Quantification of Spillovers and Recovery of the Social Network Structure: Relaxing the Assumption of Persistence

Quantification of spillovers in social network analysis requires information on the social ties between agents. As information on the social structure is rarely available, recent research such as De Paula et al. (2018) and Manresa (2016) have proposed methods that can quantify spillovers and recover the social network structure. Identification depends on the assumptions of a sparse and persistent network structure. This paper relaxes the assumption of persistence by means of structural breaks. In addition, the importance of different scenarios of structural breaks is inspected with focus on whether the break affects the spillovers. Next to that, a flexible algorithm is proposed to detect time-invariant social interactions. A simulation study has suggested that additional information on which parameters pertain the break date is valuable. Relative to a baseline model that assumes all parameters alter at the break, the flexible algorithm has the advantage in break date estimation, parameter estimation and network recovery for settings with at least partial persistence of social interactions. The detection of persistent spillovers is demonstrated to benefit predominantly the identification of key players in the network. The proposed algorithms are illustrated with an empirical application to R&D spillovers in the Electronics industry. Future research is necessary for informative inference and is recommended to explore more directions of relieving the assumption of a persistent network structure.