Market Risk With A Twist of Windows HPC 2008
Take trades (in this Proof Of Concept (POC) it will be rates trades) calculate the cashflows, PV, PV01 to input into an OLAP Cube (real-time)
Some thoughts on the problem:
- In the case of a trade change (amend) you can use the current calibrated curve and discount factors – event driven
- Portfolio size should possible influence the way we calculate. For example, for a small portfolio, it might be cheaper to calculate all the required discount factors, whereas a large portfolio may mean its best to calculate all discount factors for the next 100 years (per day)
- In the case of flows, we have a closed form solution (as pervious blogged in F#). However options/exotics we need to use Monte Carlo, which is a lot more compute intensive.
- Time of Day and currency will impact our HPC usage – e.g. Japaness YEN market
- Partitioning the HPC cluster by currency may be appropriate and assist in delivering an SLA
Sidebar: It’s logical to want to also look at Market Risk calculations using Azure. Having spend a lot of time with Azure during the beta period, the issue that still gates me is the whole Silverlight PollingDuplexHttpBinding – I still don’t see the ServiceBus being supported over HTTP/HTTPS for Silverlight which in my view is just plain silly.