Pith. sign in

REVIEW 3 major objections 5 minor 38 references

Global universal approximation with Brownian signatures

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Linear functionals on the Brownian signature can approximate any p-integrable Brownian-adapted process.

desk verdict L^p universal approximation via classical signatures is largely proved and worth referee time, but the headline 'any adapted p-integrable process' overreaches what is actually shown. read the letter →

arxiv 2512.16396 v2 pith:CKH4GGMF submitted 2025-12-18 math.PR cs.LGq-fin.MF

Mihriban Ceylan , David J. Prömel This is my paper · ORCID
classification math.PRcs.LGq-fin.MF MSC 60L1060H1060J6591G99
keywords signatureuniversalapproximationroughpathsBrownianmotionL^pdensitynon-anticipativefunctionalsweightedspacesstochasticdifferentialequations
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proves global L^p-universal approximation theorems for paths. Simple linear combinations of iterated integrals — the coordinates of the signature of a time-extended rough path — are dense, in the L^p sense, among p-integrable functionals on weighted rough path spaces, and among non-anticipative functionals on stopped rough paths. The paper then shows the Wiener measure satisfies the required exponential moment condition, which yields the advertised consequence: linear functionals on the time-extended Brownian signature can approximate any p-integrable progressively measurable process adapted to the Brownian filtration, including solutions of stochastic differential equations. A sympathetic reader would care because this supplies the theoretical foundation for signature-based models driven by Brownian noise, showing they are universal approximators in a genuine global sense rather than merely on compact sets.

What carries the argument

The signature of a path — the full collection of iterated integrals as an element of the tensor algebra — is the universal feature map; adding time as the zeroth coordinate makes the signature injective up to translation. The proofs rest on weighted spaces B_ψ(X) of functions on (stopped) rough paths with admissible weight ψ(X̂) = exp(β‖X̂‖^γ_{cc,α}), whose sublevel sets are compact in a weaker Hölder topology. A weighted Stone–Weierstrass theorem shows the algebra generated by the signature coordinates {⟨e_I, X̂_T⟩} separates points and is dense in B_ψ. For Brownian motion, Gaussian tail estimates for the Carnot–Carathéodory α-Hölder norm supply the finite exponential moment that turns weig

What would settle it

Find a progressively measurable process Y ∈ H^p for which the map ω ↦ Y_t(ω) is not a Borel function of the stopped time-extended Brownian rough path Ŵ_{[0,t]}(ω) in the topology d_Λ,α′ used on Λ^α_T; such an example would show the abstract's 'any process' statement is false, while leaving Theorems 3.4 and 3.13 intact.

Watch

Extended reading notes

Core claim

The paper's central claim: linear functionals ℓ(X̂_T) = Σ_{|I|≤N} ℓ_I ⟨e_I, X̂_T⟩ on the signature of a time-extended rough path are dense in L^p(Ĉ^α_{d,T}, ν) for any finite Borel measure ν satisfying the exponential moment condition ∫ exp(βp‖X̂‖^γ_{cc,α}) dν < ∞; the analogous stopped-rough-path version is dense in L^p(Λ^α_T, ν). The Wiener measure satisfies this condition with γ=2, so the stopped theorem gives the headline consequence: linear functionals on the time-extended Brownian signature approximate, in H^p, every p-integrable progressively measurable process adapted to the Brownian filtration, including strong solutions of Itô SDEs with continuous, linearly growing coefficients.

Load-bearing premise

The abstract's claim that every p-integrable Brownian-adapted process is approximable rests on the assumption that each such process can be represented as a Borel-measurable function of the stopped time-extended Brownian rough path; the paper does not prove this and cites it as a missing measurability analysis.

Editorial extensions

If this is right

  • Any Y ∈ H^p adapted to the Brownian filtration can be approximated in E∫_0^T |Y_t − ℓ(Ŵ_t)|^p dt by a finite linear combination of Brownian signature coordinates, with the approximation error controlled by the exponential moment of the rough path norm.
  • Solutions of Itô SDEs with continuous, linearly growing coefficients — under unique strong solution assumptions — are universal approximable by signature models, independent of the drift and diffusion structure.
  • The general theorem extends the same density to arbitrary finite Borel measures on rough path spaces with finite exponential moments, covering Gaussian processes beyond Brownian motion, such as fractional Brownian motion.
  • The non-anticipative version on stopped rough paths provides the natural functional-Itô-calculus analogue: path-dependent and causal functionals can be approximated by signature terms that only use information up to time t.
  • Because approximation is in L^p rather than uniform on compacts, the result applies to processes whose paths are almost surely unbounded, which is the case for Brownian motion.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves open the Borel-measurability step needed to pass from Corollary 4.3(ii) to 'any progressively measurable process'; if that gap is filled, the SDE-specific proof in Proposition 4.4 could be replaced by a direct argument covering a broader class of processes, including those not given as SDE solutions.
  • A quantitative version of the theorem is plausible: the rate at which the approximation error decays in N should depend on the weight moment M = ∫ ψ^p dν and on the Gaussian tail of the rough path norm; this rate is not given in the paper but could be extracted from the Stone–Weierstrass approximation bounds.
  • Since only strict monotonicity of the time coordinate is used, the same L^p universality should hold for signatures of extensions by other strictly monotone one-dimensional paths, suggesting the machinery transfers to other Gaussian rough paths and to Lévy-type drivers whenever the analogous exponential moment holds.
  • For practice, the result means that pricing, hedging, and optimal-control functionals in Brownian-driven models can be pre-computed as linear features of the path's signature — a fact that underlies recently proposed signature-based market models.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This paper develops L^p-universal approximation theorems for signature functionals on geometric rough path spaces. The main results are Theorem 3.4 (general path-dependent functionals on time-extended geometric α-Hölder rough paths) and Theorem 3.13 (non-anticipative functionals on stopped rough paths), both proved by transfer from weighted-space Stone–Weierstrass results. Section 4 specializes to Brownian motion: Corollary 4.3 shows that linear functionals of the time-extended Brownian signature are dense in L^p for functionals of the full/stopped Brownian rough path, and Proposition 4.4 shows the corresponding approximation for solutions of SDEs with continuous coefficients of linear growth. The abstract additionally claims approximation of arbitrary p-integrable Wiener-adapted processes, but this stronger statement is not proved; Remark 4.5 explicitly defers the required measurability analysis.

Significance. If the central theorems hold, they give a meaningful extension of compact-set signature universal approximation theorems to global L^p spaces, with direct applications in finance and stochastic analysis. The paper has real strengths: Theorem 3.4's truncation–Lusin–Tietze argument is clean and correct conditional on Proposition 3.3; Proposition 3.11 verifies in detail the hypotheses of the weighted Stone–Weierstrass theorem for stopped rough paths; and the Brownian exponential-moment condition is properly reduced to Gaussian tail estimates from the rough-path literature. However, the principal novelty claimed in the abstract—approximation of every p-integrable adapted process—is not actually established, and the foundation of Theorem 3.4 is delegated rather than proved. These issues are load-bearing for the paper's headline claims.

major comments (3)
  1. [Abstract, §4, Remark 4.5] The abstract states that linear functionals on the time-extended Brownian signature can approximate 'any p-integrable stochastic process adapted to the Brownian filtration'. This is not what is proved. Corollary 4.3(ii) applies only to processes of the form Y_t = f(Ŵ_{[0,t]}) with f Borel measurable on the stopped rough path space Λ^α_T, and Proposition 4.4 covers SDE solutions. The bridge from arbitrary F^W-progressively measurable Y∈H^p to such a Borel representation is explicitly deferred in Remark 4.5, which says it 'requires a careful measurability analysis ... cf. [BBH+25, Section 4.2]'. Since the abstract's headline quantifies over all p-integrable adapted processes, this is an unsupported overclaim. The theorem statements themselves are sound; the paper should either prove the factorization or revise the abstract and the statements in Section 4 to match the class actually covered
  2. [§3.1, Proposition 3.3] Proposition 3.3 is the sole foundation for the general-functional L^p theorem (Theorem 3.4), but its proof is one sentence: it 'follows line by line' from [CST25, Theorem 5.4] after replacing weakly geometric rough paths by geometric rough paths. This replacement is not automatic. The paper should verify the hypotheses of the weighted Stone–Weierstrass theorem [CST25, Theorem 3.9] for the geometric space: (i) the signature algebra separates points on the time-extended geometric rough path space; (ii) the weight ψ(X)=exp(β||X||^γ_{cc,α}) is admissible; (iii) the algebra has ψ-moderate growth. Since [CST25, Theorem 5.4] is stated for weakly geometric paths, either a precise check for geometric paths or a quotation of a result in [CST25] that already covers the geometric case is needed. This is a load-bearing gap because Theorem 3.4 cannot be used until Proposition 3.3 is justified.
  3. [§4.2, Proposition 4.4] Step 1 of Proposition 4.4 approximates the coefficients μ,σ by compactly supported smooth functions using [HS12, Proposition 1.1], then invokes [KN88, Theorem A] for the existence and uniqueness of the strong solution of the approximating SDE. The approximation argument is plausible, but the precise conditions of [KN88, Theorem A] are not stated, and it is not clear that this reference directly gives strong existence and uniqueness for smooth compact-support coefficients (as opposed to a convergence or weak-solution result). Since the rest of the proof depends on Y^ε being a well-defined strong solution and on the bound (4.2), a standard reference for strong solutions of SDEs with smooth coefficients would make the step transparent. This is fixable by citation or a short argument.
minor comments (5)
  1. [§4.1, Corollary 4.3] The statement begins 'Let α∈(1/3,1/3)', which should read α∈(1/3,1/2).
  2. [§3.2, Lemma 3.9] In the proof, the preimage is written as φ([0,T]×{bX^t_{[0,T]} ∈ bC^α_{d,T}: ψ(bX_{[0,t]})≤R}). Since ψ is defined on the stopped space Λ^α_T, the notation is confusing: the condition should be ψ(φ(t,bX))≤R, or an explicit definition of the stopping map. Please clarify.
  3. [§4.2, Step 2] The Itô–Stratonovich correction term is written with σ^ε ∂σ^ε/∂y as in the scalar case. For matrix-valued σ^ε, the term is a contraction and should be written componentwise or with a remark that the usual multi-dimensional convention is used.
  4. [§3.2, Proposition 3.11] The point-separating argument invokes [BB11, Corollary 4.24] to conclude that a continuous function is zero from vanishing integrals against polynomials. If this is the density-of-polynomials corollary, the reference is appropriate, but the property being used should be named for the reader.
  5. [§4.1] The notation Ŵ is used both for the time-extended Brownian motion and for its rough path (and the signature is denoted Ŵ as well). While the intended meaning is usually clear from context, a sentence fixing the notation would prevent confusion, especially in Corollary 4.3 and Example 4.2.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all load-bearing inputs are external benchmarks and the L^p theorems are genuine transfers.

full rationale

The derivation chain is self-contained against external results. Theorem 3.4 reduces to Proposition 3.3 (density in the weighted space) via dominated convergence, Lusin's theorem and Tietze's extension theorem; Proposition 3.3 in turn relies on [CST25, Theorem 5.4] and the weighted Stone–Weierstrass theorem of [CST25], whose authors are not the present authors. Theorem 3.13 similarly uses Proposition 3.11, whose point-separation argument invokes [BB11, Corollary 4.24], again external. The Brownian application in Corollary 4.3 uses the Gaussian tail estimates of [FH20, Propositions 3.4 and 3.5] and the integrability criterion of [FV10, Lemma A.17], and Proposition 4.4 uses classical SDE approximation ([Klo92, Theorem 4.5.3], [KN88, Theorem A], [HS12, Proposition 1.1]) together with RDE well-posedness from [FH20]. None of these load-bearing inputs is authored by Ceylan and Prömel, none is a fitted parameter relabeled as a prediction, and no uniqueness or ansatz is imported from the authors' own prior work. The stated limitation in Remark 4.5 — that making the representation Y_t = f(Ŵ_[0,t]) fully rigorous 'requires a careful measurability analysis' and is deferred — is a scope/correctness gap in the headline claim about 'any p-integrable process adapted to the Brownian filtration', but it is not a circular step: the gap does not make the theorem equal to its input by construction. The central results are genuine transfers of external weighted Stone–Weierstrass and rough-path estimates to L^p density, so the circularity score is 0.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

No data-fitting anywhere; the only empirical input is the Wiener exponential-moment condition imported from [FH20]. The hand-chosen weight parameters β, γ are existential constants, not fitted values; they are listed for completeness. The load-bearing premises are domain assumptions: the weighted Stone-Weierstrass machinery ([CST25]), the injectivity of the time-augmented signature ([HL10, BGLY16]), the Wiener exponential moment ([FH20]), and the (author-flagged) Borel functional representation on the stopped rough path space (Remark 4.5).

