Trade, Networks, and Innovations: Evidence from German Patent Data [Job Market Paper]
This paper investigates the relationship between trade and innovation using newly digitised German patent and transport data from the late 19th and early 20th centuries. Germany was highly innovative during this period of time. Innovative activity was more spread out geographically within Germany than other countries. The reasons for this geographic pattern remain an open question. The literature on this question focusses on estimating the effect of market access (measured as the distance-weighted sum of economic activity in a particular location) on innovation. The German data demonstrate the limitations of this popular approach. Many large German cities with high market access were indeed centres of innovation, but many were not. My data allow me to consider network effects not captured by standard notions of market access. I find that network measures such as betweenness and eigenvector centrality are essential to a more complete understanding of the determinants of innovation. In particular, these network measures matter for innovative activity, even controlling for market access and local fixed effects.
Constructing the network measures requires comprehensive information on interactions among economic agents - data that are rarely observable today because of privacy concerns. My data much better approximate the interactions among agents in the German economy because of the limited number of communication mechanisms at that time. Because market access is potentially endogenous, I follow prior literature, such as Donaldson and Hornbeck (2016), in using instrumental variables to eliminate such concerns. As an alternative method of identification, I also use the Chamberlain (1983) panel-data approach to static and dynamic fixed-effect models and test the overidentifying restrictions inherent in such models. All of these approaches yield similar results.
My main finding is that the estimated effects of these network measures are heterogeneous across many dimensions, most notably the sector’s degree of product differentiation. When I disaggregate the patent data by sector, the results show the importance of product differentiation increasing the network effects for many of the sectors. Market access only affects innovation in sectors with low degrees of product differentiation. In contrast, innovations of highly differentiated products take place in locations with high betweenness centrality. This disaggregation also yields new insights on the role of individual inventors (as opposed to those employed by a firm), who were very important in the US, but not in Germany: individual German inventors were mainly operating in sectors where the estimated effect of market access is relatively unimportant. However, controlling for the sector of the innovation, I find that individual inventors tend to concentrate in locations that have easy market access.
Work in Progress:
Collaborating and Signalling: An online Experiment (with D Gupta)
New insights from patterns of information flow in Germany in the early 20th century