1: \documentclass[12pt,twoside]{article}
2: \usepackage[dvips]{graphics,color}
3: \usepackage{array}
4: \topmargin -15mm
5: \textheight 235mm
6: \textwidth 160mm
7: \oddsidemargin 0mm
8: \evensidemargin 0mm
9: \renewcommand{\baselinestretch}{1.5}
10: \newcommand{\ow}{\overline{W}}
11: \newcommand{\oz}{\overline{Z}}
12: \newcommand{\own}{\overline{w}}
13: \newcommand{\ozn}{\overline{z}}
14: \begin{document}
15: \rightline{KUCP0189}
16:
17: \vspace{10mm}
18: \centerline{\Large\bfseries Small-World Effects in Wealth Distribution}
19:
20: \vspace{20mm}
21: \centerline{\large Wataru Souma
22: \footnote{e-mail: souma@isd.atr.co.jp}}
23:
24: \centerline{Information Sciences Division, ATR International,
25: Kyoto 619-0288, Japan}
26:
27: \centerline{\large Yoshi Fujiwara
28: \footnote{e-mail: yfujiwar@crl.go.jp}}
29:
30: \centerline{Keihanna Research Center, Communications Research
31: Laboratory, Kyoto 619-0289, Japan}
32:
33: \centerline{\large Hideaki Aoyama
34: \footnote{e-mail: aoyama@phys.h.kyoto-u.ac.jp}}
35:
36: \centerline{Faculty of Integrated Human Studies, Kyoto University,
37: Kyoto 606-8501, Japan}
38:
39:
40: \vspace{40mm}
41: \begin{abstract}
42: We construct a model of wealth distribution, based on
43: an interactive multiplicative stochastic process
44: on static complex networks.
45: Through numerical simulations we show
46: that a decrease in the number of links discourages
47: equality in wealth distribution, while the rewiring of links in small-world
48: networks encourages it.
49: Inequal distributions obey log-normal distributions, which
50: are produced by wealth clustering.
51: The rewiring of links breaks the wealth clustering and
52: makes wealth obey the mean field type (power law) distributions.
53: A mechanism that explains the appearance of log-normal distributions
54: with a power law tail is proposed.
55: \end{abstract}
56:
57: \newpage
58: With the advance of the theory of complex networks \cite{WS},
59: it has become clear that many social networks
60: can be understood as small-world networks \cite{NSAB}.
61: On the other hand, recent high-precision studies
62: of economic phenomena, helped by the availability of
63: high-quality data in digital format \cite{souma}, is starting
64: to reveal the basic characterics of wealth distribution,
65: which form a basis and at the same time a final product of
66: economic activities.
67: In view of these developments, we propose a stochastic model of
68: wealth distribution built on a complex network
69: and present the results of our numerical simulation in this paper.
70:
71: Several models for wealth distribution
72: have been proposed
73: with some reality and success \cite{IKR,solomon1,BM,BMetc}.
74: They, however, lack one feature we consider important:
75: In some models, each agent (economic body) goes through a stochastic
76: process of increasing and decreasing wealth, quite
77: independently from other agents.
78: In some others, each agent interacts with either randomly-selected
79: agents or neighboring agents on a fixed lattice.
80: Apparently, such a selection of economic partnership/competition
81: is far from reality, as is evident from complex network studies.
82: This is one reason we propose our model, which features both
83: the stochastic nature of economic activities and underlying
84: network structures that are expected to be close to
85: those of the relevant social activities.
86: We also note that studies of economic phenomena that build
87: upon the underlying complex networks are novel.
88:
89: In considering the mechanisms behind wealth distribution,
90: it is important to include at least two basic natures of the wealth.
91: One is a random change of the value of wealth and
92: the other is the liquidity of it.
93: As a minimal model that contains these features
94: we chose an interactive multiplicative stochastic process \cite{BM},
95: which is defined by the following evolution equation:
96: \begin{equation}
97: \label{eq:master}
98: W_i(t+1)=a_i(t)W_i(t)+\sigma^2W_i(t)+\sum_{j(\ne i)}J_{ij}(t)W_j(t)
99: -\sum_{j(\ne i)}J_{ji}(t)W_i(t),
100: \end{equation}
101: where $W_i(t)$ is the wealth of the $i$-th agent at the time $t$,
102: $a_i(t)$ is a Gaussian random variable with mean $m$
103: and variance $2\sigma^2$,
104: which describes the spontaneous increase or decrease of the wealth.
