Stochastic Control with Signatures

TL;DR

This paper introduces a novel approach to stochastic optimal control by parameterizing controls using path signatures, enabling stable, flexible, and numerically tractable solutions with applications demonstrated in mathematical finance.

Contribution

It establishes the density of signature-based controls in the class of measurable controls and develops rough path methods for stability analysis and numerical approximation.

Findings

Signature-based controls are dense in measurable controls.

Proposed methods are stable under rough path conditions.

Numerical algorithms accurately solve benchmark problems.

Abstract

This paper proposes to parameterize open loop controls in stochastic optimal control problems via suitable classes of functionals depending on the driver's path signature, a concept adopted from rough path integration theory. We rigorously prove that these controls are dense in the class of progressively measurable controls and use rough path methods to establish suitable conditions for stability of the controlled dynamics and target functional. These results pave the way for Monte Carlo methods to stochastic optimal control for generic target functionals and dynamics. We discuss the rather versatile numerical algorithms for computing approximately optimal controls and verify their accurateness in benchmark problems from Mathematical Finance.