Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Warsaw University
Warsaw, Poland
Programme
and
Book of Abstra
ts
pielaszek resear
h tel: +(4822)7420281
bonifa
ego 74/94 fax: +(4822)7420282
02-936 warszawa e-mail: bookspielaszek.
om
poland e-mail:
onferen
ess
ien
e24.
om
Prefa
e 4
S
ienti
-Organizing Committee . . . . . . . . . . . . . . . . . . . . . . . . . 4
Symposium on E
ono- and So
iophysi
s 5
Friday, November 19th . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Saturday, November 20th . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Address Book 25
3
Prefa
e
Wel
ome in behalf of physi
ists who try to look over bridge between the methods used
in physi
s and analysis of e
onomi
al and so
ial phenomena. This kind of physi
ists
initiative rapidly develops in the re
ent de
ade as it is a part of the physi
ists great
interdis
iplinary interest in
omplex systems. The promise of the s
ien
e of
omplexity
is to provide at least
ommon tools to ta
kling
omplex problems in natural and so
io-
e
onomi
s
ienti
domains. This type of resear
h a
tivity is also stimulated by the
New and Emerging S
ien
e and Te
hnology work programme as a part of the 6th
Framework Programme of the European Commission.
Our national Symposium takes pla
e one year after the international
onferen
e
Appli
ations of Physi
s in Finan
ial Analysis 4 organized by the Fa
ulty of Physi
s of
Warsaw University of Te
hnology whi
h proved that also in Poland there is an in
reas-
ing
ommunity interested in the interdis
iplinary
ooperation and studies. Moreover,
this Symposium takes pla
e ten days after the Third Nikkei Workshop and Symposium:
Pra
ti
al Fruits of E
onophysi
s, where few of us presented results of our s
ienti
studies.
The aim of our Symposium is to bring together Polish s
ientists, give them op-
portunity to dis
uss s
ienti
problems and stimulate
reation of the national so
iety
in this
ontext. The more so an initial, formal step in this dire
tion was re
ently done
as a new se
tion of the Polish Physi
al So
iety was
reated under the ambitious name
Physi
s in E
onomi
al and So
ial S
ien
es. Moreover, we hope that our Symposium
will also stimulate young resear
hers to their own eort.
4
Symposium on E
ono- and
So
iophysi
s
Warszawa, 19-20 November 2004
Janusz A. Holyst
;
1 2)
, Krzysztof Urbanowi
z
2)
Oral
1) Fa
ulty of Physi
s and Center of Ex
ellen
e for Complex Systems Resear
h, Warsaw Uni-
versity of Te
hnology, Koszykowa 75, Warszawa 00-662, Poland
2) Max-Plan
k Institute for Physi
s of Complex Systems (MPIPKS), Noethnitzer Strasse 38,
Dresden D-01187, Germany
5
FRIDAY, NOVEMBER 19TH SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
Using a re
ently developed method of noise level estimation [1℄ that makes use of properties of
the
oarse grained-entropy we have analyzed the noise level for the Dow Jones index and some
sto
ks from the New York Sto
k Ex
hange as well as Warsaw Sto
k Ex
hange. We have found
that the noise level ranges from 40 to 100 per
ent of the signal varian
e. The
ondition of
the minimal noise level has been applied to
onstru
t optimal portfolios from sele
ted shares
[2℄. We have observed that the level of noise is statisti
ally
orrelated with the sto
k pri
e
hanges and this fa
t has been used to
reate our investment strategy.
Using our method we have
al
ulated 40000 re
ommendations for portofolio optimalization
for periods 4-7 days at Warsaw Sto
k Ex
hange. The annual return re
eived in su
h a way
after substra
ting
ommissions was around 23%.
1) Polish A
ademy of S
ien
es, Institute of Nu
lear Physi
s (IFJ PAN), Radzikowskiego 152,
Kraków 31-342, Poland
2) University of Rzeszow, Institute of Physi
s, Rejtana 16, Rzeszów 35-310, Poland
In re
ent years new tools have been developed in order to investigate multifra
tal properties of
experimental and simulated data. One of them is based on the s
aling behaviour of partition
fun
tion
al
ulated from the maxima of the wavelet transform
oe
ients. This method
seems to be very useful for unfolding in the spa
e-s
ale halfplane the hierar
hi
al stru
ture of
fra
tal data. Another method proposed in this
ontext is a multifra
tal generalization of the
detrended u
tuation analysis (MF-DFA). Both methods are widely spread and
ommonly
used for estimating the multifra
tal spe
tra of signals. We
riti
ally examine validity of ea
h
method if it is applied to high-frequen
y nan
ial data and also
ompare the results obtained
for some well-known mathemati
al multifra
tals.
