1: % Template article for preprint document class `elsart'
2: % SP 2001/01/05
3:
4: \documentclass{elsart}
5:
6: % Use the option doublespacing or reviewcopy to obtain double line spacing
7: % \documentclass[doublespacing]{elsart}
8:
9: \usepackage{epsfig}
10:
11: % The amssymb package provides various useful mathematical symbols
12: \usepackage{amssymb}
13:
14: \begin{document}
15:
16: \begin{frontmatter}
17:
18: \title{On anomalous distributions in intra-day financial time series and
19: Non-extensive Statistical Mechanics}
20:
21: \author{Silvio M. Duarte Queir\'{o}s}
22:
23: \ead{sdqueiro@cbpf.br}
24: %
25: \address{
26: Centro Brasileiro de Pesquisas F\'\i sicas, \\
27: Rua Dr. Xavier Sigaud 150,22290-180 Rio de Janeiro-RJ, Brazil
28: }
29:
30: \begin{abstract}
31: In this paper one studies the distribution of log-returns (tick-by-tick)
32: in the Lisbon stock market and shows that it is well adjusted by the
33: solution of the equation, \mbox{$\frac{dp_{x}}{d\left| x\right| }=-\beta
34: _{q^{\prime }}p_{x}^{q^{\prime }}-\left( \beta _{q}-\beta _{q^{\prime
35: }}\right) p_{x}^{q}$}, which corresponds to a generalization of the
36: differential equation which has as solution the power-laws that optimise the
37: entropic form $S_{q}=-k \frac{1-\int p_{x}^{q}\ dx}{1-q}$, base of present
38: non-extensive statistical mechanics.
39: \end{abstract}
40:
41: \begin{keyword}
42: Econophysics \sep Nonextensive Statistical Mechanics \sep Complex systems
43:
44: \PACS 89.65.Gh \sep 89.75.Da \sep 05.20.-y
45: %
46: % 89.65.-s Social systems
47: % 89.65.Gh Economics, business, and financial markets
48: % 05.20.-y Classical statistical mechanics
49: % 89.75.-k Complex systems
50: % 89.75.Da Systems obeying scaling laws
51: \end{keyword}
52: \end{frontmatter}
53:
54: \section{Introduction}
55: %
56: Since Mandelbrot's work \cite{mandelbrot} on analysing cotton prices that L\'{e}vy distribution has been
57: widely used, but, frequently with L\'{e}vy scaling coefficients greater than
58: $2$, and so, out of L\'{e}vy regime\cite{stanley,pobodnik}. An alternative
59: way of describing these tailed distributions can be power laws (usually
60: called ${\it q}${\it -Gaussians }or {\it Tsallis distributions}) obtained by
61: optimising the entropic form proposed by Constantino Tsallis\cite
62: {silvio,ausloos,tsallis-borland}. Those distributions, beyond presenting the
63: advantage of finite second (and higher) order momenta (for a certain range
64: of values of its characterizing index $q$) are also the solution of various
65: kinds of non-linear Fokker-Planck equations\cite{abe,borland-pre,richmond}
66: often used as continuous time approaches to financial systems. In this
67: manuscript one shows that a generalisation of $\ $non-extensive formalism
68: permits a good adjustment to the tick-by-tick log-return distributions for
69: Lisbon Stock Market.
