Could short selling make financial markets tumble?
Extracted source text ยท arXiv cond-mat/0308548
\documentstyle[11pt,epsfig]{article}
\setlength{\topmargin}{-0.5cm}
\setlength{\headsep}{1.6cm}
\setlength{\evensidemargin}{-1cm}
\setlength{\oddsidemargin}{-1cm}
\setlength{\textheight}{23cm}
\setlength{\textwidth}{18cm}
\newcommand \be{\begin{equation}}
\newcommand \bea{\begin{eqnarray}}
\newcommand \ee{\end{equation}}
\newcommand \eea{\end{eqnarray}}
\newcommand{\lp}{\left(}
\newcommand{\rp}{\right)}
\begin{document}
\title{Could short selling make financial markets tumble?}
\author{J\o rgen Vitting Andersen$^{1,2}$\\
$^1$ U. F. R. de Sciences \'Economiques, Gestion, Math\'ematiques et
Informatique, \\ CNRS UMR7536 and Universit\'e Paris X-Nanterre,
92001 Nanterre Cedex, France \\
$^2$ Laboratoire de Physique de la Mati\`{e}re Condens\'{e}e\\ CNRS UMR6622 and
Universit\'{e} de Nice-Sophia Antipolis, 06108 Nice Cedex 2, France\\
e-mail: vitting@unice.fr\\}
\date{\today}
\maketitle
{\bf
It is suggested to consider long term trends of financial markets
as a growth phenomenon. The question that is asked is what conditions are
needed for a long term sustainable growth or contraction in a financial market?
The paper discuss the role of traditional market players
of long only mutual funds versus
hedge funds which take both short and long positions.
It will be argued that financial
markets since their very origin and only till very recently, have been
in a state of ``broken symmetry'' which favored long term growth instead of
contraction. The reason for this ``broken symmetry''
into a long term ``bull phase'' is the historical
almost complete dominance by long only players in financial markets.
Dangers connected
to short trading are illustrated by the appearence of long term
bearish trends seen in analytical results and by
simulation results of an agent based market model.
Recent short trade data of the Nasdaq Composite index show an increase in the
short activity prior to
or at the same time as dips in the market, and reveal
an steadily increase in the short trading activity, reaching
levels never seen before.
}
\vskip 0.5cm
% insert suggested PACS numbers in braces on next line \pacs{}
\vskip 0.5cm
% body of paper here
%\narrowtext
% \twocolumn[...text...]
Stock markets are known to be notoriously difficult to predict.
The impossibility of predicting has been formalized in the Efficient
Market Hypothesis (EMH) which
states that all available information is included in
prices\cite{Fama,Samuelson}. Still without attempting specific prediction, it
makes very well sense to ask which conditions are needed for a
substainable trend, e.g. what is the ``fuel'' needed for a manifestation
of a longer term bull or bear market?
In this paper it is suggested to consider long term trends of financial markets
as a growth phenomenon.
The paper will discuss the role of traditional market players
of long only mutual funds versus
the recent arrival of hedge funds which take both short and long positions.
Short selling\cite{Chen,Platt,Ackert} happens when speculators sell,
say at time $t_0$ to price $P(t_0)$, securities
that they do not
own but borrow. At a later time, say $t_1$, the speculator has to
buy the security and give it back to it's owner.
In case of a price decline between
$t_0$ and $t_1$ the speculator thereby makes the profit $P(t_0) -P(t_1)$.
This is in contrast to long only funds who are only allowed to own
securities and therefore can only profit when prices rise.
To make an analogy with a physical system, it will be argued that financial
markets since their very origin and only till very recently, have been
in a state of ``broken symmetry'' which favored long term growth instead of
contraction. The reason for this ``broken symmetry''
into a long term ``bull phase'' is the historical
almost complete dominance by long only players in financial markets. Only
with the recent arrival of investors that take up short positions is the
symmetry slowly being restored, with the implications, as will be argued
in this paper, of an increased probability for lasting decline of the markets,
i.e. appearance of a long term ``bear phase''.