6
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS FRIDAY, NOVEMBER 19TH
[1℄ H. E. Stanley et al., Self-organized
omplexity in e
onomi
s and nan
e, Pro
. Natl.
A
ad. S
i. 99-Supp, 2561-2565 (2002).
7
FRIDAY, NOVEMBER 19TH SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
We
onsider the basis of the non-linear, long-term (power-law) auto
orrelations present in
empiri
al and our syntheti
high-frequen
y nan
ial time series. The former was studied by
physi
ists sin
e more than one de
ade [1,2, and refs. therein℄ while the syntheti
time-series
was re
eived by us from the re
ently developed, hierar
hi
al, one-dimensional Continuous-
Time Random Walk model [3-8, and refs. therein℄.
This
ombined model is dened by the non-separable Weierstrass walk whi
h
an be o
asion-
ally intermitted by momentary lo
alizations (WWRIL); the lo
alizations themselves are also
des
ribed by the Weierstrass pro
ess. It should be emphasized that the steps of the walk as
well as momentary lo
alizations are un
orrelated. This approa
h makes it possible to study
by hierar
hi
al sto
hasti
simulations the whole spatial-temporal region while analyti
ally it
is possible to study only the initial, pre-asymptoti
and asymptoti
ones but not the very
important intermediate region.
The basi
ontinuous-time series obtained from this sto
hasti
simulation is shown as a se-
quen
e of ve
tors in the temporal-spatial plane. These ve
tors
onne
t the turning points
of a single realization random walk traje
tory (given time-series) expanding in the positive
temporal and spatial dire
tions as we study only the absolute values of sto
k pri
e variations.
This simulation is supported by the waiting-time distribution whi
h is the main quantity of
our two-state (walking-lo
alization) model. These states are again
hara
terized by their own
waiting-time distributions.
The syntheti
, dis
rete time-series was obtained by dis
retization of the original (basi
)
ontinuous-time series at xed time horizon. The auto
orrelation fun
tion was studied versus
time just for this dis
rete time-series. We found that the auto
orrelation exhibits a per-
sistent power-low relaxation both for the Gaussian and non-Gaussian basi
pro
esses. Our
study shows that this relaxation is the result of so-
alled 'domino ee
t' o
urring within the
dis
rete time-series. We suppose that this ee
t is responsible for the analogous long-term
auto
orrelatios
ommonly o
urring in the empiri
al nan
ial high-frequen
y time-series.
For example, by applying the Continuous-Time Weierstrass Flights model [9℄ developed in the
framework of the nonseparable Continuous-Time Random Walks formalism, we
onstru
ted a
series of diusion phase diagrams of in
reasing orders [9,6℄ on the plane dened by the spatial
and temporal fra
tional dimensions of the Weierstrass ights. To dene the risk of a given
asset the moments of in
reasing orders should be
al
ulated (by using a moving average) from
the time series represented the pri
e dynami
s of this asset. Hen
e, we are able to lo
ate the
asset on these phase diagrams and dene its global and lo
al risk of arbitrary order as well as
the level of its mobility and a
tivity again of arbitrary order [2℄.
Bibliography
[1℄ R.N. Mantegna and H.E. Stanley, An Introdu
tion to E
onophysi
s. Correlations and
Complexity in Finan
e. Cambridge Univ. Press, Cambridge 2000.
[2℄ J.-P. Bou
haud and M. Potter, Theory of Finan
ial Risks. From Statisti
al Physi
s to
Risk Management. Cambridge Univ. Press, Cambridge 2001.
[3℄ Levy Flights and Related Topi
s in Physi
s. LNP 450. Eds. M.F. Shlesinger, G.M. Za-
salavsky, U. Fris
h, Springer-Verlag, Berlin 1995.
[4℄ Anomalous Diusion. From Basi
s to Appli
ations. LNP 519. Eds. R. Kutner, A. Pkalski,
K. Sznajd-Weron. Springer-Verlag, Berlin 1999.
[5℄ W. Paul, J. Bas
hnagel, Sto
hasti
Pro
esses. From Physi
s to Finan
e. Springer-Verlag,
Berlin 1999.
[6℄ R. Kutner, F. witaªa, Study of the non-linear auto
orrelations within the Gaussian regime.
Eur. Phys. J. B 33 (2003) 495-503; R. K. and F. ., Possible origin of the non-linear long-term
auto
orrelations within the Gaussian regime, Physi
a A 330 (2003) 177-188; R. K. and F. .,
Sto
hasti
simulations of time series within Weierstrass-Mandelbrot walks, Quant. Finan
e
3 (2003) 201-211; R. K., Extreme events as foundation of Levy walks with varying velo
ity,
Chem. Phys. 284 (2002) 481-505.
