Reaction-boundary variance and adjoint-consistent local-volatility projection

arXiv:2607.05011 · q-fin.PR, q-fin.MF, q-fin.TR · Submitted 2026-07-06 · 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: "Reaction-boundary variance and adjoint-consistent local-volatility projection".

Warren: Here's the gist of the paper. We derive an operational-time variance kernel for a latent-order-book reaction boundary and use it to separate three objects usually collapsed in calendar-time volatility models The reaction boundary is the zero of a bid–ask imbalance field Derivation of the Operational Variance Kernel The paper derives an operational-time variance kernel for a latent-order-book reaction boundary and uses it to separate three objects usually collapsed in calendar-time volatility models The operational variance is obtained as a finite-scale Green-function cumulant rather than introduced as a primitive diffusion coefficient For long-memory forcing with exponent 0 < γ < 1, the operational variance has a closed asymptotic form involving effective signed-forcing intensity, liquidity slope, resilience, memory, and operational coarse-graining scale The resulting mesoscopic closure is the finitescale reduction of the detailed reaction-boundary dynamics to a (∆)u(S, u), rather than an assertion that the microscopic boundary is itself a calendar-time diffusion.

Ruth: Structural Decomposition of Volatility The construction separates three layers that are often collapsed into a single volatility coefficient The first layer is the local reaction-boundary response of the latent book The second layer is the cumulant calculation for long-memory signed forcing in operational time The third layer is the projection from operational time to calendar time.

Warren: This construction combines latent-order-book response, long-memory signed forcing, and forward–backward adjoint consistency.

Ruth: Adjoint Consistency as a Reality Constraint The paper argues that such projections are admissible only when the induced forward density operator and backward valuation operator remain adjoint on the same state space Adjoint consistency is therefore a reality constraint on operational-to-calendar time projection.

Paper summary: Ruth: So we're looking at this paper here called "Reaction-boundary variance and adjoint-consistent local-volatility projection". It’s about taking something happening at the order book level—the reaction boundary—and using that to separate three different things usually lumped together in calendar time volatility models.

Warren: Right, so what it claims is that they derive an operational time variance kernel for this reaction boundary and use it to split the problem into a structural boundary cumulant, a clock projection, and a pricing measure choice. It’s about finding a better way to look at volatility than just using calendar time models.

Iris: What I’m seeing in the abstract is that they get an operational variance as a finite-scale Green-function cumulant instead of just treating it as some kind of primitive diffusion coefficient. That sounds like they're trying to be more precise about how the noise behaves over short time scales.

Ruth: Exactly. The paper says for long memory forcing with an exponent less than one, they get a closed form for this operational variance that depends on things like effective signed-forcing intensity, liquidity slope, resilience, memory, and the operational coarse-graining scale. It’s a really detailed description of how the noise scales.

Warren: And it moves away from just saying the microscopic boundary is some kind of calendar time diffusion. Instead, they get this mesoscopic closure that's a finite-scale reduction of what happens at the reaction boundary dynamics, which is described by something like Delta u S u.

Iris: That means they are focusing on the dynamics happening in operational time before projecting it to calendar time, which is where things usually get messy. It’s about getting a more rigorous way to define what we're actually measuring there.

Ruth: The paper builds this up by decomposing the volatility into three layers: first, the local reaction-boundary response of the latent book, second, calculating the cumulant for long-memory signed forcing in operational time, and third, projecting from operational time to calendar time.

Warren: And they put a lot of emphasis on what they call adjoint consistency as a reality constraint. They argue that you can only project from operational to calendar time if the forward density operator and the backward valuation operator are adjoint on the same state space.

Paper summary: Iris: So it’s not just an arbitrary mapping; it has to satisfy this condition for it to be considered a coherent one-state pricing system, meaning its backward claim operator and forward state-price operator have to be adjoints of the same projected mechanism.

Ruth: It also relates this back to local volatility and time changes. They suggest that the deterministic clock gives you the benchmark local volatility equation, but that equation should be understood as a projection of this asymptotic kernel rather than just some simple replacement for the operational model.

Warren: The asymptotic form of this kernel is what makes that adjoint condition concrete, so you can see how it works. It suggests that even before picking a calendar time clock, you can talk about the operational kernel itself.

Iris: So they use this analytical closure to show how those structural reaction-boundary quantities look when projected into a local volatility-like surface, which helps illustrate what the parameters actually do in simulations.

