Pith

open record

sign in

arxiv: 2607.06324 · v1 · pith:JU4BM7EA · submitted 2026-07-07 · cond-mat.stat-mech · physics.data-an

Robust q-negative Multifractal Detrended Cross-Correlation Coefficient

Thiago B. Murari , Jos\'e Fernando F. Mendes , Hernane B. B. Pereira , Marcelo A. Moret This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-07-08 09:43 UTCglm-5.2pith:JU4BM7EArecord.jsonopen to challenge →

classification cond-mat.stat-mech physics.data-an
keywords multifractal cross-correlationdetrended fluctuation analysisCauchy-Schwarz inequalitynonstationary time seriesfluctuation order qbounded correlation coefficientfinancial marketsmeteorological data
0
0 comments X

The pith

Bounded multifractal cross-correlation coefficient works for all q

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

The paper introduces a new cross-correlation coefficient for nonstationary time series, called the Signed Multifractal Detrended Cross-Correlation Coefficient (ρ_SMFDCCA), that remains strictly bounded within [-1, 1] for every value of the fluctuation order parameter q, including negative q. Prior multifractal cross-correlation coefficients suffer from numerical instabilities, unbounded values, and erratic behavior when q is negative, because negative q amplifies segments with very small fluctuations, making the coefficient extremely sensitive to noise and numerical error. The existing workaround is a corrective post-processing step that inverts the coefficient whenever it exceeds the expected bounds. The authors eliminate this problem by changing what is averaged: instead of computing a global ratio of q-weighted covariance to q-weighted variance (which can diverge), they first compute a local correlation coefficient r_v in each segment, which is individually bounded by the Cauchy-Schwarz inequality, and then take a weighted average of these bounded local coefficients using amplitude-based weights. Because a weighted average of numbers each lying in [-1, 1] must itself lie in [-1, 1], the result is automatically bounded for all q, with no corrective procedures needed. The authors validate the coefficient on independent fractional Gaussian noise (no spurious cross-correlations detected), Dow Jones versus NASDAQ daily returns (stronger synchronization during large-amplitude fluctuations than small-amplitude ones, with convergence at longer scales), and meteorological data from São Paulo (heterogeneous coupling patterns between temperature and humidity variables across scales and amplitudes).

Core claim

The central discovery is a reformulation of the multifractal cross-correlation coefficient that achieves boundedness not by post-hoc correction but by construction. The key move is to normalize locally first—computing a per-segment detrended correlation coefficient r_v = C_xy(n,v) / A_v(n), where C_xy is the local detrended covariance and A_v is the geometric mean of the local detrended variances—and only then apply the q-dependent amplitude weighting. Because each r_v is individually confined to [-1, 1] by Cauchy-Schwarz, and because the weights w_v = A_v^{q/2} are non-negative, the weighted average ρ_SMFDCCA = Σ w_v r_v / Σ w_v is a convex combination of bounded quantities and is therefore

What carries the argument

Local detrended correlation coefficient r_v(n) = C_xy(n,v) / A_v(n), bounded by Cauchy-Schwarz; amplitude weights w_v(q,n) = A_v(n)^{q/2}; signed weighted average ρ_SMFDCCA(n,q) = Σ w_v r_v / Σ w_v; regularization floor ε = 10^{-12} preventing division-by-zero when local variances vanish; the Cauchy-Schwarz inequality as the structural guarantee of boundedness.