8
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS FRIDAY, NOVEMBER 19TH
[7℄ W. S
houtenes, Levy Pro
esses in Finan
e. Pri
ing Finan
ial Derivatives. Wiley Series in
Probability and Statisti
s, J. Wiley & Sons Ltd., Chi
hester 2003.
[8℄ K.I.Ilinski, Physi
s of nan
e. Gauge modeling in non-equilibrium pri
ing. J. Wiley &
Sons Ltd., New York 2001.
[9℄ R. Kutner, Hierar
hi
al spatio-temporal
oupling in fra
tional wanderings. I. Continuous-
time Weierstrass ights, Physi
a A 264 (1999) 84-106.
However, in the studies of Polish sto
k market (WSE) a lo
al, emergent and ine
ient market,
we have found at least four
hara
teristi
s that are dierent from the ones des
ribed above:
(A) The wings of distributions of returns de
ay faster. The exponent value of this de
ay is
about 4, [4℄
(B) The nonlinear short-time
orrelation in daily returns investigated by Arti
ial Insym-
metrized Patterns (AIP in short) indi
ates at the Gaussian origin of the noise while the series
from the mature markets exhibit self-similarity to the Levy noise [5℄. Moreover, we show that
the present state of this market in the AIP presentation is
ompletely dierent from famous
markets.
(C) The Polish market
rash in April 2000
an be named the anti-bubble
rash. Before that
rash the Polish sto
k market had been a growing market sin
e the Russian
rash whi
h hap-
pened in August of 1998. However, in
ontrast to the before-
rash ordinary log-periodi
pri
e
9
FRIDAY, NOVEMBER 19TH SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
development, see [6℄, the pri
e of Polish sto
ks developed in the log-periodi
style as the mar-
ket would suer from the Russian
rash. That is, the log-periodi
os
illations were initiated
by the Russian
rash. This pri
e dependen
e ended with the New Te
hnology
rash in April
2000. Su
h the after-
rash behavior observed before the
rash is
alled the anti-bubble
rash,
[7℄.
(D) The de
it in small returns is noti
able. It appears that the zero-return peaks are a
-
ompanied with valleys on the left always and sometimes on the right side. This feature is
observed not only in sto
ks of
ompanies with low
apital and low liquidity or on the verge of
bankrupt
y; the returns of liquid sto
ks from the top of the volume also exhibit the des
ribed
property, [8℄.
On poorly developed markets, su
h as emergent and lo
al markets, the investors deal with
assets of little liquidity. For example, on Polish sto
k market the majority of sto
ks (about
2/3 out of 235) are with less than 20 transa
tions per session (data from 2002 [9℄). In addition,
the investors meet in their a
tivity a strong politi
al interferen
e.
In both models the syn
hronized de
isions of many investors ee
t in a noti
eable de
it in
small returns. To our study purpose we
hoose the model of Cont and Bou
haud [12℄ modied
to a latti
e version by Stauer et al. [13℄. The intentional imitation me
hanism
an be easily
applied to other market models as, for example, Lux-Mar
hesi [14℄ or Levy-Levy-Solomon [15℄
A
knowledgement:
This work is supported by Gdansk University Proje
t: BW/5400-5-0256-4
10
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS FRIDAY, NOVEMBER 19TH
11
FRIDAY, NOVEMBER 19TH SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
To remove the bad property while retaining other good ones, Mantegna and Stanley (1994)
introdu
ed trun
ated Levy distributions. Then Koponen (1995) dened a dierent trun
ation
whi
h is better suited for analyti
al treatment. Finan
ial data analyzed in re
ent years show
however quite
ompli
ated behaviour. The empiri
al distributions of returns on real markets
are symmetri
near the origin but with mu
h fatter left tail than the right one (
f. Cont
at al (1997), Mata
z (2000)). This behaviour
annot be obtained by Koponen's family. A
satisfa
tory solution of this di
ulty has been proposed by Boyar
henko and Levendrovskii
(2000,2002).
Sin
e the distribution fun
tion of the Levy pro
ess
annot be represented analyti
ally the
dis
ussion is restri
ted to the
hara
teristi
fun
tion of the pro
ess. The starting point is the
Levy-Khint
hine formula, whi
h for a purely non-Gaussian driftless pro
ess Xt has the form
The idea of Koponen was to take a Levy measure for whi
h the dened pro
ess has bounded
variation. The measure with su
h a property
an be dened by the following density fun
tion
f(z)=
+=
exp(- |z|) |z|
1 , (2)
where
+ and
orrespond to positive and negative values of z, respe
tively. These
onstants
are responsible for the asymmetry of the Levy pro
ess and
is the
ut-o parameter, whi
h
gives nite varian
e. In parti
ular letting
to zero we obtain a standard Levy pro
ess.
