statistique, théorie et gestion de portefeuille - Docs at ISFA
statistique, théorie et gestion de portefeuille - Docs at ISFA
statistique, théorie et gestion de portefeuille - Docs at ISFA
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
172 6. Comportements mimétiques <strong>et</strong> antagonistes : bulles hyperboliques, krachs <strong>et</strong> chaos<br />
Q UANTITATIVE F INANCE Imit<strong>at</strong>ion and contrarian behaviour: hyperbolic bubbles, crashes and chaos<br />
Probability<br />
1.0<br />
Symm<strong>et</strong>ric dynamics<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
Asymm<strong>et</strong>ric dynamics<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
Effect of periodic<br />
orbits<br />
–0.5 –0.4 –0.3 –0.2 –0.1 0.0<br />
p–1/2<br />
0.1 0.2 0.3 0.4 0.5<br />
Figure 14. Cumul<strong>at</strong>ive distribution function of p − 1/2 for m = 60<br />
polled agents and param<strong>et</strong>ers ρhb = ρbh = 0.72 and<br />
ρhh = ρbb = 0.85 (dashed line) and ρhb = 0.72, ρbh = 0.74,<br />
ρhh = 0.85 and ρbb = 0.87 (continuous line). Note th<strong>at</strong> the<br />
asymm<strong>et</strong>ric dynamics has the effect of shifting the cumul<strong>at</strong>ive<br />
distribution to the right.<br />
7.1. Finite-size effects in other mo<strong>de</strong>ls<br />
We now investig<strong>at</strong>e finite-size effects resulting from a finite<br />
number N of interacting agents trading on the stock mark<strong>et</strong>.<br />
This issue of the role of the number of agents has recently been<br />
investig<strong>at</strong>ed vigorously with surprising results.<br />
Hellthaler (1995) studied the N-<strong>de</strong>pen<strong>de</strong>nce of the<br />
dynamical properties of price time series of the Levy <strong>et</strong> al<br />
mo<strong>de</strong>l (1995, 2000). Egenter <strong>et</strong> al (1999) did the same for the<br />
Kim and Markowitz (1989) and the Lux and Marchesi (1999)<br />
mo<strong>de</strong>ls. They found th<strong>at</strong>, if this number N goes to infinity,<br />
nearly periodic oscill<strong>at</strong>ions occur and the st<strong>at</strong>istical properties<br />
of the price time series become compl<strong>et</strong>ely unrealistic. Stauffer<br />
(1999) reviewed this work and others such as the Levy <strong>et</strong> al<br />
(1995, 2000) mo<strong>de</strong>l: realistically looking price fluctu<strong>at</strong>ions<br />
are obtained for N ∝ 102 , but for N ∝ 106 the prices<br />
vary smoothly in a nearly periodic and thus unrealistic way.<br />
The mo<strong>de</strong>l proposed by Farmer (1998) suffers from the same<br />
problem: with a few hundred investors, most investors are<br />
fundamentalists during calm times, but bursts of high vol<strong>at</strong>ility<br />
coinci<strong>de</strong> with large fractions of noise tra<strong>de</strong>rs. When N<br />
becomes much larger, the fraction of noise tra<strong>de</strong>rs goes to zero<br />
in contradiction to reality. On a somewh<strong>at</strong> different issue,<br />
Huang and Solomon (2001) have studied finite-size effects<br />
in dynamical systems of price evolution with multiplic<strong>at</strong>ive<br />
noise. They find th<strong>at</strong> the exponent of the Par<strong>et</strong>o law<br />
obtained in stochastic multiplic<strong>at</strong>ive mark<strong>et</strong> mo<strong>de</strong>ls is crucially<br />
affected by a finite N and may cause, in the absence of<br />
an appropri<strong>at</strong>e social policy, extreme wealth inequality and<br />
mark<strong>et</strong> instability. Another mo<strong>de</strong>l (apart from ours) where<br />
the mark<strong>et</strong> may stay realistic even for N →∞seems to be<br />
the Cont and Bouchaud percol<strong>at</strong>ion mo<strong>de</strong>l (2001). However,<br />
this only occurs for an unrealistic tuning of the percol<strong>at</strong>ion<br />
concentr<strong>at</strong>ion to its critical value. Thus, in most cases, the<br />
limit N →∞leads to a behaviour of the simul<strong>at</strong>ed mark<strong>et</strong>s<br />
Correl<strong>at</strong>ion<br />
1.2<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
–0.2<br />
Symm<strong>et</strong>ric dynamics<br />
Asymm<strong>et</strong>ric dynamics<br />
0 20 40 60 80 100 120 140 160 180 200<br />
Time lag<br />
Figure 15. Correl<strong>at</strong>ion function for m = 60 polled agents and<br />
param<strong>et</strong>ers ρhb = ρbh = 0.72 and ρhh = ρbb = 0.85 (dashed line)<br />
and ρhb = 0.72, ρbh = 0.74, ρhh = 0.85 and ρbb = 0.87 (continuous<br />
line). Note th<strong>at</strong> the correl<strong>at</strong>ion function of r<strong>et</strong>urns goes to zero<br />
(within the noise level) <strong>at</strong> time lags larger than about 100. This<br />
unrealistic fe<strong>at</strong>ure of a long-range correl<strong>at</strong>ion in the r<strong>et</strong>urns makes it<br />
unnecessary to show the even longer-range correl<strong>at</strong>ion of the<br />
absolute values of the r<strong>et</strong>urns.<br />
which becomes quite smooth or periodic and thus predictable,<br />
in contrast to real mark<strong>et</strong>s. Our mo<strong>de</strong>l, which remains<br />
(d<strong>et</strong>erministically) chaotic, is thus a significant improvement<br />
upon this behaviour. We trace this improvement on the highly<br />
nonlinear behaviour resulting from the interplay b<strong>et</strong>ween<br />
the imit<strong>at</strong>ive and contrarian behaviour. It has thus been<br />
argued (Stauffer 1999) th<strong>at</strong>, if these previous mo<strong>de</strong>ls are good<br />
<strong>de</strong>scriptions of mark<strong>et</strong>s, then real mark<strong>et</strong>s with their strong<br />
random fluctu<strong>at</strong>ions are domin<strong>at</strong>ed by a r<strong>at</strong>her limited number<br />
of large players: this amounts to the assumption th<strong>at</strong> the<br />
hundred most important investors or investment companies<br />
have much more influence than the millions of less wealthy<br />
priv<strong>at</strong>e investors.<br />
There is another class of mo<strong>de</strong>ls, the minority games<br />
(Chall<strong>et</strong> and Zhang 1997), in which the dynamics remains<br />
complex even in the limit N →∞. It has been established<br />
th<strong>at</strong> the fluctu<strong>at</strong>ions of the sum of the aggreg<strong>at</strong>e <strong>de</strong>mand<br />
have an approxim<strong>at</strong>e scaling with similar sized fluctu<strong>at</strong>ions<br />
(vol<strong>at</strong>ility/standard <strong>de</strong>vi<strong>at</strong>ion) for any N and m for the scale<br />
scaled variable 2 m /N, where m is the memory length (Chall<strong>et</strong><br />
<strong>et</strong> al 2000). In a generaliz<strong>at</strong>ion, the so-called grand canonical<br />
version of the minority game (Jefferies <strong>et</strong> al 2001), where the<br />
agents have a confi<strong>de</strong>nce threshold th<strong>at</strong> prevents them from<br />
playing if their str<strong>at</strong>egies have not been successful over the<br />
last T turns, the dynamics can <strong>de</strong>pend more sensitively on N:<br />
as N becomes small, the dynamics can become quite different.<br />
For large N, the complexity remains.<br />
The difference b<strong>et</strong>ween the limit N →∞consi<strong>de</strong>red up<br />
to now in this paper and the case of finite N is th<strong>at</strong> pt is no<br />
longer the fraction of bullish agents. For finite N, pt must<br />
be interpr<strong>et</strong>ed as the probability for an agent to be bullish.<br />
Of course, in the limit of large N, the law of large numbers<br />
275