pith. sign in

arxiv: 2605.19464 · v1 · pith:KIAUMOJWnew · submitted 2026-05-19 · ❄️ cond-mat.dis-nn · cond-mat.stat-mech

Diffusing diffusivity selects Pareto tail exponent in random growth with redistribution

Maxence Arutkin , Alexandre Vall\'ee This is my paper

Pith reviewed 2026-05-20 02:39 UTC · model grok-4.3

classification ❄️ cond-mat.dis-nn cond-mat.stat-mech
keywords diffusing diffusivityPareto tailwealth redistributionmultiplicative growthBouchaud-Mézard modelMarkov processstationary distribution
0
0 comments X

The pith

Fluctuating diffusivity selects the Pareto tail exponent in multiplicative growth with redistribution rather than the average value.

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

The paper examines random multiplicative growth combined with uniform redistribution when the noise intensity itself fluctuates in time. In the classic fixed-diffusivity case the stationary wealth distribution develops a Pareto tail whose exponent is set by the mean noise strength. When diffusivity diffuses, agents that remain in high-volatility states for extended periods contribute disproportionately to the extreme high-wealth events, so the tail exponent cannot be recovered by simply inserting the average diffusivity. For a two-state Markovian diffusivity an exact analysis produces a Pareto exponent that continuously interpolates between the slow-switching high-diffusivity limit and the fast-switching mean-diffusivity limit of the original model.

Core claim

For a geometric Brownian motion with diffusing diffusivity and redistribution, the persistence of high-diffusivity states becomes a stationary feature. Agents remaining in high-diffusivity states dominate rare large-wealth events, so the Pareto exponent is not obtained by replacing the diffusivity by its mean. For a two-state diffusivity, an exact tail analysis gives a Pareto exponent interpolating between the high-diffusivity slow-refresh limit and the mean-diffusivity fast-refresh Bouchaud-Mézard limit.

What carries the argument

two-state Markovian diffusing diffusivity coupled to geometric Brownian motion with instantaneous uniform redistribution, whose stationary wealth tail is obtained from a master equation

If this is right

  • The Pareto exponent depends explicitly on the switching rates between the two diffusivity states.
  • In the slow-switching limit the exponent approaches the value set by the high-diffusivity state alone.
  • In the fast-switching limit the exponent recovers the standard Bouchaud-Mézard result that uses only the average diffusivity.
  • The wealth distribution remains a power law, but its index is selected by the temporal structure of volatility fluctuations.

Where Pith is reading between the lines

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

  • The same selection mechanism may operate in any multiplicative process whose growth-rate variance fluctuates, such as firm-size or city-population distributions.
  • Empirical checks could compare the measured persistence time of volatility with the observed tail exponent in income or wealth data.
  • Extensions to more than two diffusivity states would produce a family of interpolating exponents controlled by the full Markov generator.

Load-bearing premise

The diffusivity evolves as a finite-state Markov process and redistribution occurs uniformly and instantaneously across all agents.

What would settle it

Numerical integration of the two-state model for long times that extracts the stationary wealth distribution and verifies whether its power-law exponent matches the exact interpolating value predicted by the master-equation analysis.

Figures

Figures reproduced from arXiv: 2605.19464 by Alexandre Vall\'ee, Maxence Arutkin.

Figure 1
Figure 1. Figure 1: FIG. 1. Standardized centered log-return distribution for [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Stationary complementary cumulative distribution [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Stationary Pareto exponent [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
read the original abstract

Random multiplicative growth with redistribution generates stationary Pareto wealth tails in the Bouchaud-M\'ezard model, but assumes a fixed multiplicative noise intensity. This is restrictive for physical and financial growth processes, where volatility (diffusivity) is often fluctuating. We replace the constant noise intensity by a diffusing diffusivity and ask how these fluctuations select the Pareto stationary tail. For a geometric Brownian motion with diffusing diffusivity, the effect is transient: log-returns show non-Gaussian short-time statistics but self-average to a Gaussian form at long times. With redistribution, the same persistence becomes stationary. Agents remaining in high-diffusivity states dominate rare large-wealth events, so the Pareto exponent is not obtained by replacing the diffusivity by its mean. For a two-state diffusivity, an exact tail analysis gives a Pareto exponent interpolating between the high-diffusivity slow-refresh limit and the mean-diffusivity fast-refresh Bouchaud-M\'ezard limit.

