We propose a model for stochastic formation of opinion clusters,
modelled by a evolving network, and herd behaviour to account for the observed
fat-tail distribution in returns of financial-price data. The only parameter
of the model is
h, the rate of information dispersion per trade,
which is a measure of herding behavior. For
h below a critical
h*the system displays a power-law distribution of the returns with exponential
cut-off. However for
h >
h* an increase in the probability
of large returns is found, and may be associated to the occurrence of large
crashes.
PACS: 87.23.Ge, 02.50.Le, 05.45.Tp, 05.65.+b
Also at LANL preprint server as paper cond-mat/9908069.