Colloquium: Statistical mechanics of money, wealth, and income

Victor M. Yakovenko
Department of Physics, University of Maryland, College Park, Maryland 20742-4111, USA

J. Barkley Rosser, Jr.

Department of Economics, James Madison University, Harrisonburg, Virginia 22807, USA

(Dated: 24 December 2009)

This Colloquium reviews statistical models for money, wealth, and income distributions devel-
oped in the econophysics literature since the late 1990s. By analogy with the Boltzmann-Gibbs
distribution of energy in physics, it is shown that the probability distribution of money is
arXiv:0905.1518v2 [q-fin.ST] 24 Dec 2009

exponential for certain classes of models with interacting economic agents. Alternative scenarios
are also reviewed. Data analysis of the empirical distributions of wealth and income reveals a
two-class distribution. The majority of the population belongs to the lower class, characterized
by the exponential (thermal) distribution, whereas a small fraction of the population in the
upper class is characterized by the power-law (superthermal) distribution. The lower part is
very stable, stationary in time, whereas the upper part is highly dynamical and out of equilibrium.

Money, its a gas. Pink Floyd, Dark Side of the Moon

PACS numbers: 89.65.Gh 89.75.Da 05.20.-y

Contents term appeared first by Stanley et al. (1996) in the pro-

ceedings of the Kolkata conference. The paper pre-
I. Historical Introduction 1 sented a manifesto of the new field, arguing that be-
II. Statistical Mechanics of Money Distribution 3
havior of large numbers of humans (as measured, e.g.,
A. The Boltzmann-Gibbs distribution of energy 4 by economic indices) might conform to analogs of the
B. Conservation of money 4 scaling laws that have proved useful in describing sys-
C. The Boltzmann-Gibbs distribution of money 5 tems composed of large numbers of inanimate objects
D. Models with debt 7 (Stanley et al., 1996). Soon the first econophysics confer-
E. Proportional money transfers and saving propensity 9
F. Additive versus multiplicative models 10
ences were organized: International Workshop on Econo-
physics, Budapest, 1997 and International Workshop
III. Statistical Mechanics of Wealth Distribution 11 on Econophysics and Statistical Finance, Palermo, 1998
A. Models with a conserved commodity 11 (Carbone et al., 2007), and the book An Introduction to
B. Models with stochastic growth of wealth 12 Econophysics by Mantegna and Stanley (1999) was pub-
C. Empirical data on money and wealth distributions 13
lished.
IV. Data and Models for Income Distribution 14 The term econophysics was introduced by analogy with
A. Empirical data on income distribution 14
B. Theoretical models of income distribution 18 similar terms, such as astrophysics, geophysics, and bio-
physics, which describe applications of physics to differ-
V. Conclusions 20 ent fields. Particularly important is the parallel with
biophysics, which studies living organisms, but they still
References 20
obey the laws of physics. Econophysics does not literally
apply the laws of physics, such as Newtons laws or quan-
tum mechanics, to humans. It uses mathematical meth-
I. HISTORICAL INTRODUCTION
ods developed in statistical physics to study statistical
properties of complex economic systems consisting of a
This Colloquium article is based on the lectures that
large number of humans. As such, it may be considered
one of us (V.M.Y.) has frequently given during the
as a branch of applied theory of probabilities. However,
last nine years, when econophysics became a popular
statistical physics is distinctly different from mathemat-
subject. Econophysics is a new interdisciplinary re-
ical statistics in its focus, methods, and results.
search field applying methods of statistical physics to
problems in economics and finance. The term econo- Originating from physics as a quantitative science,
physics was first introduced by the theoretical physi- econophysics emphasizes quantitative analysis of large
cist Eugene Stanley in 1995 at the conference Dynam- amounts of economic and financial data, which became
ics of Complex Systems, which was held in Kolkata as increasingly available with the introduction of comput-
a satellite meeting to the STATPHYS-19 conference in ers and the Internet. Econophysics distances itself from
China (Carbone et al., 2007; Chakrabarti, 2005). The the verbose, narrative, and ideological style of political
 2

economy and is closer to econometrics in its focus. Study- (Gillispie, 1963).1 This approach was further developed
ing mathematical models of a large number of interacting by Ludwig Boltzmann, who was very explicit about its
economic agents, econophysics has much common ground origins (Ball, 2004, p. 69):
with the agent-based modeling and simulation. Corre-
The molecules are like individuals, . . . and
spondingly, it distances itself from the representative-
the properties of gases only remain unaltered,
agent approach of traditional economics, which, by def-
because the number of these molecules, which
inition, ignores statistical and heterogeneous aspects of
on the average have a given state, is con-
the economy.
stant.
Another direction related to econophysics has been ad- In his book Populare Schrifen, Boltzmann (1905) praises
vocated by the theoretical physicist Serge Galam since Josiah Willard Gibbs for systematic development of sta-
early 1980 under the name of sociophysics (Galam, 2004), tistical mechanics. Then, Boltzmann says:2
with the first appearance of the term by Galam et al.
(1982). It echoes the term physique sociale pro- This opens a broad perspective, if we do not
posed in the nineteenth century by Auguste Comte, the only think of mechanical objects. Lets con-
founder of sociology. Unlike econophysics, the term so- sider to apply this method to the statistics of
ciophysics did not catch on when first introduced, but living beings, society, sociology and so forth.
it is coming back with the popularity of econophysics It is worth noting that many now-famous economists
and active support from some physicists (Schweitzer, were originally educated in physics and engineering. Vil-
2003; Stauffer, 2004; Weidlich, 2000). While the prin- fredo Pareto earned a degree in mathematical sciences
ciples of both fields have much in common, econophysics and a doctorate in engineering. Working as a civil engi-
focuses on the narrower subject of economic behavior neer, he collected statistics demonstrating that distribu-
of humans, where more quantitative data is available, tions of income and wealth in a society follow a power law
whereas sociophysics studies a broader range of social (Pareto, 1897). He later became a professor of economics
issues. The boundary between econophysics and socio- at Lausanne, where he replaced Leon Walras, also an en-
physics is not sharp, and the two fields enjoy a good rap- gineer by education. The influential American economist
port (Chakrabarti, Chakraborti, and Chatterjee, 2006). Irving Fisher was a student of Gibbs. However, most
Historically, statistical mechanics was developed in the of the mathematical apparatus transferred to economics
second half of the nineteenth century by James Clerk from physics was that of Newtonian mechanics and clas-
Maxwell, Ludwig Boltzmann, and Josiah Willard Gibbs. sical thermodynamics (Mirowski, 1989; Smith and Foley,
These physicists believed in the existence of atoms and 2008). It culminated in the neoclassical concept of mech-
developed mathematical methods for describing their sta- anistic equilibrium where the forces of supply and de-
tistical properties. There are interesting connections be- mand balance each other. The more general concept
tween the development of statistical physics and statistics of statistical equilibrium largely eluded mainstream eco-
of social phenomena, which were recently highlighted by nomics.
the science journalist Philip Ball (2002, 2004). With time, both physics and economics became more
formal and rigid in their specializations, and the social
Collection and study of social numbers, such as the origin of statistical physics was forgotten. The situation
rates of death, birth, and marriage, has been growing is well summarized by Philip Ball (Ball, 2004, p. 69):
progressively since the seventeenth century (Ball, 2004,
Today physicists regard the application of
Ch. 3). The term statistics was introduced in the eigh-
statistical mechanics to social phenomena as
teenth century to denote these studies dealing with the
a new and risky venture. Few, it seems, re-
civil states, and its practitioners were called statists.
call how the process originated the other way
Popularization of social statistics in the nineteenth cen-
around, in the days when physical science
tury is particularly accredited to the Belgian astronomer
and social science were the twin siblings of a
Adolphe Quetelet. Before the 1850s, statistics was con-
mechanistic philosophy and when it was not
sidered an empirical arm of political economy, but then
in the least disreputable to invoke the habits
it started to transform into a general method of quanti-
of people to explain the habits of inanimate
tative analysis suitable for all disciplines. It stimulated
particles.
physicists to develop statistical mechanics in the second
half of the nineteenth century. Some physicists and economists attempted to connect
the two disciplines during the twentieth century. Fred-
Rudolf Clausius started development of the kinetic the- erick Soddy (1933), the Nobel Prize winner in chemistry
ory of gases, but it was James Clerk Maxwell who made
a decisive step of deriving the probability distribution of
velocities of molecules in a gas. Historical studies show
(Ball, 2004, Ch. 3) that, in developing statistical mechan- 1 V.M.Y. is grateful to Stephen G. Brush for this reference.
ics, Maxwell was strongly influenced and encouraged by 2 Cited from Boltzmann (2006). V.M.Y. is grateful to Michael
the widespread popularity of social statistics at the time E. Fisher for this quote.
 3

for his work on radioactivity, published the book Wealth, series WEHIA/ESHIA, which deals with heterogeneous
Virtual Wealth and Debt, where he argued that the real interacting agents, regularly includes sessions on econo-
wealth is derived from the energy use in transforming raw physics. More information can be found in the reviews
materials into goods and services, and not from monetary by Farmer, Shubik, and Smith (2005); Samanidou et al.
transactions. He also warned about dangers of exces- (2007) and on the Web portal Econophysics Forum
sive debt and related virtual wealth, thus anticipating http://www.unifr.ch/econophysics/.
the Great Depression. His ideas were largely ignored at
the time, but resonate today (Defilla, 2007). The theo-
retical physicist Ettore Majorana (1942) argued in favor II. STATISTICAL MECHANICS OF MONEY
of applying the laws of statistical physics to social phe- DISTRIBUTION
nomena in a paper published after his mysterious disap-
pearance. The statistical physicist Elliott Montroll co- When modern econophysics started in the middle of
authored the book Introduction to Quantitative Aspects 1990s, its attention was primarily focused on analysis of
of Social Phenomena (Montroll and Badger, 1974). Sev- financial markets. Soon after, another direction, closer
eral economists (Blume, 1993; Durlauf, 1997; Foley, 1994; to economics than finance, has emerged. It studies the
Follmer, 1974) applied statistical physics to economic probability distributions of money, wealth, and income
problems. The mathematicians Farjoun and Machover in a society and overlaps with the long-standing line of
(1983) argued that many paradoxes in classical polit- research in economics studying inequality in a society.3
ical economy can be resolved if one adopts a prob- Many papers in the economic literature (Champernowne,
abilistic approach. An early attempt to bring to- 1953; Gibrat, 1931; Kalecki, 1945) use a stochastic pro-
gether the leading theoretical physicists and economists cess to describe dynamics of individual wealth or income
at the Santa Fe Institute was not entirely successful and to derive their probability distributions. One might
(Anderson, Arrow, and Pines, 1988). However, by the call this a one-body approach, because wealth and
late 1990s, the attempts to apply statistical physics to income fluctuations are considered independently for
social phenomena finally coalesced into the robust move- each economic agent. Inspired by Boltzmanns kinetic
ments of econophysics and sociophysics. theory of collisions in gases, econophysicists introduced
an alternative, two-body approach, where agents per-
Current standing of econophysics within the physics
form pairwise economic transactions and transfer money
and economics communities is mixed. Although an en-
from one agent to another. Actually, this approach
try on econophysics has appeared in the New Palgrave
was pioneered by the sociologist John Angle (1986,
Dictionary of Economics (Rosser, 2008a), it is fair to
1992, 1993, 1996, 2002) already in the 1980s. However,
say that econophysics has not been accepted yet by
his work was largely unknown until it was brought
mainstream economics. Nevertheless, a number of open-
to the attention of econophysicists by the economist
minded, nontraditional economists have joined this move-
Thomas Lux (2005). Now, Angles work is widely
ment, and the number is growing. Under these cir-
cited in econophysics literature (Angle, 2006). Mean-
cumstances, econophysicists have most of their papers
while, the physicists Ispolatov, Krapivsky, and Redner
published in physics journals. The journal Physica A:
(1998) independently introduced a statistical model
Statistical Mechanics and its Applications has emerged
of pairwise money transfer between economic agents,
as the leader in econophysics publications and has even
which is equivalent to the model of Angle. Soon,
attracted submissions from some bona fide economists.
three influential papers by Bouchaud and Mezard
Gradually, reputable economics journals are also start-
(2000); Chakraborti and Chakrabarti (2000);
ing to publish econophysics papers (Gabaix et al., 2006;
Dragulescu and Yakovenko (2000) appeared and
Lux and Sornette, 2002; Wyart and Bouchaud, 2007).
generated an expanding wave of follow-up publications.
The mainstream physics community is generally sym-
For pedagogical reasons, we start reviewing this subject
pathetic to econophysics, although it is not uncom-
with the simplest version of the pairwise money transfer
mon for econophysics papers to be rejected by Phys-
models presented in Dragulescu and Yakovenko (2000).
ical Review Letters on the grounds that it is not
This model is the most closely related to the traditional
physics. There is a PACS number for econophysics,
statistical mechanics, which we briefly review first. Then
and Physical Review E has published many papers on
we discuss the other models mentioned above, as well as
this subject. There are regular conferences on econo-
numerous follow-up papers.
physics, such as Applications of Physics in Financial
Interestingly, the study of pairwise money trans-
Analysis (sponsored by the European Physical Soci-
fer and the resulting statistical distribution of money
ety), Nikkei Econophysics Symposium, Econophysics Col-
loquium, and Econophys-Kolkata (Chakrabarti, 2005;
Chatterjee, Yarlagadda, and Chakrabarti, 2005). Econo-
physics sessions are included in the annual meetings of 3 See, e.g., Atkinson and Bourguignon (2000);
physical societies and statistical physics conferences. The Atkinson and Piketty (2007); Champernowne (1953);
overlap with economists is the strongest in the field of Champernowne and Cowell (1998); Gibrat (1931); Kakwani
agent-based simulation. Not surprisingly, the conference (1980); Kalecki (1945); Pareto (1897); Piketty and Saez (2003).
 4

