Measuring the market risk of complex derivatives places huge demands on computer power. RiskMetrics, a technology spinoff from JPMorgan, does it with analytics based on Microsoft .NET running on 4-way and 8-way dual-core AMD Opteron 64-bit clusters from HPC Systems.

“Our most successful penetration is with hedge funds,” said Philip Jacob, coordinator of the research and analytics groups at RiskMetrics. “Because RiskMetrics can handle a broad range of asset classes, including complex derivatives, we are probably the biggest vendor of risk systems to the hedge fund and fund of funds market.”

In addition to hedge funds, RiskMetrics’ clients include nine of the 10 largest financial institutions and half of the world’s central banks. Its tools allow users to measure and manage risk, and to communicate that risk to clients, investors, shareholders, managers and regulators.

Financial advisors at Northern Trust, which recently signed with RiskMetrics, will use the analytical tools to provide institutional and private clients with sophisticated risk and asset allocation analytics, investment planning and portfolio construction. RiskMetrics’ WealthBench lets the investment professionals create custom asset allocations based on Monte Carlo simulations, which can then be further optimized to match clients’ specific risk/return objectives. They can also perform hypothetical transaction and event stress testing as well as portfolio diversification and holdings-level risk analyses.

“We selected WealthBench for its robust analytics, multi-currency flexibility and effective presentation capabilities,” said John Skjervem, chief investment officer for Northern Trust’s Personal Financial Services division. Northern Trust will replace many of its current tools with WealthBench, he added.

Investment management has moved a long way from long-only US dollar equities, which has created expanding opportunities for RiskMetrics, said Ron Papanek, head of the hedge fund business at RiskMetrics.

“Financial instruments are getting more and more complex, so consequently the tools necessary to value and model these instruments also need to become significantly more complex. If you looked at a portfolio of equities, you can simulate those prices. You can even look at them historically and get a good idea of the volatility and risk of the securities and consequently the whole portfolio,” he added. “But when the portfolio contains derivatives, options, futures, credit derivatives and interest rate products, you can’t look at a simple historical prices series of that data. You need to re-price and value those instruments to get an idea and understanding of the risk. The world is growing dramatically more complex so we need more complex systems to monitor and manage that risk.”

At the same time, investors are beginning to think about risk as well as returns, so risk adjusted returns are an important measure for them. Papanek ticked off significant market events – Long Term Capital Management in 1998, the dot.com crash of 2000, and 9-11 in 2001.

“In the last 8 to10 years we have had a lot of once in a lifetime events, so that has put a much greater focus on risk. And it’s not just the sophisticated institutions such as banks, which have always been our clients, but a broader range of firms from hedge funds to asset managers to individual investors,” he said.

RiskMetrics incorporates multiple types of risk into a consistent methodology to provide clients with a view of their overall exposure.

“Tying it all together is where you really get value,” explained Papanek. “That is why our technology is so important. We can do simulation runs where we generate millions of simulations in a matter of minutes for a portfolio with 50,000 securities – including everything from equities to interest rate swaps. If you add one or two positions, we can recalculate that on the fly with our bank of AMD Opteron servers.”

The company operates its risk measurement as an application service provider (ASP), running client portfolios overnight to report on risk.

“We run 400 to 500 institutions, some with up to 75,000 securities in them,” Jacob said.

The firm runs thousands of scenarios to measure value at risk. It has built 50-plus securities models covering fixed income, fixed income derivatives, equities derivatives and varieties of options. It also allows clients to chain together models to measure the risk of complex structured products.

Using Microsoft Visual Studio 2005, with 64-bit targeting, RiskMetrics built its market risk engine to take full advantage of the processing power from AMD Opteron chips and hardware from HPC Systems, a small firm in southern California that is quick to take the latest from AMD and launch them into a production box within a month or two.

RiskMetrics is planning to use .NET features like remoting to facilitate the distribution of complex calculations which are currently coordinated using ‘roll your own’ coding based on a Web services model, Jacob added.

“.NET would provide the transport mechanism for describing the calculations that need to be carried out by the various worker CPUs. It is efficient, providing a high bandwidth, low latency way of interconnecting the processors. It is also easier because we can transport objects,” said Jacob. “.NET transport makes objects on a distant machine look programmatically as if they are on the local machine. We don’t have to write vast amounts of code to move objects from a local machine to a remote machine, so it dramatically simplifies the distribution of existing objected-oriented code into a distributed framework. In fact, Microsoft makes it so easy that you need to be aware that these objects are remote and you have to be sensible about accessing remote objects.”

The advantage of running risk calculations on .NET with TCP and a binary channel format is that it cuts the time lost in communication, and ups the time spent computing, he added.   

RiskMetrics had been running on 32-bit non-distributed two-way computers and began hitting memory limits.

“We couldn’t do more than 60,000 or 70,000 securities,” Jacob said.

Now it is using systems from HPC Systems with eight dual-core 64-bit AMD processors on each with 32 gigs of memory. The AMD Opteron’s hyper transport layer makes it scale well across CPUs, he added. Combined with .NET, it results in high-performance distributed computing that is also an efficient environment in which to develop.

“We can now run very large portfolios with complete stability,” added Jacob. “We can cache large amounts of intermediate data, like calibrated interest rate models, things that are the same between multiple client runs. Since .NET facilitates writing distribution code, we can write distributed computations much more quickly under .NET than when we were using Xerxes parsers.”

There are other advantages in time savings, costs and scale.

“End-to-end, processing jobs that used to take two and a half hours are now finished in half an hour or less, and the great advantage of a horizontally distributed architecture is that if a business requirement exists to deliver results more quickly, we can spend hardware dollars and scale nearly linearly,” Jacobs said.

www.amd.com

www.hpcsystems.com

www.riskmetrics.com