Various studies report that the ability of industry indexes to predict the broad market disappeared during the most recent years. I revisit this theme using more flexible switching models and imposing economically motivated constraints on the predictions. My results show that traditional constant coefficients linear models are unable to forecast the stock market over the period considered, but restricting the equity premium to be non-negative, five industries predict the market. I also show that the Markov-switching models exhibit a dismal performance, which is even worse than the ones from the constant coefficients model. Finally, I test a model with two regimes- recession and expansion- which are identified in real-time through the Arouba-Diebold-Scotti Business Conditions Index. Using this model, I find that 8 out of 33 industries can successfully forecast the market. Furthermore, a mean-variance investor who bases his decisions on it obtains sizeable utility gains, relative to another investor who uses, exclusively, the historical returns.
Socioeconomic Status and Macroeconomic Expectations
