Polynomial-Preserving Processes and Volatility Modelling in Discrete Time

arXiv:2611.10928 · q-fin.MF, econ.EM · Submitted 2026-11-07 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "Polynomial-Preserving Processes and Volatility Modelling in Discrete Time".

Warren: Here's the gist of the paper. This paper develops a general discrete-time polynomial-preserving framework for volatility modeling and derivative pricing, designed to provide exact conditional moments of arbitrary order and serial cross-moments of variance and return processes over any finite horizon using finite-dimensional matrix representations.

Ruth: This analytical tractability is maintained while accommodating flexible leverage and variance-of-variance dynamics, as well as non-Gaussian innovations.

Warren: The core concept is polynomial preservation: the conditional expectation of any polynomial in the future state must be a polynomial of no higher degree in the current state.

Ruth: The framework introduces two novel classes of stochastic volatility models that nest widely used discretetime specifications: the variance-polynomial hybrid GARCH–SV (VPGARCHSV) and the jointly polynomial hybrid GARCH–SV (JPGARCH-SV).

Title and authors: Ruth: So we're looking at this paper now called "Polynomial-Preserving Processes and Volatility Modelling in Discrete Time," and the authors are Maciej Augustyniak, Alexandru Badescu, and Zekun Zhang. Basically, they've put together a framework that lets you calculate exact conditional moments of any order for variance and return processes over any finite time horizon using these finite dimensional matrix representations.

Warren: It sounds like they're trying to solve the tractability versus flexibility trade-off in volatility modeling, which is usually a big hurdle when you try to get complex features like leverage effects or variance-of-variance dynamics into your models without losing the ability to calculate things exactly.

Iris: Right, so the main idea here is polynomial preservation, which means that for any polynomial you put in terms of the future state, its conditional expectation will always be a polynomial of no higher degree in your current state. That's a powerful constraint on what kind of dynamics you can have.

Ruth: Exactly. They introduce two specific models that fit into this framework: the variance-polynomial hybrid GARCH–SV, or VPGARCHSV, and the jointly polynomial hybrid GARCH–SV, or JPGARCH-SV. These are designed to nest widely used specifications in finance.

Warren: And they claim these models can handle non-Gaussian innovations and even distinct, correlated shocks between returns and variance, which is something a lot of standard SV models struggle with when you want exact results.

Iris: That’s the key part for me, because they establish weak convergence to continuous time polynomial diffusions or jump diffusions under two different scaling regimes. That links their discrete time moment recursions directly to those continuous time matrix exponential formulas, which is a big connection.

Ruth: And they don't just stop at theory; they provide exact conditional moments of cumulative returns and squared conditionally demeaned returns for both VPGARCHSV and JPGARCH-SV, which should be very useful for derivative valuation without needing simulation errors.

Warren: They also derived diffusion limits under the physical measure, showing that the limiting variance process is polynomial, specifically mentioning a limiting covariance matrix factorization involving Brownian motion. That gives them a concrete target for what these models are converging to in the continuous time world.

Title and authors: Iris: And they characterized the risk-neutral dynamics too, noting that those discrete time moment recursions converge to those same matrix exponential formulas of the limiting processes under specific conditions on parameters like lambda and eta.

Ruth: Moving into the practical side, they show empirical results using S andP five hundred returns and options spanning from one thousand nine hundred ninety-six to two thousand twenty-three where these specifications improve return fit by about eight to forty one percent compared to their nested benchmarks under both Gaussian and NIG innovations <ref:2611.10928#pg1,S&P 500 returns and>.

Warren: That range of improvement depends on the variance dynamics and the innovation distribution, which shows that the value of heavier tails really matters when you're looking at how you model volatility.

Iris: They also detail empirical performance metrics, showing that VPGARCH achieves the lowest mean weekly implied volatility root mean squared errors under both distributions, and JPGARCH performs particularly well over much of the pre-two thousand twenty period under NIG innovations <ref:2611.10928#pg1,mean weekly implied volatility root mean squared errors>.

Ruth: That's interesting because it suggests that depending on what you expect from the tails—Gaussian versus NIG—you might choose a different model specification to get the best practical fit.