Ruth: The main result they point to is equation twenty. It expresses the variance of boundary increments in terms of stuff that’s directly interpretable, like the effective signed-forcing intensity Aeff, liquidity slope L, resilience nu, the long memory exponent gamma, and the operational coarse-graining scale Delta.

Warren: They lay out how these components multiply together: activity multiplies this operational variance kernel; signed forcing supplies boundary fluctuations; liquidity absorbs them quadratically; resilience dampens persistent memory, and the memory exponent dictates how that observed variance changes with the coarse graining scale.

Iris: So it’s a very concrete set of structural ingredients that determine the final observed variance, rather than just fitting a generic diffusion model to data. It’s about linking the microscopic book dynamics to these measurable operational quantities.

Ruth: The paper concludes by saying this non-unique time extension preserves this operational cumulant and moves any uncertainty, or clock uncertainty, into the projection from operational to calendar time.

Paper summary: Warren: That means that even if you don't have one single calendar time clock, as long as you stick to this operational cumulant, the structure remains intact. The incompleteness is pushed into the mapping between operational and calendar time.

Iris: It’s important to note their limitation here—they state that a nonunique operational-to-calendar projection only defines a coherent one-state pricing system if its backward claim operator and forward state-price operator are adjoints of the same projected mechanism. That’s the constraint they have placed on it.

Ruth: So, in simple terms, this paper provides a structural route from those latent order book reaction boundary fluctuations to an admissible calendar time local volatility pricing representation by first deriving that operational variance kernel.

Warren: It really boils down to using these structural quantities—the effective signed-forcing intensity, the liquidity slope, the resilience—to explain how volatility behaves when you look at it through the lens of operational time before you even worry about a specific calendar clock.

Iris: And they show that this approach is valid only when you enforce that adjoint consistency between the forward and backward operators on the same state space. That’s a hard requirement for it to work as a pricing model instead of just some formal variance projection.

Ruth: So, while the paper shows how these structural quantities interact with the operational variance kernel, it stops short of showing how this directly applies to real-world trading strategies or market making in a way that bypasses all those constraints.

Warren: That’s fair; they use simulations to show which declared structural channels and their associated parameters can generate different state- and maturity-dependent surfaces. It’s illustrative, but it doesn't provide a direct plug-and-play formula for every market.

Iris: They also mention that the local volatility equation should be read as an admissible projection of this asymptotic kernel, not just a primitive replacement for the operational model, which is an important distinction for how you interpret those results.

Ruth: So the core contribution here is establishing this structural route from latent order book fluctuations to a calendar time representation by grounding it in an operational variance derived first. That’s what they’re really trying to achieve with this paper.

Conclusion: Ruth: So we've been talking about this paper, "Reaction-boundary variance and adjoint-consistent local-volatility projection." What they're really doing is taking something happening at the order book level—the reaction boundary—and building a way to connect it to what we usually call calendar time volatility.

Warren: Right, so the main idea is they derive this operational time variance kernel first. It’s not just some random noise measure; it’s a finite-scale Green-function cumulant.

Iris: And that kernel lets them separate three things that are usually lumped together in calendar time models. It breaks volatility down into the local reaction boundary response, the long memory forcing calculation in operational time, and then the projection to calendar time.

Ruth: Exactly, so it’s a structural decomposition of volatility rather than just fitting a curve to historical data. The whole thing hinges on this adjoint consistency requirement for their projections to even be valid.

Warren: That's the sticking point for me. Adjoint consistency being a reality constraint on operational-to-calendar time projection means you can only get that coherent one state pricing system if the forward density operator and backward valuation operator are adjoints on the same state space.

Iris: So it’s not just a mathematical trick; it’s a physical requirement for the model to make sense. It says if you want to map operational time results onto calendar time pricing, that mapping has to satisfy this strict adjoint condition.

Ruth: The conclusion they highlight is equation twenty, which expresses the variance of boundary increments using these directly interpretable structural quantities like effective signed-forcing intensity and resilience.

Warren: It shows how activity multiplies this kernel, how signed forcing drives the fluctuations, and how liquidity absorbs them quadratically. It gives us a clear set of ingredients that determine the final observed variance.

Iris: The paper stops short of saying this is a universal pricing formula for every market, though they use simulations to show which parameters generate different surfaces depending on the state and maturity.

Ruth: They also clarify that the deterministic local volatility equation should be seen as a projection of this asymptotic kernel, not just some simple replacement for the operational model.

Warren: So it’s about grounding that local volatility idea in this operational structure first. It suggests that even before you pick a specific calendar clock, you can talk about the operational kernel itself.