Formula (2) gives a trun
ated Levy pro
ess whi
h only partially suits the analysis of nan
ial
data. To obtain a symmetry near the origin we have to assume
+ =
. But then also tails
be
ame symmetri
. To get nonsymmetri
tails we should assume dierent
ut-o parameters
for positive and negative values of z. This was already suggested by Mata
z, who however
was unable to gure out how to insert these two parameters to the Koponen formula. The
same idea was risen also by Boyar
henko and Levendrovskii, who introdu
ed abstra
t non-
symmetri
Levy measures.
We have applied that approa
h to the density fun
tion of Koponen proposing its nonsym-
metri
version
f(z)=
+=
exp(- += |z|) |z|
1 . (3)
The obvious advantage of this formula is its simpli
ity, whi
h allows for analyti
al
al
u-
lation of the integral in formula (1). Long
al
ulations lead to the following simple expression
t (k)= t
(- )(
(
+ik) -
+
+ ) (4)
+ ( + -ik) -
+
The obtained formula has been
alibrated to the index data on Warsaw Sto
k Ex
hange
(stri
tly speaking to WIG20 data). We have found quite satisfa
tory agreement and good t
of trun
ated Levy distribution to empiri
al data.
Essentially there is no obsta
le in using that
alibration to option pri
ing. It is however
not a very reasonable proje
t for near future as options traded on WSE are not liquid enough
to ree
t noarbitrage pri
es.
Referen es
1. Boyar
henko S.I. and Levendorskii S.Z. - Option pri
ing for trun
ated Levy pro
esses,
Int.J.Theor.Appl.Finan
e 3 (2000), 549-552.
2. Boyar
henko S.I. and Levendorskii S.Z. - Non-Gaussian Merton-Bla
k-S
holes Theory,
World S
ienti
2002.
3. Cont R., Potters M. and Bou
haud J.P. - S
aling in sto
k market data: stable laws and
beyond, in S
ale Invarian
e and beyond, B.Dubrulle, F.Graner, D.Sornette Eds. Springer
1997, pp.75-85.
12
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS FRIDAY, NOVEMBER 19TH
4. Koponen I. - Analyti
approa
h to the problem of
onvergen
e of trun
ated Levy ights
towards the Gaussian sto
hasti
pro
ess, Phys.Rev. E 52 (1995), 1197-1199.
5. Mata
z A. - Finan
ial modeling and option theory with the trun
ated Levy pro
ess,
Int.J.Theor.Appl.Finan
e 3 (2000), 143-160.
6. Montegna R.N. and Stanley H.E. - Sto
hasti
pro
ess with ultraslow
onvergen
e to a
Gaussian: the trun
ated Levy ight, Phys.Rev.Lett. 73 (1994), 2946-2949.
In my presentation I am going to investigate the
lass of
opulas having regular tails whi
h
allow the uniform expansion i.e. su
h that near the origin they
an be approximated by a
homogeneous fun
tion L(u) of degree 1. Having introdu
ed the notion of the uniform tail
expansion for the multivariate
opulas I will des
ribe its main properties and determine the
set of all possible leading parts L(u). Between others I will show that L is
on
ave. Next
I will deal with the measure indu
ed by L. I will show that it is a produ
t of the Lebesque
measure on the real half line and a measure on the unit simplex.
At the end I will present the example of an appli
ation of the uniform tail expansion to
the problem of estimation of the extreme risk of the portfolio
onsisting of long positions in
risky assets. The spe
ial attention will be given to the Value at Risk (VaR).
Key words:
opulas, fat tails, tail expansions, dependen
e of extreme events, risk manage-
ment, portfolio theory.
13
SATURDAY, NOVEMBER 20TH
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
08:30 - 08:50
Coee, tea and
akes
14
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
SATURDAY, NOVEMBER 20TH
It has been widely re
ognized sin
e some time already that Nature is full of all kinds of
random networks of
omplex topology, whi
h des
ribe su
h apparently disparate systems as
biologi
al, e
onomi
al, so
iologi
al or informati
al ones. Their most
hara
teristi
feature is
the apparent s
ale-free
hara
ter of inter
onne
tions between the nodes. We shall look at then
from the information theory point of view and show that in this way we
an easily des
ribe a
wide spe
trum of possible types of distributions in
luding, in the
ase of nonextensive version
of information theory, the power-like s
ale-free distributions
hara
terist of
omplex systems.