70: %
71: \section{From Boltzmann-Gibbs-Shannon to anomalous distribution}
72: %
73: It is known that the consecrated Boltzmann-Gibbs-Shannon (BGS) entropic
74: form, $S=-\int p\left( x\right) \ \log p\left( x\right) \ dx$, is able to
75: provide a perfect description of an all range of phenomena with microscopic
76: spatial/temporal short-range interactions, but is not capable
77: to give proper answers when systems present microscopic spatial/temporal
78: long-range interactions \cite{tamarit,tsallis-celia} or
79: multifractal structure \cite{baldovin}. In order to deal with this glitch,
80: Tsallis presented the entropic form\cite{tsallis-sq},
81: \begin{equation}
82: S_{q}=-k \frac{1-\int \left[ p\left( x\right) \right] ^{q}\ dx}{1-q}
83: \end{equation}
84: $\left( q\;\epsilon \;\Re \right) $, which contains BGS as a
85: particular case when $q=1$. For this entropic form, the optimisation,
86: under the appropriate constrains $\int p\left( x\right) \ dx=1$ (unitary norm)
87: and $\int
88: x^{2}\frac{\ \left[ p\left( x\right) \right] ^{q}}{\int \left[ p\left(
89: x\right) \right] ^{q}dx}\ dx=\sigma _{q}^{2}$ (finite generalised variance),
90: yields the following power-law ($q$-Gaussian),
91: \begin{equation}
92: p\left( x\right) \;\sim \;\left[ 1-\beta _{q}\left( 1-q\right)
93: x^{2}\right] ^{\frac{1}{1-q}}\equiv e_{q}^{-\beta _{q}x^{2}},
94: \label{q-gauss1}
95: \end{equation}
96: which in the limit $q\rightarrow 1$ recovers the Gaussian distribution. The
97: $q$-Gaussian, $e_{q}^{-\beta _{q}x^{2}}$, can be rewritten, in a quite
98: interesting way, as the solution for the non-linear differential equation,
99: %
100: \begin{equation}
101: \frac{dp\left( x\right) }{d\left( x^{2}\right) }=-\beta _{q}\left[ p\left(
102: x\right) \right] ^{q}. \label{q-gauss2}
103: \end{equation}
104: From this equation one can obtain the (simmetrical)$\;q$-exponential
105: distribution changing $x^{2}$ by $\left| x\right| $ and the exponential
106: distribution considering also $q\rightarrow 1$.
107: %
108: For reasons that will be presented in the next section let one extends Eq. (%
109: \ref{q-gauss2}), for the $q$-exponential case, in the same way that was done in
110: previous works on protein re-association \cite{bemski} and flux of cosmic rays,
111: by introducing another index $q^{\prime }$,
112: \begin{equation}
113: \frac{dp\left( x\right) }{d\left| x\right| }=-\beta _{q^{\prime }}\left[
114: p\left( x\right) \right] ^{q^{\prime }}-\left( \beta _{q}-\beta _{q^{\prime
115: }}\right) \left[ p\left( x\right) \right] ^{q}. \label{gen1}
116: \end{equation}
117: (It is interesting to notice that if we consider $q^{\prime }=1$ and $q=2$
118: one obtains the expression used by Max Planck in 1900 that lead him to the
119: black-body radiation law \cite{max}). The solution of Eq. (\ref{gen1})
120: yields,
121: %
122: \begin{eqnarray}
123: \left| x\right| &=&\frac{1}{\beta _{q^{\prime }}}\left\{ \frac{\left[
124: p\left( x\right) \right] ^{-\left( q^{\prime }-1\right) }}{q^{\prime }-1}-%
125: \frac{\left( \beta _{q}/\beta _{q^{\prime }}\right) -1}{1+q-2q^{\prime }}%
126: \times \right. \\
127: &&\left. \times \left[ J\left( 1;q-2q^{\prime },q-q^{\prime },\left( \beta
128: _{q}/\beta _{q^{\prime }}\right) -1\right) -J\left( p\left( x\right)
129: ;q-2q^{\prime },q-q^{\prime },\left( \beta _{q}/\beta _{q^{\prime }}\right)
130: -1\right) \right] \right\} \nonumber
131: \end{eqnarray}
132: where $J\left( p\left( x\right) ;a,b,c\right) =\left[ p\left( x\right) %
133: \right] ^{1+a}F\left( \frac{1+a}{b};\frac{1+a+b}{c};\left[ p\left( x\right) %
134: \right] ^{b}c\right) $ and $F$ the hypergeometric function. For $1<q^{\prime }<q$
135: and $\beta _{q^{\prime }}\ll \beta _{q}$ three regions can be observed. The
136: first one (where $q$ governs) for $0\leq \left| x\right| \ll \left| x\right| ^{\ast }$, where $%
137: \left| x\right| ^{\ast }=\frac{1}{\left( q-1\right) \beta _{q}}$, a second
138: one (where both $q$ and $q^{\prime}$ govern) for $\left| x\right| ^{\ast }\ll \left| x\right| \ll \left| x\right|
139: ^{\ast \ast }\equiv \frac{\left[ \left( q-1\right) \beta _{q}\right] ^{\frac{%
140: q^{\prime }-1}{q-q^{\prime }}}}{\left[ \left( q^{\prime }-1\right) \beta
141: _{q^{\prime }}\right] ^{\frac{q-1}{q-q^{\prime }}}}$ and the third one (governed by $q^{\prime}$) for $%
142: \left| x\right| \gg \left| x\right| ^{\ast \ast }$.
143: %
144: \section{Application to Lisbon Stock Market}
145: %
146: Let one now considers the variable $x_{i}$ as the tick-by-tick log-return
147: obtained by, $x_{i}=\log \frac{\Pi _{i}}{\Pi _{i-1}}$,
148: with $\Pi _{i}$ representing the value of stock market index at tick $i$. Here
149: one neglets the time spacing between transactions which are, in mean, of
150: order of $10$ seconds and its possible influence in the magnitude of variations.
151: The description of log-return's distribution has been successfully done with $%
152: q$-Gaussians when the lag between elements at least of the order of minutes
153: (or greater), but fails when one deals with time series where lag is smaller
154: for which data seems to be better approached by exponential functions \cite
155: {yakovenko}. However, as can be seen in Fig. \ref{fig2}, for the case of the
156: tick-by-tick return in the Lisbon stock market the exponential function it
157: is only able to fit the central part of the distribution. By solving
158: numerically the Eq.(\ref{gen1}) with $q^{\prime }=1.076$, $q=1.534$, $\beta
159: _{q^{\prime }}=6.59\times 10^{3}$ and $\beta _{q}=7.47\times 10^{4}$ one
160: obtained the full line which presents a very good agreement, and clearly the
161: best compared to other distributions, for data from Lisbon stock market for
162: six decades in ordinate and two and a half decades in abcissa.
163: %
164: \begin{figure}[htb]
165: \begin{center}
166: \epsfig{figure=fig1.eps,width=0.7\textwidth,clip=}
167: \caption{{\protect\small Probability density function (PDF) for the tick by
168: tick log-return in the Lisbon Stock Market between 1}$^{st}${\protect\small %
169: \ Feb. 1996 to 28}$^{th}${\protect\small \ Jun. 2002 ( 4 million ticks
170: approx.). The full line represents the solution of Eq. (\ref{gen1}) and it is
171: clearly best approach to PDF from data (circles) when compared with the best Gaussian, }$q${\protect\small -Gausssian (}$q=2.51$%
172: {\protect\small ), exponential and q-exponential (}$q=1.59${\protect\small )
173: fits that are also plotted.}}
174: \label{fig2}
175: \end{center}
176: \end{figure}
177: %
178: For the values $q,q^{\prime },\beta _{q}$ and $\beta _{q^{\prime }}$ used, $%
179: \left| x\right| ^{\ast }\simeq 0.00003$ and $\left| x\right| ^{\ast \ast
180: }\simeq 0.004$. Unfortunately, the finiteness of data does not allow the
181: observation of the second crossover which occurs at a value of $x$ where it
182: is not possible to have a good statistics.
183: %
184: \section{Concluding remarks}
185: %
186: In conclusion one can state that like in biological and cosmological
187: phenomena, the superposition of power-laws which maximise Tsallis'
188: entropy can be an excellent approach in order to describe anomalous
189: distributions in finance, fact that could be a sign of a peculiar dynamics
190: (probably due to market characteristics like volume, number of traders)
191: different than the ones presented until now. Despite the fact that
192: aggregations were not analised here, existing works, where two regions can
193: also be observed in the probability density function of returns \cite
194: {pobodnik}, points that this distribution will relax firstly to the $q$%
195: -Gaussian \cite{tsallis-borland} and finally to the Gaussian distribution
196: \cite{silvio}. Although the study of tick-by-tick time series is usually
197: negleted in favour of, e.g.,1 minute lag time series, its analysis could be
198: important in the reach of a better understanding of the nature of these
199: (beautiful financial) systems, because all phenomena observed at major time
200: scales have their origin in this time scale dynamics.
