Computing Science: Follow the Money

Author(s): Brian Hayes

Source: American Scientist , SEPTEMBER-OCTOBER 2002, Vol. 90, No. 5 (SEPTEMBER-
OCTOBER 2002), pp. 400-405
Published by: Sigma Xi, The Scientific Research Honor Society

Stable URL: https://www.jstor.org/stable/27857715

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

is collaborating with JSTOR to digitize, preserve and extend access to American Scientist

This content downloaded from

168.195.52.137 on Sat, 11 Feb 2023 17:43:42 UTC
All use subject to https://about.jstor.org/terms
 Computing Science

Follow the Money

Brian Hayes

The rich get richer and the poor get poorer. wealth is a conserved quantity, like energy or
You've heard that before. It is a maxim so momentum. Because the total amount of wealth
often repeated, and so often confirmed by never changes, one person can get richer only if
experience, that it begins to sound like a law of another grows poorer.
nature, as familiar and irresistible as gravity. And I find it helpful to think of this miniature econ
indeed perhaps there is some physical or mathe omy in terms of a yard sale, where all the partic
matical rule governing the distribution of wealth ipants put their goods out on the lawn Saturday
in the world. No such general principle is going morning, then stroll up and down the street mak
explain the specifics of who gets rich and poor, but ing trades with their neighbors. At the end of the
it might illuminate the overall statistics. day, after all transactions are completed, an audi
This idea goes back at least a century to the tor reviews everyone's inventory and calculates
work of the Italian economist Vilfredo Pareto, their new net worth.
who tried to show that the income distribution in Many economic models assume that all trans
all cultures and countries has the same mathe actions occur at precisely the right price. Indeed,
matical form. In recent years the topic has been prices are correct by definition in such models:
taken up with renewed enthusiasm by a small Whatever price is agreed to by a willing seller
band of "econophysicists," who apply principles and a willing buyer is the value assigned to an
of statistical mechanics to questions in economic asset. Given such perfect pricing, nothing inter
theory. The essence of their approach is to study esting could ever happen in the yard-sale econo
an economy as if it were a many-body physical my. I might trade my old toaster for your broken
system such as a gas. Just as random collisions VCR, but if we negotiate the terms of the deal
between gas molecules give rise to macroscopic correctly, my net worth will not change in the
properties such as temperature and pressure, slightest, and neither will yours.
random encounters between individuals in an In practice, the assumption of perfect pricing
economic system might determine large-scale seems a little unrealistic. Some buyers are more
phenomena such as the distribution of wealth. discerning than others, and some sellers are more
Some of the computational models for explor persuasive. There are bargains to be had, and
ing these issues are remarkably easy to build and there are bad deals?concepts that could hardly
run. It takes just a few minutes' effort and a few exist if we did not agree that merchandise has a
lines of code. On the other hand, it's also remark true and proper value, which does not always
ably easy to make subtle mistakes of implemen correspond exactly to the price paid.
tation, as I'll have occasion to mention below. Even slight departures from perfect pricing
And the big challenge is not building the models bring a new dynamic to the yard-sale model. If I
but interpreting the results?deciding which buy your rusty wheelbarrow and pay more than
kinds of random encounters might represent it's worth, I am left slightly poorer after the trans
events in a real economy. action, and you are a little richer. Conversely, if I
pay less than fair value, I gain a little, and you
The Price Is Right lose. In either case there has been a transfer of
The economy being simulated in these models is wealth, typically a small fraction of the price paid.
a rather special one, based on pure, free-market These transfers are where the action is in the mod
trading. The exchange of assets is all that ever eled economy; as a matter of fact, the model can
happens here; there is no production of new ignore the transaction itself?there's no need to
wealth, and no consumption either. Leaving out talk about toasters and wheelbarrows?and sim
so much of the real economy is an obvious weak ply consider the net transfer of wealth.
ness, but there is a compensating advantage: The question is: What happens when this
What remains is a closed system. In the model, process is repeated many times? If some of the
traders are shrewder than others, you would cer
Brian Hayes is Senior Writer for American Scientist. Address: tainly expect them to do well in the long run; like
211 Dacian Avenue, Durham, NC 27701; [email protected] wise the perennial suckers are going to lose their