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 / 2 minor

Summary. The paper extends the Bouchaud-Mézard model of multiplicative wealth growth with redistribution by replacing constant diffusivity with a diffusing diffusivity process. It claims that persistence in high-diffusivity states dominates the rare large-wealth events, so the stationary Pareto exponent is not simply the mean-diffusivity value; for a two-state diffusivity, an exact tail analysis via the joint master equation yields an exponent that interpolates between the high-D slow-refresh limit and the mean-D fast-refresh limit.

Significance. If the central claim holds, the work supplies a concrete mechanism by which fluctuating volatility selects the wealth-tail exponent in a setting more realistic than fixed noise intensity. The exact two-state result is a clear strength: it is a closed-form, parameter-dependent prediction rather than a fit, and the model is directly motivated by physical and financial processes with state-dependent noise. This could influence subsequent modeling of inequality under heterogeneous agent dynamics.

major comments (2)
  1. [Two-state tail analysis (master-equation derivation)] The redistribution term in the master equation for p(w,s) (the joint density of wealth and diffusivity state) is written with the ensemble-average wealth. For a stationary power-law tail whose first moment is only marginally convergent, sample-to-sample fluctuations in the realized average are driven by the rare high-D, high-w agents; replacing the fluctuating average by its ensemble value closes the equation but requires explicit justification or a fluctuation correction to support the claimed exact exponent. This is load-bearing for the interpolation result.
  2. [Abstract and § on exact analysis] The abstract states that an 'exact tail analysis' exists, yet the provided derivation steps, boundary conditions at large w, and error estimates for the asymptotic matching are not shown in sufficient detail to verify that the Pareto exponent is obtained without post-hoc choices or truncation artifacts.
minor comments (2)
  1. [Model definition] Define the two-state switching rates and the slow/fast refresh limits explicitly in terms of the model parameters so that the interpolation formula can be reproduced without ambiguity.
  2. [Numerical results] Add direct comparison of the analytical Pareto exponent versus numerical histograms with quantified uncertainty (e.g., bootstrap errors on the fitted tail) to demonstrate agreement beyond visual inspection.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive report. The comments raise important issues regarding the mean-field closure and the presentation of the tail analysis. We respond to each point below and describe the planned revisions.

read point-by-point responses

  1. Referee: [Two-state tail analysis (master-equation derivation)] The redistribution term in the master equation for p(w,s) (the joint density of wealth and diffusivity state) is written with the ensemble-average wealth. For a stationary power-law tail whose first moment is only marginally convergent, sample-to-sample fluctuations in the realized average are driven by the rare high-D, high-w agents; replacing the fluctuating average by its ensemble value closes the equation but requires explicit justification or a fluctuation correction to support the claimed exact exponent. This is load-bearing for the interpolation result.

    Authors: We agree that the mean-field replacement of the fluctuating average wealth by its ensemble value requires explicit justification, particularly when the first moment is only marginally convergent. This closure is inherited from the original Bouchaud-Mézard model and becomes exact in the thermodynamic limit N→∞ for the bulk of the distribution; the rare high-wealth events that set the tail are already encoded in the joint density p(w,s). Fluctuation corrections to the mean are expected to be sub-dominant for the leading exponential decay of the tail. In the revision we will add a dedicated paragraph discussing the validity of this approximation, including a scaling argument that 1/N corrections do not shift the Pareto exponent at leading order. revision: partial

  2. Referee: [Abstract and § on exact analysis] The abstract states that an 'exact tail analysis' exists, yet the provided derivation steps, boundary conditions at large w, and error estimates for the asymptotic matching are not shown in sufficient detail to verify that the Pareto exponent is obtained without post-hoc choices or truncation artifacts.

