Domain specific languages and the acceleration of computational finance

Abstract

Although there is a long history of using domain specific languages (DSLs) in a variety of applications, including finance, few are used to accelerate computational finance. This talk summarizes the general advantages and difficulties of using domain specific languages and presents a sample of DSLs in computational finance. The most flexible and comprehensive of these systems focus on developing good domain models, transform specifications to a variety of target languages via intermediate representations, and are implemented in languages that combine functional, declarative, object-oriented, and pattern-matching features. This talk also includes a description of SciFinance, a domain-specific system that produces pricing codes for arbitrary financial products in OpenMP and NVIDIA CUDA as well as in pure C++ or C. Companion modules can also produce interfaces in Excel, Java, .NET, and .COM. The SciFinance system, implemented in Mathematica, tranforms systems of equations, constraints, and financial descriptors (and optional numerical method choices) into highly efficient simulation codes by applying refinement and optimization rules to a network of objects and equations. Numerical methods include Monte Carlo and partial differential equation techniques with specializations for parallel/distributed target architectures. Also covered in the talk are practical issues such as validation, and challenges for going beyond the current state-of-the-art.