ECONOPHYS-KOLKATA II
International Workshop on
Organizer
Centre for Applied Mathemati
s and Computational S
ien
e (CAMCS),
Saha Institute of Nu
lear Physi
s
Convener
Bikas K Chakrabarti
Se
retary
Arnab Chatterjee
Venue
Main Auditorium
Saha Institute of Nu
lear Physi
s
1/AF Bidhannagar, Kolkata 700 064, India
2
Abstra
ts
Invited Talks
Agent-Based Modelling with Wavelets De
omposition and an Evolutionary Arti
ial Neural Network: Appli
ations
to CAC 40 Fore
asting
Serge Hayward
Department of Finan
e, E
ole Superieure de Commer
e de Dijon
Burgundy S
hool of Business, 29 rue Sambin 21000 Dijon, Fran
e
Unlike fully revealing equilibrium of homogeneous beliefs, in the environment with heterogeneous beliefs pri
es are
driven by prevailing expe
tations of market parti
ipants. Thus, fore
asting future pri
es, one must form expe
tations
of others fore
asts. Evolution of agents' expe
tations largely governs the adaptive nature of market pri
es. Overlapping
beliefs of heterogeneous agents prevent the ee
tive examination of expe
tation formation and pri
e fore
asting by
5
traditional methods. In the approa
h proposed in this paper, a time series is de
omposed into a
ombination of
underlying series, representing beliefs of major
lusters of agents. The analysis of individual parts improves statisti
al
inferen
e, isolating ee
tively nonstationary and nonlinearly features. Emergent lo
al behaviour is also more re
eptive
to predi
tion. The overall fore
ast (weighted
ombination of individual fore
asts) is found to be determined and evolved
depending on spe
i
market
onditions. On the statisti
al level, the data generating me
hanism is
onsidered as
omplex multi-stru
tured system, with individual layers
orresponding to parti
ular frequen
ies. Re
e
ting the time
preferen
es of agents, trading strategies being homogeneous intra-type are heterogeneous inter-type for agents with
distin
t time preferen
es. Overall market a
tivity at ea
h moment, providing the dynami
feedba
k a
ross agents'
types, generates market pri
es. The frequen
y de
omposition of a time series identies the lo
al and global stru
tures
and separates short and long time dynami
s. Wavelet transforms are used for adaptive analysis of lo
al behaviour of
heterogeneous agents. The Geneti
Algorithm is applied to determine the optimal de
omposition of the signal and
representation of heterogeneous traders. The Arti
ial Neural Network is trained to learn information at the s
ale
level that is hidden in the aggregate. The resulting models seek to enhan
e the understanding of the underlying data
generating me
hanisms of nan
ial time series and to develop new approa
hes for nan
ial fore
asting.
The understanding of
omplex so
ial or e
onomi
systems is an important
hallenge of modern so
ieties. Despite
many agent based models have been investigated, in few
ases an agent based empiri
al investigation is possible due to
the la
k of data. Here we present a
omprehensive study of the Madrid Sto
k Ex
hange showing that nan
ial rms
are self-organized as an e
ology of rms with various degree of spe
ialization. Few large rms push the pri
e in a given
dire
tion demanding liquidity to the market on a long time s
ale, whereas many heterogeneous rms provide liquidity
on a short time s
ale by reverting the dire
tion of pri
e. Our analysis allows to build one of the rst empiri
ally
grounded agent based models of nan
ial markets.
6
Ex
ess Volatility and Herding in an Arti
ial Finan
ial Market: Estimation of an Agent-Based Model
Thomas Lux
Department of E
onomi
s, University of Kiel, Olshausenstr. 40, 24118 Kiel, Germany
The behavioral origins of the stylized fa
ts of nan
ial returns have been addressed in a growing body of agent-
based models of nan
ial markets. While the traditional eÆ
ient market viewpoint explains all statisti
al properties
of returns by similar features of the news arrival pro
ess, the more re
ent behavioral nan
e models explain them
as imprints of universal patterns of intera
tion in these markets. In this paper we
ontribute to this literature by
introdu
ing a very simple agent-based model in whi
h the ubiquitous stylized fa
ts (fat tails, volatility
lustering)
are emergent properties of the intera
tion among traders. The simpli
ity of the model allows us to estimate the
underlying parameters, sin
e it is possible to derive a
losed form solution for the distribution of returns. We show
that the tail shape
hara
terizing the fatness of the un
onditional distribution of returns
an be dire
tly derived from
some stru
tural variables that govern the traders' intera
tions, namely the herding propensity and the autonomous
swit
hing tenden
y.
