With the growing availability of financial data, new variables are constantly proposed to predict stock returns, although their incremental explanatory power is often limited because many capture overlapping information. While it suggests itself to extract latent factors summarizing the underlying information — e.g. consider common trends in bond yields across maturities — from these variables and to subsequently utilize these factors as predictors, the usual problems with the variables’ unknown persistence and predictive regression endogeneity resulting in spurious predictability findings still apply. To address these issues, we combine factor extraction with the IVX framework of Kostakis et al. (2015), whose instrumental variable approach is able to resolve the endogeneity issue regardless of the particular degree of persistence. Monte Carlo simulations confirm that the proposed factor-based IVX regression approach achieves good size control and, in addition, strong power should predictability be present. The empirical relevance of the approach is illustrated using S&P 500 returns and a set of commonly used predictors.
Fabian Schmidt (Thu,) studied this question.