Free-cash-flow yield — does buying cheap-on-cash beat the market?

The short version

"Buy companies that gush cash relative to their price" sounds like a sure thing — but the real result is more interesting than a yes/no. Sorting the S&P 500 on free-cash-flow yield over ten years, the relationship turns out to be U-shaped: the cheapest-on-cash names returned about +20%/yr and the most expensive returned about +19%/yr — both comfortably beat the market — while the muddled middle lagged. A simple "cheap minus expensive" bet nets to roughly nothing, not because the signal is dead, but because it shorts a tail that also wins. The part to avoid is the undifferentiated middle.

Abstract

We tested whether buying high free-cash-flow-yield stocks beats the market, on our own data: the current S&P 500, free SEC free-cash-flow / share-count / debt filings, and price history, across ten annual rebalances (2015–2024) with no look-ahead (FCF only from 10-Ks filed before each rebalance). FCF yield = trailing free cash flow ÷ market cap (and, as a refinement, ÷ enterprise value); we sort into quintiles and hold 12 months. The relationship is non-monotonic: both the high-yield quintile (+20.2%/yr) and the low-yield quintile (+19.5%/yr) beat the cap-weighted S&P (14.3%) and the equal-weight universe (17.2%), while the three middle quintiles averaged ~15.4% and trailed equal-weight in 8 of 10 years. The headline high-minus-low spread (+0.8%/yr) is near-zero only because it is a linear contrast across a U. Switching the denominator to enterprise value sharpens the cheap tail slightly (spread +1.7%/yr) but leaves the shape intact. A 50/50 barbell of the two tails returned +18.3%/yr at a 1.02 Sharpe (vs SPY's +13.6% and 0.86) — the best risk-adjusted basket we tested. The same test on S&P 600 small-caps shows the signal generalises down the size scale — the cheap tail beats the small-cap index by +7–12%/yr and the U-shape repeats — though it is higher-risk there and needs a broader basket. An early, short-window read on the UK (FTSE 350) points the same way — value beat the FTSE too — so the effect looks market-wide, not US-only (preliminary; free UK data is shallow). Verdict: a U-shaped signal — own the ends, avoid the middle; the barbell is the cleanest way to hold it. Grade B+.

1. The claim

A high free-cash-flow yield (lots of real cash relative to price) is a classic "cheap" signal — value investors' favourite, because cash is harder to fake than reported earnings. If it predicts returns, it's a low-turnover, long-only-friendly strand.

2. What the literature claims — and our scepticism

• Value (cheap beats dear) is one of the most-documented premia (Fama-French HML), and FCF-based metrics are among the better value proxies because cash flow is hard to manipulate.
• Our scepticism: value has been weak-to-negative in large caps post-2010 (HML was negative through the 2010s — see our quality study), and the premium historically lives in smaller names. Does the long-only, large-cap, recent version actually beat the market, and is the relationship even a straight line? We test it ourselves.

3. Our own analysis ← the heart of the report

Data: current S&P 500 (503 names; 499 with usable SEC fundamentals); free cash flow, shares and net debt from SEC EDGAR 10-K filings; monthly total-return prices. Method: at each 30 June 2015–2024, FCF yield = (most recent 10-K FCF, filed before the date) ÷ market cap, and separately ÷ enterprise value (market cap + net debt); sort into quintiles; hold 12 months; total return. Benchmarks: the equal-weight universe and the cap-weighted S&P (SPY).

Pooled return by quintile (mean over 10 rebalances). Q1 = lowest FCF yield (expensive on cash); Q5 = highest yield (cheap on cash):

The two extremes sit ~+5%/yr above the cap-weighted index and ~+3–5%/yr above the middle three quintiles. That is a U, not a line.

Year by year (FCF / market cap). The shape is the average over the decade — strong most years, with leadership rotating between the tails, and the occasional exception (2015, where the middle led):

