Value investing via Bayesian inference

Abstract

Classic value investing à la Graham & Dodd (Security analysis: The classic, McGrawHill, New York, 1934) focuses on selecting stocks that seem cheap relative to their intrinsic value and fundamental quality. We use Bayesian inference to account for a large amount of uncertainty within intrinsic value estimation. We find that an undervalued-minus-overvalued factor that invests in cheap quality stocks and sells expensive junk stocks selected via Bayesian inference yields high risk-adjusted returns and Sharpe ratios for equal-weighted portfolios. We also find that using value-weighted portfolios introduces size-based dilutions and shifts the focus away from actual quality characteristics like profitability, payout, safety, and past growth. Our findings suggest that while the relative benefit of accounting for uncertainty via Bayesian inference is not large over shorter holding periods, it pays off for investment horizons longer than a month.