Founder-Led Firms' 3.1× Edge: How Much Is Survivorship, How Much Is Real

The short answer. A widely-cited Bain statistic (Zook & Allen, The Founder's Mentality, 2016) says founder-led S&P 500 firms returned ~3.1× the rest (1990–2014), sold as proof that a "founder's mentality" drives superior performance. That raw number is built on current index membership, so it is survivorship- and look-ahead-prone: a zero-skill Monte-Carlo null — identical expected returns, the founder cohort merely more volatile — run through the same survive-and-be-large filter reproduces a large apparent gap. But how much it reproduces depends entirely on an assumption we did not measure: the founder cohort's volatility. At an assumed ~1.8× it recovers 76% of the excess gap (the distance from 1× to 3.1×); across a plausible range it swings from 26% to 179% (so the "% of 3.1×" column below is share-of-excess, not of the raw multiple). So "mostly survivorship" is not established — only that survivorship can inflate the raw ratio a lot. And it does not dispose of the founder question: the best controlled study (Fahlenbrach 2009) finds a founder-CEO abnormal return of ~+4.4%/yr that survives risk and characteristic controls. The honest verdict is in between — Bain's raw 3.1× is confounded and uninformative, but a smaller, real, and narrow founder premium survives.

The claim under test. That being founder-led causes ~3.1× better returns — a broad, recoverable performance edge.

Why the raw number is suspect. Bain splits the S&P 500 into "founder-led" (founder was CEO or on the board) vs "other," and compares the returns of firms in the index today. Founder-controlled firms that failed delisted and dropped out; the winners survived and get counted. Comparing the survivors of one cohort to another selects on the outcome (look-ahead inclusion) — and if the founder cohort is more volatile, its surviving tail looks spectacular even with no edge in expected return.

The null: survivorship can manufacture most of it — conditionally

Two cohorts, identical expected return (zero skill difference). The only difference: the founder cohort is more volatile and delists more. Apply the same index rule — survive the full period and be large enough at the end — then compare returns.

The mechanism is real and textbook: truncating a higher-variance sample by survival manufactures apparent outperformance even from noise (Brown, Goetzmann, Ibbotson & Ross, 1992). But read the table as a warning about our own claim, not a proof: the result is nearly linear in the one number we assumed. We picked 1.8× because it lands near three-quarters of the headline — that is a researcher degree of freedom, not a measurement.

The load-bearing weakness: we never measured founder-firm volatility

The whole "76%" rests on founder firms being ~1.8× as volatile (≈31%/yr vs ≈17%/yr). We did not measure that, and the evidence is genuinely mixed: founder-led firms skew younger/smaller/tech, which raises idiosyncratic volatility — but **family- and founder-controlled firms are repeatedly found more conservative and less volatile** than peers. If founder volatility is at or below the market's, the null reproduces far less (at 1.4× it is only 26%). (A coincidence to avoid: 1.8× also happens to be Bain's ex-tech return multiple — it is not a volatility figure.) Grounding founder-firm volatility against real return data is the open task; until then the honest reproduction is a range, not a point.

Two more honesty checks cut against a strong survivorship read:

• "Tail-driven" is not a fingerprint. Our null's mean (2.6×) far exceeds its median (1.58×), and we originally read that as the survivorship signature. But all equity returns are right-skewed (a few big winners dominate any cohort — Bessembinder 2018), so mean ≫ median is the baseline for any buy-and-hold sample. It is consistent with survivorship, not diagnostic of it.
• The controlled study finds a real edge. The decisive test is our own falsifier: a start-defined, delisting-inclusive cohort with risk controls. Fahlenbrach (2009) essentially ran it — a forward-formed, equal-weighted founder-CEO portfolio (1993–2002) earned +8.3%/yr benchmark-adjusted, and +4.4%/yr abnormal survives controls for firm characteristics, CEO characteristics, and industry. Survivorship and size are removed and the alpha persists, so a pure-survivorship story is refuted at the conditional level. Our null and Fahlenbrach test different estimands: the null shows Bain's unconditional 3.1× is uninformative; Fahlenbrach shows a conditional founder alpha nonetheless exists.

What actually survives

Put the pieces together and the truth is neither "mentality magic" nor "pure artifact":

• The raw 3.1× is inflated by survivorship + look-ahead inclusion of firms selected for having ended up large. Take it as marketing, not evidence.
• A real but smaller founder-CEO premium survives controls (~+4.4%/yr, Fahlenbrach 2009) — the survivorship null does not explain that away.
• The premium is narrow and decaying: Bain's own update puts it at ~2.1× since 2015 (down from 3.1×), and in practice it concentrates in a handful of mega-founders (Nvidia, Tesla, Meta). The one pure-play vehicle built to harvest it, the Global X Founder-Run Companies ETF (BOSS), was liquidated in 2023 — you cannot buy "founders" as a broad factor.

This is the recurring Crucible shape, but with the correction the earlier drafts missed: a headline that is partly a property of how the sample was built — like the Good to Great "leap", the nudging 2.5× ratio, and the LLM-judge length read — and a real, smaller effect underneath that the artifact story must not erase.

What this does and does not say. It does not show founder-led firms have no edge — the controlled literature says they do, if a modest and concentrated one. It does show Bain's raw 3.1× cannot support a broad causal claim: survivorship + look-ahead inclusion of a possibly-higher-variance cohort can inflate it substantially, the exact share is unmeasured, and the honest number is the controlled ~+4.4%/yr, not 3.1×.

The falsifierWe predicted that a delisting-inclusive, start-defined cohort would shrink the gap toward the tail residual. The controlled version of that test (Fahlenbrach 2009) shrinks it to ~+4.4%/yr but does not kill it. What would still move the verdict: a direct measurement of founder-firm return volatility (to pin the null's reproduction share) and a modern replication of the founder alpha post-2009 (the premium appears to be decaying).

FAQ

Is the founder's mentality a myth? No — and neither is the 3.1×'s survivorship problem. The raw 3.1× is inflated by how Bain's index was built, but a smaller founder-CEO premium (~+4.4%/yr) survives risk controls (Fahlenbrach 2009). It is real, modest, and concentrated in a few names — not a broad 3.1× factor.

What is survivorship / look-ahead inclusion here? Bain compares founder firms still in the index, so failed founder firms were deleted before counting. Selecting a cohort by its survival and end-size inflates whichever cohort has the fatter surviving tail.

So how much of the 3.1× is survivorship? We can't say precisely — it depends on how much more volatile founder firms are, which we didn't measure. Our null reproduces 26–179% of it across a plausible volatility range; the honest takeaway is that the raw ratio is confounded, and the controlled edge is ~+4.4%/yr.

Why did you soften your own earlier claim? Because the "76% is survivorship" figure rested on an unmeasured volatility assumption, the mean≫median "tell" is a property of all equity returns, and a controlled study (Fahlenbrach) finds a real founder alpha. The Crucible keeps the receipts, including the ones that qualify us.

Is this just a simulation? The null is — deliberately the smallest one that isolates survivorship + look-ahead inclusion, with all assumptions stated and swept. The empirical counterweight (Fahlenbrach) is real, peer-reviewed data. Runnable: research/probes/founder_survivorship_null.py.