105: The rest of the terms in the r.h.s. of Eq.~(\ref{eq:master})
106: describe transactions on business networks.
107: We note that in Ref.~\cite{BM}
108: the process corresponding to Eq.~(\ref{eq:master})
109: is described by a stochastic differential equation in the Stratonovich
110: sense, which generates the second term proportional
111: to $\sigma^2$ when discretized.
112: In addition, although naive discretization brings in a
113: time interval $\Delta t$ in Eq.(\ref{eq:master}),
114: it has been absorbed by the rescaling of the parameters.
115:
116: The transaction matrix $J_{ij}$ is chosen based on the underlying
117: network structure:
118: If the $i$-th agent is directly connected to the $j$-th agent by a link,
119: we take $J_{ij}=J/Z_i$,
120: where $Z_i$ is the number of agents connected to the $i$-th
121: agent (``neighbors"). Otherwise, $J_{ij}=0$.
122: The parameter $J$ is a positive number, which
123: represents the ratio of transactions and the wealth.
124: Thus, the third term in the r.h.s. of Eq.~(\ref{eq:master}) describes
125: the incoming wealth, and the fourth term describes the outgoing wealth.
126: Under this selection of transaction matrix, Eq.~(\ref{eq:master})
127: can be written as follows:
128: \begin{equation}
129: \label{eq:disc}
130: W_i(t+1)=\left\{a_i(t)+\sigma^2-J\right\}W_i(t)+J\ow_i(t),
131: \end{equation}
132: where $\ow_i(t)$ is the mean wealth of the neighbors of the $i$-th agent.
133:
134: We consider one-dimensional and static network of $N$-agents
135: with periodic boundary conditions \cite{WS,NSAB},
136: which are determined by two parameters, i.e., a rewiring
137: probability $p$ and a mean number of links per agent $\oz$.
138: (We will also use a notation $z\equiv \oz/N$.)
139: The rewiring probability $p$ is defined by
140: $p\equiv(\textrm{the number of rewired links})/
141: (\textrm{the total number of links})\in[0,1]$.
142: %$p\equiv$(the number of rewired links)/
143: %(the total number of links)$\in[0,1]$.
144: At $p=0$ we have a regular network with $Z_i=\oz$, in which the $i$-th agent is
145: connected only with $\{i-\oz/2, \cdots,i-1,i+1,i+\oz/2\}$-th agents.
146: For $p\ne0$, links are rewired to conserve the total number of links.
147: At $p=1$ we obtain completely random networks.
148:
149: In the mean field (MF) case, i.e., $z\simeq1$,
150: the stationary probability density function $p(w)$
151: is known to be the following in the large $N$ limit \cite{BM},
152: \begin{equation}
153: \label{eq:mfsolution}
154: p(w)=\frac{(\mu-1)^\mu}{\Gamma[\mu]}\frac{\exp\{-(\mu-1)/w\}}{w^{1+\mu}},
155: \end{equation}
156: where $w_i(t)$ is the wealth normalized by the mean wealth
157: $\ow(t)$; $w_i(t)=W_i(t)/\ow(t)$.
158: (We will simply call $w_i(t)$ the normalized wealth hereafter.)
159: The exponent is given by $\mu=1+J/\sigma^2$.
160: In the large $w$ range Eq.~(\ref{eq:mfsolution}) becomes a power law
161: distribution, $p(w)\propto w^{-(1+\mu)}$, and therefore the exponent $\mu$ is
162: the Pareto index in economics.
163:
164: \begin{figure}[h]
165: \begin{center}
166: \resizebox{.9\textwidth}{!}{\includegraphics{fig1.eps}}
167: \end{center}
168: \caption{\label{fig:1}
169: Some of the wealth distributions obtained by numerical simulations.
170: The rank distribution in regular networks is plotted in (a) for $z=0.1$
171: and in (b) for $z=0.001$.
172: The rank distribution in small-world networks
173: is plotted in (c) for $z=0.001$ and $p=0.01$
174: and in (d) for $z=0.001$ and $p=0.5$.}
175: \end{figure}
176:
177: We have carried out numerical simulations for $N=10^4$,
178: $m=1$, $J=0.01$ and $\sigma^2=0.01$.
179: Under this parameter set, the Pareto index in the MF case is
180: $\mu_{\textrm{mf}}=2$.
181: Some of the distributions we obtained are shown in Figs.~\ref{fig:1}~(a)-(d).
