Quality / Gross Profitability — is it still true?

The short version

Buying high-quality, profitable companies has beaten the market over the very long run — but only barely, and not reliably: in the last decade the edge was flat, and a real-world "quality" fund actually lagged the market over 13 years. So quality is worth using to avoid junk and to lean toward cheap-and-good companies — but it's not a money-maker on its own, and never a reason to overpay.

Abstract

We tested whether the profitability/quality premium still predicts returns — especially recently, and in a form we can actually use (long-only). Using Ken French's own profitability factor (RMW), 1963–2025, pulled ourselves, plus the live QUAL ETF: the premium is positive on average (full-sample +3.1%/yr, t=3.2) but is not statistically significant in any single decade, was near-flat in the 2010s (+1.4%/yr, t=0.8), and the live long-only QUAL ETF underperformed the S&P 500 by ~0.7pp/yr over 13 years. Conclusion: we keep quality, but only as a junk-exclusion screen and a "cheap + quality" up-weight / thesis-durability check — not as a standalone return signal, and never as a reason to overpay.

1. The claim (and why it matters to us)

High gross-profits-to-assets (Novy-Marx) and the broader "quality" bundle (profitable, stable, low-leverage, high-payout) are said to predict higher risk-adjusted returns. If true, it's an attractive signal for us: low-turnover (annual accounting data), long-side-viable, cheap to apply, and a natural drawdown cushion for a retail book.

2. What the literature claims — and our scepticism

• Novy-Marx (2013, JFE): gross-profits-to-assets long-short ≈ 0.31%/mo, ~same power as value; quality+value combined has a much higher Sharpe.
• Asness/Frazzini/Pedersen "Quality Minus Junk" (2019): information ratio >1, positive in 23/24 countries. (AQR sells quality strategies — an incentive worth naming.)
• Fama-French (2015): profitability factor RMW ≈ 0.25–0.34%/mo; Hou-Xue-Zhang (2020): GPA survives the replication crisis at 0.38%/mo, t=2.62 (one of the few that does).
• Our scepticism: all of these are long-short, mostly US, mostly samples ending 2010–2016, and McLean-Pontiff (2016) put average post-publication decay at ~50%. None of them answer "does the long-only leg still beat the market now?"

3. Our own analysis ← the heart of the report

Data we pulled: Ken French Data Library, F-F_Research_Data_5_Factors_2x3 (monthly RMW), and F-F_Momentum_Factor, 1963→May 2025; plus yfinance total returns (dividends reinvested) for QUAL and SPY. Method: mean monthly return, annualised (×12), t = mean ÷ (std/√n), bucketed by period. Fully reproducible from the Dartmouth zip.

RMW — Robust-minus-Weak profitability (long-short), our pull:

Live long-only check: QUAL (iShares MSCI USA Quality) vs SPY, 2013-07 → 2026-05 (12.9y): QUAL +13.63%/yr vs SPY +14.29%/yr over the same window → −0.66pp/yr (quality underperformed). Context (same pull): Value (HML) was negative in the 2010s (−2.4%/yr) and post-2010 overall; momentum post-2010 +3.6%/yr (t=1.2, also sub-significant), though the MTUM ETF beat SPY by +1.45pp/yr.

4. Findings

• The profitability premium is real over the full 60 years (t=3.2) but the significance comes from pooling — no individual decade clears t=2.
• It was weakest in the most recent full decade (2010s: +1.4%/yr, t=0.8 — effectively flat).
• The live, long-only, after-fee version (QUAL) did not beat the market over 13 years.
• Quality held up better than value (which was negative in the 2010s) but neither was a significant standalone winner recently.

5. Conclusion — keep, but as an overlay (and why)

We keep quality, demoted to a confirming overlay, because the evidence does not support it as standalone alpha at our scale, recently, long-only. Its real, defensible uses: 1. Junk-exclusion — drop low/negative gross-profits-to-assets, high leverage, unstable earnings. Avoiding junk is the most reliable long-side use of the factor. 2. "Cheap + quality" up-weight — the single best-evidenced combination (Novy-Marx); quality and value are genuinely independent families (see ../families_and_independence.md), so this is real confluence, not double-counting. 3. Durability check / drawdown cushion — this is the bridge to the house view that a trend (or long-run average) is only trustworthy if backed by quality. Quality's job here is to confirm a thesis will hold, not to time an entry.

6. Limitations & open questions

• Factor data (a diversified long-short portfolio) is not single-name selection skill.
• QUAL carries a fee (~0.15%) and MSCI's specific quality construction — not a pure GPA sort.
• We have not yet run a single-name gross-profits-to-assets sort on our HL-tradeable universe, or tested the quality+cheap combination directly on names we can hold.
• RMW is US; UK/HL applicability untested by us.

7. How we would extend this

Build a single-name gross-profits-to-assets sort on our HL universe via the verification harness; test quality-AND-cheap directly (does the combination beat either alone, recently, long-only, net of cost?). That converts "we trust the overlay" into "we measured the overlay on our own names."

Sources

• Pulled ourselves: Ken French Data Library — F-F 5-Factor (RMW), HML, Momentum (monthly, 1963→2025); yfinance total returns QUAL, SPY, MTUM.
• Cited: Novy-Marx (2013, JFE 108(1)); Asness/Frazzini/Pedersen "Quality Minus Junk" (2019, RAS 24); Fama-French (2015, JFE 116(1)); Hou-Xue-Zhang "Replicating Anomalies" (2020, RFS 33); McLean-Pontiff (2016, J. Finance).

Changelog

• v1.0 (2026-05-31) — initial study; Ken French RMW by decade + QUAL/SPY live check.