We investigate the underlying sources of gender and race differences in labor market outcomes over the life-cycle using a canonical version of Mortensen and Pissarides (1994, MP henceforth). The model exhibits learning by doing so that non-employment spells are particularly costly for human capital accumulation. We use the model to reverse engineer life-cycle human capital stocks and matching technologies, for each gender and race pair, needed to exactly match the stylized life-cycle patterns of wages and job finding rates for the unemployed and the non-participants. We find that search frictions play a central role in shaping life-cycle profiles. While wage profiles properly describe the human capital of workers in frictionless models, wage and human profiles differ significantly in the presence of frictions. Wages peak at around age 50-55 but human capital keep increasing until retirement in the MP model. Moreover, while wage differentials among race and gender first widen and then compress during the life-cycle, human capital always widen. As a result, the MP model predicts very different returns to experience than what is suggested by Mincer regressions on wages. Our exercise also indicates large differences in matching technologies, or hiring costs, by race and gender favoring males and particularly unfavorable for the old and Asian females.
