Kolmogorov-Arnold Networks in Finance: A Comparative Analysis for Derivative Pricing Models
Date
2024-09
Authors
Kricun, Fernando
relationships.isContributorOfPublication
Basaluzzo, Gabriel
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de San Andrés. Escuela de Negocios
Abstract
Kolmogorov-Arnold Networks (KANs) have recently been introduced as an alternative to traditional Multilayer Perceptrons (MLPs) for neural network representation. In this study, we investigate the application of KANs in constructing physics-informed machine learning models to solve the Black-Scholes equation for pricing derivatives. We conduct a comparative analysis between KAN-based models and those using standard MLP architectures. Our findings reveal that KANs, despite having significantly fewer parameters, achieve comparable or superior accuracy to larger MLPs, with faster convergence across fewer training epochs. These results highlight the potential of KANs as a more efficient alternative to MLPs in the context of derivative pricing models.
Description
Fil: Kricun, Fernando. Universidad de San Andrés. Escuela de Negocios; Argentina.