physics0608181/gini.tex
1: %&LaTeX
2: 
3: \documentclass[12pt]{article}
4: \usepackage{amssymb,amsmath,graphicx}
5: \usepackage{longtable}
6: \usepackage{hyperref}
7:                                
8: \newcommand{\tab}{\hspace{5mm}}
9: 
10: 
11: \begin{document}
12: 
13: \begin{center}
14: \textbf{{\Large A STRONG FACTOR FOR THE\\
15: \vspace{0.2 cm}
16: REDUCTION OF INEQUALITY}}\\
17: \vspace{0.5 cm}
18: {Diego Sa\'{a}}
19: \footnote{Escuela Polit\'{e}cnica Nacional. Quito -- Ecuador. email: dsaa@server.epn.edu.ec}\\
20: Copyright {\copyright}2006\\
21: \end{center}
22: 
23: \begin{abstract}
24: {The inequality is computed through the so-called Gini index. 
25: The population is assumed to have the variable of interest distributed 
26: according to the Gamma probability distribution. The results show that the Gini 
27: index is reduced when the population is grouped. 
28: The number of individuals in the groups is the relevant parameter, 
29: but this number does not need to be very large in order to obtain 
30: a very substantial reduction of inequality.}
31: \end{abstract}
32: 
33: \textit{PACS}:
34: 
35: 87.23.Ge Dynamics of social systems
36: 
37: 02.50.-r Probability theory, stochastic processes, and statistics
38: 
39: 65.50.+m Thermodynamic properties and entropy\\
40: 
41: 
42: \textit{Keywords}: econophysics, thermodynamics, probability distributions, Gini index, inequality, entropy\\
43: 
44: 
45: \textbf{1. INTRODUCTION}
46: 
47: 
48: The Gamma probability distribution is a powerful and flexible distribution 
49: that applies with absolute precision to a great variety of problems 
50: and systems in thermodynamics, solid state physics, economics, etc. \\
51: 
52: The present author has suggested \cite{saa} that this distribution should 
53: replace, in particular, the Planck distributions, used to describe 
54: the blackbody radiation distribution, as well as the Maxwell velocity distribution for ideal gases. \\
55: 
56: Also, in the area of econophysics, the Gamma distribution should 
57: replace profitably all of the other distributions currently used, 
58: such as the following (some of them are the same one and are 
59: instances of the Gamma distribution): Gibbs, negative exponential 
60: or simply exponential, Boltzmann, log-normal, power law, Pareto-Zipf, 
61: Erlang and Chi-squared. Simulations and applications using the 
62: Gamma distribution \cite{bhat}, \cite{patriarca},  \cite{patriarca2}, \cite{scafetta}, have shown that the Gamma distribution 
63: better fits the actual distribution of the variable of interest.\\
64: 
65: In the present paper the author develops the formula to compute 
66: the Gini index corresponding to some variable distributed in 
67: a population according to the Gamma probability distribution.\\
68: 
69: 
70: 
71: \textbf{2. THE GINI INDEX}\\
72: 
73: The Gini index, Gini ratio or Gini coefficient, is probably the 
74: most well-known and broadly used measure of inequality used in 
75: economic literature. \\
76: 
77: The Gini index derives from the Lorenz Curve. To plot a Lorenz 
78: curve, order the observations from lowest to highest on the variable 
79: of interest, such as income, and then plot the cumulative proportion 
80: of the population on the X-axis and the cumulative proportion 
81: of the variable of interest on the Y-axis. \\
82: 
83: If all individuals have the same income the Lorenz curve is a 
84: straight diagonal line, called the line of equality. If there 
85: is any inequality, then the Lorenz curve falls below the line 
86: of equality. The total amount of inequality can be summarized 
87: by the Gini index, which is the proportion of the area enclosed 
88: by the lines of equality and the Lorenz curve divided by the total triangular area under the line of equality.\\
89: 
90: In the figure below, the diagonal line represents perfect equality. The 
91: greater the deviation of the Lorenz curve from this line, the 
92: greater the inequality. The Gini index is double the area between 
93: the equality diagonal and the Lorenz curve. The minimum value 
94: of the Gini can be zero (perfect equality) and the greater can 
95: be one (the case when a single member of the population holds 
