Characteristics of Wealth-Creating Companies

Active stock selection must confront a brutal statistical reality documented by Professor Hendrik Bessembinder.¹²⁹ Across the 29,078 U.S. common stocks in the CRSP database from December 1925 through December 2023, 51.6% delivered a negative cumulative return over their entire listed lives. All of the net wealth the U.S. stock market created is attributable to roughly the top thousand stocks — about 4% of the sample — while the other 96% collectively did no better than one-month Treasury bills. Concentration inside that top slice is extreme in its own right: 86 stocks account for half of the total.

The implication is stark, and it is not the one most people draw. It is not that you should try harder to find the winners; it is that a portfolio of ten or twenty names you picked yourself is overwhelmingly likely to contain none of them, and the modal outcome of stock picking is therefore not “slightly below the index” but “materially below Treasury bills.” The index works precisely because it holds the winners by default.

There is a second funnel stacked in front of that one, and it explains why the first is so brutal. Nordhaus measured how much of the social value created by innovation the innovating firms actually keep — what he calls Schumpeterian profits — and found the answer to be roughly 2.2%.¹³⁰ The other 98% escapes to consumers as lower prices and better goods, and to imitators who copy the innovation once it is proven. So the chain running from “this company changed the world” to “this position made me money” passes through two consecutive bottlenecks: the firm captures about two cents of every dollar of value it creates, and then roughly 4% of firms account for all the net shareholder wealth inside that remainder. Being right about the technology is not the same as being right about the stock, and it is not even close. The consolation is that the missing 98% did not vanish — you collected it as a customer, whether or not you ever owned the shares.

Why the evidence does not stop anyone, including you. Numbers this stark should end the argument, and they never do, because of the self-serving bias — the well-documented tendency to attribute good outcomes to your own judgment and bad ones to circumstance. The winning position was a thesis you researched; the loser was a rate shock, a bad quarter, an analyst who lied. Both attributions feel like memory, not interpretation, which is exactly what makes the bias so durable: your track record is not stored as a track record, it is stored as a series of explanations. The only reliable antidote is mechanical. Write the thesis, the price, and the disconfirming condition down before you buy, and score the position against what you wrote instead of what you now recall having thought. A benchmark works the same way — it is a number that does not care what you meant.

The bias is worse in the people advising you, because there it is paid. Anyone earning a commission on the product they recommend will find their judgment bending toward the recommendation without experiencing anything that feels like bad faith. And do not expect the disclosure to fix it: Cain, Loewenstein, and Moore found that disclosing a conflict often makes the advice worse.¹³¹ Advisers who disclose feel morally licensed to push harder, and recipients fail to discount the advice anything like enough — so the disclosure ends up protecting the adviser, not you. Treat “I am required to tell you I am compensated by the issuer” as information about the product’s cost structure, never as a repair to it. The useful question is not whether a conflict was disclosed but whether it exists, and the only structural answer is to buy advice and products from different people.

Bessembinder’s Part III⁶² catalogs what the top wealth creators looked like while they were creating it, comparing the “Top 200” firm-decades against all others:

Rapid organic sales and asset growth

Mean annual asset growth of 88.7% and sales growth of 95.3% during the decade of peak wealth creation, against 24.7% and 32.9% for other firms, driven mainly by internal reinvestment, not acquisitions. Do not read those as a screen. They are means pulled upward by a handful of young, small, fast-scaling companies — the cross-sectional median annual asset growth is 10.5% and median sales growth 11.6%. A mature company cannot post 88%, and one that does is usually too small and too early for you to have found it.

Operating efficiency

Operating income growing faster than assets, so the income-to-asset ratio rises instead of merely holding.

High R&D commitment

Mean R&D-to-assets of 5.2%, against 3.7% for the comparison group.

Conservative leverage

Lower debt-to-assets (0.47 versus 0.53) and large cash balances, which is what let them keep reinvesting through credit contractions instead of retrenching.

The drawdown you must survive to collect this. Here is the number that should govern how you think about the whole exercise. The Top-200 firm-decades — the greatest wealth creators in the history of the U.S. market, measured during their winning decade — still suffered maximum drawdowns averaging 50.2%. Typical firms averaged 70.1%, and the worst performers 94.8%. There is no version of this where you identify a future compounder and then hold it comfortably. The binding constraint was never stock selection; it is sitting through a halving with the thesis intact, which is why section “Drawdown and Tail Risk” treats drawdown as the risk measure that matters, not volatility.

Can you screen for it in advance? Part IV asks exactly that, and the unvarnished answer is “barely.”¹³² Bessembinder opens by conceding that identifying objectively measurable characteristics that predict extreme outcomes “is challenging,” then finds some statistically significant patterns across 1960–2019. Firms that went on to the highest decade-horizon returns tended to be younger, to have had larger drawdowns in the prior decade, and to have spent more on R&D. Firms that went on to the worst returns were more levered, less profitable, spent less on R&D, and had more volatile prior-decade returns.

Note the trap in that first list, because it inverts the Part III finding. Winners had smaller drawdowns within their winning decade but larger drawdowns in the decade before it. The predictive signal is beaten-down young research spenders, not proven compounders — which is psychologically the opposite of what a screen built on Part III’s descriptive statistics would hand you, and a good deal harder to hold.