We consider a dynamical model of trading agents in which the flows of cash and goods among a large number of agents are treated as explicit variables. Each agent’s assets and individual price setting, as well as their averages over the entire trading system are traced as outputs for a long time period. We have observed that the time series of the average price exhibit two different types of oscillatory patterns, depending on the regions in the parameter space. The first type (C-type) stays chaotic throughout the period of the simulation and the second type (P-type) shifts the pattern from chaotic to periodic in the course of trading activity. The latter case accompanies the gradual loss of active agents and eventually fall into periodic motions of a small number of naturally-formed groups in which a number of agents move identically, as if the entire system tunes itself to a synchronized motion. Also the statistical property of the latter case (P-type) can be regarded as Le´vy’s stable distribution, while the former case (C-type) is more like Gaussian. ©2000 American Institute of Physics.