Local signature-based expansions
Federico M. Bandi, Roberto Renò, Sara Svaluto-Ferro · 2025
Read Local signature-based expansions by Federico M. Bandi, Roberto Renò, Sara Svaluto-Ferro (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-2504-06351.txt.
Introduction
Federico Bandi, Roberto Renò, and Sara Svaluto-Ferro expand financial processes in small time steps using the mathematics of signatures. A signature compresses a path into a tower of iterated integrals. Those integrals form a basis on which many quantities can be expanded.
The team expands the process itself, its moments, and its characteristic function. Coefficients come out explicitly, so software can grind expansions to any order. Their basis handles time, several linked random drivers, and jump processes together. One mode shrinks the step, as older work did. Another fixes the step and raises the order, opening fresh identification tricks. Derivatives pricing and risk work feed on such expansions.
Catalog record
| CREATOR | Federico M. Bandi, Roberto Renò, Sara Svaluto-Ferro |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 115 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-06351
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Local signature-based expansions Thanks: We are grateful to Christa Cuchiero (discussant) and, for their comments, to conference participants in the XXV Workshop on Quantitative Finance (Bologna, April 11-13 2024), the Barcelona Workshop in Financial Econometrics (Barcelona, May 30-31 2024), the 6 t h 6^{th} “Quantitative Finance and Financial Econometrics” Conference (Marseille, June 4-7 2024), the 4 t h 4^{th} “High Voltage Econometrics” Conference (Lecce, October 4-5 2024) and the “Financial Econometrics meets Machine Learning” Conference (Lugano, November 1-2 2024). Federico M. Bandi, Roberto Renò, Sara Svaluto-Ferro Abstract We study the local (in time) expansion of a continuous-time process and its conditional moments, including the process’ characteristic function. The expansions are conducted by using the properties of the (time-extended) Itô signature , a tractable basis composed of iterated integrals of the driving deterministic and stochastic signals: time, multiple correlated Brownian motions and multiple correlated compound Poisson processes. We show that these properties are conducive to automated expansions to any order with explicit coefficients and, therefore, to stochastic representations in which asymptotics can be conducted for a shrinking time ( t → 0 t\rightarrow 0 ), as in the extant continuous-time econometrics literature, but, also, for a fixed time (such that t < 1 t<1 ) with a diverging expansion order. The latter design opens up novel opportunities for identifying deep characteristics of the assumed process. Keywords: Brownian motion, compound Poisson process, Itô signature. JEL Classifications: C18, C22. 1 Introduction We discuss the local (in time) expansion of a stochastic process and its conditional moments. We study a triplet of expansions. First, we expand the process itself. Second, we expand “regular” moments of the process, i.e., conditional expected functions of the process with well-behaved derivatives at time 0. Finally, we turn to “irregular” moments, i.e., conditional expected functions with unbounded derivatives at time 0. A key example of the latter is the characteristic function of the standardized process. We do so by exploiting the properties of the (time-extended) Itô signature of the process’ driving shocks, a tractable basis composed of iterated integrals of the driving deterministic and random signals (namely, time, correlated Brownian motions and correlated Poisson processes) whose inherent algebraic structure will be documented to lead to automatic expansions to any order. Assume, for simplicity, an Itô semimartingale (X_{t})_{t\in[0,T]} driven by a single Brownian motion (W_{t})_{t\in[0,T]}, but general specifications with multiple Brownian shocks and compound Poisson shocks will be allowed in what follows. A process (X_{t})_{t\in[0,T]} is (W,t) -differentiable if it admits the representation X_{t}=X_{0}+\int_{0}^{t}c_{0}(s)ds+\int_{0}^{t}c_{1}(s)dW_{s}, (1.1) for square integrable processes (c_{0}(t))_{t\in[0,T]} and (c_{1}(t))_{t\in[0,T]}, the first-layer process’ “characteristics” . Should these two (stochastic) characteristics feature the same representation as (X_{t})_{t\in[0,T]} with square-integrable processes c_{00}(t),c_{10}(t) and c_{01}(t),c_{11}(t) (the second-layer process’ “characteristics” ), respectively, then (X_{t})_{t\in[0,T]} would be viewed as being twice (W,t) -differentiable, and so on for higher orders. Given n+1 (W,t) -differentiability, by iterating stochastic integrals, we show that (X_{t})_{t\in[0,T]} can, equivalently, be written as X_{t}=\langle c,\widehat{\mathbb{W}}_{t}\rangle+{\varepsilon}_{n}(t), (1.2) where the first term on the right-hand side of Eq. ( 1.2 ) is the inner product between a vector c containing time-0 values of all characteristics (up to a generic depth n ), i.e., c=(S_{0},c_{0}(0),c_{1}(0),c_{00}(0),c_{10}(0),c_{01}(0),c_{11}(0),...) (1.3) and a vector of progressively increasing (up to depth n ) iterated integrals \widehat{\mathbb{W}}_{t}=\left(1,\int_{0}^{t}ds,\int_{0}^{t}dW_{s},\int_{0}^{t}sds,\int_{0}^{t}W_{s}ds,\int_{0}^{t}sdW_{s},\int_{0}^{t}W_{s}dW_{s},...