We infer the latent social networks of investors using data on their stock holdings. We map linkages to portfolio weights using a portfolio-choice model. The precision of an investor’s private signal about firm value is assumed to increase with his connections in the city where the firm is headquartered. Using money-manager data, we find that managerial linkages to a city are overly dispersed relative to the Erdös–Rényi model of i.i.d. connections. Managers at the tail of this distribution with non-i.i.d. linkages have more university alumni in that city. Their stock holdings there outperform their holdings in other cities.

Households hold undiversified stock portfolios of firms headquartered near their city of residence. Leading explanations assign a causal role for proximity. The literature neglects that distance is endogenous. Households may locate based on unobservables such as optimism about a city’s economic prospects, which can be correlated with latent local-stock demand. We use location-choice models to account for this selection. We propose as instruments that older households prefer to locate in recreational areas for non-pecuniary reasons. Our analysis based on a widely used household data set yields significantly smaller estimates for proximity in determining portfolio choice compared to those in the literature.