If this is right

  • Any downstream method built on the multifractal cross-correlation coefficient—including the recently proposed multivariate multiple cross-correlation coefficient qDMC²_x(n)—can inherit the boundedness property by adopting the local-normalization-first construction, eliminating embedded instabilities for negative q.
  • Empirical studies that previously restricted analysis to q ≥ 0 to avoid instability can now probe the small-fluctuation regime (q < 0) with a well-defined, interpretable coefficient, potentially revealing amplitude-dependent correlation structures that were systematically missed.
  • The amplitude-stratified view of cross-correlations could serve as a diagnostic tool for regime identification in financial markets: the observed convergence of q-dependent correlations at long scales suggests a transition from heterogeneous (amplitude-dependent) to homogeneous (amplitude-independent) coupling, which could be used to characterize market maturity or stress regimes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper's acknowledgment that ρ_SMFDCCA is not designed to estimate generalized Hurst exponents or multifractal spectra raises the question of whether the local-normalization step alters the scaling information content relative to the standard q-dependent fluctuation formalism. A direct comparison of the scaling exponents derivable from the two formulations on data with known multifractal struct
  • The choice to weight by A_v^{q/2} rather than A_v^q means the weighting exponent is half that used in standard MFDFA/MFCCA. This halving may affect the effective q-range sensitivity of the observable relative to the standard formalism, and a systematic comparison of the q-dependent profiles produced by the two weighting schemes on the same data would test whether the amplitude-conditioning is quan
  • The regularization parameter ε is shown to be numerically inert over 14 orders of magnitude, but this insensitivity itself suggests that for sufficiently small local amplitudes the coefficient effectively truncates the weighting of those segments. For data with genuinely near-zero-amplitude regions (e.g., overnight periods in intraday financial data), this truncation could systematically bias the

Load-bearing premise

The claim that ρ_SMFDCCA is a meaningful multifractal observable depends on the assumption that weighting local correlation coefficients by the local fluctuation amplitude raised to the power q/2 preserves the multifractal cross-correlation structure in a way comparable to the standard q-dependent fluctuation formalism. The paper itself states that the coefficient is not designed to estimate generalized Hurst exponents, multifractal spectra, or singularity distributions, so a

What would settle it

Generate two coupled time series with a known, analytically specified multifractal cross-correlation structure (e.g., coupled binomial cascades with a known cross-correlation scaling exponent). Compute both ρ_SMFDCCA(n,q) and the standard ρ_q(n) across the full q-range. If ρ_SMFDCCA fails to reproduce the known amplitude-dependent correlation structure that ρ_q captures in the positive-q regime, or if the q-dependent profile of ρ_SMFDCCA does not correspond to any recoverable multifractal cross-correlation spectrum, then the coefficient is a bounded correlation measure but not a multifractal

Figures

Figures reproduced from arXiv: 2607.06324 by Hernane B. B. Pereira, Jos\'e Fernando F. Mendes, Marcelo A. Moret, Thiago B. Murari.

Figure 1
Figure 1. Figure 1: FIG. 1. Evaluation of [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
read the original abstract

The multifractal detrended cross-correlation coefficient $\rho_q(n)$ is widely used to investigate scale-dependent interactions, but its application to negative fluctuation orders is affected by numerical instabilities, unbounded values, and interpretational difficulties. We propose a Signed Multifractal Detrended Cross-Correlation Coefficient, $\rho_{\mathrm{SMFDCCA}}(n,q)$, an amplitude-conditioned correlation observable for multifractal detrended analysis, based on locally normalized detrended correlations and regularized fluctuation amplitudes. The proposed coefficient preserves the sign of local interactions, remains strictly bounded within $[-1,1]$ for both positive and negative values of $q$, and eliminates the corrective procedures required by previous approaches. Validation using independent fractional Gaussian noise confirms the absence of spurious cross-correlations and the numerical stability of the method. Applications demonstrate that the proposed observable resolves how cross-correlations evolve jointly with temporal scale and fluctuation amplitude, revealing scale- and amplitude-dependent correlation structures, including stronger synchronization during large fluctuations in stock-market indices and heterogeneous coupling patterns in temperature records.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 7 minor

Summary. The manuscript introduces a Signed Multifractal Detrended Cross-Correlation Coefficient, ρ_SMFDCCA(n, q), designed to remain strictly bounded within [-1, 1] for all q ∈ ℝ, including negative fluctuation orders where the standard multifractal cross-correlation coefficient ρ_q(n) becomes numerically unstable. The construction is straightforward: local detrended correlation coefficients r_v(n) are defined per segment (Eq. 9), shown to be bounded by Cauchy-Schwarz (Eq. 10), and combined via a weighted average using non-negative amplitude-based weights w_v = A_v^{q/2} (Eqs. 11, 16), yielding a bounded convex combination (Eq. 17). The method is validated on independent fGn (Fig. 1) and applied to U.S. equity indices (Fig. 2) and Brazilian weather data (Fig. 3). Code and data are publicly available.