Let us for a moment imagine a financial market could be completely
determined by the action of just one powerful investor.
This is in contrast to traditional
economic theory which view the stock market as reflecting the state of economy,
i.e. the prices of stocks should reflect present earnings as well as
the expectation value of all future earnings of a given company. An
alternative view would be to consider the stock market as the ``motor''
of the economy illustrated e.g. by the so called ``wealth effect'', where
increasing stock market prices leads to higher consumer confidence and
spending, which in turns leads to larger earnings of companies making
stocks rise, etc. ad infinitum. With this in mind, one could
ask how big a fortune
would such an investor need and which expectations
about future developments in
interest rate and dividends would ensure
the investor to maintain a given growth of the market to his own profit?
In order to answer these questions, consider first the case of a large
mutual long-only fund that tries to control the market. The wealth of
the fund, $W$, at any given time $t$ is:
\be
W(t) = n(t) P(t) + C(t) ~, \label{wealth}
\ee
where $P(t)$ is the price of the market, $n(t)$ is the number of market shares held
and $C(t)$ the cash possessed by the fund at time $t$\cite{note1}.
At each time step the dominating fund buys new market shares in
order to push up the price of the market, creating an excess
demand of market shares, $A(t)$.
$A(t)$, give rise to the following equation
for the return $r(t)$ of the market \cite{bouchaudcont,farmer}:
\be
r(t) \equiv \ln (P(t+1)) - \ln (P(t)) = A(t)/\lambda , \label{price-eq}
\ee
where $P(t)$ is the price of the market
and $\lambda$ is the liquidity of the market. The fact that the price
goes in the direction of the sign of the order imbalance $A(t)$ is
well-documented
\cite{Holthausen,Lakonishok,Chan,Maslov,ChalletStin,Plerou}.
$A(t)$ will first be taken a constant, $A(t) \equiv A$ i.e. we have
\be
r(t) \approx {d \ln{P(t)} \over dt} = A / \lambda ; \ \ P(t) = e^{A t/\lambda}, \label{P}
\ee
and
\be
{d n(t) \over dt} = A ; \ \ n(t) = A t, \label{n}
\ee
Besides expenses to keep on buying shares, the mutual fund get an income
from dividends, $d(t)$, of the shares it's already holding, and from
interest rates, $r(t)$, of its cash supply $C(t)$. As will be seen the
expectation of future dividends $d(t)$ and future interest rates $r(t)$ is
crucial in designing a profitable strategy for the mutual fund. Since
the aim of this paper is to study long term trends where changes in
dividends can be much more important compared to changes in interest rates,
$r(t)$ will be taken constant, $r(t) \equiv r$, and the focus will be the
time dependence of $d(t)$. Consider first the simplest case where also
$d$ is constant: $d(t) \equiv d$. The balance equation for the cash supply
as a function of time therefore reads\cite{note5}:
\bea
{d C(t) \over dt} & = & -{d n \over dt} P(t) + C(t) r + n(t) d
+ C_{\rm
flow}(t,r(t),d(t),P(t),...)\\
& = & - A e^{A t/\lambda} + C(t) r +A t d
+ C_{\rm
flow}(t,r(t),d(t),P(t),...)
\label{dC}
\eea
The term $ C_{\rm flow}(t,r(t),d(t),P(t),...)$ describes the possibility of
additional inflow/outflow of
money into the fund and as indicated can be thought of to depend on time,
dividends, interest rates, the price of the market
as well as many other factors like e.g. tax cuts etc\cite{Sornette1}.
In practice the inflow/outflow of liquidity often contributes a
significant part to the total wealth of a mutual fund, but for
clarity $ C_{\rm flow}(t,r(t),d(t),P(t)...)$ will be taken equal to 0 first,
before discussing the relevance of various functional forms below.