201:
202: \bigskip
203: %
204: The author is thankful to Constantino Tsallis and E.P. Borges for useful comments, Data
205: Services of Euronext Lisbon for free data providing and Funda\c{c}\~{a}o
206: para a Ci\^{e}ncia e Tecnologia (Portuguese agency) for financial support.
207: Financial support from APFA4 organising committee, which made his
208: participation possible, is deeply acknowledge.
209: %
210: \begin{thebibliography}{00}
211: %
212: \bibitem{mandelbrot} B. Mandelbrot, {\it J. Business }{\bf 36, }394{\bf \ }%
213: (1963).
214: %
215: \bibitem{stanley} R.N. Mantegna and H.E. Stanley, Nature {\bf 376}, 46 (1995).
216: %
217: \bibitem{pobodnik} B. Podobnik et al., Europhys. Lett. {\bf 50}, (2000) 711.
218: %
219: \bibitem{silvio} S. M. Duarte Queir\'{o}s, Europhys. Conf. Abs. {\bf 27C}, (2003) A92; (manuscript in preparation).
220: %
221: \bibitem{ausloos} M. Ausloos and K. Ivanova, Phys. Rev. E {\bf 68}, 046122
222: (2003).
223: %
224: \bibitem{tsallis-borland} R. Osorio, L. Borland and C. Tsallis in {\it %
225: Nonextensive Entropy - Interdisciplinary Applications, }M. Gell-Mann and C.
226: Tsallis (eds.) (Oxford University Press, 2004).
227: %
228: \bibitem{abe} S. Abe and Y. Okamoto (eds.), {\it Nonextensive Statistical
229: Mechanics and Thermodynamics, Series Lectures Notes in Physics }%
230: (Springer-Verlag, Berlin, 2001).
231: %
232: \bibitem{borland-pre} L. Borland, Phys. Rev. E {\bf 57}, 6634 (1998).
233: %
234: \bibitem{richmond} P. Richmond and L. Sabatelli, cond-mat/0309404.
235: %
236: \bibitem{tsallis-celia} C. Anteneodo and C. Tsallis, J. Math. Phys. {\bf 44}, 5194 (2003).
237: %
238: \bibitem{tamarit} M.A. Montemurro, F. Tamarit and C. Anteneodo, Phys. Rev E {\bf 67}, 031106 (2003)
239: %
240: \bibitem{baldovin} F. Baldovin, E. Brigatti and C. Tsallis, Phys. Lett. A
241: {\it in press} (2003) (cond-mat/0302559).
242: %
243: \bibitem{tsallis-sq} C. Tsallis, J. Stat. Phys. {\bf 52} , 479 (1988). A
244: regularly an updated bibliograhy on the subject is avaible at {\tt %
245: http://tsallis.cat.cbpf.br/biblio.htm}.
246: %
247: \bibitem{yakovenko} A.C. Silva and V. Yakovenko, Physica A {\bf 324}, 303
248: (2003).
249: %
250: \bibitem{bemski} C. Tsallis, G. Bemski and R.S. Mendes, Phys. Lett. A {\bf 257}%
251: , 93 (1999).
252: %
253: \bibitem{ernesto} C. Tsallis, J.C. Anjos and E.P. Borges, Phys. Lett. A {\bf %
254: 310}, 372 (2003).
255: %
256: \bibitem{max} D. ter Haar and S.G. Brush (eds.), Plancks Original Papers in
257: Quantum Physics (Taylor \& Francis, London, 1972).
258: %
259: \end{thebibliography}
260: %
261: \end{document}
262: