Which Stock Screens Actually Work? A Survivorship-Free Audit of 16 Fundamental Metrics — ROIC, Value, Momentum, and Quality — on the S&P 500 (2010–2024)

JEL classification: G11, G12, G14 Keywords: ROIC · return on invested capital · stock screens · fundamental investing · Magic Formula · Piotroski F-score · gross profitability · sustainable growth · reinvestment rate · asset-growth anomaly · earnings yield · book-to-market · value investing · quality factor · survivorship bias · point-in-time · S&P 500 · which metrics predict stock returns

Cite as: Brina, F. (2026). Which Stock Screens Actually Work? A Survivorship-Free Audit of 16 Fundamental Metrics — ROIC, Value, Momentum, and Quality — on the S&P 500 (2010–2024) (Working Paper v1.0). Zyberno Research. Zenodo: 10.5281/zenodo.20650537.

About this audit. Sixteen fundamental screens, one common survivorship-free sample (\(n=2{,}103\); §5.1), benchmark-invariant evaluation, risk-adjusted alphas (FF5 + momentum; §5.11), and Fama-MacBeth / Newey-West inference for the overlapping windows (Tables 1–2). Every result is reported, positive or null — an audit is only useful if its nulls can be trusted, and ours recover exactly the stylized facts of the era (§5.1). This paper deliberately excludes the Brina Gap and the Margin of Safety: both are valuation principles — measurements of what a price assumes — not return-prediction screens; they are treated in the companion paper (Brina 2026, The Brina Gap: A Framework for Identifying Growth Mispricing in Equity Markets, v3.0), which uses this audit as its benchmark panel.

Key findings

• On the survivorship-free S&P 500, 2010–2024, return on invested capital (ROIC) was the single most predictive fundamental metric of five-year forward returns (rank correlation +0.11; top-minus-bottom quintile spread +3.0 pp/yr), and its edge held in both halves of the period — robust to Fama-MacBeth / Newey-West inference (t = +5.9).
• The widely-used sustainable-growth screen — \(g = \text{ROIC} \times \text{reinvestment rate}\) — had essentially no predictive power (rank correlation +0.012). It fails because it multiplies a positive factor (profitability) by a negative one (reinvestment/asset growth), and the two cancel. We call this the sustainable-growth paradox.
• But after risk adjustment, no screen — not even ROIC — earns significant alpha. Each screen’s monthly long-short portfolio, regressed on the Fama-French 5 factors + momentum, has an insignificant intercept (every |t| < 1). The screens are repackaged factor exposures: value screens are HML, the momentum screen loads +1.1 on momentum, and ROIC’s raw edge is a growth/profitability tilt that paid in a quality-led regime. On the large-cap S&P 500, these screens buy factor tilts, not a free lunch.
• The famous academic quality/anomaly screens fade on large caps — ROIC is the one that survives. ROE, the most popular quality metric, is null and ROIC beats it ~17-to-1 on the same firms (+0.10 vs +0.006) — because ROE is contaminated by leverage (which hurt returns here) while ROIC is capital-structure-neutral. Gross profitability (Novy-Marx) is near-null (+0.02), and accruals, the Piotroski F-score, and asset growth all fade to ≈ 0 — consistent with these anomalies concentrating in smaller stocks.
• Correcting the sign — free-cash-flow return on capital, \(\text{FCFROIC} = \text{ROIC}\times(1-\text{reinvestment})\) — restores predictive power (+0.084) but does not beat plain ROIC.
• Composites diluted rather than improved. Blending quality with value and momentum (equal weight) underperformed ROIC alone in this period. Parsimony won.
• “Hit-rate” backtests are misleading. ~72% of S&P 500 stocks underperformed the cap-weighted index over the period, so a one-sided screen scores ~72% with no skill at all. Honest evaluation needs benchmark-invariant metrics (rank correlation, quintile spreads) — not accuracy.
• The value family was uniformly weak (book/market was the worst screen tested, rank correlation −0.12) — but this reflects the quality-led, value-hostile 2010–2024 regime, not a general verdict on value.
• Reinvestment rate was a negative predictor (−0.04) — but raw capital growth was roughly neutral (+0.01), so the negative investment signal works through reinvestment relative to earnings, not the asset-growth anomaly (which is known to concentrate in small caps, not large-cap S&P 500 names).
• Among capital-structure screens, net buyback yield was positive (+0.06 — returning capital was rewarded) and net leverage was negative (−0.08 — low-debt firms won), both consistent with a flight-to-quality regime and with the reinvestment finding.

