Every bull market eventually conjures the same anxiety. Prices run. Valuations stretch. The word "bubble" enters the conversation. And somewhere in the chorus, a strategist reminds clients that trees don't grow to the sky.
William Goetzmann, Edwin J. Beinecke Professor of Finance at the Yale School of Management 1, has a problem with that narrative — or at least with how casually it gets applied. In a new NBER working paper co-authored with Otto Manninen and James Tyler, Goetzmann2 and his team examine more than two centuries of U.S. stock market data and arrive at a finding that should reframe how advisors and clients think about booms, crashes, and the rare events that qualify as true bubbles.
Their conclusion: bubbles are extremely rare. Booms, by contrast, are far more likely to keep booming than to reverse.
The Data Behind the Claim
The study's analytical foundation is unusually deep. Goetzmann draws on a monthly price index of 100 large-cap U.S. stocks dating to 1792, supplemented by two industry-level datasets: the Fama-French 49-industry portfolios covering 1926 to 2024, and the Cowles Commission industry data spanning 1871 to 1938. The Cowles data — assembled in the 1930s by Alfred Cowles III in collaboration with Yale economist Irving Fisher and requiring 53 scholars and 1.5 million worksheet entries — had largely been forgotten until it was handed to Goetzmann by Robert Shiller, who had used it in the research cited when he won the Nobel Prize.
"Financial history helps you estimate and plan," Goetzmann says. "Five years doesn't help you with the trend. If you've got 150 years or more, that's a good foundation for understanding where you may end up."
The team defines a bubble empirically: a large price boom followed by a crash that fully reverses the prior gain. No valuation model required. No debate about fundamentals. Price dynamics only.
Booms Don't Predict Crashes
The core finding cuts against intuition. Looking at 36-month boom windows, the data shows that of 394 episodes in which the market gained at least 50% over three years, 144 were followed by another 50% gain within five years. Only 50 fully reversed. At every threshold and nearly every horizon, green bars outnumber red bars — further booms outnumber full reversals.
The pattern holds for 12-month windows too. A single-year gain of 25% or more occurred 315 times in the sample; 161 of those were followed by another 25% gain within five years, against 37 full reversals.
True bubbles — defined as a boom followed by a complete reversal — are vanishingly rare in aggregate. At the 50% three-year threshold, full reversals occurred fewer than 2% of the time across roughly 2,700 overlapping observations. At the 100% threshold, there were only 10 reversals in 232 years of market history.
"The value of our analysis is that we have a complete, clean, independent, out-of-sample dataset," Goetzmann says.
Crashes Are Even More Likely to Recover
If the boom data challenges one popular assumption, the crash data challenges another: that a big decline is the beginning of something worse.
After a 20% or greater three-year decline — 154 episodes in the sample — 95 experienced a full recovery within five years. Only 3 suffered a further 20% loss. After a 30% decline, 27 recovered and none declined further by an equivalent amount. Not a single episode of a 50% three-year decline was followed by an additional 50% decline at any horizon.
Even rapid 12-month crashes, the events most frightening to investors, tell the same story over time. At the 30% threshold, near-term risk of further loss is real — within one year, further declines slightly outnumber recoveries. But by five years, the ratio flips dramatically: 21 recoveries against 6 further declines. The dominant long-run outcome after a crash is recovery.
The Volatility Explanation
There is a nuance the paper takes care to surface. While booms do not on average predict crashes, they do predict higher subsequent volatility — and that is the mechanism through which crashes become more probable post-boom. Conditioning on a dramatic run-up effectively selects for high-volatility industries or high-volatility regimes. In those environments, both large gains and large losses become more likely. The distribution widens in both directions.
This resolves what the authors call an "apparent contradiction" — the finding from Greenwood, Shleifer, and You (2019) that booms don't predict crashes on average yet are associated with a higher subsequent crash probability. Post-boom, the tails of the return distribution expand. More crashes happen. But so do more booms.
At the industry level, the Cowles data corroborates the Fama-French findings almost exactly — a meaningful out-of-sample validation. Industries that boom average roughly 300% growth over two years. Conditional on that gain, the subsequent average price path is flat, tracking the market. Booming industries don't crash; they exhaust themselves and revert toward market returns, with elevated volatility on both sides.
What Advisors Should Take Away
Several implications flow directly from this research for client conversations and portfolio positioning.
Booms are not automatic sell signals. The popular intuition that a large run-up reliably predicts an impending crash is not supported by 232 years of U.S. data. At nearly every threshold and horizon examined, further gains outnumber full reversals following a boom.
Crashes are historically poor arguments for selling. An investor experiencing an extended market decline and considering whether to exit would find little historical support for doing so. Full recoveries dominate further declines at the five-year horizon across every crash threshold in the study.
True bubbles require two rare events in sequence. A boom followed by a full reversal is a low-probability outcome precisely because it demands two independent tail events. The less correlated those events are, the rarer the combination.
Volatility, not direction, is the real risk signal post-boom. After a dramatic run-up, the appropriate concern is not an inevitable crash but a wider distribution of outcomes — which has risk management implications for position sizing and rebalancing, not necessarily for exit.
Survivorship bias is worth acknowledging. Goetzmann flags that the U.S. market is an unusually successful historical case. Inferences might differ materially using data from pre-revolutionary Russia or Eastern European markets that experienced catastrophic disruption. For most clients, this reinforces rather than undermines the case for diversification.
The AI-driven concentration in U.S. equity markets has renewed bubble talk across the industry. Goetzmann's research doesn't dismiss those concerns — volatility risk is real and documented. But it does insist on precision. A boom is not a bubble. And history, over its longest available arc, argues that markets are more likely to recover than they are to collapse — and more likely to keep rising than to fully reverse.
That is not a reason for complacency. It is a reason for calibration.
Footnotes:
1 "Over the (Very) Long Run, Stock Bubbles Are Rare." Yale Insights, 15 June 2026.
2 “Bubbles, Booms and Crashes in the US Stock Market 1792–2024" by William N. Goetzmann, Otto Manninen, and James Tyler is available as NBER Working Paper No. 34903 (February 2026).
Copyright © AdvisorAnalyst.com