Significance. The boundedness proof is correct and derived from first principles: each r_v is bounded by Cauchy-Schwarz (Eq. 10), weights are non-negative (Eq. 16), and a weighted average of bounded quantities is bounded (Eq. 17). The regularization parameter ε = 10^{-12} is a numerical floor, not a fitted constant, and the sensitivity analysis in the End Matter confirms a broad stability plateau across fourteen orders of magnitude. The fGn null-model validation (Fig. 1) appropriately confirms near-zero values (~10^{-3}). The authors provide reproducible code and data via Zenodo, which is a strength. The paper is transparent about the observable's scope, explicitly stating it is not designed to estimate generalized Hurst exponents or multifractal spectra.

major comments (2)
  1. The central framing of the paper positions ρ_SMFDCCA as overcoming the limitations of the standard ρ_q(n) and eliminating its corrective procedures (Abstract, Introduction, Summary). However, the two quantities are structurally different observables: ρ_q(n) is a ratio of q-th-order fluctuation functions F_xy(q,n)/√(F_xx·F_yy), while ρ_SMFDCCA is an amplitude-weighted average of local Pearson-type correlation coefficients r_v. Even for q > 0 where both are well-defined, there is no a priori reason they should track the same structure (e.g., at q=2, ρ_SMFDCCA involves signed covariances C_xy in the numerator via r_v, whereas ρ_2 involves a different normalization). No head-to-head comparison of ρ_SMFDCCA and ρ_q on any dataset is provided, even in the q > 0 regime where ρ_q is stable. Without this, the claim of being a 'replacement' is unsubstantiated. The authors should either (a) add a直接
  2. The use of the term 'multifractal' for ρ_SMFDCCA requires more careful justification. The authors acknowledge (paragraph following Eq. 17) that the coefficient 'is not designed to estimate generalized Hurst exponents, multifractal spectra, or singularity distributions.' The q-dependence here is an amplitude-weighting of local correlation coefficients, not the moment-based scaling that defines multifractal analysis. The paper should more clearly delineate what 'multifractal' means in this context versus the standard usage, and whether the observable captures amplitude-dependent cross-correlation structure comparable to the multifractal formalism.
minor comments (7)
  1. Eq. (17): the subscript reads ρ_MFDCCA but should be ρ_SMFDCCA for consistency.
  2. The Limitations section uses W_i and Amplitude_i (Eq. 18) whereas the main text uses w_v and A_v. Please unify the notation.
  3. Fig. 1: the caption states maximum absolute deviations below 5×10^{-3}, but the figure itself does not clearly show this scale. Consider adding a colorbar with finer resolution or showing the maximum in an inset.
  4. The text following Eq. (8) states ε is chosen 'without loss of generality.' This is a mathematical claim that is not quite appropriate for a numerical parameter; consider rephrasing the 'without loss of generality' statement.
  5. Reference [13] is cited as Physica A 689, 131424 (2026). The volume/year appears unusual; please verify.
  6. The phrase 'proving the efficacy of the proposed observable in characterize multiscale co-movements' (U.S. equity markets section) contains a grammatical error; please revise.
  7. The Introduction contains several instances of passive voice without clear subjects (e.g., 'was proposed the DCCA cross-correlation coefficient,' 'was proposed the multifractal cross-correlation coefficient'). Please revise for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for a careful and constructive reading of our manuscript. The referee correctly identifies the boundedness proof as the core contribution and acknowledges the numerical stability, null-model validation, and reproducibility. Below we address each major comment in turn.

read point-by-point responses

  1. Referee: Major Comment 1: The central framing positions ρ_SMFDCCA as overcoming limitations of ρ_q and eliminating corrective procedures, but the two are structurally different observables. No head-to-head comparison is provided, even for q > 0 where ρ_q is stable. The claim of being a 'replacement' is unsubstantiated.