Scope: one market (S&P 500), one regime (2010–2024), five-year horizon, before transaction costs. The factor directions confirm established findings (Novy-Marx; Cooper-Gulen-Schill; Fama-French), and the sustainable-growth result extends Lockwood & Prombutr (2010); the contribution is the comprehensive, survivorship-free audit and its honest evaluation.

Abstract

Which stock screens actually work? We audit sixteen fundamental screens on the complete, survivorship-free S&P 500 (2010–2024) and find that return on invested capital (ROIC) is the most predictive of them all. ROIC’s rank correlation with five-year forward excess returns is +0.11, its top-minus-bottom quintile spread is +3.0 percentage points per year, and its edge holds in both halves of the period (Fama-MacBeth t = +5.9). The textbook sustainable-growth screen — ROIC × reinvestment rate — does not work: it multiplies a positive factor (profitability) by a negative one (reinvestment), and the two cancel; we call this the sustainable-growth paradox. Value was the weakest family of the era: book-to-market was the single worst screen tested (−0.13) — a faithful portrait of the quality-led 2010–2024 regime, not a verdict on value investing. The celebrated academic quality screens fade on large caps: ROE, gross profitability (Novy-Marx), accruals (Sloan), the Piotroski F-score, and asset growth all sit near zero, and ROIC beats ROE — the most popular quality metric — seventeen-to-one on identical firms. After risk adjustment, no screen earns significant alpha: every long-short portfolio’s intercept against the Fama-French five factors plus momentum has |t| < 1, so on the large-cap S&P 500 these screens deliver factor tilts, not a free lunch.

The audit is built to be trusted. The universe is the complete point-in-time S&P 500 — every constituent-year, including firms later acquired or delisted — with dividend-reinvested total returns, filing-date entry, and CRSP-style delisting handling. We also show why common backtests flatter themselves: roughly 72% of S&P 500 stocks underperformed the cap-weighted index, so a one-sided screen scores ~72% “accuracy” with no skill at all; honest evaluation requires benchmark-invariant statistics (rank correlations, quintile spreads), which we use throughout. The factor directions confirm established results (Novy-Marx 2013; Cooper-Gulen-Schill 2008; Fama-French 2015), and the sustainable-growth finding extends Lockwood & Prombutr (2010); the contribution is the comprehensive, survivorship-free audit of the screens investors actually use — transparent nulls included — with the full dataset and code published so every result can be reproduced, extended to other regimes, or refuted.

1. Introduction — do the screens investors use actually work?

Investors rank stocks with a familiar toolkit: profitability ratios (ROIC, ROE, gross margin), valuation multiples (earnings yield, book-to-market, EV/EBIT), growth-and-reinvestment constructs (the sustainable-growth rate), quality scores (the Piotroski F-score), and composites that blend several (the Magic Formula). Each is promoted somewhere as a way to beat the market. But most evidence on them comes either from long academic samples spanning the whole CRSP universe, or from practitioner backtests that quietly contain survivorship bias (only firms that survived to today) and look-ahead (using data before it was public).

This paper asks the practical question directly: on the universe an investor actually fishes in — the S&P 500, reconstructed point-in-time so that acquired and delisted firms are present — which of these screens carried real information about subsequent returns, and which did not?

