Correlation emergence and the Epps effect in two coupled limit order books

arXiv:2606.14182 · q-fin.TR, math.AP, q-fin.ST · Submitted 2026-06-12 · 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: "Correlation emergence and the Epps effect in two coupled limit order books".

Warren: Here's the gist of the paper. The paper provides a unified analytic account of correlation emergence and the Epps effect in two coupled limit order books because it connects reaction–diffusion order books to reduced-form accounts of the Epps effect through a discrete-to-continuum construction.

Ruth: Correlation emerges from explicit order-flow coupling between the books rather than from correlated price noise or a latent multivariate diffusion.

Warren: The model places the cross-book mechanism in γjk, with its price effect transmitted through the coupled order-density fields and moving reaction fronts and as such is sourced from the phenomenology of trading itself rather than being latent pre-specified relationships between trade or event types.

Ruth: Modeling the Coupled System The system is modeled by separating order-book dynamics from observation clocks, defining operational time steps and calendar time via a clock mapping.

Title and authors: Ruth: We've touched on what this paper is about—correlation emergence and the Epps effect in two coupled limit order books. This segment focuses on unpacking what those terms actually mean for understanding market dynamics.

Warren: It’s not just about price correlation; it’s about how that correlation builds up or fades depending on how fast we look at the data versus how fast the traders are interacting.

Iris: The title itself points to a system with two distinct order books, and then there's this Epps effect, which is basically the empirical reduction in measured cross-asset log-return correlation when you aggregate over short scales.

Ruth: That means that if you look at prices frequently, the correlation looks different than if you wait longer to average it out. The paper tries to model how that recovery happens as aggregation interval increases.

Warren: The authors set up a fixed-grid discrete random walk for order flow in operational time, and they introduce a pair-trader coupling right at the level of order creation between the books.

Iris: They also impose calendar time using separate observation clocks, which is what sets up the whole clock overlap problem they want to solve.

Ruth: Essentially, they are trying to unify how trading dynamics create correlation with how we measure that correlation across different time scales and different observation methods.

Warren: The main result is that this model successfully reduces to coupled reaction–diffusion equations when you take the continuum limit under specific conditions, like delta x approaching zero and delta u approaching zero.

Iris: That continuum limit lets them find a moving reaction boundary defining the model log-mid-price, which separates buy- and sell-dominated regions of the order book.

Ruth: So, they’re using this boundary to approximate what we see as the model log-mid-price, and that level is then used to approximate the actual observed price level in some reduced way.

Warren: It’s a bit abstract because it moves from discrete steps to continuous functions, but for modeling complex order book behavior, that’s often necessary.

Iris: The paper is really focused on showing how the clock and coupling effects combine to give you the overall correlation approximation rho comb dt being related to clock overlap and coupling response.

Ruth: So, this paper gives us a framework where we can separate the timing issue from the interaction issue when analyzing realized correlation.

Warren: It’s useful for quantitative traders because it suggests that latency and sampling frequency are not just noise; they are structural parts of how correlation manifests in the system.

Iris: And for econometricians, it offers a path to identify what variation is driving the work, which is crucial when you're trying to find a causal link.

The paper's summary: Ruth: Building on that structure, let’s talk about the core summary of this paper. It essentially says correlation emerges from explicit order-flow coupling between these two books, not some hidden multivariate diffusion.

Warren: Right, it moves the source of correlation from being some latent pre-specified relationship to being something active that happens in real time through trading activity.

Iris: The model places this cross-book mechanism specifically in gamma j k, and its price effect is transmitted through coupled order-density fields and moving reaction fronts as such, it’s sourced from the phenomenology of trading itself.

Ruth: So the phenomenology of actual trading—the creation and cancellation of orders—is what drives the correlation structure. It's not some abstract relationship we assume exists beforehand.

Warren: That means if you change the trading rules or event types in a way that changes how those order densities interact, you should see a corresponding change in the correlation emergence.

Iris: The summary also highlights that the Epps effect arises from two distinct mechanisms: one is subordination-only, and the other is coupling-only.

Ruth: The subordination mechanism is when log-prices evolve in operational time but are observed through independent clocks, and that’s why you see correlation reduced by limited clock overlap.

