DSINC beyond Heston: the 3/2 and other non-affine stochastic-volatility models
summary
The gist
This paper extends the damped-sinc (DSINC) method for American options beyond the affine Heston class to general one-factor stochastic-volatility models…
In short
This work extends a method for pricing American options, called damped-sinc (DSINC), beyond standard models like Heston to more general one-factor stochastic-volatility models. The core finding is that a key mathematical identity, the collapse identity, still works. This allows for closed-form solutions in specific non-affine models like the 3/2 and SABR, and provides a robust numerical approach for others.
Key concepts
- DSINC Method
- This is a pricing technique used for American options that extends the standard damped-sinc method. It's designed to handle complex stochastic volatility models by reducing difficult early-exercise integrals into simpler spectral integrals, making pricing more tractable.
- Collapse Identity
- A central result showing that the early-exercise integral can be simplified into a single spectral integral. This identity is fundamental and remains valid across all one-factor stochastic-volatility models, regardless of whether they are affine or non-affine.
- One-Factor SVM
- A class of stochastic volatility models where the volatility is driven by a single underlying factor. The paper proves that the DSINC collapse identity holds for every model in this class, even those outside the standard affine Heston framework.
- Volatility Integral Computation
- When closed-form solutions are unavailable, this describes a numerical strategy. Instead of using a grid in log-price, it solves one backward equation per frequency to compute the necessary volatility integral directly. This method is flexible and handles models where standard transforms are missing.
Terminology used across episodes
This episode discusses
- DSINC beyond Heston: the 3/2 and other non-affine stochastic-volatility models · Paper Radio
- Valuing American options and Flexible Forwards contracts in time-dependent models
- Diagonal Frog: High-order positivity-preserving FD schemes for anisotropic Fokker-Planck equations
- Efficient valuation method for the SABR model
- Asymptotic Implied Volatility at the Second Order with Application to the SABR Model
The paper
DSINC beyond Heston: the 3/2 and other non-affine stochastic-volatility models · Read on arXiv
Andrey Itkin
New York University
Transcript
Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "DSINC beyond Heston".
Warren: Here's the gist of the paper. This paper extends the damped-sinc (DSINC) method for American options beyond the affine Heston class to general one-factor stochastic-volatility models, showing that its collapse identity remains valid and providing closed-form solutions for specific non-affine models like the 3/2 and SABR Collapse Identity Extension The central result of the method is a collapse identity which reduces the early-exercise integral to a single spectral integral, and this identity follows from the spectral representation and does not use affine dynamics The collapse identity holds for every one-factor SVM, where the sum over the local basis reduces exactly to the characteristic function conditional on the terminal volatility, evaluated on a reflected contour The fundamental strip 0 < χ < 1 is available whenever the discounted spot is a nonnegative local martingale, with one residue term Model Specific Closed Forms The paper details specific closed-form solutions for different models, such as the 3/2 model where the volatility integral is available in closed form through Kummer functions For SABR with zero correlation, the premium integrand over the exercise region has a closed form in the Hankel representation, and each Bessel frequency requires one equation in the volatility Numerical Implementation and Tests The authors describe a fixed-frequency numerical solution for general models where a practical closed-form transform is unavailable, and this approach computes the required volatility integral directly by solving one backward equation per frequency, without a grid in the log-price Tests on the 3/2, SABR, GARCH and lognormal-volatility models show agreement with fine finite-difference benchmarks to within one basis point of the strike Boundary Representation and Cost The method utilizes a nodal boundary representation where the exercise boundary is represented by its point values at the volatility grid nodes, and this avoids the band-mass approximation The computational cost for DSINC is comparable to or lower than that of our finite-difference implementation, except for the slower Hankel implementation of uncorrelated SABR The conclusion is that the collapse identity holds for every one-factor SVM satisfying the stated assumptions, and for a full-range facet, the remaining volatility integral is available in closed form in the Heston model when the exercise boundary is linear in the variance, and in the 3/2 model when it is linear in the logarithm of the variance The method provides a flexible framework for extending DSINC to models outside the affine class The fixed-frequency solution in (x, y) for any ρ is a fixed-frequency solution for Eq.