182: In these figures the horizontal axis is the logarithm of
183: $w$ and the vertical axis is that of the rank.
184: We have numerically calculated up to $2\times 10^4$ time steps,
185: and drawn distributions at each $4\times 10^3$ step.
186: The rank distributions in the case of regular networks
187: with $z=0.1$ correspond to Fig.~\ref{fig:1}~(a).
188: The solid lines show the best fits with the MF
189: result Eq.~(\ref{eq:mfsolution}).
190: We see that all of the resulting distributions fit well with
191: the MF result with $\mu\simeq\mu_{\textrm{mf}}$.
192: Figure~\ref{fig:1}~(b) is a case of regular networks with $z=0.001$.
193: We find that the results do not agree with the MF result,
194: but instead fit well with log-normal distributions,
195: which are shown by the solid lines.
196: Note that the best fit log-normal distributions have
197: time-dependent mean value $m_{\textrm{ln}}(t)$ and
198: time-dependent variance $\sigma^2_{\textrm{ln}}(t)$.
199: On the other hand, if we rewire links with high probability
200: the MF type distributions reappear.
201: Figure~\ref{fig:1}~(c) is a case of regular networks
202: with $z=0.001$ and $p=0.5$.
203: The solid lines are the fitting by
204: MF type solutions with $\mu=1.7\sim 1.8$, which is slightly
205: smaller than $\mu_{\textrm{mf}}$.
206: Hence we observe that the change in wealth distribution occurs
207: at the intermediate value of $p$.
208: Figure~\ref{fig:1}~(d) is a case of small-world networks
209: with $z=0.001$ and $p=0.05$.
210: We have found that the distributions can not be fitted well by either the
211: log-normal functions or the MF type functions.
212: Interestingly, however, we find that
213: a combination of log-normal and power-law distributions fits
214: well with the results.
215: One such example at the $2\times 10^4$-th time step
216: is shown with a log-normal function
217: for middle and low wealth ranges (solid line)
218: and a power-law function for the high wealth range (dashed line).
219: Log-normal distributions with a power law tail like this are
220: frequently observed
221: in real world economic phenomena \cite{souma}.
222:
223: It is natural to question why distribution patterns change.
224: To answer this question, we directly consider the development
225: of the wealth.
226: Examples of numerical simulations
227: are shown in Figs.~\ref{fig:2}~(a)-(d).
228: In these figures the horizontal axis shows the agents' number
229: and the vertical axis shows time steps.
230: Though we have simulated for $N=10^4$ agents, wealth developments
231: for $10^3$ agents are shown.
232: We have calculated up to $10^4$ time steps.
233: If any agent ranks in the top $10\%$, we mark it with a black dot, and
234: if any agent ranks in the bottom $10\%$ we mark it with a red dot.
235:
236: \begin{figure}[h]
237: \begin{center}
238: \resizebox{0.45\textwidth}{!}{\includegraphics{fig2a.eps}}
239: \resizebox{0.45\textwidth}{!}{\includegraphics{fig2b.eps}}\\
240: \resizebox{0.45\textwidth}{!}{\includegraphics{fig2c.eps}}
241: \resizebox{0.45\textwidth}{!}{\includegraphics{fig2d.eps}}
242: \end{center}
243: \caption{\label{fig:2}
244: Wealth developments for regular networks with $z=0.1$ are shown in (a)
245: and with $z=0.001$ in (b).
246: The case of small-world networks with $z=0.001$
247: and $p=0.05$ is shown in (c), and $p=0.5$ in (d).}
248: \end{figure}
249: Figure~\ref{fig:2}~(a) is a case of regular networks with $z=0.1$.
250: We can see that the black and red dots are uniformly distributed
251: without notable
252: clustering.
253: Since each agent has $10^3$ neighbors,
254: each agent transacts with agents from various ranks, covering a wide range.
255: Figure~\ref{fig:2}~(b) is a case of regular networks with $z=0.001$.
256: Contrary to the case of $z=0.1$, the black and red dots cluster heavily.
257: This ``wealth clustering'' phenomenon is supported by the fact that
258: each agent has only $10$ neighbors and rich agents transact mainly
259: with rich agents and poor agents transact mainly with poor agents.
260: Hence $W_i(t)$ and $\ow_i(t)$ have strong correlations.
261: Actually, for the case of regular networks
262: with $z=0.001$, we can obtain
263: $\langle\own_i(t)w_i(t)\rangle=0.92\sim0.95$.
