An Empirical Assessment of Characteristics and Optimal Portfolios

Abstract

We show that overfitting plagues optimal portfolios obtained by linking weights directly to characteristics. We effectively mitigate overfitting (regularize) by optimizing a more concave loss function than the investor’s utility function. Significant certainty equivalent gains over benchmarks
require at least three characteristics: momentum, size, and residual volatility. Out-of-sample
utility gains are due to characteristic complementarities and depend on investor risk aversion.
For example, conditioning on momentum relieves the overfitting bias inherent in the other characteristics for our most risk-averse investor. Optimal portfolios’ returns lie largely outside the span of traditional factors and move closer to the market as risk-aversion increases.

Keywords: cross-section of stock returns; stock characteristics; optimal portfolios

JEL Classification: G10, G11