A (jump) Markov process (generalized birth-and-death process) is used to model interactions of a large number of agents subject to field-type externalities. Transition rates are (nonlinear) functions of the composition of the population of agents classified by the choices they make. The model state randomly moves from one equilibrium to another, and exhibits asymmetrical oscillations (business cycles). It is shown that the processes can have several locally stable equilibria, depending on the degree of uncertainty associated with consequences of alternative choices. There is a positive probability of transition between any pair of such basins of attraction, and mean first-passage times between equilibria can be different when different pairs of basins are calculated.