    Authors: The referee raises a valid and important point. We agree that ρ_SMFDCCA(n,q) and ρ_q(n) are structurally distinct observables: ρ_q is a ratio of q-th-order fluctuation functions F_xy(q,n)/√(F_xx(q,n)·F_yy(q,n)), whereas ρ_SMFDCCA is an amplitude-weighted convex combination of locally normalized Pearson-type correlation coefficients r_v(n). Because the normalization and averaging structures differ, there is no a priori guarantee that the two observables coincide even in the q > 0 regime, and we should not have framed ρ_SMFDCCA as a direct 'replacement' for ρ_q without empirical evidence of their relationship. We will make two changes in the revised manuscript. First, we will add a head-to-head comparison of ρ_SMFDCCA and ρ_q on the same datasets (fGn, DJI–IXIC, and the São Paulo weather pairs) in the q > 0 regime where ρ_q is well-defined and stable. This will show where the two observables agree, where they diverge, and what each reveals. Second, we will revise the framing throughout the Abstract, Introduction, and Summary to position ρ_SMFDCCA as a complementary bounded observable that addresses the specific problem of q < 0 instability, rather than as a wholesale replacement for ρ_q. The claim that ρ_SMFDCCA 'eliminates the corrective procedures required by previous approaches' will be retained in the narrow sense that, for applications requiring bounded signed cross-correlation measures across the full q-range including q < 0, our formulation does not require post-hoc inversion—but we will make clear that this applies to the specific use case of negative-q analysis, not to a claim of equivalence with ρ_q for q > 0. revision: yes

  2. Referee: Major Comment 2: The use of the term 'multifractal' for ρ_SMFDCCA requires more careful justification. The q-dependence is amplitude-weighting of local correlation coefficients, not moment-based scaling. The paper should more clearly delineate what 'multifractal' means here versus standard usage.

    Authors: We agree that the term 'multifractal' requires careful delineation. The manuscript already contains two passages that address this distinction: the paragraph following Eq. 17 states that 'the coefficient is not designed to estimate generalized Hurst exponents, multifractal spectra, or singularity distributions,' and the Introduction notes that 'the term multifractal refers to the q-dependent weighting of fluctuation amplitudes.' However, we acknowledge that these statements are not sufficiently prominent or explicit, and a reader could reasonably expect a more thorough justification of the terminology. In the revised manuscript, we will add a dedicated paragraph (or an explicit remark) that clearly states the following: (i) in standard multifractal analysis (MFDFA, MFCCA), q-dependence arises through moment-based scaling of partition functions, yielding generalized Hurst exponents and multifractal spectra; (ii) in ρ_SMFDCCA, q-dependence arises through amplitude-based weighting of bounded local correlation coefficients, producing an amplitude-stratified family of correlation observables rather than a scaling exponent; (iii) we retain the term 'multifractal' because the observable operates within the same detrended fluctuation framework and uses the same q-dependent amplitude-filtering mechanism that underlies MFCCA, but we will explicitly state that the observable does not produce multifractal spectra or scaling exponents. We believe this terminology is defensible given that the q-filtering of fluctuation amplitudes is shared with the multifractal formalism, but we agree that the distinction must be stated more prominently and unambiguously. revision: yes

Circularity Check

0 steps flagged

No circularity: boundedness is derived from Cauchy-Schwarz and convex combination; no fitted parameters or self-citation chains load the central claim.