4. Findings

• FCF yield is U-shaped, not weak. Both tails (~+20%/yr) beat the market and the equal-weight universe by ~5%/yr; the three middle quintiles averaged ~15.4% and beat equal-weight in only 2 of 10 years. The middle — large caps with neither a cheapness story nor a growth story — is the dead zone.
• The "+0.8%/yr spread" was an artefact. A high-minus-low spread is a single straight-line contrast; run across a U it subtracts one winning tail from the other and reports ≈ 0. That is exactly why a linear or rank test mislabels FCF yield as "dead" when it isn't — the same lesson as our score-vs-return work: use the full quintile profile, never a two-point contrast, on a non-monotonic signal.
• The two tails are different animals (read straight off the constituents):
• High-yield / cheap tail — classic value: banks (JPM, WFC, C, COF), energy (XOM, CVX, COP, EOG, VLO), big pharma (ABBV, BMY, PFE, GILD), cash-cow tech (AVGO, DELL, CSCO, LRCX).
• Low-yield / expensive tail — two cohorts with low FCF yield for opposite reasons: high-multiple growth (NVDA, PLTR, TSLA, AMZN, MSFT, CRWD, LLY) and heavy-capex reinvesters, where utilities dominate (DUK, SO, ETR, EXC, XEL, AEE) — low yield because capex eats the cash, not because they are dear. Both cohorts were rewarded over 2015–24.
• Leadership rotates by regime. The cheap tail rips in value years (Q5 +36.7% in 2016, +65.9% in 2020); the expensive tail wins in growth years (Q1 +15.0% vs Q5 −2.3% in 2019). The constant is that both beat the middle on average — 2015 being the visible exception, where Q3 (+14.1%) led.
• EV is the fairer denominator and modestly sharpens the cheap tail (Q5 +20.7% vs +20.2%; high-minus-low spread +1.7% vs +0.8%) without changing the shape — exactly the expected effect of penalising leverage and crediting net cash.

4a. Combining the tails — the barbell

The two tails lead in opposite regimes (value rips in 2016/2020; growth wins 2019), so we tested owning both: a 50/50 barbell of Q1 and Q5, rebuilt at monthly frequency (annual rebalance, equal-weight within quintile), 2015-07 to 2025-06.

The barbell is the best risk-adjusted basket — the highest Sharpe (1.02, above each tail, the equal-weight universe and the market), and it cuts the value tail's drawdown (−30% → −24%). The diversification is partial, not total: the two tails are +0.90 correlated in raw returns (shared market beta), but only +0.39 correlated in excess-over-market returns — so blending them lifts the information ratio to +0.75 (vs +0.62 / +0.63 for the tails alone) and cuts active risk (tracking error 8.4% → 6.1%), even though it barely changes total volatility. Owning both ends beats picking one tail and guessing the regime. (These figures are on current constituents — see Scope; the honest forward test is tracking it live.)

4b. Taking it further — concentration

A 196-name basket proves the signal but isn't something a person would hold. So we asked the practical question: how does the signal behave concentrated to a tradable number of names? Two clear results.

The value tail concentrates beautifully. Holding only the cheapest-on-FCF/EV names (top-N), monthly 2015–25, annual rebalance:

Return rises monotonically as you concentrate (the broad quintile's +19% becomes +34% at the top-3) and the Sharpe peaks around five names (1.52). The cheapest decile isn't merely cheap on average — the most extreme cheapness carried the most signal.

One correction — exclude financials. Free cash flow (CFO − capex) is not a meaningful metric for banks, insurers and REITs: their cash flow is financial flow (deposits, float, claims) or property revaluation, so FCF/EV reads spuriously huge (a bank screened at a 700%+ "yield"). They ranked artificially cheap. Excluding Financials + Real Estate is standard for an FCF screen — and it improves the concentrated sleeve rather than relying on it: top-10 ex-financials returned 31.7%/yr at a 1.43 Sharpe (vs 30.5% / 1.33 with them), shallower drawdown. So the edge is genuine, not a financials artefact. The live FCF Value product uses the ex-financials universe.

The growth tail does the opposite — it does not concentrate. Holding only the 10 lowest-yield names is a basket of cash-burners (−49% drawdown); restricting to the largest low-yield names (big utilities/telecoms in most years) returns just +8.5%/yr. The growth tail only worked as a broad ~98-name basket; concentrate it and it is either junk or a drag. So a concentrated barbell (value-10 + a growth sleeve) underperforms the value sleeve alone — concentrated, the value end is the engine.

This resolves the signal into several usable options, not one: - Broad barbell (§4a) — lowest single-name risk, Sharpe ~1.03, both ends. - Large-cap barbell — the lowest drawdown of any version (−21.7%), keeps the "own the ends" identity. - Concentrated value (the strongest) — the ten cheapest-on-FCF/EV names, Sharpe 1.33 — the natural form for a tradable, forward-tracked product.