264: Here $\own_i(t)\equiv\ow_i(t)/\ow(t)$.
265: On the other hand
266: we can obtain $\langle\own_i(t)w_i(t)\rangle=0.02\sim0.09$
267: for the case of regular networks with $z=0.1$.
268: Figure~\ref{fig:2}~(c) is a case of small-world networks
269: with $z=0.001$ and $p=0.5$.
270: This case is almost similar to Fig.~\ref{fig:2}~(a).
271: Figure~\ref{fig:2}~(d) is
272: a case of small-world networks
273: with $z=0.001$ and $p=0.05$.
274: Although wealth clustering is observed, the size of each cluster
275: is smaller than the case of regular networks with $z=0.001$.
276: From these results we find that
277: a decrease in the number of links causes wealth clustering and
278: the rewiring of links disperses it.
279:
280: If $W_i(t)$ and $\ow_i(t)$ have strong correlations,
281: Eq.~(\ref{eq:disc}) becomes a
282: pure multiplicative stochastic process, and the wealth obeys
283: log-normal distributions with mean $vt$ and variance $Dt$,
284: where
285: $v=\langle\log\{a(t)+\sigma^2\}\rangle$ and
286: $D=\langle[\log\{a(t)+\sigma^2\}]^2\rangle-
287: \langle\log\{a(t)+\sigma^2\}\rangle^2$.
288: On the other hand if $W_i(t)$ and $\ow_i(t)$ have no correlations,
289: Eq.~(\ref{eq:disc}) is regarded as a multiplicative stochastic
290: process with additive noise \cite{kesten}.
291: It is known that this process induces
292: power law distributions in large $W_i(t)$ ranges,
293: if two conditions are satisfied:
294: One condition is that the stochastic variable $a_i(t)$ and
295: the additive noise $J\ow_i(t)$ must behave independently.
296: The other condition is
297: $\langle\log\{a_i(t)+\sigma^2-J\}\rangle<0$.
298: Although the first condition does not matter in our model,
299: the parameter ranges of
300: $J$ and $\sigma^2$ are restricted by the second condition.
301: If $W_i(t)$ and $\ow_i(t)$ have no correlations
302: in the high wealth ranges and have strong correlations in the
303: middle and low wealth ranges, the wealth will obey log-normal
304: distributions with a
305: power law tail such as those observed in many real world economic phenomena
306: \cite{souma}.
307:
308: \begin{figure}[t]
309: \begin{center}
310: \resizebox{0.45\textwidth}{!}{\includegraphics{fig3a.eps}}
311: \resizebox{0.45\textwidth}{!}{\includegraphics{fig3b.eps}}
312: \end{center}
313: \caption{\label{fig:3}
314: Change of Gini coefficient in the case of regular networks (a),
315: and that of small-world networks (b).
316: }
317: \end{figure}
318: We will now study the inequality of wealth distribution
319: by using the Gini coefficient $G$ to quantify it.
320: Take
321: every possible pair of income recipients (wealth possessors) and
322: calculate the average of the absolute differences of the two incomes
323: (wealth) for each pair. $G$ is then half the ratio of the average to
324: the mean value of all the incomes (wealth). Obviously $G=0$ when
325: everyone has the same, and $G=1$ when the entire amount is
326: concentrated in a single person.
327: With our parameter set, the Gini coefficient of the MF
328: case is $G_{\textrm{mf}}=0.5$.
329:
330: The temporal changes of $G$ are shown in Figs.~\ref{fig:3} (a) and (b).
331: In these figures, the horizontal axis shows time steps
332: and the vertical axis shows the value of $G$.
333: We have numerically simulated up to $5\times 10^4$ time steps.
334: The temporal changes of $G$ in the case of
335: regular networks are shown in Fig.~\ref{fig:3}~(a).
336: In this figure we observe
337: that $G$ gradually increases with time steps
338: in the case of $z=0.01$ and $0.02$, which is
339: in accordance with the log-normal distributions with $m_{\textrm{ln}}(t)$
340: and $\sigma^2_{\textrm{ln}}(t)$.
341: We also observe that almost completely inequal distributions
342: appear in the case of $z=0.001$.
343: We find that the decrease of $z$ makes distributions inequal.
344: The temporal changes of $G$ in the case of
345: small-world networks with $z=0.001$ are shown in Fig.~\ref{fig:3}~(b).
