Regional potential and friction for technological diversification: a machine learning frontier approach

Abstract Related diversification is a pillar of industrial policy in the context of the geography of innovation. The idea that it is easier for economies to diversify into related economic activities is understandably popular and common. However, behind this idea lie some political and empirical assumptions that are rarely discussed. First, related diversification is easier than unrelated diversification. And second, the existence of related capacity is enough to diversify into new capabilities. The aim of this paper is to provide a context for these assumptions within an approach that goes beyond path dependency. We use data on patents and other socio-economic factors in European regions to demonstrate our approach. We first estimate potential capacity for each region in our data via a combination of 7051 random forest models that capture diversification possibilities agnostic to relatedness. Then we estimate the distance between potential and observed capacity as an inefficiency (or friction) measure. We then assess what contributes to its variation. Our results show that regional friction provides an additional layer of explanation for national economies and is mainly driven by unemployment and population density.