The classical value-investing factor literature reports moderate alpha magnitudes — Asness, Frazzini Piotroski (2000) reports approximately seven for the F-Score; Greenblatt (2006)approximately six. We present empirical evidence that a particular methodological combination — sub-classification of GICS sectors into thirteen fine-grained groups, Damodaran-style lifecycle stageclassification, and sector-specific score sweet-spots — does not uniformly improve factor alpha. Instead,the methodology reveals heterogeneous and asymmetric pockets of alpha that the bucketed approachcommon in factor research masks. Our strongest signal — a long position in mature-stable healthcareequities scoring 45-58 on a Beneish-Altman-Piotroski-Greenblatt composite — outperforms its regionalbenchmark by +6.10 percentage points annualized (median three-year alpha +24.0pp, mean +32.4pp,n=109, six of six cohorts positive, bootstrap 95% CI +19.6pp, +46.5pp, p<0.001), comparable to ormodestly exceeding Piotroski's lifetime F-Score alpha. A short position on mature-stable firms withcomposite score above 60 ("quality trap") underperforms benchmark by −2.95 percentage pointsannualized (mean −11.0pp three-year alpha, p=0.003). Notably, both signals remain robust during the2017-2019 cohort sub-period, which is widely characterized as the worst regime for traditional valueinvesting in modern history. Two signals previously believed to be robust (energy/materials sectorexclusion, value trap) lose statistical significance once a region-appropriate benchmark replaces SPYuniversally. Sample size n=4,507 ticker-years over six cohorts. The principal contribution ismethodological rather than magnitudinal: we document that conditional decomposition by sector andlifecycle reveals signals obscured in pooled factor analyses, with the strongest effect in healthcare andthe strongest negative effect in the quality-trap region of the score distribution. Limitations explicitlydiscussed include in-sample threshold calibration, an imperfectly mitigated survivorship bias, and theabsence of walk-forward validation.
Alejandro I. Igual Gutiérrez (Wed,) studied this question.