📝 SummarXiv

Abstract

Built to generalise classical stochastic calculus, rough path theory provides a natural and pathwise framework to model continuous non-semimartingale assets. This paper investigates the ultimate capacity of this framework to support frictionless continuous No-Free-Lunch markets \`a la Kreps-Yan. We establish a ``Rough Kreps-Yan" theorem, which links our No Controlled Free Lunch (NCFL) condition to the unbiasedness of the driver of the price process as a rough integrator. The central work of this paper is a complete classification of these unbiased rough integrators with respect to different classes of controlled paths as integrands. As the set of admissible trading strategies is enlarged to include Markovian-type and signature-type portfolios, the only admissible random rough paths must be infinitesimally close to the It\^o rough path lift of a standard Brownian motion, up to a time change. In particular, Gaussianity is no longer a model assumption, but rather a no-arbitrage market consequence. Notably, simple strategies do not appear in the theory, and if they are then reintroduced, the rough noise is further enforced to be the It\^o rough path of Brownian motion itself. Ultimately, this implies that continuous frictionless markets based on rough path theory are inevitably constrained to the semimartingale paradigm, providing a definitive answer on the limits of this approach. Our framework covers $\alpha-$H\"older continuous rough paths for $\alpha>0$ arbitrarily small.