Pith. sign in

REVIEW 4 major objections 7 minor 52 references

The Mathematics of Volatility Surfaces

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read The paper claims that a nondegenerate Gaussian shock at an active volatility-surface constraint exits the no-arbitrage set with probability 1/2, so exact static arbitrage-freeness forces degenerate noise, reflection, or submanifold…

desk verdict A real and new boundary theorem for Gaussian volatility-surface models, slightly over-advertised in the abstract; deserves serious peer review, not desk rejection. read the letter →

arxiv 2608.05198 v1 pith:AJVUYQBX submitted 2026-08-04 q-fin.MF

classification q-fin.MF MSC 91G2060H1546E3591G80
keywords impliedvolatilitysurfacestaticarbitragelocalconstrainedstochasticfieldsGaussiandynamicsboundarytheoremneuraloperatorsnormalizingflowsKarhunen–Loèveexpansion
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 treats the implied volatility surface as a single infinite-dimensional state—total variance $w_t(k,\tau)=\tau\sigma^2_t(k,\tau)$—and asks what it means for that state to stay inside the static no-arbitrage set. It proves a boundary incompatibility theorem: at any surface where a constraint (positivity, calendar monotonicity, or butterfly positivity) is active, a nondegenerate Gaussian shock exits the admissible set with probability tending to $1/2$ as the time step shrinks, and in continuous time exact invariance forces degenerate normal noise, reflection, or confinement to an arbitrage-free submanifold. The same geometry yields practical tools: Dupire local variance is the ratio $\partial_\tau w/g[w]$, a Karhunen–Loève expansion with exact truncation error, a closed-form minimum-variance vega-field hedge, and three viable model templates (reflected price dynamics, tangent-projected innovations, parameter maps). The result matters because Gaussian factor/PCA surface models, the standard descriptive picture since the late 1990s, cannot be exactly arbitrage-free at the boundary; model design must build the constraints into the architecture.

What carries the argument

The load-bearing object is the total-variance field $w_t(k,\tau)=\tau\sigma^2_t(k,\tau)$ as an element of a weighted Sobolev space $H=H^2_\rho(D)$, evolving under an $H$-valued Itô equation $dw_t=\mu_t\,dt+\Sigma\,dW_t$. Its admissible set $K_m$ is cut out by three constraint families: positivity $w\ge m$, calendar monotonicity $\partial_\tau w\ge 0$, and the Gatheral–Jacquier butterfly inequality $g[w]\ge 0$, where $g[w]=\bigl(1-\frac{k w_k}{2w}\bigr)^2-\frac{w_k^2}{4}\bigl(\frac1w+\frac14\bigr)+\frac{w_{kk}}2$. Two mechanisms carry the argument. First, the boundary-layer scaling: a Gaussian shock has normal component of order $\sqrt{\Delta}$ and drift of order $\Delta$, so at an active constraint the signed violation $\Phi(w_0+\sqrt{\Delta}\xi)$ collapses to a centered normal, giving the $1/2$ exit probability; the continuous-time trichotomy follows from the Itô formula together with Dambis–Dubins–Schwarz and the Brownian law of the iterated logarithm. Second, the chart geometry: the pointwise Black–Scholes map $B$ makes the price set $K_{c,m,M}$ closed and convex while $K_m$ is nonconvex in total-variance coordinates, and the ratio $a=\partial_\tau w/g[w]$ turns the constraint functionals into the numerator and denominator of Dupire local variance.

What would settle it

Compute, on real option data, the one-step violation frequency of an unconstrained Gaussian KL generator conditional on small pre-step butterfly margin as the time step shrinks: the theorem predicts convergence to $1/2$ along the curve $N(\Phi/(\sqrt{\Delta}\sigma_\Phi))$, so a frequency that stays near zero, or a finite counterexample surface with a nondegenerate Gaussian direction that never exits $K_m$, would falsify it.

