Gatheral's Conjecture: General sufficient conditions for refutation
summary
The gist
In this paper we give general sufficient conditions for a stochastic volatility model to be a counterexample for Gatheral’s conjecture.
In short
The paper provides general conditions showing how a stochastic volatility model can disprove Gatheral's conjecture. By analyzing integrated variance derivatives and Wiener chaos expansions, it establishes that for rough volatility models, refutation occurs if the correlation exceeds a threshold dependent on the Hurst index H. Similar conditions are found for Markovian models based on specific parameter relationships in their defining SDEs.
Key concepts
- Gatheral's Conjecture
- This conjecture posits that the lowest possible value of a volatility option is achieved when using local volatility dynamics calibrated to European option prices. The paper investigates whether general stochastic volatility models violate this principle.
- Hurst Index (H)
- The Hurst index describes the long-term memory or roughness of a variance process, particularly in rough volatility models. A value of H less than 1/2 indicates a more irregular or 'rough' process, which is crucial for determining refutation conditions.
- Wiener Chaos Expansions
- These are mathematical tools used to analyze stochastic processes by expanding them into a series involving multiple Wiener processes. They allow researchers to compare the integrated variance of the stochastic model against that of the local variance model to find violations.
- Rough Volatility Models
- These are a specific class of models where the variance process follows a Volterra SDE with a Riemann-Liouville kernel and Hurst index H <= 1/2. The paper focuses on these models to establish general refutation conditions based on their asymptotic properties.
Terminology used across episodes
This episode discusses
The paper
Gatheral's Conjecture: General sufficient conditions for refutation · Read on arXiv
Elisa Alós, ʓscar Burs, Vladimir Lucic
Universitat Pompeu Fabra and Barcelona School of Economics · Universitat de Barcelona · Imperial College London
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: "Gatheral's Conjecture: General sufficient conditions for refutation".
Warren: Here's the gist of the paper. In this paper we give general sufficient conditions for a stochastic volatility model to be a counterexample for Gatheral’s conjecture.
Ruth: The conditions, achieved by asymptotic properties of integrated variance derivatives and Wiener chaos expansions, show that a general class of rough volatility models are generally counterexamples to Gatheral’s conjecture.
Warren: Gatheral's Conjecture Context Gatheral’s conjecture states that the lowest possible value of a volatility option is achieved by assuming local volatility dynamics given the prices of European options.
Ruth: Earlier practitioner’s observations reported higher fair volatility swap strikes in a calibrated local volatility model rather than a Heston model, which was in line with Gatheral’s conjecture.
Title and authors: Ruth: So we're looking at this paper today, "Gatheral's Conjecture: General sufficient conditions for refutation." It’s about finding general rules under which a stochastic volatility model fails Gatheral's conjecture.
Warren: Yeah, it’s about showing that for a whole class of rough volatility models, the conjecture isn't true unless you have really specific parameter settings.
Iris: The authors use asymptotic properties of integrated variance derivatives and Wiener chaos expansions to figure this out. It sounds like they are using some heavy machinery to see how the variance behaves over time as we get really close to zero maturity.
Ruth: Exactly, they are looking at the short-time behavior of things—the integrated variance derivatives—to create conditions where a model breaks the conjecture if you keep your time to maturity small enough.
Warren: And for those rough volatility models, they found that if the Hurst index H is less than one half and the correlation has a certain absolute value, say greater than two H plus three fourths, then Gatheral's conjecture is refuted.
Iris: That specific threshold, that two H plus three fourths for when the correlation is large enough to break it—that sounds like a very concrete condition we can check in practice.
Ruth: It means that if you are working with a rough process and your correlation is strong enough, you can expect the conjecture to fail when the option expires soon.
Warren: For those Markovian variance models, where H equals one half and the correlation is one in absolute value, they found that a specific relationship between the coefficients of the stochastic differential equation defining variance has to be met for it to refute Gatheral's conjecture.
Iris: They look at Wiener chaos expansions here to compare the integrated variance of this stochastic model against that of a local variance model, which is where things get technical.
Ruth: It’s about comparing how the integrated variance scales in both cases, which is what lets them see if one is always lower than the other as time gets very small.
Title and authors: Warren: They also showed how these refutation conditions extend to broader classes of models by relaxing some of those initial strong hypotheses that were needed for their main theorem.
Iris: So they're not just looking at one specific model structure, but finding a general way to look at any variance process—rough or Markovian—and predicting if it will be a counterexample based on its parameters.
Ruth: The examples they give, like the rough Heston model and the SABR model, show that these refutation conditions are actually achievable in some specific cases.
Warren: For instance, they show that for the SABR model with parameters alpha equal to one point two zero and rho equal to negative one with a time to maturity of zero point zero one, the difference between those quadratic payoffs is negative.
Iris: That means for that particular setup, the integrated variance under the stochastic model is lower than what you'd get from the local variance model.
Ruth: But they also noted that for that same SABR setup with alpha equal to one point two zero and rho equal to positive one, the short-time behavior of those payoffs is consistent with Gatheral's conjecture.
Warren: So it depends entirely on the correlation sign and magnitude, even within the same model family.
Iris: And they gave another condition for rough processes: if you have a variance process satisfying a different set of hypotheses, say H prime one, then the conjecture fails if your correlation is greater than or equal to two H plus three fourths.
Ruth: That two H plus three fourths threshold is what we need to keep in mind when analyzing these rough volatility dynamics.
Warren: For Markovian variances with a half Hurst index and a correlation of one, they look at the constant K naught alpha beta rho and if it's negative, Gatheral's conjecture fails under certain conditions as well.
Iris: So for the Markovian case, it comes down to finding that specific relationship on the coefficients of that SDE defining variance.
Ruth: This paper really gives us a framework where we can use these asymptotic expansions to check if a model is likely to be a counterexample before we even run a full simulation.
Title and authors: Warren: It lets us use the structure of the integrated variance derivatives to predict outcomes based on the parameters of the volatility process itself.
Iris: The limitation they mention is that they need very strong hypotheses on those coefficients alpha and beta when they are proving their main result, so if you want to apply it to a wider variety of models, you might need to relax those initial assumptions.
Ruth: And that’s the trade-off—you get general conditions, but sometimes the proof requires a lot of setup first.
Warren: So we see that for rough volatility models, it's about the Hurst index and correlation threshold, and for Markovian models, it's about specific coefficient relationships.
Iris: It’s a clear separation based on whether you have a rough or Markovian structure for your variance process.
Ruth: To wrap up this paper on "Gatheral's Conjecture: General sufficient conditions for refutation," the main point is that we now have general conditions—based on short-time properties and chaos expansions—to show when a stochastic volatility model will act as a counterexample to Gatheral's conjecture.
Warren: Specifically, for rough volatility models, it won't hold if the correlation exceeds that threshold dependent on the Hurst index H.
Iris: And for Markovian variances, the failure condition is determined by a specific relationship between the coefficients of the SDE defining variance when you look at small time to maturity.
Ruth: The implication for us in markets is that we can use these analytical tools to check if a model structure we are using might be violating this conjecture under short-term trading horizons.
Warren: It gives us a way to test if the complexity of a stochastic volatility model is actually leading to pricing discrepancies that contradict Gatheral's expectation.
Iris: For those who only listen, remember that this paper provides the mathematical tools—the integrated variance derivatives and Wiener chaos expansions—to systematically check if your chosen volatility process is behaving in a way that refutes the conjecture under certain conditions.
The paper's summary: Ruth: So, this paper provides general rules for when a stochastic volatility model breaks Gatheral's conjecture by looking at how integrated variance derivatives and Wiener chaos expansions behave in rough volatility settings.
Warren: Right, so if you want to know if a model is going to be a counterexample, you look at its parameters against these specific conditions derived from those asymptotic expansions.
Iris: And they’re showing that for rough models, the correlation coefficient and the Hurst index dictate whether the conjecture holds or fails under small time-to-maturity conditions.
Ruth: It means we can use these math tools to predict if a model structure is going to contradict Gatheral's expectation before we even run a full simulation on it.
Warren: Exactly, it gives us a way to test if the complexity of that stochastic volatility model is actually leading to pricing discrepancies that go against what Gatheral predicted.
Iris: And for Markovian models, they give us a condition based on the coefficients of the SDE defining variance when the correlation is one and time to maturity is small enough.
Ruth: So, it’s a clear distinction between rough and Markovian processes, with different tests depending on which structure you have.
Warren: The paper does show examples, like with the SABR model, where you can actually see these refutation conditions in action for specific parameter choices.
Iris: It's important to remember that these conditions come from looking at very short-time behavior of those derivatives and expansions, so they might not hold true for long time horizons.
Ruth: And the authors admit that they have to relax some of their original strong hypotheses—those H one through three—to apply these refutation conditions more broadly across different types of variance processes.
Warren: So, it’s less about a single perfect model and more about finding general thresholds that work for a whole class of them.
Iris: The main conclusion is that for rough volatility models, the conjecture can't hold if the correlation gets too high relative to the Hurst index.
Ruth: And for Markovian variances, it fails when there’s a specific negative relationship between those SDE coefficients under certain conditions.
Warren: So what this means practically is that we have a mathematical filter now to check if our current models are behaving in a way that contradicts Gatheral's expectation on short time scales.
Iris: It gives us the framework to systematically check for these violations using the tools of integrated variance derivatives and chaos expansions.
Ruth: Next up, we’re going to look at how these specific conditions translate into actionable checks for the different types of volatility models we use every day.
The paper's improvements: Tom: So, we’re talking about how this paper improves on the original work by showing that these refutation conditions aren't just for one specific setup but generalize across broader classes of models.
Ruth: Right, so they took those initial strong hypotheses from the first version and showed how you can relax them—those H one through three things—and still get valid refutation results.
Warren: So, if you have a variance process that satisfies a slightly weaker condition, like hypothesis H prime one, you still get that threshold based on the Hurst index and correlation.
Iris: That means the result isn't just locked into one specific type of volatility structure; it applies more generally to any process meeting those relaxed criteria.
Ruth: It’s about making the findings more robust so they don't just apply to a few textbook examples, but really cover a whole range of real market models.
Warren: And for the Markovian case, they showed that the condition isn't just about being at H equals one half and correlation equal one; it’s about finding that specific negative relationship between those SDE coefficients.
Iris: So, even in the more constrained Markovian world, the identification of what variation is doing the work remains crucial to finding that specific coefficient constraint.
Ruth: It also introduced these examples, like the SABR model, which lets you actually see how those refutation conditions play out when you plug in concrete numbers for alpha and rho.
Warren: Those examples are interesting because they show that sometimes for one set of parameters, the conjecture actually holds true under short-time behavior.
Iris: But other parameter choices immediately push it over that threshold, confirming the general condition they derived is actually working as intended.
Ruth: The whole point of these improvements is moving from a narrow finding to a set of general sufficient conditions for refutation across rough and Markovian dynamics.
Warren: It gives us a more comprehensive toolset for testing volatility models against Gatheral's conjecture than we had before.
Iris: And while the paper successfully relaxes some of those initial strong hypotheses, it still flags that these results are tied to the very short-time behavior of those derivatives and expansions.
Ruth: So, the implication is that we have a better way to analytically check if a model structure is likely to be a counterexample based on its parameters.
Warren: This sets up the next part of our discussion: how these specific conditions translate into actionable checks for the different types of volatility models we use every day.
Conclusion: Tom: So to wrap up this discussion on "Gatheral's Conjecture: General sufficient conditions for refutation," the core idea is that we have now established general mathematical conditions for when a stochastic volatility model fails Gatheral's conjecture based on its structure and parameters.
Ruth: Right, so what this means for us in economics is that we can use these analytic tools to check if our current models are behaving in a way that contradicts what Gatheral predicted under short time scales.
Warren: It gives us a way to test if the complexity of a stochastic volatility model is actually leading to pricing discrepancies that go against Gatheral's expectation.
Iris: We have shown how the behavior of integrated variance derivatives and Wiener chaos expansions allows us to systematically check for these violations in both rough and Markovian cases.
Ruth: The main finding is that for rough volatility models, the conjecture won't hold if the correlation exceeds a threshold that depends on the Hurst index H.
Warren: And for Markovian variances, it fails when there’s a specific relationship between those SDE coefficients under certain conditions when time to maturity is small enough.
Iris: It’s important to remember that these refutation conditions come from looking at very short-time behavior of those derivatives and expansions, so they might not hold true for long time horizons.
Ruth: The authors also showed how these refutation conditions extend to broader classes of models by relaxing some of their original strong hypotheses, which makes the findings more applicable.
Warren: So, it’s less about a single perfect model and more about finding general thresholds that work for a whole range of them when you look at the short term.
Iris: The implication is that we have a better way to analytically check if your chosen volatility process is behaving in a way that refutes the conjecture based on its mathematical structure.
Ruth: We’ve summarized how the asymptotic properties of those variance processes lead us to these specific conditions for when Gatheral's conjecture fails.
Warren: It gives us a more comprehensive toolset for testing volatility models against Gatheral's expectation than we had before.
Iris: And while the paper successfully relaxes some of those initial strong hypotheses, it still flags that these results are tied to the very short-time behavior of those derivatives and expansions.
Ruth: We’ve summarized how the asymptotic properties of those variance processes lead us to these specific conditions for when Gatheral's conjecture fails.
Warren: It gives us a more comprehensive toolset for testing volatility models against Gatheral's expectation than we had before.
Iris: This work sets up a clear way to use the math—the integrated variance derivatives and chaos expansions—to systematically check if your chosen volatility process is behaving in a way that refutes the conjecture.
Ruth: Next up, we’re going to look at how these specific conditions translate into actionable checks for the different types of volatility models we use every day.
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