96: all of the variable). \cite{hale}\\
97: 
98: \begin{figure}[htbp]
99: \begin{center}
100: % \includegraphics[viewport=-50 0 800 320,width=15cm,clip]
101: \includegraphics[bb = 20 70 592 345, scale=0.80]
102: %[viewport=-50 0 700 420,width=16cm,clip]
103: %[bb = 20 20 592 449, scale=1.00]
104: {gini.eps} 
105: \caption{The Gini index}
106: \end{center}
107: \end{figure}
108: 
109: \vspace{3.5 cm}
110: 
111: \textbf{3. THE GAMMA DISTRIBUTION}\\
112: 
113: 
114: The author of this paper has proposed \cite{saa} that the Gamma distribution 
115: seems to be the correct distribution of blackbody energy radiation, 
116: money and other variables from comparable continuous systems.\\
117: 
118: 
119: The Gamma probability distribution function (\textit{pdf}) \cite{saa} has the form
120: 
121: 
122: \begin{equation}
123: P(p,\lambda ,x)=\frac{\lambda ^{p} \cdot x^{p-1} }{Exp(\lambda \cdot
124: x)\cdot \Gamma (p)}
125: \end{equation}
126: 
127: 
128: The parameters of this distribution are called the shape parameter 
129: (\textit{p}) and the scale parameter ($\lambda $).\\
130: 
131: The Gamma function satisfies:
132: 
133: \begin{equation}
134: \Gamma (p)=\int\nolimits_{0}^{\infty }x^{p-1} e^{-x} dx 
135: \end{equation}
136: 
137: This expression is the equivalent, for the Gamma distribution, of the \textit{partition function} defined in classical thermodynamics for the Boltzmann distribution.\\
138: 
139: The incomplete Gamma function $\Gamma (p,\lambda \cdot x)$, used in the following, has a similar integrand, and the only difference is that the lower limit of integration is $\lambda \cdot x$ instead of zero.\\
140: 
141: 
142: If the value of the variable \textit{p} is particularized to an integer 
143: value then this distribution converts into the Erlang distribution.\\
144: 
145: If the variable \textit{p} has the value 1, the Gamma distribution 
146: converts into the negative exponential distribution, also called 
147: Boltzmann, Gibbs, Boltzmann-Gibbs or simply exponential law.\\
148: 
149: 
150: 
151: In the area of econophysics, the use of the so-called 
152: Pareto or power law distribution is very common, although it is obvious that 
153: it is not a probability distribution (where the sum or integral of probabilities is equal to the unity) because its integral does 
154: not converge. It should be profitably replaced by the Gamma 
155: distribution with the proper parameters.\\
156: 
157: The average of a quantity \textit{x} distributed according to the Gamma distribution is
158: \begin{equation}
159: \left\langle x\right\rangle =\frac{p}{\lambda }
160: \end{equation}
161: 
162: As a result, if we keep the average of \textit{x} equal to the unity 
163: then \textit{p} must\\
164: be equal to $\lambda $. Nevertheless the larger the values of $p = \lambda$ the smaller the variance since this is given by:
165: 
166: \begin{equation}
167: \sigma ^{2} =\frac{p}{\lambda ^{2} } 
168: \end{equation}
169: 
170: The maximum of the Gamma distribution is at the position
171: 
172: 
173: \begin{equation}
174: x_{\max } =\frac{p-1}{\lambda } =\left\langle x\right\rangle
175: -\frac{1}{\lambda } 
176: \end{equation}\\
177: 
178: \vspace{2cm}
179: 
180: \textbf{4. INCOME OF GROUPS OF EARNERS}\\
181: 
182: Dragulescu and Yakovenko \cite{dragulescu1} compute the distribution of the 
183: combined income of two earners and get the following formula, 
184: which they find in excellent agreement with the income data of the USA population. They assume that the income \textit{r} of two earners is the sum 
185: of the individual incomes: \textit{r}=\textit{r}{\small 1}+\textit{r}{\small 2}. Hence, the total 
186: income \textit{pdf}, $P_{2}(r)$, is given by the convolution
187: 
188: 
189: \begin{equation}
190: P_{2} (r)=\int\nolimits_{0}^{r}P_{1} (x) \cdot P_{1} (r-x)dx=\frac{r}{R^{2} }
191: e^{-r/R}
192: \end{equation}
193: 
194: 
195: 
196: where the individual incomes \textit{r}{\small 1} and \textit{r}{\small 2} are assumed 
197: to be uncorrelated and to have exponential distributions of the form: \textit{P}$_{\mathit{1}}$\textit{(r)=e}$^{\mathit{-r/R}}$\textit{/R}, 
198: where \textit{R} is the average income of the population. 