American Scientist, Volume 90

This content downloaded from

168.195.52.137 on Sat, 11 Feb 2023 17:43:42 UTC
All use subject to https://about.jstor.org/terms
shirts. But suppose that everyone is equally skill
ful, so that who wins and who loses is purely a
matter of chance. The amount of gain and loss is
also determined at random?but it's always less
than the total wealth of the poorer agent, so that
traders never risk losing more than they own.
Before reading on, you might try to predict
what will happen in such an economy If every
one starts out with the same bankroll, how will
the assets be distributed after many random ex
changes? Will the levels of wealth remain uni
form? Perhaps the system will evolve toward a
Gaussian distribution, with most people having a
middling amount of money, while a few are very
poor and a few are rich?
Here is the answer given by the computer ex
periment: If trading continues long enough, es
sentially all the wealth winds up in the hands of
one person. The yard-sale economy, as formulat
ed in this model, is a winner-take-all lottery. The
traders might just as well put all their goods in
one big pile, and then roll the dice to decide who
keeps it all. (Strictly speaking, one trader gets all
the goods only if wealth is quantized?if there is
some smallest unit of value below which one's
worth falls to zero. If wealth can be subdivided
indefinitely, the winner's share comes arbitrarily
close to 100 percent but never quite gets there.)
This condensation of all property in the hands
of one individual is an economic catastrophe?
something like the formation of a black hole in as
trophysics. It's obviously bad news for the major
ity of the people, who are left penniless. But even
if you happen to be the big winner, your victory
may prove hollow. Although you have all the rich
es in the world, you can't buy a thing, because no
one else has goods to sell. And you can't sell any
thing either, because no one has money to buy
with. The whole economy is frozen.

Molecular Economics
I first began experimenting with the yard-sale
simulation after reading an article, "Wealth dis
tributions in asset exchange models," by Slava
Ispolatov, Paul L. Krapivsky and Sidney Redner
of Boston University. The computer models de
scribed there seemed both intriguing and easy to
re-create, and so I wrote a quick-and-dirty pro
gram to play with some of them. I was perplexed
when my results were quite different from those
reported in the article. A second look revealed
that I had misread a key equation, so that my
model differed from theirs in a small but crucial
way. Later I found a paper by Anirban Chakra
borti of the Saha Institute of Nuclear Physics in
India that describes essentially the same model I
had accidentally created.
At least two other groups of physicists have re Figure 1. Wealth flows through a model economy in which vertical
cently published work on related themes. In channels represent individual traders and horizontal channels show
France, Jean-Philippe Bouchaud of the Centre the gain or loss in transactions between traders. In this specific mod
d'?tudes de Saclay and Marc M?zard of the Ecole el (the "yard-sale" economy), gain or loss is limited to the wealth of
normale sup?rieure have described "wealth con the poorer trading partner. All traders start out equal at the top of the
densation" in a somewhat different model. And diagram, but wealth becomes more concentrated with time.

2002 September-October 401

This content downloaded from

168.195.52.137 on Sat, 11 Feb 2023 17:43:42 UTC
All use subject to https://about.jstor.org/terms
 0.001 0.01 0.1 1 10 100 1,000 0.001 0.01 0.1 1 10 100 1,000 0 0.2 0.4 0.6 0.8 1.0
wealth of individuals wealth of individuals fraction of population
Figure 2. Distribution of wealth in the yard-sale economy grows steadily more extreme. In each of these graphs the yellow line indicates the sta
ing configuration: 1,000 traders, each of wealth 1. Curves in progressively darker shades of blue show the distribution after increasing number
transactions, from 1,000 to 10 million. The graph at left plots the number of people in each wealth category. The middle graph shows the
wealth of all those in each category. The right graph is a Lorenz curve (named for the Italian economist Max O. Lorenz), which gives the p
tion of the population that controls a given proportion of the wealth. All three graphs reveal the collapse of the economy as one trader accumu
essentially all the wealth. The curves are averages over several thousand runs, but in some cases statistical fluctuations are still conspicuou