Finan
ial time series in general exhibit average behavior at "long" time s
ales and sto
hasti
behavior at "short"
time s
ales. As in statisti
al physi
s, the two have to be modelled using dierent approa
hes - deterministi
for trends
and probabilisti
for
u
tuations about the trend. In this talk, we will des
ribe a new wavelet based approa
h to
separate the trend from the
u
tuations in a time series. A determin- isti
(non-linear regression) model is then
onstru
ted for the trend using geneti
algorithm. We thereby obtain an expli
it analyti
model to des
ribe dynami
s
of the trend. Further the model is used to make predi
tions of the trend. We also study statisti
al and s
aling
properties of the
u
tuations. The
u
tuations have non- Gaussian probability distribution fun
tion and show multi
s
aling behavior. Thus, our work results in a
omprehensive model of trends and
u
tuations of a nan
ial time series.
Random matrix approa
h to
u
tuations and s
aling in
omplex systems : Appli
ations in market analysis
M. S. Santhanam
Theoreti
al Physi
s and Complex Systems Division
Physi
al Resear
h Laboratory, Navrangpura, Ahmedabad 380 009, India
Random matrix theory (RMT) had its origins in the statisti
al analysis of nu
lear spe
tra. In re
ent times, it
has emerged as an important paradigm to study and understand multivariate
orrelations in empiri
al settings. For
instan
e, sto
k-sto
k
orrelations in a market
an be subje
ted to RMT type analysis. Thus, in prin
iple, any
multivariate time series data
an be analysed within the RMT framework. We review some of these results. However,
many important questions about universality and s
aling require only a single time series, for example, a series of sto
k
market index. Can RMT be a model for su
h pro
esses ? We argue that we
an treat the appropriately transformed
eigenvalues of a quantum system to be a time series. Then, we
an use RMT to model
orrelations and
u
tuations
arising in univariate series. We show that detrended
u
tuation analysis, often used to study s
aling, is equivalent
to delta-3 statisti
s widely used in RMT
ontext. Using this
onne
tion we obtain theoreti
al estimates for Hurst
exponents. In general, an anti-persitent time series
an be modelled by the 'spe
tral time series' drawn from the
standard Wigner-Dyson ensemble of RMT. We illustrate these ideas using nan
ial data from the markets.
The Power (Law) of Indian Markets: Analysing NSE and BSE trading statisti
s
Sitabhra Sinha
Institute of Mathemati
al S
ien
es, C.I.T. Campus, Taramani, Chennai 600 113, INDIA
We have re
ently analysed both ti
k-by-ti
k as well as daily
losing pri
es for sto
ks listed in the NSE, and daily
losing pri
es for sto
ks listed in BSE. We have found that regardless of the ex
hange, and the time resolution of
the data, the return distribution shows a long tail, suggestive of a power law with exponent approximately equal to
-3. This runs
ounter to some re
ent studies that have
laimed that the Indian market behaves
loser to a random
walk than sto
k markets in the developed world. We have also analysed the distribution of the number of trades and
volume, as well as the volatility behavior. Finally, we have looked at how
orrelated the movement of sto
ks are in
NSE, by
on
entrating on the NIFTY-50 sto
ks.
The Apparent "Madness of Crowds": Irrational
olle
tive behavior as an emergent out
ome of networked
intera
tions among individually rational agents
Sitabhra Sinha
Institute of Mathemati
al S
ien
es, C.I.T. Campus, Taramani, Chennai 600 113, INDIA
The o
urren
e of abnormal events in markets su
h as "bubbles" and "
rashes" have often been as
ribed to irrational
behavior of traders. In normal
ir
umstan
es, agents are thought to be rational beings whose de
isions are entirely
governed by desire to maximise their gain with minimum risk. However, in
ertain
ir
umstan
es, agents seem to take
part in an extraordinary "mass delusion", where ea
h individual follows the mass of traders behaving as a herd to a
ourse of a
tion whi
h seems irrational (su
h as ex
hanging a tulip for a house, during the height of the Tulipomania
buuble in 17th
entury Holland). In this talk we argue that su
h apparently irrational behavior is an emergent behavior
of a network of intera
ting agents, where ea
h agent is behaving rationally. Our argument is based on the fa
t that
every agent has a pro
essing
ost of verifying ea
h pie
e of information that it re
eives. To minimize this
ost it has to
trust a few other agents on the basis of previous experien
e. This "web of trust"
reates an intera
tion network among
agents, in whi
h one agent
an free-ride on the pro
essing performed by another agent by simply
opying whatever the
latter is doing. While in normal
ir
umstan
es this will maximize the gain of an agent, its su
ess depends
ru
ially
on the assumption that someone down the
hain of trust is a
tually pro
essing the information to
ome up with the
orre
t a
tion. If this assumption is violated, e.g., by the formation of a
losed feedba
k "loop of trust", apparently
irrational behavior su
h as bubbles and
rashes ensue.
Posters
There had been lot of fo
us on modelling dynami
s of sto
k market. Approa
hes
onsidered normally have an
underlying agent-agent intera
tion network. Empiri
al data to verify su
h models is diÆ
ult to obtain. We present a
model where there is no dire
t intera
tion among agents and they intera
t only via market. They use their rationale
to minimize risk and maximize gains. This model
aptures basi
features of sto
k market like formation of expe
tation
bubbles,
rashes and volatility
lustering.
Modeling the bouts of a
tivity in a sto
k market based on trading tenden
ies
Varsha Kulkarni
Department of Physi
s and Astrophysi
s, University of Delhi, Delhi - 110007.
The term `Sto
k Market' in an abstra
t way explains the me
hanism by whi
h trading of
ompany sto
ks takes pla
e.
If we were to tra
e the
ourse of sto
k market
orrelations, we would nd that linkages between pri
es and indi
es
tighten during periods of nan
ial
risis or `highly volatile situations'. Volatility is basi
ally a measure of market
u
tuations and a volatile situation is
aused not only by sudden, unanti
ipated sho
ks, news-bursts, e
onomi
and
politi
al
y
les but also by a kind of trading pattern between agents that emerges as a
umulative ee
t of traders'
psy
hology having no obvious pre
ursors (like Bla
k Friday of 1987). The study of latter forms an integral part of
the eld of Behavioral E
onomi
s whi
h applies s
ienti
resear
h on human and so
ial
ognitive biases to better
understand e
onomi
de
isions and their ee
ts on pri
es, returns et
. Presented here is a model based on a simple
me
hanism of trading between agents in an arti
ial sto
k market to bring out intermittent bursts of a
tivity. From
the results it may be observed that the bursts in
orrelations and volumes of transa
tion are not only
on
urrent
but
ome in
lusters. As the system runs, it gets into volatile situations and returns to its normal state on its
own. Although this model presents a highly idealized pi
ture of a sto
k market, with its rules rendering the traders
irrational, it may nevertheless be thought of as a modest representation of the self-organizing behavior of the market.
11
Multifra
tal analysis of nan
ial time series using Wavelet Transform Modulus Maxima Method
A. N. Sekar Iyengar, Md. Nurujjaman, and R. Narayanan
Plasma Physi
s Division, Saha Institute of Nu
lear Physi
s, 1/AF Bidhannagar, Kolkata 700064
Multifra
tal analysis using WTMM is a well established te
hnique in applied mathemati
s and physi
s and is gaining
popularity in nan
ial resear
h also. This has a lot of advantages over the stru
ture fun
tion method and in this
paper we will dis
uss the multifra
tality of the nan
ial time series of various
ompanies using the WTMM method.
Re
ently, we have developed a wavelet-based method for
hara
terizing the s
aling properties of non-stationary time
series [Phys. Rev. E 72, 046120 (2005)℄. In the present work, we report the presen
e of long-range
orrelation and
multi-s
aling behavior of National Sto
k Ex
hange (NSE) S&P CNX index values. Fourier analysis is also performed
to study the s
aling behavior of time series and the results are
ontrasted with wavelet analysis.