We develop a theoretical model based on efficient bargaining, where both log outside productivity and log productivity in the current job follow a random walk. This setting allows the application of real option theory. We derive the efficient worker-firm separation rule. We show that wage data from completed job spells are uninformative about the true tenure profile. The model is estimated on the Panel Study of Income Dynamics. It fits the observed distribution of job tenures well. Selection of favorable random walks can account for the concavity in tenure profiles. About 80% of the estimated wage returns to tenure is due to selectivity in the realized outside productivities.