Iris: And that’s what makes their asymptotic form concrete, so it shows exactly how those adjoint conditions actually play out. The way the variance scales with coarse graining is determined by the memory exponent gamma.

Ruth: The authors conclude by saying this non-unique time extension preserves that operational cumulant and pushes any clock uncertainty into the projection from operational to calendar time.

Warren: So, even if you don't have one perfect calendar clock, as long as you stick to this operational cumulant, the underlying structure remains intact. The incompleteness is just moved into that mapping process.

Iris: It’s a strong result because it provides a rigorous way to handle the inherent uncertainty in time scales when moving from operational dynamics to real-world pricing measures.

Ruth: This leads us nicely into how this framework might change how we think about risk across different time horizons, and what kind of modeling assumptions we need to check on the underlying agents.

School of Mathematics and Statistics, University of New South Wales · Department of Statistical Sciences, University of Cape Town

q-fin.PR, q-fin.MF, q-fin.TR

Submitted: 2026-07-06

Updated: 2026-11-05

Comments: 16 pages, 12 figures, 1 table; minor clarifications, updated references, expanded discussion and limitations, results unchanged; Code reproducibility see:https://github.com/timgebbie/reaction-boundary-vol-surface-reproducibility

Code: https://github.com/timgebbie/reaction-boundary-vol-surface-reproducibility

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 86/100

The gist: We derive an operational-time variance kernel for a latent-order-book reaction boundary and use it to separate three objects usually collapsed in calendar-time volatility models The reaction boundary…

Key concepts

Operational Variance Kernel
This is a specific mathematical tool derived from the zero of the bid-ask imbalance field in operational time. It is not just a diffusion coefficient but a finite-scale Green-function cumulant that captures how volatility changes over discrete operational steps, rather than continuous calendar time.
Structural Decomposition of Volatility
Volatility is broken down into three distinct layers: the immediate reaction of the latent order book, the effect of long-memory signed forcing in operational time, and the projection from this operational view onto calendar time. This separates complex market behavior into manageable parts.
Adjoint Consistency
This is a crucial constraint for projecting results from operational time to calendar time. It requires that the forward density operator and backward valuation operator remain adjoint on the same state space. If this condition isn't met, a coherent one-state pricing system cannot be defined.
Operational Coarse-Graining Scale ($Δ$)
This scale represents how much operational time is grouped together. The paper shows that the resulting variance depends on this scale, indicating that the observed volatility structure changes depending on whether you look at very fine or coarse operational time increments.

Terminology

Summary

We derive an operational-time variance kernel for a latent-order-book reaction boundary and use it to separate three objects usually collapsed in calendar-time volatility models The reaction boundary is the zero of a bid–ask imbalance field

Derivation of the Operational Variance Kernel

The paper derives an operational-time variance kernel for a latent-order-book reaction boundary and uses it to separate three objects usually collapsed in calendar-time volatility models The operational variance is obtained as a finite-scale Green-function cumulant rather than introduced as a primitive diffusion coefficient For long-memory forcing with exponent 0 < γ < 1, the operational variance has a closed asymptotic form involving effective signed-forcing intensity, liquidity slope, resilience, memory, and operational coarse-graining scale The resulting mesoscopic closure is the finitescale reduction of the detailed reaction-boundary dynamics to a (∆)u(S, u), rather than an assertion that the microscopic boundary is itself a calendar-time diffusion.

Structural Decomposition of Volatility

The construction separates three layers that are often collapsed into a single volatility coefficient The first layer is the local reaction-boundary response of the latent book The second layer is the cumulant calculation for long-memory signed forcing in operational time The third layer is the projection from operational time to calendar time. This construction combines latent-order-book response, long-memory signed forcing, and forward–backward adjoint consistency. p. 1

Where the paper says it

In summary: the purpose here is to derive a finite-scale operational variance coefficient for a reaction-boundary price process and to identify the conditions under which its projection into calendar time defines a local-volatility pricing model. The construction combines latent-orderbook response, long-memory signed forcing, and forward– backward adjoint consistency.