The concentrated numbers are the most survivorship-flattered of all (today's cheap survivors are exactly the names that didn't go bust), so treat the levels as an optimistic ceiling — the live forward track is the real test.

4c. Beyond large-caps — does it hold in smaller US firms?

The value premium should bite harder where the market is less efficient, so we ran the same test (FCF/EV, ex-financials, annual rebalance, monthly 2015–25) on the S&P 600 small-cap index (~430 names with usable SEC filings).

Three findings: - The signal holds, with a bigger raw edge over the index. Cheap-on-cash small-caps (Q5) beat the small-cap index by +7%/yr (15.3% vs 8.0%); a 20-name value basket beat it by +12%/yr. And the U-shape replicates — both tails beat the muddled middle, exactly as in large caps. - But it is a higher-risk expression. Lower Sharpe than large-cap value (0.61 vs ~1.0) and much deeper drawdowns (−37% to −47%) — small-caps are simply more volatile. - Concentration inverts. In large caps the top-3/5 cheapest were the best (Sharpe ~1.5); in small caps that flips — the top-3 is a −49% wreck, and you need ~20 names to diversify the noise (top-20 is the best of the concentrated set). Small-cap value wants breadth, not concentration.

So the value signal generalises down the size scale — cheap-on-cash beats the small-cap index too — but the small-cap version is a broader-basket, higher-volatility play, not the tight concentrated one that worked in large caps. Whether it holds in other markets (e.g. the UK) is the open question — a fair test needs longer-history non-US fundamentals we don't yet have cleanly (see Scope).

4d. Does it hold in the UK? (preliminary)

Is this a US phenomenon, or a feature of markets generally? We ran the same FCF/EV test on the FTSE 350 (excluding financials and investment trusts — see the data note). Up front: this is a short, ~3-year window (2023–26) — directional, not definitive — because free UK fundamentals are shallow.

Even on this short window the shape repeats: both tails (Q1 +17.9%, Q5 +16.3%) beat the muddled middle (Q2 was −0.1%), and the cheap value tail beat the FTSE by ~+5–8%/yr at lower volatility than the growth tail. The early read: FCF/EV value looks market-wide, not a US-only effect — though we hold this lightly until we have deeper data.

Data note — why the UK is harder. The US arms use SEC EDGAR: 10+ years of standardised, machine-readable filings, free. The UK has no free equivalent. yfinance carries only ~4–5 years of UK statements; Companies House holds the depth, but files large companies' accounts as PDFs, not tagged data — so a clean decade-long extraction would mean error-prone PDF scraping. Rather than lower the accuracy bar, we report the honest short-window result and leave the deep UK test for when better data is available. (We also had to exclude investment trusts explicitly: free cash flow is meaningless for them, and their NAV-vs-price quirks otherwise corrupt the data.)

5. Conclusion — non-monotonic: own the ends, avoid the middle

In large caps, FCF yield is not a straight-line "cheaper is better" factor — it is U-shaped. The cheap tail (value: banks, energy, pharma, cash-cow tech) and the expensive tail (high-multiple growth and heavy-capex reinvesters like utilities) both beat the market and the equal-weight universe by ~5%/yr; the middle quintiles lagged in 8 of 10 years. A simple high-minus-low spread reads ≈ 0% precisely because it shorts a tail that also wins — which is why the metric looks dead to a linear test and isn't. The usable read: FCF yield separates the market into "has a reason to be owned" — either genuinely cheap, or growing / reinvesting enough to justify the price — versus an undifferentiated middle that underperforms. Using enterprise value instead of market cap sharpens the cheap tail slightly but leaves the shape intact. As a standalone monotonic long/short it is still a poor factor; as a map of where the returns are — the ends, not the middle — it is genuinely informative. The cleanest practical expression is the 50/50 barbell (§4a): it beat the market by ~4.7%/yr at a higher Sharpe (1.02 vs 0.86) across both regimes — which is why we grade FCF yield B+. Concentrated to a tradable basket (§4b), the value end is the engine — the ten cheapest-on-FCF/EV names ran at a 1.33 Sharpe — which is the version we put live, forward-tracked, as the FCF Value portfolio in Portfolio Lab. And the shape is not a large-cap quirk — it repeats in US small-caps (§4c), and an early, short-window read on the UK (§4d) points the same way (value beat the FTSE too). So it looks like a feature of markets generally rather than a US artefact — though the UK leg is preliminary, gated on the deeper data the UK simply doesn't make free.