346: From this figure we find that the rewiring of links in
347: small-world networks makes distributions equal.
348:
349: The results of this study are summarized in two parts.
350: First, we have carried out a simulation
351: for a wide range of values of $p$ and $z$,
352: which is summarized by the phase diagram shown in Fig.~\ref{fig:4}.
353: This diagram has two extreme distributions,
354: the log-normal distributions with
355: $m_{\textrm{ln}}(t)$ and $\sigma^2_{\textrm{ln}}(t)$
356: and the MF type distributions represented by Eq.~(\ref{eq:mfsolution}).
357: These two ranges are interpolated by
358: a decrease in the number of links or the rewiring of the links.
359: In the intermediate region, the wealth distribution may be fitted well
360: with log-normal distributions with a power law tail,
361: which are observed in real world economic phenomena \cite{souma}.
362: Lastly, from the study of the Gini coefficient
363: we have found that a decrease in the number of links
364: makes wealth distribution inequal and the rewiring of links in
365: small-world networks makes it equal.
366: Hence society will be protected against exceeding inequalities
367: in wealth by small-world effects in economic networks.
368:
369: \begin{figure}[t]
370: \begin{center}
371: \resizebox{0.5\textwidth}{!}{\includegraphics{fig4}}
372: \end{center}
373: \caption{\label{fig:4}
374: Patterns of wealth distribution in the case of $N=10^4$.
375: }
376: \end{figure}
377:
378: We note that it would be very interesting to carry out an analysis of
379: models similar to the ones studied here, but built on other kinds of
380: networks. For example, the network can be dynamic, allowing
381: rewiring to be either independent from or dependent on wealth.
382: Other novel networks, like scale-independent networks,
383: would also be worth studying.
384: We believe that the work presented here forms a basis for the
385: study of general stochastic processes on complex networks.
386:
387: The authors would like to thank Dr.~K.~Shimohara (ATR-ISD)
388: for his continuous encouragement and warm support.
389:
390: \begin{thebibliography}{99}
391: \bibitem{WS}
392: D.~J.~Watts and S.~H.~Strogatz, Nature {\bfseries 393}, 440 (1998);
393: D.~J.~Watts, {\itshape Small Worlds: The Dynamics of Networks
394: between Order and Randomness.} (Princeton University Press, Princeton,
395: New Jersey 1999)
396: \bibitem{NSAB}
397: M.~E.~Newman, cond-mat/0001118;
398: S.~H.~Strogatz, Nature {\bfseries 410}, 268 (2001);
399: R.~Albert and A.-L.~Barab\'asi, cond-mat/0106096;
400: S.~N.~Dorogovtsev and J.~F.~F.~Mendes, cond-mat/0106144.
401: \bibitem{souma}
402: H.~Aoyama, W.~Souma, Y.~Nagahara, H.~P.~Okazaki,
403: H.~Takayasu and M.~Takayasu, Fractals {\bfseries 8}, 293 (2000);
404: W.~Souma, to appear in Fractals {\bfseries 9}, (2001), cond-mat/0011373.
405: \bibitem{IKR}
406: S.~Ispolatov, P.~L.~Krapivsky and S.~Redner, Eur. Phys. Jour.
407: B {\bfseries 2}, 267 (1998).
408: \bibitem{solomon1}
409: S.~Solomon, {\itshape Computational Finance 97}, ed. A-P.~N.~Refenes,
410: A.~N.~Burgess, J.~E.~Moody,
411: (Kluwer Academic Publications, Dordecht 1998),
412: cond-mat/9803367.
413: \bibitem{BM}
414: J.~P.~Bouchaud and M.~M\'ezard, Physica A {\bfseries 282}, 536 (2000).
415: \bibitem{BMetc}
416: Z.~Burda,
417: D.~Johnston, J.~Jurkiewicz, M.~Kami\'nski, M.~A.~Nowak,
418: G.~Papp and I.~Zahed, cond-mat/0101068.
419: \bibitem{kesten}
420: H.~Kesten, Acta Math. {\bfseries 131}, 207 (1973);
421: D.~Sornette and R.~Cont, J. Phys. I {\bfseries 7}, 431 (1997);
422: H.~Takayasu, A.-H.~Sato and M.~Takayasu,
423: Phys. Rev. Lett. {\bfseries 79}, 966 (1997);
424: D.~Sornette, Phys. Rev. E {\bfseries 57}, 4811 (1998).
425: \end{thebibliography}
426: \end{document}
427: