Case Studies

Speeding up pricing compex instruments in the cloud

Reval is a leading derivative risk management and hedge accounting software-as-a-service (SaaS) provider. In order to quickly develop new structured products and vastly speed up Monte Carlo based derivatives for its SaaS customers, Reval chose SciFinance® from SciComp, used in conjunction with GPU hardware from NVIDIA.

Reval has been adding complex structured instruments to it's flagship SaaS product, Reval®. These instruments included Dual Currency bonds, Power Reverse Dual Currency bonds, Inverse Floaters and CMS Steepeners, all with embedded caps and floors and call/put options as well as Range Accruals and Principal Protected Notes. Reval needed to find a way to improve the development time for these instruments.

These types of derivatives are priced using Monte Carlo simulation, which requires running between 20,000 to 50,000 trials and uses a great deal of CPU time. Implementing a Monte Carlo framework in a SaaS environment adds extra stress to CPU usage at month-end closing and at times of high system usage. A 50,000 trial simulation of a complex instrument can take up to 60 seconds to run in a conventional hosted environment. Even with dedicated analytics servers, this can cause unacceptable response times as analytics servers are shared across hundreds of clients.

Reval teamed with SciComp to implement a combined hardware and software solution. The results were a faster derivatives model development cycle and accelerations of 100X faster than serial code for derivative valuations in the SaaS environment.

Promoting a focus on the modeling tasks

In a detailed case study coauthored with SciComp, analysts at Merrill Lynch indicate that software synthesis makes it much easier to handle complex problems and allows them to focus on the problem and modeling choices, rather than on programming and debugging. In doing research, they can now solve within a day or two problems that appeared too complex to solve in a reasonable time using conventional techniques. In addition, quick turnaround gives their busy analysts the time to experiment with alternative techniques and fine-tune production codes. The analysts have also found that automatically generating codes ensures a consistent set of assumptions about the valuation of a portfolio and a consistent style across all models, even when those models are generated by different people over an extended period of time.

Increasing code accuracy and development speed

Dr. Anastasios Politis, a quantitative analyst at KBC Financial Products, uses SciFinance to generate codes for pricing new options (both for research and production) and to determine whether closed-form solutions are precise enough. Dr. Politis says that SciFinance makes code development faster and easier for him, and that his bank benefits from more accurate models and fewer deals lost because of slow pricing. SciFinance allows Dr. Politis to develop many models within a single day rather than over the course of a week. Using conventional manual programming methods to develop finite difference codes of the variety SciFinance produces, he says, is immensely time-consuming (exactly how time-consuming depends on the resemblance to existing codes). By putting the correct numerical elements, such as solvers, at his disposal, SciFinance enables Dr. Politis to develop some new models in just a few hours. For generating certain types of models (those for American-style options and barrier options) SciFinance has become Dr. Politis' preferred method. He finds codes generated by SciFinance to be superior to the more traditional lattice-based codes. In addition, because certain features can be expressed with a single ASPEN specification statement, SciFinance greatly facilitates his pricing of the varied complex features of options such as convertible bonds.

Gaining confidence in models

Fortis Bank analysts use SciFinance primarily to gain confidence in their existing models and to test new modeling approaches. They expect to generate production pricing models in the future. With SciFinance, analysts have been able to rapidly generate a variety of accurate, PDE-based codes to validate existing pricing products. Also, they can more quickly and confidently test new pricing models, which helps bring new exotic-option products to market faster. Fortis Bank analysts also use SciFinance to research new pricing approaches and conduct experiments that give them a better feel for more sophisticated models.

Reducing labor in a risk-control environment

Dr. Raymond Hawkins, Associate Director of Risk Control at Bear, Stearns Securities Corporation, used SciFinance in a risk-control rather than trading environment. The risk-control department performs risk analysis for clearance of client portfolios on a daily basis, repricing every single security within a portfolio and doing a variety of stress tests to determine the portfolio risk. For each security, the department first develops an ASPEN specification that incorporates the terms and conditions of the security, then develops a pricing tool from the code that SciFinance generates. Bear, Stearns Securities previously depended on proprietary models for the pricing tools, but moved to SciFinance because they felt it would be an extremely cost-effective approach. For Dr. Hawkins, an important feature of software synthesis is its ability to reduce labor while producing consistent, highly accurate programs. With program synthesis, highly trained analysts can focus their energy on the analysis and risk control, not on programming.

Excerpted from "SciFinance: A Program Synthesis Tool for Financial Modeling." Proceedings of the Twelfth Innovative Applications of Artificial Intelligence Conference (IAAI-2000), Austin, Texas, July 30-August 3, 2000. R. Akers, I. Bica, E. Kant, C. Randall, and R. Young. American Association for Artificial Intelligence.