Hendrik Bessembinder, at Arizona State, has spent years asking a question that sounds almost silly until you look at the data: if the stock market creates enormous wealth in aggregate, why do most individual stocks fail? His 2018 paper answered it for 1926 to 2016. In March 2026 he extended the work to a full century — 29,754 common stocks, January 1926 through December 2025.
The average lifetime buy-and-hold return across all of them is over 30,000%.
The median is −6.9%.
Read those two numbers next to each other. The typical stock in the greatest wealth-creating machine in economic history lost money over its listed life.
That is not a paradox, and it is not a reason to stop picking stocks. But it is a constraint, and it has a specific shape. Understanding that shape changes two things: how many positions you hold, and how long you hold them. Neither of those is a stock-picking decision. Both of them matter more than most stock-picking decisions.
(One note before we go further: the 2026 update is an initial draft and has not been peer-reviewed. The methodology is identical to the 2018 paper, which was published in the Journal of Financial Economics and is one of the most-cited findings in the field. Bessembinder also publishes the underlying spreadsheet, so every figure below is checkable by you rather than takeable on trust.)
Start with the headline figures, because they are unusually clean.
ONE HUNDRED YEARS OF US COMMON STOCKS, 1926–2025
─────────────────────────────────────────────────────────────────
Stocks in the sample 29,754
Mean lifetime buy-and-hold return >30,000%
Median lifetime buy-and-hold return −6.9%
─────────────────────────────────────────────────────────────────
Generated a positive lifetime return 48%
Beat one-month Treasury bills ~41%
Beat the broader market 28%
─────────────────────────────────────────────────────────────────
Total net shareholder wealth creation $90.96 trillion
Firms accounting for HALF of it 46
Firms accounting for ALL of it 1,082 (3.72%)
Firms that destroyed wealth vs T-bills 59%
─────────────────────────────────────────────────────────────────
Source: Bessembinder, One Hundred Years in the U.S. Stock Markets,
working paper, March 2026.
The gap between the mean and the median is the whole story. Returns are not distributed symmetrically. A stock can rise by ten thousand percent; it can only fall by one hundred. A small number of extraordinary outcomes drag the average far above anything the typical holding delivers.
Statisticians call this positive skewness. For an investor it means something more practical: the headline “stocks return about 10% a year” is a fact about the market, not a fact about stocks. It describes what happens to a portfolio that owns everything. It does not describe what happens to a portfolio that owns twelve things.
And the concentration is intensifying rather than easing. In the 2018 study, 89 firms accounted for half of $42.6 trillion of net wealth creation. In the update, total wealth creation more than doubled to $91 trillion — and the number of firms responsible for half of it fell to 46. Apple and Nvidia together account for roughly a tenth. Nvidia alone created about $4.51 trillion of shareholder wealth after 2016, close to a tenth of everything created in that period.
The haystack grew. The needles got fewer.
This is the part I found most useful, and it is rarely quoted.
Bessembinder breaks the century into decades. The deterioration in individual stock outcomes since the mid-1980s is stark — and it happened while the market as a whole was doing extremely well.
DECADE-HORIZON OUTCOMES FOR INDIVIDUAL STOCKS
─────────────────────────────────────────────────────────────────
1926–1985 1986–2025
(first six) (last four)
─────────────────────────────────────────────────────────────────
Median decade return 63.6% 5.8%
Share beating T-bills ~61% 48%
─────────────────────────────────────────────────────────────────
Meanwhile: the value-weighted market portfolio performed
strongly across 1986–2025.
Two different things happened at once. The index did well. The median stock inside it did not.
The most likely explanation is compositional rather than mysterious. Fama and French documented that the 1970s and 1980s brought a wave of younger, smaller, more speculative companies onto US exchanges. Many were short-lived. They dragged the middle of the distribution down while the largest winners pulled further away from it.
The practical implication is uncomfortable and worth sitting with: whatever hit rate you assume for your own selection, the base rate against which you should be measuring it has been falling for forty years.
Here is a complete method, not a teaser. It takes about two minutes and it will probably change your position count.
If 3.72% of all listed firms generated the entirety of net wealth creation, then the first question for any concentrated portfolio is not “are my picks good?” It is: what is the probability my portfolio contains any of them at all?