199: 
200: As is well known, the exponential distribution, also called Boltzmann 
201: or Gibbs distribution, is a special case of the Gamma distribution 
202: when the parameter \textit{p} is equal to 1. Whereas the resulting expression 
203: (6) is a Gamma distribution with parameter \textit{p}=2. This expression describes the 
204: income distribution of a population of groups of two earners.
205: 
206: By generalizing this idea and maintaining constant the scale 
207: parameter $\lambda $ we can verify that the convolution of two Gamma distributions, 
208: with respective shape parameters \textit{p1} and \textit{p2}, produces another 
209: Gamma distribution with shape parameter \textit{p1+p2}:
210: 
211: \begin{equation}
212: P(p1+p2,\lambda ,r)=\int\nolimits_{0}^{r}P(p1,\lambda ,x)\cdot  P
213: (p2,\lambda ,r-x)dx
214: \nonumber
215: \end{equation}
216: 
217: \begin{equation}
218: P(p1+p2,\lambda ,r)=\frac{\lambda ^{p1+p2} }{\Gamma (p1+p2)} r^{p1+p2-1}
219: e^{-\lambda \cdot r} 
220: \end{equation}
221: 
222: If we interpret the shape parameter \textit{p} as the number of earners 
223: then equation (7) simply says that the income distribution of 
224: the sum of incomes of two groups, with respective number of earners \textit{p1} 
225: and \textit{p2}, is given by the (Gamma) income distribution of the sum of earners. The parameters \textit{p} can have any positive real values.\\
226: 
227: \vspace{2.0cm}
228: 
229: \textbf{5. THE GINI INDEX OF GROUPED EARNERS}\\
230: 
231: The horizontal axis of the Lorenz curve, \textit{x}(\textit{r}), represents 
232: the cumulative fraction of population with income below \textit{r}, 
233: and the vertical axis \textit{y}(\textit{r}) represents the fraction of 
234: income this population accounts for. \\
235: The respective values for these fractions are given by the following 
236: formulas \cite{dragulescu3}:
237: \begin{equation}
238: x(r)=\int\nolimits_{0}^{r}P(p,\lambda ,r')dr' =1-\frac{\Gamma (p,\lambda
239: \cdot r)}{\Gamma (p)} 
240: \end{equation}
241: 
242: \begin{equation}
243: y(r)=\frac{\int\nolimits_{0}^{r}r'\cdot P(p,\lambda ,r')dr'
244: }{\int\nolimits_{0}^{\infty }r'\cdot P(p,\lambda ,r')dr' } =1-\frac{\Gamma
245: (p+1,\lambda \cdot r)}{\Gamma (p+1)} 
246: \end{equation}\\
247: 
248: The range of these variables is between 0 and 1.
