Physica A 274 (1999) 216221

www.elsevier.com/locate/physa

Applications of statistical mechanics to nance

Rosario N. Mantegnaa;b; , Zoltan Palagyic , H. Eugene Stanleyd;e
a Istituto

Nazionale per la Fisica della Materia, Unita di Palermo, Palermo, I-90128, Italy
di Energetica ed Applicazioni di Fisica, Universita di Palermo, Applicazioni di Fisica,
Viale delle Scienze, Palermo, I-90128, Italy
c Department of Mathematics, Budapest University of Economic Sciences, F
ovam ter 8,
1093 Budapest, Hungary
d Center for Polymer Studies, Boston, MA 02215, USA
e Department of Physics, Boston University, Boston, MA 02215, USA

b Dipartimento

Abstract
We discuss some apparently universal aspects observed in the empirical analysis of stock
price dynamics in nancial markets. Specically we consider (i) the empirical behavior of the
return probability density function and (ii) the content of economic information in nancial time
c 1999 Elsevier Science B.V. All rights reserved.
series.
PACS: 89.90.+n
Keywords: Econophysics

1. Introduction
The analyses and modeling of nancial markets started in 1900 with the pioneering
work of the French mathematician Bachelier [1]. Since the 1950s, the analysis and
modeling of nancial markets have become an important research area of economics
and nancial mathematics [2]. The researches pursued have been very successful, and
nowadays a robust theoretical framework characterizes these disciplines [3 6]. In parallel to these studies, starting from the 1990s a group of physicists became interested
in the analysis and modeling of nancial markets by using tools and paradigms of
their own discipline (for an overview, consider, for example, [711]). The interest of
physicists in such systems is directly related to the fact that predictability has assumed
a meaning in physics over the years, which is quite dierent from the one originally
associated with the predictability of, for example, a Newtonian linear system. The

Corresponding author. Fax: +39-091-236306.

E-mail address: [email protected] (R.N. Mantegna)

c 1999 Elsevier Science B.V. All rights reserved.

0378-4371/99/$ - see front matter
PII: S 0 3 7 8 - 4 3 7 1 ( 9 9 ) 0 0 3 9 5 - 7

R.N. Mantegna et al. / Physica A 274 (1999) 216221

217

degree of predictability of physics systems is nowadays known to be essentially limited in nonlinear and complex systems. This makes the physical prediction less strong,
but on the other hand the area of research covered by physical investigations and of
its application may increase [12].
The research approach of physicists to nancial modeling aims to be complementary
to the ones of nancial mathematicians and economists. The main goals are to (i)
contribute to a better understanding and modeling of nancial markets and (ii) promote
the use of physical concepts and expertise in the multidisciplinary approach to risk
management.
In this communication, we review some results of our work on (i) the statistical
properties of returns in nancial markets and (ii) the characterization of the simultaneous dynamics of stock prices in a nancial market. Specically, we recall statistical
properties of price changes empirically observed in dierent markets worldwide. In
particular, we discuss studies performed on the New York stock exchange [13,14], the
Milan stock exchange [15] and the Budapest stock exchange [16]. The communication
is organized as follows, we rst discuss the results of empirical analyses performed in
these dierent markets and then we address the problem concerning the presence of
economic information in a nancial time series.

2. Statistical properties of price dynamics

The knowledge of the statistical properties of price dynamics in nancial markets is
fundamental. It is necessary for any theoretical modeling aiming to obtain a rational
price for a derivative product issued on it [17] and it is the starting point of any
valuation of the risk associated with a nancial position [18]. Moreover, it is needed
in any eort aiming to model the system. In spite of this importance, the modeling of
such a variable is not yet conclusive. Several models exist which show partial successes
and unavoidable limitations. In this research, the approach of physicists maintain the
specicity of their discipline, namely to develop and modify models by taking into
account the results of empirical analysis.
Several models have been proposed and we will not review them here. Here, we
wish to focus only on the aspects which are universally observed in various stock
price and index price dynamics.
2.1. Short- and long-range correlations
In any nancial market either well established and highly active as the New
York stock exchange, emerging as the Budapest stock exchange, or regional as
the Milan stock exchange the autocorrelation function of returns is a monotonic
decreasing function with a very short correlation time. High-frequency data analyses
have shown that correlation times can be as short as a few minutes in highly traded
stocks or indices [14,19]. A fast decaying autocorrelation function is also observed in

