Non-Linear and Meta-Stable Dynamics in Financial Markets: Evidence from High Frequency Crypto Currency Market Makers

Igor Halperin

REVIEW 4 major objections 5 minor 13 references

Non-Linear and Meta-Stable Dynamics in Financial Markets: Evidence from High Frequency Crypto Currency Market Makers

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Using high-frequency crypto data, this paper claims log-prices follow strongly non-linear drifts that produce non-quadratic potentials, switching between single-well and double-well shapes with sampling frequency and market regime.

desk verdict Worth a serious look, but the meta-stability claim is not yet statistically secured: the paper shows clear nonlinear drifts in Uniswap v3 data, but the double-well potentials appear only in 2-month windows and come without error bars or a null-model test. read the letter →

arxiv 2509.02941 v1 pith:II5POA7S submitted 2025-09-03 q-fin.ST physics.comp-phq-fin.GNq-fin.TR

Igor Halperin This is my paper · ORCID

classification q-fin.STphysics.comp-phq-fin.GNq-fin.TR MSC 62P0560J6091G80

keywords non-lineardriftdouble-wellpotentialKramers-Moyalexpansioncryptocurrencymarketmicrostructureautomatedmakermetastabledynamicshigh-frequencyfinanceLangevinequation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tests a long-standing hypothesis that financial prices are not driven by linear diffusion but by non-linear drifts that can be read directly from data. It estimates the drift and volatility of crypto log-prices non-parametrically with a Kramers-Moyal expansion, then integrates the drift to recover an effective potential for the price. The central finding is that the drift is strongly non-linear at intraday sampling, the potential is non-quadratic, and its shape flips between a single well (stable regime) and a double well (metastable, stress-like regime) depending on the look-back window and sampling frequency. If correct, this gives direct empirical support to the author's earlier theory that market dynamics live in non-quadratic potentials, and offers a data-driven way to detect market stress from the topology of the recovered potential.

What carries the argument

The Kramers-Moyal expansion, which estimates the first two jump coefficients $K1(x)$ and $K2(x)$ from empirical time series without assuming a parametric model. These are converted to drift $\mu(x)=K1(x)/\Delta t$ and diffusion $\sigma^2(x)/2=K2(x)/(2\Delta t)$, then to the potential $U(x)$ by numerically integrating $\mu(x)=-\partial U/\partial x$. The shape of $U(x)$, single-well versus double-well, is the diagnostic that carries the argument.

What would settle it

Run the identical estimation pipeline on synthetic data generated from a linear Ornstein-Uhlenbeck process with the same irregular transaction timestamps and sample sizes; if linear inputs routinely produce comparable non-quadratic drifts or double wells, the reported nonlinearity is an artifact of the estimation procedure rather than genuine market structure.

Watch

Extended reading notes

Core claim

The paper claims that model-free estimates of the drift of crypto log-prices reject a linear force and instead show cubic-or-higher non-linearities, so the implied potential $U(x) = -\int \mu(x) dx$ is genuinely non-quadratic. In two-month windows of a dollar-anchored pool, the potential frequently has a double-well shape at sub-hour sampling, while shorter one-month windows show single wells, which the paper attributes to data insufficiency in the tails. Non-linearity weakens as sampling frequency drops, and a crypto-cross pair tends to single-well but still non-quadratic potentials. The double-well cases are interpreted as metastable dynamics: the 'particle' log-price sits in a local minimum

Load-bearing premise

Everything rests on treating the log-price as a one-dimensional Markov diffusion whose drift depends only on the current price, and on converting irregular transaction timings into a single average time step when turning estimated coefficients into physical drift and volatility.