249: The Gini index for the Gamma probability distribution is obtained replacing (8) and (9) into the following integral:
250: 
251: 
252: \begin{equation}
253: G=2\cdot \int\nolimits_{0}^{1}(x-y)\cdot dx
254: \end{equation}
255: 
256: or, as a function of r:
257: 
258: \begin{equation}
259: G=1-2\cdot \int\nolimits_{0}^{\infty}y(r) \frac{dx(r)}{dr}\cdot dr
260: \end{equation}
261: 
262: Whose result is:
263: 
264: 
265: \begin{equation}
266: G=\frac{\Gamma (2 p+1)}{4^p \cdot \Gamma (p+1)^2}
267: \end{equation}\
268: 
269: 
270: This shows that the gini index is independent of the values of $\lambda$ but depends on the number of individuals in the groups, p.
271: For example, if we instantiate the parameter \textit{p} to 1, this formula gives 
272: the Gini index for the exponential distribution, which is 1/2. This is the Gini for one earner and for any value 
273: of the parameter $\lambda $.\\
274: 
275: Let us compute the Gini for the Gamma distribution for a few 
276: earners (integer values of the parameter \textit{p}). \\
277: 
278: The following table shows the Gini index corresponding 
279: to each value of \textit{p} between 1 and 5, and the proportion of 
280: the first Gini relative to the second, etc.\\
281: 
282: 
283: 
284: \begin{tabular}{c|c|c}
285: \hline
286: p	& Gini(p) &	Proportion\\
287:  & & Gini(p)/Gini(p+1)\\
288: \hline
289: 1	& 1/2      = 0.500 &	1.33\\
290: 2	& 3/8      = 0.375	& 1.20\\
291: 3	& 5/16      = 0.3125	& 1.14\\
292: 4	& 35/128 = 0.2734	& 1.11\\
293: 5	& 63/256 = 0.2461	& \\
294: \hline
295: \end{tabular}\\
296: 
297: Table1. Gini index as a function of the number of earners (p)\\
298: 
299: This table shows an important reduction of the Gini index, of
300: 0.125 points when the number of earners in the groups passes from 1 to 2 
301: and of an additional 0.0625 when the number of earners 
302: rises from 2 to 3. The proportion between the Gini index corresponding 
303: to a given number \textit{p} of earners and the following, \textit{p}+1, 
304: tends to 1 as the number of earners in the groups grows.\\
305: 
306: The following simple, but approximate, formula provides values up to around 11.4\% lower than the previous exact formula:
307: 
308: \begin{equation}
309: G=\frac{1}{2 \cdot \sqrt {p}}
310: \nonumber
311: \end{equation}
312: 
313: For example for p=4, this formula gives the value 0.25, whereas the exact value, shown in the previous table, is around 0.2734; for p=100 this formula provides 0.05, but the exact value is close to 0.05635.\\
314: 
315: It is important to know this mechanism for the reduction of inequality. 
316: But, of course, the next more important issue would be how to 
317: form the groups and achieve the redistribution, of the individual 
318: income of each one of the individuals that constitute the group 
319: of earners, among all of them. This point is addressed very briefly here and should be addressed more deeply by 
320: other investigators.\\
321: 
322: The persons that constitute the groups must be selected randomly 
323: from the entire population, which is assumed to have a Gamma 
324: probability distribution of the income. Otherwise I would prefer 
325: to ``share my wealth'' with Gates and Rockefellers.\\
326: 
327: More seriously, the financial institutions, welfare, non-governmental 
328: organizations, etc. should prefer to finance and help groups 
329: instead of to particular individuals. There already are many 
330: forms of organizations in the world that procure this kind of 
331: behavior, such as cooperatives, kibbutz, comunas (from common), 
332: families, etc., which have demonstrated to be a very good mechanism 
333: for the redistribution of the income and consequent reduction 
334: of poverty and inequality.\\
335: 
336: \textbf{6. ENTROPY OF THE GAMMA DISTRIBUTION}\\
337: 
338: 
339: The entropy of the Gamma distribution was defined by the present 
340: author in other paper \cite{saa} in the form:
341: 
342: \begin{equation}
343: S(p,\lambda ,x)=1-\frac{\Gamma (p,\lambda \cdot x)}{\Gamma (p)} 
