Yard-Sale Wealth Exchange Model

• Yard-Sale Model is a stochastic wealth-exchange framework where transactions are based on the poorer agent’s wealth, leading to wealth condensation.
• The model employs kinetic-theory methods such as Boltzmann and Fokker–Planck equations to capture how local fairness can produce global inequality.
• Extensions including redistribution, wealth-attained advantage, risk limits, and network constraints help regulate and reveal diverse outcomes in wealth concentration.

The Yard-Sale Model (YSM) is a stochastic multiplicative wealth-exchange model in which pairwise transactions transfer an amount proportional to the wealth of the poorer agent. In its baseline form, two agents interact at random, total wealth is conserved in each exchange, and the winner is chosen with equal probability; yet the long-run dynamics amplify inequality and, without redistribution, drive the system toward wealth condensation or oligarchy. Within kinetic-theory treatments, the YSM admits Boltzmann and Fokker–Planck descriptions, while probabilistic analyses establish almost sure condensation in the fully connected case. With redistribution, the same framework produces stationary wealth distributions with a sharp low-wealth cutoff and a high-wealth tail resembling Pareto’s law (Boghosian, 2012, Boghosian, 2014, Boghosian et al., 2014).

1. Microscopic exchange rule

The defining mechanism of the YSM is that the stake in any transaction is set by the poorer participant’s wealth. In the continuous-wealth formulation, if agents with wealths $w$ and $w'$ transact, the transferred amount is

$A(w,w',r)=\beta r \min(w,w'),$

where $r=\pm 1$ with equal probability and $\beta$ is small. Equivalently,

$$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$

so that after the exchange

$$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$

This rule preserves nonnegativity of wealth and conserves total wealth in each interaction (Boghosian, 2014).

Discrete-time share formulations use the same logic. If $X_\ell^i$ denotes agent $i$’s share of total wealth after $\ell$ trades, the poorer agent $w'$0 and richer agent $w'$1 update according to

$w'$2

with $w'$3 fair in the unbiased model. The amount at stake is therefore always a fraction of the poorer agent’s wealth, not of the richer agent’s wealth or of the loser’s wealth (Börgers et al., 2024).

A recurrent point in the literature is that local fairness does not preclude global concentration. In the original unbiased model, each agent’s expected wealth is conserved and, in the normalized formulation, $w'$4 for all $w'$5; nonetheless the process condenses. In the complete-graph case, the probability that agent $w'$6 eventually holds all wealth is exactly its initial wealth share,

$w'$7

This identifies the YSM as a model in which unbiased microscopic transfers generate maximally unequal macroscopic outcomes (Börgers et al., 2023).

2. Kinetic and continuum descriptions

Boghosian’s kinetic-theory treatment formulates the YSM as a closed economy with one-agent density $w'$8 and, before closure, a two-agent density $w'$9. Under the random-agent approximation,

$A(w,w',r)=\beta r \min(w,w'),$0

the gain–loss dynamics become a Boltzmann equation for wealth exchange. This equation preserves both the number of agents

$A(w,w',r)=\beta r \min(w,w'),$1

and total wealth

$A(w,w',r)=\beta r \min(w,w'),$2

The kinetic analogy is explicit: agent transactions play the role of collisions, and wealth replaces molecular energy or momentum as the exchanged quantity (Boghosian, 2012).

In the small-transaction regime, the YSM becomes a small-step stochastic process. The standard Fokker–Planck form is

$A(w,w',r)=\beta r \min(w,w'),$3

For the baseline YSM, the mean step vanishes,

$A(w,w',r)=\beta r \min(w,w'),$4

while the mean squared step is

$A(w,w',r)=\beta r \min(w,w'),$5

with

$A(w,w',r)=\beta r \min(w,w'),$6

This yields the nonlinear Fokker–Planck equation

$A(w,w',r)=\beta r \min(w,w'),$7

whose coefficients depend self-consistently on the wealth distribution itself (Boghosian, 2014).