Watch

Extended reading notes

Core claim

The paper's central discovery is an incompatibility between Gaussian dynamics and the boundary of the static no-arbitrage set $K_m=\{w\ge m,\ \partial_\tau w\ge 0,\ g[w]\ge 0\}$. Theorem 7.3 shows that for any smoothed constraint functional $\Phi$ that is active at $w_0$ and any trace-class Gaussian increment with nonzero variance along the constraint normal, the Euler shock $w_0+\Delta\mu+\sqrt{\Delta}\xi$ violates $\Phi$ with probability tending to $1/2$ as $\Delta\to 0$, independent of drift, because normal noise is order $\sqrt{\Delta}$ while drift is order $\Delta$. Corollary 7.5 sharpens this to continuous time: an exactly invariant $K_m$-valued Gaussian field must have vanishing normal noise at every active constraint, so the only exits from the Gaussian class are degenerate diffusion, reflection, or evolution on a submanifold contained in $K_m$. Complementing the negative theorem, the interior is exponentially safe (Borell–TIS bound), the price chart is convex while the total-variance chart is not, and the Dupire identity $a=\partial_\tau w/g[w]$ makes the same two constraint functionals observable through local volatility. The paper claims this geometry reorganizes the model universe into three viable constructions and makes no-arbitrage an architectural requirement for learned generators.

Load-bearing premise

The argument is set on a fixed compact quoted window in moneyness and maturity, with time to maturity bounded away from zero, and the paper assumes that on this window the Lee wing constraints are dominated by the butterfly constraint; if infinite-domain wing limits impose material constraints that the finite window misses, the arbitrage-free labels attached to generated surfaces on the window would not transfer to real tradable surfaces outside it.

Editorial extensions

If this is right

  • No unconstrained Gaussian or PCA-style surface model, however the drift is chosen, can be exactly statically arbitrage-free at an active constraint; the exit probability tends to $1/2$ as the time step shrinks.
  • Any exactly arbitrage-free surface model must implement one of three mechanisms: reflected dynamics in the convex price chart, tangent projection of Gaussian proposals, or diffusion on a parameterized submanifold whose image lies in $K_m$.
  • On a fixed quote grid, normally reflected price-coordinate dynamics have a unique strong solution and the projected Euler scheme converges, so static constraints can be enforced exactly pathwise.
  • Learned generators inherit the half-law if their noise is Gaussian with nonvanishing constraint-normal component; feasibility must be architectural, for example through simplex heads in price coordinates or cone heads in local-volatility coordinates.
  • The minimum-variance vega-field hedge has the closed form $\alpha^*=(H^*CH)^{-1}H^*C\nu$, with the increment covariance operator as the metric, and its residual variance is a computable audit statistic.

Reading between the lines

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

  • The same boundary-layer argument should apply to any stochastic field with smooth equality constraints and nondegenerate Gaussian noise—yield curves, survival curves, or density surfaces—so the trichotomy of degeneracy, reflection, or submanifold confinement is likely a general template for shape-constrained infinite-dimensional dynamics.
  • The ratio $a=\partial_\tau w/g[w]$ suggests a direct diagnostic: on days with extreme local volatility, checking whether the numerator or denominator approaches zero attributes the behavior to calendar versus butterfly activity, and this diagnostic is only tested indirectly in the paper's empirical design.
  • For model comparison, violation frequency conditional on pre-step margin—not unconditional violation rate—is the metric that separates Gaussian from constrained generators, because the paper's interior survival bound shows both types coincide away from the boundary.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper develops a unified infinite-dimensional state-space framework for implied total variance surfaces, viewed as H^2_rho(D)-valued random fields constrained by positivity, calendar monotonicity, and a butterfly differential inequality. It establishes topological and geometric properties of the admissible set (closedness, convexity in price coordinates, nonconvexity in total-variance coordinates, tangent cones), proves a boundary incompatibility theorem: at an active constraint detected by window-averaged functionals, a nondegenerate Gaussian shock exits the admissible set with probability tending to 1/2 as the time step shrinks (Theorem 7.3), with a continuous-time trichotomy (degenerate normal noise, reflection, or submanifold confinement) in Corollary 7.5. It then develops Hilbert-space dynamics with exact modal reduction, Karhunen-Loève expansions, a vega-field hedging theory with a closed-form minimum-variance hedge, a local-volatility chart via the Dupire ratio, three viable model constructions (reflected price dynamics, tangent-projected innovations, parameter-map/submanifold dynamics), neural-operator heads, arbitrage-free normalizing flows, fading-signature fields, and a pre-registered empirical protocol with nine models, six metric families, and seven falsifiable hypotheses.