    Authors: We accept that the current presentation of the tail analysis is insufficiently detailed. The claimed exactness holds within the mean-field master equation; the solution proceeds by assuming an exponential tail form for large w, substituting into the stationary joint equation, and solving the resulting algebraic condition for the decay rate. In the revised manuscript we will expand the relevant section to display the full substitution steps, state the boundary conditions imposed at large w (vanishing probability current and regularity), and include a brief error estimate for the asymptotic matching. These additions will allow direct verification that no post-hoc truncation is used. revision: yes

Circularity Check

0 steps flagged

Derivation of Pareto exponent follows directly from master equation on joint density without self-referential reduction or fitted inputs.

full rationale

The paper sets up a two-state Markov diffusivity process coupled to multiplicative growth plus redistribution, then solves the stationary tail of the joint density p(w,s) via the master equation. The resulting exponent interpolates between the high-D slow-refresh limit and the mean-D Bouchaud-Mézard limit by construction from those equations; no parameter is fitted to the target exponent and then called a prediction, nor is the result defined in terms of itself. The redistribution term is taken from the standard mean-field form of the model, which is an explicit modeling choice rather than a hidden self-definition. No load-bearing self-citation chain or ansatz imported from prior author work is required for the central claim. This is the normal case of a self-contained derivation from stated assumptions.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The model rests on the standard geometric-Brownian-motion growth plus redistribution framework plus the new assumption of a diffusing diffusivity process; no explicit free parameters are named in the abstract, but the two-state values and switching rates are implicit model choices.

free parameters (1)
  • two-state diffusivity values and switching rates
    The specific high and low diffusivity levels and the transition rates between them are parameters of the two-state model used for the exact analysis.
axioms (2)
  • domain assumption Wealth evolves as geometric Brownian motion whose diffusivity itself follows a Markov process
    Standard extension of the Bouchaud-Mézard setup invoked to generate the stationary distribution.
  • domain assumption Redistribution is instantaneous and uniform across agents
    Core modeling choice that produces the stationary Pareto tail.

pith-pipeline@v0.9.0 · 5694 in / 1407 out tokens · 42682 ms · 2026-05-20T02:39:24.392106+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

45 extracted references · 45 canonical work pages

  1. [1]

    Drăgulescu and V

    A. Drăgulescu and V. M. Yakovenko, Exponential and power-law probability distributions of wealth and income in the united kingdom and the united states, Physica A: Statistical Mechanics and its Applications299, 213 (2001)

    work page 2001

  2. [2]

    V. M. Yakovenko and J. B. Rosser Jr, Colloquium: Sta- tistical mechanics of money, wealth, and income, Reviews of modern physics81, 1703 (2009)

    work page 2009

  3. [3]

    Charpentier and E

    A. Charpentier and E. Flachaire, Pareto models for top incomes and wealth, The Journal of Economic Inequality 20, 1 (2022)

    work page 2022

  4. [4]

    Gabaix, Zipf’s law for cities: an explanation, The Quarterly journal of economics114, 739 (1999)

    X. Gabaix, Zipf’s law for cities: an explanation, The Quarterly journal of economics114, 739 (1999)

    work page 1999

  5. [5]

    Gabaix, Power laws in economics and finance, Annu

    X. Gabaix, Power laws in economics and finance, Annu. Rev. Econ.1, 255 (2009)

    work page 2009

  6. [6]

    Sornette and R

    D. Sornette and R. Cont, Convergent multiplicative pro- cesses repelled from zero: power laws and truncated power laws, Journal de Physique I7, 431 (1997)

    work page 1997

  7. [7]

    D. H. Zanette and S. Manrubia, Fat tails and black swans: Exact results for multiplicative processes with re- sets, Chaos: An Interdisciplinary Journal of Nonlinear Science30(2020)

    work page 2020

  8. [8]

    Bouchaud and M

    J.-P. Bouchaud and M. Mézard, Wealth condensation in a simple model of economy, Physica A: Statistical Me- 7 chanics and its Applications282, 536 (2000)

    work page 2000

  9. [9]

    Biham, O

    O. Biham, O. Malcai, M. Levy, and S. Solomon, Generic emergence of power law distributions and levy-stable in- termittent fluctuations in discrete logistic systems, Phys- ical Review E58, 1352 (1998)

    work page 1998

  10. [10]

    Levy and S

    M. Levy and S. Solomon, Power laws are logarithmic boltzmannlaws,InternationalJournalofModernPhysics C7, 595 (1996)

    work page 1996

  11. [11]