full rationale

The paper's central claim is that ρ_SMFDCCA(n,q) ∈ [-1,1] for all q ∈ ℝ. The derivation is self-contained and non-circular: (1) each local correlation r_v(n) = C_xy(n,v)/A_v(n) is bounded by Cauchy-Schwarz (Eq. 9-10), a standard external mathematical fact; (2) weights w_v(q,n) = A_v(n)^{q/2} are non-negative by construction (Eq. 11, 16); (3) ρ_SMFDCCA is defined as a normalized weighted average of the r_v (Eq. 13), so boundedness follows from the elementary property that a convex combination of quantities in [-1,1] remains in [-1,1] (Eq. 17). No parameter is fitted to the target result. The regularization ε = 10^{-12} is a numerical floor, not a fitted constant, and the sensitivity analysis (End Matter) confirms independence over 14 orders of magnitude. The empirical applications use external datasets (fGn, DJI/NASDAQ, weather) and do not feed back into the mathematical proof. Self-citations (Ref. [1], [7], [9], [10]) appear in the introduction and context, not as load-bearing premises for the boundedness derivation. The paper acknowledges that ρ_SMFDCCA does not estimate Hurst exponents or multifractal spectra, which is a scope limitation, not a circularity. No step in the derivation chain reduces to its own inputs by construction, fitting, or self-citation. The correct concern about whether ρ_SMFDCCA captures the same multifractal information as ρ_q is a question of correctness and comparability, not circularity, and the paper does not claim equivalence to ρ_q by definition. The derivation is self-contained against external mathematical facts and external benchmarks, so the circularity score is 0.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

The method introduces no new physical entities or postulated objects. The free parameter ε is a numerical safeguard with demonstrated insensitivity. The polynomial order m is a standard DFA parameter left unspecified. The axioms are either standard mathematical inequalities or a domain assumption about the interpretability of amplitude-weighted local correlations.

free parameters (2)
  • ε (regularization constant) = 1e-12
    Numerical floor to prevent division by zero when local detrended variances vanish (Eq. 8). Not fitted to data; sensitivity analysis in End Matter confirms zero RMSE across ε ∈ [10^-20, 10^-6].
  • m (polynomial detrending order) = Not explicitly stated in text
    Standard parameter in DFA/DCCA methods; the paper references polynomial trends of order m (Eq. 2) but does not specify the value used for the empirical results.
axioms (3)
  • standard math Cauchy-Schwarz inequality applied to detrended residuals
    Invoked in Eq. (10) to establish that each local correlation coefficient r_v(n) is bounded in [-1, 1]. This is the mathematical foundation of the boundedness claim.
  • standard math Non-negativity of amplitude-based weights
    Invoked in Eq. (16) to establish that w_v(q,n) = A_v(n)^{q/2} ≥ 0, ensuring that the weighted average of bounded r_v values is itself bounded.
  • domain assumption Weighting by A_v^{q/2} preserves multifractal cross-correlation information
    The paper assumes that using local fluctuation amplitudes raised to q/2 as weights (Eq. 11) provides a meaningful amplitude-conditioned characterization of cross-correlations. The paper partially disclaims this by stating the coefficient is 'not designed to estimate generalized Hurst exponents' but still frames it as multifractal.

pith-pipeline@v1.1.0-glm · 12490 in / 2766 out tokens · 460770 ms · 2026-07-08T09:43:55.111864+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

  1. [1]

    Murari, T

    T.DiMatteo, M.A.Moret, H.B.deBarrosPereira, T.B. Murari, T. M. da Rocha Filho, and J. F. F. Mendes, The European Physical Journal B98, 171 (2025)

    work page 2025

  2. [2]

    C.-K. Peng, S. V. Buldyrev, S. Havlin, M. Simons, H. E. Stanley, and A. L. Goldberger, Physical Review E49, 1685 (1994)

    work page 1994

  3. [3]

    Podobnik and H

    B. Podobnik and H. E. Stanley, Physical Review Letters 100, 084102 (2008)

    work page 2008

  4. [4]

    Havlin, A

    J.W.Kantelhardt, S.A.Zschiegner, E.Koscielny-Bunde, S. Havlin, A. Bunde, and H. E. Stanley, Physica A: Sta- tistical Mechanics and its Applications316, 87 (2002)

    work page 2002

  5. [5]

    Zhou, Physical Review E77, 066211 (2008)

    W.-X. Zhou, Physical Review E77, 066211 (2008)

    work page 2008

  6. [6]

    G. F. Zebende, Physica A: Statistical Mechanics and its Applications390, 614 (2011)

    work page 2011

  7. [7]

    Nascimento Filho, E

    A. Nascimento Filho, E. Pereira, P. Ferreira, T. Murari, and M. Moret, Physica A: Statistical Mechanics and its Applications508, 550 (2018)

    work page 2018

  8. [8]

    Ferreira, L

    P. Ferreira, L. Kristoufek, and E. J. d. A. L. Pereira, Physica A: Statistical Mechanics and its Applications 545, 123803 (2020)

    work page 2020

  9. [9]

    G. A. Azevedo, R. R. Sampaio, A. S. N. Filho, M. A. Moret, and T. B. Murari, Scientific reports11, 791 (2021)

    work page 2021

  10. [10]

    J. B. Oliveira, T. B. Murari, A. S. Nascimento Filho, H. Saba, M. A. Moret, and C. A. L. Cardoso, Science of The Total Environment860, 160491 (2023)

    work page 2023

  11. [11]

    M. B. Figueredo, R. L. S. Monteiro, A. do Nasci- mento Silva, J. R. de Araújo Fontoura, A. R. da Silva, and C. A. P. Alves, Scientific Reports13, 7512 (2023)

    work page 2023

  12. [12]

    Kwapień, P

    J. Kwapień, P. Oświęcimka, and S. Drożdż, Physical Re- view E92, 052815 (2015)

    work page 2015

  13. [13]

    A. M. da Silva Filho, A. P. N. de Castro, and E. F. Guedes, Physica A: Statistical Mechanics and its Appli- cations689, 131424 (2026)

    work page 2026

  14. [14]

    Sierra-Porta, Journal of Atmospheric and Solar- Terrestrial Physics266, 106407 (2025)

    D. Sierra-Porta, Journal of Atmospheric and Solar- Terrestrial Physics266, 106407 (2025)

    work page 2025

  15. [15]

    Wątorek, S

    M. Wątorek, S. Drożdż, J. Kwapień, L. Minati, P. Oświęcimka, and M. Stanuszek, Physics Reports901, 1 (2021)

    work page 2021

  16. [16]

    Jiang, W.-J

    Z.-Q. Jiang, W.-J. Xie, W.-X. Zhou, and D. Sornette, Reports on Progress in Physics82, 125901 (2019)

    work page 2019

  17. [17]

    Di Matteo, Quantitative Finance7, 21 (2007)

    T. Di Matteo, Quantitative Finance7, 21 (2007)

    work page 2007

  18. [18]

    Barunik, T

    J. Barunik, T. Aste, T. Di Matteo, and R. Liu, Physica A: Statistical Mechanics and its Applications391, 4234 (2012)

    work page 2012

  19. [19]

    R. N. Mantegna, The European Physical Journal B- Condensed Matter and Complex Systems11, 193 (1999)

    work page 1999

  20. [20]

    T. W. Epps, Journal of the American Statistical Associ- ation74, 291 (1979)

    work page 1979

  21. [21]

    Drożdż, J

    S. Drożdż, J. Kwapień, P. Oświęcimka, and R. Rak, New Journal of Physics12, 105003 (2010)

    work page 2010

  22. [22]

    Instituto Nacional de Meteorologia (INMET),Banco de Dados Meteorológicos para Ensino e Pesquisa (BDMEP), https://bdmep.inmet.gov.br/(accessed June 1, 2026)

    work page 2026