1.1 Contribution

1. A clean, survivorship-free audit of the standard screens on the S&P 500, with honest, benchmark-invariant evaluation. The ranking is useful in itself; the demonstration that raw hit-rate accuracy is inflated by the cap-weighted benchmark is a cautionary contribution for anyone who backtests screens.
2. The sustainable-growth paradox: the textbook construct \(g=\text{ROIC}\times\text{retention}\), embedded in countless DCFs and screens, is self-defeating as a ranking signal because it is the product of a positive factor (profitability) and a negative one (reinvestment/asset growth). Lockwood & Prombutr (2010) first linked the sustainable-growth rate to returns and found its profitability component dominant; we give the modern factor-language account — on the large-cap S&P 500 the ROIC-based product nets to ≈ 0 — and show a cash-flow correction (FCFROIC).
3. A parsimony result: across single screens, corrected constructs, and composites, nothing in the toolkit reliably beats plain ROIC, whose edge is stable across both sub-periods.

The factor directions — profitability rewarded, aggressive investment penalized — replicate Novy-Marx (2013) and Cooper-Gulen-Schill (2008) / Fama-French (2015), and the oppositely-signed structure is the Hou-Xue-Zhang (2015) q-factor model. The sustainable-growth result extends Lockwood & Prombutr (2010) — who first linked the sustainable-growth rate to returns and found its profitability component dominant — into modern factor language, rather than discovering a new effect. Our genuine value-added is the audit itself — in the tradition of large-scale anomaly audits and multiple-testing scrutiny (Harvey, Liu & Zhu 2016; McLean & Pontiff 2016; Hou, Xue & Zhang 2020): the comprehensive synthesis on a clean, survivorship-free, point-in-time universe, with honest benchmark-invariant evaluation, transparent nulls, and head-to-head comparisons.

2. Data: the survivorship-free S&P 500

Point-in-time membership, 2010–2024. We reconstruct index membership for each year from the published constituent list and its change history — 771 distinct firms across all constituent-years (year-end counts 503–507, matching the index). Firms later acquired, merged, or delisted are present in the years they were genuine members, removing survivorship bias.

Fundamentals come from SEC filings (XBRL Company Facts), as of each fiscal year-end. Prices are dividend-reinvested total returns; the benchmark is SPY (total return). Entry is fiscal year-end + 90 days, a conservative proxy for the 10-K filing date that removes look-ahead. Delisting is handled CRSP-style: firms acquired mid-window are carried to the deal price and proceeds reinvested in the index; bankruptcies go to terminal value.

Outcome: five-year forward total return in excess of SPY, annualized, winsorized at ±50 pp/yr. Subset: because invested-capital metrics are not meaningful for banks, insurers, and REITs, the audit runs on the non-financial subset (~2,000 firm-years; exact \(n\) reported per analysis).

3. Which screens did we test?

Completed (with results below):

Future work — two screens need data not in the current pipeline, and both are expected to mirror results already in Table 1: full shareholder yield (needs dividends; the buyback arm is reported) and net debt / EBITDA (needs D&A; net-debt / market-cap is reported). (EV/EBIT ≈ earnings yield, already tested.)

Deliberately excluded: the Brina Gap and the Margin of Safety. Both are valuation principles — they measure what a market price assumes, not a return-ranking signal — so judging them as screens would be a category error. They are treated, with full validity testing, in the companion paper (Brina, 2026).

4. Are stock-screen “hit rates” misleading? (Yes — here is why)

The most common way to judge a screen is its directional hit rate: how often a high score precedes a positive excess return. We avoid it, because it is inflated by the cap-weighted benchmark. Over 2010–2024, roughly 72% of S&P 500 stocks underperformed the cap-weighted index — the index was carried by a handful of mega-caps. A rule that mostly predicts “underperform” therefore scores ~72% with no skill, and any predominantly one-signed screen inherits a flattering accuracy that reflects the base rate, not information.