free parameters (2)
  • γ (weight exponent) = 2
    Hand-chosen in ψ(X̂) = exp(β||X̂||^γ_{cc,α}); needs γ ≥ ⌊1/α⌋ = 2 and the boundary case γ = 2 is used because the Gaussian tail estimate in [FH20] gives E[exp(η| ||Ŵ|| |²_α)] < ∞. Not data-fitted; an existential choice.
  • β (weight scale) = β ∈ (0, η/(C^γ p)]
    Hand-chosen small enough that ∫ ψ^p dν < ∞ via E[exp(βp C^γ | ||Ŵ|| |^γ_α)] ≤ E[exp(η| ||Ŵ|| |²_α)] < ∞ (Cor 4.3 proof). Existence parameter, not fitted.
assumptions (6)
  • domain assumption Weighted Stone-Weierstrass theorem [CST25, Thm 3.9] applies to the algebra generated by the time-augmented signature coordinates on spaces of geometric α-Hölder rough paths (and stopped versions)
    Load-bearing for Propositions 3.3 and 3.11; Prop 3.3 transfers it from weakly geometric to geometric paths without proof.
  • domain assumption The time-augmented signature map X̂ ↦ (⟨e_I, X̂_T⟩)_I is injective on Ĉ^α_{d,T} and separates stopped paths on Λ^α_T (time-augmentation kills tree-like equivalence)
    Remark 3.1; cited to [HL10, BGLY16]; needed for point separation in the Stone-Weierstrass hypotheses.
  • domain assumption Exponential moment condition for the Stratonovich-enhanced time-extended Brownian rough path: ∃η > 0 with E[exp(η | ||Ŵ|| |_α²)] < ∞, plus equivalence of ||·||_{cc,α} and | ||·|| |_α
    Section 4.1, proof of Cor 4.3(i)-(ii); cited to [FH20, Props 3.4/3.5], [FV10, Lemma A.17]; the hinge that makes the Brownian application valid.
  • domain assumption Functional representation: every F^W-progressively measurable Y ∈ H^p equals f(Ŵ_{[0,·]}) for a Borel f on (Λ^α_T, d_{Λ,α'})
    Remark 4.5 explicitly flags this as not fully rigorous; needed for the abstract's 'any adapted p-integrable process' claim.
  • standard math Well-posedness and L^p bounds for Itô SDEs with linear-growth coefficients; smooth compact-support approximation of coefficients; RDE well-posedness for C^3_b vector fields
    Proposition 4.4 steps 1-2; cited to [Klo92, Thm 4.5.3], [HS12, Prop 1.1], [KN88, Thm A], [FH20, Thms 8.3/9.1].
  • domain assumption Compact embedding of geometric α-Hölder rough paths into geometric α'-Hölder for α' < α (with d_{cc,α'})
    Lemma 3.9 and Remark 2.1; cited to [CST25, Remark A.7]; needed for ψ to be an admissible weight.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Global universal approximation with Brownian signatures." pith.science (2026). https://pith.science/paper/CKH4GGMF

@misc{pith2026251216396,
  author       = {Pith},
  title        = {Pith review of: Global universal approximation with Brownian signatures},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CKH4GGMF}},
  note         = {Machine review of arXiv:2512.16396}
}
abstract

We establish $L^p$-universal approximation theorems for general path-dependent and non-anticipative functionals on suitable rough path spaces, showing that linear functionals acting on signatures of time-extended rough paths are dense with respect to the $L^p$-distance. To that end, we derive global universal approximation theorems for weighted rough path spaces. We demonstrate that these $L^p$-universal approximation theorems apply to Gaussian processes, in particular, to fractional Brownian motion. As a consequence, linear functionals on the signature of the time-extended Brownian motion can approximate any $p$-integrable stochastic process adapted to the Brownian filtration, including solutions to stochastic differential equations.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 2 linked inside Pith

  1. [1]

    Imanol Perez Arribas, Cristopher Salvi, and Lukasz Szpruch, Sig- SDE s model for quantitative finance , Proceedings of the First ACM International Conference on AI in Finance (New York, NY, USA), ICAIF '20, Association for Computing Machinery, 2021

    2021

  2. [2]

    2, Springer, 2011

    Haim Brezis and Haim Br \'e zis, Functional analysis, sobolev spaces and partial differential equations, vol. 2, Springer, 2011

    2011

  3. [3]

    Hager, Sebastian Riedel, and Tobias Nauen, Stochastic control with signatures, SIAM J

    Peter Bank, Christian Bayer, Paul P. Hager, Sebastian Riedel, and Tobias Nauen, Stochastic control with signatures, SIAM J. Control Optim. 63 (2025), no. 5, 3189--3218

    2025

  4. [4]

    Christian Bayer, Goncalo dos Reis, Blanka Horvath, and Harald Oberhauser, Signature M ethods in F inance , first ed., Springer Finance, Springer Cham, 2025

    2025

  5. [5]

    Financial Math

    Erhan Bayraktar, Qi Feng, and Zhaoyu Zhang, Deep signature algorithm for multidimensional path-dependent options, SIAM J. Financial Math. 15 (2024), no. 1, 194--214

    2024

  6. [6]

    Horatio Boedihardjo, Xi Geng, Terry Lyons, and Danyu Yang, The signature of a rough path: uniqueness, Adv. Math. 293 (2016), 720--737

    2016

  7. [7]

    Hager, Sebastian Riedel, and John Schoenmakers, Optimal stopping with signatures , The Annals of Applied Probability 33 (2023), no

    Christian Bayer, Paul P. Hager, Sebastian Riedel, and John Schoenmakers, Optimal stopping with signatures , The Annals of Applied Probability 33 (2023), no. 1, 238--273

    2023

  8. [8]

    29 (2025), no

    Christian Bayer, Luca Pelizzari, and John Schoenmakers, Primal and dual optimal stopping with signatures, Finance Stoch. 29 (2025), no. 4, 981--1014

    2025

Show all 38 references
  1. [9]

    Rama Cont and David-Antoine Fournié, Functional I t \^o calculus and stochastic integral representation of martingales , The Annals of Probability 41 (2013), no. 1

    2013

  2. [10]

    Finance 35 (2025), no

    Christa Cuchiero, Guido Gazzani, Janka M \"o ller, and Sara Svaluto-Ferro, Joint calibration to SPX and VIX options with signature-based models , Math. Finance 35 (2025), no. 1, 161--213

    2025

  3. [11]

    3, 910--957

    Christa Cuchiero, Guido Gazzani, and Sara Svaluto-Ferro, Signature-based models: Theory and calibration, SIAM Journal on Financial Mathematics 14 (2023), no. 3, 910--957

    2023

  4. [12]

    London Math

    Kuo-Tsai Chen, Iterated integrals and exponential homomorphisms, Proc. London Math. Soc. (3) 4 (1954), 502--512

    1954

  5. [13]

    4, 1239--1303

    Christa Cuchiero and Janka M \"o ller, Signature methods in stochastic portfolio theory, SIAM Journal on Financial Mathematics 16 (2025), no. 4, 1239--1303

    2025

  6. [14]

    01, 192--202

    Ilya Chevyrev, Vidit Nanda, and Harald Oberhauser, Persistence P aths and S ignature F eatures in T opological D ata A nalysis , IEEE Transactions on Pattern Analysis & Machine Intelligence 42 (2020), no. 01, 192--202

    2020

  7. [15]

    176, 1--42

    Ilya Chevyrev and Harald Oberhauser, Signature moments to characterize laws of stochastic processes, Journal of Machine Learning Research 23 (2022), no. 176, 1--42

    2022

  8. [16]

    Christa Cuchiero, Philipp Schmocker, and Josef Teichmann, Global universal approximation of functional input maps on weighted spaces, arXiv preprint arXiv:2306.03303v5 (2025)

    2025

  9. [17]

    Zdzisław Denkowski, Stanislaw Migórski, and Nikolas Papageorgiou, An introduction to nonlinear analysis: Theory, SpringerLink Bücher, Boston, MA, 2003

    2003

  10. [18]

    Finance 19 (2019), no

    Bruno Dupire, Functional I t\^o calculus , Quant. Finance 19 (2019), no. 5, 721--729

    2019

  11. [19]

    Friz and Martin Hairer, A course on rough paths: with an introduction to regularity structures, second edition ed., Universitext, Cham, Switzerland, 2020

    Peter K. Friz and Martin Hairer, A course on rough paths: with an introduction to regularity structures, second edition ed., Universitext, Cham, Switzerland, 2020

    2020

  12. [20]

    Folland, Real analysis: modern techniques and their applications, 2

    Gerald B. Folland, Real analysis: modern techniques and their applications, 2. ed. ed., A Wiley-Interscience publication, New York Weinheim, 1999

