seminar similarities a d differe ces betwee turbule ce a...
TRANSCRIPT
University of Ljubljana
Faculty of Mathematics and Physics
Jadranska 19, Ljubljana
Seminar
SIMILARITIES A�D DIFFERE�CES
BETWEE� TURBULE�CE A�D FI�A�CIAL
MARKETS
April 2009
Author: Matjaž Ivančič
Mentor: doc. dr. Primož Ziherl
Abstract: The comparative analysis of the statistical properties on financial markets and the
velocity of the air in a fully turbulent state shows that an interaction between economics and
statistical physics may be useful - i.e., it may be fruitful to pursue analogies and differences
between two various stochastic models developed in economics and the approaches used in
turbulence theory. The exchange of concepts, models and techniques of data analysis offers an
opportunity to characterize qualitatively and quantitatively analogies and differences between
these two stochastic processes.
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
2
Introduction
One of the criticisms of physicists working with economic systems is that this kind of activity
cannot be a branch of physics because the equation of motion of the process is unknown. But
if this criterion were so important, several current research fields in physics would be
disqualified, such as the modelling of friction and many studies in the area of granular matter.
Moreover, a number of problems in physics that are described by a well-defined equation,
such as turbulence, are not analytically solvable even with sophisticated mathematical and
physical tools [1].
On a qualitative level, turbulence and financial markets are attractively similar. The complex
statistical behaviour of velocity increments on a certain length scale in turbulent flows is
assumed to be due to a cascading process. The energy, which is fed into the system on large
scales, is continuously transported towards smaller scales due to the inherent instability of
vortices of a given scale towards perturbations on smaller scales. Finally, the energy is
dissipated at the smallest scale. A similar mechanism has been proposed for financial markets,
where the energy cascade was replaced by a flow of information. Initially, the assumption of a
cascading process in financial markets was based on similarities in the empirical description
of the probability density functions (PDFs) of price and velocity increments. The analogy
between turbulence and finance (Table I) has inspired many further studies, but has also been
criticized for being too superficial [1, 2].
Hydrodynamic Turbulence Financial Markets
Energy Information
Spatial distance Time delay
Intermittency (laminar periods interrupted by
turbulent burst)
Volatility clustering
Energy cascade in space hierarchy Information cascade in time hierarchy
Table I) Postulated correspondence between fully developed three-dimensional turbulence and financial markets [2]
In this seminar, we discuss the fully developed turbulence hydrodynamics in fluid flow in
comparison with stochastic modelling of stock prices. Our aim is to show that cross-
fertilization between the turbulence and the economic index might be useful, but we shall find
that the formal correspondence between turbulence and financial systems is not supported by
qualitative calculations.
Financial markets
A financial market is a mechanism that allows people to easily buy and sell financial
securities, commodities and other replaceable items of value at low transaction costs and at
prices that reflect the efficient-market hypothesis. Financial markets have evolved
significantly over several hundred years and are undergoing constant innovation to improve
liquidity.
University of Ljubljana
Faculty of Mathematics and Physics
Both general markets and specialized markets
buyers and sellers in one place, thus making it easier for them to find each other. An economy
which relies primarily on interactions between buyers and sellers to allocate resources is
known as a market economy in contrast either to a command economy or to a non
economy such as a gift economy. The financial markets can be divided into different subtypes
[3]:
• Foreign exchange markets
• Commodity markets, w
• Capital markets, which consist of:
o Stock markets,
common stock, and enable the subsequent trading thereof.
o Bond markets, which provide financing through t
enable the subsequent trading thereof.
• Money markets, which provide short term debt financing and investment.
• Derivatives markets, which provide instruments for the management of financial risk.
o Futures markets
products at some future date; see also forward market.
• Insurance markets, which facilitate the redistribution of various risks.
If we look at different economic indexes [Fig. 1] we can find that different types of mar
have a similar complex price dynamic. Although, we can presume that a study of one index
will give us a general view of all financial markets.
The availability of high-frequency data for financial markets has made it possible to study
market dynamics on timescales of less than a day. For stock market and foreign exchange
rates have shown that there is a net flow of information
Fig. 1/a) Charts derived from Ref. [4] for a 2 year period sh
exchange rate (foreign exchange market).
