2016 CARD Award for Best Ph.D. Dissertation in Agricultural, Environmental, and Energy Economics and Policy
Combinatorial Innovation, Evidence from Patent Data, and Mandated Innovation
This paper presents an original model of knowledge production, and tests several predictions of the model using a novel dataset built from 8.3 million US patents. In this model, new ideas are built by combining pre-existing technological building blocks into new combinations. The outcome of research is always stochastic, but firms are Bayesians who learn which sets of technological building blocks tend to yield useful discoveries and which do not. Consistent with this model’s prediction, I show that the number of patents granted in a particular technology class increases in the years after new useful combinations of technology first appear in the class. Moreover, after new combinations first appear, I show subsequent patents are more likely to draw on the same combination of technology, consistent with firms learning the technologies can be fruitfully combined. Patents are also more likely to combine technologies that have already been combined with many of the same (other) technologies, even if they have never been combined with each other. Finally, I show that the probability of using a combination declines over time, and that the total number of patents granted in a technology class also declines over time, if there are not new connections between technologies continuously discovered. This is consistent with the model’s predictions about firms using up all the useful ideas that can be built from a fixed set of technological building blocks.