Warren: And the coupling mechanism is when you have calendar time aggregation interacting with the finite response of the pair-trader coupling. This dampens short-term correlations due to trader interaction latency.

Iris: They approximate this combined effect using a product formula, rho comb dt being approximated by rho sub dt times f(kappa delta t). That’s the combined mechanism approximation we discussed earlier.

Ruth: So, the summary boils down to separating timing issues from interaction issues when you try to explain why correlations are observed how they are.

Warren: For a trader, this means understanding the system requires looking at both the sampling rate and the speed of information transfer between assets.

Iris: It’s a way to identify which variation is doing the work by isolating these mechanisms, which helps in identifying what's driving correlation in real-time.

The paper's improvements: Ruth: Now for some of the suggested improvements the authors propose for this model. They suggest enhancing correlation modeling by explicitly separating clock overlap attenuation from finite coupling response effects.

Warren: That sounds like a direct improvement for practical application, allowing us to predict how correlation decays under different sampling frequencies and coupling strengths.

Iris: They want to develop a model incorporating the "subordination-only mechanism," which models where correlation is reduced purely by limited clock overlap, focusing on observation timing dynamics.

Ruth: That lets an AI system specifically model cross-asset dynamics governed purely by how we observe them, not necessarily the physical coupling strength.

Warren: And they want a "coupling-only mechanism" where correlation is modeled as a function of calendar time aggregation and the coupling response rate to quantify how trader interaction latency dampens short-term correlations.

Iris: They also suggest creating this "combined mechanism" approximation that multiplies the clock overlap factor with the coupling response factor, enabling AI to predict realized correlations when both are present.

Ruth: That combined approximation is basically a leading-order separable approximation, requiring the clock-overlap and local coupling-response terms to separate in expectation.

Warren: And they want fractional clock dynamics integrated into correlation models using Mittag-Leffler functions, which could capture sublinear variance growth under anomalous calendar-time sampling.

Iris: That suggests a way to model more complex temporal dependencies if you see non-linear behavior in the log-price observations under unusual sampling conditions.

Conclusion: Ruth: So we wrap up with the conclusion of this paper on "Correlation emergence and the Epps effect in two coupled limit order books." They get an analytical explanation from a discrete-to-continuum construction and a regularised representation of pair-trader coupling.

Warren: It confirms that correlation emerges from explicit order flow coupling, not some latent noise or pre-specified relationships.

Iris: They successfully distinguish attenuation due to incomplete clock overlap from attenuation due to the finite response of the pair-trader coupling.

Ruth: So, when we put it all together, we have a unified analytical account that separates those two things clearly for anyone looking at correlation dynamics in coupled systems.

Warren: It’s a useful tool because it allows us to see exactly where the modeling breaks down or where the fundamental mechanisms are at play.

Iris: For me, the ability to separate those two effects is what makes this work, and that’s what you can use to build better identification strategies.

Ruth: It gives us a solid foundation for understanding market dynamics by dissecting correlation emergence into its timing and interaction components.

Warren: We've seen how the paper tackles the Epps effect in two coupled limit order books, so that’s where we are with this one.

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

q-fin.TR, math.AP, q-fin.ST

Submitted: 2026-06-12

Updated: 2026-09-05

Comments: 16 pages, 1 figure, 7 appendices. Revised numerical implementation to conform to the fixed operational-time lattice and explicit calendar-time subordination, corrected numerical source and coupling treatment; the analytical results and main conclusions are unchanged. Reproducibility code: https://github.com/timgebbie/correlation-emergence-reproducibility

Journal ref: Physica A: Statistical Mechanics and its Applications, 2026, 132172

DOI: 10.1016/j.physa.2026.132172

Code: https://github.com/timgebbie/correlation-eme

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

Importance score: 87/100

The gist: The paper provides a unified analytic account of correlation emergence and the Epps effect in two coupled limit order books because it connects reaction–diffusion order books to reduced-form…

Key concepts

Reaction–Diffusion Order Books
This model treats the order book dynamics as a reaction-diffusion system. It describes how changes in price (reaction) spread through the order book structure (diffusion). This framework is used to analyze how order flow propagates and influences correlated prices across different books.
Epps Effect
The Epps effect describes the attenuation of realized correlation when observing prices from different, asynchronous clocks. The paper decomposes this effect into two parts: one caused by limited clock overlap and another caused by the finite response time of the coupling between traders.
Operational Time vs. Calendar Time
The model distinguishes between 'operational time,' which is a fixed grid based on trade events, and 'calendar time,' which is when prices are observed. This separation allows researchers to analyze how correlation changes based on the sampling frequency of the observation clocks relative to the underlying trading activity.