Turbulence in Financial M
April 2008
3
Both general markets and specialized markets exist. Markets work by placing many interested
buyers and sellers in one place, thus making it easier for them to find each other. An economy
which relies primarily on interactions between buyers and sellers to allocate resources is
my in contrast either to a command economy or to a non
economy such as a gift economy. The financial markets can be divided into different subtypes
Foreign exchange markets, which facilitate the trading of foreign exchange
, which facilitate the trading of commodities.
which consist of:
which provide financing through the issuance of shares or
common stock, and enable the subsequent trading thereof.
, which provide financing through the issuance of bonds, and
enable the subsequent trading thereof.
, which provide short term debt financing and investment.
, which provide instruments for the management of financial risk.
Futures markets, which provide standardised forward contracts for trading
products at some future date; see also forward market.
, which facilitate the redistribution of various risks.
If we look at different economic indexes [Fig. 1] we can find that different types of mar
have a similar complex price dynamic. Although, we can presume that a study of one index
will give us a general view of all financial markets.
frequency data for financial markets has made it possible to study
on timescales of less than a day. For stock market and foreign exchange
rates have shown that there is a net flow of information from long to short timescales [3
] for a 2 year period showing a moderate complexity of the chart
).
Turbulence in Financial Markets
Matjaž Ivančič
exist. Markets work by placing many interested
buyers and sellers in one place, thus making it easier for them to find each other. An economy
which relies primarily on interactions between buyers and sellers to allocate resources is
my in contrast either to a command economy or to a non-market
economy such as a gift economy. The financial markets can be divided into different subtypes
he trading of foreign exchange.
which provide financing through the issuance of shares or
he issuance of bonds, and
, which provide short term debt financing and investment.
, which provide instruments for the management of financial risk.
dardised forward contracts for trading
, which facilitate the redistribution of various risks.
If we look at different economic indexes [Fig. 1] we can find that different types of markets
have a similar complex price dynamic. Although, we can presume that a study of one index
frequency data for financial markets has made it possible to study
on timescales of less than a day. For stock market and foreign exchange
from long to short timescales [3, 4, 5].
chart dynamics of USD-CHF
University of Ljubljana
Faculty of Mathematics and Physics
Fig. 2/b) Charts derived from Ref. [4] for a 2 year period showing a moderate complexity of the price dynamics
(commodity market).
Fig. 3/c) Charts derived from Ref. [4] for a 2 year period showing a moderate complexity of the
Apple Inc. (capital market). Due to similar influences and volatility we can presume that the whole
described by a single equation of motion.
Turbulence
Turbulence is a well-defined physical problem which
physics. Among the approaches that have been
arguments based on dimensional analysis, statistical modelling
Consider a simple system that exhibits turbulence, a fluid of kinematic viscosity
with velocity � in a pipe of diameter
complexity of this flowing fluid is the Reynolds number,
When �� reaches a particular threshold value, the complexities of the fluid ex
suddenly becomes turbulent (Fig. 2
Turbulence in Financial M
April 2008
4
] for a 2 year period showing a moderate complexity of the price dynamics
] for a 2 year period showing a moderate complexity of the
Apple Inc. (capital market). Due to similar influences and volatility we can presume that the whole
equation of motion.
defined physical problem which remains one of the great challenges in
physics. Among the approaches that have been proposed are analytical approaches, scaling
uments based on dimensional analysis, statistical modelling and numerical simulations [
Consider a simple system that exhibits turbulence, a fluid of kinematic viscosity
in a pipe of diameter �. The control parameter whose value determines the
complexity of this flowing fluid is the Reynolds number,
�� � ��� �
reaches a particular threshold value, the complexities of the fluid ex
(Fig. 2) [7].
Turbulence in Financial Markets
Matjaž Ivančič
] for a 2 year period showing a moderate complexity of the price dynamics of gold
] for a 2 year period showing a moderate complexity of the shares price dynamics of
Apple Inc. (capital market). Due to similar influences and volatility we can presume that the whole financial market is
one of the great challenges in
are analytical approaches, scaling
and numerical simulations [5].
