Abstract
We investigate the role of information about dividends in expectation formation and price dynamics. We introduce a novel experimental design within the learning-toforecast framework that combines predictions about dividends, modelled as an exogenous stationary variable, with predictions about prices endogenously determined by a positive feedback mechanism. We vary the salience of the dividend across the treatments, with participants able to observe its mean value only, its realized series, or required to forecast it. Our results suggest that the way dividend information is presented shapes whether market prices remain anchored to fundamentals or depart from them. Increasing the salience of the information about dividends significantly changes market dynamics: large and persistent bubbles shrink substantially and prices remain surprisingly stable around the fundamental.