Editorial extensions

If this is right

Intraday models that assume linear drifts miss the dominant source of non-linearity; state-dependent mean reversion matters at sub-hour scales.
The same market can appear stable at hourly sampling and metastable at ten-minute sampling, so potential topology is inherently time-scale dependent.
A double-well potential implies metastable quasi-equilibria whose lifetime is set by barrier height; rapid transitions between wells may be misattributed to elevated Brownian volatility if ignored.
The single-versus-double-well diagnostic works without parametric assumptions and can be applied to any high-frequency financial series.
Shorter observation windows (one month or less) may under-detect double wells due to tail insufficiency, so the choice of window drives qualitative conclusions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The conversion of irregular transaction intervals into a single average time step in Eq. (4) is a place where the pipeline could manufacture apparent non-linearity; re-estimating with arrival-time-weighted schemes or testing on synthetic data would clarify this.
Because the double-well structure appears in specific two-month windows, a natural next step is to align those windows with identifiable market-stress events such as volatility spikes or liquidity crunches in the underlying crypto market.
The same non-parametric machinery could be applied to intraday equity or FX data to see whether non-quadratic potentials are a general market feature or specific to automated-market-maker microstructure.
If the instanton interpretation is right, short-window options on the underlying should show related signatures, such as bimodal return distributions, offering an independent test of the metastable picture.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

Short version: this is a legitimate empirical test of a specific theory, applied to a new public dataset, and it produces a clear qualitative result—nonlinear drifts and non-quadratic potentials are the norm in intraday Uniswap v3 crypto data. But the paper's flashiest claim, the double-well (meta-stable) potentials, is built on a comparison that could easily be a regime-pooling artifact, and the absence of any uncertainty quantification makes it hard to know what to trust.

What's genuinely new: Kramers-Moyal analysis of high-frequency AMM data, using public transaction data from Uniswap v3 on Arbitrum. Unlike Wand et al. (2024) on stocks, who found support for quadratic drift but not cubic/quartic, this paper finds strong nonlinearity in the drift at short lags, and the weakening with sampling frequency is a nice empirical regularity. The author is also honest that one-month windows give single wells, and he flags the stochastic-drift alternative in Section 4.0.2. That's the kind of engagement you want.

The soft spots are real. First, no error bars anywhere. The drift and potential curves are presented as exact functions, and the difference between a single well and a double well can be a few points in the tails. A simple bootstrap over blocks or a null model—for example, simulate a linear drift with regime switching and run the same estimator—would test whether the double-well is real or an artifact of pooling heterogeneous periods. The paper does neither. Second, the window-length dependence: two-month windows show double wells, one-month windows show single wells, and the dismissal of the one-month result as 'data insufficiency' is not supported. The stress-test's alternative—that shorter windows are better approximations of a single regime—is equally plausible and never tested. Third, the conversion of Kramers-Moyal coefficients to physical drift using a single average dt for irregularly spaced data is a crude approximation; the paper doesn't discuss its effect on the shapes of the potentials.

None of this is fatal to the program. The non-quadratic, single-well potentials are already a solid empirical contribution. But the meta-stability (double-well/instanton) conclusion is not established. Who should read it: anyone working on market microstructure or stochastic models of crypto prices, and anyone who wants an example of how easy it is to over-read a non-parametric estimate. I'd send it to a serious referee—the data are public, the method is standard, and the weaknesses are addressable in revision—but my own verdict would be conditional, not accept.

Best, [You]

Referee Report

4 major / 5 minor

Summary. The paper analyzes high-frequency Uniswap v3 log-price data (USDC-WETH and WBTC-WETH on Arbitrum) using a non-parametric Kramers–Moyal expansion. From estimated conditional drift and volatility, it derives effective potentials U(x) for lags from 10 minutes to 6 hours and across two-month and one-month windows. The central claim is that the drift is strongly non-linear and the potential is non-quadratic, sometimes double-well, indicating metastable market states. The paper interprets double-well potentials as evidence supporting the author's earlier theoretical framework of non-linear market dynamics and instanton transitions.

Significance. If substantiated, the finding would provide empirical support for non-linear drift and metastable potentials in intraday crypto markets, which would be of interest for AMM microstructure, market-stress detection, and model specification. The non-parametric approach and the use of granular decentralized-exchange data are appropriate and commendable. However, the paper currently lacks the statistical machinery needed to distinguish genuine double-well dynamics from estimation artifacts, regime pooling, or finite-lag bias. The claimed evidence is therefore not yet secured, although the questions addressed are significant.

major comments (4)

