How large are geographic “knowledge spillovers”? That is, when firms are located near firms in the same area, how much does knowledge of new invention spill across firm boundaries? Zvi Griliches, in particular, wrote a great deal of theory on this question before he passed away, but empirical estimates are, unsurprisingly, difficult to uncover. Clearly, an answer to this question will also depend on how spillovers are defined, a point I’ll return to later.
Jeff Furman and his coauthors attempt to provide some semblance of an answer by examining the pharma industry. They construct an enormous database of firm patents coded by theraputic class (an indicator of what area of medicine the patent concerns) and lab location. They also construct a database of research articles on those classes, also coded by location. In order to account for lags in knowledge spillover, they assume knowledge depreciates at 15% per year, and estimate stocks of knowledge about a given class at each location. A straightforward set of regressions finds evidence that an increased stock of research about a theraputic class in a given urban area increases patents from other private firms in that area. In particular, 1000 more local publications about a theraputic class leads to .19 marginal patents, whereas 1000 more global publications about that class leads to only .11 marginal patents. Further, only research by public firms spills over. Private research from other nearby firms does not.
A worry here comes from the fact that estimates are wholly reduced form. There is no micro data, no data on licensing, no data on researchers hired at both universities and firms, etc. Because of this, there is no structural model that might hint at why, exactly, such spillovers occur, or why private research does not spill over. In particular, a lack of private research spillover would seem to contradict the common urban econ assumption of spillover among closely-located firms which is quite important for generating the pattern of industrial clustering that occurs in many industries, and especially in R&D intensive ones. Another worry is that I, for one, don’t really know what a spillover is supposed to be. If a spillover is an externality, then it must occur outside of the market. But this paper does not have data which might tell me whether the increased patenting is due, say, to joint work among firms, or among firms and universities, as some part of a contractual relationship. Therefore, perhaps this paper should be read only as another (particularly data-intensive) piece of evidence supporting the geographic importance of knowledge spread – the exact micromechanisms need further evidence.