A dynamic herding model with interactions of trading volumes is introduced.
At time $t$, an agent trades with a probability, which depends on the ratio of
the total trading volume at time $t-1$ to its own trading volume at its last
trade. The price return is determined by the volume imbalance and number of
trades. The model successfully reproduces the power-law distributions of the
trading volume, number of trades and price return, and their relations.
Moreover, the generated time series are long-range correlated. We demonstrate
that the results are rather robust, and do not depend on the particular form of
the trading probability.