Consider a one step forward looking self-referential model with a one dimensional state variable where the information set available to agents when they make their forecast for the next period includes the current realization of an extrinsic random process but does not include the current value of the state variable. Agents thus iterate twice on their beliefs about the law of motion of the system to generate their forecasts; this is as in models of learning and in contrast to the specification considered in the literature on sunspot equilibria where the extrinsic random process is the state variable. This paper demonstrates (and characterizes the conditions for) the existence of self-fulfilling stochastic equilibria with bounded fluctuations driven purely by extrinsic beliefs for the model described above; furthermore, the existence of these equilibria is shown to be independent of the determinacy properties of the perfect foresight dynamics of the model. (Traditional sunspot equilibria appear as a special case of the formulation of the paper). The paper indicates that the problem of multiplicity of rational expectations equilibria in these models is more severe than believed erstwhile.