We therefore use three benchmark-invariant statistics, which compare firms to each other rather than to a no-skill base rate:

1. Pooled rank correlation (Spearman) of the screen with five-year forward excess return.
2. Within-sector-year rank correlation — the same, inside (sector × year) groups of ≥4 comparable firms, averaged with a t-statistic across groups (the fairest “tiebreaker among comparables”).
3. Top-minus-bottom quintile spread (Q5 − Q1) in mean annualized excess return.

A joint cross-sectional regression (size controlled) reports each screen’s standardized coefficient. (§5.11 reports the FF5+momentum factor-model alphas; Tables 1–2 report Fama-MacBeth (1973) / Newey-West (1987, 4-lag) standard errors across the 12 entry-year cohorts, the proper correction for the overlapping five-year windows.)

5. The audit results

5.1 The horse race — which screens carried information?

All sixteen screens are evaluated on one common sample of \(n = 2{,}103\) non-financial firm-years (those with the core fields, a 5-year forward total-return excess, and 12-month momentum all available). Five screens that need extra line items are reported on labelled subsets of it — net buyback yield, gross profitability, ROE, accruals, and the Piotroski F-score (coverage in the table notes; quality screens discussed in §5.10). Table 1 ranks them all by pooled rank correlation.

Table 1 — Audit results (one common sample; ranked by pooled rank-corr with 5-yr forward excess return).

† \(n=2{,}068\) (needs prior-year shares). ‡ \(n=1{,}644\) (needs a reported gross margin; see §5.10). § labelled subsets requiring extra line items: ROE \(n=1{,}843\), accruals \(n=1{,}906\), Piotroski \(n=946\). FM t = Newey-West (4-lag) t-statistic of the mean annual rank correlation across the 12 entry-year cohorts — proper inference for the overlapping five-year windows. ROIC and FCFROIC are robustly positive; sustainable growth is null; reinvestment, leverage, and book/market robustly negative.

The test recovers exactly the stylized facts of the era — quality up, value down, momentum modest — which is itself a validation: a procedure that correctly detects what did work lends credibility to its nulls.

5.2 Does ROIC predict stock returns? (Yes — and durably)

ROIC is the most informative single screen on every column: the best rank correlation (+0.112), the widest quintile spread (+3.0 pp/yr), and a positive within-sector tilt. In the joint regression it is the strongest fundamental factor (Fama-MacBeth +0.66 pp/yr per sd, t = +2.40; Table 2). And it is durable — across the two halves of the sample its rank correlation is +0.142 (entries ≤2014) and +0.100 (entries ≥2015), clearly positive in both — and that consistency, under proper Fama-MacBeth / Newey-West inference for the overlapping windows, yields a t of +5.94 (Table 1). That stability is the strongest evidence that ROIC’s edge is real rather than a fluke of one window.

5.3 Why does the sustainable-growth screen fail? (The sustainable-growth paradox)

The construct \(g = \text{ROIC}\times\text{retention}\) — the sustainable-growth rate (Higgins 1977) — is everywhere: the engine of two-stage DCFs and a popular quality-growth screen. Yet it sits near zero (ρ = +0.012), far below ROIC. The reason is that reinvestment is a negative factor. In the multivariate Fama-MacBeth regression (Table 2, §5.8), ROIC enters strongly positive (+0.66, NW t = +2.40) while reinvestment enters significantly negative (−0.91, NW t = −3.09); the two are nearly independent (rank correlation −0.07). Their product blends a winner with a loser, and the signal cancels. A practitioner ranking stocks on the sustainable-growth rate is mixing two oppositely-signed factors into noise. This is the sustainable-growth paradox. Lockwood & Prombutr (2010) first documented that the sustainable-growth rate relates to returns — negatively in their 1964–2007 full-universe sample, which they attribute to risk — and that its profitability component dominates; our account here is the modern factor-language version, in which the ROIC-based product nets to ≈ 0 on the large-cap S&P 500 because its profitability arm (+) and investment arm (−) are oppositely signed.