Equivalent small-transaction equations also appear in the earlier Boltzmann-based derivation. One common form is

$A(w,w',r)=\beta r \min(w,w'),$8

with $A(w,w',r)=\beta r \min(w,w'),$9 and $r=\pm 1$0 interpreted as cumulative and partial-moment functionals of $r=\pm 1$1. In either notation, the continuum equation is nonlocal in wealth space, conservative in both $r=\pm 1$2 and $r=\pm 1$3, and mathematically analogous to diffusion equations with distribution-dependent mobility (Boghosian, 2012).

3. Condensation, oligarchy, and monotone inequality

The untaxed YSM has a robust asymptotic tendency toward complete concentration of wealth. In the kinetic formulation without redistribution, Boghosian proved that the Gini coefficient is an $r=\pm 1$4-function, or Lyapunov functional, for both the Boltzmann and Fokker–Planck equations. Writing $r=\pm 1$5 for the Gini functional, the key statement is

$r=\pm 1$6

In the Fokker–Planck limit,

$r=\pm 1$7

so inequality increases monotonically. A variational argument then shows that the maximizer under fixed $r=\pm 1$8 and $r=\pm 1$9 has

$\beta$0

corresponding to oligarchy: a singular state in which one agent asymptotically holds all wealth (Boghosian et al., 2014).

Probabilistic proofs sharpen this conclusion. For the discrete normalized process, the squared Euclidean norm

$\beta$1

serves as a concentration measure. The one-step estimate

$\beta$2

implies summability of $\beta$3, and Borel–Cantelli yields $\beta$4 almost surely. From this it follows that the maximum wealth

$\beta$5

satisfies

$\beta$6

The asymptotic state is therefore winner-takes-all on the complete graph (Börgers et al., 2023).

A common misconception is that symmetric win/loss probability should prevent concentration. The YSM literature rejects that implication. The unbiased model is fair only at the level of immediate expected transfer; the multiplicative dependence on the poorer agent’s wealth generates long-run inequality amplification. This same conclusion persists in several extensions with wealth-acquired advantage, and in some formulations it persists even under certain poverty-acquired-advantage conditions, provided the concentration statistic retains positive drift (Boghosian et al., 2014, Börgers et al., 2024).

4. Redistribution and Pareto-like stationary states

Redistribution enters the YSM as a deterministic tax-and-rebate drift. If wealth is taxed at rate $\beta$7 per transaction, each agent of wealth $\beta$8 loses $\beta$9, the total tax collected is $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$0, and equal redistribution across $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$1 agents gives the net change

$$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$2

Since this term is not random, it contributes drift but not diffusion. Writing $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$3, the full Fokker–Planck equation becomes

$$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$4

This preserves both the number of agents and total wealth while introducing a restoring drift toward mean wealth (Boghosian, 2014).

Stationary states are obtained by setting $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$5. Near very small $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$6, if $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$7 is tiny, then $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$8 and $$A(w,w',r)=\beta r \big[w\Theta(w'-w)+w'\Theta(w-w')\big],$$9, reducing the steady-state equation to

$$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$0

with solution

$$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$1

This function is extremely small as $$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$2, non-analytic there, and effectively gives a sharp lower cutoff. For larger $$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$3, the full nonlinear equation yields a heavy tail that numerically matches Pareto-like form (Boghosian, 2014).

The earlier kinetic treatment reaches the same qualitative conclusion. With taxation,

$$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$4

Near the origin, the steady asymptotic form is

$$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$5

which vanishes faster than any power as $$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$6. For large wealth, the same analysis suggests an approximate power law, and numerical fits reported Pareto indices increasing with tax rate, approximately $$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$7 at $$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$8, $$w_{\text{final}} = w + A(w,w',r), \qquad w'_{\text{final}} = w' - A(w,w',r).$$9 at $X_\ell^i$0, $X_\ell^i$1 at $X_\ell^i$2, and $X_\ell^i$3 at $X_\ell^i$4 (Boghosian, 2012).

