Computational Finance (E1)
Project PI: Patrick Jaillet
PROJECT DESCRIPTION
This project will test the capabilities of advanced Intel platforms for solving high-end computational finance problems in the area of pricing, portfolio optimization, and risk management. Leading-edge financial institutions routinely use advanced techniques from mathematics, engineering, and computer science to create, price, and manage risk for new instruments. High-end computers are now powerful enough to allow analyses and decisions based on real-time data feeds (tick-by-tick), as well as on huge real-time databases (data warehousing).
Specific computationally demanding problems to be addressed in this proposal will include (1) solutions to support activities linked to data warehousing and data mining; (2) fixed-income securities and derivative businesses; (3) valuation of securities with multiple sources of "noise," in multi-factor models (e.g., term structure models, FX models, and convertible securities); and (4) valuation of path-dependent instruments, especially mortgage-backed. This program is to develop and test distributed and parallel computing solution procedures for these most computationally demanding problems based on exclusive Intel Architecture - Microsoft NT system. The design techniques are aimed at providing a lower cost of ownership for the solution of the most computationally demanding finance problems.
TECHNICAL CHALLENGES
The complexity and fast resolution of this class of computational finance problems routinely require the use of very advanced techniques and algorithms derived from mathematics, engineering, and computer science. The need for high performance computing, high-levels of systems integrity, systems availability, and "fast-cycle" solutions becomes even more critical when "real-time" analyses and capital allocation decisions are needed. The computational finance decisional systems must be made based on very large streams of real-time data that is increasingly derived from electronic trading activity. A second challenge will be the integration of such procedures within hybrid networks allowing both NT machines and supercomputers/superservers (e.g. Cray).
IMPACT
Traditionally, "back-office'' calculations have been developed on networks of UNIX workstations with access to fast superservers and supercomputers. Recently however, Window NT workstations have increasingly replaced the "front-office" infrastructure in financial markets (e.g., London, Wall Street), and have started to gain access to the "back-offices" as well. It is thus essential to evaluate and demonstrate the (1) "at the limit" performance of parallel and distributed computing Pentium-Windows NT architecture for such computationally intensive financial problems, and (2) the ease of their integration with traditional supercomputers. Proof of concept will increase the exposure of Intel Architecture - Microsoft NT system to the financial markets, and more generally into financial industries at large.
EQUIPMENT
Cluster NT Systems, Systems Lab with 32 dual processors workstations, fast communications between computers with 100 Mbit/s Fast Ethernet (to be shared with Project E3)
RESOURCES
This project will be centered in the Institute for Computational Finance (ICF) and the Texas Institute for Computational Mathematics (TICAM). The 35 affiliate faculty of ICF span three colleges (Business, Engineering, and Natural Sciences) and eight departments. Resources will also include facilities such as the EDS Financial Trading and Technology Center (FTTC), and the IC2 Institute.
BENEFITS TO INTEL
As discussed above, this project is to evaluate and demonstrate the "at the limit" performance of parallel and distributed computing Pentium-Windows NT architecture for computationally intensive finance problems. Proof of concept will increase the exposure of Intel Architecture - Microsoft NT system into the financial markets, and more generally into financial industries at large.
RELATED WEB SITES
Follow this link to the Texas Business School Enterprise Systems and Management Web sites.