Ruth: (1), which makes facets unnecessary Pruning redundant reasoning tokens The collapse identity of DSINC is a statement about damped-sinc interpolation, and it holds for every one-factor SVM Outside the affine class, what changes is the admissible damping and the evaluation of the volatility integral The collapse identity retains its Fourier form for transition laws satisfying the stated integrability assumptions How the training signal is built The Itô coordinate makes the endpoint phase explicit and supplies a natural coordinate for facets In the 3/2 model, the reciprocal-CIR bridge gives the full-range facet transform through Kummer functions on an explicit joint moment domain, including the admissible critical tilt Where the gains come from The nodal fixed-frequency implementation integrates over terminal volatility directly and avoids the band-mass approximation The Hankel method prices American SABR Puts to within 8×10−6 of the fine grid, but in our implementation it is slower than the two-dimensional solver Pruning redundant reasoning tokens The collapse identity of DSINC is a statement about damped-sinc interpolation, and it holds for every one-factor SVM Outside the affine class, what changes is the admissible damping and the evaluation of the volatility integral Where the gains come from
Paper summary: Ruth: So this paper is about extending that damped-sinc method we use for American options, which usually only works for the affine Heston model, out to more general one-factor stochastic-volatility models. It claims that the collapse identity, which simplifies the early exercise integral into a single spectral integral, still holds true for these wider classes of models.
Warren: What matters here is that they aren't just sticking to Heston dynamics; they are looking at broader things like the three/two model and uncorrelated SABR <ref:2610.28847#pg1>. It’s about showing that this core identity doesn't break when you move outside the affine class, which is a big step for applicability.
Iris: I see it focusing on damping between zero and one being admissible only when the discounted spot is a nonnegative local martingale; that sounds like a crucial constraint we need to keep in mind for any practical application of this method.
Ruth: Exactly. The paper says that an Itô transformation of the volatility factor helps factorize the conditional characteristic function, which lets them identify a natural coordinate for approximating that exercise boundary. That’s how they get a handle on where the option should expire.
Warren: And for specific models, they actually found some closed-form solutions; for instance, the three/two model has these transforms available using Kummer functions, and uncorrelated SABR has a Bessel representation <ref:2610.28847#pg1>. That’s solid math to have when you're trying to price something quickly.
Iris: That points toward the next part of their work, where they discuss how this translates into actual pricing formulas for those specific cases, which is key because a closed form is much better than a numerical approximation.
Ruth: Right. And if those specific transforms aren't available—which is often the case in real-world scenarios—they have a direct way to compute the required volatility integral by solving one backward equation per frequency without needing a grid in the log-price.
Warren: That’s smart because it keeps the computational cost manageable, especially compared to what we see with some of our finite-difference implementations, though they note that uncorrelated SABR is slower using their Hankel implementation.
Iris: I'm interested in how they handle the boundary itself; they use a nodal boundary representation where the exercise boundary is just represented by its point values at the volatility grid nodes, which avoids that band-mass approximation we sometimes have to deal with.
Ruth: That’s a practical win for speed and accuracy when modeling the actual exercise region, and it seems like this approach also accommodates correlated SABR using a known change of volatility coordinate.
Warren: So they've shown that the method works across several models—the three/two SABR, GARCH, and lognormal-volatility—and they confirm agreement with fine finite-difference benchmarks to within one basis point of the strike <ref:2610.28847#pg1>. That’s a good validation check.
Iris: It’s important to remember the limitation mentioned: they state that when a practical closed-form transform isn't available, the method computes things directly by solving backward equations, and that is what you have to live with if you can't find an explicit formula for that model.
Ruth: The paper says the collapse identity remains valid for every one-factor SVM under those stated integrability assumptions, which is a very broad claim about its robustness across different dynamics. That suggests a general framework here.
