Abstract
Existing research on return predictability traditionally employs aggregate, market-level information. To investigate the applicability of return predictability at a finer level, we examine out-of-sample time-series return predictability at the characteristic-based portfolio level, using predictive regressions with portfolio-level predictors and a sum-of-the-parts approach. In addition to rejecting the null of no predictability at the market level, we detect statistically and economically significant out-of-sample predictability amongst particular portfolios. Notably, we show that large growth portfolios exhibit return predictability, consistent with predictions drawn from prior literature, while we fail to consistently detect predictability for all remaining size and book-to-market portfolios. Our results reveal a significant (relative) forecast error R-squared of 0.65 % for large-growth stocks, translating into an annualised certainty equivalent gain of 1.37 %.