Simulation of Correlated Asset for Risk Purposes for Risk Calculation Purposes
What was the project about?
For compliance purposes, financial institutions are required to report on several financial risk measures. Any non-trivial calculation of risk, including the calculation of CVA, value-at-risk, or classic Greek risk factors are associated with a particular portfolio of assets and financial derivatives is based on a specification and simulation of the underlying assets and credit information about the involved. For portfolios with multiple such underlying assets, the processes that drive the asset quote generation are correlated and how they are correlated can be estimated from historical data.
Risk calculations are often associated with performance problems, due to portfolios containing non-trivial assets and due to certain risk calculations, that require certain assets, to be treated as a derivative that optionally can default. These performance problems are particularly caused by the inherent need for nested simulation. These problems are apparent both for smaller institutions that serve as asset managers on behalf of private investors and for larger institutions and have difficulties to fulfill compliance obligations.
Project scope and results
To help solve the above-mentioned performance problems, the involved researchers based on input from the industrial partners managed to create tooling for analyzing asset values (creating covariance matrices, etc) and for creating so-called historic simulations of assets in the specification language Haskell and in Futhark, a language for parallel programming on GPUs.
Moreover specifications and design of a prototype allowing investors to balance portfolio returns and CVaR was defined and design and implemention of a library for solving linear programming problems in parallel, a key component for the prototype was published as open source.
The project resulted in the establishment of several new BSc projects and one MSc project on the topic of balancing portfolio return, portfolio risk, and portfolio sustainability, which also was supported by Copenhagen Fintech in a follow up project (see more here).
Martin Elsman Professor, Department of Computer Science
Fine Analytics and KH Investment Partner
A key component of a solution has been published as an open source package for solving linear programming problems in parallel (parallel revised simplex method) using the programming language Futhark. Find it here
Additional background material
The Futhark Programming Language developed at DIKU. Find here