We examine the effect of information externalities in an on-the-job search model with multiple agents. In each period agents earn a wage and can choose to engage in costly search for a better paying position. Three different information spillovers are analyzed: no information, full information and partial information spillover. The equilibrium strategies of the agents are identified for each information structure and the model is tested in a series of experiments using human subjects. The experiment results indicate that when subjects have no information regarding the other subjects' wages or employing firms, they tend to search less than the model predicts. With full information spillover, when information about others' wages and employing firms is revealed, search becomes less frequent, as predicted by model. However, evidence suggests that subjects tend to free ride on the choice of others more than predicted by the equilibrium. Theory implies that with partial information spillover, i.e. when only information about wages earned is revealed, agent search choice should be identical to the no information case. However, contradictory to theoretical prediction, some subjects are more likely to search, especially those with lower wages than their peers. We trace this divergence to a comparison effect, also know as "catching up with the Joneses" effect. This paper shows that when examining search behavior, one needs to take into consideration and identify the information externalities facing the searchers, as these change search choices in significant ways.
Nash equilibrium can be interpreted as a steady state of a game where players hold correct beliefs about the other players' behavior and act rationally. In this paper, we experimentally examine the process that leads to this steady state. We find: (1) For non-dominance solvable games the existence of a teacher greatly facilitates (is necessary for) convergence to Nash equilibrium. (2) For dominance solvable games, while a teacher is again a facilitator for convergence when the time horizon of the game is relatively short (20 periods), if the horizon is long enough subjects appear to be able to converge by iteratively eliminating dominated strategies, even if teaching is difficult. (3) The ability to see one's opponent's payoffs is essential for convergence. (4) A successful model of belief formation should include a component taking a player's opponent's payoffs into account.