Supervised Learning In Asset Pricing

Abstract

This chapter looks at many prediction problems in asset pricing that are of high-dimensional nature in a large number of observable variables that could have useful predictive information. For example, in stock return prediction a huge number of variables could potentially be relevant as predictors. It covers firm characteristics from accounting data, signals extracted from textual information in corporate disclosures, variables summarizing the history of price and trading volume, information in media reports, and many other variables that could potentially contain predictive information. The chapter describes the existing literature in asset pricing that has dealt with this high-dimensionality by imposing ad hoc sparsity. Rather than considering large numbers of predictors simultaneously, researchers often consider small sets of predictors in isolation.