    Chatterjee and B

    A. Chatterjee and B. K. Chakrabarti, Kinetic exchange models for income and wealth distributions, The Euro- pean Physical Journal B60, 135 (2007)

    work page 2007

  12. [12]

    Greenberg and H

    M. Greenberg and H. O. Gao, Twenty-five years of ran- dom asset exchange modeling, The European Physical Journal B97, 69 (2024)

    work page 2024

  13. [13]

    B. M. Boghosian, Kinetics of wealth and the pareto law, Physical Review E89, 042804 (2014)

    work page 2014

  14. [14]

    Garlaschelli and M

    D. Garlaschelli and M. I. Loffredo, Effects of network topology on wealth distributions, Journal of Physics A: Mathematical and Theoretical41, 224018 (2008)

    work page 2008

  15. [15]

    Ichinomiya, Bouchaud-mézard model on a random network, Physical Review E—Statistical, Nonlinear, and Soft Matter Physics86, 036111 (2012)

    T. Ichinomiya, Bouchaud-mézard model on a random network, Physical Review E—Statistical, Nonlinear, and Soft Matter Physics86, 036111 (2012)

    work page 2012

  16. [16]

    Ichinomiya, Power-law exponent of the bouchaud- mézard model on regular random networks, Physical Re- view E—Statistical, Nonlinear, and Soft Matter Physics 88, 012819 (2013)

    T. Ichinomiya, Power-law exponent of the bouchaud- mézard model on regular random networks, Physical Re- view E—Statistical, Nonlinear, and Soft Matter Physics 88, 012819 (2013)

    work page 2013

  17. [17]

    Bernard, J.-P

    M. Bernard, J.-P. Bouchaud, and P. Le Doussal, Mean- field theory for heterogeneous random growth with redis- tribution, Physical Review E113, L032101 (2026)

    work page 2026

  18. [18]

    Kesten, Random difference equations and renewal the- ory for products of random matrices, Acta Mathematica 131, 207 (1973)

    H. Kesten, Random difference equations and renewal the- ory for products of random matrices, Acta Mathematica 131, 207 (1973)

    work page 1973

  19. [19]

    B. Wang, J. Kuo, S. C. Bae, and S. Granick, When brow- nian diffusion is not gaussian, Nature materials11, 481 (2012)

    work page 2012

  20. [20]

    M. V. Chubynsky and G. W. Slater, Diffusing diffusivity: a model for anomalous, yet brownian, diffusion, Physical review letters113, 098302 (2014)

    work page 2014

  21. [21]

    A. V. Chechkin, F. Seno, R. Metzler, and I. M. Sokolov, Brownian yet non-gaussian diffusion: from superstatis- tics to subordination of diffusing diffusivities, Physical Review X7, 021002 (2017)

    work page 2017

  22. [22]

    Tyagi and B

    N. Tyagi and B. J. Cherayil, Non-gaussian brownian dif- fusion in dynamically disordered thermal environments, The Journal of Physical Chemistry B121, 7204 (2017)

    work page 2017

  23. [23]

    Sposini, A

    V. Sposini, A. V. Chechkin, F. Seno, G. Pagnini, and R. Metzler, Random diffusivity from stochastic equa- tions: comparison of two models for brownian yet non- gaussian diffusion, New Journal of Physics20, 043044 (2018)

    work page 2018

  24. [24]

    Jain and K

    R. Jain and K. L. Sebastian, Diffusion in a crowded, rear- ranging environment, The Journal of Physical Chemistry B120, 3988 (2016)

    work page 2016

  25. [25]

    Lanoiselée, N

    Y. Lanoiselée, N. Moutal, and D. S. Grebenkov, Diffusion-limitedreactionsindynamicheterogeneousme- dia, Nature communications9, 4398 (2018)

    work page 2018

  26. [26]

    Gopikrishnan, V

    P. Gopikrishnan, V. Plerou, L. A. N. Amaral, M. Meyer, and H. E. Stanley, Scaling of the distribution of fluctua- tions of financial market indices, Physical Review E60, 5305 (1999)

    work page 1999

  27. [27]

