Rough kernel hedging

Nicola Muça Cirone, Cristopher Salvi · 2025

Read Rough kernel hedging by Nicola Muça Cirone, Cristopher Salvi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-09683.txt.

Introduction

Nicola Muca Cirone and Cristopher Salvi tackle risk hedging in rough markets in a 2025 paper. Hedging means placing trades that offset losses when prices move. Many real price paths are jagged and depend on the whole past path, not just the last step. The authors build their method on kernels, a math tool for comparing paths, and on signature kernels that sum up a path's shape.

Their method makes few guesses about how markets move, since it treats paths in a broad rough form. A key theorem shows a single best answer exists for the hedging task. The answer can be written in closed form for a wide range of loss rules. Extra inputs such as trading signals or past hedge choices can be folded in too.

Catalog record

CREATORNicola Muça Cirone, Cristopher Salvi
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE86 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-09683

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Rough kernel hedging Nicola Muça Cirone, Cristopher Salvi Abstract Building on the functional-analytic framework of operator-valued kernels and un-truncated signature kernels [ 38 ] , we propose a scalable, provably convergent signature-based algorithm for a broad class of high-dimensional, path-dependent hedging problems. We make minimal assumptions on market dynamics by modelling them as general geometric rough paths, yielding a fully model-free approach. Moreover, by means of a representer theorem, we provide theoretical guarantees on the existence and uniqueness of a global minimum of the resulting optimization problem, and derive an analytic solution under highly general loss functions. Similar to the popular deep hedging [ 5 ] —but in a more rigorous fashion—our method can also incorporate additional features by means of the underlying operator-valued kernel, such as trading signals, news analytics, and past hedging decisions, aligning closely with true machine-learning practice. ∎ MSC 46C07 60L10 60L20 1 Introduction In idealized, complete, and frictionless markets, it is theoretically possible to perfectly hedge financial derivatives, thereby eliminating risk through appropriate hedging strategies. However, real markets are incomplete due to transaction costs, market impact, liquidity constraints, and other frictions, making perfect hedging generally unattainable. Consequently, traders aim to find approximate hedging strategies that minimize a cost function selected according to their risk preferences. There exists an extensive body of literature on hedging strategies in market models that include frictions, highlighting the complex nature of the problem. In markets where trading activities temporarily impact asset prices, [ 37 ] analyses a one-dimensional Black-Scholes model where trading a security affects its price. The optimal trading strategy in this setting is derived by solving a system of three coupled nonlinear PDEs. Extending this framework, [ 2 ] addresses a more general tracking problem, which encompasses the temporary price impact hedging issue, within a Bachelier model. The authors provide a closed-form solution for the strategy that involves conditional expectations of a time integral over the optimal frictionless hedging strategy. In the context of proportional transaction costs, [ 41 ] shows that in a Black-Scholes market, the minimal superhedging price for a European call option is equal to the current spot price of the underlying asset. This result implies that super-replication strategies are of limited practical interest in one-dimensional cases due to their prohibitive cost and become numerically intractable in higher-dimensional settings. Mean-variance hedging has been another area of focus in the literature. [ 21 ] tackles the mean-variance hedging problem for vanilla options on Lévy processes, offering a semi-explicit solution. Their approach is further developed in [ 17 ] , whose authors present an algorithm for the mean variance hedging of vanilla options. In a frictionless framework, [ 24 ] uses BSDEs to derive the optimal hedge for the mean-variance problem. However, finding the optimal hedge remains challenging in practical applications. Machine learning techniques have also been applied to hedging problems. [ 22 ] employs neural networks to price and hedge European options. Building on this, [ 5 ] extends the methodology to exotic derivatives by approximating the optimal hedging strategy using deep learning. Despite its potential, this approach generally finds only local minima rather than global ones, and the computational cost of training can be significant. An alternative method involves the use of truncated signatures from the rough path theory [ 33 ] to linearize the general hedging problem. The motivation for using signatures in the context of path-dependent financial optimization problems stems from the well-known fact that linear functionals acting on the signature form a unital algebra of real-valued functions on paths that separates points [ 28 ] ; thus, by the classical Stone-Weierstrass theorem, such linear functionals of signatures are universal approximators for functions on (compact sets of unparameterized) paths. Hence, signature coefficients form an ideal set of features for machine learning applications dealing with sequential data [ 13 , 39 , 7 ] . In fact, signature methods have become very popular in recent years in quantitative finance [ 1 , 40 , 20 , 36 ] as well as in the broader data science communities, and have been applied in a variety of contexts including deep learning [ 25 , 35 , 9 , 18 , 10 , 23 , 3 ] , kernel methods [ 38 , 27 , 26 , 34 ] , cybersecurity [ 11 ] , and computational neuroscience [ 19 ] . Going back to the hedging problem, [ 31 ] proposes to use this approach, but it requires several stringent assumptions: 1) payoffs are approximated by signature payoffs; 2) the method is limited to low-dimensional market dynamics due to the exponential growth of signature features; 3) the theory is developed for one-dimensional market dynamics and assumes that the augmented process with its quadratic covariation (i.e., lead-lag transformation) forms a geometric rough path—a property that holds almost surely for semimartingales as shown in [ 14 ] , but not for general rough paths; 4) it is restricted to polynomial or exponential hedging losses. Contributions In this paper, we utilize the well-established functional-analytic framework of operator-valued kernels together with untruncated signature kernels [ 38 ] to provide a scalable and provably convergent signature-based algorithm for a broad class of path-dependent hedging problems that resolves all the limitations mentioned above in the previous work and is applicable to high-dimensional financial data streams. We impose minimal assumptions on the market dynamics by modelling them as general geometric rough paths. Our approach is thus entirely model-free. We present a novel representer theorem that guarantees the existence and uniqueness of a global minimum, and provide an analytic solution to a broad class of hedging problems with highly general loss functions. Similar to deep hedging, but in a more rigorous manner, our method considers not only the prices of hedging instruments, but can also incorporate features captured by the operator-valued kernel. These features can include additional information like trading signals, news analytics, or past hedging decisions—quantitative data that a human trader might utilize—in line with true machine learning practices. Outline of the paper In Section 2 , we summarize the theoretical foundations and numerical implementation of our method in the context of a simple quadratic hedging problem which we use as a running example throughout the paper. In Section 3 , we introduce signature kernels within our framework and explain how this class of kernels provide the necessary guarantees to solve the hedging problem. In Section 4 , we develop a general and flexible theory for kernel hedging, going well beyond the quadratic case. Finally, in Section 5 , we present simple experimental results that validate our theoretical findings. 2 Motivating Example - Quadratic Hedging For demonstration purposes, we summarise the key ideas of our method for solving a quadratic hedging problem. We assume familiarity of the reader with the theory of rough paths; a brief primer on the subject is given in Appendix A . We refer to [ 15 , 32 ] for an in-depth treatment of the topic. Let V=\mathbb{R}^{d} be the space in which the market prices X take values and fix a temporal horizon T>0 . Since the profit-and-loss (P&L) of a trading strategy is given by an integral against X we need to fix, a priori, a

Read the complete text