Friend-of-a-Friend in Production Networks: Micro Estimates and Macro Implications

Abstract

We quantify the friend-of-a-friend effect in production networks, where shared partners act as relational capital facilitating new links. We develop a general equilibrium (GE) model that endogenizes link formation and incorporates the friend-of-a-friend mechanism. We show the model reduces to a dyad-level logit specification, enabling estimation via a quadruple-based conditional logit. Using Japanese firm-to-firm transaction data, we find a sizable friend-of-a-friend effect, comparable in magnitude to the effects of geographical distance and sectoral proximity. Counterfactual simulations in a calibrated GE model indicate that eliminating the mechanism reduces aggregate output by 0.6% and reshapes firm-level shock propagation by changing network topology.