Adjoint Consistency as a Reality Constraint

The paper argues that such projections are admissible only when the induced forward density operator and backward valuation operator remain adjoint on the same state space Adjoint consistency is therefore a reality constraint on operational-to-calendar time projection. A non-unique operational-to-calendar projection defines a coherent one-state pricing system only if its backward claim operator and forward state-price operator are adjoints of the same projected mechanism. p. 4

Where the paper says it

(34) Proposition 1 (Adjoint consistency as a reality constraint). A non-unique operational-to-calendar projection defines a coherent one-state pricing system only if its

Relation to Local Volatility and Time Changes

The deterministic clock gives the benchmark local-volatility equation The operational kernel can be discussed before any calendar-time clock is selected The asymptotic form of Ξ is what makes the adjoint condition concrete. The local-volatility equation should therefore be read as an admissible projection of the asymptotic kernel, not as a primitive replacement for the operational model. p. 5

Where the paper says it

Thus two clocks can agree on a scalar projected variance at a point while implying different forward laws once hidden activity, filtration, or clock-risk information is taken into account. The local-volatility equation should therefore be read as an admissible projection of the asymptotic kernel, not as a primitive replacement for the operational model.

Reaction-Boundary Volatility Surfaces

The analytic closure above can also be used to illustrate how the structural reaction-boundary quantities appear after projection into a local-volatility-like surface The simulations are therefore illustrative. The figures show which declared structural channels and there associated parameters can generate different state- and maturity-dependent surfaces. p. 8

Where the paper says it

They do not estimate L(m), ν(m), Aeff(m), γ(m) or α(T) from market data, and the plotted object should not be read as a calibrated or arbitrage-free Dupire local-volatility surface. Rather, the figures show which declared structural channels and there associated parameters can generate different state- and maturity-dependent surfaces.

Conclusion

The key result is Eq. (20) It expresses the variance of boundary increments in terms of directly interpretable structural quantities: the effective signed-forcing intensity Aeff, the local liquidity slope L, the resilience ν, the long-memory exponent γ, and the operational coarse-graining scale ∆ Activity multiplies the operational variance kernel; signed forcing supplies boundary fluctuations; liquidity absorbs them quadratically; resilience damps persistent memory; and the memory exponent determines how the observed variance changes with coarse graining. p. 9

Where the paper says it

The key result is Eq. It expresses the variance of boundary increments in terms of directly interpretable structural quantities: the effective signed-forcing intensity Aeff, the local liquidity slope L, the resilience ν, the long-memory exponent γ, and the operational coarse-graining scale ∆.

The nonunique-time extension preserves this operational cumulant and moves incompleteness, or clock uncertainty, into the projection from operational to calendar time. p. 9
Where the paper says it

This limit is not meant to be a complete empirical model, but it is the clearest summary of the mechanism: activity amplifies variance, signed forcing supplies fluctuations, liquidity absorbs them quadratically, resilience damps persistent memory, and the memory exponent determines how the observed variance changes with coarse graining. The nonunique-time extension preserves this operational cumulant and moves incompleteness, or clock uncertainty, into the projection from operational to calendar time.

Code and Reproducibility

The simulations and original figure outputs are archived in the versioned v1.0.0 reproducibility release The figures used here are square-format renderings of the same calculations and parameter settings. p. 9

Where the paper says it

0 reproducibility release [1]. The figures used here are square-format renderings of the same calculations and parameter settings.

The code is available at https://github. p. 9
Where the paper says it

University of Cape Town. URL: https://github.

com/timgebbie/reaction-boundary-vol-surface-reproducibility/releases/tag/v1. p. 9
Where the paper says it

URL: https://github. com/timgebbie/ reaction-boundary-vol-surface-reproducibility/ releases/tag/v1.

0.0.

Acknowledgments

During preparation of this manuscript, the authors used a generative-AI tool for language editing and consistency checks The authors reviewed and edited all outputs and take full responsibility for the content of the manuscript.

References

The references listed are [1] through [33]. The paper cites sources such as Bouchaud, J. p. 0

Where the paper says it

Trading activity arrives through irregular sequences of orders, trades, quotes and cancellations, so it is often useful to describe price formation first on an activity, or operational, clock and only afterwards relate that evolution to calendar time [13, 12, 2]. In the present paper, operational time u indexes the local market activity on which the reaction-boundary dynamics evolve, whereas calendar time t carries the maturity, discounting and contractual timing of the option-pricing problem.

P., Farmer, J. p. 2
Where the paper says it

Use the regularised kernel g (τ0) ν,D (r) = exp[−ν(r + τ0)] p 4πD(r + τ0) 1r≥0 (11) and a locally stationary long-memory forcing covariance Cm(τ ; S, u) = Am(S, u)(|τ | + τ0) −γ(S,u) , (12) where 0 < γ < 1. The sign-memory exponent may be interpreted through the hidden-order splitting mechanism of Lillo, Mike and Farmer [27].