has virtually no counterpart in modern economics, so not change. This quantity is given by the combinatorial
econophysicists initiated a new direction here. Only formula in terms of the factorials
the search theory of money (Kiyotaki and Wright, N!
1993) is somewhat related to it. This theory was W = . (3)
N1 ! N2 ! N3 ! . . .
an inspiration for the early econophysics paper by
Bak, Nrrelykke, and Shubik (1999) studying dynamics The logarithm of multiplicity is called the entropy S =
of money. However, a probability distribution of money ln W . In the limit of large numbers, the entropy per
among the agents was only recently obtained within particle can be written in the following form using the
the search-theoretical approach by the economist Miguel Stirling approximation for the factorials
Molico (2006). His distribution is qualitatively similar to X Nk  Nk 
the distributions found by Angle (1986, 1992, 1993, 1996, S X
= ln = Pk ln Pk . (4)
2002, 2006) and by Ispolatov, Krapivsky, and Redner N N N
k k
(1998), but its functional form is unknown, because it
was obtained only numerically. Now we would like to find what distribution of particles
among different energy states has the highest entropy,
i.e., the highest multiplicity,
P provided the total energy of
the system, E = k Nk k , has a fixed value. Solution of
A. The Boltzmann-Gibbs distribution of energy
this problem can be easily obtained using the method of
Lagrange multipliers (Wannier, 1987), and the answer is
The fundamental law of equilibrium statistical me-
given by the exponential distribution (1).
chanics is the Boltzmann-Gibbs distribution. It states
The same result can be also derived from the er-
that the probability P () of finding a physical system or
godic theory, which says that the many-body system oc-
subsystem in a state with the energy is given by the
cupies all possible states of a given total energy with
exponential function
equal probabilities. Then it is straightforward to show
(Lopez-Ruiz et al., 2008) that the probability distribu-
P () = c e/T , (1)
tion of the energy of an individual particle is given by
Eq. (1).
where T is the temperature, and c is a normalizing con-
stant (Wannier, 1987). Here we set the Boltzmann con-
stant kB to unity by choosing the energy units for mea- B. Conservation of money
suring the physical temperature T . Then, the expecta-
tion value of any physical variable x can be obtained as The derivations outlined in Sec. II.A are very general
P k /T
and only use the statistical character of the system and
k xk e the conservation of energy. So, one may expect that the
hxi = P
, (2)
k /T
ke exponential Boltzmann-Gibbs distribution (1) would ap-
ply to other statistical systems with a conserved quantity.
where the sum is taken over all states of the system. The economy is a big statistical system with millions
Temperature is equal to the average energy per particle: of participating agents, so it is a promising target for ap-
T hi, up to a numerical coefficient of the order of 1. plications of statistical mechanics. Is there a conserved
Eq. (1) can be derived in different ways (Wannier, quantity in the economy? Dragulescu and Yakovenko
1987). All derivations involve the two main ingredients: (2000) argued that such a conserved quantity is money
statistical character of the system and conservation of m. Indeed, the ordinary economic agents can only re-
energy . One of the shortest derivations can be sum- ceive money from and give money to other agents. They
marized as follows. Let us divide the system into two are not permitted to manufacture money, e.g., to print
(generally unequal) parts. Then, the total energy is the dollar bills. Let us consider an economic transaction be-
sum of the parts: = 1 + 2 , whereas the probability tween agents i and j. When the agent i pays money
is the product of probabilities: P () = P (1 ) P (2 ). The m to the agent j for some goods or services, the money
only solution of these two equations is the exponential balances of the agents change as follows
function (1).
A more sophisticated derivation, proposed by Boltz- mi mi = mi m,
mann, uses the concept of entropy. Let us consider N mj mj = mj + m. (5)
particles with the total energy E. Let us divide the en-
ergy axis into small intervals (bins) of width and The total amount of money of the two agents before and
count the number of particles Nk having the energies after transaction remains the same
from k to k + . The ratio Nk /N = Pk gives the mi + mj = mi + mj , (6)
probability for a particle to have the energy k . Let us
now calculate the multiplicity W , which is the number i.e., there is a local conservation law for money. The rule
of permutations of the particles between different energy (5) for the transfer of money is analogous to the trans-
bins such that the occupation numbers of the bins do fer of energy from one molecule to another in molecular
 5

collisions in a gas, and Eq. (6) is analogous to conser- is also a common idealization in some economic liter-
vation of energy in such collisions. Conservative models ature (Kiyotaki and Wright, 1993; Molico, 2006). This
of this kind are also studied in some economic literature means that money balances of the agents cannot go be-
(Kiyotaki and Wright, 1993; Molico, 2006). low zero: mi 0 for all i. Transaction (5) takes place
We should emphasize that, in the model of only when an agent has enough money to pay the price:
Dragulescu and Yakovenko (2000) [as in the economic mi m, otherwise the transaction does not take place.
models of Kiyotaki and Wright (1993); Molico (2006)], If an agent spends all money, the balance drops to zero
the transfer of money from one agent to another repre- mi = 0, so the agent cannot buy any goods from other
sents payment for goods and services in a market econ- agents. However, this agent can still receive money from
omy. However, the model of Dragulescu and Yakovenko other agents for delivering goods or services to them. In
(2000) only keeps track of money flow, but does not keep real life, money balance dropping to zero is not at all
track of what goods and service are delivered. One reason unusual for people who live from paycheck to paycheck.
for this is that many goods, e.g., food and other supplies, Enforcement of the local conservation law (6) is the
and most services, e.g., getting a haircut or going to a key feature for successful functioning of money. If the
movie, are not tangible and disappear after consump- agents were permitted to manufacture money, they
tion. Because they are not conserved, and also because would be printing money and buying all goods for noth-
they are measured in different physical units, it is not ing, which would be a disaster. The physical medium of
very practical to keep track of them. In contrast, money money is not essential here, as long as the local conser-
is measured in the same unit (within a given country vation law is enforced. The days of gold standard are
with a single currency) and is conserved in local transac- long gone, so money today is truly the fiat money, de-
tions (6), so it is straightforward to keep track of money clared to be money by the central bank. Money may
flow. It is also important to realize that an increase in be in the form of paper currency, but today it is more
material production does not produce an automatic in- often represented by digits on computerized bank ac-
crease in money supply. The agents can grow apples on counts. The local conservation law (6) is consistent with
trees, but cannot grow money on trees. Only a central the fundamental principles of accounting, whether in the
bank has the monopoly of changing the monetary base single-entry or the double-entry form. More discussion of
Mb (McConnell and Brue, 1996). (Debt and credit issues banks, debt, and credit will be given in Sec. II.D. How-
are discussed separately in Sec. II.D.) ever, the macroeconomic monetary policy issues, such as
Unlike, ordinary economic agents, a central bank or a money supply and money demand (Friedman and Hahn,
central government can inject money into the economy, 1990), are outside of the scope of this paper. Our goal
thus changing the total amount of money in the system. is to investigate the probability distribution of money
This process is analogous to an influx of energy into a sys- among economic agents. For this purpose, it is appropri-
tem from external sources, e.g., the Earth receives energy ate to make the simplifying macroeconomic idealizations,
from the Sun. Dealing with these situations, physicists as described above, in order to ensure overall stability of
start with an idealization of a closed system in thermal the system and existence of statistical equilibrium in the
equilibrium and then generalize to an open system sub- model. The concept of equilibrium is a very common
ject to an energy flux. As long as the rate of money idealization in economic literature, even though the real
influx from central sources is slow compared with relax- economies might never be in equilibrium. Here we ex-
ation processes in the economy and does not cause hy- tend this concept to a statistical equilibrium, which is
perinflation, the system is in quasi-stationary statistical characterized by a stationary probability distribution of
equilibrium with slowly changing parameters. This sit- money P (m), as opposed to a mechanical equilibrium,
uation is analogous to heating a kettle on a gas stove where the forces of demand and supply match.
slowly, where the kettle has a well-defined, but slowly in-
creasing, temperature at any moment of time. A flux of
money may be also produced by international transfers
C. The Boltzmann-Gibbs distribution of money
across the boundaries of a country. This process involves
complicated issues of multiple currencies in the world
and their exchange rates (McCauley, 2008). Here we use Having recognized the principle of local money con-
an idealization of a closed economy for a single country servation, Dragulescu and Yakovenko (2000) argued that
with a single currency. Such an idealization is common the stationary distribution of money P (m) should be
in economic literature. For example, in the two-volume given by the exponential Boltzmann-Gibbs function anal-
Handbook of Monetary Economics (Friedman and Hahn, ogous to Eq. (1)
1990), only the last chapter out of 23 chapters deals with
an open economy. P (m) = c em/Tm . (7)
Another potential problem with conservation of money
is debt. This issue will be discussed in Sec. II.D. As a Here c is a normalizing constant, and Tm is the money
starting point, Dragulescu and Yakovenko (2000) consid- temperature, which is equal to the average amount of
ered simple models, where debt is not permitted, which money per agent: T = hmi = M/N , where M is the total
 6

money, and N is the number of agents.4 5

N=500, M=5*10 , time=4*10 .
5
18
To verify this conjecture, Dragulescu and Yakovenko
(2000) performed agent-based computer simulations of 16
m, T
money transfers between agents. Initially all agents were 14
given the same amount of money, say, $1000. Then, a 3
pair of agents (i, j) was randomly selected, the amount 12

Probability, P(m)

log P(m)
2
m was transferred from one agent to another, and the 10
process was repeated many times. Time evolution of 1
8
the probability distribution of money P (m) is illustrated
in computer animation videos by Chen and Yakovenko 6
0
0 1000 2000 3000
(2007) and by Wright (2007). After a transitory pe- Money, m
4
riod, money distribution converges to the stationary form
shown in Fig. 1. As expected, the distribution is well fit- 2
ted by the exponential function (7).
0
Several different rules for m were considered by 0 1000 2000 3000
Money, m
4000 5000 6000
Dragulescu and Yakovenko (2000). In one model, the
transferred amount was fixed to a constant m = $1. FIG. 1 Histogram and points: Stationary probability dis-
Economically, it means that all agents were selling their tribution of money P (m) obtained in agent-based computer
products for the same price m = $1. Computer ani- simulations. Solid curves: Fits to the Boltzmann-Gibbs law
mation (Chen and Yakovenko, 2007) shows that the ini- (7). Vertical line: The initial distribution of money. From
tial distribution of money first broadens to a symmet- Dragulescu and Yakovenko (2000).
ric Gaussian curve, characteristic for a diffusion pro-
cess. Then, the distribution starts to pile up around
the m = 0 state, which acts as the impenetrable bound- (7), as in the first model. Computer animation for this
ary, because of the imposed condition m 0. As a re- model is also given by Chen and Yakovenko (2007).
sult, P (m) becomes skewed (asymmetric) and eventu- The final distribution is universal despite differ-
ally reaches the stationary exponential shape, as shown ent rules for m. To amplify this point further,
in Fig. 1. The boundary at m = 0 is analogous to Dragulescu and Yakovenko (2000) also considered a toy
the ground-state energy in statistical physics. Without model, where m was taken to be a random fraction
this boundary condition, the probability distribution of of the average amount of money of the two agents:
money would not reach a stationary state. Computer m = (mi + mj )/2. This model produced the same
animations (Chen and Yakovenko, 2007; Wright, 2007) stationary distribution (7) as the two other models.
also show howPthe entropy of money distribution, defined The models of pairwise money transfer are attrac-
as S/N = k P (mk ) ln P (mk ), grows from the initial tive in their simplicity, but they represent a rather
value S = 0, where all agents have the same money, to primitive market. Modern economy is dominated
the maximal value at the statistical equilibrium. by big firms, which consist of many agents, so
While the model with m = 1 is very simple and Dragulescu and Yakovenko (2000) also studied a model
instructive, it is not realistic, because all prices are with firms. One agent at a time is appointed to be-
taken to be the same. In another model considered by come a firm. The firm borrows capital K from another
Dragulescu and Yakovenko (2000), m in each transac- agent and returns it with interest hK, hires L agents and
tion is taken to be a random fraction of the average pays them wages , manufactures Q items of a prod-
amount of money per agent, i.e., m = (M/N ), where uct, sells them to Q agents at a price p, and receives
is a uniformly distributed random number between 0 profit F = pQ L hK. All of these agents are ran-
and 1. The random distribution of m is supposed to domly selected. The parameters of the model are opti-
represent the wide variety of prices for different products mized following a procedure from economics textbooks
in the real economy. It reflects the fact that agents buy (McConnell and Brue, 1996). The aggregate demand-
and consume many different types of products, some of supply curve for the product is taken in the form p(Q) =
them simple and cheap, some sophisticated and expen- v/Q , where Q is the quantity consumers would buy at
sive. Moreover, different agents like to consume these the price p, and and v are some parameters. The pro-
products in different quantities, so there is a variation in duction function of the firm has the traditional Cobb-
the paid amounts m, even when the unit price of the Douglas form: Q(L, K) = L K 1 , where is a param-
same product is constant. Computer simulation of this eter. Then the profit of the firm F is maximized with
model produces exactly the same stationary distribution respect to K and L. The net result of the firm activ-
ity is a many-body transfer of money, which still satis-
fies the conservation law. Computer simulation of this
model generates the same exponential distribution (7),
4 Because debt is not permitted in this model, we have M = Mb , independently of the model parameters. The reasons for
where Mb is the monetary base (McConnell and Brue, 1996). the universality of the Boltzmann-Gibbs distribution and
 7

