Networked instrumental variable estimation: The case of Hausman-style instruments

In this paper, I argue that in situations of complex network dependence, the traditional and widely used Hausman-style instrumental variable estimation may not be valid for causal identification. This is the case for inter-regional migration networks when evaluating place-based labor market policies, and for correlated unobserved consumer tastes in the product and geographic space in demand estimation. I build an economic model for these two cases, respectively, to derive the estimating equation and to shed light on the fallacy---omitted variable bias and the resulting violation of exclusion restriction---of the traditional econometric framework. I then build an alternative econometric framework and propose a new approach to estimation that exploits higher-order network neighbors and, then, I establish its desirable properties. I conduct Monte Carlo simulations and two empirical analyses that each correspond to the two economic models to validate this new approach of estimation.