For randomly selected holdings, that is a single line of arithmetic:
P(at least one winner) = 1 − (1 − p)^N
where p = 0.0372 (the 3.72% that produced all net wealth creation)
N = number of positions
─────────────────────────────────────────────────────────────────
POSITIONS P(AT LEAST ONE P(AT LEAST ONE OF
HELD OF THE 3.72%) THE 46, p = 0.0016)
─────────────────────────────────────────────────────────────────
1 3.7% 0.2%
5 17.3% 0.8%
10 31.6% 1.6%
12 36.6% 1.9%
20 53.2% 3.2%
30 67.9% 4.7%
50 85.0% 7.7%
100 97.7% 14.8%
─────────────────────────────────────────────────────────────────
A twelve-position portfolio, selected at random, has roughly a one-in-three chance of containing a single one of the companies that produced all of the market’s net gains.
Now the two adjustments that make this usable rather than merely bleak.
Adjustment one: this assumes random selection. That is the entire point. The table is not your fate — it is the hurdle. If you believe you have an edge, this is the number that edge has to beat. Most investors have never written that number down, which means they have never actually specified what their edge would need to accomplish. A process that lifts your hit rate from 3.72% to 8% is doing genuine work. A process that lifts it to 4.5% is not, and you should probably know which one you have.
Adjustment two: this is a backward-looking universe. The 3.72% is drawn from every firm that has ever listed in the US over a century — including thousands that no longer exist, many of which were small, speculative and short-lived from the outset. If you are picking from the roughly four thousand companies listed today, and particularly if you exclude the smallest and least established, your starting distribution is genuinely different. Better, probably. Nobody knows by how much. Anyone who tells you they do is guessing.
What to do with it. Run the calculation on your own position count tonight. Then ask yourself an honest question: is my portfolio concentrated because I have an edge that justifies fewer, larger bets — or is it concentrated because holding thirty names felt like too much work?
Those are very different portfolios that look identical on a statement.
Suppose you clear the first hurdle. You own one.
Here the data gets genuinely difficult, and it comes from a different source: Morgan Stanley’s Counterpoint Global has studied the drawdowns experienced by the biggest wealth creators.
Six companies — Apple, Microsoft, Nvidia, Alphabet, Amazon and ExxonMobil — added $17.1 trillion of shareholder wealth. Their average maximum drawdown was 80.3%.
That figure is roughly the same as the average maximum drawdown across the full sample of stocks. Which is to say: owning the best companies of the century did not spare you the experience of owning the worst.
WHAT HOLDING A WINNER ACTUALLY FELT LIKE
─────────────────────────────────────────────────────────────────
Amazon −95% Dec 1999 → Oct 2001. Took until Oct 2009 —
more than eight years — to make a new high.
Lifetime wealth creation from its 1997 IPO
to end-2024: about $2.1 trillion.
Apple −82% Twice. Once 1991–1997, again 2000–2003.
Nvidia −89.7% Bottoming Oct 2002; recovery took roughly
1,032 trading sessions. It has fallen about
90% on two separate occasions. Average time
to recover from a major drawdown: ~41 months.
Longest: 103 months.
Microsoft −65% 2007–2008.
─────────────────────────────────────────────────────────────────
Average maximum drawdown, six largest wealth creators: 80.3%
─────────────────────────────────────────────────────────────────
The arithmetic of recovery is worth stating plainly, because it is the reason most people do not survive these episodes. A 50% loss requires a 100% gain to get back to even. An 80% loss requires 400%. A 90% loss requires 900%.
And you have to hold through it while every available signal — the price, the press, your own spreadsheet, possibly your clients — tells you that you were wrong. In the case of Amazon, for more than eight years.
So the 4% problem has two halves, and almost all of the attention goes to the first one:
Your outcome = P(you owned one) × P(you still owned it at the end)
──────────────── ──────────────────────────────
selection retention
(endlessly discussed) (barely discussed at all)
I think retention is where more money is actually lost. Selection failures are visible and get analysed. Retention failures get filed under “I took profits,” which sounds like a decision rather than an error.
Now the part that changed how I read all of the above.
You would expect the greatest wealth creators of the century to be the fastest compounders. They are not.
Among Bessembinder’s top 30 stocks by cumulative return, the median annualised return was 13%. Not 40%. Not 25%. Thirteen percent — a good return, comfortably above the long-run market average, and completely unremarkable in any single year. Nobody would have written a newsletter about it.
What made those outcomes extraordinary was duration. Those stocks were listed for an average of 93.9 years.