In the paper we present the
onne
tivity and distan
e matrix evolution for dierent types
of networks: exponential, s
ale-free and random ones. Statisti
al and spe
tral properties of
these matri
es are in
luded as well as topologi
al features of the networks.
The GDP values for ea
h of these
ountries are rst normalised to their 1990 value given
in US dollars as published by the Groningen Growth and Development Center on their web
page [2℄. The data
over the period between 1950 and 2003, i.e. 54 points for ea
h
ountry.
The GDP yearly in
rement given by Eq.(1) is
onsidered and its statisti
al properties are
presented
15
SATURDAY, NOVEMBER 20TH
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
d: Ln x Ln ->[0,1℄,
where the distan e d takes the value 0 for orrelated time series and 1 for anti orrelated series.
The distan
es between
ountry are illustrated for a few (5y, 15y, 25y, 35y, 45y) time win-
dows, where y denotes years. It means that
orrelations between su
h
ountries are measured
within very short, short, medium, long and very long time windows.
In order to obtain some quantitative information on the
ountry
orrelations, we have looked
for
lusters or stru
tures formation. Classi
al way to sear
h for
luster is nd subgraphs with
high
lustering
oe
ient [3℄. The alternative way is to build a well dened stru
ture e.g.
Minimal Spanning Tree (MST) and look for a stru
tures repeating in
onse
utive time win-
dows or to nd a set of nodes
onne
ted by short links. For the sake of simpli
ity minimal
length path algorithm (MLP), whi
h is a 1-D modi
ation of the MST algorithm is used.
This algorithm emphasizes the strongest
orrelation between entities with the
onstraint that
the item is atta
hed only on
e to the network. This results in a la
k of loops in the tree.
The
onstru
tion of more elaborate networks is left for further studies. Two dierent graphs:
the unidire
tional (with a given initial point) and bidire
tional minimal length paths (UMLP
and BMLP respe
tively) are
onstru
ted, as a fun
tion of time and for moving time windows
of various sizes. The size of time window is
onstant during the displa
ement.
[BMLP℄: The algorithm begins with sear
hing for the pair of
ountries whi
h has the shortest
distan
e between them. Then these
ountries be
ome the root of a
hain. In the next step
the
loset
ountry for both ends of the
hain is sear
hed. Being sele
ted it is atta
hed to the
appropriate end. Next a sear
h is made for the
losest neighbour of the new ends of the
hain.
Being sele
ted, the entity is atta
hed, a.s.o.
Considering dierent time windows a sort of
riti
al
orrelation time has been found. It
is pointed out that the size of the time window for whi
h the
orrelations are well seen should
not be shorter then 15y, but the most appropriate is 25y time window. This means that on
the level of global e
onomy
orrelations are well seen in the medium length time window and
o-operations between
ountries form a stable relationship. In the
ase of medium and long
time window formation of
lusters, understood as a set of
ountries with highly
orrelated
GDP in
rement is observed.
The properties of UMLP and BMLP algorithms are
ompared and it is found that BMLP
algorithm is more sensitive to sear
hing for a
lustering patterns among
onsidered entities,
while UMLP is suitable for ranking
ountries (
ompanies) and
ould be useful in solving port-
folio problems.
A new method for estimating a realisti
minimal time window to observe
orrelations in
ma
roe
onomy is thus suggested. This method
ould be also applied to a sto
k market anal-
ysis. The mean distan
es analysis is expe
ted to be useful in estimating the shortest time
window in analyzing
orrelations on a sto
k market as well. In su
h a
ase it should be
om-
pared to moving average windows.
Bibliography
16
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
SATURDAY, NOVEMBER 20TH
where W1 ,W2 ,..., are i.i.d. random variables, representing yearly premium less yearly ag-
gregate
laims, and the initial surplus S0 is xed. The model is intended to produ
e answers
on
erning the event of ruin (probability of ruin, time of ruin, de
it when ruin o
urs et
.).
Typi
ally it is assumed that the premium
omponent of Wt is
onstant, and the distribution
of Wt is known.
In real life however, premium is written in advan
e to
over
laims over the
oming expo-
sure period that are often reported and paid a number of periods later. The inadequa
y is
even more obvious in the
ase of the
ontinuous-time model, where the time elapsed between
re
eiving premium and paying (eventually)
ompensations is totally negle
ted. In order to
17
SATURDAY, NOVEMBER 20TH
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
restore
orresponden
e of the model to real life pro
esses, we
ould
hange the interpretation
of variables involved. Premium inow less
laims outlays appearing in the model as Wt
ould
be interpreted as
orresponding to a
ounting
on
epts of premium earned (premium written
less in
rement of the premium reserve) less
laims o
urred (
laims paid plus in
rement of
the outstanding
laims reserve). This leads to interpreting the surplus as the amount of free
assets, and
onsequently the ruin as insolven
y. Under this interpretation the surplus model
is meaningful for pra
ti
e, as in fa
t it fo
uses on phenomena of
ru
ial importan
e for all
involved parties: shareholders, tax authority, poli
yholders, and insuran
e supervision. How-
ever, the problem arises when we take into a
ount that:
I. outstanding
laims amount is a random variable, and the
orresponding reserve is in fa
t
its point predi
tor, based on information available at the a
ounting date.
Additional problem that makes predi
tions
omplex is that:
II. in real life the risk parameter (
hara
terising the
laims pro
ess) is not xed, so its pre-
di
tions are needed as well for premium setting as for reserving purposes.
The paper
on
erns on in
orporating the two above mentioned
ompli
ations into the model
of the insurers surplus. It is shown that (at the
ost of
ertain simplifying assumptions) the in-
orporation
ould be presented as su
h reinterpretation of the surplus St itself and the variable
Wt , that leaves
lassi
al relationships between these variables unae
ted. So, in a way the
paper is fo
used on
alibrating of the above elements of the model to the empiri
al eviden
e.
Te
hniques used in the paper resembles in general those used by S
heike (1992), who intro-
du
ed the notion of fair premium for the
laim pro
ess with dependent in
rements, applying
to this purpose the Doob-Meyer de
omposition for sub-martingales. Assumptions have been
hosen so as to enable
asting the model into the state-spa
e form, whi
h allows for expli
it
expressions of premium and reserves as based on predi
tors of respe
tive
laim payments.
Although using Kalman ltering te
hniques for reserving purposes is not a new idea, their
appli
ation for restoring the
orresponden
e of the simple surplus model to real life pro
esses
is, to my best knowledge, new.
We assume that 'a', 'b' are bounded
ontinuous ve
tor fun
tions, and '
', 'd' are bounded
ontinuous matrix fun
tions of suitable dimensions. Additionally we shall assume that the
matrix dd
T (the supers
ript 'T' stands for transponse) is nondegenerate. Noti
e that equation
(2)
orresponds to dis
retization of a diusion pro
ess. The set of fa
tors may in
lude divident
yields, pri
e - earning rations, short term interest rates, the rate of ination see e.g. [1℄. The
dynami
s of su
h fa
tors is usually modeled using diusion, frequently linear equations eg.
in the
ase when we assume following [1℄ that 'a' is a fun
tion of spot interest rate governed
by the Vasi
ek pro
ess. Our assumptions
on
erning boundedness of ve
tor fun
tions 'a' and
'b' may be relaxed allowing linear growth, however in su
h
ase we shall need other more
ompli
ated assumptions.
Assume that starting with an initial apital V(0) we invest in assets. Let hi (n) be the part
18
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
SATURDAY, NOVEMBER 20TH
of the wealth pro
ess lo
ated in the i-th asset at time n, whi
h is assumed to be nonnegative.
The
hoi
e of hi (n) depends on our observation of the asset pri
es and e
onomi
al fa
tors up
to time 'n'. Denoting by V(n) the wealth pro
ess at time 'n' and by h(n)=(h1 (n), ...,hm (n))
our investment strategy at time 'n', we have that h(n) belongs to U={(h1 ,...,hm ), hi >=0,
i=1 m hi =1} and
with
<0. Using (3) we
an write the
ost fun
tionals (4) and (5) in the more
onve-
nient forms. Namely,
n 1 ln ( m h (t) (x(t),h(t))}}/n}
x (h(n))= lim infn >oo {{Ex {t=0
0
J i=1 i i
=lim infn >=infty {{Ex {t=0
n 1
(x(t),h(t))}}/n}, (6),
m
with
(x,h)=E{ln(i=1 hi i (x,W(0)))}. It is
lear that risk neutral
ost fun
tional J depends
0
on un
ontrolled Markov pro
ess (x(n)) and we pra
ti
ally maximize the
ost fun
tion
itself.
Consequently an optional
ontrol is of the form
ontrol (u'(x(n)), where suph
(x,h)=
(x,
u'(x)) and fun
tion Borel measurable u': R
k -> U exists by
ontinuity of
for xed x belong-
ing
k
to R .
This
ontrol does not depend on asset pri
es and is a time independent fun
tion of
urrent values of the fa
tors x only. The Bellman equation
orresponding to the risk neutral
ontrol problem is of the form
h;
n (!):=
t=0 n 1 exp{
ln ( i=1 m hi (t) i (x(t),W(t)))}(E{exp{
ln (i=1 m hi (t) i (x(t),W(t))}|Ft }) 1
P jn h; = h; n (!)=Pjn (d! ).
Then
n 1 ln( m h (t) (x(t),
Jx (h(n))={lim sup n >oo {ln Ex {exp{
t=0 i=1 i i W(t)))}}/n}}/
={lim supn
h;
n 1
>oo {ln E x {exp{ t=0
(x(t),h(t))}}/n}}/ , (8)
with
and where
/
orresponds to optimal value of the
ost fun
tional (8). Noti
e that un-
der measure P
h;
the pro
ess (x(n)) is still Markov but with
ontrolled transition operator
P
h;
(x,dy). Following [5℄ we shall show that
/ -> 0
19
SATURDAY, NOVEMBER 20TH
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
whenever ->0.
The study of risk sensitive portfolio optimization has been originated in [1℄ and then
on-
tinued in a number of papers in parti
ular in [12℄. Risk sensitive
ost fun
tional was studied
in papers [9℄, [5℄, [6℄, [3℄, [4℄, [8℄, [2℄, [7℄ and referen
es therein. Using splitting of Markov pro-
esses arguments (see [11℄) we study Poisson equation for additive
ost fun
tional the solution
of whi
h is also a solution to risk neutral Bellman equation. We
onsider then risk sensitive
portfolio optimization with risk fa
tor
lose to 0. We generalize the result of [12℄, where uni-
form ergodi
ity of fa
tors was required and using [7℄ show the existen
e of Bellman equation
for small risk in a more general ergodi
ase. The proof of that nearly optimal
ontinuous
risk neutral
ontrol fun
tion is also nearly optimal for risk sensitive
ost fun
tional with risk
fa
tor
lose to 0 is based on modi
ation of the arguments of [5℄ using some results from the
theory of large deviations.
Referen es
[1℄ T.R. Biele
ki, S. Pliska, Risk sensitive dynami
asset management, JAMO, 39 (1999),
pp.337-360.
[2℄ V.S. Borkar, S.P. Meyn, Risk-Sensitive Optimal Control for Markov De
ision Pro
esses
with Monotone Cost, Math. Meth. Oper. Res., 27 (2002), pp. 192-209.
[5℄ G.B. Di. Masi, L. Stettner, Risk sensitive
ontrol of dis
rete time Markov pro
esses
with innite horizon, SIAM J. Control Optimiz., 38 (2000), pp. 61-78.
[6℄ G.B. Di Masi, L. Stettner, Innite horizon risk sensitive
ontrol of dis
rete time Markov
pro
esses with small risk, Sys. Control Lett., 40 (2000), pp. 305-321.
[7℄ G.B. Di Masi, L. Stettner, Innite horizon risk sensitive
ontrol of dis
rete time Markov
pro
esses under minorization property, submitted.
[9℄ D. Hernandez-Hernandez, S.J. Mar
us, Risk sensitive
ontrol of Markov pro
esses in
ountable state spa
e, Sys. Control Letters, 29 (1996), pp. 147-155.
[10℄ I. Kontoyiannis, S.P. Meyn, Spe
tral Theory and Limit Theorems for Geometri
ally
Ergodi
Markov Pro
ess, Ann. Appl. Prob. 13 (2003), 304-362.
[11℄ S.P. Myen, R.L. Tweedie, Markov Chains and Sto hasti Stability, Springer 1996.
[12℄ L. Stettner, Risk sensitive portfolio optimization, Math. Meth. Oper. Res. 50 (1999),
463-474.
20
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
SATURDAY, NOVEMBER 20TH
attitude with regard to the personal area spreads in a dierent way than that with regard to
the e
onomi
area. Thus, we assume that ea
h agent tries to inuen
e its neighbors, but in
the personal area the information ows inward from the neighborhood (like in most opinion
dynami
models), whereas in the e
onomi
area the information ows outward from the agent
or group of agents to the neighborhood (like in the Sznajd model [1℄).
In 2000 we proposed a new model [1℄ of opinion formation named by Stauer the Sznajd
model (SM). The model has found many appli
ations - from politi
s to nan
e and market-
ing. For the review of the model and its appli
ations see [2-7℄ and refs. therein. We assumed
that individual opinion is represented by an Ising spin (yes or no), like in many opinion dy-
nami
s models. The really new thing that we introdu
ed was the dynami
s of spins. The
motivation to propose this new dynami
s was a phenomenon
alled by so
ial psy
hologists
the So
ial Validation - one fundamental way that we de
ide what to do in a situation is to
look to what others are doing. A group of people sharing the same opinion inuen
es the
neighborhood mu
h easier than isolated individuals. On the other hand, it seems that so
ial
validation phenomena works mu
h weaker in private aspe
ts of life, su
h as religion. These
attitudes are mostly inuen
ed by the family or friends. For this reason we believe that to
model the evolution of attitude to personal freedom one should use Glauber dynami
s.