5.4 Can the sustainable-growth screen be fixed? (Partly — meet FCFROIC)

If the problem is the sign on reinvestment, flip it. Replacing RR with (1 − RR) gives free-cash-flow return on capital:

\[\text{FCFROIC} = \text{ROIC}\times(1-\text{RR}) = \frac{\text{Free Cash Flow}}{\text{Invested Capital}}\]

— now the product of two positive factors (high profitability × high cash conversion), and the more honest “quality” number: cash generated per dollar of capital after reinvestment. It works (\(n=2{,}103\)): ρ = +0.084 (t = +2.82, Q5−Q1 +2.9 pp), far above the dead construct. But it does not beat plain ROIC and adds nothing incremental to it (excess ~ ROIC + FCFROIC + size: ROIC t = +3.82, FCFROIC t = +0.65). The repair recovers ROIC’s signal; it does not improve on it.

5.5 Do composites help? (No — they diluted)

An equal-weight blend of FCFROIC, earnings yield, and momentum scored ρ = +0.040 — worse than ROIC alone — because in this regime the value and momentum arms dragged the quality arm down. Adding factors diluted the signal. Parsimony won.

5.6 Does value still work? (Not in this regime)

Book/Market — the classic value factor (Fama & French 1993) — is the worst screen in the table (−0.126, Q5−Q1 −4.5 pp); Earnings Yield and EPV (Greenwald et al. 2001) hover around zero. Across both sub-periods the value family stayed weak (Book/Market −0.158 then −0.123). This is a faithful portrait of the quality-led, value-hostile 2010–2024 regime — not evidence that value is generally uninformative. A genuine value regime (e.g., 2000–2007) is outside this sample; see §7.

5.7 Is the Magic Formula still effective? Does momentum work in the S&P 500?

The Magic Formula (ROIC + earnings yield; Greenblatt 2005) lands mid-pack (ρ = +0.057, Q5−Q1 +2.1 pp) — its edge is essentially the ROIC component plus a cross-sector tilt; within sector it is flat (−0.002). Momentum (Jegadeesh & Titman 1993) is modest pooled (+0.043) but is the strongest within-sector screen (t = +2.5), consistent with industry-relative momentum.

5.8 Which screens survive when tested together?

Table 2 — Multivariate Fama-MacBeth regression (excess ~ standardized factors; mean coefficient in pp/yr per 1 sd, Newey-West 4-lag t).

Profitability survives, investment is a drag, size dominates. The two value measures are collinear and split the value load across specifications — here earnings yield carries it (−1.46) while book/market’s small positive residual is a collinearity artifact, not a value premium; univariate, book/market is the most negative screen in the audit (Table 1). Plain pooled OLS agrees on ROIC (+2.30), reinvestment (−2.24), and size (−5.35).

5.9 Does buying back stock help? Does leverage hurt? Does capital growth predict returns?

Three capital-allocation screens in Table 1 round out the picture. Buying back stock helped — net buyback yield is positive (ρ = +0.057), among the stronger screens — reinforcing the central finding that the market rewarded returning capital and penalized reinvesting it. High leverage hurt — net debt / market cap is sharply negative (ρ = −0.086; −3.1 pp top-vs-bottom), consistent with a flight-to-quality era in which cash-rich firms led and levered firms lagged. And raw capital growth was roughly neutral (ρ = +0.015) — the classic asset-growth anomaly (Cooper-Gulen-Schill) does not replicate cleanly on large-cap S&P 500 names; the negative investment signal operates specifically through the reinvestment rate (relative to earnings), not the growth of the asset base — consistent with such anomalies concentrating in smaller, less liquid stocks.

5.10 ROIC vs the academic quality screens — which survive on large caps?

ROIC’s rivals are the literature’s other quality and accounting screens. Tested head-to-head, none survives on the large-cap S&P 500 while ROIC does.