Terminology

Summary

The paper provides a unified analytic account of correlation emergence and the Epps effect in two coupled limit order books because it connects reaction–diffusion order books to reduced-form accounts of the Epps effect through a discrete-to-continuum construction. p. 0

Where the paper says it

Correlation emergence and the Epps effect in two coupled limit order books Chris Angstmanna , Tim Gebbieb aSchool of Mathematics and Statistics, University of New South Wales, Sydney, NSW 2052, Australia bDepartment of Statistical Sciences, University of Cape Town, Rondebosch 7701, Western Cape, South Africa Abstract We give a unified analytic account of correlation emergence and the Epps effect in two coupled limit order books. The Epps effect is the empirical reduction in measured cross-asset log-return correlation at short aggregation scales, or equivalently the recovery of measured correlation as the aggregation interval increases.

Correlation emerges from explicit order-flow coupling between the books rather than from correlated price noise or a latent multivariate diffusion. p. 4

Where the paper says it

We connect these strands by embedding correlation emergence directly into a coupled reaction–diffusion representation of multiple order books. Correlation emerges from explicit order-flow coupling between the books rather than from correlated price noise or a latent multivariate diffusion.

The model places the cross-book mechanism in γjk, with its price effect transmitted through the coupled order-density fields and moving reaction fronts and as such is sourced from the phenomenology of trading itself rather than being latent pre-specified relationships between trade or event types. p. 4
Where the paper says it

The cross-dependence is nevertheless structural in the latent event model 2 . Our model places the cross-book mechanism in γjk, with its price effect transmitted through the coupled order-density fields and moving reaction fronts and as such is sourced from the phenomenology of trading itself rather than 2For example, in the four-dimensional Hawkes price model of Chang et al.

Modeling the Coupled System

The system is modeled by separating order-book dynamics from observation clocks, defining operational time steps and calendar time via a clock mapping.

(A) The operational-time model log-mid-price for book j is the unique reaction-boundary zero crossing p(j)(u) = 0. p. 1

Where the paper says it

We focus on two books, although the ordered-pair notation extends directly to more assets. Moving reaction boundary and calendar clock The operational-time model log-mid-price for book j is the unique reaction-boundary zero crossing p (j) (u) = {x : φ (j) (x, u) = 0}.

The common operational lattice is defined by fixed log-price grid spacing and operational time step. p. 1

Where the paper says it

The coupled limit order books We separate the order-book dynamics from the clocks through which they are observed. Let xi = i∆x, un = n∆u, n ∈ N, (1) where ∆x > 0 is the fixed log-price-grid spacing, ∆u > 0 is the fixed operational-time step, u is operational time and t is reserved for calendar time.

The total source increment includes single-book order source, external shock, and ordered-pair source terms (3). p. 1

Where the paper says it

The exponent αu belongs to the order-density dynamics in operational time; the calendar-time clock exponent αt is introduced separately below. The total source increment is then: c i(j) n = s i(j) n + δ i(j) n + X k̸=j ℓ i(j,k) ε,n , (3) where s (j) is the one-book order source, δ (j) is an external shock and ℓ (j,k) ε is the ordered-pair source acting on book j in response to book k, with regularisation parameter ε.

Decomposition of the Epps Effect

The paper demonstrates that the Epps effect arises from two distinct mechanisms.