In this paper we will show that building
onsensus in personal life is mu
h more di
ult
that in e
onomi
s. On the other hand, the dynami
s
onne
ted with e
onomi
freedom is
mu
h more
ompli
ated than the one
onne
ted with personal freedom. The model
an be
used to des
ribe the formation of so
ial attitude regarding e
onomi
freedom, i.e., whether
the so
iety prefers an open market or a state
ontrolled market. We will also show how the
personal attitudes and toleran
e inuen
e this formation pro
ess, a problem that is dis
ussed
in the behavioral nan
e literature [8℄.
Referen es
[1℄ K. Sznajd-Weron, J. Sznajd, Opinion evolution in
losed
ommunity. Int. J. Mod. Phys.
C 11 (6), 1157-1165 (2000).
[2℄ D. Stauer, So
iophysi
s - A review of re
ent Monte Carlo simulations. Fra
tals 11, 313-
318 (2003).
[3℄ D. Stauer, So
iophysi
s: the Sznajd model and its appli
ations. Comput. Phys. Com-
mun. 146 (1), 93-98 (2002).
[4℄ P. Ri
hmond, L. Sabatelli, Langevin pro
esses, agent models and so
io-e
onomi
systems.
Physi
a A 336, 27-38 (2004).
[5℄ A.T. Bernards, U.M.S. Costa, A.D. Araujo, D. Stauer, Damage spreading,
oarsening
dynami
s and distribution of politi
al votes in Sznajd model on square latti
. Int. J. Mod.
Phys. C 12 (2), 159-167 (2001).
[6℄ A.T. Bernardes, D. Stauer, J. Kertesz, Ele
tion results and the Sznajd model on Barabasi
network, Eur. Phys. J. B 25 (1), 123-127 (2002).
[7℄ C. Shulze, Advertising in the Sznajd marketing model. Int. J. Mod. Phys. C 14 (1), 95-98
(2003).
[8℄ D. Kahneman, A. Tversky, (eds.), Chaos, Values and Frames. Cambridge Univ. Press,
2000.
21
SATURDAY, NOVEMBER 20TH
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
14:15 - 14:40
The Heider balan
e and so
ial distan
e
Oral 1) 1) 1)
Krzysztof Kuªakowski , Przemysªaw Gawro«ski , Piotr Gronek
1) AGH University of S
ien
e and Te
hnology, Fa
ulty of Physi
s and Applied Computer S
i-
en
e (AGH), Mi
kiewi
za 30, Kraków 30-059, Poland
The Heider balan
e is usually dis
ussed in terms of
ognitive dissonan
e. Here we stress its
role for theory of
oni
ts. The model system is a group of N persons, represented by nodes
of a fully
onne
ted graph. A set of dierential equations is proposed for the time evolution
of the so
ial distan
e between the group members. This distan
e
an be measured in the
Bogardus s
ale. In the Heider approa
h, the distan
e r(i,j) is positive if i and j a
ept ea
h
other, and negative in the opposite. After some time T(N), the system rea
hes the Heider
balan
e, i.e. two subgraphs appear. Within both subgraphs, all distan
es are positive; others
are negative. We dis
uss an inuen
e of limitations of the allowed range of the so
ial distan
e
on the system dynami
s.
[2℄ Z. Wang and W. Thorngate, Sentiment and So
ial Mitosis: Impli
ations of Heider's Bal-
an
e Theory, JASSS vol 6 No 3, http://jasss.so
.surrey.a
.uk/6/3/2.html
22
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
SATURDAY, NOVEMBER 20TH
SAS provides a layer of unique business intelligen
e whi
h in
ludes the SAS
r Industry In-
telligen
e Solutions (xIS). xIS build on the
ustomer's operational environments and help
transform organizations into Intelligent Enterprises by surfa
ing know-how based on their
own real-world data.
xIS bring together what has been learnt from many proje
ts that SAS has been involved
in worldwide to address industry's hottest business topi
s. They represent a quality, val-
idated intelligen
e solution with a standard ar
hite
ture that is so exible that it
an be
applied to most organizations.
23
SATURDAY, NOVEMBER 20TH
SYMPOSIUM ON ECONO- AND SOCIOPHYSICS
24
Address Book
25
ADDRESS BOOK
26
ADDRESS BOOK
27
ADDRESS BOOK
28
Index
29