344: \end{equation}
345: 
346: where \textit{p} and $\lambda $ are the shape and scale parameters of the Gamma distribution 
347: and \textit{x} is the variable being distributed. Note that this expression 
348: is identical to (8).\\
349: 
350: Expression (13) is the definition of the ``non-extensive'' entropy, in the sense that it does not have units and is precisely the cumulative distribution function (CDF) of the Gamma probability distribution. The incomplete Gamma function alone, which is the numerator in this expression, can be considered as the corresponding ``extensive entropy''.\\
351: 
352: Litchfield \cite{litchfield} compares several measures of inequality and exposes, 
353: following Cowell, that any member of the Generalized Entropy 
354: (GE) class of inequality measures satisfies five axioms, which 
355: we now try to apply to the Gamma entropy:\\
356: 
357: 
358: \textit{The Pigou-Dalton Transfer Principle}. An income transfer from 
359: a poorer person to a richer person should register as a rise 
360: (or at least not as a fall) in inequality and an income transfer 
361: from a richer to a poorer person should register as a fall (or 
362: at least not as an increase) in inequality. 
363: 
364: This axiom does not apply, since the Gamma distribution is obtained 
365: from an equilibrium equation among actors with different incomes. 
366: Any transfer between them should maintain the Gamma distribution 
367: and hence also the equilibrium. The entropy of the Gamma distribution 
368: does not depend on the individual incomes but on the complete 
369: statistical distribution.\\
370: 
371: \textit{Income Scale Independence.} This requires that the inequality 
372: measure be invariant to changes in scale as happens say when changing 
373: currency unit. 
374: 
375: The Gamma distribution passes this test because the parameter $\lambda $ works as an average that suppresses any additional factor in the variable \textit{x}.\\
376: 
377: \textit{Principle of Population}. This principle requires inequality 
378: measures to be invariant to replications of the population: merging 
379: two identical distributions should not alter inequality. 
380: 
381: Again, the Gamma distribution is a statistical distribution and 
382: therefore is not affected by the number of individuals in the 
383: population.\\
384: 
385: \textit{Anonymity.} This axiom, sometimes also referred to as `\textit{Symmetry}', 
386: requires that the inequality measure be independent of any characteristic 
387: of individuals other than their income.
388: 
389: The Gamma distribution satisfies this axiom trivially. \\
390: 
391: \textit{Decomposability.} This requires overall inequality to be related 
392: consistently to constituent parts of the distribution, such as 
393: population sub-groups. For example if inequality is seen to rise 
394: amongst each sub-group of the population then we would expect 
395: inequality overall to also increase. 
396: 
397: The Gamma distribution satisfies this axiom through the Gini 
398: index associated with the Gamma distribution, as was proved in 
399: section 5. The parameter \textit{p} of the Gamma entropy also takes 
400: into account the number of members in the groups, but with a 
401: more compact expression.\\
402: 
403: 
404: \textbf{7. CONCLUSIONS}\\
405: 
406: 
407: The analysis shown in section 4 proves that the Gamma entropy solves the so-called \textit{Gibbs}' 
408: \textit{paradox}. Current Physics assumes that the entropy should 
409: not change as a result of mixing two amounts of identical gases. In the present paper it has been proved that this assumption does not hold when we use \textit{non-extensive} definitions of entropy, such as the cumulative or normalized Gamma entropy. It is also doubtful that the entropy will not change for the extensive case.\\
410: 
411: For example, in his ``Thermodynamics Lecture Notes'' \cite{melrose}, Prof. Professor 
412: Donald B. Melrose, Director, RCfTA and Head of Theoretical Physics, 