its limitations are discussed in Sec. II.F. negative (debt D) contributions: M D = Mb . After
Well after the paper by Dragulescu and Yakovenko spending some cash in binary transactions (5), the agent
(2000) appeared, the Italian econophysicists still has the debt obligation (negative money), so the to-
Patriarca et al. (2005) found that similar ideas had tal money balance mi of the agent (net worth) becomes
been published earlier in obscure Italian journals by negative. We see that the boundary condition mi 0,
Eleonora Bennati (1988, 1993). It was proposed to discussed in Sec. II.B, does not apply when debt is per-
call these models the Bennati-Dragulescu-Yakovenko mitted, so m = 0 is not the ground state any more. The
game (Garibaldi et al., 2007; Scalas et al., 2006). The consequence of permitting debt is not a violation of the
Boltzmann distribution was independently applied to conservation law (which is still preserved in the gener-
social sciences by the physicist Jurgen Mimkes (2000); alized form for net worth), but a modification of the
Mimkes and Willis (2005) using the Lagrange principle boundary condition by permitting agents to have neg-
of maximization with constraints. The exponential ative balances mi < 0 of net worth. A more detailed
distribution of money was also found by the economist discussion of positive and negative money and the book-
Martin Shubik (1999) using a Markov chain approach keeping accounting from the econophysics point of view
to strategic market games. A long time ago, Benoit was presented by the physicist Dieter Braun (2001) and
Mandelbrot (1960, p 83) observed: Fischer and Braun (2003a,b).
Now we can repeat the simulation described in Sec.
There is a great temptation to consider the II.C without the boundary condition m 0 by allowing
exchanges of money which occur in economic agents to go into debt. When an agent needs to buy a
interaction as analogous to the exchanges of product at a price m exceeding his money balance mi ,
energy which occur in physical shocks be- the agent is now permitted to borrow the difference from
tween gas molecules. a bank and, thus, to buy the product. As a result of
this transaction, the new balance of the agent becomes
He realized that this process should result in the expo-
negative: mi = mi m < 0. Notice that the local con-
nential distribution, by analogy with the barometric dis-
servation law (5) and (6) is still satisfied, but it involves
tribution of density in the atmosphere. However, he dis-
negative values of m. If the simulation is continued fur-
carded this idea, because it does not produce the Pareto
ther without any restrictions on the debt of the agents,
power law, and proceeded to study the stable Levy distri-
the probability distribution of money P (m) never stabi-
butions. Ironically, the actual economic data, discussed
lizes, and the system never reaches a stationary state. As
in Secs. III.C and IV.A, do show the exponential distri-
time goes on, P (m) keeps spreading in a Gaussian man-
bution for the majority of the population. Moreover, the
ner unlimitedly toward m = + and m = . Because
data have a finite variance, so the stable Levy distribu-
of the generalized conservation law discussed above, the
tions are not applicable because of their infinite variance.
first moment hmi = Mb /N of the algebraically defined
money m remains constant. It means that some agents
become richer with positive balances m > 0 at the ex-
D. Models with debt pense of other agents going further into debt with nega-
tive balances m < 0, so that M = Mb + D.
Now let us discuss how the results change when debt is
Common sense, as well as the experience with the cur-
permitted.5 From the standpoint of individual economic
rent financial crisis, tells us that an economic system can-
agents, debt may be considered as negative money. When
not be stable if unlimited debt is permitted.7 In this case,
an agent borrows money from a bank (considered here as
agents can buy any goods without producing anything in
a big reservoir of money),6 the cash balance of the agent
exchange by simply going into unlimited debt. Arguably,
(positive money) increases, but the agent also acquires
the current financial crisis was caused by the enormous
a debt obligation (negative money), so the total balance
debt accumulation in the system, triggered by subprime
(net worth) of the agent remains the same. Thus, the act
mortgages and financial derivatives based on them. A
of borrowing money still satisfies a generalized conserva-
widely expressed opinion is that the current crisis is not
tion law of the total money (net worth), which is now
the problem of liquidity, i.e., a temporary difficulty in
defined as the algebraic sum of positive (cash M ) and
cash flow, but the problem of insolvency, i.e., the inher-
ent inability of many participants pay back their debts.
Detailed discussion of the current economic situation
5 The ideas presented here are quite similar to those by Soddy is not a subject of this paper. Going back to the idealized
(1933). model of money transfers, one would need to impose some
6 Here we treat the bank as being outside of the system consisting sort of modified boundary conditions in order to prevent
of ordinary agents, because we are interested in money distribu-
tion among these agents. The debt of agents is an asset for the
bank, and deposits of cash into the bank are liabilities of the bank
(McConnell and Brue, 1996). We do not go into these details in
order to keep our presentation simple. For more discussion, see 7 In qualitatively agreement with the conclusions by McCauley
Keen (2008). (2008).
 8

5 5
N=500, M=5*10 , time=4*10 .
18

16 40

Probability,P(m) (1x10 )
-3
14

)
-3
10

log(P(m)) (1x10
30
12 Model with debt, T=1800
Probability, P(m)

10 20
0.1

8 Model without debt, T=1000

-50 0 50 100 150

10 Monetary Wealth,m
6

4
0

2
-50 0 50 100 150
0
0 2000 4000 6000 8000 Monetary Wealth,m
Money, m

FIG. 2 Histograms: Stationary distributions of money with

FIG. 3 The stationary distribution of money for the required
and without debt. The debt is limited to md = 800.
reserve ratio R = 0.8. The distribution is exponential for
Solid curves: Fits to the Boltzmann-Gibbs laws with the
positive and negative money with different temperatures
money temperatures Tm = 1800 and Tm = 1000. From
T+ and T , as illustrated by the inset on log-linear scale.
Dragulescu and Yakovenko (2000).
From Xi, Ding, and Wang (2005).

unlimited growth of debt and to ensure overall stability money come from? It comes from the increase in the to-
of the system. Dragulescu and Yakovenko (2000) con- tal debt in the system. The maximal total debt is given
sidered a simple model where the maximal debt of each by D = Mb /RMb and is limited by the factor R. When
agent is limited to a certain amount md . This means the debt is maximal, the total amounts of positive, Mb /R,
that the boundary condition mi 0 is now replaced by and negative, Mb (1 R)/R, money circulate among the
the condition mi md for all agents i. Setting inter- agents in the system, so there are two constraints in the
est rates on borrowed money to be zero for simplicity, model considered by Xi, Ding, and Wang (2005). Thus,
Dragulescu and Yakovenko (2000) performed computer we expect to see the exponential distributions of positive
simulations of the models described in Sec. II.C with the and negative money characterized by two different tem-
new boundary condition. The results are shown in Fig. peratures: T+ = Mb /RN and T = Mb (1 R)/RN .
2. Not surprisingly, the stationary money distribution This is exactly what was found in computer simula-
again has the exponential shape, but now with the new tions by Xi, Ding, and Wang (2005), as shown in Fig.
boundary condition at m = md and the higher money 3. Similar two-sided distributions were also found by
temperature Td = md + Mb /N . By allowing agents to go Fischer and Braun (2003a).
into debt up to md , we effectively increase the amount of
However, in reality, the reserve requirement is not
money available to each agent by md . So, the money tem-
effective in stabilizing total debt in the system, be-
perature, which is equal to the average amount of effec-
cause it applies only to deposits from general public,
tively available money per agent, increases correspond-
but not from corporations (O Brien, 2007).8 More-
ingly.
over, there are alternative instruments of debt, includ-
Xi, Ding, and Wang (2005) considered another, more ing derivatives and various unregulated financial inno-
realistic boundary condition, where a constraint is im- vations. As a result, the total debt is not limited in
posed not on the individual debt of each agent, but practice and sometimes can reach catastrophic propor-
on the total debt of all agents in the system. This is tions. Here we briefly discuss several models with non-
accomplished via the required reserve ratio R, which stationary debt. Thus far, we did not consider the in-
is briefly explained below (McConnell and Brue, 1996). terest rates. Dragulescu and Yakovenko (2000) studied
Banks are required by law to set aside a fraction R of a simple model with different interest rates for deposits
the money deposited into bank accounts, whereas the into and loans from a bank. Computer simulations found
remaining fraction 1 R can be loaned further. If the that money distribution among the agents is still expo-
initial amount of money in the system (the money base) nential, but the money temperature slowly changes in
is Mb , then, with repeated loans and borrowing, the to-
tal amount of positive money available to the agents in-
creases to M = Mb /R, where the factor 1/R is called
the money multiplier (McConnell and Brue, 1996). This 8 Australia does not have reserve requirements, but China actively
is how banks create money. Where does this extra uses reserve requirements as a tool of monetary policy.
 9

time. Depending on the choice of parameters, the total 5

N=500, M=5*10 , =1/3.
amount of money in circulation either increases or de- 16

creases in time. A more sophisticated macroeconomic 14

model was studied by the economist Steve Keen (1995,
2000). He found that one of the regimes is the debt- 12
induced breakdown, where all economic activity stops

Probability, P(m)
under the burden of heavy debt and cannot be restarted 10

without a debt moratorium. The interest rates were 8

fixed in these models and not adjusted self-consistently.
Cockshott and Cottrell (2008) proposed a mechanism, 6
where the interest rates are set to cover probabilistic
withdrawals of deposits from a bank. In an agent-based 4

simulation of the model, Cockshott and Cottrell (2008) 2

found that money supply first increases up to a certain
limit, and then the economy experiences a spectacular 0
0 1000 2000 3000 4000 5000
crash under the weight of accumulated debt. Further Money, m
studies along these lines would be very interesting. In
the rest of the paper, we review various models without FIG. 4 Histogram: Stationary probability distribution of
debt proposed in literature. money in the multiplicative random exchange model (8) for
= 1/3. Solid curve: The exponential Boltzmann-Gibbs law.
From Dragulescu and Yakovenko (2000).
E. Proportional money transfers and saving propensity
much justification for the proportionality law (8), Angle
In the models of money transfer discussed in Sec. II.C,
(1986) connected this rule with the surplus theory of so-
the transferred amount m is typically independent of
cial stratification (Engels, 1972), which argues that in-
the money balances of the agents involved. A differ-
equality in human society develops when people can pro-
ent model was introduced in physics literature earlier by
duce more than necessary for minimal subsistence. This
Ispolatov, Krapivsky, and Redner (1998) and called the
additional wealth (surplus) can be transferred from origi-
multiplicative asset exchange model. This model also sat-
nal producers to other people, thus generating inequality.
isfies the conservation law, but the transferred amount of
In the first paper by Angle (1986), the parameter was
money is a fixed fraction of the payers money in Eq.
randomly distributed, and another parameter gave a
(5):
higher probability of winning to the agent with the higher
m = mi . (8) money balance in Eq. (5). However, in the following pa-
pers, he simplified the model to a fixed (denoted as by
The stationary distribution of money in this model, com- Angle) and equal probabilities of winning for higher- and
pared in Fig. 4 with an exponential function, is similar, lower-balance agents, which makes it completely equiv-
but not exactly equal, to the Gamma distribution: alent to the model of Ispolatov, Krapivsky, and Redner
(1998). Angle (2002, 2006) also considered a model where
P (m) = c m em/T . (9) groups of agents have different values of , simulating
the effect of education and other human capital. All
Eq. (9) differs from Eq. (7) by the power-law prefactor of these models generate a Gamma-like distribution, well
m . From the Boltzmann kinetic equation (discussed in approximated by Eq. (9).
Sec. II.F), Ispolatov, Krapivsky, and Redner (1998) de- Another model with an element of proportionality was
rived a formula relating the parameters and in Eqs. proposed by Chakraborti and Chakrabarti (2000).9 In
(8) and (9): this model, the agents set aside (save) some fraction of
their money mi , whereas the rest of their money balance
= 1 ln 2/ ln(1 ). (10)
(1)mi becomes available for random exchanges. Thus,
When payers spend a relatively small fraction of their the rule of exchange (5) becomes
money < 1/2, Eq. (10) gives > 0. In this case,
mi = mi + (1 )(mi + mj ),
the population with low money balances is reduced, and
P (0) = 0, as shown in Fig. 4. mj = mj + (1 )(1 )(mi + mj ). (11)
The economist Thomas Lux (2005) brought to the at- Here the coefficient is called the saving propensity, and
tention of physicists that essentially the same model,
called the inequality process, had been introduced and
studied much earlier by the sociologist John Angle
(1986, 1992, 1993, 1996, 2002), see also the review 9 This paper originally appeared as a follow-up e-print
by Angle (2006) for additional references. While cond-mat/0004256 on the e-print cond-mat/0001432 by
Ispolatov, Krapivsky, and Redner (1998) did not give Dragulescu and Yakovenko (2000).
 10

the random variable is uniformly distributed between F. Additive versus multiplicative models
0 and 1. It was pointed out by Angle (2006) that, by
the change of notation (1 ), Eq. (11) can be The stationary distribution of money (9) for the mod-
transformed to the same form as Eq. (8), if the random els of Sec. II.E is different from the simple exponential
variable takes only discrete values 0 and 1. Computer formula (7) found for the models of Sec. II.C. The origin
simulations by Chakraborti and Chakrabarti (2000) of of this difference can be understood from the Boltzmann
the model (11) found a stationary distribution close to kinetic equation (Lifshitz and Pitaevskii, 1981; Wannier,
the Gamma distribution (9). It was shown that the pa- 1987). This equation describes time evolution of the dis-
rameter is related to the saving propensity by the for- tribution function P (m) due to pairwise interactions:
mula = 3/(1) (Patriarca, Chakraborti, and Kaski,
dP (m)
ZZ
2004a,b; Patriarca et al., 2005; = {f[m,m ][m,m +] P (m)P (m ) (12)
Repetowicz, Hutzler, and Richmond, 2005). For 6= 0, dt
agents always keep some money, so their balances never +f[m,m+][m,m ] P (m )P (m + )} dm d.
drop to zero, and P (0) = 0, whereas for = 0 the
distribution becomes exponential. Here f[m,m ][m,m +] is the probability of transfer-
ring money from an agent with money m to an agent
with money m per unit time. This probability, multi-
plied by the occupation numbers P (m) and P (m ), gives
the rate of transitions from the state [m, m ] to the state
[m , m + ]. The first term in Eq. (12) gives the de-
population rate of the state m. The second term in Eq.
In the subsequent papers by the Kolkata school (12) describes the reversed process, where the occupation
(Chakrabarti, 2005) and related papers, the case number P (m) increases. When the two terms are equal,
of random saving propensity was studied. In these the direct and reversed transitions cancel each other sta-
models, the agents are assigned random parameters tistically, and the probability distribution is stationary:
drawn from a uniform distribution between 0 and dP (m)/dt = 0. This is the principle of detailed balance.
1 (Chatterjee, Chakrabarti, and Manna, 2004). It In physics, the fundamental microscopic equations of
was found that this model produces a power-law tail motion obey the time-reversal symmetry. This means
P (m) 1/m2 at high m. The reasons for stability of that the probabilities of the direct and reversed processes
this law were understood using the Boltzmann kinetic are exactly equal:
equation (Chatterjee, Chakrabarti, and Stinchcombe,
2005; Das and Yarlagadda, 2005; f[m,m ][m,m +] = f[m,m +][m,m ] . (13)
Repetowicz, Hutzler, and Richmond, 2005),
but most elegantly in the mean-field theory When Eq. (13) is satisfied, the detailed balance condi-
(Bhattacharyya, Chatterjee, and Chakrabarti, 2007; tion for Eq. (12) reduces to the equation P (m)P (m ) =
Chatterjee and Chakrabarti, 2007; Mohanty, 2006). P (m )P (m + ), because the factors f cancels out.
The fat tail originates from the agents whose saving The only solution of this equation is the exponential func-
propensity is close to 1, who hoard money and do not tion P (m) = c exp(m/Tm ), so the Boltzmann-Gibbs
give it back (Patriarca, Chakraborti, and Germano, distribution is the stationary solution of the Boltzmann
2006; Patriarca et al., 2005). A more rigor- kinetic equation (12). Notice that the transition prob-
ous mathematical treatment of the problem was abilities (13) are determined by the dynamical rules of
given by During, Matthes, and Toscani (2008); the model, but the equilibrium Boltzmann-Gibbs distri-
During and Toscani (2007); Matthes and Toscani bution does not depend on the dynamical rules at all.
(2008). An interesting matrix formulation of the This is the origin of the universality of the Boltzmann-
problem was presented by Gupta (2006). Relaxation Gibbs distribution. We see that it is possible to find the
rate in the money transfer models was studied by stationary distribution without knowing details of the dy-
During, Matthes, and Toscani (2008); Gupta (2008); namical rules (which are rarely known very well), as long
Patriarca et al. (2007). Dragulescu and Yakovenko as the symmetry condition (13) is satisfied.
(2000) considered a model with taxation, which also has The models considered in Sec. II.C have the time-
an element of proportionality. The Gamma distribution reversal symmetry. The model with the fixed money
was also studied for conservative models within a transfer has equal probabilities (13) of transferring
simple Boltzmann approach by Ferrero (2004) and, money from an agent with the balance m to an agent
using more complicated rules of exchange motivated with the balance m and vice versa. This is also true
by political economy, by Scafetta, Picozzi, and West when is random, as long as the probability distribution
(2004a,b). Independently, the economist Miguel Molico of is independent of m and m . Thus, the stationary
(2006) studied conservative exchange models where distribution P (m) is always exponential in these models.
agents bargain over prices in their transactions. He However, there is no fundamental reason to expect
found stationary Gamma-like distributions of money in the time-reversal symmetry in economics, where Eq. (13)
numerical simulations of these models. may be not valid. In this case, the system may have a
 11

