WBS 5th Annual Initial Margin, XVA and KVA Conference
23 March 2016
Industry speakers shared practical advice on SIMM, dynamic margin modelling and the challenges of delivering ever greater numbers of XVA calculations.
At the WBS 5th Annual Initial Margin, XVA and KVA Conference (16-18 March 2016) it was great to see so many industry speakers sharing practical advice on SIMM, dynamic margin modelling and the challenges of delivering ever greater numbers of XVA calculations.
On the technical side there was a general discussion around the advantages of hardware versus software to crack the computation nut. For our part, GPU is not something we see attempted very often and algorithmic methods seem to be the most commonly used.
There was some discussion of re-coding versus overlay methods for AAD and there was general sharing of how to deal with the resulting memory challenges. It was nice to see the nested checkpoint method recommended and we would agree with this. It is easier said than done of course, but at least there is some agreement on the problem that needs to be solved.
On the methodology side there was a dizzying array of maths to be digested – particularly on KVA and margin simulation. As engineers, simplicity is the key for us and transparent, tractable methods are always the most appealing. Our partners at Global Valuation are one example and we continue to like their large matrix method for co-simulating risk factors. Another great example of this is Danske Bank, and Nicki Rasmussen gave a very clear walk through of its cash flow method following Jesper Andreasen's renowned CVA on an iPad Mini paper from a couple of years back.
Some of the approaches covered were less easy to digest but the opportunity to explore every aspect of the topic in depth was tremendously useful. Another highlight was Citibank's Youssef Elouerkhaoui, who presented a practical and detailed approach to modelling CCP risk.
One final discussion was more philosophical and covered the issue of whether or not to charge marginal capital cost (KVA) to trading P&L. Riskcare’s view on this is pretty straightforward. As a profit centre in any business, shareholders expect you to deliver a target return on their equity and that translates to a return on capital. Anything below that and you are in trouble, anything above and they are happy.
While this is the hurdle rate for happy investors, you wouldn't expect the profitability of your unit only to be measured versus target return – after all, profit is profit and what matters here is cost. So we would subtract only the cost of capital, to keep traders focused on capital consumption and to keep P&L volatility down.
As with every challenge, there are numerous third-party companies offering remedies in the XVA space and a wide array of products now provide fast on-demand calculation as a library, as a service, or as a tool set. Some bold claims are being made here about compute times and it remains to be seen how effective these are in the field.
We do expect this segment to continue to expand, however, as underlying models become more standardised and the commercial argument for outsourcing grows. Either way, the conclusion is that evaluation of XVA metrics can and must continue to be made to be computationally cheaper.
Overall, it was a lively and useful few days and a good opportunity for the XVA community to come together for some mutual problem solving. While there are definitely some frustrations – particularly around uncertainties with bi-lateral margining – there will be plenty to discuss over the coming months.