(A) The analytical treatment separates the reaction-boundary dynamics in operational time from the asynchronous clocks through which prices are observed, this identifies the attenuation due to clock overlap separately from the finite response of the pair-trader coupling. p. 0

Where the paper says it

Our construction places the cross-book mechanism directly with a mesoscopic pair-trader coupling whose price effect passes through the order-density fields and moving reaction fronts. Because the analytical treatment separates the reaction-boundary dynamics in operational time from the asynchronous clocks through which prices are observed, this identifies the attenuation due to clock overlap separately from the finite response of the pair-trader coupling.

The realized correlation is decomposed into a subordination-only mechanism and a coupling-only mechanism. p. 3

Where the paper says it

Subordination-only mechanism If log-prices evolve in operational time but are observed through independent clocks, the correlation is reduced by limited clock overlap. One finds ρ sub ∆t = ρu A12(∆t), (20) where A12(∆t) is an overlap factor and ρu is the instantaneous correlation of the two Brownian innovations in operational time.

(A) If log-prices evolve in operational time but are observed through independent clocks, the correlation is reduced by limited clock overlap. p. 3

Where the paper says it

Subordination-only mechanism If log-prices evolve in operational time but are observed through independent clocks, the correlation is reduced by limited clock overlap. One finds ρ sub ∆t = ρu A12(∆t), (20) where A12(∆t) is an overlap factor and ρu is the instantaneous correlation of the two Brownian innovations in operational time.

Mechanism Analysis

The realized correlation is approximated by a leading-order product form combining the clock and coupling effects.

(C) The product formula is a leading-order separable approximation, this requires the clock-overlap and local coupling-response terms to separate in expectation. p. 5

Where the paper says it

0 50 100 150 200 250 300 350 400 Calendar aggregation scale t [s] 0. 0 Normalized covariance response (a) Clock only Simulation 95% band Equal-rate clock theory F( clk ) Clock-only simulation 0 50 100 150 200 250 300 350 400 Calendar aggregation scale t [s] (b) Translation-mode coupling only Simulation 95% band F( ) theory Coupling-only simulation 0 50 100 150 200 250 300 350 400 Calendar aggregation scale t [s] (c) Clock and translation-mode coupling Combined simulation 95% band Leading-order product F( clk )F( ) Same-clock conditional reference Combined simulation Clock, coupling and combined Epps curves: theory and simulation Figure 1: Clock, coupling and the emergence of correlation.

The combined approximation is given by ρcomb∆t ≃ ρ∞A12(∆t)f(κ∆t). p. 3

Where the paper says it

Since the two boundaries move in opposite directions when the spread is nonzero, both motions reduce the same spread. Combined mechanism When both effects are present, a leading-order separability approximation gives ρ comb ∆t ≃ ρ∞A12(∆t)f(κ∆t).

The clock-only benchmark yields ρsub∆t ≃ ρuf(λclk12 ∆t), which approaches the instantaneous correlation as aggregation time increases (A. p. 3

Where the paper says it

One finds ρ sub ∆t = ρu A12(∆t), (20) where A12(∆t) is an overlap factor and ρu is the instantaneous correlation of the two Brownian innovations in operational time. For independent Poisson refresh clocks, using the pooled rate in the continuous exponential-lag aggregation form gives the tractable approximation ρ sub ∆t ≃ ρuf(λ clk 12 ∆t), (21) where the ordinary aggregation response is f(x) := 1 − 1 − e −x x (22) for x ≥ 0, and f(0) := 0.

21).

Numerical Validation and Correspondence

The numerical implementation follows the same separation used in the analysis, evolving books on a common operational lattice.

(5) The numerical implementation here now directly follows the current separation. p. 6

Where the paper says it

This provides an analytical counterpart and explanation to the prior numerical simulation studies [5, 9] which used a non-uniformly sampled grid implementation were the time mixing with the finite coupling response could not easily be separated. The numerical implementation here now directly follows the current separation.

p. 6
Where the paper says it

This provides an analytical counterpart and explanation to the prior numerical simulation studies [5, 9] which used a non-uniformly sampled grid implementation were the time mixing with the finite coupling response could not easily be separated. The numerical implementation here now directly follows the current separation.