413: School of Physics, University of Sydney says: ``It follows that 
414: the entropy increases in this case and it is not difficult to 
415: see that the entropy change as a result [of] mixing is always positive. If they [the gases] are identical then the change in entropy must 
416: be zero and yet the calculation seems to imply that there is 
417: a change in entropy. This is referred to as the Gibbs paradox. There 
418: is no simple physical resolution of the Gibbs paradox within 
419: the framework of classical statistical mechanics.''\\
420: 
421: It is clear that the entropy associated with the distribution (7), which 
422: is the distribution of the combined income of two earners, is 
423: given by the Gamma entropy with parameter 
424: (\textit{p1}+\textit{p2}), whereas the income distributions of each one of the earners have associated individual Gamma entropies with respective parameters \textit{p1} and \textit{p2}. Therefore, the entropy associated with the sum of a 
425: certain variable belonging to two or more actors must change 
426: even though the actors were identical. The original entropies 
427: are recovered if, for the studied variable, the individual incomes of the actors are again considered independently. \\
428: 
429: In both cases the population is the same but the values for the studied variable are different. The different income values depend on the grouping 
430: of individuals and on the corresponding averaging of the variable.\\ 
431: 
432: The non-extensive entropy computed for groups with the same number of individuals must not change for a new population obtained combining two populations that have the same statistical properties. The statistical (non-extensive) properties of the combined population, such as average temperature, are conserved; however, the corresponding extensive properties, such as the total energy or money of the system and even the particular values corresponding to each particle or individual of the population, necessarily change. 
433: 
434: 
435: \begin{thebibliography}{99}
436: 
437: \bibitem{bhat}
438: Bhattacharya, Mukherjee \& Manna. Detailed simulation results 
439: for some wealth distribution models in Econophysics. arXiv: physics/0504161 
440: v1, Apr 2005. 
441: 
442: 
443: \bibitem{dragulescu1}
444: Dragulescu, Adrian \&Yakovenko, Victor. Statistical Mechanics 
445: of Money, Income, and Wealth: A Short Survey. arXiv: cond-mat/0211175. 
446: Nov. 2002. 
447: 
448: \bibitem{dragulescu3}
449: Dragulescu, Adrian. Applications Of Physics To Economics 
450: And Finance: Money, Income, Wealth, And The Stock Market. arXiv: 
451: cond-mat/0307341 v2, 16 Jul. 2003. 
452: 
453: \bibitem{hale}
454: Hale, Travis. The Theoretical Basics of Popular Inequality 
455: Measures. University of Texas Inequality Project.\\ http://utip.gov.utexas.edu/tutorials/theo\_basic\_ineq\_measures.doc
456: 
457: \bibitem{litchfield}
458: Litchfield, Julie A. Inequality: Methods and Tools. March 
459: 1999.\\
460: http://www.worldbank.org/poverty
461: 
462: \bibitem{melrose}
463: Melrose, Donald B. Lecture 2: Maxwell Distribution: Ideal 
464: Gases. http://ckw.phys.ncku.edu.tw/public/pub/WebSources/Melrose/\\
465: www.physics.usyd.edu.au/rcfta/thermo.html
466: 
467: \bibitem{patriarca}
468: Patriarca, Chakraborti \& Kaski. Gibbs versus non-Gibbs distributions 
469: in money dynamics. Physica A 340 (2004) 334-339. Elsevier.
470: 
471: \bibitem{patriarca2}
472: Patriarca, Chakraborti \& Germano. Influence of saving propensity on the power law tail of wealth distribution. arXiv: physics/0506028 v1, Jun 2005. 
473: 
474: \bibitem{saa}
475: Sa\'{a}, Diego. On An Improvement Of The Planck Radiation Energy 
476: Distribution. In: http://arxiv.org/abs/physics/0603117 v3, Jul 2006.
477: 
478: \bibitem{scafetta}
479: Scafetta, Picozzi \& West. A trade-investment model for distribution 
480: of wealth. Physica D 193 (2004) 338-352. Elsevier.
481: 
482: 
483: \end{thebibliography}
484: 
485: 
486: 
487: \end{document}
488: