GROMM: Generative Risk-neutral Option Market Models

summary

The gist

We present a statically and dynamically arbitrage-free European option market model with smooth interpolation between option prices, based on the SANOS [Buehler et al., 2026b] framework…

In short

This model creates a dynamically arbitrage-free European option market by combining a discrete local volatility process with a smoothing process, based on the SANOS framework. It provides an explicit construction of risk-neutral dynamics and proves the existence of a unique strong solution. The model fits observed option prices well and allows for estimating physical dynamics and market price of risk.

Key concepts

SANOS Framework
This is the core architectural approach used to model the stock price as a product of two processes: a time-and-state discrete process (X) and a smoothing process (Y). This structure is used to encode option market information using discrete local volatilities, which are then represented in the option surface.
Local Volatilities (DLV)
These are discrete representations of local volatility derived from the option price surface. The model uses these as a convenient way to represent the market without being strictly a local volatility model. Their positivity is key to ensuring the market is free of static arbitrage.
Risk-Neutral Dynamics
Under a risk-neutral measure (Q), the densities of future stock prices must behave as martingales, meaning their expected future value equals their current value. The model constructs specific drifts and volatilities based on these martingale properties to ensure the resulting dynamics are free of dynamic arbitrage.
Strong Solution
This mathematical concept guarantees that there is one unique, well-defined path for the process over a given time interval. Proving the existence of a strong solution means that the model's evolution under risk-neutral conditions is mathematically guaranteed to be unique and consistent.

Terminology used across episodes

This episode discusses

The paper

GROMM: Generative Risk-neutral Option Market Models · Read on arXiv

Magnus Wiese, #, #

Mathematical Institute, University of Oxford · #Oxford-Man Institute, University of Oxford

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: "GROMM: Generative Risk-neutral Option Market Models".

Warren: Here's the gist of the paper. We present a statically and dynamically arbitrage-free European option market model with smooth interpolation between option prices, based on the SANOS

Buehler et al., 2026b: framework, and prove that its dynamics admit a unique strong solution Model Architecture The core architecture follows the SANOS approach

Buehler et al., 2026b: , which models the stock price as the product of a time-and-state discrete process X and a smoothing process Y, in our case Black & Scholes <ref:2610.16732#pg6>.

Ruth: The former is used to encode the option market using a discrete form of local volatilities as code book <ref:2610.16732#pg3>.

Warren: This representation does not mean that the model is a local volatility model – we merely use the implied local volatilities as a convenient representation of the option surface <ref:2610.16732#pg4>.

Ruth: Compared to using implied volatilities the advantage is that mere positivity of the local volatilities is sufficient to ensure the option market is free of static arbitrage at all times <ref:2610.16732#pg4>.

Title and authors: Ruth: The title GROMM Generative Risk-neutral Option Market Models tells us that the focus isn't just on fitting prices, but on generating a whole market model that respects risk-neutral dynamics through this stochastic local volatility lens.

Warren: It sounds like they are tackling the fundamental issue of making an option surface representation work not just as a static snapshot, but as a living process that evolves without introducing arbitrage opportunities.

Iris: The paper summarizes their approach by stating they present a model that is statically and dynamically arbitrage-free, based on the SANOS framework, and they prove its dynamics have a unique strong solution.

Ruth: That uniqueness part is significant because it means there isn't some messy ambiguity about how to evolve this market model forward in time under the risk-neutral measure Q.

Warren: They show their strength using S andP Index Option data from January two thousand twenty-two to August two thousand twenty-five and the initial fit covers more than ninety-five percent of relevant market option prices even with a simple grid architecture <ref:2610.16732#pg1,S&P Index Option data from January 2022 to August 2025>.

Iris: The data they use is quite recent, covering about eighteen months of trading days, which gives us some real-world context for testing these complex dynamics.

Ruth: It sets a high bar there; fitting ninety-five percent of the quotes on such recent data suggests the underlying structure they've captured is quite robust.

Warren: So, to summarize this part, GROMM aims to provide a model that is both statically and dynamically arbitrage-free by using a stochastic local volatility representation derived from SANOS, and they demonstrate its validity on recent S andP Index option data.

The paper's summary: Ruth: They summarize the paper by explaining that the model takes the discrete process X and a smoothing process Y, where X encodes local volatilities as a code book, which is how they represent the option surface.

Warren: The key implication here is that this representation doesn't have to be interpreted as a local volatility model; it just uses implied local volatilities as a convenient way to describe the surface.

Iris: They emphasize that the mere positivity of those local volatilities is enough to guarantee that the option market is free of static arbitrage at all times, which is a big simplification over other approaches.

Ruth: That's a strong statement because it means we don't need complicated checks just to ensure no arbitrage exists when looking at the static surface itself.

Warren: But their main contribution, as they frame it, is showing how to train the dynamics under P and Q and prove that this model is actually free of arbitrage while fitting market dynamics well.

Iris: They also propose a range of BHW-tests to test for absence of arbitrage on this dynamic model, which shows they are checking the property rigorously across different setups.

Ruth: So, the summary boils down to moving beyond static representations and proving that these stochastic dynamics maintain the no-arbitrage property even when you consider the time evolution under both physical and risk-neutral measures.

The paper's improvements: Warren: The paper points out several improvements they made to their previous models, specifically focusing on how they construct the model from data and how they calibrate it to ensure it fits observed bid-ask prices well.

Iris: They highlight that their implementation of the Tangent Lévy benchmark fails the no-arbitrage tests in their setup, so this new GROMM approach seems designed to address those failures directly.

Ruth: The authors also discuss comparing their model against other benchmarks like CRC, and they find that Model D has a richer factor representation and a finer time discretization, which is what they lean on for better performance.

Warren: And the rank-sixteen system they propose has maximum and 99th percentiles of coordinate statistics at three hundred forty-two point four seven zero and three hundred twenty-eight point eight one one, which suggests it captures a lot of variance compared to their previous systems.

Iris: They use diagnostic tests like the squared diagnostic and the entropy diagnostic, showing that these metrics remain consistent with risk neutrality on both the restricted universe and the full instrument grid under their model.

Ruth: So, they are essentially showing that by increasing complexity, specifically by using a rank-sixteen system, they can maintain consistency with risk neutrality across various checks while still fitting the market data reasonably well <ref:2610.16732#pg2>.

Conclusion: Ruth: To wrap up on GROMM, the authors have given us an explicit construction of a continuous-time risk-neutral measure for this SANOS discrete local volatility representation.

Warren: They extend the static SANOS surface from Buehler et al. 2026b to a dynamically arbitrage-free model, giving us that closed-form risk-neutral drift for every discrete local volatility coordinate and spot diffusion <ref:2610.16732#pg1>.

Iris: They also proved that the DLV dynamics admit a unique strong solution under the assumption of a square-integrable bound on the vol of vol, which is Theorem two point two in their paper <ref:2610.16732#pg1,dynamics admit a unique strong solution>.

Ruth: So, we have this explicit construction, which is much more concrete than using numerical martingale measures for pricing than what was done before.

Warren: And they showed that when you estimate the physical dynamics and the implied market price of risk, they recover a stable low-rank factor structure for the vol of vol.

Iris: The final part is that they show their rank-sixteen model has coordinate statistics at three hundred forty-two point four seven zero and three hundred twenty-eight point eight one one, and their conditional diagnostics remain consistent with the risk-neutral null on the full instrument grid under this GROMM model.

Ruth: So, GROMM is a significant step because it moves from static models to dynamically arbitrage-free ones by providing an explicit construction of the continuous-time risk-neutral measure for this specific type of option market representation.

Warren: It gives us a model where we can actually test for dynamic arbitrage using their BHW tests on restricted universes and full instrument grids, which is valuable information for risk management.

Iris: We just need to keep in mind their limitation, though they flag that the model't rely on a square-integrable bound on the vol of vol for that unique strong solution proof to hold.

Ruth: So, GROMM gives us a much more rigorous way to understand how option prices move dynamically, and we're ready to see what the next paper throws at us.

More episodes

← Home