Significance. If the central claims hold, the paper supplies a rigorous geometric foundation for volatility-surface modeling in which static no-arbitrage is a constraint-set property rather than an afterthought. The half-law in Theorem 7.3 is a clean and honest result: the proof via Taylor expansion plus Slutsky is elementary and appears correct, and the continuous-time trichotomy is a standard and well-executed Dambis-Dubins-Schwarz/LIL argument. The paper is unusually transparent: every formal result is tiered [Proved]/[Conditional]/[Conjectural], the nonconvexity certificate in Proposition 3.3 is verified with interval arithmetic and ships with code, and Section 15 is a genuinely pre-registered design with explicit falsification conditions. The hedging formula (53) and the modal truncation error (46) are concrete, usable outputs. The main weakness is that the headline scope—'at any surface with an active constraint'—is wider than the proved statement, which covers only positive-measure active windows in H^2; this gap affects the abstract, the introduction, and the empirical hypothesis H1, and requires either a proof extension or a scope correction.

major comments (4)
  1. [§7.3, Theorem 7.3; Remark 7.2; Abstract; §1] Theorem 7.3 establishes the 1/2 exit limit only for the window-averaged functionals (30), i.e., when the constraint is active on a set of positive measure inside the window. The abstract's 'at any surface with an active constraint' and the introduction's 'Gaussian field models ... are incompatible with the boundary' are broader, and Remark 7.2 itself concedes that an isolated point contact is invisible to the L^2 order structure. Since butterfly constraints in practice often bind on small sets (a single strike or short-dated wing), the gap is load-bearing. The finite-grid Theorem 7.6 gives only tangency and inward-drift conditions, not an exit probability, so it does not fill the gap. The paper should either restrict the advertised claim to 'positive-measure active window' or promote the asserted pointwise version (for H^s with s>3) from a remark to a proved theorem.
  2. [§7.1 and Corollary 7.5] Corollary 7.5 is proved for the additive-noise equation (28), d w_t = μ_t dt + Σ dW_t. The trichotomy is then stated in the abstract and Section 1 without this restriction, but state-dependent Gaussian-type factor models d w_t = μ_t dt + B(w_t) dW_t, which are the natural lift of PCA-style models after learning, are outside the proof. The same argument plausibly extends to Lipschitz B(w), but the extension is not written down; as it stands, the advertised 'exact invariance therefore requires degeneracy, reflection, or confinement' overstates the proved scope. The proof should be extended or the statement qualified to the additive class.
  3. [§15.4, Hypothesis H1] Hypothesis H1 predicts that the unconstrained Gaussian model's one-step violation frequency 'follows the standardized-margin curve ... increasing toward 0.5 as the conditioning margin and simulation step shrink.' This prediction is based on Theorem 7.3, but that theorem does not apply to pointwise or isolated active constraints on the empirical grid, where Theorem 7.6 only yields tangency and drift conditions. As written, H1 may test a stronger statement than the proved half-law supports. The design should either condition on positive-measure active regions (e.g., using window functionals on a neighborhood of the margin state) or explicitly formulate a separate, weaker prediction for isolated contacts.
  4. [§3.1 and §16] The paper repeatedly justifies the compact window D=[k_min,k_max]×[τ_min,τ_max] by stating that Lee wing constraints are 'dominated by the butterfly constraint on this window' and are omitted from K. This dominance claim is asserted but not demonstrated; a counterexample on the boundary of D or a brief argument (or reference) showing that any surface satisfying the butterfly inequality on D can be extended to an arbitrage-free surface on the full half-plane would make the scope of the 'arbitrage-free' labels precise. As it stands, the skeptic's concern that finite-window feasibility may not transfer outside D is left open.
minor comments (7)
  1. [§4.2, Remark 4.3] The equality T_K(w)=L(w) is tiered [Conditional] on a Robinson-type constraint qualification with the pullback of a Slater direction through B^{-1}. Since this equality is not used in the proof of Theorem 7.3, the conditional tier is acceptable, but a one-line example of a w where the pullback Slater direction is known to work would make the remark more useful.
  2. [§3.1] The sentence 'On the compact window D they are dominated by the butterfly constraint' would benefit from a pointer to the precise sense of domination (e.g., pointwise bound on the Lee moment slope in terms of g), since it is the only place where the omission of Lee wings is justified.
  3. [§2.2, Proposition 2.1(iii)] The paper is careful that w_kk only exists in L^2, but the butterfly functional g is then only an L^2 element; the notation g[w]≥0 a.e. is clear, though the text could state explicitly that pointwise positivity of g is not claimed for general w∈H.
  4. [§5, Theorem 5.1] The proof of (19) divides two identities and assumes the denominator is nonzero; the subsequent Corollary 5.2 treats the zero-pole geometry, but the theorem statement could explicitly include the assumption ∂_τ w>0 and g[w]>0 to avoid ambiguity.
  5. [§13.3] The discussion of quasi-invariance and the Ramer framework is accurate and appropriately hedged; however, the phrase 'generic coordinatewise neural flows need not satisfy these conditions' could be strengthened by a concrete example (e.g., a nonlinear shift map on ℓ^2 that sends the reference Gaussian to a singular measure) to make the warning more tangible.
  6. [Throughout] The paper is very long and contains several notational transitions (w, c, a, S, etc.). A consolidated notation table near the end of Section 2 would help the reader navigate the coordinate changes, especially the duality between the variance chart and the price chart.
  7. [References] The citations to 'Noguer i Alonso 2026a,b' are to unpublished working papers by the author; in a journal version, these should be marked as forthcoming or available online, and the text should clarify their role (they are used for covariance repair taxonomy and market-mode analysis, not for the main proofs).

