The ability of an economy to generate, recombine and diffuse new ideas has a profound influence on its ability to sustain growth. Agglomeration, by eliminating the physical space between people, firms and ideas, affects not only the rate of innovation, but also the type of idea recombinations that take place. While the role of co-location on innovation has been extensively studied, the micro-foundations of knowledge transfer and generation still remain undeveloped.

Co-location, by increasing the frequency of interactions with very low expected value, qualitatively transforms the knowledge that is recombined, allowing for high levels of "exploration". This is particularly important when the uncertainty surrounding idea quality is high (e.g. early-stage research) and the outcome distribution is very skewed (e.g. scientific breakthroughs).

Leveraging a natural experiment, I analyze how geographic proximity impacts the recombination of ideas using a novel dataset of department relocations at UPMC Paris (1995-2010).

Collaborations across departments pairs are compared before and after co-location, a change that disproportionately affects low expected value interactions. Results are consistent with the idea that proximity has a profound influence both on the rate and the direction of innovative activity.