Consider a simple system that exhibits turbulence, a fluid of kinematic viscosity � flowing
. The control parameter whose value determines the
(1)
reaches a particular threshold value, the complexities of the fluid explode as it
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
5
Fig. 4) Solution of the 2D vorticity equation proposed by P. Bartosz from McMaster University, Hamilton, Canada. As we
can see above, increasing the fluid velocity causes a major complexity in the motion of the fluid past the obstacle. In a) where
the Re = 10 the flow is laminar. As the Re increases turbulence occurs. From b) (Re = 75) through c) (Re = 100) to d) (Re =
150) the flow becomes turbulent and the von Karman vortex street appears. For Re = 250 (e) and 500 (f) where the turbulent
cascade is clearly seen [8].
The equations describing the time evolution of an incompressible fluid have been known
since Navier’s work published in 1823 [4], which led to the Navier-Stokes equation,
�� ��, � + ���, � ∙ ∇���, � � −∇� + �∇���, �, (2)
Usually one assumes that the fluid is incompressible:
∇ ∙ ��, � � 0� (3)
Here ��, � is the velocity at position � and time and � is the pressure. The Navier-Stokes
equation completely characterizes the fully developed turbulence, a technical term indicating
turbulence at high Reynolds number. For very large values of Re no analytical or even
numerical solutions of Eqs. (1) and (2) are known.
a)
b)
c)
d)
e)
f)
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
6
In 1942, a breakthrough in the description of fully developed turbulence was achieved by
Kolmogorov [1]. He showed that in the limit of infinite Reynolds numbers, in the inertial
range (the range of length scales over which energy is transferred and dissipation due to
molecular viscosity is negligible), the mean square velocity increment
��Δ������� � ����� + �� − ������� (4)
behaves approximately as
���������~�� �� (5)
where � is the distance of two points where the velocity is measured. The distances are smaller
than the overall dimension where the fluid turbulent behaviour occurs and larger than the
typical length below the kinetic energy is dissipated into heat.
Although the Kolmogorov theory describes well the second-order of velocity increment
�������2� and provides the exact relation for the third-order �������3� moments observed in
experiments, its fails to describe the higher moments and the intermittent behaviour of
velocity increments. In the experimental studies of fully developed turbulence,
experimentalists usually measure the velocity as a function of time [1].
Parallel analysis of price dynamics and fluid velocity
Thanks to the amounts of data from stock markets it is possible to make a good evaluation of
the price dynamics and compare it to measurements of a turbulent fluid. Those data can
provide us with a better comprehension of the speculative price dynamics similar to the
energy cascade of a hydrodynamic turbulence.
Several statistical techniques (e.g. the measure of the probability density function, the
measure of the spectral density, etc.) commonly used in the study of the stochastic processes
have been used for a long time in turbulence [6]. Moreover, recently there have been many
attempts to identify stochastic processes whose statistical properties are close to those
observed in turbulence. In their paper, Mantegna and Stanley [6] report analogies and
differences between the quantitative measures of fluctuations in an economic index and the
fluctuations in velocity of a fluid in a fully turbulent state. They observe non-Gaussian
statistics and intermittency for both processes but the time evolution of the second moment
and the shape of the probability density functions stock market dynamics and fluid turbulence
are not the same.
The economic data set studied consists of all 1447514 records of the S&P 500 cash index
recorded during the period January 1984 December 1989 [7]. In this analysis, the "trading
time" extends from the opening until the closing of the day, and then continues with the
opening of the next trading day. The time intervals between successive records are not fixed:
the average value between successive records is close to 1 min during 1984 and 1985 and
close to 15 s during the period 1986-1989. From this database is selected the complete set of
non-overlapping records separated by a time interval ∆ ± $Δ (where $ is the tolerance and
always less than 0.035 s). We denote the value of the S&P 500 as %�� [Fig. 3(a)], and the
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
7
successive variations of the S&P index is &'(�� � %�� − %� − Δ� [Fig. 3(b)] [6]. The
complexity of the price motion clearly shows a strange behaviour with smaller and bigger
variations around the uptrend.
Fig. 5) a) Time evolution of the S&P 500, sampled with a time resolution ∆) � *+, over the period January 1984 –
December 1989, where we can notice the big leap slightly above the 6000 trading time. b) Hourly variations of the S&P 500
index in the 6-year period January 1984 – December 1989, which emphasise the big leap at t = 6000 h [6].