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the half-law and invariance theorems derive from the stated Gaussian dynamics and constraint geometry; same-author citations are auxiliary and not load-bearing.

full rationale

I found no load-bearing circularity in the derivation chain. Theorem 7.3 derives the 1/2 exit law from a first-order Taylor expansion of the active constraint functional, the scaling <ell, xi>_H = N(0, sigma_Phi^2), drift of order Delta, and Slutsky; the active-window functional is the theorem's input, not its output. Corollary 7.5 is a genuine Ito-calculus argument using the Hilbert-space Ito formula, Dambis-Dubins-Schwarz, and the Brownian law of the iterated logarithm. The viability trichotomy follows from the proved vanishing of the normal noise, not from an assumed conclusion. The local-volatility chart, modal reduction, KL expansion, hedge formula, neural-operator feasibility theorem, normalizing-flow likelihoods, and signature ODE are each derived from stated assumptions and standard external tools (Gatheral-Jacquier, Cepa, Slominski, Ramer, Friedman, etc.). The empirical section is explicitly pre-registered: it fixes models, metrics, and thresholds before data contact and reports no estimates, so no fitted input is renamed as a prediction. The only same-author citations are Noguer i Alonso (2026a,b), used in Section 9.1 and Remark 10.5 as an auxiliary taxonomy for covariance repair and market-mode interpretation; these are not invoked to prove the boundary theorem or any central invariance claim. Footnote 1 even warns that the author's three working drafts should not be treated as independent contributions, which reduces rather than creates circularity risk. The paper's own scope limitations - Remark 7.2 limiting the half-law to window activity of positive measure, and Section 16 limiting claims to the compact window, uniformly elliptic local chart, finite-grid likelihoods, and fixed-grid reflected construction - narrow what is proved but do not make any stated result depend on itself.

Assumptions & free parameters 3 free parameters · 8 assumptions · 1 invented entities

The central boundary theorem carries no fitted parameters: the floor m, upper cap M, and weight rho are modeling or design constants, not numbers fit to data. The main assumptions are imported from cited no-arbitrage characterizations and standard infinite-dimensional stochastic analysis; the paper explicitly tiers the additional constraint-qualification and reflected-SDE hypotheses as conditional. The only genuinely new postulated object, the fading-signature field, is equipped with a falsifiable ablation (H6).

