Does a Crash Prove a Bubble? Testing Jeff Yass's Expected-Value Argument
Jeff Yass says a stock crash is just uncertainty resolving. Give his $100 example a 10-year wait and an 8% hurdle, and his 15% long shot needs 21.6% odds.
On November 6, 2026, the Wall Street Journal published an opinion piece by Jeff Yass, a co-founder of Susquehanna International Group, titled “There’s No Such Thing as a Stock-Market Bubble.” His argument is that a stock price is the market’s best guess at every possible future, each weighted by its probability. When prices fall, “it isn’t a delusion correcting. It’s uncertainty resolving.”
His example is a 1999 internet stock trading at $100. If the company becomes the next giant, the stock goes to $1,000. If it becomes the next Pets.com, it goes to zero. Give the good outcome a 15% chance and the expected value is $150, so at $100, he writes, the stock “isn’t expensive. Buying it is a sensible bet,” even though the most likely result is a total loss. When it goes to zero, Yass argues, nobody was irrational: the buyers took a 15% shot and the 85% came in.
Yass is right that a crash, taken alone, says nothing about whether the buyers were rational, and the long-run evidence supports his doubts about bubble callers. His arithmetic leaves out two things: how long the payoff takes to arrive, and what investors require for bearing the risk while they wait. Once those are added, his own example is a sensible bet only if the payoff comes quickly. And the logic that keeps a crash from proving a bubble also keeps a recovery from disproving one.
The short version
- A 15% chance at $1,000 is worth $150 only if the $1,000 arrives today. The op-ed does not say when it arrives. If it takes ten years, $150 discounted at 5.07%, the 10-year Treasury yield on November 18, 2026, is worth about $91, less than the $100 price, before any reward for risk. At an 8% required return and a ten-year wait, the price needs a 21.6% chance of the good outcome.
- For an index fund investor, the op-ed and the research reach the same practical answer. For the market as a whole, price booms have predicted higher volatility more reliably than they have predicted crashes, so a bubble call is a poor reason to sell. Single industries are different: after big run-ups, crashes became much more likely.
- A recovery does not settle the debate. In our calculation from Ken French’s data, $1 put into the tech sector at the March 2000 close was worth $7.56 by September 2026. The same dollar in the total US market was worth $8.49.
- Yass’s sportsbook analogy suggests a test he does not run: add up the odds that rival companies’ prices require and see whether the sector needs more winners than its market can hold.
What the argument gets right
The gap between the expected outcome and the most likely outcome is the strongest part of the op-ed. A bet can be sound and still lose most of the time. Summitward’s options guide works through a nearly identical bet, $100 that becomes $1,100 or nothing, and shows that the median result is a total loss even when the expected value is attractive.
Judging a decision by how it turned out is a common mistake, and Yass illustrates it with a recent game. The New York Knicks came back from 29 points down to beat the San Antonio Spurs in Game 4 of the 2026 NBA Finals, a Finals record. FanDuel’s live odds on the Knicks reached +3000 during the game, according to Covers, an implied chance of 3.2%. A 3% shot that comes in does not show that the 3% price was wrong.
Stock returns are skewed in the way his example assumes. Hendrik Bessembinder’s 2026 update covers 29,754 US common stocks from 1926 through 2025. The median stock’s lifetime buy-and-hold return was −6.9%. Of 29,081 firms, 59% returned less than one-month Treasury bills over the months they were listed, and 46 firms account for half of the $91 trillion of net wealth the market created. Most single bets lose, and a few large winners pay for the rest. The T-bills guide covers that research in full.
Finance research also gives his point about uncertainty a formal version. Ľuboš Pástor and Pietro Veronesi showed that “the fundamental value of a firm increases with uncertainty about average future profitability,” and that the uncertainty needed to match Nasdaq valuations at their 2000 peak was “high but plausible.” In their later model of technological revolutions, the bubble-like pattern in prices is “observable ex post but unpredictable ex ante.”
The long-run record supports his skepticism about bubble calls. William Goetzmann, Otto Manninen and James Tyler counted US stock market booms from 1792 to 2024, using a price index without dividends. Of 394 overlapping three-year windows in which the market rose at least 50%, 144 were followed within five years by another 50% gain and 50 by a full reversal. Their summary: “Booms do not reliably predict crashes, but they do predict higher subsequent volatility, increasing the likelihood of both large gains and large losses.” Industries tell a different story, covered in the sportsbook section below.