It is preferable to express (\ref{dC}) in terms of the growth rate of
the financial market, $\alpha \equiv A/\lambda$ and the cash in
terms of the market liquidity, $\bar{C} \equiv C/ \lambda$ in which
case (\ref{dC}) become:
\be
{d \bar{C}(t) \over dt}
= - \alpha e^{\alpha t} + \bar{C}(t) r + \alpha t d \label{dC1}
\ee
The solution of (\ref{dC1}) reads:
\be
\bar{C}(t) = {\alpha e^{\alpha t} \over r - \alpha}
- {\alpha d t \over r}
- {\alpha d \over r^2}
+ { e^{r t} \over r - \alpha} \{- \alpha -{\alpha^2 d \over r^2}
+{\alpha d \over r}
+ C_0 (r - \alpha) \} \label{C1}
\ee
The condition to determine whether the mutual fund can impose
a given growth of the stock market is that $C(t) > P(t) \ \ \forall t$
, i.e. the cash supply should always be large enough that yet another
market share can be bought.
The time dependence of (\ref{C1}) is dominated by the
factors $e^{\alpha t}$ and $e^{r t}$ expressing the growth
due to the stock market and interest rates respectively. For
a stock market with a growth rate larger than the interest
rate, $\alpha > r$ the first term on the r.h.s. of (\ref{C1})
dominates and becomes negative. In this case the growth of the
stock market is not sustainable, and depending on the values
of $r,d, C_0$ the investor will inevitably face a critical time, $t_c$,
without enough money to secure further growth of the stock
market. This situation is illustrated by the liquidity curve b) in Fig.~1 for
the case $\alpha = 0.2, r=0.1$ and $d=0.02$ \cite{note6}:
after approximately 750
days the liquidity of the investor falls below the price of the
stock market given by curve a) and he can no longer secure the 20\%
growth of the stock market. Had the investor received four times
more in dividend, illustrated by curve c), $t_c$ would only be delayed
by some few tens of days. Increasing the initial amount of liquidity, curve
d), push $t_c$ further to the future, but eventually,
for a ``super'' interest
rate growth of the stock market, the investor will fall short
of money and the growth have to stop. On the other hand for
``sub'' interest rate growth $\alpha < r$ the term $e^{r t}$
dominates and a sustainable growth depends on the values of
$(\alpha,r,d)$. A $\alpha=7 \%$ growth, curve e), with interest rates
at $r= 10 \%$ and dividends $d= 2\%$ can be sustained if the initial amount of money is
large enough curve f) whereas lowering the initial liquidity the
growth no longer becomes sustainable, curve g).
The Eq.~(\ref{C1}) illustrates the case that the expectation of a
constant dividend $d$ is not sufficient for a super interest
rate market growth as e.g. was seen in the booming stock market growth
of the ninetees. Clearly in that case the rise of the stock market
was closely related to the sky rocketing expectations (justified or not)
of future dividends. Using the Efficient Market Hypothesis an
investor would expect
the dividend to follow the growth of the stock market,
and as will be seen super
interest rate growth become possible in that case.
If one take the dividend to increase proportional to the
price $d(t) \equiv d_0 P(t)$ (\ref{dC1}) takes the form:
\be
{d \bar{C}(t) \over dt}
= - \alpha e^{\alpha t} + \bar{C}(t) r
+ \alpha t e^{\alpha t} d_0 \label{dC_dP}
\ee
with the solution:
\be
\bar{C}(t) = \alpha e^{\alpha t} \{
{t d_0 - 1 \over \alpha - r}
- {d_0 \over (\alpha - r )^2} \}
+ e^{r t} \{ {-r \alpha + \alpha^2 + \alpha d_0
\over (\alpha - r )^2}
+ C_0 \} \label{C_dP}
\ee
Curve b) in Fig.~2 illustrates a solution of (\ref{C_dP}) where a given
amount of initial liquidity and a interest rate $r$ of 10\%,
a dividend $d_0$ of 2\% is not enough to sustain a stock market
growth of 20\% (curve a)). On the other hand, for a higher initial
dividend $d_0$ of 8\% and same initial amount of liquidity, super
interest growth of 20\% does become sustainable (i.e. $C(t) \geq P(t)
\ \ \forall
t$) as illustrated
by curve c). Sufficient initial funds are however required as illustrated
by curve d) where a high initial dividend of 8\% is not sufficient
to avoid that the investor runs out of money ($C(t) < P(t)$ for $t \approx
2800$). For each set of values $(d_0,C_0)$ there exists a range of
growth potentials, $\alpha$, for the investor $r < \alpha_{\rm low}
< \alpha < \alpha_{\rm high}$. If the investor chooses to push up the
market at a too slow growth rate $\alpha < \alpha_{\rm low}$ he eventually
will run out of money at large time scales, whereas for a too fast growth $\alpha >
\alpha_{\rm high}$ the investor will miss money to continue buying shares at
short time scales. For the case illustrated by curve c) in Fig.~2, $\alpha_{\rm
low}=0.18, \alpha_{\rm high}=0.32$.
The discussion so far has been on a large mutual long-only fund that
tries to benefit from a constant growth in a stock market. Eqs.(\ref{dC}) and
(\ref{dC_dP}) also describe the more morbid situation where a hedge fund
tries to benefit from a constant decline in a stock market by short selling.
This situation
is described in (\ref{dC_dP})
taking $\alpha$ and $d_0$ negative. $d_0$ should be taken negative since a
short seller has to pay dividend to the owner from which the stock
was borrowed. Instead of the condition $C(t) > P(t)$ that was needed
at all times for a mutual fund to keep on buying shares, a hedge fund
would need to ensure that $W(t) > |n| P(t)$, i.e. the wealth of the
hedge fund has to be sufficient that it can at any time buy back
all the shares $n$ that it has borrowed (and sold) at an earlier
time\cite{note2}. Figure~3 illustrates the case c) of Figure~2
with an interest rate of $r=10 \%$ and dividend $d_0= 8\%$ and the
same initial liquidity $C_0=10$ but now $\alpha=-20\%$ (curve a).
Since the wealth $W(t)$, curve c), is above $|n| P(t)$, curve b), it
is indeed sustainable (and very profitable (!) as seen from the wealth
curve) for a dominating hedge fund to force a continuous decline of
the stock market. Notice that the case of a dominating long-only
mutual fund that pushes the prices up versus the case of a dominating
short-only hedge fund that makes the market tumble is only
symmetric on short time scales.
This is illustrated in curve d) in Fig.~3 which shows the
wealth of a long-only fund with same parameters as the short-only
fund curve c), except for the signs of $\alpha$ and $d$. Since
the difference is however small on time scales relevant for
trading (e.g. the difference in wealth is less than 2\% after
5 years for the example given in Fig.~3) the difference will
be ignored in the following.
One can think of many different and relevant expressions of the
flow term $C_{\rm flow}(t,r(t),d(t),P(t),...)$. The simplest case
where customers add/subtract a constant flow of money per time unit,
$C_{\rm flow}(t,r(t),d(t),P(t),...)= \mu$, would not change anything
of the analysis presented of (\ref{dC}) and (\ref{dC_dP}) since
$e^{\alpha t}$ and $e^{r t}$ would still be the dominant factors.
Taking into account the interest rate in the inflow/outflow,
$C_{\rm flow}(t,r(t),d(t),P(t),...)= \mu e^{r t}$, would
create a dominant $\mu t e^{r t}$ term in the case of a constant
dividend (\ref{dC}). However in the more relevant case where the
dividend follows the price, $d =d_0 P(t)$ (\ref{dC_dP}) super
interest market growth would still be dominated by the
${t d_0 \over \alpha - r}\alpha e^{\alpha t}$ term and
the conclusions presented above would remain
unchanged. \ \\
The eqs. (\ref{dC}) and (\ref{dC_dP}) express the case of one single
powerful investor that has enough liquidity to drive the market. This
is clearly not a very realistic situation, but as will be seen it helps
understanding the much more realistic and complex case of
a group of investors that all try to drive the market for their own
profit. One very much studied market model with competing agents is
the so called ``Minority Game'' (MG) introduced by Challet and Zhang
\cite{minoritygames-standard}. In the ``Minority Game'' the
first approximation is to include only the direction of
a market move by representing a financial time series as
a binary time series $B(t)$ with 0 corresponding to a down
move and 1 to an up move.
Having in total $N$ agents, each agent $i$ possesses a
memory of the last $m$ digits of $B(t)$.
A strategy gives a prediction for the next
outcome of $B(t)$ based on the history of the last $m$
digits of $B$. Since there are $2^m$ possible histories, the total number
of strategies is given by $S=2^{2^m}$. Each agent holds the same number $s$
(in general different) of strategies among the $S$ possible strategies.
At each time $t$, every agent uses her most successful strategy (in
terms of payoff, see below) to decide whether to buy or sell
an asset. The agent takes
an action $a_i(t) = \pm 1$ where $1$ is interpreted as
buying an asset and $-1$ as selling an asset. The excess demand,
$A(t)$, at time $t$ is
therefore given as $A(t) = \sum_{i=1}^{N} a_i(t)$.
The return and the price of the market is as before given by (\ref{price-eq})
In the MG the payoff function of a strategy, $g^{\rm MG}_i(t)$,
was rewarding strategies that a each time step
belong to the minority decision:
$ g^{\rm MG}_i(t) = -a_i(t) A(t)$.
However as stressed in another paper,
(\cite{Vitting}),
in real markets, the driving force underlying the competition between investors
is not a struggle to be in the minority at each time step, but rather a fierce
competition to gain money. Introducing profit as the selection mechanism for
the strategies the payoff function was shown to take the form\cite{Vitting}:
$g^\$_i(t+1) = a_i(t) A(t+1)$
The same payoff function was also used in another market model by Giardina and
Bouchaud\cite{Giardina1,Giardina2}.
Consequently the market model with this payoff function was called the
$\$$-game as a reminder that profit drives the selection of strategies.
Figure~4 illustrates one price history (fat solid line) of the $\$$-game
with $N=20$ agents, $r=10\%$, $d_0=8\%$ assuming like before that the dividend
follows the market price $d(t)=d_0 P(t)$. The parameter values used were $s=4,
m=8, C_0=50$. Stochasticity was introduced via the initial strategies hold
by the agents as well as by $N$
additional ``noise'' agents
who
at each time step made a random decision to either sell or buy one
market share. The liquidity parameter, $\lambda$, in (\ref{price-eq})
was determined using the assumption that an univocal decision of
all the agents to sell an asset should lead to a ``crash'' of 10\%, i.e.
from the condition $r(t)=-2N/\lambda=-10\%$. The 10\% was chosen to
mimic some of the very largest crashes seen in financial markets.
Thin dotted lines represent the 5\%,
50\% and 95\% quantiles (from bottom to top) respectively, i.e. at every
time $t$ out of
the 1000 different initial configurations only 50 got below the 5\%
quantile line, 500 got below the 50\% quantile and 950 below the 95
\% quantile. The {\em average} behavior of an
agent\footnote {when using the word ``agent''
hereafter is meant to refer to the $N$ agents that use strategies in
the decision to buy or sell}
can now be
understood using the analysis of eq.'s (\ref{wealth})-(\ref{C_dP})
All the
$N$ agents in figure~4 were long-only, i.e. they correspond to the
case represented in figure~2 curve c). As predicted from the analysis
of (\ref{dC_dP}) the price $P(t)$ is {\em in average}
tilted towards positive returns.
Figure~5 represents simulations with the same parameter values as in
figure~4 except that a fraction, $\rho =0.20$ of the agents can be
both short and long. As can be seen from the 5\% quantile, the
introduction of agents that can take short positions clearly
increases the probability significantly for a lasting bearish
trend. Increasing $\rho$ to 0.4 amplifies this tendency, again
in accordance with (\ref{dC_dP}) as illustrated by the short-only
hedge fund figure~3 curve c. \ \\
Figure~7 shows the Nasdaq Composite index (fat solid line) as well as
statistics of short selling over the time period
1/1 2000 - 1/8 2003. Dotted line represents the fraction
of change in shares shorted, $n_s(t)$, defined
as the monthly total number of
change in shares shorted divided by the total monthly
share volume\cite{note4}. The dashed line represents the cumulative fraction
of change in shares shorted, $Q_s(t) = \sum_{t'=0}^t n_s(t')$.
Both $n_s(t)$ and $Q_s(t)$ have been multiplied with a factor
$10^5$ in order to make possible a comparison with the temporal
evolution of the Nasdaq index.
An interesting quantity in the present context
would be the short activity, $A_s(t)$, simply defined as the
total volume of short transactions divided by the total volume
of transactions. Notice that $n_s(t)$ gives the lower bound
on $A_s(t)$ since short positions opened and closed (or vice versa)
within a month are not counted in $n_s(t)$ whereas they would
be in $A_s(t)$. Since $n_s(t)$ typically varies
between 0.1-0.6 \% it is not unreasonable to assume $A_s(t)$ to
be one order of magnitude larger, i.e. some few \%. This means
that the short activity on the Nasdaq index could be in the
neighborhood that lead to a high probability of a lasting
bearish trend seen in e.g. Fig.~4-6.
Simple visual inspection
of the short trade data in Fig.~7 reveals an increase in the
short activity $n_s(t)$ prior to
or at the same time as dips in the market
(see e.g. spring and fall 2000, summer 2001, fall 2002).
Looking at the overall activity clearly confirms this pattern
with a steadily increase in the cumulative short trading activity
$Q_s(t)$ over the time period 2000 till summer 2001 and a
steadily decrease of the market in the same time period. It is
puzzling however, that one see an
continuous increase of short activity so far in 2003
where the market also has made an increase.
{\em Either} the market is right and short sellers will have to
cover their positions in further progress of the market (i.e.
$Q_s(t)$ should decrease in the future), {\em or} the
short sellers got it right and the market is in for a future
dip\cite{Sornette2}. \ \\
It is clearly not easy to estimate precisely what impact
short trading has on the evolution of financial markets. Still
this paper has tried to point out some of the dangers connected
to short trading as illustrated by the appearing of long term
bearish trends seen in the analysis of (\ref{dC_dP}) and by
the simulation results in Fig.~4-6. It should be noted from
Fig.~4-6 that even as the fraction of agents that use short
trading increase, the probability for a long term ``bullish''
phase remain unchanged as illustrated by the 95\% quantile lines
which are unchanged in all three cases. This highlights
that the danger with short selling is not when the markets are
already in a long term bullish phase, since short sellers then
are forced to cover their positions. The real danger is instead
when a downwards spiral of the markets has begun, in which case
an increase in short trading activity will only increase the
downward trend. Looking at the bad performance of the markets
over the past years together with the increase in short selling
activity, which was virtually absent just few years ago, the present
study suggest the possibility that financial markets could have entered
a new era not seen so far. \ \\
The author thanks D. Sornette for useful discussions.
\begin{thebibliography}{}
\bibitem{Fama}
E.F. Fama, Journal of Finance, {\bf 25}, 383 (1970).
\bibitem{Samuelson}
P. A. Samuelson, Industrial Management Review, {\bf 6}, 41 (1965).
\bibitem{Chen}
``Role of Speculative Short Sales in Price Formation:
Case of the Weekend Effect'', H. Chen, V. Singal, preprint (2001);
\bibitem{Platt}
``A Fuller Theory of Short Selling'' H. D. Platt, preprint (2002).
\bibitem{Ackert}
``Bubbles in Experimental Asset Markets: Irrational Exuberance No More'',L. F.
Ackert, N. Charupat, B. K. Church and R. Deaves, Federal Reserve Bank of
Atlanta Working Paper (2002).
\bibitem{note1}
By a market share is understood a portfolio of stocks from
which the financial market index is composed.
\bibitem{bouchaudcont} Bouchaud, J.-P. and R. Cont,
Eur. Phys. J. B {\bf 6}, 543 (1998).
\bibitem{farmer} Farmer, J.D., Int. J. Theo. Appl. Fin. {\bf 3}, 425 (2000).
\bibitem{Holthausen} Holthausen, R.W., R.W. Leftwich and D. Mayers,
J. Fin. Econ {\bf 19}, 237 (1987).
\bibitem{Lakonishok} Lakonishok, J. A. Shleifer, R. Thaler and R.W.
Vishny, J. Fin. Econ. {\bf 32}, 23 (1991).
\bibitem{Chan} Chan, L.K.C. and J. Lakonishok,
J. Fin. Econ. {\bf 33}, 173 (1995).
\bibitem{Maslov} Maslov, S. and M. Mills,
Physica A {\bf 299}, 234 (2001).
\bibitem{ChalletStin} Challet, D. and R. Stinchcombe,
Physica A {\bf 300}, 285 (2001).
\bibitem{Plerou} V. Plerou, P. Gopikrishnan, X. Gabaix and
H. E. Stanley, Phys. Rev. E {\bf 66}, 027104 (2002).
\bibitem{note5}
A similar set of eq.'s [1-5] were used in \cite{Giardina2} to
explain bubble formation in an agent based model.
\bibitem{Sornette1}
``Evidence of Fueling of the 2000 New Economy Bubble by
Foreign Capital Inflow: Implications for the Future of the US
Economy and its Stock Market'', D. Sornette and W.-X. Zhou,
preprint, Cond.-Mat/0306496
(2003).
\bibitem{Giardina2}
I. Giardina and J.-P. Bouchaud, Eur. Phys. J. B {\bf 31} 421 (2003).
\bibitem{note6}
Typical values of $(\alpha,r,d)$ will be used rather than
trying to model a specific market and time period.
\bibitem{note2}
For simplicity the possibility of leverage is ignored.
\bibitem{minoritygames-standard} For an introduction to the Minority Game see e.g.
the papers: D. Challet and Y.-C. Zhang, Physica A {\bf 246},
407 (1997); ibid {\bf 256} 514 (1998); R. Savit, R. Manuca and R. Riolo,
Phys. Rev. Lett. {\bf 82}, 2203 (1999);
A. Cavagna, Phys. Rev. E {\bf 59} R3783, (1999).
See also the special webpage on
the Minority Game by D. Challet at
$www.unifr.ch/econophysics/minority/minority.html$
For more specific applications of the Minority Game to financial markets see
e.g. D. Challet et al., Quant. Fin {\bf 1}, 168-176 (2001);
M. Marsili, Physica A {\bf 299}, 93, (2001);
N. F. Johnson et al., Int. J. Theor. Appl. Fin. {\bf 3}, 443 (2001);
Johnson N. F. et al.,
Physica A {\bf 299}, 222 (2001); Lamper, D., S. D. Howison and N. F. Johnson,
Phys. Rev. Letts. {\bf 88}, 017902, U190-U192.
\bibitem{Vitting} J. Vitting Andersen and D. Sornette, Eur. Phys. J. B. {\bf
31} 141 (2003);
\bibitem{Giardina1} I. Giardina, J.-P. Bouchaud, Physica A {bf 299} 28 (2001);
\bibitem{note4}
The short date can be retrieved directly without charge at the
Nasdaq Composite home page:
$www.nasdaqnews.com/MarketData/News\_mrktdata\_home.htm$
\bibitem{Sornette2}
Such a dip has actually been predicted to happen:
D. Sornette and W.-X. Zhou, Quantitative Finance {\bf 2}, 468 (2002); See also
D. Sornette and W.-X. Zhou, ``Predictability of Large Future Changes in Complex
System'', preprint at cond-mat/034601, D. Sornette ``Why Stock Markets Crash''
(Critical Events in Complex Financial Systems) (Princeton University Press,
Princeton, NJ 2003).
