MVA Optimisation with Machine Learning Algorithms
15 Pages Posted: 23 Feb 2017 Last revised: 14 Dec 2017
Date Written: October 23, 2017
MVA is becoming a dominant XVA component in interdealer derivatives trading in the post Margin Reform environment. Unlike FVA which can be either a funding cost or a funding benefit, MVA due to BCBS IOSCO IM is always a cost because of non-rehypothecability of the initial margin posted under the margin rules covering non-centrally cleared derivatives. This prompts dealers to investigate MVA optimisation solutions. The dealers face a complex non-linear optimisation problem: not only MVA is a non-linear function in continuous trade parameters such as tenor and notional, it is also a function of discrete variables such as counterparty and underlying asset. Many standard optimisation algorithms based on the continuity or convexity of the objective function are no longer suitable due to the non-linear and the mixed discrete and continuous nature of the problem. At the same time we observe, mainly on the buy side, increasingly widespread usage of Machine Learning techniques in quantitative finance. This paper is about novel applications of Genetic Algorithm and Particle Swarm Optimisation to the problem of MVA optimisation.
Keywords: MVA, Machine Learning, Genetic Algorithm, Particle Swarm Optimisation, MVA Optimisation
JEL Classification: C61, C63, G13, G15
Suggested Citation: Suggested Citation