“What is sort of universally accepted in academic circles (is that) there has not been a development of statistical mathematical framework for comparing the values of similar CDOs (collateralized debt obligations) … that enables one to model in a forward looking way the market risk of these instruments, “ says Cotton.

Currently, the private research firm is developing a mathematical framework for creating credit models in the future and has turned to supercomputing to do the intense number crunching.
“The timing is coincidental. We had identified this problem well before the last couple of weeks played out,” says Nicholas de Jong, VP of engineering and operations at Julius Finance.

On Aug. 21, Julius Finance announced it was using Interactive Supercomputing’s Star-P software to develop more realistic credit models from which banks, hedge funds and other financial institutions can make more accurate predictions about a portfolio’s potential. Star-P enables researchers to easily code algorithms and models on their desktops using popular high –level languages like MATLAB, Python and R, automatically transforming the programs to run on parallel HPCs, stated the release. Star-Plus eliminates the need to re-program, the applications in complex languages such as C, Fortran or MPI (message passing interface) to run on Julius Finance’s Linux-based cluster, according to the release.

While most of the problems during the credit crisis had to do with
CDOs that provided exposure to bonds and mortgage backed products with sub-prime loans, Julius Finance is focusing on synthetic CDOs that reference credit default swaps on corporate bonds, rather than the physical underlying bonds.

Part of the problem today is that industry standard models are applied to CDOs on a deal-by-deal basis and every deal is different because of different levels of subordination, different maturities and different portfolios of assets that are referenced, says Cotton, who previously worked in Morgan Stanley’s synthetic CDO business where he was responsible for several innovations in credit derivatives. “People have tended to use models to price the portfolios in isolation,” he says. Another flaw is that the models fell short in pricing illiquid instruments. “The methods they used in deals which are not similar to a liquid traded instrument like CDX or iTRAXX (both are leading credit derivatives indexes) that sort of mathematics has not been developed at all,” says Cotton.
Though the researchers at Julius Finance say they are not experts on mortgage modeling, their work could benefit other types of CDOs in the future.

“We’re trying to create one model for many products whereas today people are using a different model for every product,” says Cotton. “When we put a value on an illiquid security, we’re better placed to take in more data points in the market than people would otherwise use and also to draw a more sensible interpolation between those data points,” says the CEO.

“We’re trying to prototype new models,” says de Jung, who contends the existing models have been used for years because they are simple and quick to price deals. “Unfortunately, they’re not necessarily very realistic. To experiment with more plausible models, which might better reflect reality is more computationally intensive,” de Jung explains. This entails doing as much testing and research in a short amount of time, which is why it’s using supercomputing. But that is not the “secret sauce,” adds de Jung.

At the moment, Julius Finance is “flying a little bit under the radar,” says de Jung, but it expects the firm to offer something by the end of the year. The business model is a subscription service where banks and hedge funds would access to a daily feed of data, and Julius Finance would generate all the computation that is required, he says.