Industry Interdependencies and Cross-Industry Return Predictability

We use the adaptive LASSO from the statistical learning literature to identify economically connected industries in a general predictive regression framework. The framework permits complex industry interdependencies, including both direct and indirect sectoral links. Consistent with gradual information diffusion across economically connected industries, we find extensive evidence that lagged returns of interdependent industries are significant predictors of individual industry returns. Using network analysis, we detect a significant relation between an industry’s importance in the U.S. production network and the pervasiveness of the predictive power of the industry’s lagged return. We also compute out-of-sample industry return forecasts based on the lagged returns of interdependent industries and show that crossindustry return predictability is economically valuable: an industry-rotation portfolio that goes long (short) industries with the highest (lowest) forecasted returns exhibits limited exposures to a variety of equity risk factors, delivers substantial alpha, and performs very well during the recent Global Financial Crisis. JEL classifications: C22, C58, G11, G12, G14