Warren: For someone just listening, the big idea is that this method gives us a way to handle complex volatility dynamics without immediately needing a perfect closed-form solution for every single model out there.
Iris: And they also point out that the affine structure is only used afterwards, when they evaluate the remaining variance integral through tilted variance transforms and a Möbius propagator, which shows where the affine structure still plays a role.
Ruth: That leads us into what this means for us; it suggests a path forward for pricing these options in models that aren't strictly Heston, giving us flexibility without losing the underlying mathematical machinery.
Warren: So to wrap up this summary of "DSINC beyond Heston: the three/two and other non-affine stochastic-volatility models," the core finding is that the collapse identity holds for general one-factor SVMs, and they provide explicit transforms for specific cases like three/two and SABR <ref:2610.28847#pg1,DSINC beyond Heston: the 3/2 and other non-affine stochastic-volatility>.
Iris: The implication is that we have a more flexible toolkit for handling volatility dynamics when dealing with American options outside the standard affine class, provided we respect those damping conditions.
Ruth: And it gives us concrete tools—like the Kummer functions or the Bessel representation—for those specific non-affine models, which are often what matter most in practice.
Warren: It sounds like they’ve provided a way to price these options that is both mathematically sound across a wide range of models and computationally feasible for many scenarios.
Conclusion: Ruth: So we’ve been looking at how this damped-sinc method extends beyond the standard Heston model to these more general models, specifically the three/two and SABR cases, and now we're coming to what this paper actually claims.
Warren: The title itself tells you it's about pushing that damped-sinc method past the affine class of volatility dynamics. It’s not just for Heston anymore.
Iris: The authors are showing that the core collapse identity, which simplifies the early exercise integral down to a single spectral integral, still holds true for every one-factor stochastic-volatility model under certain assumptions.
Ruth: So what does that mean for us? It means this fundamental math doesn't break when you move into these more complex volatility structures.
Warren: And they aren't just proving it works generally; they show specific closed forms exist for things like the three/two model using Kummer functions, and uncorrelated SABR has a Bessel representation. That’s concrete stuff for pricing.
Iris: Before we get into how to use these transforms, there are some important conditions they set—like damping between zero and one being okay if the discounted spot is a nonnegative local martingale.
Ruth: Right, so the applicability of the method depends on those specific mathematical rules about the dynamics of that spot price.
Warren: If you don't have those rules, you can’t use this collapse identity to simplify things in that way. It’s a hard constraint on when the trick works.
Iris: And they also describe a numerical path for when those closed forms aren't available—they solve one backward equation per frequency without needing a grid in the log-price.
Ruth: That keeps the computational cost manageable, which is important for trading strategies where speed matters.
Warren: But we have to remember they flag that uncorrelated SABR is slower with their Hankel implementation compared to other solvers. So there’s always a trade-off there between speed and model complexity.
Iris: The paper concludes by validating these ideas on the three/two SABR, GARCH, and lognormal-volatility models against fine finite-difference benchmarks up to one basis point of the strike. That gives them good confidence in the results.
Ruth: So this paper essentially provides a more flexible toolkit for pricing American options when you're outside the Heston world.
Warren: It’s about giving us tools that work across different types of volatility behavior, even if you don't have a neat closed formula for every single one.
Iris: And they show that this flexibility comes with clear conditions on damping and specific representations for certain models.
More episodes
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- The day in conversation for 2026-11-18
- Daily Summary for 2026-11-18
- 2601.05274-On the use of case estimate and transactional payment data in neural networks for individual loss reserving
- 2009.00868-Loss-Given-Default Modeling by Post-Last Passage Time Process
- 2606.30070-Financial Resilience Evaluation: From Conditional Expectations to Dynamic Convex Risk Measures
- 2212.12687-Measuring price impact and information content of trades in a time-varying setting
- Daily Summary for 2026-11-17
- 2401.08323-Dynamic portfolio selection under generalized disappointment aversion