Sentiments, strategic uncertainty, and information structures in coordination games

We study experimentally how changes in the information structure affect behavior in coordination games with incomplete information (global games). We find two systematic departures from the theory: (1) the comparative statics of equilibrium thresholds and signal precision are reversed, and (2) as information becomes very precise subjects' behavior approximates the efficient equilibrium of the game, not the risk dominant one. We hypothesize that sentiments in the perception of strategic uncertainty could drive our results. To formalize this hypothesis we extend the standard model by introducing sentiments and we test this mechanism experimentally by eliciting beliefs. We find empirical support for our hypothesis: Subjects are over-optimistic about the actions of others when the signal precision is high and over-pessimistic when it is low. Thus, we show how changes in the information structure can give rise to sentiments that drastically affect outcomes in coordination games.