5. Generalizations and control parameters

A major generalization is the inclusion of wealth-attained advantage (WAA), in which the richer trader is more likely to win. In one continuous-bias formulation,

$X_\ell^i$5

and the smooth distribution is augmented by a singular oligarchic component,

$X_\ell^i$6

In the coexistence regime $X_\ell^i$7, the condensed wealth fraction obeys

$X_\ell^i$8

with asymptotic fraction

$X_\ell^i$9

The transition at $i$0 is described as second-order, with a coexistence phase above criticality (Boghosian et al., 2016).

Another extension adds economic growth. In the Growth, Exchange, and Distribution model, the Yard-Sale exchange rule is retained and an amount $i$1 of new wealth is injected after each block of $i$2 exchanges, distributed according to

$i$3

The critical point is $i$4. For $i$5, the rescaled wealth distribution reaches a steady state, the wealth metric satisfies $i$6, and rank correlations decay to zero; for $i$7, no steady state is reached, mobility disappears, and wealth condenses. Fixed-$i$8 simulations gave the critical-exponent estimates

$i$9

whereas the mean-field theory at fixed Ginzburg parameter predicts

$\ell$0

The model therefore organizes the competition between exchange-driven concentration and growth allocation through a sharp critical point (Liu et al., 2021).

Other modifications target the exchange kernel itself. One variant replaces the classical stake with

$\ell$1

leading to a continuum equation with diffusion coefficient

$\ell$2

For this modified model,

$\ell$3

and, more strongly,

$\ell$4

With redistribution, the steady-state asymptotics match the classical taxed YSM: a depleted origin at small wealth and a Gaussian tail at large wealth (Cohen et al., 2023).

Risk-limited variants modify how much wealth can be exposed in each trade. In one recent model, the transferred amount is

$\ell$5

with winning probability

$\ell$6

and risk parameters drawn from

$\ell$7

For $\ell$8, the system still approaches $\ell$9 regardless of $w'$00. For $w'$01, however, smaller $w'$02 yields smaller stationary inequality, and the simulations identify an optimal risk

$w'$03

This suggests that limiting individual exposure weakens the multiplicative instability responsible for extreme concentration (Giordano et al., 8 Aug 2025).

6. Networks, local condensation, and structured populations

Restricting exchanges to a graph qualitatively changes the asymptotic state. In graph-based formulations with adjacency matrix $w'$04, the relevant condensation set is

$w'$05

The generalized convergence statement is

$w'$06

Thus the networked YSM converges toward local wealth condensation: wealthy agents may remain, but their trading neighbors become impoverished (Börgers et al., 2024).

Earlier numerical work on rings, square lattices, and random graphs found that the stable wealth-sharing phase is almost unchanged by topology, whereas the unstable phase is strongly network-dependent. In mean field, the critical line is determined by

$w'$07

That same critical interface was reported numerically for all analyzed network topologies. Beyond criticality, however, wealth no longer condenses onto a single agent except in the complete-graph limit; instead it freezes onto an extensive set of locally rich agents, whose density depends on topology and links naturally to coalescence of immobile reactants (Bustos-Guajardo et al., 2012).

Network structure remains decisive in extended models with redistribution and WAA. On Erdős–Rényi random networks, the networked Extended Yard Sale model exhibits local wealth condensation rather than the global condensation of the fully connected case. Using the temperature-like ratio

$w'$08

the transition occurs near $w'$09, and the order parameter

$w'$10

behaves continuously in Monte Carlo simulations. A central structural observation is that rich agents tend not to be adjacent: if wealthy agents are connected, one tends to lose wealth to the other. The scaling analysis further implies that a nonzero macroscopic richest-agent share in the thermodynamic limit requires the graph to become fully connected. This confirms that structured populations replace global oligarchy by topology-constrained local concentration (Bibow et al., 7 May 2025).