    1999

  13. [21]

    Avner Friedman, Foundations of modern analysis, Courier Corporation, 1982

    1982

  14. [22]

    Theory Related Fields 136 (2006), no

    Peter Friz and Nicolas Victoir, A note on the notion of geometric rough paths, Probab. Theory Related Fields 136 (2006), no. 3, 395--416

    2006

  15. [23]

    Friz and Nicolas B

    Peter K. Friz and Nicolas B. Victoir, Multidimensional stochastic processes as rough paths: theory and applications, Cambridge studies in advanced mathematics; 120, Cambridge, 2010

    2010

  16. [24]

    Benjamin Graham, Sparse arrays of signatures for online character recognition, ArXiv Preprint arXiv:1308.0371 (2013)

    2013 arXiv

  17. [25]

    Ben Hambly and Terry Lyons, Uniqueness for the signature of a path of bounded variation and the reduced path group, Ann. of Math. (2) 171 (2010), no. 1, 109--167

    2010

  18. [26]

    Martina Hofmanov \'a and Jan Seidler, On weak solutions of stochastic differential equations, Stoch. Anal. Appl. 30 (2012), no. 1, 100--121

    2012

  19. [27]

    Patrick Kidger, Patric Bonnier, Imanol Perez Arribas, Cristopher Salvi, and Terry Lyons, Deep signature transforms, 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), 32 (2019)

    2019

  20. [28]

    Financial Math

    Jasdeep Kalsi, Terry Lyons, and Imanol Perez Arribas, Optimal execution with rough path signatures, SIAM J. Financial Math. 11 (2020), no. 2, 470--493

    2020

  21. [29]

    Kloeden, Numerical solution of stochastic differential equations, Applications of Mathematics, Stochastic Modelling and Applied Probability; 23, Berlin, Heidelberg, 1992

    Peter E. Kloeden, Numerical solution of stochastic differential equations, Applications of Mathematics, Stochastic Modelling and Applied Probability; 23, Berlin, Heidelberg, 1992

    1992

  22. [30]

    Kaneko and S

    H. Kaneko and S. Nakao, A note on approximation for stochastic differential equations, S\' e minaire de P robabilit\' e s, XXII , Lecture Notes in Math., vol. 1321, Springer, Berlin, 1988, pp. 155--162

    1988

  23. [31]

    Kir \'a ly and Harald Oberhauser, Kernels for sequentially ordered data, J

    Franz J. Kir \'a ly and Harald Oberhauser, Kernels for sequentially ordered data, J. Mach. Learn. Res. 20 (2019), Paper No. 31, 45

    2019

  24. [32]

    Shreve, Brownian motion and stochastic calculus, second ed., Graduate Texts in Mathematics, vol

    Ioannis Karatzas and Steven E. Shreve, Brownian motion and stochastic calculus, second ed., Graduate Texts in Mathematics, vol. 113, Springer-Verlag, New York, 1991

    1991

  25. [33]

    Lyons, Michael Caruana, and Thierry L\' e vy, Differential equations driven by rough paths, Lecture Notes in Mathematics, vol

    Terry J. Lyons, Michael Caruana, and Thierry L\' e vy, Differential equations driven by rough paths, Lecture Notes in Mathematics, vol. 1908, Springer, Berlin, 2007, Lectures from the 34th Summer School on Probability Theory held in Saint-Flour, July 6--24, 2004, With an intro...

    1908

  26. [34]

    Daniel Levin, Terry Lyons, and Hao Ni, Learning from the past, predicting the statistics for the future, learning an evolving system, ArXiv Preprint arXiv:1309.0260 (2013)

    2013 arXiv

  27. [35]

    Terry Lyons, Sina Nejad, and Imanol Perez Arribas, Numerical method for model-free pricing of exotic derivatives in discrete time using rough path signatures, Appl. Math. Finance 26 (2019), no. 6, 583--597

    2019

  28. [36]

    , Non-parametric pricing and hedging of exotic derivatives, Appl. Math. Finance 27 (2020), no. 6, 457--494

    2020

  29. [37]

    Andrew McLeod and Terry Lyons, Signature methods in machine learning, EMS Surv. Math. Sci., 2025

    2025

  30. [38]

    Schell and R

    A. Schell and R. Alaifari, Nonparametric R egression of S tochastic P rocesses via S ignatures , ETH Zurich, Research Report No. 2023-45 (2023)

    2023

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.