Topology and Formation of Production Input Interlinkages: Evidence from Japanese microdata

Recent studies have emphasized the role of production input interlinkages in explaining a wide range of economic phenomena. In the vast majority of these studies, however, the input-output architecture is fixed and exogenously given, and hence relatively little is known about the evolution of the production network and the underlying mechanisms shaping its dynamics. Here we seek to fill this gap in the extant literature by studying the evolution of production input interlinkages on the most granular level of economic activity, building on a diverse set of more than 80,000 companies sampled across nearly all industries of the Japanese economy. We find that several network properties with a pronounced impact on shock propagation and the emergence of aggregate fluctuations are remarkably stable over time and invariant under the local link formation mechanism. To estimate the mechanism inducing this stability, we employ a stochastic actor-oriented model that resolves the problem of interdependent observations inherent to networked environments. This model approaches the dynamics of the production network from the perspective of individual firms whose myopic decisions to change their suppliers take into account the effect of direct connections and link externalities. Building on this approach, we find that topological features of the network such as network distance and the current number of relationships are a main driver of network dynamics in subsequent periods, and are quantitatively more important than productivity differentials.