\end{thebibliography}
\begin{figure}[h]
\includegraphics[width=16cm]{figure1.eps}
\caption{\protect\label{Fig1}
a) Price $P(t)$ of (\ref{P}) as a function of time $t$ for
$\alpha = 0.2$.
b) Cash $C(t)$ solution of (\ref{dC}) for
the case $\alpha = 0.2, r=0.1, C_0=2$ and $d=0.02$.
c) Similar to curve b) but with $d=0.08$. d) similar to c) but
with $C_0=5$.
e) Price $P(t)$ of (\ref{P}) for
$\alpha = 0.07$.
f) Cash $C(t)$ solution of (\ref{dC}) for
the case $\alpha = 0.07, r=0.1, C_0=5$ and $d=0.02$.
c) Similar f) but with $C_0=2$.
}
\end{figure}
\begin{figure}[h]
\includegraphics[width=16cm]{figure2.eps}
\caption{\protect\label{Fig2}
a) Price $P(t)$ of (\ref{P}) as a function of time $t$ for
$\alpha = 0.2$.
b) Cash $C(t)$ solution of (\ref{dC_dP}) for
the case $\alpha = 0.2, r=0.1, C_0=10$ and $d=0.02$.
c) Similar to curve b) but with $d=0.08$. d) similar to c) but
with $C_0=5$.
}
\end{figure}
\begin{figure}[h]
\includegraphics[width=16cm]{figure3.eps}
\caption{\protect\label{Fig3}
a) Price $P(t)$ of (\ref{P}) as a function of time $t$ for
$\alpha = -0.2$.
b) $|n|P(t)$ for
the case $\alpha = -0.2, r=0.1, C_0=10$ and $d=-0.08$.
c) Wealth, $W(t)$ as a solution of (\ref{dC_dP}) and (\ref{wealth})
for the parameters of curve b). Same as curve b) except
for signs of $\alpha, d$, $\alpha = 0.2$ and $d=0.08$.
}
\end{figure}
\begin{figure}[h]
\includegraphics[width=16cm]{figure4.eps}
\caption{\protect\label{Fig4}
Fat solid line: price $P(t)$ from one configuration of the $\$$-game
with $N=20$ agents, $r=10\%$, $d_0=8\%$ (assuming dividend
$d(t)=d_0 P(t)$). The parameter values used were $s=4,
m=8, C_0=50$. Stochasticity was introduced via $N$
additional ``noise'' agents. The fraction of agents allowed
to take short positions, $\rho =0$.
Liquidity parameter $\lambda=0.0025$.
Thin dotted lines represent the 5\%,
50\% and 95\% quantiles (from bottom to top) respectively, i.e. at every
time $t$ out of
the 1000 different initial configurations only 50 got below the 5\%
quantile line and similarly for the other quantile lines.
}
\end{figure}
\begin{figure}[h]
\includegraphics[width=16cm]{figure5.eps}
\caption{\protect\label{Fig5}
Same as figur~4 but with $\rho=0.2$
}
\end{figure}
\begin{figure}[h]
\includegraphics[width=16cm]{figure6.eps}
\caption{\protect\label{Fig6}
Same as figur~4 but with $\rho=0.4$
}
\end{figure}
\begin{figure}[h]
\includegraphics[width=16cm]{figure7.eps}
\caption{\protect\label{Fig7}
Fat solid line represents the Nasdaq Composite index.
Dotted line represents the fraction
of change in shares shorted, $n_s(t)$, defined
as the monthly total number of
change in shares shorted divided by the total monthly
share volume. The dashed line represents the cumulative fraction
of change in shares shorted, $Q_s(t) = \sum_0^t n_s(t)$.
Both $n_s(t)$ and $Q_s(t)$ have been multiplied with a factor
$10^5$.
}
\end{figure}
\end{document}