free parameters (3)
  • Variance floor m = unspecified, any m>0
    Introduced in Section 3.2 to make the admissible set closed; the boundary theorem holds for every fixed m>0, so the value is not fitted and does not carry the result.
  • Upper variance cap M = unspecified, M>m
    Used in the price-chart band K_{c,m,M} in Section 11.1 to keep the Black-Scholes inverse uniformly regular; a design cap, not fitted.
  • Sobolev weight rho = unspecified
    Choice of H^2 inner product in Section 2.2; Lemma 10.3 shows hedge weights and residual variance are invariant, so the choice is inessential for hedging claims.
assumptions (8)
  • standard math Sobolev embedding, Rellich-Kondrachov compactness, and Riesz representation on the compact rectangle D
    Used in Proposition 2.1 to get C^0 embeddings and kernel sections, and in Proposition 3.1 for weak closedness by compact embedding.
  • domain assumption Gatheral-Jacquier characterization that g[w]>=0 is equivalent to no butterfly arbitrage, and Carr-Madan/Davis-Hobson/Roper characterization of call-price constraints
    Imports the definition of the admissible set K, including the butterfly functional (7), from cited literature rather than proving it; if the equivalence fails on the compact window, K is not the true no-arbitrage set.
  • domain assumption Risk-neutral pricing measure exists and discounted fixed-contract call prices must be local martingales
    Used in Section 6 to separate static feasibility from dynamic no-arbitrage and to state the drift restriction; the paper does not construct the measure.
  • standard math Cylindrical Wiener process on Hilbert space with trace-class covariance and the Hilbert-space Ito formula
    Foundation of Sections 7 and 8 dynamics; standard from Da Prato and Zabczyk but assumed as background.
  • ad hoc to paper Assumption 7.7: smooth Gaussian noise with C^2 sample paths and finite mean C^2 norm, e.g. KL modes in C^{2,alpha} with absolutely summable weighted norms
    Explicitly assumed for the interior-survival Borell-TIS bound in Proposition 7.8 and Corollary 9.2; the boundary half-law does not need it.
  • domain assumption Compact quoted window D with tau_min>0, and Lee wing constraints dominated by butterfly constraints on this window
    Sections 2.1 and 16 state this economic choice; the model concerns the quoted window, not the full infinite domain.
  • ad hoc to paper Constraint qualification for tangent cone equality K=L is assumed [Conditional] via a Robinson-type Slater direction pulled back through B^{-1}
    Remark 4.3; not needed for Theorem 7.3 but needed for full tangent-cone geometry.
  • domain assumption Well-posedness of reflected SDEs on the infinite-dimensional closed convex set K_{c,m,M} is imported from Haussmann-Pardoux and Nualart-Pardoux and tiered [Conditional]
    Proposition 11.1 leaves verification of the structural hypotheses for this specific set to future work.
invented entities (1)
  • Fading-signature field independent evidence
    purpose: Encodes surface history as exponentially weighted iterated integrals so surface dynamics become finite-dimensional and path-dependent (Section 14).
    Hypothesis H6 in Section 15.4 is a falsifiable ablation: a signature-level field with M>=1 must beat the zero-signature ablation on out-of-sample Sobolev RMSE and calm-to-stress transfer; no physical entity is postulated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Mathematics of Volatility Surfaces." pith.science (2026). https://pith.science/paper/AJVUYQBX

@misc{pith2026260805198,
  author       = {Pith},
  title        = {Pith review of: The Mathematics of Volatility Surfaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AJVUYQBX}},
  note         = {Machine review of arXiv:2608.05198}
}
abstract

This paper develops a unified mathematical theory of implied, local, and learned volatility surfaces. Total variance $w_t(k,\tau)=\tau\sigma_t^2(k,\tau)$ is an infinite-dimensional state constrained by positivity, calendar monotonicity, and the butterfly differential inequality. We establish the topology and tangent geometry of this arbitrage set and prove that a nondegenerate Gaussian shock at an active constraint exits with probability tending to one half. Exact invariance therefore requires tangency, reflection, or confinement to an arbitrage-free manifold. We separate this static invariance problem from dynamic no-arbitrage, derive the Musiela maturity-transport identity, and identify the additional fixed-contract martingale restriction. We formulate Hilbert-space dynamics, prove an exact modal reduction with closed-form truncation error, derive Karhunen--Lo\`eve factors, identify the portfolio derivative as a vega field, and obtain the covariance-optimal hedge $\alpha^\ast=(H^\ast C H)^{-1}H^\ast C\nu$. The local-volatility chart completes the geometry: Dupire local variance is the ratio $a=\partial_\tau w/g[w]$ of the calendar and butterfly constraint functionals. Neural operators provide arbitrage-free universal approximation through simplex and cone heads. Normalizing-flow maps then add tractable conditional densities: we derive exact change-of-variables formulae for exponential local-variance flows and invertible stick-breaking price-simplex flows, while stating the quasi-invariance conditions required in genuine function space. Finally, fading-signature fields encode surface history and yield autonomous finite-dimensional controlled dynamics. The result is one framework for representation, dynamics, arbitrage, dimension reduction, likelihood-based learning, simulation, and hedging, together with a falsifiable empirical protocol.