The clearest illustration in the data is a comparison between two unglamorous businesses:
THE COST OF TWO PERCENTAGE POINTS, OVER A CENTURY
─────────────────────────────────────────────────────────────────
Altria (formerly Philip Morris) 16.53% annualised
$1 → about $4.42 million
Vulcan Materials 14.35% annualised
─────────────────────────────────────────────────────────────────
Altria's annual rate was 1.18× Vulcan's.
Altria's cumulative outcome was 8.81× as large.
─────────────────────────────────────────────────────────────────
Roughly two extra percentage points a year, sustained for a hundred years, produced an eightfold difference in the final number. That is compounding doing what compounding does, and it is almost impossible to feel in real time, because in any given year the gap between 14.35% and 16.53% is invisible.
And then there is the observation that I keep coming back to:
Not one of the thirty highest-annualised-return stocks also appears among the thirty highest cumulative-return stocks.
Not one. The stocks that compounded fastest and the stocks that created the most wealth are two entirely disjoint sets. The fast ones did not last. The lasting ones were never fast.
If you have spent any time screening for high growth rates, that sentence deserves a few minutes of thought.
The most common reading of this research is that it is a case for indexing, and that reading is intellectually serious. Bessembinder himself noted in 2018 that the results help explain why active strategies, which tend to be poorly diversified, most often underperform. William Sharpe’s arithmetic of active management makes the same point from first principles: before costs, the average actively managed dollar must match the market, because active investors collectively are the market. After costs, it must trail. Bogle’s version was blunter — don’t look for the needle, buy the haystack.
I am not going to pretend that argument is weak. It isn’t. Anyone who picks individual stocks should be able to explain why they are accepting a structurally worse base rate, and “I enjoy it” is a legitimate answer as long as it is stated honestly.
But there is a second reading, and it is the one this letter is built on.
If wealth creation is concentrated in a tiny number of companies, and if what distinguishes those companies is duration rather than velocity, then the useful analytical question is not “what will grow fastest over the next three years.” It is: which of these businesses is still going to be earning an above-average return on capital in 2045, and what would have to be true for that to hold?
That is a different question from the one the market spends its time on. It is not a forecasting question — nobody can forecast twenty years of anything. It is a structural question about what makes a competitive position degrade slowly rather than quickly.
It is also, conveniently, the question that almost nobody is competing on. Sell-side research is a machine for producing next year’s earnings estimate. There is no consensus estimate for how long.
Two numbers. Both take under ten minutes.
One. Count your positions. Look up the inclusion probability in the table in section 3. Write it down.
Two. Take your largest holding and find its worst peak-to-trough decline over the past ten years. Then ask, honestly: had I owned it at the top, at the size I currently own it, would I still own it today?
Most people find that their portfolio is built for a distribution that does not exist — concentrated enough to need winners, but managed in a way that would have ejected every winner in the table above somewhere around month fourteen.
Everything above is the diagnosis, and the diagnosis is public. You can download Bessembinder’s spreadsheet yourself.
What the data does not give you is a method. It tells you that duration is the variable that mattered. It does not tell you how to identify duration in advance, which is the only version of this problem worth solving.
That is what the premium section is for.
The Duration Screen. Seven observable characteristics that separate businesses whose advantage degrades slowly from those whose advantage degrades fast — built from what the century’s actual survivors had in common, not from theory. Each with the specific line item where you can check it, and the threshold I use.
The Drawdown Pre-Commitment. A one-page document you write before you buy, specifying the decline you are agreeing to hold through and the three conditions — and only those three — that would release you from it. This is the single change that most improved my own retention, and I explain exactly why the “three conditions” constraint does the work.
The Survivorship Audit. How to run the base-rate arithmetic on your actual portfolio, including the adjustment for a modern listed universe rather than a hundred-year one. With a worked example on a real twelve-position book.
The velocity trap, in detail. Why none of the thirty fastest compounders made the thirty largest wealth creators, what happened to them instead, and the four screening habits that push you toward the fast list and away from the durable one. If you screen on growth, this section is aimed at you.
Where duration is observable today. Not tickers as gifts — a walk through the specific places in this market where the duration characteristics are present and being priced as though they are not, and three places where they appear present and, on my reading, are an illusion.
And the position that taught me this. Named, with the entry, the exit, the return I actually got, and the return the same holding produced for anyone who did nothing. The gap between those two numbers is the most expensive thing I have ever paid to learn.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.