Warren: If we look at risk-neutral measures, they found specific conditions on parameters, like lambda and eta, needed for polynomial preservation when innovations are correlated, and for NIG innovations specifically, they characterize polynomial preservation through Hermite extensions.

Iris: And under the risk-neutral measure Q with correlated innovations, they found that for VPGARCHSV with a correlation parameter greater than zero, polynomial preservation requires a cancellation of a non-polynomial leverage term when lambda plus one half is not zero.

Ruth: That’s quite specific. It means there are very precise conditions on the parameters required to maintain this polynomial structure under the risk-neutral world.

Warren: And for JPGARCHSV, joint polynomiality requires the product of rho and eta to be zero under the physical measure, which is a different restriction from what they found for VPGARCHSV.

Iris: For JPGARCHSV, its joint operator is polynomial if and only if xi two of x times v sub perp JP of x is affine under the physical measure. That gives us a way to characterize these joint polynomial structures beyond just affine specifications.

Title and authors: Ruth: So, to wrap up this paper "Polynomial-Preserving Processes and Volatility Modelling in Discrete Time," the main implication is that we have a route to characterizing the polynomial moment structure of these complex volatility models and developing alternatives beyond just simple affine ones.

Warren: It gives us tools for exact calculation of moments and serial cross-moments without resorting to simulation errors, which is a huge computational saving for anyone doing derivative pricing.

Iris: And we have established how these discrete time models link up to continuous time processes under specific scaling regimes, which opens up new ways to connect finite difference methods with continuous stochastic volatility theory.

Ruth: The empirical results show that these models offer tangible improvements in fit and error reduction for things like S andP five hundred returns, depending on the specific dynamics and innovations you're dealing with <ref:2611.10928#pg1>.

Warren: It seems like the choice between VPGARCHSV and JPGARCH-SV isn't just about which one fits best overall, but depends on whether you prioritize fitting under Gaussian or NIG innovation scenarios.

Iris: What they didn't fully cover in this work is the exact nature of those non-polynomial leverage terms when those conditions aren't met, so the limitation is that their framework relies heavily on these specific structural requirements for polynomial preservation to hold.

Ruth: Exactly, so while we get exact moments under these specific structures, we still need to understand what happens when the underlying dynamics don't satisfy those exact algebraic conditions for polynomiality.

Warren: So, in summary of "Polynomial-Preserving Processes and Volatility Modelling in Discrete Time," they give us a robust framework for finding tractable models that retain high flexibility, provided you stick to the polynomial preservation constraints.

Iris: It’s a solid piece of work because it bridges the gap between complex, flexible volatility dynamics and the exact moment calculations we need for pricing.

Ruth: It really does make you think about how we can build more sophisticated models that aren't just approximations but have verifiable structural properties like polynomiality.

Warren: We'll see if this framework helps us push the boundaries on modeling asset returns in the coming months.

The paper's summary: Ruth: So, to recap, these authors are proposing a way to build volatility models in discrete time that keep a specific mathematical structure—polynomial preservation—so you can calculate exact moments for any order without needing simulations or approximations.

Warren: Right, they’re saying this polynomial preservation thing lets them handle things like variance dynamics and even non-Gaussian innovations while still being able to get those exact conditional means of returns over any time horizon.

Iris: What that means is they've got these two specific model classes, the VPGARCHSV and the JPGARCH-SV, which are designed to fit into what we already use in finance, but they have this strong mathematical property where if you take a polynomial function of your future state and find its expected value now, it stays a polynomial of the same degree in your current state.

Ruth: It’s about that exact calculation. Imagine you need the conditional expectation of something complicated involving three steps into the future; this framework says that instead of getting stuck with an approximation, you can get an exact polynomial answer based only on what's happening right now.

Warren: That sounds computationally heavy, though, because they’re using finite dimensional matrix representations to handle all those moments and cross-moments for variance and return processes over any finite horizon.

Iris: It is intensive math, but the paper shows they can link these discrete time recursions to continuous time polynomial diffusions under specific scaling regimes, which is a huge connection between how we model things on a computer versus how the actual market moves in real time.

Ruth: And then they test it. They look at S andP five hundred returns and options spanning over twenty-seven years and found these models actually improve the fit of the returns by about eight to forty one percent compared to their existing benchmarks, depending on whether you're looking at Gaussian or NIG innovations.

Warren: That range is pretty wide. So it suggests that which specification—the VPGARCHSV or the JPGARCH-SV—you choose really depends on whether you anticipate heavier tails in the data, like those found with NIG innovations versus Gaussian ones.

Iris: They also looked at the risk-neutral world, where things get a bit trickier because of correlations between returns and variance. They showed that for polynomial preservation to hold under those conditions with correlated innovations, there are very specific requirements on parameters like lambda and eta.

Ruth: It’s a tight constraint. And they found some interesting conditions—like needing a cancellation of a non-polynomial leverage term if the drift isn't just right—to keep that polynomial structure intact when moving to the risk-neutral measure.

Warren: So, what this means for us on the trading side is that if we’re running complex strategies, we need to be careful about how much flexibility we want in our volatility inputs because this framework puts a hard mathematical boundary on what kind of dynamics actually allows for these exact moment calculations.

Iris: And they point out their limitation too. The analysis relies heavily on these specific algebraic conditions for polynomial preservation to work perfectly; if the underlying dynamics don't meet those exact requirements, you lose that nice clean polynomial structure.

Ruth: So this paper gives us a very precise tool to build models that are both flexible enough for real-world data and rigid enough to give us exact answers on moments, but we still need more research on what happens when those structural assumptions break down.

The paper's improvements: Ruth: We’re looking at how this work actually makes things better than what we have now, and the main improvement is that they show how these models can be built to handle those tricky correlations between returns and volatility shocks.

Warren: They've introduced two novel classes of models, the VPGARCHSV and the JPGARCH-SV, which are specifically structured to accommodate non-Gaussian innovations and distinct shocks that aren't perfectly correlated, something standard models often struggle with.

Iris: That means they’ve identified restrictions on conditional means and innovation loadings that allow for polynomial preservation of variance alone or jointly with returns, which is a big step in flexibility.

Ruth: And this leads to the next improvement: they show how these discrete time models can be linked to continuous time polynomial diffusions under certain scaling regimes, meaning they’re not just theoretical constructs but have a way to connect them back to continuous stochastic volatility theory.

Warren: That connection is key for us on the trading side because it gives us a bridge from the discrete moment recursions—what we calculate step-by-step—to those continuous time matrix exponential formulas that describe the real process.

Iris: And they provide exact conditional moments of cumulative returns and squared conditionally demeaned returns for both VPGARCHSV and JPGARCH-SV, which is a huge improvement because it means you get those exact numbers without having to run a simulation or deal with approximation errors.

Ruth: That’s valuable for derivative pricing because it removes that layer of uncertainty from the moment calculations, giving us something much more solid than we usually get.

Warren: But there’s a catch on how good these results are. They show empirical improvements on things like S andP five hundred returns over twenty-seven years, where the error reduction is between eight and forty one percent compared to what you get from your standard nested benchmarks, depending on the innovation distribution.

Iris: That range tells us that the benefit of using this framework really depends on whether you expect heavier tails in your data, like with NIG innovations versus Gaussian ones.

Ruth: So the implication is that if you’re dealing with real-world data that has heavy tails, these models can give you a significant edge in fitting the return process and reducing volatility errors.

Warren: And they also found that the VPGARCHSV tends to achieve lower mean weekly implied volatility root mean squared errors under both distributions, while the JPGARCH-SV performs particularly well over much of the pre-two thousand twenty period when using NIG innovations.

Iris: That’s a practical piece of information for model selection; it suggests you might pick one specification over the other based on what kind of innovation noise you think is dominating your data.

Ruth: What they did not fully address is exactly what happens when those structural requirements for polynomial preservation aren't met, so the method only works perfectly when the underlying dynamics satisfy those very specific algebraic conditions.

Warren: That means we still have to be careful; this framework is powerful, but it’s a tool for a specific class of volatility dynamics.

Iris: Exactly, so while we get exact moments under these specific structures, we still need more research on what happens when the underlying dynamics don't meet those exact algebraic conditions for polynomiality to hold.

Ruth: It’s a solid piece of work because it bridges the gap between complex, flexible volatility dynamics and the exact moment calculations we need for pricing, but it definitely leaves room for further investigation into those edge cases.

Conclusion: Ruth: So we’re wrapping up this discussion on "Polynomial-Preserving Processes and Volatility Modelling in Discrete Time" by Augustyniak, Badescu, and Zhang. Basically, they gave us a framework to calculate exact conditional moments of any order for variance and return processes over any finite time horizon using these matrix representations.

Warren: They showed how this polynomial preservation concept allows them to handle flexible leverage and variance-of-variance dynamics while still keeping the math tractable through these finite dimensional matrix representations.

Iris: The big point is that they establish a link between this discrete time framework and continuous time polynomial diffusions or jump diffusions under two different scaling regimes, which is really connecting finite difference methods to continuous stochastic volatility theory.

Ruth: They also provided exact conditional moments of cumulative returns and squared conditionally demeaned returns for both the VPGARCHSV and JPGARCH-SV models, which is a big win for derivative valuation because it means no simulation errors there.

Warren: The empirical results show that these specifications improve return fit by about eight to forty one percent relative to their nested benchmarks under both Gaussian and NIG innovations, depending on the variance dynamics you choose.

Iris: They also characterized the risk-neutral dynamics, showing how those discrete time moment recursions converge to the matrix exponential formulas of the limiting processes under specific parameter conditions.

Ruth: It really shows that even in complex models, we can find these exact moment structures if we impose enough constraints, like polynomial preservation.

Warren: So this paper gives us a robust way to build models with high flexibility that still have verifiable structural properties like polynomiality for calculating moments without the usual computational burden.

Iris: What they didn't fully cover is the exact nature of those non-polynomial leverage terms when those specific conditions for polynomial preservation don't hold.

Ruth: That means while this framework is great for finding exact results under certain assumptions, we still need to understand what happens when the underlying dynamics break those rules.

Warren: It’s a lot of heavy math to keep track of all those scaling regimes and parameter conditions, but the payoff is getting exact moments without relying on costly simulations.

Iris: Overall though, this work provides a clear route to characterizing the polynomial moment structure of these complex volatility models and developing alternatives beyond just simple affine specifications.

Ruth: This paper "Polynomial-Preserving Processes and Volatility Modelling in Discrete Time" gives us a powerful tool for exact calculation of moments and serial cross-moments without resorting to simulation errors.

Warren: It seems like the choice between the VPGARCHSV and JPGARCH-SV isn't just about which one fits best overall, but depends on whether you prioritize fitting under Gaussian or NIG innovation scenarios.

Iris: I think this work is important because it gives us a way to connect these discrete time models to continuous time processes, which opens up new ways to link finite difference methods with continuous stochastic volatility theory.

Maciej Augustyniak, Alexandru Badescu, Zekun Zhang

University of Montreal · University of Calgary

q-fin.MF, econ.EM

Submitted: 2026-11-07

Updated: 2026-11-07

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 91/100

The gist: This paper develops a general discrete-time polynomial-preserving framework for volatility modeling and derivative pricing…

Key concepts

Polynomial Preservation
This core concept means that the conditional expectation of any polynomial in the future state of a system must result in a polynomial of no higher degree when expressed in terms of the current state. This property ensures that complex moment calculations remain analytically manageable, avoiding computational errors associated with high-order approximations.
VPGARCHSV and JPGARCH-SV
These are two new classes of stochastic volatility models designed to be 'polynomial-preserving.' They extend standard GARCH–SV models by incorporating polynomial dynamics for variance. They are characterized by their ability to maintain this preservation property, allowing them to accurately model complex return and variance relationships under various innovation types.
Weak Convergence
This concept connects the discrete-time models developed in the paper to continuous-time stochastic processes. It shows that as the time steps become infinitesimally small, the discrete moment recursions converge to formulas derived from continuous-time matrix exponentials, providing a bridge between discrete calculations and established continuous-time theory.

Terminology

Summary

This paper develops a general discrete-time polynomial-preserving framework for volatility modeling and derivative pricing, designed to provide exact conditional moments of arbitrary order and serial cross-moments of variance and return processes over any finite horizon using finite-dimensional matrix representations. p. 0

Where the paper says it

Polynomial-Preserving Processes and Volatility Modelling in Discrete Time Maciej Augustyniak∗ Alexandru Badescu† Zekun Zhang‡ Abstract This paper develops a discrete-time framework for volatility modelling and derivative pricing based on polynomial preservation. The framework provides exact moments of arbitrary order and serial cross-moments of variance and return processes over any finite horizon through finite-dimensional matrix representations.

This analytical tractability is maintained while accommodating flexible leverage and variance-of-variance dynamics, as well as non-Gaussian innovations. p. 0
Where the paper says it

The framework provides exact moments of arbitrary order and serial cross-moments of variance and return processes over any finite horizon through finite-dimensional matrix representations. This analytical tractability is maintained while accommodating flexible leverage and variance-of-variance dynamics and non-Gaussian innovations.

The core concept is polynomial preservation: the conditional expectation of any polynomial in the future state must be a polynomial of no higher degree in the current state. p. 1
Where the paper says it

Relative to their nested benchmarks, the proposed models improve the fit to returns and a large panel of equity index options. Polynomial preservation means that the conditional expectation of any polynomial in the future state is a polynomial of no higher degree in the current state.

The framework introduces two novel classes of stochastic volatility models that nest widely used discretetime specifications: the variance-polynomial hybrid GARCH–SV (VPGARCHSV) and the jointly polynomial hybrid GARCH–SV (JPGARCH-SV). p. 0

Where the paper says it

We introduce two novel classes of stochastic volatility models that nest widely used discretetime specifications and study their statistical properties. Under two distinct scaling regimes, we establish weak convergence to polynomial diffusions and jump-diffusions and convergence of polynomial moments, linking discrete-time moment recursions to continuous-time matrixexponential formulas.

These models are characterized by polynomial preservation, meaning that the conditional expectation of any polynomial in the future state is a polynomial of no higher degree in the current state. p. 1
Where the paper says it

Relative to their nested benchmarks, the proposed models improve the fit to returns and a large panel of equity index options. Polynomial preservation means that the conditional expectation of any polynomial in the future state is a polynomial of no higher degree in the current state.

The framework establishes weak convergence to continuous-time polynomial diffusions and jump-diffusions under two distinct scaling regimes, linking discrete-time moment recursions to continuous-time matrix-exponential formulas. p. 0
Where the paper says it

We introduce two novel classes of stochastic volatility models that nest widely used discretetime specifications and study their statistical properties. Under two distinct scaling regimes, we establish weak convergence to polynomial diffusions and jump-diffusions and convergence of polynomial moments, linking discrete-time moment recursions to continuous-time matrixexponential formulas.

The paper provides exact conditional moments of cumulative returns under JPGARCH-SV and VPGARCH-SV with independent innovations at every finite order and horizon. p. 0

Where the paper says it

Polynomial-Preserving Processes and Volatility Modelling in Discrete Time Maciej Augustyniak∗ Alexandru Badescu† Zekun Zhang‡ Abstract This paper develops a discrete-time framework for volatility modelling and derivative pricing based on polynomial preservation. The framework provides exact moments of arbitrary order and serial cross-moments of variance and return processes over any finite horizon through finite-dimensional matrix representations.

Furthermore, it derives diffusion limits for these models under the physical measure, showing that the limiting variance process is polynomial, exemplified by a limiting covariance matrix factorization involving BM(x). p. 2
Where the paper says it

, 2026), we derive two limiting regimes under the physical measure. Both regimes preserve polynomial variance dynamics and yield joint polynomiality in log price and variance for JPGARCH-SV.

The analysis also characterizes risk-neutral dynamics, noting that the discrete-time moment recursions converge to the matrix-exponential formulas of the limiting processes. p. 2
Where the paper says it

For the corresponding polynomial states, moment recursions converge to continuous-time matrix-exponential formulas. These results provide a new discrete-time construction of continuous-time stochastic volatility models with polynomial variance dynamics.

Connections between discrete and continuous time are established, where discrete observations of a continuous-time polynomial process form a polynomial state-space model, and increment cumulants are expressed as finite polynomials in tau. p. 1
Where the paper says it

Only recently have Kallsen and Richert (2026a,b) developed a general discrete-time polynomial state-space framework for filtering and estimation. Their framework includes the standard symmetric GARCH(1,1) model as a special case, but their primary focus is on inference rather than on characterizing the dynamics that preserve polynomiality in discrete-time SV models with leverage effects.

Empirical analysis using S&P 500 returns and options spanning over 27 years shows that both specifications improve return fit and reduce mean weekly implied volatility root mean squared errors by approximately 8–41% relative to their nested benchmarks under both Gaussian and NIG innovations. p. 0

Where the paper says it

Under two distinct scaling regimes, we establish weak convergence to polynomial diffusions and jump-diffusions and convergence of polynomial moments, linking discrete-time moment recursions to continuous-time matrixexponential formulas. An empirical analysis of S&P 500 returns and a large panel of options spanning 1996–2023 shows that the proposed specifications improve return fit and reduce mean weekly implied volatility root mean squared errors by approximately 8–41% relative to their nested benchmarks, depending on the variance dynamics and innovation distribution.

VPGARCH achieves the lowest mean weekly pricing error under both distributions, while JPGARCH performs particularly well over much of the pre-2020 period under NIG innovations. p. 3
Where the paper says it

VPGARCH achieves the lowest mean weekly pricing error under both distributions. NIG innovations improve option fit much more for HNGARCH and JPGARCH than for NGARCH and VPGARCH, suggesting that the value of heavier innovation tails depends on the variance dynamics.

The paper also details empirical performance metrics, noting that VPGARCH reduces mean weekly IVRMSE and every reported error quantile relative to nested benchmarks under both innovation distributions. p. 3
Where the paper says it

Both proposed specifications substantially improve the fit relative to their nested benchmarks. JPGARCH reduces the mean weekly implied volatility root mean squared error (IVRMSE) relative to HNGARCH by approximately 41% under Gaussian innovations and 8% under NIG innovations; the corresponding gains of VPGARCH over NGARCH are 11% and 14%, respectively.

The analysis delves into polynomial preservation under different measures, including the risk-neutral measure Q(n). p. 3

Where the paper says it

Appendix B studies the pricing kernel and risk-neutral preservation. 2 Polynomial-Preserving Hybrid GARCH–SV Models This section characterizes hybrid GARCH–SV dynamics with analytically tractable moments under the physical measure.

It provides specific conditions on parameters (lambda(tau), eta(tau), etc. p. 2
Where the paper says it

3 Such higher-order and cross-date moments can be cumbersome to derive even for standard GARCH specifications. For the proposed classes, the model-specific task is to construct the one-step transition matrices from the model parameters and innovation moments; the same matrix formulas then apply across specifications and forecast horizons.

) required for polynomial preservation under Q(n) for VPGARCH-SV and JPGARCH-SV when innovations are correlated. p. 2
Where the paper says it

The framework accommodates non-Gaussian innovations and distinct, correlated return and variance shocks. In the Gaussian case, the first class, variance-polynomial hybrid GARCH–SV (VPGARCHSV), extends the NGARCH dynamics of Engle and Ng (1993), used for option valuation, for example, by Duan (1995), while the second, jointly polynomial hybrid GARCH–SV (JPGARCH-SV), extends the affine HNGARCH model of Heston and Nandi (2000) and its SV counterpart studied by Heston et al.

For NIG innovations, polynomial preservation is characterized via Hermite extensions, requiring specific properties of the loading functions (psi i). The analysis of diffusion limits under Q(n) shows that for VPGARCH-SV with > 0, polynomial preservation requires cancellation of a non-polynomial leverage term when lambda(tau) + 1/2 not equal to 0. p. 0
Where the paper says it

An empirical analysis of S&P 500 returns and a large panel of options spanning 1996–2023 shows that the proposed specifications improve return fit and reduce mean weekly implied volatility root mean squared errors by approximately 8–41% relative to their nested benchmarks, depending on the variance dynamics and innovation distribution. Keywords: Polynomial-preserving processes; hybrid GARCH–stochastic volatility models; weak limits; option pricing.

The final conclusion regarding the polynomiality of the limiting operators states that for VPGARCH-SV, the marginal variance operator is polynomial if and only if its drift in Proposition SM. p. 2

Where the paper says it

, 2026), we derive two limiting regimes under the physical measure. Both regimes preserve polynomial variance dynamics and yield joint polynomiality in log price and variance for JPGARCH-SV.

