The benchmark test dataset contained a wide variety of derivative products including: interest rate swaps of various types; caps/floors; European, American and barrier options; and credit default swaps. The composition of the portfolio, maturity profile, number and range of risk factors, and complexity of netting and collateral agreements were chosen to represent a realistic trading portfolio of a Tier 1 trading bank. The choice of 5,000 scenarios and 125 time steps represents an emerging best practice view from regulators and market practitioners of the number of scenarios and time steps that will be required in future to control sampling error and capture roll-off risk respectively.
For Algorithmicsâ clients, the commercial impact of this benchmark promises to be twofold: lower capital and operating costs and, for the first time, very sophisticated risk assessment available real-time and pre-deal.
Neil Bartlett, Chief Technology Officer, Algorithmics, commented: âThe traditional solution to support large-scale credit exposure simulations such as this is to scale out with hundreds or even thousands of CPUs. By working closely with Intel and explicitly taking advantage of the Intel Xeon processor L5400, we have been able to get performance on a single server that ordinarily would require a large cluster with many hundreds of CPUs. This leads us to expect considerably lower capital and operating costs for our clients and makes Monte Carlo-based pre-deal credit exposure measurement and limits checking a practical and compelling alternative to more simplistic add-on approaches.â
Nigel Woodward, Global Director, Financial Services at Intel, commented: âWe are aware that risk management is now more than ever a priority across financial services. These tests prove that where new investment is required to enable firms to prove they have effective policies in place, deep risk computation can be achieved on industry-standard, low-cost hardware. This promises a cost-effective approach in terms of capital expenditures and ongoing operational costs due to lower energy consumption in the data center. Using Intel's fasterLAB, Intelâs engineers worked with Algorithmics to ensure optimized performance from the processor level in the architecture.â
Dr Michael Zerbs, President and COO of Algorithmics, commented: âThis benchmark demonstrates the strength of our innovation and R&D in improving software performance and reducing total cost of ownership for our clients. Earlier this year the Counterparty Risk Management Policy Group reported that all banks should be able to understand their counterparty exposure to all major trading counterparties within a couple of hours. Our breakthrough improvements in computational performance now mean that global institutions can cost-effectively perform full risk simulations for their largest trading counterparties in a few minutes instead of hours, and pre-deal, what-if risk profiles for plain vanilla and exotic derivatives at the transaction, portfolio, and counterparty levels can be completed in milli- or sub-seconds.â