Add time and risk to the $100 example
The op-ed compares $150 of expected value with a $100 price without saying when the $1,000 arrives. It arrives years later if at all, and a dollar in ten years is worth less than a dollar now. Discount the expected payoff at an annual required return over years and you get what the stock is worth today, :
Here is the chance of the good outcome and and are the two payoffs. Setting equal to the market price and solving for the chance gives the break-even probability, the odds the price needs:
Start with the safest rate available. On November 18, 2026, the 10-year Treasury par yield was 5.07%, according to the Treasury’s daily yield curve. Assume the payoff takes ten years. Discounted at that rate, $150 is worth about $91 today, so at 15% odds the stock is overpriced at $100 even for a buyer who wants no reward for risk. At 5.07%, the 15% shot covers its price only if the payoff arrives within about 8.2 years. Stock investors also want a reward for risk, and Summitward’s equity hurdle guide estimates its current size. At an 8% required return, $150 in ten years is worth $69.48, and the price needs a 21.6% chance of the good outcome.
| Years until the outcome | 5% | 8% | 12% |
|---|---|---|---|
| 3 | 11.6% | 12.6% | 14.0% |
| 5 | 12.8% | 14.7% | 17.6% |
| 10 | 16.3% | 21.6% | 31.1% |
| 20 | 26.5% | 46.6% | 96.5% |
The wait matters as much as the odds. At 8%, a 15% chance at $1,000 is worth $100 only if the payoff arrives within about 5.3 years. If the good outcome would take twenty years to show up, the same price asks for odds close to a coin flip.
A single required return is a shortcut. In asset-pricing theory, a dollar that arrives when investors are struggling is worth more than a dollar that arrives when they are flush, so the probabilities embedded in market prices mix how likely an outcome is with how much people care about it. A Federal Reserve Bank of Minneapolis paper on market-based probabilities makes the point with insurance: “a household may be willing to pay a lot for insurance against the possibility of job loss or a flood, even if the household sees the outcome as highly unlikely.”
Try your own long shot
Enter a price, two payoffs, your own odds, the wait and the return you require. The calculator reports what the stock is worth at your odds, the chance the price needs, and how many winners a group of identically priced rivals would need between them. Real valuations have many scenarios and interim cash flows; the stock valuation guide has a fuller model that solves for the growth a price implies.
Whose probabilities are in the price?
Yass argues that hype is already in the price, because the speculators buying and selling know about it and factor it in with their own money at risk. Part of the market fits that description. David Thesmar and Emil Verner digitized Value Line’s expected-return forecasts back to 1956 and found that they rose when the earnings-to-price ratio was high, fell after strong returns, and predicted future returns. Investor surveys, from households to corporate finance chiefs, behave differently. Robin Greenwood and Andrei Shleifer found that those expectations were “strongly negatively correlated with model-based expected returns”: they rose after good years, when later returns tended to be lower.
Prices come from trades between both groups, and the sophisticated side does not always lean against the crowd. Markus Brunnermeier and Stefan Nagel found that hedge funds “did not exert a correcting force on stock prices during the technology bubble”; they rode technology stocks up and cut their positions before prices fell. Andrei Shleifer and Robert Vishny explained one reason: professional arbitrageurs invest other people’s money and can be forced to sell when a mispricing widens before it closes. Hype is in the price in the sense that every trade at that price included it. Whether the resulting price reflects accurate probabilities is the question bubble arguments are about. The guide to who moves the stock market covers how fund flows push prices around.
Option prices include an insurance premium
Yass calls options insurance policies and reads expensive options as the market announcing in public that a crash is possible. Insurance prices include more than the odds: they include how much buyers dislike the outcome. Joshua Coval and Tyler Shumway found that S&P 500 put options earned average returns below the risk-free rate, the pattern you would expect from insurance, and Peter Carr and Liuren Wu found that the variance risk premium on the S&P 500 is strongly negative: option prices have implied more volatility than later showed up. Expensive options therefore mix a forecast with a fee, and a high price for protection does not by itself tell you how likely a crash is.
That bears on the op-ed’s closing rule: if you think AI is a bubble, buy puts, and if you won’t, don’t call anyone irrational. A put pays only if the fall arrives before expiry and is large enough to cover the premium. From June 1986 to December 2018, Cboe’s index that holds the S&P 500 and buys a put 5% below the market each month rose 708%, while its index that sells puts rose 1,835%, so over that stretch the premium flowed from buyers of protection to sellers, per a Cboe summary of research by Oleg Bondarenko. Roni Israelov found that in typical use “put options are quite ineffective at reducing drawdowns” compared with simply holding less stock. An investor can be right that prices are too high and still lose money on puts, so declining to buy them says little about what that investor believes. The tail-risk hedging guide compares puts with owning less stock.