Adrian Dr?gulescu and Victor M. Yakovenkoleast

of it's not an all-or-nothing economy. Further
more, although the shape of the distribution is sta
the University of Maryland have written on "the
statistical mechanics of money." ble, individuals do not remain stationary within it:
A source of ideas for most of these models is are many rags-to-riches-to-rags stories in
There
the analogy between market economics andsuch the a society. The gap between rich and poor
kinetic theory of gases. The molecules of aseems gas less unfair if people have a reasonable
are continually colliding with one another chance and of moving between these categories.
exchanging energy, in much the way that ran An exponential distribution of wealth is clear
domly chosen buyers and sellers in an economic ly preferable to a winner-take-all outcome, and
model exchange sums of money. Yet gases doan economic model based on the kinetic theory
not
follow the evolutionary path of the yard-sale of gases may also have a certain aesthetic ap
economy. An economic collapse, where onepeal?at per least to physicists. Nevertheless, the in
son sucks in all the money, corresponds to aterpretation
gas of the model is problematic. There is
where one molecule has all the kinetic energy, no obvious reason to expect economic agents to
and the rest are standing still. Don't hold your act like colliding molecules, and indeed the ran
breath waiting for that to happen. dom repartitioning of kinetic energy is a fairly
Where the yard-sale model departs from strange the template for mercantile transactions. Ap
kinetic theory of gases is in the details of the plied
ex in the yard-sale context, it suggests that
change of wealth or energy. When two gas when mole Bill Gates comes to browse among my
cules collide, they can reapportion their energy lawn
in ornaments, he and I will pool all our assets
any way that leaves the total unchanged. Ifand thethen randomly split up the pot.
molecules have energies a and b just before they One kind of financial transaction that might fit
collide, afterward they can have any combina the pattern of the kinetic gas theory is marriage
followed by divorce: This is a case where the par
tion of energies that add up to a + b. Translating
ties do combine their holdings and later redivide
this energy-redistribution process into financial
terms yields a market in which the parties them,
to a although perhaps not quite randomly. In
transaction combine their wealth and then the
ran corporate world, mergers and spin-offs might
domly divide the total. A simulated economy produce similar results.
based on this rule does not collapse the way the
yard-sale model does; wealth remains spread Crime Doesn't Pay
The two models described so far lie at opposite
throughout the population, although not uni
poles along an axis defined by the amounts the
formly so. The distribution follows an exponen
trading
tial curve: The number of people with wealth w is parties put at risk. In the yard-sale model,
the most that can be won or lost is the total wealth
proportional to e-wlJ, where is the temperature.
(In the economic context, Dr?gulescu and Yakoof the poorer partner. Since this model evolves to
ward a state where nearly everyone is impover
venko identify the temperature with the average
ished, the typical transaction is extremely small.
amount of money available to the participants.)
In the marriage-divorce model, in contrast, the
An exponential distribution crowds most of the
entire fortunes of both partners are up for grabs.
people into the lower economic strata, but com
Here is a recipe for a third model that occu
pared with the lopsided outcome of the yard-sale
pies
model, the degree of inequity is fairly mild. At a middle ground. As in the yard-sale algo