The comparison between analytical and numerical fields shows that the correspondence is at the level of the reaction-boundary motion, not pointwise equality of the two source fields (D). p. 4

Where the paper says it

The translation-mode implementation is given by Equation (29); its projection onto the reactionboundary motion is described in Appendix D, with the analytical rate construction given in Appendices Appendix B and Appendix E. The comparison tests the reduced reaction-front response, not pointwise equality of the analytical and numerical source fields.

Conclusion

The paper obtains an analytical explanation of correlation emergence and the Epps effect from a discrete-to-continuum construction and a regularised representation of the pair-trader coupling.

(7) We obtain an analytical explanation of correlation emergence and the Epps effect from the discrete-tocontinuum construction of coupled order books and a regularised representation of the pair-trader coupling. p. 6

Where the paper says it

This analysis complements arguments that the Epps effect reflects the discrete-event nature of markets [8] by showing how that discreteness appears after the diffusion limit and which additional assumptions are required. Conclusion We obtain an analytical explanation of correlation emergence and the Epps effect from the discrete-tocontinuum construction of coupled order books and a regularised representation of the pair-trader coupling.

Separating reaction-boundary dynamics in operational time from asynchronous observation clocks distinguishes attenuation due to incomplete clock overlap from attenuation due to the finite coupling response. p. 6

Where the paper says it

Conclusion We obtain an analytical explanation of correlation emergence and the Epps effect from the discrete-tocontinuum construction of coupled order books and a regularised representation of the pair-trader coupling. Separating reaction-boundary dynamics in operational time from the asynchronous observation clocks distinguishes attenuation due to incomplete clock overlap from attenuation due to the finite coupling response.

The numerical implementation here now directly follows the current separation.

The component calculations recover the clock and coupling responses over the tested numerical regime (Panel (c)). p. 6

Where the paper says it

The two books evolve on a common operational lattice, the ordered coupling acts through the receiving reaction front, and the completed boundary paths are observed through separate previousrefresh clocks. Figure 1 shows that the component calculations recover the clock and coupling responses over the tested numerical regime.

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. Enhanced correlation modeling in trading systems by explicitly separating clock overlap attenuation from finite coupling response effects, allowing for a more nuanced understanding of Epps effect origins. This enables AI systems to better predict correlation decay under different sampling frequencies and coupling strengths.

  2. Development of a reaction diffusion order book model that incorporates the subordination-only mechanism where the correlation is reduced by limited clock overlap, enabling AI to model cross-asset dynamics governed purely by observation timing.

  3. Implementation of a coupling-only mechanism where correlation is modeled as a function of calendar time aggregation and coupling response rate, allowing AI to quantify how trader interaction latency dampens short-term correlations.

  4. Creation of a combined mechanism approximation that multiplies the clock overlap factor with the coupling response factor, enabling AI to predict realized correlations when both asynchronous observation and finite coupling times are present.

  5. Integration of fractional clock dynamics into correlation models using Mittag-Leffler functions, allowing AI to capture sublinear variance growth in log-price observations under anomalous calendar-time sampling.

  6. Development of a numerical implementation that separates operational time evolution from calendar time observation using distinct previous refresh clocks, providing a robust framework for simulating coupled order books with realistic clock dependencies.

  7. Calibration of coupling parameters by mapping them to the reaction boundary response rate kappajk, which allows AI systems to infer structural source parameters from observed correlation dynamics without requiring pointwise equality between analytical and numerical fields.

Abstract

We give a unified analytic account of correlation emergence and the Epps effect in two coupled limit order books. The Epps effect is the empirical reduction in measured cross-asset log-return correlation at short aggregation scales, or equivalently the recovery of measured correlation as the aggregation interval increases. The model starts from a fixed-grid discrete random walk for order flow in operational time, with creation, cancellation and diffusion. A pair-trader coupling between the books is introduced at the level of order creation, and calendar time is imposed through separate observation clocks. We clarify how the operational-time model reduces to coupled reaction--diffusion equations with a moving reaction boundary defining the model log-mid-price. Using a regularised local-response representation of the coupling, we derive approximate closed-form expressions for realised correlations as a function of aggregation time. Here the Epps effect is shown to arise from two distinct mechanisms: asynchronous observation clocks (subordination), finite coupling response times, and their combination.

Sources

Related papers