[Sect. 2.2, Eq. (4)] The conversion K1(x)/Δt and K2(x)/(2Δt) uses a single average Δt for data that are irregularly spaced and averaged over 15 transactions, with typical spacing of 100 seconds. For non-equidistant observations, this estimator is not justified and has unknown bias. Since the potential U(x) is obtained by integrating μ(x), every reported well and barrier depends on this conversion. Please either use an estimator that accounts for irregular sampling, or provide a quantitative bound on the bias, or validate the procedure on simulated diffusions with known drift and irregular sampling.
[Sect. 3.1, Figs. 1 and 2] The double-well topology appears only for windows of at least two months and disappears for one-month windows. The paper dismisses the one-month result as 'data insufficiency in the tails' without supporting evidence. An equally plausible explanation is regime pooling: E[Δx|x] estimated over a long window mixes distinct market regimes, and a mixture of linear drifts can produce multiple zero crossings even when each regime's drift is linear. A null-model test using regime-switching linear diffusions, or a bootstrap that resamples within shorter sub-windows, is needed to show that the double wells are not an aggregation artifact.
[Figs. 1, 3, 4; Sect. 4.0.1] No uncertainty quantification is provided. The drift, volatility, and potential curves are shown as single lines, and the 'small local minimum' in Fig. 4 could easily be within sampling noise. Confidence bands from bootstrap or sub-sampling should be reported, and the barrier height separating a local from a global minimum should be compared with the estimation uncertainty. Without this, the metastable-state interpretation is not statistically supported.
[Sect. 4.0.1] The observed weakening of non-linearity with increasing lag is not by itself evidence of scale-dependent dynamics. For any diffusion with non-linear drift, finite-lag conditional moments generally become less non-linear as the lag grows, because the transition density is smoothed by diffusion. To support the multi-scale claim, the paper should compare the estimated lag dependence against that predicted by the estimated infinitesimal drift on simulated data, or against an explicit finite-lag formula.

minor comments (5)

[Figs. 3 and 4 captions] The captions refer to 'UBTC-WETH' while the text discusses 'WBTC-WETH'. Also, Fig. 4 is described in Sect. 3.3 as the USDC-WETH 2025 pool, but the caption says 'UBTC-WETH 2025'. Please correct the naming to avoid confusion.
[Sect. 3.1] Typo: 'onbtained' should be 'obtained'.
[Sect. 2.1] The statement that each observation is the log-price 'averaged over 15 consecutive transactions' is ambiguous. Clarify whether this is an average of log-prices or of transaction prices, and how the timestamp for the averaged observation is defined.
[Sect. 2.2] The conversion to 'annualized' rates is not described. Please give the scaling factors used and clarify whether the same scaling is applied at all lags.
[Figs. 1–4] The figures contain many overlapping curves. The paper would benefit from clearer line styles, error bands, or separate panels so that the reader can identify individual lags and windows.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical estimation is self-contained; self-citations are motivational/interpretive.

full rationale

The paper's core empirical claim is obtained by non-parametric Kramers–Moyal estimation of drift and diffusion from price data, followed by numerical integration to obtain a potential. This is a direct data-driven calculation; the prior theory is not used as an input to the estimation. Equation (4) defines the physical drift as K1/Δt, and Equation (3) defines the potential as the negative integral of the drift, so the potential is a mathematical transform of the estimated drift, not a fitted prediction of the theory. The self-citations [7]–[12] are used only to motivate the hypothesis and to interpret the results (e.g., instantons, marketron language), not to derive the observed potential shapes. The choice of two-month windows and the interpretation of one-month results as data insufficiency are methodological judgments that could affect robustness, but they do not make the derivation circular. No equation or parameter is constructed from the target conclusion; the Kramers–Moyal estimator is model-free and could in principle have produced linear drifts and purely quadratic potentials. Therefore no circular step can be exhibited.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The analysis introduces no fitted parameters or new entities; its load-bearing inputs are the diffusion assumption and the deterministic-drift assumption. Preprocessing choices (0.5% trimming, 15-transaction averaging) and window selection are not formal parameters but affect results.

assumptions (3)

domain assumption The log-price follows a one-dimensional Markov diffusion process (Langevin equation) with Gaussian white noise.
Invoked in Sect. 2.2, Eqs. (1)-(2). If the true process has jumps or non-Markovian memory, the Kramers-Moyal estimated drift is not the physical drift.
domain assumption The drift is a deterministic function of the current log-price only.
Stated in Sect. 2.2 and discussed in Sect. 4.0.2. Hidden stochastic factors in the drift are considered but assumed to be slow-varying or averaged out.
domain assumption The Kramers-Moyal coefficients estimated from discrete, irregularly spaced data with a single average time step are unbiased estimators of the continuous-time drift and diffusion.
Used in Eq. (4). The data have a typical time step of about 100 seconds but are not evenly spaced, and no bias correction is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Non-Linear and Meta-Stable Dynamics in Financial Markets: Evidence from High Frequency Crypto Currency Market Makers." pith.science (2026). https://pith.science/paper/II5POA7S