Neither a crash nor a recovery settles it
Eugene Fama, who doubts that bubbles exist, has long pointed out that every test of whether prices are right is also a test of a model of what returns should be, so a surprising result can be blamed on the market or on the model (Fama 1991). In his Nobel lecture he defined a bubble as an “irrational strong price increase that implies a predictable strong decline,” as Greenwood, Shleifer and Yang You quote him. Yass applies that skepticism to crashes. It applies equally to recoveries: a stock that falls 90% and later comes back has not shown that its old price was fair.
Dot-com history has cases that fit either story. Cisco closed at $80.06 on March 27, 2000, the day it briefly became the most valuable public company in the world, and its price did not close higher until December 10, 2025, according to Bloomberg. Priceline’s highest close in 1999 was $162.375 and its lowest in late 2002 was $1.10, a 99.3% fall, according to its annual reports. The company survived and renamed itself Booking Holdings in 2018. A believer in bubbles reads Cisco as a sound business bought at a bubble price; a believer in Yass’s view reads it as a fair bet on a bigger outcome that did not come in. Both readings fit the same price history, so the price history cannot choose between them.
The op-ed makes one claim that data can test: that investors who held the whole internet sector long enough came out ahead. No clean index of 1999 internet stocks exists, so we used the closest long public series, Ken French’s HiTec industry portfolio. It holds business equipment, software and telecom stocks; most e-commerce firms sit in a different portfolio. Returns include dividends and exclude fees and taxes. For comparison, the Nasdaq Composite closed at 5,048.62 on March 10, 2000, fell 77.9% to 1,114.11 by October 9, 2002, and first closed higher on April 23, 2015 (FRED, price only).
From the March 2000 close, the tech sector fell 77.9% by September 2002. It got back above its starting value for good in July 2016, moved ahead of T-bills for good in July 2017, and recovered for good after inflation in January 2019. By September 2026, $1 had grown to $7.56, a 7.9% annual return. The total US market turned the same dollar into $8.49. A buyer at the peak came out ahead of cash after waiting 17 years and was still behind a total market index fund more than 26 years later.
| Bought at the close of | Tech sector | Total market | Ahead of the market for good from |
|---|---|---|---|
| December 1998 | $14.34 | $11.06 | May 2023 |
| December 1999 | $8.53 | $8.83 | Still behind |
| March 2000 | $7.56 | $8.49 | Still behind |
| December 2000 | $13.80 | $9.99 | March 2023 |
The answer depends on the entry date. Buyers before the 1999 run-up and buyers after the first leg of the crash are ahead of the market today, but both moved ahead for the last time in 2023, 22 to 24 years after they bought. Peak buyers are still waiting. Weighted equally, which gives the many small companies of that era more say, the sector turned $1 at the March 2000 close into $5.81, 32% less than the market.
A sportsbook test for a whole sector
Yass uses sportsbooks to argue that a market price beats any single bettor. Sportsbooks also show how to check whether a set of prices hangs together. A book that offers both teams at −110 implies a 52.4% chance for each, 104.8% in total. The extra 4.8 points are the overround that pays the bookmaker, and nobody reads them as real probability.
The same addition works on stocks. Suppose seven companies are racing for one winner-take-most market and each trades at $100 with the example’s payoffs. At an 8% required return, each price needs a 21.6% chance of winning, so together the seven prices need about 1.5 winners. If the market has room for one, the prices cannot all be right, even though each looks defensible on its own.
| Rivals priced this way | Implied winners at 5% | Implied winners at 8% |
|---|---|---|
| 3 | 0.49 | 0.65 |
| 5 | 0.81 | 1.08 |
| 7 | 1.14 | 1.51 |
| 10 | 1.63 | 2.16 |
Economists call the underlying error competition neglect. Greenwood and Samuel Hanson found it in dry bulk shipping from 1976 to 2011: high earnings brought more ship orders and forecast low returns. Their model fits only if firms both over-extrapolate demand and “partially neglect the endogenous investment responses of their competitors.” Each shipowner’s plan made sense alone; the fleet that all of them ordered together did not.