ROIC vs ROE. Return on equity is the single most widely used quality metric — yet on identical firms (\(n=1{,}843\)) it is null (ρ = +0.006) while ROIC is +0.102: ROIC beats ROE roughly seventeen-to-one. The reason is mechanical and consistent with §5.9 — ROE is contaminated by leverage (ROE ≈ ROIC scaled by a leverage factor), and high-leverage firms underperformed this period; ROIC is capital-structure-neutral and stays clean.

Table 3 — ROIC vs its quality rivals (identical-subset head-to-heads).

The accounting-quality screens fade too. Accruals (Sloan 1996) carry the expected negative sign but only marginally (ρ = −0.012, FM t = −1.70); the Piotroski (2000) F-score is null (ρ = +0.006). (Gross profitability and the F-score need a reported gross margin / a full balance sheet, so they run on 78% and 45% of firms respectively — pure-services names are necessarily thinned.)

Put together, this is the paper’s most repeated pattern: five celebrated screens — gross profitability, ROE, accruals, the Piotroski F-score, and asset growth — all fade to ≈ null on the large-cap S&P 500, consistent with the evidence that such anomalies concentrate in smaller, less liquid stocks (Hou-Xue-Zhang 2020). ROIC — a capital-efficiency measure that nets out leverage and the cost structure below the gross line — is the one quality signal that survives.

5.11 Do any screens earn alpha, or just factor exposure?

The correlations and spreads above are raw. The decisive test is whether a screen earns returns beyond the known factors. For each screen we form a monthly-rebalanced long-short portfolio (top-minus-bottom quintile, equal-weight, screens known at formation — no look-ahead) and regress its 171-month return series on the Fama-French five factors plus momentum.

Table 4 — Risk-adjusted alphas (monthly long-short regressed on FF5 + momentum).

The result is unambiguous and, for an honest audit, reassuring: no screen earns statistically significant alpha — every t(α) is below 1 in magnitude. Each screen is a repackaging of known factor exposures. The value screens load heavily on HML (+0.58) and carry negative alpha in this value-hostile period; ROIC and FCFROIC are growth-and-profitability tilts; and — a clean validation that the machinery works — the momentum screen loads +1.10 on the momentum factor with no residual. The raw cross-sectional edge of ROIC documented above is therefore factor exposure — a tilt that paid in a quality-led regime — not a free lunch. An investor applying these screens to the large-cap S&P 500 is buying factor tilts, not alpha.

Caveat: a large-cap-only (~450-stock) universe gives equal-weight quintile long-shorts limited statistical power; the point estimates are nonetheless economically small, and the loadings cleanly identify each screen’s factor content.

6. What does this mean for investors?

Parsimony beats complexity. In this universe and era, the simplest fundamental screen — ROIC — is the one to use. Multiplying it by reinvestment kills it; correcting that to a cash-flow basis ties it; blending it with value and momentum dilutes it. Concretely: (i) do not rank on the sustainable-growth screen \(\text{ROIC}\times\text{retention}\) — it cancels two real factors; (ii) if you want one quality number, ROIC (or FCFROIC) is hard to beat; (iii) be wary of equal-weight composites that dilute a strong factor with weak ones; (iv) distrust hit-rate backtests against the cap-weighted index — judge screens on cross-sectional spreads.

7. Limitations and scope

• One market, one regime. Both sub-periods sit inside a single quality-led, value-hostile era; we never observe a genuine value regime. The value family’s weakness is therefore period-bound, and its behaviour elsewhere is an open question requiring out-of-era or international data.
• No transaction costs / capacity. Screens are evaluated as cross-sectional signals, not implementable strategies (the five-year horizon implies low turnover, but costs are unmodeled).
• Risk adjustment — done. §5.11 reports each screen’s FF5+momentum alpha; none is significant, so the raw spreads reflect factor exposure, not alpha.
• Sample — harmonized. All sixteen screens are evaluated on one common sample (\(n = 2{,}103\); §5.1), with five (net buyback yield, gross profitability, ROE, accruals, Piotroski) on labelled subsets.
• Inference — done. Headline significances are reported with Fama-MacBeth / Newey-West (4-lag) standard errors across the 12 entry-year cohorts (Tables 1–2), the proper correction for the overlapping five-year windows; the findings survive (and several strengthen).

