Tradiens
EssaysSign in

Tradiens Research · Data essay · Equities

Pick a Stock at Random and the odds say it will lose to cash

Most US stocks have done worse than cash, which here means short-term US government bills. So who made the $34.8 trillion by which the stock market as a whole beat cash?

It is December 1999. You can put your savings into one stock and hold it for as long as you like. Three stocks keep coming up:

  • a giant: one of the largest industrial companies in the world, worth about $508 billion as the year ends;
  • the most widely held stock in America;
  • an online store that started out selling books and lost $720 million that year.

Which one do you buy, and could you have known which one to hold? Keep your answer in mind, because this essay returns to it. Before it does, look at what any single pick was up against, in the record of every stock listed in the US over nine decades.

Version 3.4First published Sep 26, 2026Last updated Oct 3, 2026

01 The ninety-year record

One professor followed every US stock over a ninety-year span and compared each with cash.

Hendrik Bessembinder, a finance professor at Arizona State University, followed all 25,967 common stocks listed in the US between 1926 and 2016, each from its first day of trading to its last, or to the end of 2016, with dividends included. He measured every one of them against one-month Treasury bills, which is roughly what cash earns, because cash is the simplest alternative there is: a stock that earned less than Treasury bills over its own life did not pay its owner anything for the risk of holding it.

The figure puts each stock into one of four rows, by how its whole life turned out. Each bar is a share of all the stocks, so a full bar would be every one of them. Start with the three amber rows.

02 Behind cash

The typical stock lost a little, and one stock in eight lost nearly everything.

Add up the three amber rows, and 57.4% of all stocks ended their lives behind Treasury bills. Most of them lost money outright, which is worse than merely trailing cash. The middle outcome was a small loss: line all the stocks up from worst to best, and the one halfway along returned about −2.3% over its entire life, not per year.

The most common single outcome was losing nearly everything, which is the top amber row, about one stock in eight. To see how both can be true, picture each stock’s result as a pin on a long ruler. The near-total losses all stand on the same spot, which makes it the most crowded spot on the ruler, while the other pins are scattered along its length, and the pin halfway along the row sits at a small loss.

03 Two famous losers

Some of the biggest losses came from the best-known companies of their day.

A list like that sounds like a list of small, forgotten companies, but the two example lines under the amber rows belong to two of the best-known names of their time. A dollar put into Lucent Technologies when it listed in 1996 had shrunk to 45 cents when the stock ended ten years later. WorldCom is the stranger case, because it made money and still ended behind: over the two decades of its stock’s record, a dollar in its shares became $1.65, while a dollar in Treasury bills became $3.86. That can happen because cash compounds too, so the bar a stock has to clear rises with every year it stays listed.

Figure 1 · Every US common stock, 1926–2016 · each stock’s whole-life return against cash

Most stocks ended behind cash.

Of all 25,967 stocks, measured against cash (one-month US Treasury bills):

  • 11.8%lost essentially everything3,071 stocks. They lost all, or nearly all, of their value.
  • 38.7%lost money, but not all of itFrom small losses to large ones. The typical (median) stock is here: −2.3% over its whole life, not per year.A large loss: Lucent, 1996–2006. $1 in the stock became $0.45; $1 in cash became $1.47.
  • 6.9%made money, but less than cashWorldCom, 1980–2002: $1 in the stock became $1.65; $1 in cash became $3.86.
  • Together, 57.4% of all stocks ended behind cash.
  • 42.6%beat cash

Each bar is a share of all 25,967 stocks, so a full bar would be every stock. Amber bars ended behind cash; the grey bar beat it. Each stock is compared with cash over its own years on the market.

Why so many fell behind

More than half fell behind cash mainly because so many stocks were removed by their exchange.

How does a company the size of WorldCom end up behind cash? Its own board answered in an investigation report in 2003. WorldCom was a telephone and data company that grew by buying other companies with its own shares, a strategy which, the investigators wrote, “depended on a consistently increasing stock price”. They found that it had kept more than $7 billion of its largest cost out of its published results, and that the purpose was to go on reporting fast growth when the real growth was much lower. In June 2002 WorldCom announced that those results had been wrong, and within a month it filed for bankruptcy protection.

Most stocks that fell behind cash did so in a plainer way, and they belong to a far larger group: the 9,187 stocks, about a third of all of them, that their exchange removed from trading. Were they losers because they were removed, or removed because they had already lost? By the end of its listed life the typical removed stock had lost 92%, so for the typical one the loss was already there when the listing ended. Why each of those companies failed is not in the record.

Nearly all of the removed stocks ended behind cash, and they are more than half of all the stocks that did: of the roughly 14,900 stocks that ended behind cash, about 8,600 were ones their exchange had removed. Bessembinder draws the same conclusion. The finding that most stocks trail Treasury bills, he writes, “is primarily attributable to the stocks that were removed from listing by the stock exchanges”, and he adds that this is of limited use unless one can predict in advance which group a stock will end up in. So the count has its explanation: most stocks fell behind cash mainly because so many of them were removed by their exchange, most of their value already gone. What that does not explain is how the market as a whole came out ahead.

The same market in dollars

Measured in dollars, the same market beat cash by trillions.

So far, holding stocks looks like a losing game. Yet over the same ninety years the stock market as a whole made its owners far richer than cash would have, and the difference lies in what has been counted. Counting stocks gives a tiny listed company exactly the same weight as ExxonMobil, because each of them is a single entry in the list whatever its size.

