Financial Simulations On a Massively Parallel Connection Machine

The valuation of interest rate contingencies relies on simulation methods that integrate models of the term structure of interest rates with estimations of the contingent cash flows. Such simulations are quite complex and computationally int.ensive. In this paper we describe the development of massively parallel procedures for financial modeling simulations on systems like the Connection Mlachine. Several primitives generic among multiple applications are first presented, and the discussion culminates in an option adjusted analysis model for mortgage-backed securities. The primitives achieve sustained computing rates in the range 30 120 MFLOPS on a 4K processor Connection Machine, model CMI-2a. The response time for a complete option adjusted analysis (1 2 seconds) makes the model usable for real time analysis and trading applications.