402 American Scientist, Volume 90

This content downloaded from

168.195.52.137 on Sat, 11 Feb 2023 17:43:42 UTC
All use subject to https://about.jstor.org/terms
 rithm, pick two trading partners at random, and coercion or deception. In other words, what is be
also randomly choose which of the partners is to ing modeled here is theft and fraud.
lose (the donor) and which is to gain (the recipi When the theft-and-fraud model is allowed to
ent). But instead of setting the size of the trade as run for many iterations, there is no economic col
a random fraction of the poorer player's wealth, lapse. The wealth distribution reaches an equilib
make it a random fraction of the wealth of the rium on an exponential curve much like that seen
donor. This rule still satisfies the commonsense in the marriage-and-divorce model. (I have no
constraint that you can never be made to pay comment on this evidence that marriage and di
more than you have. In each transaction you risk vorce have the same economic impact as larceny,
losing a random fraction of your own wealth, but nor will I speculate on why a world populated by
you have a chance to gain a random fraction of bank robbers winds up with a fairer distribution
the other person's fortune. of wealth than an economy of honest merchants.)
What kinds of real-world transactions might be
described by this model? No doubt there are Beyond the Dreams of Avarice
many plausible interpretations, but here is one The recent publications on asset-exchange mod
that I find intriguing. A distinctive characteristic of els describe many more variations. Dr?gulescu
the trading scheme is that the richer party always and Yakovenko mention a family of models that
has more to lose and the poorer more to gain. Un differ among themselves only in the rule for
der these terms, any sensible person would try to choosing an amount of money to transfer. In one
do business only with wealthier partners, and no case it is a small fixed quantity; in another it is a
one would ever willingly choose to trade with a random fraction of the trading pair's average
less-affluent person (assuming traders can gauge wealth; in a third model the amount is a random
the wealth of their partners). Thus if trading be fraction of the average wealth of the entire popu
tween nonequals takes place at all, it must be by lation. To avoid putting traders into debt or bank

0.001 0.001
0.001 0.1 1 10 100 1,000 0.001 0.1 1 10 100 1,000 0.2 0.4 0.6 0.8 1.0
wealth of individuals wealth of individuals fraction of population

Figure 3. A subtle change in the economic model yields a totally different distribution of wealth. The change alters the rule for setting the amount
of each transaction: The upper limit is the loser's wealth rather than the wealth of the poorer partner. Under these conditions no individual ac
quires more than a few percent of the total wealth. Moreover, the distribution is stable: After the first few thousand transactions, the shape of the
curves does not change. Ironically, the kinds of transactions that produce this pleasant outcome could be interpreted as theft and fraud.

1,000

0.001 0.001
0.001 0.1 1 10 100 1,000 0.001 0.1 1 10 100 0.2 0.4 0.6 0.8 1.0
wealth of individuals wealth of individuals fraction of population

Figure 4. Taxes and welfare forestall the collapse of the yard-sale model. Here the underlying transaction m
but after every trade a randomly chosen trader is taxed a randomly chosen amount, and the proceeds are
traders. The result is a stable distribution with a comparatively narrow span between richest and poorest.