None of these is expected to overturn the central findings, which are stable across every cut tried.

8. Conclusion — which screens actually work?

On the complete point-in-time S&P 500 over 2010–2024, return on invested capital is the most informative fundamental screen, and durably so, while reinvestment is a drag — so the textbook sustainable-growth screen, which multiplies the two, cancels itself out (the sustainable-growth paradox). Correcting the sign recovers ROIC’s signal but does not exceed it, and composites dilute it. The lesson is parsimony. And once risk-adjusted against the Fama-French factors plus momentum, none of these screens earns alpha — what looks like a quality edge is a factor tilt — so the honest takeaway for a large-cap investor is about which factor exposures a screen delivers, not which screen “beats the market.” We report these results — including the nulls — on a survivorship-clean universe with open code, so each can be checked, extended to other regimes and markets, or refuted.

Data availability and reproducibility

The point-in-time membership reconstruction, the firm-year dataset, the corporate-event records, and the full computation pipeline are published as companion files so every number can be independently reproduced. A plain-language companion (Q&A form) is posted at Zyberno.

References

• Brina, F. (2026). The Brina Gap: A Framework for Identifying Growth Mispricing in Equity Markets. Working Paper v3.0, Zyberno Research. Zenodo: 10.5281/zenodo.20510518; SSRN 6361659. (Companion paper: the valuation-principles side of this research program.)
• Cooper, M., Gulen, H., & Schill, M. (2008). Asset Growth and the Cross-Section of Stock Returns. Journal of Finance, 63(4), 1609–1651.
• Fama, E., & French, K. (1993). Common Risk Factors in the Returns on Stocks and Bonds. Journal of Financial Economics, 33(1), 3–56.
• Fama, E., & French, K. (2015). A Five-Factor Asset Pricing Model. Journal of Financial Economics, 116(1), 1–22.
• Fama, E., & MacBeth, J. (1973). Risk, Return, and Equilibrium: Empirical Tests. Journal of Political Economy, 81(3), 607–636.
• Greenblatt, J. (2005). The Little Book That Beats the Market. Wiley.
• Greenwald, B., Kahn, J., Sonkin, P., & van Biema, M. (2001). Value Investing: From Graham to Buffett and Beyond. Wiley.
• Harvey, C., Liu, Y., & Zhu, H. (2016). … and the Cross-Section of Expected Returns. Review of Financial Studies, 29(1), 5–68.
• Higgins, R. (1977). How Much Growth Can a Firm Afford? Financial Management, 6(3), 7–16.
• Hou, K., Xue, C., & Zhang, L. (2015). Digesting Anomalies: An Investment Approach (the q-factor model). Review of Financial Studies, 28(3), 650–705.
• Hou, K., Xue, C., & Zhang, L. (2020). Replicating Anomalies. Review of Financial Studies, 33(5), 2019–2133.
• Jegadeesh, N., & Titman, S. (1993). Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency. Journal of Finance, 48(1), 65–91.
• Lockwood, L., & Prombutr, W. (2010). Sustainable Growth and Stock Returns. Journal of Financial Research, 33(4), 519–538.
• McLean, R. D., & Pontiff, J. (2016). Does Academic Research Destroy Stock Return Predictability? Journal of Finance, 71(1), 5–32.
• Newey, W., & West, K. (1987). A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix. Econometrica, 55(3), 703–708.
• Novy-Marx, R. (2013). The Other Side of Value: The Gross Profitability Premium. Journal of Financial Economics, 108(1), 1–28.
• Piotroski, J. (2000). Value Investing: The Use of Historical Financial Statement Information to Separate Winners from Losers. Journal of Accounting Research, 38(Suppl.), 1–41.
• Sloan, R. (1996). Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings? The Accounting Review, 71(3), 289–315.