D., Lillo, F. p. 2
Where the paper says it

Use the regularised kernel g (τ0) ν,D (r) = exp[−ν(r + τ0)] p 4πD(r + τ0) 1r≥0 (11) and a locally stationary long-memory forcing covariance Cm(τ ; S, u) = Am(S, u)(|τ | + τ0) −γ(S,u) , (12) where 0 < γ < 1. The sign-memory exponent may be interpreted through the hidden-order splitting mechanism of Lillo, Mike and Farmer [27].

, 2009, and Carr, P., Wu, L., 2004. The references include foundational texts like Black, F. p. 6
Where the paper says it

Reaction-boundary volatility surfaces The analytic closure above can also be used to illustrate how the structural reaction-boundary quantities appear after projection into a local-volatility-like surface. We consider stylised liquidity, resilience-response and signed-forcing profiles across log-moneyness and maturity, using the finite-scale operational variance derived in Sections 4–5 and its deterministic calendar-time projection from Section 6.

, Scholes, M. p. 0
Where the paper says it

The calculation below then asks how fluctuations of this interface generate a finitescale variance before any calendar-time pricing model is imposed. Volatility is often introduced at the level of a calendartime diffusion, either postulated directly as in Black– Scholes–Merton or inferred from option prices in localvolatility models [6, 29, 19, 23].

, 1973 and Merton, R. p. 0
Where the paper says it

The calculation below then asks how fluctuations of this interface generate a finitescale variance before any calendar-time pricing model is imposed. Volatility is often introduced at the level of a calendartime diffusion, either postulated directly as in Black– Scholes–Merton or inferred from option prices in localvolatility models [6, 29, 19, 23].

C., 1973. The list of references is provided at the end of the document. p. 9
Where the paper says it

The authors reviewed and edited all outputs and take full responsibility for the content of the manuscript. References [1] Angstmann, C.

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. Operational variance kernel derivation allows AI systems to move beyond postulating primitive diffusion coefficients by deriving a structural decomposition of local volatility into boundary variance, activity, and pricing-measure choice. This enables AI to model volatility not as an external parameter but as a consequence of internal market dynamics like order flow and liquidity slope.

  2. Adjoint consistency serves as a reality constraint for projecting operational-to-calendar time, meaning AI systems can identify where incompleteness enters in time projections by checking if the induced forward density operator and backward valuation operator remain adjoint on the same state space. This prevents the creation of formal coefficient projections that are not coherent one-state pricing models.

  3. The framework allows for modeling market uncertainty as a choice between candidate clock projections, where only those belonging to Ad(U) define a coherent representation. AI systems can use this to distinguish between formal coefficient projection rather than a pricing model and true pricing systems based on the adjoint-real condition in Equation (34).

  4. The asymptotic closure provides an analytical bridge from microstructure response to local volatility by deriving the operational variance kernel under explicit local and asymptotic assumptions, relating it directly to structural quantities like effective signed-forcing intensity Aeff, the local liquidity slope L, the resilience nu, the long-memory exponent gamma, and the operational coarse-graining scale Delta.

  5. The simulation results provide a structured way for AI systems to compare different structural channels by showing how they generate surface morphology across different regimes (low-z vs high-z response). This allows AI to diagnose which structural components—liquidity, resilience, or forcing—are the leading source of state dependence in specific market conditions.

Abstract

We derive an operational-time variance kernel for a latent-order-book reaction boundary and use it to separate three objects usually collapsed in calendar-time volatility models: a structural boundary cumulant, a clock projection, and a pricing-measure choice. The reaction boundary is the zero of a bid--ask imbalance field. For a locally linear book, signed order-flow perturbations displace this zero through a damped Abel response kernel, so the variance of boundary increments is obtained as a finite-scale Green-function cumulant rather than introduced as a primitive diffusion coefficient. For long-memory forcing with exponent 0<γ<1, the operational variance has a closed asymptotic form involving effective signed-forcing intensity, liquidity slope, resilience, memory, and operational coarse-graining scale. A deterministic activity clock gives the benchmark local-volatility projection. More general, non-unique clocks generate candidate calendar-time pricing systems. We argue that such projections are admissible only when the induced forward density operator and backward valuation operator remain adjoint on the same state space. Adjoint consistency is therefore a reality constraint on operational-to-calendar time projection: it disciplines non-unique time and identifies where incompleteness enters.

Sources

Related papers