation, what kind of optimizations should people be thinking about in terms in order to achieve the type of accuracy required for these calculations?

Issakov: Well, because hybrid models may become very big in this computation, you have to apply some optimizations to satisfy memory and time requirements, and one of the most important optimizations is the application of factor models in this type of computation. And it’s very nontrivial to apply the idea factors models for exposures and CVA computation. And what we’ve came up with recently is, we call it parallelization of the hybrid model.

First you find primary and secondary factors in your model. And then you’re able to break the big hybrid model that might consist of thousands of factors, into smaller pieces, compute exposures by those smaller pieces and aggregate at the end. And the final result of the calculation is exactly the same as the result that could be obtained by the original big hybrid model, and that was a nontrivial step to achieve.

Jockle: And then looking at the presentation, you walk through a particular use case where you had 118 different factors built-in and you were still able to come down and achieve sub-millisecond pricing and risk evaluation on that.

Issakov: Right, that’s correct.

Jockle: And one would presume over, not just for a single instrument but when you start to multiplying over thousands and thousands of instruments and trying to meet those real-time objectives even that optimization at a core aggregation is critical to achieve those time horizons for trading.

Issakov: Yes that’s true. I also want to mention another type of optimization which we call Shared model, where you actually build the model once, you calibrate it once, and you create all Monte Carlo paths once, and then, you share the same model across all the computations in your portfolio. So that the time that’s spent on creating Monte Carlo paths which might be substantial for vanilla instruments. Pricing for itself for vanilla instruments is very quick its milliseconds, microseconds, but creation of Monte Carlo paths takes time. So this optimization which is called Shared model offers a way to achieve real time CVA calculations.

Jockle: And one would presume in terms of concepts around Algorithmic Exposure the optimization and Share models that there’s a real mitigation of model risk in terms of applying a hybrid model against that portfolio, is that accurate?

Issakov: Yes that’s correct because the hybrid model basically, one feature/benefits of the modern approach is that you compute exposures in CVA by exactly the same models that you use for pricing. That’s a very important point and a few years back it was only a dream, even in big banks, that it’s possible to do that. But it’s possible now because normally exposures in counterparty risk for computation, some simple models were used. But this approach, with optimizations, allows us to use exactly the same models as for pricing which provides a very high level of accuracy for CVA computation. 

Jockle: Serguei thank you so much for joining us today.

Issakov: Thank you Jim.

Jockle: If you’d like to see the presentation that was given by Serguei or any of the reference papers (available on SSRN - vanillas and exotics) back from Risk USA please visit, where they can be downloaded and of course, join in the conversation @nxanalytics on twitter as well as look at our blog going forward. Well see you next time. Thank you.

Blog Post - Apr 05, 2010

The Art and Science of Risk Management for Today's Markets

Need Assistance?

Want More From Numerix? Subscribe to our mailing list to stay current on what we're doing and thinking

Want More from Numerix?

Subscribe to our mailing list to stay current on what we're doing and thinking at Numerix

Subscribe Today!