To make an evaluated comparison between the financial markets and the fluid turbulence
accurate data of fully developed turbulent flow is needed.
The turbulence data were provided by P. Kaylashnat and coworkers [9]. Measurements were
made in the atmospheric surface layer about 6m above a wheat canopy in the Connecticut
Agricultural Research Station. The Reynolds number ,- expressed in the Taylor microscale,
where the dissipation begins to affect the eddies, was around 1500. Velocity fluctuations were
measured using the standard hot-wire velocimeter operated in the constant temperature mode
on a DISA 55M01 anemometer. The file consists of 130000 velocity records ��� digitized
and linearized before processing [Fig. 4(a)]. The associated velocity differences .'(�� ���� − �� − Δ� are shown in Fig. 4(b).
Fig. 6) a) Time evolution of the wind velocity recorded in the atmosphere at very high Reynolds number; the Taylor
microscale Reynolds number around 1500. The time units are given in arbitrary units. (b) Velocity differences of the time
series given in (a) where we can notice the same kind of leaps that in Fig. 3 (b) [9].
%��
&��
trading time (hours) trading time (hours)
���
.��
time (hours) time (hours)
400
300
200
100
0 2000 4000 6000 8000 10000
20
10
0
-10
-20
0 2000 4000 6000 8000 10000
20
10
0
-10
-20
6
3
0
-3
-6
0 5000 100000 0 5000 100000
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
8
We focus attention on the dynamics of the index variation &'(�� and on the dynamics of the
velocity difference .'(��, and denote by ��&� and ��.� the associated probability density
functions (PDFs).
By measuring the time dependence of the standard deviations 01�� and 02�� of ��&� and
��.�, we find that:
• In the case of the S&P 500 index variations [Fig. 5(a)] the time dependence of the
standard deviation, when Δ > ~ 15 min fits well the behaviour [5]
0&�Δ� ∝ Δ0�53. (6)
The exponent is close to the typical value of 0.5 observed in random processes with
independent increments.
• The velocity difference of the fully turbulent fluid shows a time dependence of the
standard deviation, fitting the behaviour [Fig. 5(b)] [6]
0.�Δ� ∝ Δ0�33 (7)
which is observed in short-time anti-correlated random processes, meaning the more
positive differences occur, the more probable is a negative difference in direction.
Fig. 7) a) Log-log plot of the standard deviation :;�<)� of the probability distribution =�;� characterizing the increments ;<)�)� as a function of <) for the S&P 500 time series. After a time intervaI of superdiffusive behaviour, when the standard
deviation is not yet linear �> < <) ≤ *A BCD� a diffusive behavior close to the one expected for a random process with
independent identically-distributed increments is observed; the measured diffusion exponent >� AE is close to the theoretical
value * F� characteristic for normal diffusion. (b) Standard deviation :G�<)� of the probability distribution =�G�
characterizing the velocity increments G<)�)� plotted double logarithmically as a function of <) for the velocity difference
time series in turbulence. After a time interval of superdiffusive behaviour �> < <) ≤ *>�, a diffusive behaviour close to the
one expected for a fluid in the inertial range is observed (the measured diffusion exponent 0.33 is close to the theoretical
value * E� predicted by Kolmogorov) [6].
Similar conclusions are reached if we measure the spectral density of the time series %�� and
���. Economic data [Fig. 6(a)] have the spectral density typical of a Brownian motion,
H�I� ∝ IJ�. For turbulence data [Fig. 6(b)] the spectral density shows a wide inertial range
(the range of length scales over which energy is transferred and dissipation due to molecular
viscosity is negligible) as H�I� ∝ IJK �� [6].
0 1�∆�
0 2�∆�
trading time (minutes) time (a.u.)
slope = 0.33 slope = 0.53
10�
10L
10M
10JL
10J�
10M 10L 10� 10� 10N
10�
10L
10M
10JL
10J�
10M 10L 10� 10� 10N
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
9
Fig. 8) a) Spectral density of the S&P 500 time series. The O−F power-law behaviour expected for a random process with
increments independent and identically distributed is observed over a frequency interval of more than 4 orders of magnitude.
b) Spectral density of the velocity time series. The O−A E� inertial range (low frequency) and the dissipative range (high
frequency) are clearly observed [6].
