Interactions in Asset Pricing

We propose a linearization of rule-based algorithms that reveals the most important interactions between characteristics and macroeconomic variables when explaining future stock returns. Our results suggest that the two types of predictors are intertwined, which implies that the relationships between returns and standard asset pricing characteristics are not only non-linear but also strongly dependent on the state of the economy. From an investment standpoint, our interpretable machine learning approach produces portfolios that outperform out-of-sample