Share Buybacks — only the real ones, only when cheap

The short version

Companies buying back their own shares is supposed to be bullish — but a basket of the biggest buyers actually lagged the market, and the old "pop" on a buyback announcement stopped working years ago. It only helps in a narrow case: when the share count genuinely shrinks and the stock is cheap. A vague "they're buying back stock" is not a reason to buy.

Abstract

We tested whether buying companies that repurchase stock still beats the market. Our own pull of the buyback ETF vs the S&P shows the buyback basket has underperformed recently (PKW −0.7pp/yr over 19 years, ~−2pp/yr over the last 10), and the academic announcement drift is documented as gone post-2002. What survives is narrow: a value-conditioned drift, and only for buybacks that genuinely shrink the share count (not cosmetic ones offsetting stock-comp). We verified that real reductions do happen on our names (WEX −20%, AutoZone −36%, Keller −5%, QinetiQ −8%). Conclusion: keep it only as the rule we already use — real share-count reduction plus cheapness — never a generic "they're buying back stock" buy.

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

Companies that buy back stock — especially cheap ones genuinely retiring shares — are said to earn positive long-run drift. If true, it's a low-turnover, long-side strand.

2. What the literature claims — and our scepticism

• Ikenberry, Lakonishok & Vermaelen (1995): +12.1% abnormal over 4 years, +45.3% for value stocks — but 1980–90 data, and ~0% for "glamour" names.
• Fu & Huang (2016, Management Science): the long-run abnormal return disappears for events announced 2003–2012 — the classic drift decayed as the market got more efficient.
• Our scepticism: the headline magnitudes are old, long-short, gross, and authored partly by the signal's own champions. We check the recent, long-only, after-fee reality ourselves.

3. Our own analysis ← the heart of the report

Data we pulled: yfinance total returns (dividends reinvested) for the buyback ETF (PKW) vs SPY; and raw shares-outstanding histories to verify real count reduction.

• PKW (Buyback Achievers) vs SPY, 2006→2026 (19y): PKW +10.31%/yr vs SPY +10.99%/yr → −0.69pp/yr (the buyback basket underperformed). Independent 10-year figures put it nearer −2pp/yr.
• Real share-count check (the test we actually use): WEX −19.6%, AutoZone −36.0%, Keller −4.9%, QinetiQ −7.9% over ~2018→2026 — genuine reductions. (Honesty flag: yfinance returned AAPL +209% and BKNG +1573% — these are split-adjustment artifacts, not real issuance; discarded, not reported as fact.)

4. Findings

• The buyback basket has not beaten the market recently (PKW −0.7 to −2pp/yr).
• The classic announcement drift is gone post-2002 (Fu & Huang).
• Real share-count reduction is verifiable and is the right gate — but it only carries return conditioned on cheapness (value, not glamour; Ikenberry).

5. Conclusion — keep, but only the narrow rule we already use

A buyback is a strand only when (1) diluted shares actually fall year-on-year (net of stock-comp — not a board authorisation, not a dilution-offset), and (2) the name already screens cheap. That is exactly the test we applied to WEX/Keller/QinetiQ. A generic "they're buying back stock" is not a buy reason, and the buyback basket as a whole is not an edge. Note this is the same capital-allocation factor as net share issuance — never count both (see families map).

6. Limitations & open questions

• PKW is one ETF with its own construction/fee; a clean test is a value-conditioned, net-buyback- yield single-name sort on our universe — which needs point-in-time fundamentals + survivorship- free prices we don't yet have.
• Share-count history from yfinance is noisy (the split artifacts above); free EDGAR financial-statement filings make it clean.

7. How we would extend this

Build a net-buyback-yield × cheapness single-name sort — share counts from free SEC financial-statement filings, prices from the history we hold — and test whether real-reduction + cheap still earns drift recently. All on free data.

Sources

• Pulled ourselves: yfinance total returns (PKW, SPY); shares-outstanding histories.
• Cited: Ikenberry, Lakonishok & Vermaelen (1995, JFE 39); Fu & Huang (2016, Management Science 62(4)).

Changelog

• v1.0 (2026-06-01) — PKW-vs-SPY pull + share-count verification. Single-name net-yield sort pending data.