\right), (1.4) i.e., the truncated (to the order n ) time-extended Itô signature of W . The second term on the right-hand side of Eq. ( 1.2 ) is an approximation error whose statistical order will be made explicit. The elements of the time-extended Itô signature satisfy a key property: their products can always be expressed as linear maps of other elements of the signature. We formalize this property and use it extensively in our proofs. We employ it, in particular, to transition from an expansion of the process (as in Eq. ( 1.2 )) to expansions of the local moments of the process, both for the case in which nonlinear transformations of the process have bounded derivatives at zero (the “regular” case) and for the case when these transformations have unbounded derivatives at zero (the “irregular” case). As we emphasize below, the irregular case - which includes the characteristic function of the standardized process - addresses the recent continuous-time econometrics literature directly. The notion of Itô signature traces back to Chen (1957) ; Chen (1977) and plays an important role in the context of rough path theory, initiated by Lyons (1998) . The usefulness of the concept is easily justified by existing universal approximation theorems. It is known that continuous (relative to specific variation distances) functionals of continuous as well as càdlàg paths can be approximated on compact sets of paths by linear functionals of the time-extended Itô signature (c.f. Cuchiero et al., 2022 ). We contribute to three largely separate (to date) lines of work: the growing recent literature in continuous-time econometrics on the use of short-term expansions for inference and pricing, the literature on signature-based methods in mathematical finance and the literature on short-term expansions in probability and process theory. Regarding the first literature, local expansions of the process’ characteristic function have been used to identify the characteristics (e.g., the volatility, c_{1}(0), and the volatility of volatility, c_{11}(0) ) of a process of interest on its first two layers (c.f., Jacod and Todorov, 2014 , Bandi and Renò, 2017 , Todorov, 2019 , Todorov, 2021 , and Chong and Todorov, 2024 ) as well as to price structured financial products with short expirations ( Bandi et al., 2023 ). Expanding locally conditional moments, like the process’ characteristic function, is known to be a cumbersome procedure, even when the expansion is low (first or second) order. As a result, existing expansions in the literature are low order. Low-order expansions, in turn, are empirically informative - for specific problems - but may be insufficient more generally, i.e., when time is not overly short and/or when econometric interest is in deeper characteristics that low-order expansion would not reveal. In our case, the expansions are automated and may be of arbitrary accuracy ( t^{n} with t<1 and any value of n ), thereby opening up new possibilities, from inference on the deep characteristics of the càdlàg process of interest (as n\rightarrow\infty ) to essentially exact (closed-form) solutions of short-term pricing problems. We turn to the second literature. Signature-based methods have experienced recent attention in mathematical finance (e.g., Perez Arribas et al., 2020 ) with applications largely focused on derivative pricing (e.g., Lyons et al., 2020 , Bayer et al., 2023 , Cuchiero et al., 2023 , Abi Jaber and Gérard, 2024 , and Cuchiero et al., 2024 ) and trading/portfolio allocation (e.g. Kalsi et al., 2020 , Akyildirim et al., 2023 , Cuchiero and Möller, 2023 , and Ning et al., 2023 ). Some studies have also exploited the potential of signature-based methods for financial machine learning (e.g., Buehler et al., 2020 , and Lemahieu et al., 2023 ). We introduce the notion of Itô signature to the continuous-time econometrics literature, a literature that has seen considerable growth over the last two decades thanks to the increased availability of both (high-frequency) data and computational power. Specifically, we exploit the approximation properties of the Itô signature to derive expansions of flexible continuous-time models with