The test has limits. Real races can have several winners, payoffs differ by company, and market values do not come labeled with probabilities. Its use is as a consistency check: when you can name the rivals for one prize, ask whether their prices together assume more success than the prize can supply. Industry history adds a warning. Greenwood, Shleifer and You found that after an industry beat the market by 100% over two years, the chance of a 40% crash within two years rose to 53%, against 14% for all industries, even though average returns over the following two years were close to zero. Run-ups were more likely to end in a crash when they came with rising volatility, more stock issuance, price acceleration and outsized gains for newer firms. The AI bubble guide covers their crash odds in detail.
What to do with this
- Index fund investors: the op-ed and the research point to the same action, which is none. For the whole market, booms have forecast volatility better than crashes, so the useful response to bubble talk is to confirm that your stock allocation could survive a deep drawdown and to keep rebalancing on schedule.
- Holders of one large stock or a pile of RSUs: a single long shot’s most likely outcome is a loss, even when its expected value is high. An index owns every long shot, including the few that pay for the rest. The concentration risk and position sizing guides cover how much of one bet to hold.
- Stock pickers: before buying a story stock, write down the payoffs and the wait, choose the return you require, and compute the chance the price needs. If it is higher than the chance you believe, the stock is expensive for you even if the story comes true.
- Anyone considering puts: holding less stock has been a cheaper way to reduce crash exposure than buying protection month after month. The tail-risk hedging guide compares the two.
Frequently asked questions
Are stock prices probability-weighted averages of possible outcomes?
Roughly, with two adjustments. Each outcome has to be discounted for the time until it arrives, and outcomes are weighted by how much investors care about them as well as by how likely they are. That is why a market price can be read as a risk-adjusted bet but not as a direct forecast of probabilities.
Do option prices tell you the probability of a crash?
They tell you the price of crash protection. Buyers pay extra to insure against bad outcomes, so S&P 500 puts have earned less than the risk-free rate on average, and option-implied volatility has tended to exceed the volatility that followed.
Is AI a bubble?
For the market as a whole, nobody has shown a reliable way to tell in advance, which is the point of both the Goetzmann, Manninen and Tyler record and the Pástor and Veronesi models. For a single industry, sharp run-ups with rising volatility and issuance have raised the odds of a crash. What you can check is what current prices require: the payoffs, how long they take, and how many companies can win. The AI bubble guide applies William Bernstein’s checklist, and the value ETF guide shows how much AI exposure ordinary funds already hold.
Should I buy puts if I think the market is overvalued?
Only if you accept that a put needs the fall to arrive before it expires and to be larger than the premium. Puts have lost money on average, as insurance tends to, and holding less stock has reduced drawdowns more cheaply in the research cited above.
Did investors who held dot-com stocks eventually come out ahead?
In dollars and against cash, yes, after a long wait. Against the total market, it depended on when they bought. In our calculation, $1 put into the tech sector at the March 2000 close was worth $7.56 in September 2026, against $8.49 for the total market and $1.64 for T-bills.
Does tax change the math?
A little. In a taxable account, a winning long shot held more than a year is taxed at long-term capital gains rates, which shrinks the good outcome. A losing one can offset other gains and, per IRS Topic 409, up to $3,000 a year of other income ($1,500 if married filing separately), with the rest carried forward. Both effects narrow the gap between the outcomes without changing how the break-even probability works.
Key takeaways
- A 90% loss does not prove the buyers were irrational, and a recovery does not prove the old price was fair. Price history alone cannot settle whether a bubble existed.
- Expected value needs discounting. If the op-ed’s $1,000 takes ten years to arrive, its $150 expected value is worth about $91 at the 5.07% 10-year Treasury yield (November 18, 2026), below the $100 price; at an 8% required return the stock needs a 21.6% chance of the good outcome.
- Option prices include an insurance premium: implied volatility has run above the volatility that followed, and S&P 500 puts have earned less than the risk-free rate. Buying puts tests timing and cost more than belief.
- From the March 2000 close, the tech sector took until July 2016 to get back to even for good and still trailed the total market in September 2026, in our calculation from Ken French’s data.
- When several rivals chase one prize, add up the odds their prices need. A total well above the number of possible winners is a concrete sign that the sector is priced for more success than it can have.
Related guides
- A 1,000% Upside and a 100% Downside Does Not Make It a Good Bet for expected value against the median outcome.
- Are We in an AI Bubble? William Bernstein’s Four Signs and Why Investors Should Stay Humble for the bubble-detection research.
- Cliff Asness’s 10 Investing Pet Peeves, Revisited for how he restricts the word “bubble.”
- Most Stocks Lose to T-Bills. The Market Still Wins. for the skew in individual stock returns.
