Count Models of Social Networks in Finance

Abstract

Social networks are thought to be important for the investment and performance of mutual fund managers. We propose a measure of whether a manager is part of a network using only data on the distribution of the number of stocks headquartered in a given city that are held by managers. For some cities, the count distribution is roughly Poisson. However, for a significant fraction of cities, the count distribution is highly overdispersed Poisson --- where most managers have a couple of picks but a few managers have many picks. We show that the degree of overdispersion is a theoretically well-motivated measure of network influence and that managers with concentrated stock picks in a city are likely to be part of a network in that city. These managers indeed significantly outperform other managers by around 1.6% per annum.

Keywords: Social Networks, Poisson Regressions, Investor Behavior

JEL Classification: G1, G2, G3