A different kind of analysis of the PDFs ��&� and ��.� turns out to be quite powerful for the
description of experimental results. We analyze the point of each PDF that is least affected by
the noise introduced by the finiteness of the data set – ��0� varies with Δ [6].
Figs. 7(a) and (b) show log-log plots of ��& � 0� and ��. � 0� as functions of the time
interval Δ between successive observations. The deviation from a Gaussian process is shown
by plotting on the same figure the value of �P�0� determined from the measured values of
0�Δ� by assuming that the process is Gaussian – using the equation
�Q�0� � 1R2S0�Δ� � (8)
The clear difference between ��0� and �P�0� seen in Fig. 7 shows that both PDFs have a non-
Gaussian distribution and that the detailed shapes and the scaling properties of the two PDFs
are different. Recent empirical studies of stock market indices show that financial markets can
be described by a Lévy-stable distribution commonly used in stochastic processes�11, 6�.
Sp�ctra
l D�nsit
y�dB�
Sp�ctra
l D�nsit
y�dB�
10 logLM �I� 10 logLM �I�
experimental data
slope = - 1.62
experimental data
slope = - 1.98
40
20
0
-20
-40
-60
-80
-100 -60 -50 -40 -30 -20 -10 0
40
20
0
-20
-40
-60
-80
-100 -50 -40 -30 -20 -10 0
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
10
Fig. 9) a) S&P 500 index. Probability of return to the origin: =�; � >� (circles) and =c�; � >� (filled squares) [see Eq. (8)]
as functions of the time sampling interval <). The two measured quantities differ in the full interval implying that the profile
of the PDF must be non-Gaussian. A power-law behaviour is observed for the entire time interval spanning three orders of
magnitude. The slope of the best linear fit is −>� d*e ± >� >FA [6]. b) Velocity of the fully turbulent fluid. Probability of
return to the origin: =�>� (circles) and =c�>� (filled squares) (see Eq. (8)) as functions of the time sampling interval <).
Again, the two measured quantities differ across the full interval, implying that the profile of the PDF must be non-Gaussian.
However in this case, a single sealing power-law behaviour does not exist for the entire time interval spanning three orders of
magnitude. The slope of the best linear fit (which is of quite poor quality) is −>� Afe ± >� ** [6].
The PDF of any stochastic process can be defined by the Fourier transform of its
characteristic function:
��g� � 12S h I��i−jgk+∞
−∞ � (9)
The characteristic function for the Lévy-stable distribution is
lnI�� � jm − noop q1 + jr s oot tg upS
2 vw � (10)
The characteristic function tells us that Lévy-stable distributions have four parameters: p, r, n
and m. The location parameter is m, and if p is greater than 1, m is equal to the expectation of
mean of the distribution. The scale parameter is n, while the parameter r is an index of
skewness, which can take any value in the interval −1 ≤ r ≤ 1. When r � 0 the distribution
is symmetric. When r > 0 (and 1 < p < 2), the distribution is skewed right (i.e., has a long
tail to the right) and the degree of skewness increases in the interval 0 < r ≤ 1 as r
approaches 1. Similarly, when r < 0 (and 1 < p < 2), the distribution is skewed left, with
the degree of skewness increasing in the interval −1 ≤ r < 0 as r approaches −1 [11].
Of the four parameters of a Lévy-stable distribution, the characteristic exponent p is the most
important for the purpose of comparing the fit. The character exponent p determines the
height of the extreme tails of the distribution, and can take any value in the interval 0 < p ≤2. When p � 2, the relevant Lévy-stable distribution is the normal (Gaussian) distribution.
The total probability in the extreme tails increases as p moves away from 2 and toward 0.
