Rory Mullen

Technology is central to economics, but current technology datasets fail to meet the needs of economists because they lack scale, scope, span, or specificity. We apply natural language processing and positive and unlabeled machine learning to patent descriptions and business descriptions from U.S. public firms over three decades to produce a firm-level technology dataset that offers an unrivaled combination of these characteristics. We identify the core technologies of patent non-owning firms, estimate the value to these firms of innovations to the core technologies they use, and show how stock market participants inefficiently process technological information on these firms, leading to profitable trading strategies.

We examine the impact of investment completion risk on the stock market performance of clean energy firms, isolating the completion risk channel in a theoretical model and testing the channel using project-level data from the U.S. nuclear and wind industries. We find that completion risk substantially raises the cost of capital for clean energy investments. For the nuclear industry, a 1 percentage point rise in completion risk corresponds to a 36 basis point rise in stock returns, after controlling for investment cost. In a case study, we interpret heightened regulatory scrutiny following the 1979 Three Mile Island nuclear accident as a shock to completion risk and find that completion risk strengthens as a return predictor after the accident. In Fama-MacBeth regressions, we find that nuclear completion risk is priced, particularly for firms with complex technologies. Evidence from the renewable wind industry supports our main findings from the nuclear industry.