So Bessembinder also measured every company in dollars. For each one he asked how much richer its shareholders ended up than if the same money had earned Treasury-bill interest, counting the dividends (the cash a company pays its shareholders), the money it paid out by buying back its own shares, and the money shareholders put in when it sold new ones. This measure treats companies very differently from the count, because a company’s size and its years of growth now matter, while a company that failed can subtract no more than the money put into it and the interest that money would have earned in Treasury bills. Add the results up across every company, subtracting what the losers destroyed from what the winners created, and the US stock market earned its owners $34.8 trillion more than cash would have between 1926 and 2016.

Figure 2 · 1926–2016 · By count: 25,967 stocks · By dollars: $34.8T of net wealth above T-bills, in December-2016 dollars

All of the market’s gain above cash came from the top 4.3% of companies.

The top 4.3% are the 1,092 companies drawn here: the ten named, then the 11th to 90th, then the 91st to 1,092nd.

By count: share of all 25,967 stocks — Figure 1 in two rows

  • 57.4%behind cash
  • 42.6%beat cash

By dollars · companies

Bar length is wealth created above cash, on one scale throughout. Green: wealth created above cash; red: wealth destroyed.

That answers the question of who made the money. The whole gain, after every loss has been subtracted, came from a small group: the top 4.3% of companies, 1,092 of them. Half of it came from just 90 companies. (The dollar measure counts companies, of which there are slightly fewer than stocks, because a company that issued more than one class of share is counted once.)

It would be wrong to picture the other 95.7% as companies that earned nothing. Many of them created wealth above cash, about $6.0 trillion in all, and the others destroyed about as much, so the two sums cancel, and the group as a whole ended about level with Treasury bills.

That leaves two pictures of one market. Counted stock by stock, most of them ended behind cash, and added up in dollars, the same stocks ended far ahead of it.

How can both pictures be true?

Why both pictures are true

A loss stops at the money you put in, but a gain can keep compounding for decades.

The first half of the answer was that most stocks trailed cash largely because so many of them were removed by their exchange. The other half, which explains the dollar side, is a plain piece of arithmetic that applies to every stock. A share price can fall only as far as zero. However badly a company fails, its owners cannot lose more than the money they put in: put in a dollar, and losing that dollar is the worst that can happen. On the way up there is no such stop. A company that keeps growing earns each year’s gain on top of all the gains of the years before, so the longer it lasts, the further it pulls away from where it started.

How far can that go? A dollar put into the tobacco company now called Altria at the start of 1926, with the dividends reinvested, had become about $4.4 million by the end of 2025. Against a worst case of one lost dollar, the two sides of the bet are plainly not the same size.

The same imbalance appears when Bessembinder measures whole companies in dollars. The worst result in his study is WorldCom’s $95.5 billion of wealth destroyed, and the best is the just over a trillion dollars that ExxonMobil created, so the best company added more than ten times what the worst one took away.

Bessembinder names two causes for this shape. The first is that stock returns are tilted even over a single month, with a few large gains among many small results. Fewer than half of all the monthly stock returns in his record, 47.8%, beat the Treasury bill of the same month, and yet the average monthly stock return was about three times the bill’s. That combination is possible only if the months that beat cash beat it by more than the other months fell short, which means that a minority of large gains was already lifting the average.

The second cause is compounding, which means that each period’s gain or loss is applied to whatever the period before left behind, and it pushes most results below the average even when nothing is stacked against the stock. In the paper’s own illustration, a stock either rises 20% or falls 20% in each period, with equal chance, so that its average return in any one period is zero. After two periods there are four equally likely paths: two rises give +44%, a rise and a fall in either order give −4%, and two falls give −36%. The average of the four is still zero, but three of the four paths end in a loss, because a rise followed by a fall of the same percentage leaves less than the starting amount. A single path out of four carries the whole average, which means that the average describes almost nobody’s result.

So the losses are many and each of them is limited, while the gains are few and have no limit, which is how most stocks can end behind cash while the market as a whole ends trillions of dollars ahead of it.

Back to December 1999

All three stocks fell hard, and only one of them was back within ten years.

Knowing that a few companies make the market’s fortune does not tell you which ones they are, or what it is like to own one. So return to your three stocks. The online store was Amazon, the giant was General Electric, and the most widely held stock in America was Lucent. Apple was not one of your three, but its own fall came two years before your choice.

Figure 3 · The three stocks offered in December 1999 at the start of this essay: the giant, the most widely held and the online store · Years until each closed above its December 1999 level again

All three fell hard, and only one was back within ten years.

Amazon the online storeStarts at its peak close: $106.69 on Dec 10, 1999
9.9 years back above on Oct 23, 20099.9 years back above, Oct 2009Sep 2001 low: −94% ($5.97)Sep 2001 low: −94% ($5.97)
General Electric the giantStarts at its last close of 1999
25.4 years back above on Jun 2, 202525.4 years back above, Jun 2025Mid-2018 market value: at least $116B, from $508BMid-2018 market value:
at least $116B, from $508B
Lucent the most widely heldStarts Nov 30, 1999
Never back above. Nov 2006: merged into Alcatel; the shares stopped tradingNever back above. Nov 2006: merged
into Alcatel; the shares stopped tradingLate 2002 value: $6.3B, from $233BLate 2002 value: $6.3B, from $233B

The amber line is the years from each row’s starting point, named under the company, until its shares first closed above that level again (dot). Lucent’s line stops at the bar. Short ticks mark a dated value; only Amazon’s is its low. Sources and dates: see Methodology.

