This paper develops a search and matching model with heterogeneous firms, on-the-job

search by workers, Nash bargaining over wages and adaptive learning. We assume that

workers are boundedly rational in the sense that they do not have perfect foresight about the

outcome of wage bargaining. Instead workers use a recursive OLS learning mechanism and

base their forecasts on the linear wage regression with the firm's productivity and worker's

current wage as regressors. For a restricted set of parameters we show analytically that the

Nash bargaining solution in this setting is unique. We embed this solution into the agentbased

simulation and provide a numerical characterization of the Restricted Perceptions

Equilibrium. The simulation allows us to collect data on productivities and wages which is

used for updating workers' expectations. The estimated regression coefficient on productivity

is always higher than the bargaining power of workers, but the difference between the two

is decreasing as the bargaining power becomes larger and vanishes when workers are paid

their full productivity. In the equilibrium a higher bargaining power is associated with

higher wages and larger wage dispersion, in addition, the earnings distribution becomes

more skewed. Moreover, our results indicate that a higher bargaining power is associated

with a lower overall frequency of job-to-job transitions and a lower fraction of inefficient

transitions among them. Our results are robust to the shifts of the productivity distribution