A scaling compatible with a Lévy-stable process is observed for economic data [Fig. 7(a)] and
indeed a Lévy distribution reproduces quite well the central part of the distribution of the S&P
10�
10L
10M
10JL
∆�a� u� � ∆�minut�s�
��0�
��0�
P(0)
Pg(0)
slope = - 0.59
P(0)
Pg(0)
slope = - 0.71
10�
10L
10M
10JL
10M 10L 10� 10� 10M 10L 10� 10�
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
11
500 index variations [Fig. 8(a)] [4]. The Lévy-stable modelling with p � 1�40, r � 0,
n � 0�00375 and m � 0, obtained from the best fit of the probability of return to the origin
data, describes the data well over a three order of magnitude time interval (ranging from 1 to
1000 minutes). The tails deviate from the Lévy profile when & ≥ 0�3, ensuring a finite
variance to the stochastic process. The deviantion show us have a more complex distribution
broken at & � 0�3 [10].
No similar scaling exists for turbulence data over a wide time interval [Fig. 7(b)] [5]. By
using the measured values of ��0� and 02 and hypothesizing a stretched exponential PDF, it
is possible to describe quite well the experimental PDF of the velocity difference with a
stretched exponential distribution
��.� � |2�1 |� Γ�1 |� � �xp �− o.o|
� �� (11)
characterized by a (time-dependent) stretching exponent | and a scale factor �. Fig. 8(b)
shows the experimental probability density function measured for Δ � 1, together with a
stretched exponential distribution characterized by the parameter | � 0�61 [6].
Fig. 10) a) Experimental PDF =�;� of the S&P 500 index variations =∆)�;� observed at time intervals ∆) � * BCD (circles).
The symmetrical Lévy stable distribution of index � � *� e> and scale factor � � >� >>EdA is plotted as a solid line. The
parameters characterizing the stable distribution are obtained from the analysis of the scaling properties of the experimental
data on the probability of return to the origin =�; � >� [5]. b) Experimental PDF =�G� of the velocity difference G∆)�)� of a
fluid in fully developed turbulence observed at the highest temporal resolution available ∆) � * BCD (circles). In the figure is
also plotted as a solid line the symmetrical stretched exponential distribution of index � � >� �* and scale factor � � >� >�Ae (solid line). The characterizing parameters of the stretched exponential distribution are obtained starting from the
experimental value of the probability of return to the origin =�G � >� �6�.
Conclusion
The parallel analysis of velocity fluctuations in turbulence and index changes in financial
markets shows that the same statistical methods can be used to investigate systems with
known, but unsolvable equations of motion, and systems for which a basic mathematical
description of the process is still unknown. Despite that, in the two phenomena we find both
10 log
LM ��.�
10 log
LM ��&�
& .
experimental data
Lévy stable experimental data
streched exp.
2
0
-2
-4
-1.0 -0.5 0.0 0.5 1.0
2
1
0
-1
-2
-3
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
University of Ljubljana Turbulence in Financial Markets
Faculty of Mathematics and Physics Matjaž Ivančič
April 2008
12
• similarities: intermittency, non-Gaussian PDFs, the phenomena of information or
energy cascade, and
• differences: the PDFs have different shapes in the two systems and the probability of
return to the origin shows different behaviour. Moreover, velocity fluctuations are
anti-correlated whereas index fluctuations are essentially uncorrelated.
References
[1] R. N. Mantegna and H. E. Stanley, An introduction to econophysics: correlations and
complexity in finance (Cambridge University Press, Cambridge, 2000).
[2] J. Voit, The Statistical Mechanics of Financial Markets (Springer, Berlin, 2005).
[3] M. Levinson, Guide to Financial Markets (Bloomberg Press, New York, 2003).
[4] http://finance.yahoo.com/ (31. 3. 2009).
[5] U. Frish, Turbulence: The Legacy of A. ". Kolmogorov (Cambrige University Press,
Cambrige, 1995).
[6] T. Mizuno, S. Kurihara, M. Takayasu and H. Takayasu, Physica A 234, 296 (2003).
[7] R. N. Mantegna in H. E. Stanley, Physica A 239, 255 (1997).
[8] http://www.math.mcmaster.ca/ (31. 3. 2009).
[9] P. Kailasnath, K.R. Sreenivasan and G. Stolovitzky, Phys. Rev. Lett. 68, 2766 (1992).
[10] M. F. Shlesinger, G. M. Zaslavskya and U. Frisch, Lévy Flights and Related Topics in
Physics (Springer, Berlin, 1995).
[11] N. Oman, Study and Modeling of Price Variations in Financial Markets (Seminarji na
FMF, Ljubljana, 2008/2009).