You Don't Have a Concentrated Portfolio. You Have an Unmeasured One.

Every conversation about diversification eventually meets the same wall. Someone leans back, smiles, and says: "If I'd just held NVDA from 2016, I wouldn't need a factor model." And they're right. They wouldn't. A single position, held with conviction, would have outperformed nearly any optimized portfolio you could have built. This is the multibagger fantasy, and it has a powerful gravitational pull on how investors think about their portfolios.

The fantasy isn't wrong. Concentrated bets do create generational wealth. The data is unambiguous on this, and we'll get to it shortly. But the fantasy is also incomplete in ways that quietly shape how investors construct portfolios. Because the question isn't whether holding NVDA from 2016 would have made you rich. It's whether the process that leads someone to hold NVDA from 2016 would have also led them to hold Cisco from 1998, or GE from 2000, or Peloton from 2020. With the same conviction. The same thesis structure. The same emotional certainty.

Stock picking works for some people. But the data on how rare the winning stocks actually are is worth sitting with. And once you do, what gets called "stock picking" often turns out, on closer look, to be something different.

The shape of the distribution

In 2018, Hendrik Bessembinder published a study that quietly reframed the entire active-versus-passive debate. He looked at every common stock listed in the CRSP database (the Center for Research in Security Prices, hosted at the University of Chicago - the standard academic database of U.S. stock prices and returns going back to 1926) from 1926 onward and asked a deceptively simple question: how many of them, over their lifetime as public companies, beat one-month Treasury bills?

The 2024 update of his work, covering 29,078 stocks from 1925 through 2023, found that 51.6% of all stocks had negative cumulative returns over their lifetime. The median stock didn't just underperform the market. It lost money outright, with a median lifetime return of negative 7.41% per year. The average stock remained publicly listed for only 11.6 years before being delisted, acquired, or going to zero. Just 31 stocks survived the full 98-year window.

And yet, over the same period, $1 invested in the broad U.S. stock market in 1926 grew to $229.40 by 2023. The market worked. The math, on its face, doesn't reconcile.

The reconciliation is in the shape of the distribution. Almost all of that wealth creation came from a vanishingly small number of stocks. Just 4% of listed stocks account for the entire net wealth creation of the U.S. stock market above Treasury bills. The other 96% collectively matched cash. At the extreme, 17 stocks delivered cumulative returns greater than five million percent. Altria Group sits at the top with a cumulative return of roughly 266 million percent.

This isn't a quirk of ancient data. JP Morgan's recurring Agony & Ecstasy studies, focused on the modern era, find that roughly 40% of all stocks ever in the Russell 3000 since 1980 suffered a "catastrophic loss". They define this as a 70%+ decline from peak from which the stock never recovered. The median stock underperformed the index by 54% over its lifetime. Two-thirds underperformed.

The conclusion isn't that stock picking doesn't work. It's that stock picking is a game with extremely skewed odds. The base rate of picking a winner that meaningfully drives portfolio returns is genuinely low. Not because investors are unintelligent, but because the underlying distribution is brutal. Most of the wealth is in the tail. Miss the tail, and you've spent years matching cash.

The survivorship trick your memory plays

The multibagger fantasy gets its power from the way memory works. We don't reason about stocks the way the data does. We reason about them the way memory does.

Walk through the actual cognitive sequence of "I would have held NVDA." It usually looks something like this. NVIDIA exists today, has obvious dominance in AI, has a chart that goes up and to the right, and a coherent narrative about chips and parallel computing that anyone can articulate now. Working backward from a known outcome, the conviction feels obvious.

But the same cognitive process, applied to a different ten-year window, produces:

"If I'd just held Cisco from 1998, when it was the dominant infrastructure play of the internet build-out..."

"If I'd just held GE from 2000, when it was the most admired company in America..."

"If I'd just held Intel from 2020, when it had unassailable share in the most important silicon category in the world..."

"If I'd just held Peloton from 2020, when at-home fitness was the obvious post-COVID secular trend..."

Each of these was, at the time, exactly the same kind of thesis. Dominant company, obvious narrative, conviction that anyone with attention to the market could have. Each is now a cautionary tale. The investor who holds NVDA from 2016 doesn't feel lucky. They feel smart. The investor who held Cisco from 1998 with identical reasoning feels stupid. Same process. Different draw from a wildly skewed distribution.

This is survivorship bias, but a particular flavor of it. The kind that operates in the first person. We retroactively credit ourselves with conviction we would have applied uniformly to any of the names that fit the same pattern at the time.

And there's a second mechanism worth being aware of. We don't just rewrite our past convictions to flatter ourselves. We selectively forget the bets that didn't work.

The behavioral finance literature has a name for the first part: hindsight bias, or the "knew-it-all-along effect," documented by Baruch Fischhoff in 1975. Once an outcome is known, our memory reconstructs the prior probability we assigned to it, inflating our sense of foresight. The dot-com bubble, the 2008 crisis, the rise of AI: in retrospect, each feels like it was obviously predictable. At the time, almost no one predicted them.

The second mechanism is more recent, and more specific to investing. A 2023 study published in the Journal of Behavioral and Experimental Finance by Li and Rong asked real investors to recall their best- and worst-performing stocks. The finding was direct: investors were systematically more likely to forget their worst-performing stocks than their best. The proportion who remembered their winners but forgot their losers was substantially higher than the reverse. The bias was stronger for less experienced investors and stronger for those who paid more attention to stock prices.

Put the two together, and the picture gets fuller. The past gets reorganized around the winners. The losers fade. The investing history we recall ends up cleaner and more decisive than the one that actually happened. It is worth noticing this when imagining what we would have done in 2016.

This doesn't mean concentration is wrong. It means concentration is a high-variance bet, and the variance isn't always what investors picture in their heads when they imagine it.

What "stock picking" usually turns out to be

There's another layer worth looking at, and it tends to surface in unexpected ways.

When a holdings-based factor model is run on a typical "high-conviction stock-picker" portfolio (eight or ten names, mostly large-cap tech and consumer growth, accumulated over a few years), the result is rarely "this investor has unique stock-selection skill." More commonly, the decomposition looks something like the following hypothetical example:

Beta of 1.3 to the market. Heavy tilt toward the momentum factor, the growth factor, and the low-volatility-of-the-bull-market factor. Negative exposure to value. Sector concentration of 60%+ in technology, much of which is correlated to the same underlying drivers (rates, AI capex cycle, semiconductor demand). The "uncorrelated picks" turn out to be 0.85+ correlated to QQQ.

The specific numbers vary from portfolio to portfolio. The pattern, in our experience, repeats.

Often, what looks like stock-picking skill is, on decomposition, a leveraged bet on a small number of well-known factor exposures that happened to work for a particular decade. The concentration is real. What gets missed is what the concentration is actually in. Ten different tickers can express a single underlying factor bet.

This isn't an indictment. It's a measurement question. And it has a real cost. If a portfolio is, in effect, a leveraged growth-momentum bet, the regimes that hurt growth-momentum will hurt it harder than expected. The drawdowns can feel idiosyncratic ("why is everything down at once?") when they are actually the predictable behavior of a single factor exposure unwinding.

Concentration isn't the problem. Unmeasured concentration is.

The honest framing

None of this argues against concentrated investing. There are investors with genuine edge. Fundamental analysts who do work others don't. Traders with informational advantages. Allocators with patient capital who can underwrite long drawdowns. For them, concentrated bets, sized appropriately, are how edge gets monetized.

The question is whether you know which one you are.

The investors with genuine edge tend to share a habit. Before the trade, they can articulate what kind of market environment will hurt the position, what the factor exposures look like, what the correlation to the existing book is, and what would falsify the thesis. The story comes first, the trade follows. The reverse pattern, where the explanation gets assembled after the price has moved, is more common than people tend to admit.

The data from Bessembinder and Cembalest doesn't tell you not to make concentrated bets. It tells you the population-level distribution you're sampling from. It tells you the base rate. It says: here is the prior. What's your evidence that you're sampling from a different distribution?

Where this leaves the multibagger conversation

The investor who held NVDA from 2016 didn't do anything wrong. They participated in something genuinely rare and were rewarded for it. The reward is real. The wealth is real. The story they tell themselves about why they were rewarded is the part worth examining.

If the answer is "I had non-public insight into transformer architectures and TSMC fab capacity that the market hadn't priced," that's edge, and concentration was the right way to express it. If the answer is "I read about AI and bought the obvious leader," that's a draw from a distribution that produced Cisco for an entire generation of investors who reasoned identically in 1999.

Both can be true at different times. Neither investor is required to apologize. But only one is in a position to do it again.

The investors who compound capital across decades tend to be the ones who, after picking NVDA, look at the rest of the portfolio and ask: what bet am I actually making now? What kills it? How big should this be? Without an honest answer to that question, it becomes hard to separate gains that came from skill from gains that came from a particular regime being kind. Both can feel the same from inside the portfolio. They are not the same thing.

When was the last time you looked at your portfolio and asked, honestly, what bet you were making. Not stock by stock, but in aggregate?

---

References

Bessembinder, H. (2018). Do Stocks Outperform Treasury Bills? Journal of Financial Economics, 129(3), 440-457.

Bessembinder, H. (2024). Which U.S. Stocks Generated the Highest Long-Term Returns? SSRN Working Paper. CRSP data through December 2023; 29,078 stocks.

Cembalest, M. (2014, 2021, 2024). The Agony & The Ecstasy: The Risks and Rewards of a Concentrated Stock Position. J.P. Morgan Eye on the Market, recurring publication.

Fischhoff, B. (1975). Hindsight != Foresight: The Effect of Outcome Knowledge on Judgment Under Uncertainty. Journal of Experimental Psychology: Human Perception and Performance, 1(3), 288-299.

Li, K. K., & Rong, K. (2023). Real-life Investors' Memory Recall Bias: A Lab-in-the-Field Experiment. Journal of Behavioral and Experimental Finance, 37.