Figures

Figures reproduced from arXiv: 2608.05198 by the authors.

Figure 1
Figure 1. Schematic geometry of the three surface charts. The total-variance chart can be nonconvex [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. The universal Gaussian boundary layer. Panel (a) plots the exact limit [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Spectral truncation and its auditable error. The display uses the normalized reference [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Field hedging as covariance-weighted projection. A deterministic schematic book field [PITH_FULL_IMAGE:figures/full_fig_p026_4.png]
Figure 5
Figure 5. Figure 5: The two approximation axes of Theorem 14.3. The illustration uses normalized λi ∝ i −2 and ε 2 q,M = 0.48e −0.9M to display the bound P i>q λi + ε 2 q,M. Rank q controls omitted surface directions; depth M controls path-memory approximation. The values are illustrative…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 45 canonical work pages

  1. [1]

    Review of Derivatives Research , year =

    Skiadopoulos, George and Hodges, Stewart and Clewlow, Les , title =. Review of Derivatives Research , year =

  2. [2]

    Dynamics of Implied Volatility Surfaces , journal =

    Cont, Rama and da Fonseca, Jos. Dynamics of Implied Volatility Surfaces , journal =. 2002 , volume =

  3. [3]

    A Market Model for Stochastic Implied Volatility , journal =

    Sch. A Market Model for Stochastic Implied Volatility , journal =. 1999 , volume =

  4. [4]

    Mathematical Finance , year =

    Schweizer, Martin and Wissel, Johannes , title =. Mathematical Finance , year =

  5. [5]

    Finance and Stochastics , year =

    Schweizer, Martin and Wissel, Johannes , title =. Finance and Stochastics , year =

  6. [6]

    Local Volatility Dynamic Models , journal =

    Carmona, Ren. Local Volatility Dynamic Models , journal =. 2009 , volume =

  7. [7]

    Kallsen, Jan and Kr. On a. Finance and Stochastics , year =

  8. [8]

    Finance and Stochastics , year =

    Durrleman, Valdo , title =. Finance and Stochastics , year =

Show all 52 references
  1. [9]

    , title =

    Carr, Peter and Madan, Dilip B. , title =. Finance Research Letters , year =

  2. [10]

    Davis, Mark H. A. and Hobson, David G. , title =. Mathematical Finance , year =

  3. [11]

    Roper, Michael , title =

  4. [12]

    Quantitative Finance , year =

    Gatheral, Jim and Jacquier, Antoine , title =. Quantitative Finance , year =

  5. [13]

    , title =

    Lee, Roger W. , title =. Mathematical Finance , year =

  6. [14]

    , title =

    Fengler, Matthias R. , title =. Quantitative Finance , year =

  7. [15]

    Nonparametric Option Pricing under Shape Restrictions , journal =

    A. Nonparametric Option Pricing under Shape Restrictions , journal =. 2003 , volume =

  8. [16]

    and Reisinger, Christoph and Wang, Sheng , title =

    Cohen, Samuel N. and Reisinger, Christoph and Wang, Sheng , title =. Applied Mathematical Finance , year =

  9. [17]

    Journal of Computational Finance , year =

    Hendriks, Sebas and Martini, Claude , title =. Journal of Computational Finance , year =

  10. [18]

    Consistency Problems for

    Filipovi. Consistency Problems for. 2001 , address =

  11. [19]

    2014 , address =

    Da Prato, Giuseppe and Zabczyk, Jerzy , title =. 2014 , address =

  12. [20]

    Set-Valued Analysis , publisher =

    Aubin, Jean-Pierre and Frankowska, H. Set-Valued Analysis , publisher =. 1990 , address =

  13. [21]

    Annali della Scuola Normale Superiore di Pisa, Classe di Scienze , volume =

    Aubin, Jean-Pierre and Da Prato, Giuseppe , title =. Annali della Scuola Normale Superiore di Pisa, Classe di Scienze , volume =

  14. [22]

    Tyrrell and Wets, Roger J.-B

    Rockafellar, R. Tyrrell and Wets, Roger J.-B. , title =. 1998 , address =

  15. [23]

    and Taylor, Jonathan E

    Adler, Robert J. and Taylor, Jonathan E. , title =. 2007 , address =

  16. [24]