6 is affine, since vVP is quadratic. p. 9
Where the paper says it

We use the same conditional mean return specification in both models to isolate the effects of their variance dynamics. 11For stationary versions of both specifications with finite mean variance, the conditionally demeaned returns y (n) (k+1)τ − ψ0 h (n) kτ ; τ belong to the SR-SARV(1) class of Meddahi and Renault (2004), since their conditional variance τh(n) kτ has affine conditional mean dynamics.

Joint polynomiality additionally requires eta = 0 under the physical measure. p. 2
Where the paper says it

, 2026), we derive two limiting regimes under the physical measure. Both regimes preserve polynomial variance dynamics and yield joint polynomiality in log price and variance for JPGARCH-SV.

For JPGARCH-SV, its joint operator is therefore polynomial if and only if xi 2(x)v JP(x) is affine. p. 2
Where the paper says it

The framework accommodates non-Gaussian innovations and distinct, correlated return and variance shocks. In the Gaussian case, the first class, variance-polynomial hybrid GARCH–SV (VPGARCHSV), extends the NGARCH dynamics of Engle and Ng (1993), used for option valuation, for example, by Duan (1995), while the second, jointly polynomial hybrid GARCH–SV (JPGARCH-SV), extends the affine HNGARCH model of Heston and Nandi (2000) and its SV counterpart studied by Heston et al.

The framework offers a route to characterizing their polynomial moment structure and developing more flexible alternatives beyond affine specifications. p. 9
Where the paper says it

11For stationary versions of both specifications with finite mean variance, the conditionally demeaned returns y (n) (k+1)τ − ψ0 h (n) kτ ; τ belong to the SR-SARV(1) class of Meddahi and Renault (2004), since their conditional variance τh(n) kτ has affine conditional mean dynamics. However, our additional restrictions yield polynomial conditional moments of variance at every degree and, under the joint restrictions, polynomial mixed return–variance moments; these properties are not implied by the SR-SARV(1) structure.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

1 Polynomial preservation allows exact computation of conditional moments of any order and serial cross-moments of variance and return processes over any finite horizon through finite dimensional matrix representations. This provides analytical tractability for complex stochastic volatility models without simulation error or computational burden.

2 The framework accommodates non Gaussian innovations and distinct, correlated return and variance shocks by identifying restrictions on conditional means and innovation loadings that preserve polynomial moments of variance alone or jointly with returns. This allows for the construction of two novel classes of hybrid GARCH-SV models that nest widely used specifications.

3 The models can be linked to continuous time polynomial diffusions and jump-diffusions under specific scaling regimes, providing a new discrete time construction of continuous time stochastic volatility models with polynomial variance dynamics. This connection is established through weak convergence to these limiting processes.

4 The framework provides exact conditional moments of cumulative returns and squared conditionally demeaned returns for hybrid GARCH-SV and JPGARCH-SV, which are valuable for derivative valuation without simulation error or the associated computational burden. These results allow for exact moment calculations under JPGARCH-SV, where exact conditional moments of cumulative returns are provided.

5 The framework enables empirical analysis of S&P 500 returns and options by showing that the proposed specifications improve return fit and reduce mean weekly implied volatility root mean squared errors by approximately 8–41% relative to their nested benchmarks, depending on the variance dynamics and innovation distribution. This suggests that the value of heavier innovation tails depends on the variance dynamics.

Abstract

This paper develops a discrete-time framework for volatility modelling and derivative pricing based on polynomial preservation. The framework provides exact moments of arbitrary order and serial cross-moments of variance and return processes over any finite horizon through finite-dimensional matrix representations. This analytical tractability is maintained while accommodating flexible leverage and variance-of-variance dynamics and non-Gaussian innovations. We introduce two novel classes of stochastic volatility models that nest widely used discrete-time specifications and study their statistical properties. Under two distinct scaling regimes, we establish weak convergence to polynomial diffusions and jump-diffusions and convergence of polynomial moments, linking discrete-time moment recursions to continuous-time matrix-exponential formulas. An empirical analysis of S&P 500 returns and a large panel of options spanning 1996--2023 shows that the proposed specifications improve return fit and reduce mean weekly implied volatility root mean squared errors by approximately 8--41% relative to their nested benchmarks, depending on the variance dynamics and innovation distribution.

Related papers