non-exponential stationary distribution or no stationary over the Internet. There is no particular empirical ev-
distribution at all. In the model (8), the time-reversal idence for the proportional rules (8) or (11). However,
symmetry is broken. Indeed, when an agent i gives a the difference between the additive (7) and multiplicative
fixed fraction of his money mi to an agent with bal- (9) distributions may be not so crucial after all. From the
ance mj , their balances become (1 )mi and mj + mi . mathematical point of view, the difference is in the im-
If we try to reverse this process and appoint the agent j plementation of the boundary condition at m = 0. In
to be the payer and to give the fraction of her money, the additive models of Sec. II.C, there is a sharp cutoff
(mj + mi ), to the agent i, the system does not return for P (m) 6= 0 at m = 0. In the multiplicative models of
to the original configuration [mi , mj ]. As emphasized by Sec. II.E, the balance of an agent never reaches m = 0, so
Angle (2006), the payer pays a deterministic fraction of P (m) vanishes at m 0 in a power-law manner. But for
his money, but the receiver receives a random amount large m, P (m) decreases exponentially in both models.
from a random agent, so their roles are not interchange- By further modifying the rules of money transfer and
able. Because the proportional rule typically violates introducing more parameters in the models, one can ob-
the time-reversal symmetry, the stationary distribution tain even more complicated distributions (Saif and Gade,
P (m) in multiplicative models is typically not exponen- 2007; Scafetta and West, 2007). However, one can ar-
tial.10 Making the transfer dependent on the money bal- gue that parsimony is the virtue of a good mathematical
ance of the payer effectively introduces Maxwells demon model, not the abundance of additional assumptions and
into the model. Another view on the time-reversal sym- parameters, whose correspondence to reality is hard to
metry in economic dynamics was presented by Ao (2007). verify.
These examples show that the Boltzmann-Gibbs dis-
tribution does not necessarily hold for any conservative
III. STATISTICAL MECHANICS OF WEALTH
model. However, it is universal in a limited sense. For DISTRIBUTION
a broad class of models that have time-reversal symme-
try, the stationary distribution is exponential and does In the econophysics literature on exchange models, the
not depend on details of a model. Conversely, when terms money and wealth are often used interchange-
the time-reversal symmetry is broken, the distribution ably. However, economists emphasize the difference be-
may depend on details of a model. The difference be- tween these two concepts. In this section, we review the
tween these two classes of models may be rather sub- models of wealth distribution, as opposed to money dis-
tle. Deviations from the Boltzmann-Gibbs law may oc- tribution.
cur only if the transition rates f in Eq. (13) explicitly
depend on the agents money m or m in an asymmetric
manner. Dragulescu and Yakovenko (2000) performed a A. Models with a conserved commodity
computer simulation where the direction of payment was
randomly fixed in advance for every pair of agents (i, j). What is the difference between money and wealth?
In this case, money flows along directed links between Dragulescu and Yakovenko (2000) argued that wealth wi
the agents: i j k, and the time-reversal symme- is equal to money mi plus the other property that an
try is strongly violated. This model is closer to the real agent i has. The latter may include durable material
economy, where one typically receives money from an property, such as houses and cars, and financial instru-
employer and pays it to a grocery store. Nevertheless, ments, such as stocks, bonds, and options. Money (paper
the Boltzmann-Gibbs distribution was still found in this cash, bank accounts) is generally liquid and countable.
model, because the transition rates f do not explicitly de- However, the other property is not immediately liquid
pend on m and m and do not violate Eq. (13). A more and has to be sold first (converted into money) to be
general study of money exchange models on directed net- used for other purchases. In order to estimate the mon-
works was presented by Chatterjee (2009). etary value of property, one needs to know its price p.
In the absence of detailed knowledge of real mi- In the simplest model, let us consider just one type of
croscopic dynamics of economic exchanges, the semi- property, say, stocks s. Then the wealth of an agent i is
universal Boltzmann-Gibbs distribution (7) is a nat- given by
ural starting point. Moreover, the assumption of
Dragulescu and Yakovenko (2000) that agents pay the wi = mi + p si . (14)
same prices m for the same products, independent of
their money balances m, seems very appropriate for the It is assumed that the price p is common for all agents
modern anonymous economy, especially for purchases and is established by some kind of market process, such
as an auction, and may change in time.
It is reasonable to startPwith a model where both
the totalPmoney M = i mi and the total stock
10 However, when m is a fraction of the total money mi +mj of the S = s
i i are conserved (Ausloos and Pekalski,
two agents, the model is time-reversible and has the exponential 2007; Chakraborti, Pradhan, and Chakrabarti, 2001;
distribution, as discussed in Sec. II.C. Chatterjee and Chakrabarti, 2006). The agents pay
 12

money to buy stock and sell stock to get money, theless W remains constant on average, because the
and so on. Although P M and S are conserved, the total money M and stock S are conserved. A dif-
total wealth W = i wi is generally not conserved ferent model for wealth distribution was proposed by
(Chatterjee and Chakrabarti, 2006), because of price Bouchaud and Mezard (2000). In this model, time evo-
fluctuation in Eq. (14). This is an important difference lution of the wealth wi of an agent i is given by the
from the money transfers models of Sec. II. The wealth stochastic differential equation
wi of an agent i, not participating in any transactions,
may change when transactions between other agents dwi X X
= i (t) wi + Jij wj Jji wi , (15)
establish a new price p. Moreover, the wealth wi of an dt
j(6=i) j(6=i)
agent i does not change after a transaction with an agent
j. Indeed, in exchange for paying money m, the agent where i (t) is a Gaussian random variable with the mean
i receives the stock s = m/p, so her total wealth hi and the variance 2 2 . This variable represents growth
(14) remains the same. Theoretically, the agent can or loss of wealth of an agent due to investment in stock
instantaneously sell the stock back at the same price and market. The last two terms describe transfer of wealth
recover the money paid. If the price p never changes, between different agents, which is taken to be propor-
then the wealth wi of each agent remains constant, tional to the wealth of the payers with the coefficients Jij .
despite transfers of money and stock between agents. So, the model (15) is multiplicative and invariant under
We see that redistribution of wealth in this model the scale transformation wi Zwi . For simplicity, the
is directly related to price fluctuations. A math- exchange fractions are taken to be the same for all agents:
ematical model of this process was studied by Jij = J/N for all i 6= j, where N is the total number of
Silver, Slud, and Takamoto (2002). In this model, the agents. In this case, the last two terms in P Eq. (15) can
agents randomly change preferences for the fraction of be written as J(hwi wi ), where hwi = i wi /N is the
their wealth invested in stocks. As a result, some agents average wealth per agent. This case represents a mean-
offer stock for sale and some want to buy it. The price p field model, where all agents feel the same environment.
is determined from the market-clearing auction matching It can be easily shown that the average wealth increases
supply and demand. Silver, Slud, and Takamoto (2002) 2
in time as hwit = hwi0 e(hi+ )t . Then, it makes more
demonstrated in computer simulations and proved ana- sense to consider the relative wealth wi = wi /hwit . Eq.
lytically using the theory of Markov processes that the (15) for this variable becomes
stationary distribution P (w) of wealth w in this model
is given by the Gamma distribution, as in Eq. (9). dwi
Various modifications of this model considered by Lux = (i (t) hi 2 ) wi + J(1 wi ). (16)
dt
(2005), such as introducing monopolistic coalitions, do
not change this result significantly, which shows robust- The probability distribution P (w, t) for the stochastic
ness of the Gamma distribution. For models with a con- differential equation (16) is governed by the Fokker-
served commodity, Chatterjee and Chakrabarti (2006) Planck equation
found the Gamma distribution for a fixed saving propen-
[J(w 1) + 2 w]P
 
sity and a power-law tail for a distributed saving propen- P (wP )
= + 2 w . (17)
sity. t w w w
Another model with conserved money and stock was
studied by Raberto et al. (2003) for an artificial stock The stationary solution (P/t = 0) of this equation is
market, where traders follow different investment strate- given by the following formula
gies: random, momentum, contrarian, and fundamen- 2

talist. Wealth distribution in the model with random eJ/ w

P (w) = c . (18)
traders was found have a power-law tail P (w) 1/w2 w2+J/2
for large w. However, unlike in other simulations, where The distribution (18) is quite different from the
all agents initially have equal balances, here the initial Boltzmann-Gibbs (7) and Gamma (9) distributions. Eq.
money and stock balances of the agents were randomly (18) has a power-law tail at large w and a sharp cutoff at
populated according to a power law with the same ex- small w. Eq. (15) is a version of the generalized Lotka-
ponent. This raises the question whether the observed Volterra model, and the stationary distribution (18) was
power-law distribution of wealth is an artifact of the ini- also obtained by Solomon and Richmond (2001, 2002).
tial conditions, because equilibration of the upper tail The model was generalized to include negative wealth by
may take a very long simulation time. Huang (2004).
Bouchaud and Mezard (2000) used the mean-field ap-
proach. A similar result was found for a model with
B. Models with stochastic growth of wealth pairwise interaction between agents by Slanina (2004).
In his model, wealth is transferred between the agents
Although the total wealth W is not exactly con- following the proportional rule (8), but, in addition, the
served in the models considered in Sec. III.A, never- wealth of the agents increases by the factor 1 + in each
 13

transaction. This factor is supposed to reflect creation United Kingdom, IR data for 1996
100%
of wealth in economic interactions. Because the total
wealth in the system increases, it makes sense to con-
sider the distribution of relative wealth P (w). In the

Cumulative percent of people

limit of continuous trading, Slanina (2004) found the 10% BoltzmannGibbs

same stationary distribution (18). This result was re-

produced using a mathematically more involved treat- 100%
ment of this model by Cordier, Pareschi, and Toscani 1% Pareto
(2005); Pareschi and Toscani (2006). Numerical sim-
ulations of the models with stochastic noise by
Scafetta, Picozzi, and West (2004a,b) also found a power 0.1%
law tail for large w. Equivalence between the models with
10%
pairwise exchange and exchange with a reservoir was dis- 0 20 40 60 80 100
Total net capital, kpounds
cussed by Basu and Mohanty (2008).
0.01%
We now contrast the models discussed in Secs. III.A 10 100 1000
Total net capital (wealth), kpounds
and III.B. In the former case, where money and com-
modity are conserved, and wealth does not grow, the FIG. 5 Cumulative probability distribution of net wealth in
distribution of wealth is given by the Gamma distribu- the UK shown on log-log (main panel) and log-linear (inset)
tion with the exponential tail for large w. In the latter scales. Points represent the data from the Inland Revenue,
models, wealth grows in time exponentially, and the dis- and solid lines are fits to the exponential (Boltzmann-Gibbs)
tribution of relative wealth has a power-law tail for large and power (Pareto) laws. From Dragulescu and Yakovenko
w. These results suggest that the presence of a power- (2001b).
law tail is a nonequilibrium effect that requires constant
growth or inflation of the economy, but disappears for a
closed system with conservation laws. bution among the living. Using an adjustment procedure
The discussed models were reviewed based on the age, gender, and other characteristics of the
by Chatterjee and Chakrabarti (2007); deceased, the UK tax agency, the Inland Revenue, recon-
Richmond, Hutzler, Coelho, and Repetowicz (2006); structed the wealth distribution of the whole population
Richmond, Repetowicz, Hutzler, and Coelho (2006); of the UK (Her Majesty Revenue and Customs, 2003).
Yakovenko (2009) and in the popular article by Fig. 5 shows the UK data for 1996 reproduced from
Hayes (2002). Because of lack of space, we Dragulescu and Yakovenko (2001b). R The figure shows

omit discussion of models with wealth condensa- the cumulative probability C(w) = w P (w ) dw as a
tion (Bouchaud and Mezard, 2000; Braun, 2006; function of the personal net wealth w, which is composed
Burda et al., 2002; Ispolatov, Krapivsky, and Redner, of assets (cash, stocks, property, household goods, etc.)
1998; Pianegonda et al., 2003), where a few agents accu- and liabilities (mortgages and other debts). Because sta-
mulate a finite fraction of the total wealth, and studies of tistical data are usually reported at non-uniform intervals
wealth distribution on complex networks (Coelho et al., of w, it is more practical to plot the cumulative proba-
2005; Di Matteo, Aste, and Hyde, 2004; Hu et al., 2006, bility distribution C(w) rather than its derivative, the
2007; Iglesias et al., 2003). So far, we discussed the probability density P (w). Fortunately, when P (w) is an
models with long-range interaction, where any agent can exponential or a power-law function, then C(w) is also
exchange money and wealth with any other agent. A an exponential or a power-law function.
local model, where agents trade only with the nearest The main panel in Fig. 5 shows a plot of C(w) on the
neighbors, was studied by Bak, Nrrelykke, and Shubik log-log scale, where a straight line represents a power-
(1999). law dependence. The figure shows that the distribu-
tion follows a power law C(w) 1/w with the expo-
nent = 1.9 for the wealth greater than about 100 k.
C. Empirical data on money and wealth distributions The inset in Fig. 5 shows the same data on the log-
linear scale, where a straight line represents an expo-
It would be interesting to compare theoretical results nential dependence. We observe that, below 100 k,
for money and wealth distributions in various models the data are well fitted by the exponential distribution
with empirical data. Unfortunately, such empirical data C(w) exp(w/Tw ) with the effective wealth temper-
are difficult to find. Unlike income, which is discussed in ature Tw = 60 k (which corresponds to the median
Sec. IV, wealth is not routinely reported by the majority wealth of 41 k). So, the distribution of wealth is char-
of individuals to the government. However, in some coun- acterized by the Pareto power law in the upper tail of
tries, when a person dies, all assets must be reported for the distribution and the exponential Boltzmann-Gibbs
the purpose of inheritance tax. So, in principle, there ex- law in the lower part of the distribution for the great
ist good statistics of wealth distribution among dead peo- majority (about 90%) of the population. Similar results
ple, which, of course, is different from the wealth distri- are found for the distribution of income, as discussed in
 14

Sec. IV. One may speculate that wealth distribution in United States, IRS data for 1997
100%
the lower part is dominated by distribution of money, be-
cause the corresponding people do not have other signif-
icant assets (Levy and Levy, 2003), so the results of Sec.

Cumulative percent of returns

BoltzmannGibbs
II give the Boltzmann-Gibbs law. On the other hand,
10%
the upper tail of wealth distribution is dominated by in-
vestment assess (Levy and Levy, 2003), where the results 100%
of Sec. III.B give the Pareto law. The power law was Pareto
studied by many researchers (Klass et al., 2007; Levy,
2003; Levy and Levy, 2003; Sinha, 2006) for the upper- 1%
10%
tail data, such as the Forbes list of 400 richest people. On
the other hand, statistical surveys of the population, such
0 20 40 60 80 100
as the Survey of Consumer Finance (Diaz-Gimenez et al., AGI, k$
1997) and the Panel Study of Income Dynamics (PSID), 0.1%
1 10 100 1000
give more information about the lower part of the wealth Adjusted Gross Income, k$
distribution. Curiously, Abul-Magd (2002) found that
the wealth distribution in the ancient Egypt was consis- FIG. 6 Cumulative probability distribution of tax returns for
USA in 1997 shown on log-log (main panel) and log-linear
tent with Eq. (18). Hegyi et al. (2007) found a power-law
(inset) scales. Points represent the Internal Revenue Service
tail for the wealth distribution of aristocratic families in data, and solid lines are fits to the exponential and power-law
medieval Hungary. functions. From Dragulescu and Yakovenko (2003).
For direct comparison with the results of Sec. II, it
would be interesting to find data on the distribution of
money, as opposed to the distribution of wealth. Making IV. DATA AND MODELS FOR INCOME DISTRIBUTION
a reasonable assumption that most people keep most of
their money in banks, one can approximate the distribu- In contrast to money and wealth distributions, more
tion of money by the distribution of balances on bank empirical data are available for the distribution of in-
accounts. (Balances on all types of bank accounts, such come r from tax agencies and population surveys. In
as checking, saving, and money manager, associated with this section, we first present empirical data on income
the same person should be added up.) Despite imperfec- distribution and then discuss theoretical models.
tions (people may have accounts in different banks or
not keep all their money in banks), the distribution of
balances on bank accounts would give valuable informa- A. Empirical data on income distribution
tion about the distribution of money. The data for a
large enough bank would be representative of the distri- Empirical studies of income distribution have a long
bution in the whole economy. Unfortunately, it has not history in the economic literature.11 Many articles
been possible to obtain such data thus far, even though on this subject appear in the journal Review of In-
it would be completely anonymous and not compromise come and Wealth, published on behalf of the In-
privacy of bank clients. ternational Association for Research in Income and
The data on the distribution of bank accounts balances Wealth. Following the work by Pareto (1897), much
would be useful, e.g., to the Federal Deposits Insurance attention was focused on the power-law upper tail
Company (FDIC) of the USA. This government agency of income distribution and less on the lower part.
insures bank deposits of customers up to a certain max- In contrast to more complicated functions discussed
imal balance. In order to estimate its exposure and the in the economic literature (Atkinson and Bourguignon,
change in exposure due to a possible increase in the limit, 2000; Champernowne and Cowell, 1998; Kakwani, 1980),
FDIC would need to know the probability distribution of Dragulescu and Yakovenko (2001a) demonstrated that
balances on bank accounts. It is quite possible that FDIC the lower part of income distribution can be well
may already have such data. fitted with the simple exponential function P (r) =
c exp(r/Tr ), which is characterized by just one
Measuring the probability distribution of money would parameter, the income temperature Tr . Then
be also very useful for determining how much people Dragulescu and Yakovenko (2001b, 2003) showed that
can, in principle, spend on purchases (without going into the whole income distribution can be fitted by an expo-
debt). This is different from the distribution of wealth, nential function in the lower part and a power-law func-
where the property component, such as a house, a car,
or retirement investment, is effectively locked up and, in
most cases, is not easily available for consumer spending.
Thus, although wealth distribution may reflect the dis- 11 See, e.g., Atkinson and Bourguignon (2000);
tribution of economic power, the distribution of money Atkinson and Piketty (2007); Champernowne and Cowell
is more relevant for immediate consumption. (1998); Kakwani (1980); Piketty and Saez (2003).
 15

Adjusted gross income in 2001 dollars, k$ US, IRS data for 1983 and 2000
100% 4.017 40.17 401.70 4017 100%
100% 1 4%
1990, 27.06 k$ Gini from IRS data
1990 1991, 27.70 k$ 90 19%
s 1992, 28.63 k$ Gini=(1+f)/2
10% 1993, 29.31 k$ 10%
BoltzmannGibbs 1994, 30.23 k$ 80
1995, 31.71 k$
Cumulative percent of returns

Cumulative percent of income

1996, 32.99 k$
1997, 34.63 k$ 70 0.5
100% 1998, 36.33 k$ 1%
1999, 38.00 k$
2000, 39.76 k$ 60
19
80 2001, 40.17 k$
s
10% 0.1% 50
1983, 19.35 k$ 0
1984, 20.27 k$ 1980 1985 1990 1995 2000
40 Year
1% 1985, 21.15 k$ 0.01%
1986, 22.28 k$ 30 1983
1987, 24.13 k$ Pareto
1988, 25.35 k$ 2000
0.1% 20
1989, 26.38 k$
10
0.01%
0.1 1 10 100 0
Rescaled adjusted gross income 0 10 20 30 40 50 60 70 80 90 100%
Cumulative percent of tax returns

FIG. 7 Cumulative probability distribution of tax returns

FIG. 8 Main panel: Lorenz plots for income distribu-
plotted on log-log scale versus r/Tr (the annual income r nor-
tion in 1983 and 2000. The data points are from the
malized by the average income Tr in the exponential part of
IRS (Strudler, Petska, and Petska, 2003), and the theoret-
the distribution). The IRS data points are for 19832001, and
ical curves represent Eq. (20) with the parameter f de-
the columns of numbers give the values of Tr for the corre-
duced from Fig. 7. Inset: The closed circles are the IRS
sponding years. From Silva and Yakovenko (2005).
data (Strudler, Petska, and Petska, 2003) for the Gini coef-
ficient G, and the open circles show the theoretical formula
G = (1 + f )/2. From Silva and Yakovenko (2005).
tion in the upper part, as shown in Fig. 6. The straight
line on the log-linear scale in the inset of Fig. 6 demon-
strates the exponential Boltzmann-Gibbs law, and the
straight line on the log-log scale in the main panel illus-
class is extremely stable and does not change in time, de-
trates the Pareto power law. The fact that income distri-
spite gradual increase in the average income in nominal
bution consists of two distinct parts reveals the two-class
dollars. This observation suggests that the lower-class
structure of the American society (Silva and Yakovenko,
distribution is in statistical thermal equilibrium.
2005; Yakovenko and Silva, 2005). Coexistence of the ex-
ponential and power-law distributions is also known in On the other hand, as Fig. 7 shows, income distribu-
plasma physics and astrophysics, where they are called tion of the upper class does not rescale and significantly
the thermal and superthermal parts (Collier, 2004; changes in time. Silva and Yakovenko (2005) found that
Desai et al., 2003; Hasegawa et al., 1985). The boundary the exponent of the power law C(r) 1/r decreased
between the lower and upper classes can be defined as the from 1.8 in 1983 to 1.4 in 2000. This means that the up-
intersection point of the exponential and power-law fits per tail became fatter. Another useful parameter is the
in Fig. 6. For 1997, the annual income separating the two total income of the upper class as the fraction f of the to-
classes was about 120 k$. About 3% of the population tal income in the system. The fraction f increased from
belonged to the upper class, and 97% belonged to the 4% in 1983 to 20% in 2000 (Silva and Yakovenko, 2005).
lower class. However, in year 2001, increased and f decreased, in-
Silva and Yakovenko (2005) studied time evolution of dicating that the upper tail was reduced after the stock
income distribution in the USA during 19832001 using market crash at that time. These results indicate that
the data from the Internal Revenue Service (IRS), the the upper tail is highly dynamical and not stationary. It
government tax agency. The structure of income dis- tends to swell during the stock market boom and shrink
tribution was found to be qualitatively the same for all during the bust. Similar results were found for Japan
years, as shown in Fig. 7. The average income in nom- (Aoyama et al., 2003; Fujiwara et al., 2003; Souma, 2001,
inal dollars has approximately doubled during this time 2002).
interval. So, the horizontal axis in Fig. 7 shows the nor-
malized income r/Tr , where the income temperature Although relative income inequality within the lower
Tr was obtained by fitting of the exponential part of the class remains stable, the overall income inequality in the
distribution for each year. The values of Tr are shown USA has increased significantly as a result of the tremen-
in Fig. 7. The plots for the 1980s and 1990s are shifted dous growth of the income of the upper class. This is
vertically for clarity. We observe that the data points in illustrated by the Lorenz curve and the Gini coefficient
the lower-income part of the distribution collapse on the shown in Fig. 8. The Lorenz curve (Kakwani, 1980) is a
same exponential curve for all years. This demonstrates standard way of representing income distribution in the
that the shape of the income distribution for the lower economic literature. It is defined in terms of two coordi-
 16

60
diagonal line, divided by the area of the triangle be-

Percentage of total income in tail

30%
neath the diagonal line (Kakwani, 1980). Time evo-
50
lution of the Gini coefficient, as computed by the IRS

Average income in k$
25% (Strudler, Petska, and Petska, 2003), is shown in the in-
40
r set of Fig. 8. Dragulescu and Yakovenko (2001a) derived
20% analytically the result that G = 1/2 for a purely exponen-
30
15% T tial distribution. In the first approximation, the values
r
of G shown in the inset of Fig. 8 are indeed close to the
20
10% theoretical value 1/2. If we take into account the upper
f
tail using Eq. (20), the formula for the Gini coefficient be-
5% 10
comes G = (1 + f )/2 (Silva and Yakovenko, 2005). The
0% 0
inset in Fig. 8 shows that this formula gives a very good
1985 1990 1995 2000 2005 fit to the IRS data for the 1990s using the values of f
Years deduced from Fig. 7. The values G < 1/2 in the 1980s
FIG. 9 Historical evolution of the parameters hri, Tr , and the cannot be captured by this formula, because the Lorenz
fraction of income f going to the upper tail, as defined in Eq. data points are slightly above the theoretical curve for
(21). From Banerjee (2008). 1983 in Fig. 8. Overall, we observe that income inequal-
ity has been increasing for the last 20 years, because of
swelling of the Pareto tail, but decreased in 2001 after
nates x(r) and y(r) depending on a parameter r: the stock market crash.
Rr It is easy to show that the parameter f in Eq. (20) and
r in Fig. 8 is given by
0 r P (r ) dr
Z

x(r) = P (r ) dr , y(r) = R
. (19)
0 0 r P (r ) dr
hri Tr
f= , (21)
The horizontal coordinate x(r) is the fraction of the pop- hri
ulation with income below r, and the vertical coordinate
y(r) is the fraction of the income this population accounts where hri is the average income of the whole popula-
for. As r changes from 0 to , x and y change from 0 to tion, and the temperature Tr is the average income in
1 and parametrically define a curve in the (x, y) plane. the exponential part of the distribution. Eq. (21) gives
Fig. 8 shows the data points for the Lorenz a well-defined measure of the deviation of the actual in-
curves in 1983 and 2000, as computed by come distribution from the exponential one and, thus,
the IRS (Strudler, Petska, and Petska, 2003). of the fatness of the upper tail. Fig. 9 shows historical
Dragulescu and Yakovenko (2001a) analytically derived evolution of the parameters hri, Tr , and f given by Eq.
the Lorenz curve formula y = x + (1 x) ln(1 x) for (21).12 We observe that Tr has been increasing, essen-
a purely exponential distribution P (r) = c exp(r/Tr ). tially, monotonously (most of this increase is inflation).
This formula is shown by the upper curve in Fig. 8 and In contrast, hri had sharp peaks in 2000 and 2006 coin-
describes the 1983 data reasonably well. However, for ciding with the speculative bubbles in financial markets.
year 2000, it is essential to take into account the fraction The fraction f , which characterizes income inequality,
f of income in the upper tail, which modifies for the has been increasing for the last 20 years and reached
Lorenz formula as follows (Dragulescu and Yakovenko, maxima of 20% in the years 2000 and 2006 with a sharp
2003; Silva and Yakovenko, 2005; Yakovenko and Silva, drop in between. We conclude that the speculative bub-
2005) bles greatly increase the fraction of income going to the
upper tail, but do not change income distribution of the
y = (1 f )[x + (1 x) ln(1 x)] + f (x 1). (20) lower class. When the bubbles inevitably collapse, in-
come inequality reduces.
The last term in Eq. (20) represent the vertical jump of Thus far we discussed the distribution of individual in-
the Lorenz curve at x = 1, where a small percentage of come. An interesting related question is the distribution
population in the upper class accounts for a substantial P2 (r) of family income r = r1 + r2 , where r1 and r2 are
fraction f of the total income. The lower curve in Fig. 8 the incomes of spouses. If the individual incomes are
shows that Eq. (20) fits the 2000 data very well. distributed exponentially P (r) exp(r/Tr ), then
The deviation of the Lorenz curve from the straight
diagonal line in Fig. 8 is a certain measure of income Z r
inequality. Indeed, if everybody had the same income, P2 (r) = dr P (r )P (r r ) = c r exp(r/Tr ), (22)
0
the Lorenz curve would be the diagonal line, because
the fraction of income would be proportional to the frac-
tion of the population. The standard measure of income
inequality is the Gini coefficient 0 G 1, which is 12 A similar plot was constructed by Silva and Yakovenko (2005)
defined as the area between the Lorenz curve and the for an earlier historical dataset.
 17

United States, Bureau of Census data for 1996 PSID data for families, 1999
6% 100

5%

Labor income of another earner, k$

80

4%
Probability

60
3%

2% 40

1%
20

0%
0 10 20 30 40 50 60 70 80 90 100 110 120
Annual family income, k$ 0
0 20 40 60 80 100
Labor income of one earner, k$
FIG. 10 Histogram: Probability distribution of family in-
come for families with two adults (US Census Bureau data).
FIG. 11 Scatter plot of the spouses incomes (r1 , r2 ) and
Solid line: Fit to Eq. (22). From Dragulescu and Yakovenko
(r2 , r1 ) based on the data from the Panel Study of Income
(2001a).
Dynamics (PSID). From Dragulescu and Yakovenko (2003).

United States, Bureau of Census data for 19471994

where c is a normalization constant. Fig. 10 shows 100%
1
that Eq. (22) is in good agreement with the family 90
0.8
income distribution data from the US Census Bureau Cumulative percent of family income
3
80 0.6 Gini coefficient
(Dragulescu and Yakovenko, 2001a). In Eq. (22), we as- 8
sumed that incomes of spouses are uncorrelated. This 70 0.375
simple approximation is indeed supported by the scat- 60
0.2
ter plot of incomes of spouses shown in Fig. 11. Each 0
50 1950 1960 1970 1980 1990
family is represented in this plot by two points (r1 , r2 ) Year
and (r2 , r1 ) for symmetry. We observe that the den- 40
sity of points is approximately constant along the lines
30
of constant family income r1 + r2 = const, which indi-
cates that incomes of spouses are approximately uncor- 20
related. There is no significant clustering of points along 10
the diagonal r1 = r2 , i.e., no strong positive correlation
0
of spouses incomes. 0 10 20 30 40 50 60 70 80 90 100%
Cumulative percent of families
The Gini coefficient for the family income dis-
tribution (22) was analytically calculated by FIG. 12 Main panel: Lorenz plot for family income calculated
Dragulescu and Yakovenko (2001a) as G = 3/8 = 37.5%. from Eq. (22), compared with the US Census data points.
Fig. 12 shows the Lorenz quintiles and the Gini co- Inset: The US Census data points for the Gini coefficient
efficient for 19471994 plotted from the US Census for families, compared with the theoretically calculated value
Bureau data (Dragulescu and Yakovenko, 2001a). The 3/8=37.5%. From Dragulescu and Yakovenko (2001a).
solid line, representing the Lorenz curve calculated from
Eq. (22), is in good agreement with the data. The
systematic deviation for the top 5% of earners results this average, nations or groups of nations may have quite
from the upper tail, which has a less pronounced effect different Gini coefficients that persist over time due
on family income than on individual income, because to specific historical, political, or social circumstances
of income averaging in the family. The Gini coefficient, (Rosser and Rosser, 2004). The Nordic economies, with
shown in the inset of Fig. 12, is close to the calculated their famously redistributive welfare states, have G
value of 37.5%. Notice that income distribution is in the mid-20%, while many of the Latin American
very stable for a long period of time, which was also countries have G over 50%, reflecting entrenched social
recognized by economists (Levy, 1987). Moreover, the patterns inherited from the colonial era.
average G for the developed capitalist countries of North Income distribution has been examined in
America and western Europe, as determined by the econophysics papers for different countries:
World Bank, is also close to the calculated value 37.5% Japan (Aoyama et al., 2003; Ferrero, 2004,
(Dragulescu and Yakovenko, 2003). However, within 2005; Fujiwara et al., 2003; Nirei and Souma,
 18

2007; Souma, 2001, 2002; Souma and Nirei, transitions from one income to another. In the case where
2005), Germany (Clementi and Gallegati, the income r changes by a small amount r over a time
2005a; Clementi, Gallegati, and Kaniadakis, period t, the Markov process can be treated as income
2007), the UK (Clementi and Gallegati, diffusion. Then one can apply the general Fokker-Planck
2005a; Clementi, Gallegati, and Kaniadakis, equation (Lifshitz and Pitaevskii, 1981) to describe evo-
2007; Ferrero, 2004, 2005; lution in time t of the income distribution function P (r, t)
Richmond, Hutzler, Coelho, and Repetowicz, (Silva and Yakovenko, 2005)
2006), Italy (Clementi, Di Matteo, and Gallegati,
h(r)2 i
 
2006; Clementi and Gallegati, 2005b; P (BP ) hri
= AP + , A= , B= .
Clementi, Gallegati, and Kaniadakis, 2007), t r r t 2t
the USA (Clementi and Gallegati, 2005a; (23)
Rawlings et al., 2004), India (Sinha, 2006), The coefficients A and B in Eq. (23) are determined by
Australia (Banerjee, Yakovenko, and Di Matteo, the first and second moments of income changes per unit
2006; Clementi, Di Matteo, and Gallegati, 2006; time. The stationary solution t P = 0 of Eq. (23) obeys
Di Matteo, Aste, and Hyde, 2004), and New Zealand the following equation with the general solution
(Ferrero, 2004, 2005). The distributions are qualitatively  Z r
A(r )

similar to the results presented in this section. The (BP ) c
= AP, P (r) = exp dr .
upper tail follows a power law and comprises a small r B(r) B(r )
fraction of population. To fit the lower part of the (24)
distribution, different papers used the exponential, For the lower part of the distribution, it is reasonable
Gamma, and log-normal distributions. Unfortunately, to assume that r is independent of r, i.e., the changes
income distribution is often reported by statistical in income are independent of income itself. This pro-
agencies for households, so it is difficult to differentiate cess is called the additive diffusion (Silva and Yakovenko,
between one-earner and two-earner income distributions. 2005). In this case, the coefficients in Eq. (23) are the
Some papers used interpolating functions with different constants A0 and B0 . Then Eq. (24) gives the expo-
asymptotic behavior for low and high incomes, such as nential distribution P (r) exp(r/Tr ) with the effec-
the Tsallis function (Ferrero, 2005) and the Kaniadakis tive income temperature Tr = B0 /A0 .13 The coincidence
function (Clementi, Gallegati, and Kaniadakis, 2007). of this result with the Boltzmann-Gibbs exponential law
However, the transition between the lower and upper (1) and (7) is not accidental. Indeed, instead of con-
classes is not smooth for the US data shown in Figs. sidering pairwise interaction between particles, one can
6 and 7, so such functions would not be useful in derive Eq. (1) by considering energy transfers between
this case. The special case is income distribution in a particle and a big reservoir, as long as the transfer
Argentina during the economic crisis, which shows a process is additive and does not involve a Maxwell-
time-dependent bimodal shape with two peaks (Ferrero, demon-like discrimination (Basu and Mohanty, 2008).
2005). Although money and income are different concepts, they
may have similar distributions, because they are gov-
erned by similar mathematical principles. It was shown
explicitly by Cordier, Pareschi, and Toscani (2005);
B. Theoretical models of income distribution Dragulescu and Yakovenko (2000); Slanina (2004) that
the models of pairwise money transfer can be described
Having examined the empirical data on income distri- in a certain limit by the Fokker-Planck equation.
bution, let us now discuss theoretical models. Income On the other hand, for the upper tail of income dis-
ri is the influx of money per unit time to an agent i. tribution, it is reasonable to expect that r r, i.e.,
If the money balance mi is analogous to energy, then income changes are proportional to income itself. This is
the income ri would be analogous to power, which is the known as the proportionality principle of Gibrat (1931),
energy flux per unit time. So, one should conceptually and the process is called the multiplicative diffusion
distinguish between the distributions of money and in- (Silva and Yakovenko, 2005). In this case, A = ar and
come. While money is regularly transferred from one B = br2 , and Eq. (24) gives the power-law distribution
agent to another in pairwise transactions, it is not typi- P (r) 1/r+1 with = 1 + a/b.
cal for agents to trade portions of their income. Never- Generally, the lower-class income comes from wages
theless, indirect transfer of income may occur when one and salaries, where the additive process is appropriate,
employee is promoted and another demoted while the to- whereas the upper-class income comes from bonuses,
tal annual budget is fixed, or when one company gets a investments, and capital gains, calculated in percent-
contract whereas another one loses it, etc. A reasonable ages, where the multiplicative process applies (Milakovic,
approach, which has a long tradition in the economic
literature (Champernowne, 1953; Gibrat, 1931; Kalecki,
1945), is to treat individual income r as a stochastic pro-
cess and study its probability distribution. In general, 13 Notice that a meaningful stationary solution (24) requires that
one can study a Markov process generated by a matrix of A > 0, i.e., hri < 0.
 19

2005). However, the additive and multiplicative pro- 0

10
cesses may coexist. An employee may receive a cost-of-
living raise calculated in percentages (the multiplicative 1
process) and a merit raise calculated in dollars (the addi- 10 2005
tive process). Assuming that these processes are uncorre-
lated, we have A = A0 +ar and B = B0 +br2 = b(r02 +r2 ), 2
10

Cumulative Distribution Function

where r02 = B0 /b. Substituting these expressions into Eq.
(24), we find 3
10
(r0 /Tr ) arctan(r/r0 )
e
P (r) = c . (25)
[1 + (r/r0 )2 ]1+a/2b 4
10

The distribution (25) interpolates between the exponen-

tial law for low r and the power law for high r, be- 5
10
cause either the additive or the multiplicative process
dominates in the corresponding limit. The crossover be-
tween the two regimes takes place at r = r0 , where
1995
the additive and multiplicative contributions to B are
equal. The distribution (25) has three parameters: the
income temperature Tr = A0 /B0 , the Pareto expo-
nent = 1 + a/b, and the crossover income r0 . It is
a minimal model that captures the salient features of
2 1 0 1 2
the empirical income distribution. Eq. (25) was obtained 10 10 10 10 10
Annual Income/T
by Yakovenko (2009), and a more general formula for
correlated additive and multiplicative processes was de-
rived by Fiaschi and Marsili (2009) for a sophisticated FIG. 13 Fits of the IRS data for income distribution using
economic model. Fits of the IRS data using Eq. (25) Eq. (25). Plots for different years are shifted vertically for
are shown in Fig. 13 reproduced from Banerjee (2008). clarity. From Banerjee (2008).
A mathematically similar, but more economically ori-
ented, model was proposed by Nirei and Souma (2007);
Souma and Nirei (2005), where labor income and assets tion (Gibrat, 1931). However, the width of this distri-
accumulation are described by the additive and multi- bution increases in time, so the distribution is not sta-
plicative processes correspondingly. A general stochastic tionary. This was pointed out by Kalecki (1945) a long
process with additive and multiplicative noise was stud- time ago, but the log-normal distribution is still widely
ied numerically by Takayasu et al. (1997), but the sta- used for fitting income distribution, despite this funda-
tionary distribution was not derived analytically. A sim- mental logical flaw in its justification. In the classic pa-
ilar process with discrete time increments was studied by per, Champernowne (1953) showed that a multiplicative
Kesten (1973). Besides economic applications, Eq. (25) process gives a stationary power-law distribution when
may be also useful for general stochastic processes with a boundary condition is imposed at r0 6= 0. Later, this
additive and multiplicative components. result was rediscovered by econophysicists (Levy, 2003;
To verify the multiplicative and additive hypotheses Levy and Solomon, 1996; Sornette and Cont, 1997). In
empirically, it is necessary to have data on income mo- Eq. (25), the exponential distribution of the lower class
bility, i.e., the income changes r of the same people effectively provides such a boundary condition for the
from one year to another. The distribution of income power law of the upper class. Notice also that Eq. (25)
changes P (r|r) conditional on income r is generally not reduces to Eq. (18) in the limit r0 0 with B0 = 0, but
available publicly, although it can be reconstructed by A0 6= 0.
researchers at the tax agencies. Nevertheless, the mul- There are alternative approaches to income distribu-
tiplicative hypothesis for the upper class was quantita- tion in economic literature. One of them, proposed by
tively verified by Aoyama et al. (2003); Fujiwara et al. Lydall (1959), involves social hierarchy. Groups of peo-
(2003) for Japan, where such data for the top taxpayers ple have leaders, which have leaders of the higher order,
are publicly available. and so on. The number of people decreases geometri-
The power-law distribution is meaningful only when cally (exponentially) with the increase in the hierarchical
it is limited to high enough incomes r > r0 . If all level. If individual income increases by a certain factor
incomes r from 0 to follow a purely multiplicative (i.e., multiplicatively) when moving to the next hierar-
process (A0 = 0 and B0 = 0), then one can change chical level, then income distribution follows a power law
to a logarithmic variable x = ln(r/r ) in Eq. (23) and (Lydall, 1959). However, this original argument of Ly-
show that it gives a Gaussian time-dependent distribu- dall can be easily modified to produce the exponential
tion Pt (x) exp(x2 /2 2 t) for x, i.e., the log-normal distribution. If individual income increases by a certain
distribution for r, also known as the Gibrat distribu- amount, i.e., income increases linearly with the hierarchi-
 20

cal level, then income distribution is exponential. The distribution in a society. Although social classes have
latter process seems to be more realistic for moderate been known in political economy since Karl Marx, real-
annual incomes below 100 k$. A similar scenario is the ization that they are described by simple mathematical
Bernoulli trials (Feller, 1966), where individuals have a distributions is quite new. Very interesting work was
constant probability of increasing their income by a fixed done by the computer scientist Ian Wright (2005, 2009),
amount. We see that the deterministic hierarchical mod- who demonstrated emergence of two classes in an agent-
els and the stochastic models of additive and multiplica- based simulation of initially equal agents. This work
tive income mobility represent essentially the same ideas. has been further developed in the upcoming book by
Cottrell, Cockshott, Michaelson, Wright, and Yakovenko
(2009), integrating economics, computer science, and
V. CONCLUSIONS physics.
Econophysics may be also useful for teaching of sta-
The invasion of physicists into economics and fi- tistical physics. If nothing else, it helps to clarify the
nance at the turn of the millennium is a fascinating foundations of statistical physics by applying it to non-
phenomenon. It generated a lively public debate about traditional objects. Practitioners of statistical physics
the role and future perspectives of econophysics, covering know very well that the major fascinating attraction of
both theoretical and empirical issues.14 The econophysi- this field is the enormous breadth of its applications.
cist Joseph McCauley proclaimed that Econophysics
will displace economics in both the universities and
boardrooms, simply because what is taught in economics References
classes doesnt work (Ball, 2006). Although there is
some truth in his arguments (McCauley, 2006), one may Abul-Magd, A. Y., 2002, Wealth distribution in an ancient
consider a less radical scenario. Econophysics may be- Egyptian society, Physical Review E 66, 057104.
come a branch of economics, in the same way as game Anderson, P. W., K. J. Arrow, and D. Pines, 1988, Eds.,
theory, psychological economics, and now agent-based The Economy as an Evolving Complex System (Addison-
modeling became branches of economics. These branches Wesley, Reading).
have their own interests, methods, philosophy, and jour- Angle, J., 1986, The surplus theory of social stratification
and the size distribution of personal wealth, Social Forces
nals. When infusion of new ideas from a different field
65, 293326.
happens, the main contribution often consists not in Angle, J., 1992, The inequality process and the distribution
answering old questions, but in raising new questions. of income to blacks and whites, Journal of Mathematical
Much of the misunderstanding between economists and Sociology 17, 7798.
physicists happens not because they are getting different Angle, J., 1993, Deriving the size distribution of personal
answers, but because they are answering different ques- wealth from the rich get richer, the poor get poorer,
tions. Journal of Mathematical Sociology 18, 2746.
The subject of income and wealth distributions and Angle, J., 1996, How the Gamma Law of income distribution
social inequality was very popular at the turn of another appears invariant under aggregation, Journal of Mathe-
century and is associated with the names of Pareto, matical Sociology 21, 325358.
Angle, J., 2002, The statistical signature of pervasive com-
Lorenz, Gini, Gibrat, and Champernowne, among others.
petition on wage and salary incomes, Journal of Mathe-
Following the work by Pareto, attention of researchers matical Sociology 26, 217270.
was primarily focused on the power laws. However, Angle, J., 2006, The Inequality Process as a wealth maxi-
when physicists took a fresh look at the empirical data, mizing process, Physica A 367, 388414.
they found a different, exponential law for the lower part Anglin, P., 2005, Econophysics of wealth distributions:
of the distribution. Demonstration of the ubiquitous a comment, in Chatterjee, Yarlagadda, and Chakrabarti
nature of the exponential distribution for money, wealth, (2005), pp. 229238.
and income is one of the new contributions produced by Ao, P., 2007, Boltzmann-Gibbs distribution of fortune and
econophysics.15 The motivation, of course, came from broken time reversible symmetry in econodynamics, Com-
the Boltzmann-Gibbs distribution in physics. Further munications in Nonlinear Science and Numerical Simula-
studies revealed a more detailed picture of the two-class tion 12, 619626.
Aoyama, H., W. Souma, and Y. Fujiwara, 2003, Growth and
fluctuations of personal and companys income, Physica A
324, 352358.
Atkinson, A. B., and F. Bourguignon, 2000, Eds., Handbook
14 See, for example, Anglin (2005); Ball (2006); Carbone et al. of Income Distribution (Elsevier, Amsterdam).
(2007); Gallegati, Keen, Lux, and Ormerod (2006); Atkinson, A. B., and T. Piketty, 2007, Eds., Top Incomes over
Keen (2008); Lux (2005, 2009); McCauley (2006);
the Twentieth Century (Oxford University Press, Oxford).
Richmond, Chakrabarti, Chatterjee, and Angle (2006); Rosser
(2006, 2008b); Stauffer (2004); Yakovenko (2009). Ausloos, M., and A. Pekalski, 2007, Model of wealth and
15 The exponential distribution is also ubiquitous in the probabil- goods dynamics in a closed market, Physica A 373, 560
ity distributions of financial returns (Kleinert and Chen, 2007; 568.
McCauley and Gunaratne, 2003; Silva, Prange, and Yakovenko, Bak P., S. F. Nrrelykke, and M. Shubik, 1999, Dynamics of
2004) and the growth rates of firms. money, Physical Review E 60, 25282532.
 21

Ball, P., 2002, The physical modelling of society: a historical Chatterjee, A., and B. K. Chakrabarti, 2006, Kinetic market
perspective, Physica A 314, 114. models with single commodity having price fluctuations,
Ball, P., 2004, Critical Mass: How One Thing Leads to An- The European Physical Journal B 54, 399404.
other (Farrar, Straus and Giroux, New York). Chatterjee, A., and B. K. Chakrabarti, 2007, Kinetic ex-
Ball, P., 2006, Econophysics: Culture Crash, Nature 441, change models for income and wealth distributions, The
686688. European Physical Journal B 60, 135149.
Banerjee, A., 2008, Ph.D. thesis (Department of Physics, Uni- Chatterjee, A., B. K. Chakrabarti, and S. S. Manna, 2004,
versity of Maryland). Pareto law in a kinetic model of market with random sav-
Banerjee, A., V. M. Yakovenko, and T. Di Matteo, 2006, A ing propensity, Physica A 335, 155-163.
study of the personal income distribution in Australia, Chatterjee, A., B. K. Chakrabarti, and R. B. Stinchcombe,
Physica A 370, 5459. 2005, Master equation for a kinetic model of a trading
Basu, U., and P. K. Mohanty, 2008, Modeling wealth distri- market and its analytic solution, Physical Review E 72,
bution in growing markets, The European Physical Jour- 026126.
nal B 65, 585589. Chatterjee, A., S. Yarlagadda, and B. K. Chakrabarti, 2005,
Bennati, E., 1988, Un metodo di simulazione statistica Eds., Econophysics of Wealth Distributions (Springer, Mi-
per lanalisi della distribuzione del reddito, Rivista Inter- lan).
nazionale di Scienze Economiche e Commerciali 35, 735 Chen, J., and V. M. Yakovenko, 2007, Com-
756. puter animation videos of money-transfer models,
Bennati, E., 1993, Il metodo di Montecarlo nellanalisi eco- http://www2.physics.umd.edu/~ yakovenk/econophysics/animation.
nomica, Rassegna di Lavori dellISCO (Istituto Nazionale Clementi, F., T. Di Matteo, and M. Gallegati, 2006, The
per lo Studio della Congiuntura), Anno X, No. 4, pp. 3179. power-law tail exponent of income distributions, Physica
Bhattacharyya, P., A. Chatterjee, and B. K. Chakrabarti, A 370, 4953.
2007, A common mode of origin of power laws in models Clementi, F., and M. Gallegati, 2005a, Paretos
of market and earthquake, Physica A 381, 377382. law of income distribution: evidence for Ger-
Blume, L. E., 1993, The statistical mechanics of strategic many, the United Kingdom, the United States, in
interaction, Games and Economic Behavior 5, 387424. Chatterjee, Yarlagadda, and Chakrabarti (2005), pp.
Boltzmann, L., 1905, Populare Schriften (Barth, Leipzig), p. 314.
S.360f. Clementi, F., and M. Gallegati, 2005b, Power law tails in
Boltzmann, L., 2006, Ludwig Boltzmann 18441906 (Aus- the Italian personal income distribution, Physica A 350,
trian Central Library for Physics, Vienna), ISBN 3-900490- 427438.
11-2, p. 113. Clementi, F., M. Gallegati, and G. Kaniadakis, 2007, -
Bouchaud, J.-P., and M. Mezard, 2000, Wealth condensation generalized statistics in personal income distribution, The
in a simple model of economy, Physica A 282, 536545. European Physical Journal B 57, 187193.
Braun, D., 2001, Assets and liabilities are the momentum of Cockshott, P., and A. Cottrell, 2008, Probabilistic
particles and antiparticles displayed in Feynman-graphs, political economy and endogenous money, talk pre-
Physica A 290, 491500. sented at the conference Probabilistic Political Econ-
Braun, D., 2006, Nonequilibrium thermodynamics of wealth omy, Kingston University (UK), July 2008, available at
condensation, Physica A 369, 714722. http://www.dcs.gla.ac.uk/publications/PAPERS/8935/probpolecon
Burda, Z., D. Johnston, J. Jurkiewicz, M. Kaminski, M. A. Coelho, R., Z. Neda, J. J. Ramasco, and M. A. Santos, 2005,
Nowak, G. Papp, and I. Zahed, 2002, Wealth condensation A family-network model for wealth distribution in soci-
in Pareto macroeconomies, Physical Review E 65, 026102. eties, Physica A 353, 515528.
Carbone, A., G. Kaniadakis, and A. M. Scarfone, 2007, Collier, M. R., 2004, Are magnetospheric suprathermal par-
Where do we stand on econophysics? Physica A 382, ticle distributions ( functions) inconsistent with maximum
xixiv. entropy considerations? Advances in Space Research 33,
Chakrabarti, B. K., 2005, Econophys-Kolkata: a short 21082112.
story, in Chatterjee, Yarlagadda, and Chakrabarti (2005), Cordier, S., L. Pareschi, and G. Toscani, 2005, On a kinetic
pp. 225228. model for a simple market economy, Journal of Statistical
Chakrabarti, B. K., A. Chakraborti, and A. Chatterjee, 2006, Physics 120, 253277.
Eds., Econophysics and Sociophysics: Trends and Perspec- Cottrell, A. F., P. Cockshott, G. J. Michaelson, I. P. Wright,
tives (Wiley-VCH, Berlin). and V. M. Yakovenko, 2009, Classical Econophysics (Rout-
Chakraborti A., and B. K. Chakrabarti, 2000, Statistical ledge, Oxford).
mechanics of money: how saving propensity affects its dis- Das, A., and S. Yarlagadda, 2005, An analytic treatment of
tribution, The European Physical Journal B 17, 167170. the Gibbs-Pareto behavior in wealth distribution, Physica
Chakraborti, A., S. Pradhan, and B. K. Chakrabarti, 2001, A 353, 529538.
A self-organising model of market with single commodity, Defilla, S., 2007, A natural value unit Econophysics as
Physica A 297, 253259. arbiter between finance and economics, Physica A 382,
Champernowne, D. G., 1953, A model of income distribu- 4251.
tion, The Economic Journal 63, 318351. Desai, M. I., G. M. Mason, J. R. Dwyer, J. E. Mazur, R. E.
Champernowne, D. G., and F. A. Cowell, 1998, Economic Gold, S. M. Krimigis, C. W. Smith, and R. M. Skoug, 2003,
Inequality and Income Distribution (Cambridge University Evidence for a suprathermal seed population of heavy ions
Press, Cambridge). accelerated by interplanetary shocks near 1 AU, The As-
Chatterjee, A., 2009 Kinetic models for wealth exchange on trophysical Journal 588, 11491162.
directed networks, The European Physical Journal B 67, Diaz-Gimenez, J., V. Quadrini, and J. V. Rios-Rull, 1997,
593598. Dimensions of inequality: Facts on the U.S. distributions
 22

of earnings, income, and wealth, Federal Reserve Bank of Monetary Economics (North-Holland, Amsterdam), Vol. 1
Minneapolis Quarterly Review 21 (2), 3-21. and Vol. 2.
Di Matteo, T., T. Aste, and S. T. Hyde, 2004, Exchanges Fujiwara, Y., W. Souma, H. Aoyama, T. Kaizoji, and M.
in complex networks: income and wealth distributions, Aoki, 2003, Growth and fluctuations of personal income,
in The Physics of Complex Systems (New Advances and Physica A 321, 598604.
Perspectives), edited by F. Mallamace and H. E. Stanley Gabaix, X., P. Gopikrishnan, V. Plerou, and H. E. Stanley,
(IOS Press, Amsterdam), p. 435. 2006, Institutional investors and stock market volatility,
Dragulescu, A. A., and V. M. Yakovenko, 2000, Statistical The Quarterly Journal of Economics 121, 461504.
mechanics of money, The European Physical Journal B Galam, S., Y. Gefen, Y. Shapir, 1982, Sociophysics: a
17, 723729. new approach of sociological collective behaviour. I. Mean-
Dragulescu, A. A., and V. M. Yakovenko, 2001a, Evidence behaviour description of a strike, Journal of Mathematical
for the exponential distribution of income in the USA, The Sociology 9, 113.
European Physical Journal B 20, 585589. Galam, S., 2004, Sociophysics: a personal testimony, Phys-
Dragulescu, A. A., and V. M. Yakovenko, 2001b, Exponen- ica A 336, 4955.
tial and power-law probability distributions of wealth and Gallegati, M., S. Keen, T. Lux, and P. Ormerod, 2006, Wor-
income in the United Kingdom and the United States, rying trends in econophysics, Physica A 370, 16.
Physica A 299, 213221. Garibaldi, U., E. Scalas, and P. Viarengo, 2007, Statistical
Dragulescu, A. A., and V. M. Yakovenko, 2003, Statistical equilibrium in simple exchange games II: The redistribution
mechanics of money, income, and wealth: a short survey, game, The European Physical Journal B 60, 241246.
in Modeling of Complex Systems: Seventh Granada Lec- Gibrat, R., 1931, Les Inegalites Economiques (Sirely, Paris).
tures, edited by P. L. Garrido and J. Marro, American In- Gillispie, C. C., 1963, Intellectual factors in the background
stitute of Physics (AIP) Conference Proceedings 661, pp. of analysis by probabilities, in Scientific Change, edited
180183. by A. C. Crombie (Heinemann, London), pp. 431453.
During, B., D. Matthes, and G. Toscani, 2008, Kinetic equa- Gupta, A. K., 2006, Money exchange model and a general
tions modelling wealth redistribution: A comparison of ap- outlook, Physica A 359, 634640.
proaches, Physical Review E 78, 056103. Gupta, A. K., 2008, Relaxation in the wealth exchange mod-
During, B., and G. Toscani, 2007, Hydrodynamics from ki- els, Physica A 387, 68196824.
netic models of conservative economies, Physica A 384, Hasegawa, A., K. Mima, and M. Duong-van, 1985, Plasma
493506. distribution function in a superthermal radiation field,
Durlauf, S. N., 1997, Statistical mechanics approaches to Physical Review Letters 54, 26082610.
socioeconomic behavior, in The Economy as a Complex Hayes, B., 2002, Follow the money, American Scientist 90,
Evolving System II, edited by W. B. Arthur, S. N. Durlauf, 400405.
and D. A. Lane (Addison-Wesley, Redwood City), pp. 81 Hegyi, G., Z. Neda, and M. A. Santos, 2007, Wealth distribu-
104. tion and Paretos law in the Hungarian medieval society,
Engels, F., 1972, The Origin of the Family, Private Property Physica A 380, 271277.
and the State, in the Light of the Researches of Lewis H. Her Majesty Revenue and Customs, 2003,
Morgan (International Publishers, New York). Distribution of Personal Wealth,
Farjoun, E., and M. Machover, 1983, Laws of Chaos: A Prob- http://www.hmrc.gov.uk/stats/personal_wealth/wealth_oct03.pdf
abilistic Approach to Political Economy (Verso, London). Hu, M. B., W. X. Wang, R. Jiang, Q. S. Wu, B. H. Wang, and
Farmer, J. D., M. Shubik, and E. Smith, 2005, Is economics Y. H. Wu, 2006, A unified framework for the Pareto law
the next physical science? Physics Today 58 (9), 3742. and Matthew effect using scale-free networks, The Euro-
Feller, W., 1966, An Introduction to Probability Theory and pean Physical Journal B 53, 273277.
Its Applications (John Wiley, New York), Vol. 2, p. 10. Hu, M. B., R. Jiang, Q. S. Wu, and Y. H. Wu, 2007, Sim-
Ferrero, J. C., 2004, The statistical distribution of money ulating the wealth distribution with a Richest-Following
and the rate of money transference, Physica A 341, 575 strategy on scale-free network, Physica A 381, 467472.
585. Huang, D. W., 2004, Wealth accumulation with random re-
Ferrero, J. C., 2005, The monomodal, polymodal, equi- distribution, Physical Review E 69, 057103.
librium and nonequilibrium distribution of money, in Iglesias, J. R., S. Goncalves, S. Pianegonda, J. L. Vega, and
Chatterjee, Yarlagadda, and Chakrabarti (2005), pp. 159 G. Abramson, 2003, Wealth redistribution in our small
167. world, Physica A 327, 1217.
Fiaschi, D., and M. Marsili, 2009, Distribu- Ispolatov S., P. L. Krapivsky, and S. Redner, 1998, Wealth
tion of Wealth and Incomplete Markets: The- distributions in asset exchange models, The European
ory and Empirical Evidence, working paper, Physical Journal B 2, 267276.
Kakwani, N., 1980, Income Inequality and Poverty (Oxford
http://www.dse.ec.unipi.it/persone/docenti/fiaschi/Lavori/distributionWealthIncompleteMarkets.pdf.
Fischer, R., and D. Braun, 2003a, Transfer potentials shape University Press, Oxford).
and equilibrate monetary systems, Physica A 321, 605 Kalecki, M., 1945, On the Gibrat distribution, Economet-
618. rica 13, 161170.
Fischer, R., and D. Braun, 2003b, Nontrivial bookkeeping: Keen, S., 1995, Finance and economic breakdown: Modeling
a mechanical perspective, Physica A 324, 266271. Minkys financial instability hypothesis, Journal of Post
Foley, D. K., 1994, A statistical equilibrium theory of mar- Keynesian Economics 17, 607635.
kets, Journal of Economic Theory 62, 321345. Keen, S., 2000, The nonlinear economics of debt deflation,
Follmer, H., 1974, Random economies with many interacting in Commerce, Complexity, and Evolution, edited by W. A.
agents, Journal of Mathematical Economics 1, 5162. Barnett et al. (Cambridge University Press, Cambridge),
Friedman, B. M., and F. H. Hahn, 1990, Eds., Handbook of pp. 87117.
 23

Keen, S., 2008, Conservation laws in econophysics, McConnell, C. R., and S. L. Brue, 1996, Economics: Princi-
and the non-conservation of money, talk presented ples, Problems, and Policies (McGraw-Hill, New York).
at the conference Probabilistic Political Economy, Milakovic, M., 2005, Do we all face the same constraints? in
Kingston University (UK), July 2008, available at Chatterjee, Yarlagadda, and Chakrabarti (2005), pp. 184
191.
http://www.debtdeflation.com/blogs/wp-content/uploads/2007/09/KeenNonConservationMoney.pdf.
Kesten, H., 1973, Random difference equations and renewal Mimkes, J., 2000, Society as a many-particle system, Jour-
theory for products of random matrices, Acta Mathemat- nal of Thermal Analysis and Calorimetry 60, 10551069.
ica 131, 207248. Mimkes, J., and G. Willis, 2005, Lagrange principle of wealth
Kiyotaki, N., and R. Wright, 1993, A search-theoretic ap- distribution, in Chatterjee, Yarlagadda, and Chakrabarti
proach to monetary economics, The American Economic (2005), pp. 6169.
Review 83, 6377. Mirowski, P., 1989, More Heat than Light: Economics as So-
Klass, O. S., O. Biham, M. Levy, O. Malcai, and S. Solomon, cial Physics, Physics as Natures Economics (Cambridge
2007, The Forbes 400, the Pareto power-law and efficient University Press, Cambridge).
markets, The European Physical Journal B 55, 143147. Mohanty, P. K., 2006, Generic features of the wealth distri-
Kleinert, H., and X. J. Chen, 2007, Boltzmann distribution bution in ideal-gas-like markets, Physical Review E 74,
and market temperature, Physica A 383, 513518. 011117.
Levy, F., 1987, Changes in the distribution of American fam- Molico, M., 2006, The distribution of money and prices in
ily incomes, 1947 to 1984, Science 236, 923927. search equilibrium, International Economic Review 47,
Levy, M., 2003, Are rich people smarter? Journal of Eco- 701722.
nomic Theory 110, 4264. Montroll, E. W., and W. W. Badger, 1974, Introduction to
Levy, M., and H. Levy, 2003, Investment talent and the Quantitative Aspects of Social Phenomena, (Gordon and
Pareto wealth distribution, Review of Economics and Breach, New York).
Statistics 85, 709725. Nirei, M., and W. Souma, 2007, A two factor model of
Levy, M., and S. Solomon, 1996, Power laws are logarithmic income distribution dynamics, Review of Income and
Boltzmann laws, International Journal of Modern Physics Wealth 53, 440459.
C 7, 595751. O Brien, Y.-Y. C., 2007, Reserve requirement sys-
Lifshitz, E. M., and L. P. Pitaevskii, 1981, Physical Kinetics tems in OECD countries, Finance and Eco-
(Pergamon, Oxford). nomics Discussion Series, Divisions of Research
Lopez-Ruiz, R., J. Sanudo, and X. Calbet, 2008, Geometrical & Statistics and Monetary Affairs of the Fed-
derivation of the Boltzmann factor, American Journal of eral Reserve Board, Washington, D.C., 2007-54,
Physics 76, 780781. http://www.federalreserve.gov/pubs/feds/2007/200754/200754abs
Lux, T., 2005, Emergent statistical wealth distributions in Pareto, V., 1897, Cours dEconomie Politique (LUniversite
simple monetary exchange models: a critical review, in de Lausanne).
Chatterjee, Yarlagadda, and Chakrabarti (2005), pp. 51 Patriarca, M., A. Chakraborti, and G. Germano, 2006, In-
60. fluence of saving propensity on the power-law tail of the
Lux, T., 2009, Applications of statistical physics in finance wealth distribution, Physica A 369, 723736.
and economics, in Handbook of Research on Complexity, Patriarca, M., A. Chakraborti, E. Heinsalu, and G. Germano,
edited by J. B. Rosser (Edward Elgar, Cheltenham, UK 2007, Relaxation in statistical many-agent economy mod-
and Northampton, MA). els, The European Physical Journal B 57, 219224.
Lux, T., and D. Sornette, 2002, On rational bubbles and fat Patriarca, M., A. Chakraborti, and K. Kaski, 2004a, Gibbs
tails, Journal of Money, Credit, and Banking 34, 589610. versus non-Gibbs distributions in money dynamics, Phys-
Lydall, H. F., 1959, The distribution of employment in- ica A 340, 334339.
comes, Econometrica 27, 110115. Patriarca, M., A. Chakraborti, and K. Kaski, 2004b, Statis-
Majorana, E., 1942, Il valore delle leggi statistiche nella fisica tical model with a standard Gamma distribution, Physical
e nelle scienze sociali, Scientia 36, 5866 [English trans- Review E 70, 016104.
lation by R. N. Mantegna, 2006, in Ettore Majorana Sci- Patriarca, M., A. Chakraborti, K. Kaski, and G. Ger-
entific Papers, edited by G. F. Bassani (Springer, Berlin), mano, 2005, Kinetic theory models for the distribution
pp. 250260]. of wealth: Power law from overlap of exponentials, in
Mandelbrot, B., 1960, The Pareto-Levy law and the distribu- Chatterjee, Yarlagadda, and Chakrabarti (2005), pp. 93
tion of income, International Economic Review 1, 79106. 110.
Mantegna, R. N., and H. E. Stanley, 1999, An Introduction Pareschi, L., and G. Toscani, 2006, Self-similarity and power-
to Econophysics: Correlations and Complexity in Finance like tails in nonconservative kinetic models, Journal of Sta-
(Cambridge University Press, Cambridge). tistical Physics 124, 747779.
Matthes, D., and G. Toscani, 2008, On steady distributions Pianegonda, S., J. R. Iglesias, G. Abramson, and J. L. Vega,
of kinetic models of conservative economies, Journal of 2003, Wealth redistribution with conservative exchanges,
Statistical Physics 130, 1087-1117. Physica A 322, 667675.
McCauley, J. L., 2006, Response to Worrying Trends in Piketty, T., and E. Saez, 2003, Income inequality in the
Econophysics, Physica A 371, 601609. United States, 1913-1998, The Quarterly Journal of Eco-
McCauley, J. L., 2008, Nonstationarity of efficient finance nomics 118, 139.
markets: FX market evolution from stability to instability, Raberto, M., S. Cincotti, S. M. Focardi, and M. Marchesi,
International Review of Financial Analysis 17, 820837. 2003, Traders long-run wealth in an artificial financial
McCauley, J. L., and G. H. Gunaratne, 2003, An empirical market, Computational Economics 22, 255272.
model of volatility of returns and option pricing, Physica Rawlings, P. K., D. Reguera, and H. Reiss, 2004, Entropic
A 329, 178-198.
 24

basis of the Pareto law, Physica A 343, 643652. Sinha, S., 2006, Evidence for power-law tail of the wealth
Repetowicz, P., S. Hutzler, and P. Richmond, 2005, Dynam- distribution in India, Physica A 359, 555562.
ics of money and income distributions, Physica A 356, Slanina, F., 2004 Inelastically scattering particles and wealth
641654. distribution in an open economy, Physical Review E 69,
Richmond, P., B. K. Chakrabarti, A. Chatterjee, and J. 046102.
Angle, 2006, Comments on Worrying Trends in Econo- Smith, E., and D. K. Foley, 2008, Classical thermodynam-
physics: income distribution models, in Econophysics of ics and economic general equilibrium theory, Journal of
Stock and Other Markets, edited by A. Chatterjee and B. Economic Dynamics and Control 32, 765.
K. Chakrabarti (Springer, Milan), pp. 244253. Soddy, F., 1933, Wealth, Virtual Wealth and Debt, 2nd ed.
Richmond, P., S. Hutzler, R. Coelho, and P. Repetow- (Dutton, New York).
icz, 2006, A review of empirical studies and Solomon, S., and P. Richmond, 2001, Power laws of wealth,
models of income distributions in society, in market order volumes and market returns, Physica A 299,
Chakrabarti, Chakraborti, and Chatterjee (2006). 188197.
Richmond, P., P. Repetowicz, S. Hutzler, and R. Coelho, Solomon. S., and P. Richmond, 2002, Stable power laws
2006, Comments on recent studies of the dynamics and in variable economies; Lotka-Volterra implies Pareto-Zipf,
distribution of money, Physica A, 370, 4348. The European Physical Journal B 27, 257261.
Rosser, J. B. Jr., 2006, The nature and future of econo- Sornette, D., and R. Cont, 1997, Convergent multiplicative
physics, in Econophysics of Stock and Other Markets, processes repelled from zero: power laws and truncated
edited by A. Chatterjee, and B. K. Chakrabarti (Springer, power laws, Journal de Physique I (France) 7, 431444.
Milan), pp. 225234. Souma, W., 2001, Universal structure of the personal income
Rosser, J. B. Jr., 2008a, Econophysics, in New Palgrave distribution, Fractals 9, 463470.
Dictionary of Economics, edited by L. E. Blume and S. N. Souma, W., 2002, Physics of personal income, in Empirical
Durlauf (Macmillan, London), 2nd edition. Science of Financial Fluctuations: the Advent of Econo-
Rosser, J. B. Jr., 2008b, Debating the role of econophysics, physics, edited by H. Takayasu (Springer, Tokyo), pp. 343
Nonlinear Dynamics, Psychology, and Life Sciences 12, 352.
311323. Souma, W., and M. Nirei, 2005, Empiri-
Rosser, J. B. Jr., and M. V. Rosser, 2004, Comparative Eco- cal study and model of personal income, in
nomics in a Transforming World Economy (MIT Press, Chatterjee, Yarlagadda, and Chakrabarti (2005), pp.
Cambridge), 2nd edition. 3442.
Saif, M. A., and P. M. Gade, 2007, Emergence of power-law Stanley, H. E., et al., 1996, Anomalous fluctuations in the
in a market with mixed models, Physica A 384, 448456. dynamics of complex systems: from DNA and physiology
Samanidou, E., E. Zschischang, D. Stauffer, and T. Lux, to econophysics, Physica A 224, 302321.
2007, Agent-based models of financial markets, Reports Stauffer, D., 2004, Introduction to statistical physics outside
on Progress in Physics 70, 409450. physics, Physica A 336, 15.
Scafetta, N., S. Picozzi, and B. J. West, 2004a, An out-of- Strudler, M., T. Petska, and R. Petska, 2003, An analysis
equilibrium model of the distributions of wealth, Quanti- of the distribution of individual income and taxes, 1979
tative Finance 4, 353364. 2001, The Internal Revenue Service, Washington D.C.,
Scafetta, N., S. Picozzi, and B. J. West, 2004b, A trade- http://www.irs.gov/pub/irs-soi/03strudl.pdf.
investment model for distribution of wealth, Physica D Takayasu, H., A. H. Sato, and M. Takayasu, 1997, Stable in-
193, 338352. finite variance fluctuations in randomly amplified Langevin
Scafetta, N., and B. J. West, 2007, Probability distributions systems, Physical Review Letters 79, 966969.
in conservative energy exchange models of multiple inter- Wannier, G. H., 1987, Statistical Physics (Dover, New York).
acting agents, Journal of Physics Condensed Matter 19, Weidlich, W., 2000, Sociodynamics: A Systematic Approach
065138. to Mathematical Modeling in the Social Sciences (Harwood
Scalas, E., U. Garibaldi, and S. Donadio, 2006, Statistical Academic, Amsterdam).
equilibrium in simple exchange games I: Methods of solu- Wright, I., 2005, The social architecture of capitalism,
tion and application to the Bennati-Dragulescu-Yakovenko Physica A 346, 589620.
(BDY) game, The European Physical Journal B 53, 267 Wright, I., 2007, Computer simulations of sta-
272. tistical mechanics of money in Mathematica,
Schweitzer, F., 2003, Brownian Agents and Active Particles: http://demonstrations.wolfram.com/StatisticalMechanicsOfMoney
Collective Dynamics in the Natural and Social Sciences Wright, I., 2009, Implicit microfoundations for
(Springer, Berlin). macroeconomics, Economics (e-journal) 3, 2009-19,
Shubik, M., 1999, The Theory of Money and Financial Insti- http://www.economics-ejournal.org/economics/journalarticles/2
tutions (The MIT Press, Cambridge), Vol. 2, p. 192. Wyart, M., and J.-P. Bouchaud, 2007, Self-referential be-
Silva, A. C., R. E. Prange, and V. M. Yakovenko, 2004, Ex- haviour, overreaction and conventions in financial mar-
ponential distribution of financial returns at mesoscopic kets, Journal of Economic Behavior & Organization 63,
time lags: a new stylized fact, Physica A 344, 227235. 124.
Silva, A. C., and V. M. Yakovenko, 2005, Temporal evolution Xi, N., N. Ding, and Y. Wang, 2005, How required reserve
of the thermal and superthermal income classes in the ratio affects distribution and velocity of money, Physica
USA during 1983-2001, Europhysics Letters 69, 304310. A 357, 543555.
Silver, J., E. Slud, and K. Takamoto, 2002, Statistical equi- Yakovenko, V. M., 2009, Econophysics, statistical mechanics
librium wealth distributions in an exchange economy with approach to, in Encyclopedia of Complexity and System
stochastic preferences, Journal of Economic Theory 106, Science, edited by R. A. Meyers (Springer, Berlin).
417435. Yakovenko, V. M., and A. C. Silva, 2005, Two-
 25

class structure of income distribution in the Chatterjee, Yarlagadda, and Chakrabarti (2005), pp.
USA: Exponential bulk and power-law tail, in 1523.