@misc{pith2026250902941,
  author       = {Pith},
  title        = {Pith review of: Non-Linear and Meta-Stable Dynamics in Financial Markets: Evidence from High Frequency Crypto Currency Market Makers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/II5POA7S}},
  note         = {Machine review of arXiv:2509.02941}
}

read the original abstract

This work builds upon the long-standing conjecture that linear diffusion models are inadequate for complex market dynamics. Specifically, it provides experimental validation for the author's prior arguments that realistic market dynamics are governed by higher-order (cubic and higher) non-linearities in the drift. As the diffusion drift is given by the negative gradient of a potential function, this means that a non-linear drift translates into a non-quadratic potential. These arguments were based both on general theoretical grounds as well as a structured approach to modeling the price dynamics which incorporates money flows and their impact on market prices. Here, we find direct confirmation of this view by analyzing high-frequency crypto currency data at different time scales ranging from minutes to months. We find that markets can be characterized by either a single-well or a double-well potential, depending on the time period and sampling frequency, where a double-well potential may signal market uncertainty or stress.

Figures

Figures reproduced from arXiv: 2509.02941 by the authors.

**Figure 2.** Comparison of estimated drift, volatility, and potential [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Comparison of estimate drift, volatility, and potential [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Comparison of estimated drift, volatility, and potential [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

**Figure 1.** Figure 1: For a sufficiently high barrier, such metastable state can live long, making escape from such state [PITH_FULL_IMAGE:figures/full_fig_p010_1.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

[1]

Uniswap v3 Core

H. Adams, N. Zinsmeister, M. Salem, R. Keefer, and D. Robinson, “Uniswap v3 Core”. Uniswap Labs, (2021)

work page 2021
[2]

The Pricing of Options and Corporate Liabilities

Black, F., & Scholes, M. (1973). “The Pricing of Options and Corporate Liabilities”. Journal of Political Economy, 81(3), 637-654

work page 1973
[3]

Dash, Quantitative Finance and Risk Management: a Physicist’s Approach , Second Edition, World Scientific (2016)

J.W. Dash, Quantitative Finance and Risk Management: a Physicist’s Approach , Second Edition, World Scientific (2016)

work page 2016
[4]

Gardiner, Handbook of Stochastic Methods, Third Ed., Springer (2004)

C.M. Gardiner, Handbook of Stochastic Methods, Third Ed., Springer (2004)

work page 2004
[5]

A Disruptive Research Playbook for Studying Disruptive Innovations

Georgiev, G.“Implied volatility in decentralized finance pool”. arXiv preprint arXiv:2402.13329, (2024)

work page Pith review arXiv 2024
[6]

kramersmoyal: Kramers--Moyal coefficients for stochastic processes

L.R. Gorj ˜ao and F. Meirinhos, “kramersmoyal: Kramers–Moyal Coefficients for Stochastic Processes”, https://arxiv.org/abs/1912.09737 Github: https://github.com/LRydin/KramersMoyal (2019)

work page Pith review arXiv 1912
[7]

Quantum Equilibrium-Disequilibrium: Asset Price Dynam- ics, Symmetry Breaking, and Defaults as Dissipative Instantons

I. Halperin and M.F. Dixon, “Quantum Equilibrium-Disequilibrium: Asset Price Dynam- ics, Symmetry Breaking, and Defaults as Dissipative Instantons”, Physica A 537, 122187, https://doi.org/10.1016/j.physa.2019.122187 (2020)

work page arXiv 2019
[8]

The Inverted Parabola World of Classical Quantitative Finance: Non-Equilibrium and Non-Perturbative Finance Perspective

I. Halperin, “The Inverted World of Classical Quantitative Finance: a Non-Equilibrium and Non- Perturbative Finance Perspective”, https://arxiv.org/abs/2008.03623 (2020)

work page Pith review arXiv 2008

Show all 13 references

[9]