218

R.N. Mantegna et al. / Physica A 274 (1999) 216221

the empirical analysis of data recorded transaction by transaction. By using as time

index the number of transactions emanating from a selected origin, a time memory as
short as a few transactions has been detected in the dynamics of most traded stocks of
the Budapest emerging nancial market [16].
The short-range memory between returns is directly related to the necessity of
absence of continuous arbitrage opportunities in ecient nancial markets. In other
words, if correlation were present between returns (and then between price changes)
this would allow one to devise trading strategies that would provide a net gain continuously and without risk. The continuous search for and the exploitation of arbitrage
opportunities from traders focused on this kind of activity drastically diminish the
redundancy in the time series of price changes. Another mechanism reducing the redundancy of stock price time series is related to the presence of the so-called noise
traders. With their action, noise traders add into the time series of stock price information, which is unrelated to the economic information decreasing the degree of
redundancy of the price changes time series.
It is worth pointing out that not all the economic information present in stock price
time series disappears due to these mechanisms. Indeed the redundancy that needs to
be eliminated concerns only price change and not any of its nonlinear functions [20].
The absence of time correlation between returns does not mean that returns are
identically distributed over time. In fact, dierent authors have observed that nonlinear
functions of return such as the absolute value or the square are correlated over a time
scale much longer than a trading day. Moreover, the functional form of this correlation
seems to be power-law up to at least 20 trading days approximately [19,2126].
A nal observation concerns the degree of stationary behavior of the stock price
dynamics. Empirical analysis shows that returns are not strictly sense stationary stochastic processes. Indeed the volatility (standard deviation of returns) is itself a stochastic
process. Although a general proof is still lacking, empirical analyses performed on
nancial data of dierent nancial markets suggest that the stochastic process is locally
non-stationary but asymptotically stationary. By asymptotically stationary we mean that
the probability density function (pdf) of returns measured over a wide time interval
exists and it is uniquely dened. A paradigmatic example of simple stochastic processes
which are locally non-stationary but asymptotically stationary is provided by ARCH
[27] and GARCH [28] processes.
2.2. The distribution of returns
The pdf of returns shows some universal aspects. By universal aspects we mean
that they are observed in dierent nancial markets at dierent periods of time provided
that a suciently long time period is used in the empirical analysis. The rst of these
universal or stylized facts is the leptokurtic nature of the pdf. A leptokurtic pdf
characterizes a stochastic process having small changes and very large changes more
frequently than in the case of Gaussian distributed changes. Leptokurtic pdfs have
been observed in stocks and indices time series by analyzing both high-frequency and

R.N. Mantegna et al. / Physica A 274 (1999) 216221

219

daily data. Thanks to the recent availability of transaction-by-transaction data, empirical

analyses on a transaction time scale have also been performed. One of these studies
performed by analyzing stock prices in the Budapest stock exchange shows that return
pdf is leptokurtic down to a transaction time scale [16].
The origin of the observed leptokurtosis is still debated. There are several models trying to explain it. Just to cite (rather arbitrarily) a few of them: (i) a model
of Levy stable stochastic process [29]; (ii) a model assuming that the non-Gaussian
behavior occurs as a result of the uneven activity during market hours [30]; (iii) a
model where a geometric diusive behavior is superimposed on Poissonian jumps [31];
(iv) a quasi-stable stochastic process with nite variance [32]; and (v) a stochastic
process with rare events described by a power-law exponent not falling into the Levy
regime [3335]. The above processes are characterized by nite or innite moments.
In an attempt to nd the stochastic process that best describes stock price dynamics, it
is important to try to preliminarily conclude about the niteness or inniteness of the
second moment.
The above answer is not simply obtained [38] and careful empirical analyses must be
performed to reach a reliable conclusion. It is our opinion that an impressive amount of
empirical evidence has been recently found supporting the conclusion that the second
moment of the return pdf is nite [13,3337]. This conclusion has a deep consequence on the stability of the return pdf. The niteness of the second moment and the
independence of successive returns imply that the central limit theorem asymptotically
applies. Hence, the form expected for the return pdf must be Gaussian for very long
time horizons. We then have two regions at short time horizons we observe leptokurtic distributions whereas at long time horizons we expect a Gaussian distribution.
A complete characterization of the stochastic process needs an investigation performed
at dierent time horizons. During this kind of analyses, non-Gaussian scaling and its
breakdown has been detected [13,14].

3. Collective dynamics
In the previous sections we have seen that universal facts suggest that the stock
price change dynamics in nancial markets is well described by an unpredictable time
series. However, this does not imply that the stochastic dynamics of stock price time
series is a random walk with independent identically distributed increments. Indeed
the stochastic process is much more complex than a customary random walk.
One key question in the analysis and modeling of a nancial market concerns the
independence of the price time series of dierent stocks traded simultaneously in the
same market. The presence of cross-correlations between pairs of stocks has been
known since a long time and it is one of the basic assumptions of the theory for
the selection of the most ecient portfolio of stocks [39]. Recently, physicists have
also started to investigate theoretically and empirically the presence of such crosscorrelations.

220

R.N. Mantegna et al. / Physica A 274 (1999) 216221

It has been found that a meaningful economic taxonomy may be obtained by starting
from the information stored in the time series of stock price only. This has been
achieved by assuming that a metric distance can be dened between the synchronous
time evolution of a set of stocks traded in a nancial market and under the essential
ansatz that the subdominant ultrametric associated with the selected metric distance is
controlled by the most important economic information stored in the time evolution
dynamics [40].
Another kind of study is devoted to the detection of the statistical properties of eigenvalues and eigenvectors of the covariance matrix of n stocks simultaneously traded.
Also, with this approach the hypothesis that the dynamics of stock price in a portfolio
of n stocks is described by independent random walks is falsied [41 43]. Moreover,
information about the number of terms controlling eigenvectors can be detected.
The observation of the presence of a certain degree of statistical synchrony in the
stock price dynamics suggests the following conclusion. Consideration of the time
evolution of only a single stock price could be insucient to reach a complete modeling
of all essential aspects of a nancial market.

4. Summary
This communication brie y discusses some of the stylized universal facts that are
observed in nancial markets and are considered robust by several researchers working
in the eld. Starting from these results, one can devise studies trying to enrich and
expand this knowledge to provide theoreticians and computer scientists the empirical
facts that need to be explained by their models progressively proposed. The ultimate
goal is to contribute to the search for the best model describing a nancial market,
one of the most intriguing complex systems.

Acknowledgements
R.N.M. wishes to thank INFM and MURST for nancial support. Z.P. thanks Cashline Broker Securities Ltd. for nancial support, and H.E.S. thanks the NSF.
References
[1] L. Bachelier, Theorie de la speculation [Ph.D. Thesis in mathematics], Ann. Sci. de lEcole Normale
Superieure III-17 (1900) 2186.
[2] P.H. Cootner (Ed.), The Random Character of Stock Market Prices, MIT Press, Cambridge, MA, 1964.
[3] J.E. Ingersoll Jr., Theory of Financial Decision Making, Rowman & Littleeld, Savage, MD, 1987.
[4] R.C. Merton, Continuous-Time Finance, Blackwell, Cambridge, MA, 1990.
[5] I. Karatzas, S.E. Shreve, Brownian Motion and Stochastic Calculus, 2nd Edition, Springer, Berlin, 1991.
[6] J.Y. Campbell, A.W. Lo, A.C. MacKinlay, The Econometrics of Financial Markets, Princeton University
Press, Princeton, 1997.

R.N. Mantegna et al. / Physica A 274 (1999) 216221

221

[7] J. Kertesz, I. Kondor (Eds.), Econophysics: Proceedings of the Budapest Workshop, Kluwer Academic
Press, Dordrecht (in press).
[8] R.N. Mantegna (Ed.), Proceedings of the International Workshop on Econophysics and Statistical
Finance, Physica A (special issue) 269 (1999).
[9] K. Lauritsen, P. AlstrHm, J.-P. Bouchaud (Eds.), Proceedings of the International Conference on
Application of Physics in Financial Analysis, Int. J. Theor. Appl. Finance (special issue) (in press).
[10] J.-P. Bouchaud, M. Potters, Theories des Risques Financiers, Alea-Saclay, Paris, 1997.
[11] R.N. Mantegna, H.E. Stanley, An Introduction to Econophysics: Correlations and Complexity in Finance,
Cambridge University Press, Cambridge, 1999.
[12] G. Parisi, Physica A 263 (1999) 557.
[13] R.N. Mantegna, H.E. Stanley, Nature 376 (1995) 46.
[14] R.N. Mantegna, H.E. Stanley, Nature 383 (1996) 587.
[15] R.N. Mantegna, Physica A 179 (1991) 232.
[16] Z. Palagyi, R.N. Mantegna, Physica A 269 (1999) 132.
[17] J.C. Hull, Options, Futures, and Other Derivatives, 3rd Edition, Prentice-Hall, Upper Saddle River, NJ,
1997.
[18] D. Due, J. Pan, J. Derivatives, spring issue (1997) 7.
[19] Y. Liu, P. Gopikrishnan, P. Cizeau, M. Meyer, C.-K. Peng, H.E. Stanley, Phys. Rev. E 60 (1999) 1390.
[20] R. Baviera, M. Pasquini, M. Serva, D. Vergni, A. Vulpiani, Eciency in foreign exchange markets,
cond-mat 9901225.
[21] M.M. Dacorogna, U.A. Muller, R.J. Nagler, R.B. Olsen, O.V. Pictet, J. Int. Money Finance 12 (1993)
413.
[22] R. Cont, M. Potters, J.-P. Bouchaud, in: B. Dubrulle, F. Graner, D. Sornette (Eds.), Scale Invariance
and Beyond, Springer, Berlin, 1997.
[23] P. Cizeau, Y. Liu, M. Meyer, C.-K. Peng, H.E. Stanley, Physica A 245 (1997) 441.
[24] Y. Liu, P. Cizeau, M. Meyer, C.-K. Peng, H.E. Stanley, Physica A 245 (1997) 437.
[25] M. Pasquini, M. Serva, Physica A 269 (1999) 140.
[26] M. Raberto, E. Scalas, G. Cuniberti, M. Riani, Physica A 269 (1999) 148.
[27] R.F. Engle, Econometrica 50 (1982) 987.
[28] T. Bollerslev, J. Econometrics 31 (1986) 307.
[29] B.B. Mandelbrot, J. Bus. 36 (1963) 394.
[30] P.K. Clark, Econometrica 41 (1973) 135.
[31] R.C. Merton, J. Financial Econom. 3 (1976) 125.
[32] R.N. Mantegna, H.E. Stanley, Phys. Rev. Lett. 73 (1994) 2946.
[33] T. Lux, Appl. Financial Econom. 6 (1996) 463.
[34] T. Lux, J. Econ. Behav. Organ. 33 (1998) 143.
[35] P. Gopikrishnan, M. Meyer, L.A.N. Amaral, H.E. Stanley, Eur. Phys. J. B 3 (1998) 139.
[36] P. Gopikrishnan, V. Plerou, L.A.N. Amaral, M. Meyer, H.E. Stanley, Scaling of the distributions of
uctuations of nancial market indices, Phys. Rev. E 60 (1999) in press, cond-mat 9905305.
[37] V. Plerou, P. Gopikrishnan, L.A.N. Amaral, M. Meyer, H.E. Stanley, Scaling of the distribution of price
uctuations of individual companies, Phys. Rev. E, cond-mat 9907161 (submitted for publication).
[38] A.L. Tucker, J. Bus. Econom. Stat. 10 (1992) 73.
[39] H. Markowitz, Portfolio Selection: Ecient Diversication of Investment, Wiley, New York, 1959.
[40] R.N. Mantegna, Eur. Phys. J. B 11 (1999) 193.
[41] S. Galluccio, J.-P. Bouchaud, M. Potters, Physica A 259 (1998) 449.
[42] L. Laloux, P. Cizeau, J.-P. Bouchaud, M. Potters, Phys. Rev. Lett. 83 (1999) 1468.
[43] V. Plerou, P. Gopikrishnan, B. Rosenow, L.A.N. Amaral, H.E. Stanley, Phys. Rev. Lett. 83 (1999)
1471.