Statistical Discrimination and E ciency

This paper addresses a fundamental, yet unresolved, question: is statistical discrimination a market failure? I contrast a competitive model of statistical discrimination with the allocation chosen by a social planner who must respect the same informational and technological constraints as the rms in the decentralized model. Equilibria are always contrained ine cient due to \free riding" in human capital investments, but this market failure is present also when the market coordinates on an equilibrium without discrimination and does not mean that statistical discrimination is ine cient as a phenomenon. On the contrary, there are e ciency gains from discrimination in terms of reduced \mismatch" between workers and jobs. Whether the solution to the planning problem involves discrimination depends on the trade-o between the informational gains of specialization and the losses in terms of increased investment costs.