    Cont, Empirical properties of asset returns: stylized facts and statistical issues, Quantitative finance1, 223 (2001)

    R. Cont, Empirical properties of asset returns: stylized facts and statistical issues, Quantitative finance1, 223 (2001)

    work page 2001

  28. [28]

    S. L. Heston, A closed-form solution for options with stochastic volatility with applications to bond and cur- rency options, The review of financial studies6, 327 (1993)

    work page 1993

  29. [29]

    Z. Ding, C. W. Granger, and R. F. Engle, A long mem- ory property of stock market returns and a new model, Journal of empirical finance1, 83 (1993)

    work page 1993

  30. [30]

    M.ArutkinandS.Reuveni,Doublystochasticcontinuous time random walk, Physical Review Research6, L012033 (2024)

    work page 2024

  31. [31]

    E. W. Montroll and G. H. Weiss, Random walks on lat- tices. ii, Journal of Mathematical Physics6, 167 (1965)

    work page 1965

  32. [32]

    R. N. Mantegna and H. E. Stanley, Scaling behaviour in the dynamics of an economic index, Nature376, 46 (1995)

    work page 1995

  33. [33]

    Kontoyiannis and S

    I. Kontoyiannis and S. Meyn, Large Deviations Asymp- totics and the Spectral Theory of Multiplicatively Regu- lar Markov Processes, Electronic Journal of Probability 10, 61 (2005)

    work page 2005

  34. [34]

    B. K. Beare and A. A. Toda, Determination of pareto exponents in economic models driven by markov multi- plicative processes, Econometrica90, 1811 (2022)

    work page 2022

  35. [35]

    Medo, Breakdown of the mean-field approximation in a wealth distribution model, Journal of Statistical Me- chanics: Theory and Experiment2009, P02014 (2009)

    M. Medo, Breakdown of the mean-field approximation in a wealth distribution model, Journal of Statistical Me- chanics: Theory and Experiment2009, P02014 (2009)

    work page 2009

  36. [36]

    J. Hur, M. Ha, and H. Jeong, Interplay of network struc- ture and talent configuration on wealth dynamics, Phys- ical Review E110, 024312 (2024)

    work page 2024

  37. [37]

    Burda, D

    Z. Burda, D. Johnston, J. Jurkiewicz, M. Kamiński, M. A. Nowak, G. Papp, and I. Zahed, Wealth conden- sation in pareto macroeconomies, Physical Review E65, 026102 (2002)

    work page 2002

  38. [38]

    Derrida and H

    B. Derrida and H. Spohn, Polymers on disordered trees, spin glasses, and traveling waves, Journal of Statistical Physics51, 817 (1988)

    work page 1988

  39. [39]

    M.Guéneau, S.N.Majumdar,andG.Schehr,Largedevi- ations in switching diffusion: from free cumulants to dy- namical transitions, Physical Review Letters135, 067102 (2025)

    work page 2025

  40. [40]

    Bouchaud and R

    J.-P. Bouchaud and R. E. Farmer, Self-fulfilling prophe- cies, quasi nonergodicity, and wealth inequality, Journal of Political Economy131, 947 (2023)

    work page 2023

  41. [41]

    O. S. Klass, O. Biham, M. Levy, O. Malcai, and S. Solomon, The forbes 400, the pareto power-law and efficient markets, The European Physical Journal B55, 143 (2007)

    work page 2007

  42. [42]

    Benhabib, A

    J. Benhabib, A. Bisin, and S. Zhu, The distribution of wealth and fiscal policy in economies with finitely lived agents, Econometrica79, 123 (2011)

    work page 2011

  43. [43]

    R. L. Axtell, Zipf distribution of us firm sizes, science 293, 1818 (2001)

    work page 2001

  44. [44]

    Bouchaud, Power laws in economics and finance: Some ideas from physics, Quantitative finance1, 105 (2001)

    J.-P. Bouchaud, Power laws in economics and finance: Some ideas from physics, Quantitative finance1, 105 (2001)

    work page 2001

  45. [45]

    Gabaix, J.-M

    X. Gabaix, J.-M. Lasry, P.-L. Lions, and B. Moll, The dynamics of inequality, Econometrica84, 2071 (2016)

    work page 2071