    Inventiones Mathematicae , year =

    Borell, Christer , title =. Inventiones Mathematicae , year =

  17. [25]

    , title =

    Bogachev, Vladimir I. , title =. 1998 , address =

  18. [26]

    White Noise Driven Quasilinear

    Nualart, David and Pardoux,. White Noise Driven Quasilinear. Probability Theory and Related Fields , year =

  19. [27]

    and Pardoux,

    Haussmann, Ulrich G. and Pardoux,. Stochastic Variational Inequalities of Parabolic Type , journal =. 1989 , volume =

  20. [28]

    C. Probl. The Annals of Probability , year =

  21. [29]

    Euler's Approximations of Solutions of

    S. Euler's Approximations of Solutions of. Stochastic Processes and their Applications , year =

  22. [30]

    1990 , address =

    Kunita, Hiroshi , title =. 1990 , address =

  23. [31]

    and Silverman, Bernard W

    Ramsay, James O. and Silverman, Bernard W. , title =. 2005 , address =

  24. [32]

    Inference for Functional Data with Applications , publisher =

    Horv. Inference for Functional Data with Applications , publisher =. 2012 , address =

  25. [33]

    Simulation of Arbitrage-Free Implied Volatility Surfaces , journal =

    Cont, Rama and Vuleti. Simulation of Arbitrage-Free Implied Volatility Surfaces , journal =

  26. [34]

    Risks , year =

    Cuchiero, Christa and Khosrawi, Wahid and Teichmann, Josef , title =. Risks , year =

  27. [35]

    Quantitative Finance , year =

    Gatheral, Jim and Jaisson, Thibault and Rosenbaum, Mathieu , title =. Quantitative Finance , year =

  28. [36]

    Quantitative Finance , year =

    Guyon, Julien and Lekeufack, Jordan , title =. Quantitative Finance , year =

  29. [37]

    Noguer i Alonso, Miquel , title =

  30. [38]

    arXiv preprint arXiv:2606.28869 , year =

    Noguer i Alonso, Miquel , title =. arXiv preprint arXiv:2606.28869 , year =

  31. [39]

    Journal of Machine Learning Research , volume =

    Kovachki, Nikola and Li, Zongyi and Liu, Burigede and Azizzadenesheli, Kamyar and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , title =. Journal of Machine Learning Research , volume =

  32. [40]

    Nature Machine Intelligence , volume =

    Lu, Lu and Jin, Pengzhan and Pang, Guofei and Zhang, Zhongqiang and Karniadakis, George Em , title =. Nature Machine Intelligence , volume =

  33. [41]

    Annals of Mathematics , volume =

    Hambly, Ben and Lyons, Terry , title =. Annals of Mathematics , volume =

  34. [42]

    Journal of Machine Learning Research , volume =

    Chevyrev, Ilya and Oberhauser, Harald , title =. Journal of Machine Learning Research , volume =

  35. [43]

    Risk , volume =

    Dupire, Bruno , title =. Risk , volume =

  36. [44]

    Friedman, Avner , title =

  37. [45]

    Gatheral, Jim , title =

  38. [46]

    and Ingersoll, Jonathan E

    Cox, John C. and Ingersoll, Jonathan E. and Ross, Stephen A. , title =. Econometrica , volume =

  39. [47]

    Mathematical Finance , volume =

    Bj\"ork, Tomas and Svensson, Lars , title =. Mathematical Finance , volume =

  40. [48]

    and St rmer, Erling , title =

    Powers, Robert T. and St rmer, Erling , title =. Communications in Mathematical Physics , volume =

  41. [49]

    Proceedings of the 32nd International Conference on Machine Learning , series =

    Rezende, Danilo Jimenez and Mohamed, Shakir , title =. Proceedings of the 32nd International Conference on Machine Learning , series =

  42. [50]

    Journal of Machine Learning Research , volume =

    Papamakarios, George and Nalisnick, Eric and Rezende, Danilo Jimenez and Mohamed, Shakir and Lakshminarayanan, Balaji , title =. Journal of Machine Learning Research , volume =

  43. [51]

    Journal of Functional Analysis , volume =

    Ramer, Roald , title =. Journal of Functional Analysis , volume =

  44. [52]

    Revuz, Daniel and Yor, Marc , title =

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.