If you picked the online store, you picked Amazon, and for the next two years everything a shareholder could see looked like failure. The shares fell 94% from their peak in December 1999 to their low in September 2001, and they did not close above that peak again until October 2009. The company lost more than a billion dollars in 2000 and ended the year with negative equity, which means that on its books it owed more than it owned. Two online retailers in which it held significant stakes, Pets.com and living.com, shut down that year, and in January 2001 it cut about 15% of its workforce. Jeff Bezos opened his next letter to shareholders with the word “Ouch”, and went on: “As of this writing, our shares are down more than 80% from when I wrote you last year.”

The sales figures offered little comfort either. In every quarter of the fall Amazon sold more than it had a year earlier, but the gain shrank from 95% at the start of 2000 to almost nothing by the quarter in which the share price reached its low, and 2001 as a whole was another year of losses. So the shareholder at the bottom owned a business that was still losing money and had almost stopped growing, and nothing in those figures marked it out as a future winner. Growth then returned, and eight years later so did the price. Someone who bought at the listing in 1997 and held until 2025 earned 31.1% a year, which is the reward of the earliest owners and not the return of a buyer at the 1999 peak.

If you picked the giant, you picked General Electric, and its story shows that a winner of the ninety-year record can still lose most of its value late in its life. It ranks fourth among all US companies for wealth created between 1926 and 2016. Its own reports put its market value at $508 billion at the end of 1999 and at about a quarter of that in the middle of 2018. The part of the decline that has a reason on record came late. In 2018 GE disclosed large losses on insurance policies that it no longer sold but still had to pay out on, wrote down the value of its power division, and cut its dividend almost to nothing. The US market regulator, the Securities and Exchange Commission, later wrote that in 2017 and 2018 the share price “fell almost 75% as challenges in its power and insurance businesses were disclosed to the public”. That June GE was replaced in the Dow Jones Industrial Average, the best-known index of large American companies, of which it had been a member without a break since 1907. Its shares did not get back to their level of the end of 1999 until June 2025 (a data provider’s figures; see Methodology).

If you picked the most widely held stock in America, you picked Lucent, which made equipment for telephone and data networks. “Most widely held” means that more investors owned it than any other stock, and it was large as well as popular: at the end of November 1999 the shares held by outside investors were worth $233 billion. Five weeks later, on January 6, 2000, the company warned that its quarter would fall short, and the first of the reasons it gave was that customers had moved faster than it foresaw to its newest optical equipment, which it could not yet make and install quickly enough. That evening, in late trading, the shares fell 28%.

Then the customers themselves cut back. In its annual report for 2001 Lucent wrote that the telephone companies it sold to faced slower growth and “reduced access to capital”, and were “reducing capital expenditures and rethinking their plans to expand their networks”. When they cut back on building, Lucent’s sales fell by about a quarter in a single year, and the loss for that year was $16.2 billion. Over the following year Lucent cut its workforce by more than a third, and by the end of November 2002, three years after the choice, the shares held by outside investors were worth $6.3 billion.

Some of the growth before the fall was later disputed too. In 2004 the Securities and Exchange Commission alleged that Lucent had improperly counted more than a billion dollars of revenue in the year of the warning, and Lucent settled the case without admitting or denying it. At the end of November 2006 what remained of the company became part of the French company Alcatel. By Bessembinder’s count it left its shareholders $80.0 billion poorer than cash, the third-worst result in the study.

Apple was not one of your three choices. It is here because in 1997 it looked finished, and by 2025 it stood first in the whole record. In two years its sales had fallen by more than a third and it had lost nearly two billion dollars. The company’s own annual report gave the reason, which was that customers had “concerns regarding the Company’s strategic direction, financial condition and future prospects, and the viability of the Macintosh platform”, the Macintosh being its computer. In plain words, buyers doubted that the company and its computer would survive. Its debt had been cut to junk, the grade given to borrowers who are thought to be at real risk of not repaying. In August 1997 Microsoft invested $150 million in it, and in September Steve Jobs was named interim chief executive.

The following year showed what the start of a recovery looks like from the inside, and it did not look like growth. Sales fell again, by 16%, and yet Apple reported a profit in every quarter. With sales shrinking, the profit came from the cost side: the large one-off costs of the crisis year, for an acquisition, for restructuring and for ending a licence agreement, were not repeated, its running expenses fell by 32%, and it held far less unsold stock, 6 days’ supply instead of 31.

The products that Apple is known for today came after that. The iPod was introduced in October 2001 and the iPhone in June 2007, and by 2016 the iPhone made up 63% of Apple’s sales. By the end of that year Apple had created more wealth for its shareholders than any company except ExxonMobil, and in the ranking that runs to 2025 it was first.

So the question you were asked in December 1999 has its answer. All three of your stocks fell hard. Amazon was back above its old price within ten years, General Electric needed a quarter of a century, and Lucent’s stock ended in 2006 without returning. Could you have known which one to hold? Nothing shown here would have told you, because in December 1999 the only one of the three that was losing money was the one that came back first. At their lowest points the winners looked like the loser, because Amazon, Apple and Lucent were each reporting large losses and each cutting back. Selling Apple in 1997, or Amazon in 2001, would therefore not have been foolish. Researchers at Morgan Stanley published a study in 2025 of the stocks of more than 6,500 US companies between 1985 and 2024. They found that the typical one fell 85% from its peak at some point, and that more than half never climbed back to that peak. On that record, a stock that had lost nearly all of its price resembled the many that never came back far more than it resembled a future giant.

The winner is not hard to own because it is rare. It is hard to own because, for years at a time, it looks exactly like a loser.

Try it yourself

Holding more stocks raises the odds of owning one of the few, but only slowly.

If you cannot tell the few from the rest in advance, one thing is still in your hands, and that is how many stocks you hold. The figure below lets you try it. Choose a number and draw a random portfolio, in which each pick has a 4.3% chance of being one of the few, because that is the share of all companies that together made the market’s whole gain above cash. Because every pick is a separate small chance, the probability of missing all of them shrinks by the same small factor with each stock you add, which is why the odds climb slowly. With twenty stocks the chance of holding at least one of them is about 58%, and it takes many more before that becomes close to certain.

Figure 4 · A teaching model · The top 4.3% (“the few”) are the 4.3% of companies that together made the market’s whole gain above cash, 1926–2016 · Each pick is an independent random draw

How many stocks before you own one of the few?

58%of the time, over many draws, 20 random stocks contain at least one of the top 4.3%
42%of the time they contain none of them

in the top 4.3% · not

One random draw. Press “Draw again” for another.

The odds are exact for this model: 1 − (1 − 0.043)N.

The model counts only whether one of the few companies is among your picks. It says nothing about returns, because it does not model how much any stock earned or how long it was held, and owning one of the few would have paid only if you had also held it through years like Amazon’s, whose shares took almost ten years to climb back above their 1999 peak.

What the record tells you

Owning one stock and owning the whole market are two different bets.

The record shows that stocks as a group beat cash, so the lesson is about how they are held: a single stock was a very different bet from the market. The market’s gain came from a few companies that this record gives no way of naming in advance, and even those looked like mistakes for years at a time. Holding many stocks is therefore the plain, mechanical way to make owning some of them close to certain, and it asks for no judgment about which ones they will be. Holding a few is a bet of another kind, namely that you can find them early and sit through falls like Amazon’s without selling. In a simulation, Bessembinder tried the most concentrated version of that bet, one stock at a time, drawn at random again every month for the full ninety years, and it ended behind Treasury bills in 73% of the runs.

Putting your savings into one stock in December 1999 was that second bet. Either bet can be reasonable. Holding many is reasonable because it does not depend on foresight. Holding a few is reasonable because a single winner, found early and kept, was worth so much. The ninety years show what someone who holds many stocks takes on and what someone who holds few takes on. They do not reveal which companies will be the exceptional winners next.

After 2016

Nvidia ranked 129th in Bessembinder's count to 2016 and second in his count to 2025, so the list of the few keeps changing.

That record closes at the end of 2016, but Bessembinder has since carried the count forward to 2025, and the updated ranking has a new name near the top. Nvidia, a designer of computer chips, stood 129th in the ranking that ends in 2016 and second in the one that ends in 2025. It, too, had once looked like a mistake, when its shares fell about 90% in nine months of 2002. Over the years between the two rankings its business changed as well: at the start of that period most of its sales came from its gaming business, and by the end the great majority came from data centers, the buildings full of computers that run online services.

So the companies near the top of one edition are not simply the ones near the top of the next, and knowing the last list does not give you the next one. Nor is the group that carries half of the wealth getting any larger. By our count from the author’s data files, it took 90 companies in the edition that ends in 2016 and 46 in the one that ends in 2025, and every edition in between needed fewer than the one before it. Why that number keeps falling, we could not establish from any source.

Figure 5 · Companies that made half of all US stock-market wealth · five editions of the study · 1 dot = 1 company

Each update of the study needs fewer companies to account for half of the wealth.

1926–201690 companies made half

1926–201983 companies made half

1926–202272 companies made half

1926–202452 companies made half

1926–202546 companies made half

Each total is in dollars of its own end year, so the dollar figures are not directly comparable; the company counts are. The totals: $34.8T to 2016, $55.1T to 2022 and about $91T to 2025; the 2019 and 2024 editions have no verified total here. The count for 2016 is the study’s own; the later counts are our count from the author’s data files.

So anyone choosing a stock today is where you were in December 1999. Then the giant was General Electric, the stock that everyone owned was Lucent, and the one that looked like a mistake was Amazon. Today’s three have other names, and this record offers no way to tell, from the inside, which of them the next edition will count among the few.

How we did this

Sources & method.

Every figure about the whole market comes from Hendrik Bessembinder’s peer-reviewed study, which uses the CRSP database, the standard academic record of US stock prices, to follow every US common stock from July 1926 to December 2016,1 and from the data files he has published with it.3 Figures from his later updates carry their own date range. The company stories come from the companies’ own filings with the Securities and Exchange Commission, from the Commission’s releases, from the WorldCom board’s investigation report and from the index provider’s announcement; where a figure comes from the press instead, the text says so. Nothing is estimated unless it says so.

Methodology: data, measures, figures, limits
Source
Bessembinder (2018),1 its working paper,2 and his published company-level data files,3 from which the company ranks and dollar figures are read. Later windows: 1926–20224 and 1926–2025.5 We report published figures; we do not re-run CRSP.
Stocks vs companies

The return figures count stocks (25,967 common stocks): 42.6% beat one-month Treasury bills over their lifetime, 57.4% did not; the median lifetime return is −2.29%; the most common outcome (5% bins) is −100%, with 3,071 stocks (11.8%) losing essentially everything; 9,187 stocks were removed by their exchange (CRSP delisting codes 500–599; mergers and exchanges of shares are a separate group. The study notes that the reason for a delisting is not always reported. Where one is, 1,378 stocks had insufficient capital, 1,071 a price below the acceptable level, 1,004 were late in filing or had not paid fees, and 974 did not meet the exchange’s financial guidelines: 4,427 of the 9,187 between them.), with a median return of −91.95%. The wealth figures count companies (25,332): the top 4.3% (1,092) account for all $34.8 trillion of net wealth; 90 companies (0.36%) account for half (50.15%); the other 95.7% together roughly match Treasury bills.

The most common band of lifetime returns, a loss of essentially 100% when returns are rounded to the nearest 5%, holds 3,071 stocks (11.8%). Of the 9,187 stocks removed by their exchange, 6.8% beat one-month Treasury bills. “Of the roughly 14,900 stocks that ended behind cash, about 8,600 were ones their exchange had removed” is our arithmetic: 57.4% of 25,967 = 14,905, and 9,187 × 93.2% = 8,562. The median stock was listed for seven and a half years (90 months), and 36 stocks were present for all ninety years; listing also ends when a company is taken over, so these two figures describe how long stocks stayed listed, not how many failed. The study states that the result that most stocks trail Treasury bills “is primarily attributable to the stocks that were removed from listing by the stock exchanges”, and adds that this is of limited use unless one can predict in advance which group a stock will end up in. The 25,967 stocks were issued by 25,335 companies, because a company that issued more than one class of share is counted once; for three of them the wealth measure could not be calculated, which leaves the 25,332 companies of the dollar figures. Of the 24,240 companies outside the top 1,092, 9,577 created $6.0 trillion and 14,663 destroyed $6.0 trillion; these counts are from the author’s company file, and the article’s text gives 9,579 and 14,661. ExxonMobil created $1,002.1 billion and Apple $745.7 billion in the 1926–2016 ranking; “more than ten times” is 1,002.1 / 95.5.

Wealth
Dollars by which a company’s shareholders ended up richer (or poorer) than if the same money had earned one-month Treasury-bill interest over its listed life, counting dividends, buybacks and share issues; 1926–2016 figures in December-2016 dollars. The same measure grows in later windows partly because the Treasury-bill alternative keeps compounding, so dollar figures from different windows are not compared. Because the measure is wealth above Treasury bills, a failed company subtracts the money put into it and the bill interest that money would have earned, and the amount grows as the window lengthens (WorldCom: −$95.5B, −$102.0B and −$114.5B in the editions to 2016, 2022 and 2025).
Illustrations and simulations
The example of a stock that rises or falls 20% with equal chance (+44%, −4%, −4%, −36% after two periods; median −4%) is the study’s own teaching example, not measured data. The single-stock strategy is a simulation in the study, not an observation of real investors: for each month from July 1926 to December 2016 one stock is picked at random and the monthly returns are linked, 20,000 times over; over the full ninety years it ended behind Treasury bills in 73% of runs and behind the market in 96%. Measured: of 3,575,216 monthly stock returns from 1926 to 2016, 47.8% exceeded the one-month Treasury bill of the same month; the mean monthly stock return was 1.13% and the mean bill return 0.37%. The share of stocks that beat Treasury bills does not fall steadily as the horizon lengthens (monthly 47.8%, yearly 51.6%, by decade 49.5%, lifetime 42.6%), so the essay makes no such claim. Figure 4 is a model built on the published 4.3%; the 58% for twenty stocks is 1 − (1 − 0.043)20 = 58.5%.
Companies

WorldCom: all details from the report of the Special Investigative Committee of its board (March 31, 2003): more than $9 billion of false or unsupported accounting entries, 1999 to the first quarter of 2002; line costs (the cost of carrying a call or data transmission from its starting point to its ending point, about half of all expenses from 1999 to 2001) understated by more than $7 billion; reported at about 42% of revenue (a target, not a measured figure for any one year) while typically above 50%; the report states the objective as “to hold reported line costs to approximately 42% of revenues…, and to continue reporting double-digit revenue growth when actual growth rates were generally substantially lower”; it places the harder market conditions in 2000 and the glut of network capacity in 2001; MCI acquired in September 1998 for a total price approaching $40 billion; share price $8.17 (early January 1994) to $47.91 (end of September 1999), adjusted for share splits by the report; restatement announced June 25, 2002; Chapter 11 filing July 21, 2002.16 The return of the stock in Bessembinder’s data file (November 1980 to June 2002) was +65.2%, $1 to $1.65, against $3.86 in Treasury bills.3,15 That record of the share begins before the company itself was organized in 1983 and became public through a merger in 1989, so its first years belong to a predecessor listing.

Lucent: $233B, $25.4B and $6.3B are the market value of shares held by non-affiliates on November 30 of 1999, 2001 and 2002, from the cover pages of its annual reports;11,24 “most widely held” is from contemporary press reports;12 the January 6, 2000 warning is from the company’s own filing, which names several factors, the shift to its newest optical products first, and which forecast revenue of $9.8 to $9.9 billion and earnings of 36 to 39 cents a share against 48 cents a year earlier;23 the 28% fall, in late trading on January 6, is from contemporary press reports, and no share price for those days is printed because the reports disagree;12 customer-financing commitments of $8.1B at September 30, 2000; fiscal 2001 (to September 30, 2001) revenue $21.3B against $28.9B (−26%, our calculation) and net loss $16.2B; about 77,000 employees at September 30, 2001 and about 47,000 a year later; lowest price $0.55 in the quarter that ended December 31, 2002, from its annual report;24 the SEC’s 2004 allegation concerns about $1.148 billion of revenue in fiscal 2000;17 the merger with Alcatel was completed on November 30, 2006, when Lucent became a subsidiary of the company renamed Alcatel-Lucent.23 Lifetime return (April 1996 to November 2006) −54.8%, $1 to $0.45, against $1.47 in Treasury bills. The stock returns are from Bessembinder’s data files;3 the Treasury-bill totals are our compounding of the monthly one-month bill rate in the Ken French Data Library.15 Neither WorldCom nor Lucent is claimed to be among the 9,187.

Amazon: 1999 net loss $720M;9 closing prices $106.69 (December 10, 1999) and $5.97 (September 28, 2001), before the 2022 split (−94.4%, our calculation); first close above the 1999 peak on October 23, 2009;10 2000 net loss $1.41B, long-term debt $2.1B and negative shareholders’ equity at the end of 2000; 1,300 jobs, about 15% of the workforce, cut on January 30, 2001. Quarterly net sales against the same quarter a year earlier, our calculation from the annual report for 2001: +95%, +84%, +79% and +44% in 2000; +22%, +16%, +0.2% and +15% in 2001; from one quarter to the next, sales fell through the first three quarters of 2001. The full year 2001 was a net loss of $567M, and the small net profit of its fourth quarter would, by the company’s own annual report, have been a loss without a currency gain, so the essay does not call it a first profit. Net sales were $1.12B in the fourth quarter of 2001 and $5.45B in the third quarter of 2009 (release of October 22, 2009).25,26 31.1% a year is the compound return of the stock from May 1997 to December 2025 in Bessembinder’s data file.5

General Electric: fourth for wealth created, 1926–2016 ($608.1B). Market value $508.3B at the end of 1999 and at least $116.2B on the last business day of the second quarter of 2018, both from its annual reports;19 “about a quarter” is 116.2 / 508.3 = 22.9%, and the 2018 figure is a lower bound. Events of 2018: a $6.2 billion after-tax charge on its run-off insurance business (January 16); a $22 billion write-down of GE Power and the quarterly dividend cut from 12 cents to 1 (October 30); replaced in the Dow Jones Industrial Average on June 26.20,22 The return of the share price: adjusted for share splits and spin-offs, the first close above the price of the end of 1999 came on June 2, 2025, and the first close above the peak of August 28, 2000 on September 16, 2025; these are from a commercial data provider’s price series (Yahoo Finance), not from a primary source, and the adjustment factors are the provider’s. Earlier events not told in the text: on October 1, 2008 it announced that it would sell at least $12 billion of new shares and $3 billion of preferred stock to Berkshire Hathaway; on February 27, 2009 it cut the quarterly dividend from 31 cents to 10; on March 12, 2009 Standard & Poor’s lowered its rating from AAA to AA+. All from GE’s own releases and filings and the S&P Dow Jones Indices release of June 19, 2018.19,20,22 “Fell almost 75%” in 2017 and 2018 is the SEC’s statement; it gives no start or end day.18

Apple: net sales $11.1B, $9.8B and $7.1B in fiscal 1995, 1996 and 1997 (“more than a third” is 7,081 / 11,062 = −36%); net losses of $816M and $1,045M (“nearly two billion” is their sum, $1,861M); Macintosh unit sales −27% in fiscal 1997; debt ratings cut to non-investment grade over two years, to B3 (Moody’s) and B− (Standard & Poor’s) in 1997; Microsoft’s $150M purchase of non-voting preferred stock, August 1997; fiscal 1998 net sales $5.9B (−16%) and net income $309M, a swing of $1,354M of which one-off charges ($667M in fiscal 1997, $7M in fiscal 1998) are the largest single part; operating expenses −32%; gross margin 19% to 25%; inventory 31 days to 6; iPhone 63% of net sales in fiscal 2016; all from Apple’s annual reports.7,8,27 Wealth created $745.7B (1926–2016, second) and $5.02T (1926–2025, first).3,5

Nvidia: “dropped 90 percent” from January 4 to October 8, 2002 is the drawdown study’s figure, on intraday prices (on closing prices our check gives −89.2%), and the same study reports that the PHLX Semiconductor Sector Index fell 65% over the same period, which is context and not a cause;6 129th in 1926–2016 ($54.8B), second in 1926–2025 ($4.58T), two figures in dollars of different end years that are not subtracted from each other.3,5 Revenue for the fiscal year that ended on January 29, 2017: $6.91B, of which gaming $4.06B (nearly 60%) and data center $830M (about 12%); for the fiscal year that ended in January 2026: $215.9B, of which data center $193.7B (about nine-tenths); the shares are our division. Nvidia designs its chips and has them manufactured by others.28

Altria: 16.5% a year, $1 to about $4.4M, January 1926 to December 2025.5

Drawdowns: Mauboussin and Callahan, Drawdowns and Recoveries (Morgan Stanley, May 21, 2025): median maximum drawdown 85% for the stocks of more than 6,500 US companies, 1985–2024; about 54% never returned to their prior peak by the end of 2024. The study measures price changes, not total returns; it uses intraday prices; and it excludes companies that were delisted for cause or went bankrupt.6

Half of the wealth, by edition: 90 companies (to 2016; the study’s own figure), 83 (to 2019), 72 (to 2022), 52 (to 2024) and 46 (to 2025). The counts after 2016 are our calculation from the author’s published data files.3

The figures
Figure 1 puts each of the 25,967 stocks in exactly one of four rows; every bar is a share of all of them, drawn to two decimals: 11.83%, 38.67%, 6.90% and 42.60%, which add up to 100%. Two rows are published: 3,071 stocks (11.8%) are in the most common band of lifetime returns, a loss of essentially 100% when returns are rounded to the nearest 5% (the article does not give the band’s exact edge, so the figure says “all, or nearly all” and states no threshold), and 42.6% beat one-month Treasury bills. The article does not print the other two rows; we calculate them from two figures it does print. It reports that 49.5% of stocks had a positive lifetime return. So 50.5% did not, and taking away the 11.8% that lost essentially everything leaves 38.7% that lost money but not all of it (the few stocks that returned exactly zero are counted here). And taking the 42.6% that beat Treasury bills away from the 49.5% with a positive return leaves the 6.9% that made money but less than Treasury bills. The row with the typical stock runs from small losses to large ones: the median, −2.29%, and Lucent, −54.8%, are both in it. The article publishes no finer breakdown of lifetime returns for this period, so none is drawn. The by-count side of Figure 2 is Figure 1 in two rows. Figure 2, by dollars: every bar is drawn to one linear scale on which ExxonMobil’s $1,002.1B fills the track. The top ten are single companies; General Motors there is the company delisted in 2009, not today’s. Companies 11 to 90 ($11.9T) and 91 to 1,092 ($17.4T, from $74.3B down to $5.3B each) are laid end to end in rank order and folded into lines of $1T on that scale, one company per segment; the top 90 hold 50.15% of the $34.8T. Of the other 24,240 companies, 9,577 created $6.0T and 14,663 destroyed $6.0T, netting to about +$0.6B, counted from Bessembinder’s company file;3 the article’s text gives 9,579 and 14,661. Figure 3 measures time on one ruler: the years from each stock’s starting point until its shares first closed above that level again. The starting points differ. Amazon’s is its peak close ($106.69 on December 10, 1999); General Electric’s is its last close of 1999; Lucent’s is November 30, 1999, the date of the $233B market value in its annual report, because no December 1999 Lucent share price is printed (the press reports disagree). Lucent’s line ends on November 30, 2006, when its shares stopped trading. General Electric’s date of June 2, 2025 comes from a data provider’s share prices adjusted for splits and spin-offs, not from a primary source. The ticks are dated values, and only Amazon’s is its low: General Electric’s market value of at least $116.2B in mid-2018 and Lucent’s $6.3B at the end of November 2002 come from the companies’ annual reports (Lucent: shares held by outside investors). The waits, 9.9, 25.4 and 7.0 years, are our calculation from those dates. Figure 4 is exact for its model; the hand is a real random draw each time. Figure 5: one dot per company.
Color
Green and red appear only on Bessembinder’s dollars of shareholder wealth created or destroyed. Shares of stocks, probabilities, share prices, market values and company losses are shown in neutral ink, amber or violet: a percentage describes; it is not money.
Allegations, not findings of a court
The SEC’s 2004 statement that Lucent improperly recognized about $1.148 billion of revenue in fiscal 2000 is an allegation; Lucent settled without admitting or denying it.17
What we left out, and why
The reason General Electric’s shares fell after their peak in 2000: we found no source we could rely on, so the essay gives none, and it draws no line between the events of 2008–2009 and those of 2018. Whether Amazon’s business had “recovered” by 2001: sales growth had slowed to almost nothing and the year was a loss, so the essay says only that. A causal account of Apple between 1998 and 2016: the filings give dated milestones and an end state, not a chain of causes. The reason for Nvidia’s fall in 2002. Why Nvidia’s rank changed: its revenue shifted from gaming to data centers over the same years, but revenue belongs to the company and wealth to its shareholders, and the figures do not prove the link. Why each update of the study needs fewer companies for half of the wealth: we found no explanation we could source. What became of Lucent’s loans to its customers. The number of Lucent’s shareholders: the sources disagree. Whether WorldCom’s or Lucent’s stock is among the 9,187 removed by their exchange: our sources do not say.
Limits
A descriptive record of US common stocks in one period, not a forecast, a valuation signal or advice. Figures shift with the exact window and with later updates. This is an evergreen essay about closed studies, and it carries no live data.

References

  1. Bessembinder, H. (2018). Do stocks outperform Treasury bills? Journal of Financial Economics, 129(3), 440–457. https://doi.org/10.1016/j.jfineco.2018.06.004
  2. Bessembinder, H. (2017). Do Stocks Outperform Treasury Bills? SSRN Working Paper No. 2900447. papers.ssrn.com
  3. Bessembinder, H. Wealth-creation data files (1926–2016, 1926–2019, 1926–2022, 1926–2024, 1926–2025). W. P. Carey School of Business, Arizona State University. wpcarey.asu.edu
  4. Bessembinder, H. (2023). Shareholder Wealth Enhancement, 1926 to 2022. SSRN Working Paper No. 4448099. papers.ssrn.com
  5. Bessembinder, H. (2026). One Hundred Years in the U.S. Stock Markets. SSRN Working Paper No. 6438198. papers.ssrn.com
  6. Mauboussin, M. J., & Callahan, D. (2025, May 21). Drawdowns and Recoveries. Morgan Stanley Counterpoint Global. morganstanley.com (archived copy: web.archive.org)
  7. Apple Computer, Inc. (1997). Form 10-K for the fiscal year ended September 26, 1997. SEC EDGAR. sec.gov
  8. Microsoft. (1997, August 6). Microsoft and Apple affirm commitment to build next generation software for Macintosh. news.microsoft.com
  9. Amazon.com, Inc. (2000). Form 10-K for the fiscal year ended December 31, 1999. SEC EDGAR.
  10. CNNMoney. (2009, October 23). Amazon shares hit all-time high. Archived copy: web.archive.org
  11. Lucent Technologies Inc. (1999). Form 10-K405 for the fiscal year ended September 30, 1999. SEC EDGAR.
  12. CNNMoney. (2000, January 6 and 7). Lucent warns of 1Q earnings dip; Lucent’s stock crashes. Archived copies: web.archive.org, web.archive.org
  13. CNNMoney. (2018, June 19). GE is getting kicked out of the Dow. Archived copy: web.archive.org
  14. CNNMoney. (2018, September 26). GE’s market value falls below $100 billion. Archived copy: web.archive.org
  15. French, K. R. Data Library: Fama/French factors (monthly one-month Treasury bill rate). Tuck School of Business, Dartmouth College. mba.tuck.dartmouth.edu
  16. Special Investigative Committee of the Board of Directors of WorldCom, Inc. (2003, March 31). Report of Investigation. Form 8-K, Exhibit 99.1. SEC EDGAR. sec.gov
  17. U.S. Securities and Exchange Commission. (2004, May 17). Press Release 2004-67 (Lucent Technologies Inc.). sec.gov
  18. U.S. Securities and Exchange Commission. (2020, December 9). Press Release 2020-312 (General Electric Company). sec.gov
  19. General Electric Company. Annual reports for 1999 (Form 10-K405, accession no. 0000040545-00-000023), 2009 (Form 10-K, 0000040545-10-000010) and 2018 (Form 10-K, 0000040545-19-000014). SEC EDGAR.
  20. General Electric Company. Press releases of October 1, 2008, February 27, 2009 (Form 8-K, Exhibit 99), January 16, 2018 (Form 8-K) and October 30, 2018.
  21. Clough, R. (2018, October 1). Bloomberg News, as published by Financial Advisor magazine (fa-mag.com).
  22. S&P Dow Jones Indices. (2018, June 19). Press release on the Dow Jones Industrial Average: Walgreens Boots Alliance replaces General Electric, effective before the open on June 26, 2018.
  23. Lucent Technologies Inc. Form 8-K filed January 7, 2000 (accession no. 0000950123-00-000151) and Form 8-K filed December 6, 2006 (accession no. 0000898822-06-001443). SEC EDGAR.
  24. Lucent Technologies Inc. Annual reports for fiscal 2001 (Form 10-K405), 2002 (Form 10-K) and 2003 (Form 10-K, accession no. 0001206774-03-000885). SEC EDGAR.
  25. Amazon.com, Inc. Annual reports on Form 10-K for 2000 (accession no. 0001032210-01-500087) and 2001 (accession no. 0001032210-02-000059). SEC EDGAR.
  26. Amazon.com, Inc. Press releases of January 30, 2001 and October 22, 2009 (Form 8-K, Exhibit 99.1); J. P. Bezos, 2000 Letter to Shareholders.
  27. Apple Computer, Inc. / Apple Inc. Annual reports for fiscal 1998 (Form 10-K405, accession no. 0001047469-98-044981), 2002 (0001047469-02-007674), 2007 (0001047469-07-009340) and 2016 (0001628280-16-020309). SEC EDGAR.
  28. NVIDIA Corporation. Form 10-K for the fiscal year ended January 29, 2017 (accession no. 0001045810-17-000027); press release of February 25, 2026 on results for fiscal 2026 (Form 8-K, Exhibit 99.1). SEC EDGAR.

How to cite this essay

Tradiens Research. (2026). The 4% that carry everyone (Version 3.4). Tradiens Data Essays. https://tradiens.com/essays/bessembinder-4-percent

@misc{tradiens2026bessembinder4pct,
  author       = {{Tradiens Research}},
  title        = {The 4\% That Carry Everyone},
  year         = {2026},
  version      = {3.4},
  howpublished = {Tradiens Data Essays},
  note         = {Based on Bessembinder (2018), J. Financial Economics 129(3), 440--457},
  url          = {https://tradiens.com/essays/bessembinder-4-percent}
}

Tradiens

Measure your own results the way this study measures the market.

Tradiens is a trading journal that counts every trade, losers included, and shows the real spread of your results, so you can tell a repeatable edge from one lucky win. It does not tell you what to buy.

See how Tradiens works →

Every figure is sourced; see Methodology and References above. The draw in Figure 4 is a teaching model built on the published 4.3%.

Tradiens Research · Data essay · v3.4 Trade like you know.