6. Scope and limitations

• US is deep; the UK leg is short. US large-cap (S&P 500) and small-cap (S&P 600) are both fully backtestable on SEC filings (10+ years). The UK (§4d) is only a ~3-year, directional read: free UK fundamentals are shallow (yfinance ~4–5y) and the deep source (Companies House) files large companies' accounts as PDFs, not machine-readable data — so a clean decade-long UK test isn't available for free.
• Current constituents (survivorship). Today's S&P 500 applied back inflates all returns — and most of all the growth tail, since survivors are disproportionately the growth winners. A point-in-time membership set would be cleaner and would likely trim the low-yield tail more than the high-yield one.
• Equal-weight within quintile; long-only. A cap-weighted or long/short construction would read differently.
• Annual fiscal-year FCF dated at year-end with a filing lag; monthly prices.

7. How we would extend this

1. Track the barbell forward, out-of-sample, in Portfolio Lab — the live test that survivorship can't flatter, and the natural home for a generic rules-based version.
2. A deep UK / non-US test — the early read (§4d) is positive but short-window. A decade-long UK test is gated on data: Companies House files large-cap accounts as PDFs (not tagged), so it needs PDF extraction or a paid feed — revisit when better data is worth the cost.
3. Decompose the low-yield tail — separate genuine growth from heavy-capex reinvesters (utilities); they carry low FCF yield for different reasons and need not persist together.
4. Cheap × quality within the cheap tail — the best-evidenced value combination.

Sources

• Pulled ourselves: S&P 500 + S&P 600 constituents; SEC EDGAR 10-K free cash flow, shares & net debt; total-return prices (incl. IJR, SPY); FTSE 350 constituents + yfinance UK fundamentals/prices (incl. ISF) for the preliminary UK leg.
• Context: Fama-French value (HML); our own quality study (value/quality weak in large caps post-2010).

Changelog

• v1.5 (2026-06-08) — added a preliminary UK leg (§4d): on a short ~3-year window (FTSE 350 ex-trusts), the U-shape repeats and the cheap value tail beat the FTSE by ~+5–8%/yr at lower vol — so the effect looks market-wide, not US-only. Flagged directional: free UK fundamentals are shallow and Companies House files large-cap accounts as PDFs, so a deep UK test isn't freely available (documents the UK data challenge).
• v1.4 (2026-06-08) — extended beyond large-caps (§4c): ran the same FCF/EV test on the S&P 600 small-cap index. The signal holds — the cheap tail beats the small-cap index by +7%/yr (a 20-name value basket +12%/yr) and the U-shape replicates — but it is higher-risk (Sharpe 0.61, drawdowns to −47%) and needs breadth (top-3 blows up; ~20 names is the sweet spot), the opposite of large-caps. UK remains the open frontier (data-limited — yfinance current-only, Companies House the deep-but-raw route).
• v1.3 (2026-06-07) — added the concentration analysis (§4b): the value tail concentrates strongly (top-10 Sharpe 1.33, top-5 1.52; return rises to +34%/yr at top-3), while the growth tail does not concentrate (naive bottom-10 = cash-burners, −49% DD; large-cap low-yield +8.5%/yr). Resolves the signal into several options; the concentrated value sleeve is the live Portfolio Lab product.
• v1.2 (2026-06-07) — added the monthly barbell test (§4a): a 50/50 of the two tails returned +18.3%/yr at a 1.02 Sharpe (best of the baskets; SPY 0.86), information ratio +0.75, drawdown −24% vs the value tail's −30%; the tails are +0.39 correlated in excess-over-market returns. Grade raised C+ → B-.
• v1.1 (2026-06-07) — re-examined the shape using the full Q1–Q5 profile (the v1.0 run kept only Q1/Q5). FCF yield is U-shaped — both tails beat the market and equal-weight by ~+4.5%/yr vs the middle; the +0.8% high-minus-low spread is a linear artefact. Added FCF/EV (spread +1.7%, shape unchanged) and constituent evidence for the two tails. Conclusion changed from "weak value overlay" to "non-monotonic — own the ends, avoid the middle."
• v1.0 (2026-06-07) — first full run: S&P 500, 10 annual rebalances 2015–24, no look-ahead. Reported the Q5−Q1 spread (+0.8%/yr) and read it as a weak large-cap signal / value overlay.