Non-Equilibrium Skewness, Market Crises, and Option Pricing: Non-Linear Langevin Model of Markets with Supersymmetry

I. Halperin, “Non-Equilibrium Skewness, Market Crises, and Option Pricing: Non-Linear Langevin Model of Markets with Supersymmetry”, Physica A 594, 127065 (2022)

2022
[10]

Phases of MANES: Multi-Asset Non-Equilibrium Skew Model of a Strongly Non- Linear Market with Phase Transitions

I. Halperin, “Phases of MANES: Multi-Asset Non-Equilibrium Skew Model of a Strongly Non- Linear Market with Phase Transitions”. Reviews in Modern Quantitative Finance, ed. A. Itkin, World Scientific, vol.1, pp. 97–150 (2024). 11

2024
[11]

Marketron games: Self-propelling stocks vs dumb money and metastable dynamics of the Good, Bad and Ugly markets

I. Halperin and A. Itkin, “Marketron games: Self-propelling stocks vs dumb money and metastable dynamics of the Good, Bad and Ugly markets”, https://arxiv.org/abs/2501.12676 (2025)

2025 arXiv
[12]

Marketron Through the Looking Glass: From Equity Dynamics to Option Pricing in Incomplete Markets

I. Halperin and A. Itkin, “Marketron Through the Looking Glass: From Equity Dynamics to Option Pricing in Incomplete Markets”,https://arxiv.org/abs/2508.09863 (2025)

2025 arXiv
[13]

Estimating Stable Fixed Points and Langevin Potentials for Financial Dynamics

T. Wand, T. Wiedemann, J. Harren, and O. Kamps, “Estimating Stable Fixed Points and Langevin Potentials for Financial Dynamics”. Physical Review E, 109, 024226. arXiv:2309.12082 (2024). 12

2024 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Uniswap v3 Core

H. Adams, N. Zinsmeister, M. Salem, R. Keefer, and D. Robinson, “Uniswap v3 Core”. Uniswap Labs, (2021)

work page 2021

[2] [2]

The Pricing of Options and Corporate Liabilities

Black, F., & Scholes, M. (1973). “The Pricing of Options and Corporate Liabilities”. Journal of Political Economy, 81(3), 637-654

work page 1973

[3] [3]

Dash, Quantitative Finance and Risk Management: a Physicist’s Approach , Second Edition, World Scientific (2016)

J.W. Dash, Quantitative Finance and Risk Management: a Physicist’s Approach , Second Edition, World Scientific (2016)

work page 2016

[4] [4]

Gardiner, Handbook of Stochastic Methods, Third Ed., Springer (2004)

C.M. Gardiner, Handbook of Stochastic Methods, Third Ed., Springer (2004)

work page 2004

[5] [5]

A Disruptive Research Playbook for Studying Disruptive Innovations

Georgiev, G.“Implied volatility in decentralized finance pool”. arXiv preprint arXiv:2402.13329, (2024)

work page Pith review arXiv 2024

[6] [6]

kramersmoyal: Kramers--Moyal coefficients for stochastic processes

L.R. Gorj ˜ao and F. Meirinhos, “kramersmoyal: Kramers–Moyal Coefficients for Stochastic Processes”, https://arxiv.org/abs/1912.09737 Github: https://github.com/LRydin/KramersMoyal (2019)

work page Pith review arXiv 1912

[7] [7]

Quantum Equilibrium-Disequilibrium: Asset Price Dynam- ics, Symmetry Breaking, and Defaults as Dissipative Instantons

I. Halperin and M.F. Dixon, “Quantum Equilibrium-Disequilibrium: Asset Price Dynam- ics, Symmetry Breaking, and Defaults as Dissipative Instantons”, Physica A 537, 122187, https://doi.org/10.1016/j.physa.2019.122187 (2020)

work page arXiv 2019

[8] [8]

The Inverted Parabola World of Classical Quantitative Finance: Non-Equilibrium and Non-Perturbative Finance Perspective

I. Halperin, “The Inverted World of Classical Quantitative Finance: a Non-Equilibrium and Non- Perturbative Finance Perspective”, https://arxiv.org/abs/2008.03623 (2020)

work page Pith review arXiv 2008