An Empirical Examination of Shift-Share Instruments

Bartik's (1991, 1993) approach to identifying shocks in demand to regional economies has been used extensively for nearly thirty years. We chronicle the development of Bartik-type shift-share instruments and examine the empirical performance of alternative versions that use different combinations of national shift and local share variables in their construction. We offer three main findings. First, instruments constructed from shares that omit employment in non-traded sectors empirically dominate versions that include total employment. Second, industrial sectors with high average shares and low variation across areas are more likely to be non-traded and endogenous. This suggests placing large weights on non-traded sector shares worsens both relevance and potential endogeneity. Finally, we demonstrate national shifters other than employment, such as prices and wages, can be used to construct instruments with unique and relevant explanatory power