2002 September-October 40

This content downloaded from

168.195.52.137 on Sat, 11 Feb 2023 17:43:42 UTC
All use subject to https://about.jstor.org/terms
 ruptcy, Dr?gulescu and Yakovenko apply the distribution of wealth reaches some stable equi
meta-rule that if the loser cannot pay, the entire librium. What is the root of the difference?
transaction is canceled. In all these models the Dr?gulescu and Yakovenko point out that
equilibrium distribution has an exponential form, transactions like those in the yard-sale model
and there is no economic collapse. break time-reversal symmetry. For an example of
Ispolatov, Krapivsky and Redner look at a transaction rule that is reversible, consider the
greedy or exploitative rules, where the wealthier marriage-and-divorce model. Lumping together
party always wins the exchange (perhaps reflect two fortunes and then splitting the sum is a
ing a situation where the poor have less bargain process that works the same both forward and
ing power). When the amount transferred is a backward. If two traders report that they have $5
random fraction of the poorer's agent's wealth and $3 at one moment, and $7 and $1 at another
(as in the yard-sale model), the result is economic moment, with a single transaction between these
collapse, with all funds gravitating toward one states, you can't tell which report is earlier and
person. Of course it's hardly a surprise that sys which is later. The lumping-and-splitting rule
tematic greed yields a harsh outcome. The sur could apply in either direction. In the yard-sale
prise is that this obviously biased rule is no worse model, on the other hand, the crucial step is tak
than the symmetrical rule in the yard-sale model. ing the minimum of the two amounts, and re
Chakraborti looks at the effect of savings, al versing this operation cannot always restore the
lowing traders to hold back some of their capital initial configuration. A transaction carried out un
from the market. In the yard-sale economy, sav der the yard-sale rule can go from the $5-and-$3
ings cannot forestall a collapse. Reserving a fixed state to the $7-and-$l state, but not the other way. .*
sum of money shifts the minimum wealth up The irreversibility of the yard-sale rule acts as a
from zero but does not alter the dynamics of the kind of ratchet: Once the economy wanders into a
model. Saving a fixed fraction of wealth slows the state with an unbalanced distribution of wealth, it
collapse, but the winner still takes all in the end. takes a long while to find its way out again. To
Several authors mention the effects of taxes, see more clearly how the ratchet works, consider
welfare and other explicit means of redistributing an even simpler model?an economy pared down
income. Imposing a tax on wealth prevents the to just two participants. Now the changing for
implosion of the yard-sale economy (see Figure 4), tunes of either trader can be represented by a ran
but the effects of an income tax are not so clear. I dom walk along a line extending from zero to the
experimented with income taxes by collecting a total wealth available. All activity stops if the trad
percentage of each transaction and redistributing er reaches either end of this line. A random walk
the proceeds in equal shares to all traders. A low that takes steps of uniform length is guaranteed to
tax rate does not protect against collapse, but hit an end point sooner or later (a fate known as
models with tax rates higher than about 15 percent gambler's ruin). But this is not what is going on in
do seem to survive ^definitely. If there is a sharp the yard-sale model. There the steps are not of
threshold between these regimes, I have been un fixed size; because transactions are limited to the
able to identify it. lesser of the trading partners' assets, the steps get
smaller as the walk approaches either end point. If
Trading with Zeno there is no smallest unit of currency, the random
The various economic models discussed here dif walk becomes a "Zeno walk," which spends most
fer in many details, but they can be classified in of its time in the neighborhood of an end point
two broad families: those where the economy but never actually gets there.
falls into a black hole, with one trader acquiring To simplify the model still further, we can take
nearly everything of value, and those where the a Zeno walk on the interval from 0 to 1, choosing

0 20 40 60 80 100
transactions (average per trader)

Figure 5. Economic mobility varies as dramatically as the distribution

is highly stratified: Whoever gets to the top after the first few transacti
model (lower band) ifs rags to riches from start to finish. The sequence o

