Stationary and nonstationary stochastic processes P. Jung, Rev. Mod.
Phys. 234, 175 (1993) occur in a variety of phenomena as different
as Brownian motion A. Einstein, Ann. Phys. 17, 549 (1905),
Johnson noise J. Johnson, Phys. Rev. 32, 97 (1928), stellar
dynamics S. Chandrasekhar, Rev. Mod. Phys. 15, 1 (1943), and
quantum optics H. Risken, in Progress in Optics.
Besides in physics, stochastic processes have been successfully applied in
economics for modeling and thus explaining diverse levels of economics
systems, ranging from the ``micro'' level of company products to the
``macro'' level of company sizes and even national economies. Recently, Fu
et al.\ D. Fu et al., Proc. Natl. Acad. Sci. USA\/ 102, 18801
(2005) show that for different economic variables from both the micro and
the macro level, the distribution of logarithmic growth rates are
approximately (i) exponential in the central part, (ii) power-law decaying
in the tails, and that there is (iii) a monotonically decreasing power-law
relation between the company sales and the standard deviation of logarithmic
growth rates.
Fu et al.\ propose a process recently cited in the Handbook of Industrial
Organization Volume 3, edited by Robert Porter and Mark Armstrong for
modeling the empirical observations (i) and (ii), but this model fails to
reproduce observation (iii).
For modeling observations (i)--(iii), we propose the multiplicative
stochastic process of logarithmic growth rates
equation
R_t łnłeft(S_tS_t-1\right)=\mu_0\Delta t +
(S_t-1)^\gamma\sigma_0\eta_t\Delta t,
equation
where $\sigma$, $\gamma$, and $\mu$ are three parameters, $\eta_t$ is an
i.i.d.\ Gaussian noise, and $S_t $ is the random variable.
When the parameter $\gamma$ introduced for modeling the dependence of the
standard deviation $\sigma(R_t)$ on the size $S_t$ is set equal to zero, the
stochastic process reduces to geometric Brownian motion, the most
widely employed stochastic process in finance. The process can
also be related to the Ornstein-Uhlenbeck process, a well-known stochastic
process introduced in physics.
For different time series of logarithmic growth rates $R_t$ with
$\gamma=-0.15$, we calculate the average size $S \rangle$ and the
standard deviation $\sigma(R_t)$. Fig.~1(a) shows that, due to $< 0$,
$\sigma(R_t)$ versus $S\rangle$ scales as a power law $\sigma(R_t)
S\rangle^\beta$, where $\beta=\gamma$.
We find in Fig.~1(b) that for $\gamma=-0.15$ the central part of
distribution $P(R_t|S_0)$ can be approximated by an exponential
distribution, and Fig.~1(c) shows that the far tails of $P(R_t|S_0)$ can be
approximated by power-laws, where the parameter $\sigma$ controls the
power-law exponent.
We also find in Fig.~1(d) that four important macroeconomic
variables, (export, import, debt, and investments) exhibit the same
properties (i)-(iii).