Gay, Sebastien and Karger, Ezra, Patent Citations and Stock Performance: Constructing a Dynamic Industry Classification (September 15, 2014). Available for download at SSRN: http://ssrn.com/abstract=2496414

“Researchers in academia and the private sector use industry classifications to compare growth across and between industries, to construct industry indices, and to control for industry-level correlations of stocks over time. But commonly used industry classifications do not reliably predict stock co-movement because companies change their core structure quickly relative to the slow and haphazard updating of the widely-used subjective industry classifications. We propose an objective industry classification that clusters companies based on a network of patent citations to better measure changes in the relationships between companies over time. Our cluster-based industry classification predicts daily stock co-movement between 5% and 25% better than the standard SIC and NAICS classifications. Our classification is also a statistically significant addition to a standard 3-factor Fama-French model and outperforms SIC and NAICS classifications in its ability to explain variance in daily and monthly stock returns. While the ability of Standard Industry Classification (SIC) codes and the North American Industry Classification System (NAICS) to group together companies with similar daily stock returns has deteriorated over time, our clustering framework consistently groups together companies with similar daily stock returns in each year between 1947 and 2010.”