404 American Scientist, Volume 90

This content downloaded from

168.195.52.137 on Sat, 11 Feb 2023 17:43:42 UTC
All use subject to https://about.jstor.org/terms
to go left or right at random but letting the step Rather than trying to match the output of the
size always be half the distance to the nearer end models to economic statistics, it might be more
point (rather than a random fraction of this dis fruitful to examine real-world economic practices
tance). If we begin at the point % the initial step for signs of the basic mechanisms that underlie
size is Suppose the first move is to the right, the models. In particular, the fatal feature of the
reaching the point %. Now the step size is %. If we yard-sale model is the rule limiting the size of a
turn back to the left, we do not return to our transaction to the wealth of the lesser trading
starting point but instead stop at %. Where will partner. The rule appears to be perfectly fair and
we wind up after steps? The probability distri symmetrical, and yet it has the effect that the far
bution for this process has an intricate fractal ther you fall through the economic strata, the
structure, so there is no simple answer, but the harder youTl find it to climb back up.
likeliest landing places get steadily closer to the Is such a rule likely to be enforced in everyday
end points of the interval as increases. This commerce? Not always. It is clearly violated in
skewed probability distribution is the ratchetlike many forms of gambling and speculation, where
mechanism that drives the yard-sale model to the whole point of the transaction is the hope of
states of extreme imbalance. gaining more than you put at risk. Doubtless there
are other exceptions as well. For the most part,
Fair Trade though, those of us with less money are limited to
Models of the market economy may lead to some smaller-scale buying and selling. And the lower
cute mathematics, but do they have the slightest the ceiling on your economic activity, the slower
connection with the price of peas in the real your progress up through the ranks. When I buy a
world? Can they predict the actual distribution new car, I have little chance?no matter how
of wealth observed in human societies? shrewdly I bargain?of significantly altering the
As it happens, the shape of the actual distribu balance of assets between me and General Motors.
tion is uncertain and controversial. Most of the Explaining the distribution of wealth among
available data concern the distribution of income, individuals is not the only possible application of
which is not quite the same as the distribution of the trading models. They might in fact be better
wealth. Pareto, 100 years ago, argued that the in suited to describing relations among companies,
come distribution obeys a power law, so that the where a sudden consolidation of wealth could
proportion of people whose income is at least be interpreted as the emergence of a monopoly.
varies as x~a; Pareto believed that the exponent a Beyond the corporate world, there is the ques
is a universal constant with a value of about 2.5. tion of whether the models might have anything
Other economists have proposed a log-normal to say about commerce among nations, and the
income curve, meaning that the distribution of ongoing debate over free markets, fair trade and a
the logarithm of income is Gaussian. "level playing field." If some mechanism like that
The model of Bouchaud and M?zard (which of the yard-sale model is truly at work, then mar
includes investment earnings as well as trade) kets might very well be free and fair, and the
yields a Pareto-like power law for the wealth dis playing field perfectly level, and yet the outcome
tribution. Some of the "greedy" models of Ispola would almost surely be that the rich get richer
tov, Krapivsky and Redner also appear to fit a and the poor get poorer. You7ve heard that before.
power-law curve. But the models drawn most di
rectly from the kinetic theory of gases predict an Bibliography
exponential distribution of wealth. Dr?gulescu Bouchaud, Jean-Philippe, and Marc M?zard. 2000. Wealth
and Yakovenko argue that the middle part of the condensation in a simple model of economy. Physica A
282:536-545.
actual wealth distribution is indeed exponential,
Chakraborti, A., and . K. Chakrabarti. 2000. Statistical me
with a "Pareto tail" in the highest tax brackets. chanics of money: how saving propensity affects its dis
All the computational models are so crude, how tribution. The European Physical Journal B17:167-170.
ever, and the empirical measurements are so un Chakraborti, Anirban. 2002. Distributions of money in
certain, that curve-fitting inspires little confidence. model markets of economy. http://arXiv.org/abs/
Also unclear is whether events comparable to cond-mat/0205221
the collapse of the yard-sale model can happen in Dr?gulescu, A., and V. M. Yakovenko. 2000. Statistical me
chanics of money. The European Physical Journal B17-.723-729.
a real economy. Societies where a small elite con
Dr?gulescu, A., and V. M. Yakovenko. 2001. Evidence for
trols almost all the property, while the rest of the
the exponential distribution of income in the USA. The
people are destitute, are all too common. But European Physical Journal B20:585-589.
does this situation result from a mathematical in
Ispolatov, S., P. L. Krapivsky and S. Redner. 1998. Wealth
stability in the system of trade, or is there a sim distributions in asset exchange models. The European
pler explanation, such as mere malice and greed? Physical Journal B2:267-276.
In any case, economic collapse seems never to go Lambert, Peter J. 1993. The Distribution and Redistribution of
to completion in the real world, as it does in the Income: A Mathematical Analysis. Second edition. Man
chester, U.K.: Manchester University Press.
models. Tycoons amass immense fortunes, but
Mandelbrot, Benoit . 1997. Fractals and Scaling in Finance:
no ever one goes home with all the marbles. (Bill
Discontinuity, Concentration, Risk. New York: Springer.
Gates holds much less than 1 percent of the Pareto, Vilfredo. 1896, 1897 (reprinted 1964). Cours
world's wealth.) d'Economie Politique. Geneva: Librairie Droz.

2002 September-October 405

This content downloaded from

168.195.52.137 on Sat, 11 Feb 2023 17:43:42 UTC
All use subject to https://about.jstor.org/terms