- Tail-Risk Hedging: Puts, CAOS, HIDE, and the Managed-Futures Alternative for the cost of crash protection.
- Growth Stocks Do Not Mean Higher Expected Returns for why the price paid sets the expected return.
Sources
- Yass, J., “There’s No Such Thing as a Stock-Market Bubble,” Wall Street Journal, November 6, 2026. wsj.com
- Author’s calculations: dot-com sector returns and the break-even tables, with code and data, in summitward-research/does-a-crash-prove-a-bubble. See the method note below.
- Bessembinder, H., “One Hundred Years in the U.S. Stock Markets,” working paper, draft of September 27, 2026. SSRN 6438198. Firm-level data: W. P. Carey School of Business.
- Goetzmann, W., Manninen, O., and Tyler, J., “Bubbles, Booms and Crashes in the US Stock Market 1792-2024,” NBER Working Paper 34903, February 2026. nber.org
- Greenwood, R., Shleifer, A., and You, Y., “Bubbles for Fama,” Journal of Financial Economics 131(1), 2019; NBER Working Paper 23191. nber.org
- Pástor, Ľ., and Veronesi, P., “Was There a Nasdaq Bubble in the Late 1990s?” Journal of Financial Economics 81(1), 2006 (NBER w10581), and “Technological Revolutions and Stock Prices,” American Economic Review 99(4), 2009 (NBER w11876).
- Fama, E., “Efficient Capital Markets: II,” Journal of Finance 46(5), 1991 (doi), and “Two Pillars of Asset Pricing,” American Economic Review 104(6), 2014 (doi).
- Thesmar, D., and Verner, E., “Beliefs and Stock Market Fluctuations: New Evidence from the Past Seven Decades,” NBER Working Paper 35286, May 2026. nber.org
- Greenwood, R., and Shleifer, A., “Expectations of Returns and Expected Returns,” Review of Financial Studies 27(3), 2014. nber.org
- Brunnermeier, M., and Nagel, S., “Hedge Funds and the Technology Bubble,” Journal of Finance 59(5), 2004. gsb.stanford.edu
- Shleifer, A., and Vishny, R., “The Limits of Arbitrage,” Journal of Finance 52(1), 1997. nber.org
- Greenwood, R., and Hanson, S., “Waves in Ship Prices and Investment,” Quarterly Journal of Economics 130(1), 2015. nber.org
- Coval, J., and Shumway, T., “Expected Option Returns,” Journal of Finance 56(3), 2001. doi
- Carr, P., and Wu, L., “Variance Risk Premiums,” Review of Financial Studies 22(3), 2009. doi
- Israelov, R., “Pathetic Protection: The Elusive Benefits of Protective Puts,” Journal of Alternative Investments, Winter 2019. aqr.com
- Cboe, “White Paper Shows Volatility Risk Premium Facilitated Higher Risk-Adjusted Returns for PUT Index,” May 29, 2019. cboe.com
- Federal Reserve Bank of Minneapolis, “Market-Based Probabilities: A Tool for Policymakers,” January 7, 2015, section 2.1. minneapolisfed.org
- Market history: Cisco record close (Bloomberg, December 10, 2025); priceline.com 10-K405 for 1999 and 10-K for 2002, Item 5 closing price ranges (1999, 2002); Nasdaq Composite closes (FRED NASDAQCOM); 10-year Treasury par yield (U.S. Treasury); Knicks comeback (NBA.com, Covers).
Method
Sector returns: monthly value- and equal-weighted returns for the HiTec portfolio in Ken French’s 5 industry portfolios (business equipment, telephone and television transmission; file built from the 202609 CRSP database). Market: Fama-French Mkt-RF plus RF; T-bills: RF. Real returns use BLS CPI-U, not seasonally adjusted (CUUR0000SA0); October 2025, which BLS did not publish because of the 2025 lapse in appropriations, is filled with the geometric mean of September and November. Entry dates were chosen before running the numbers: the close of December 1998, December 1999, March 2000 (the HiTec value-weighted month-end peak) and December 2000. “For good” dates are the month after the last month-end below the comparison. Break-even tables compound annually; the 5.07% Treasury figure is a par yield used as an annual rate. No fees, trading costs or taxes.
More in Investing & Portfolio
Browse all investing & portfolio guidesGet new guides by email
Evidence-based, no jargon. At most two emails a month. Unsubscribe any time.
Try it in Summitward
See portfolio factor analysis in action with your own financial data. Free to start, no credit card required.