Fewer bubble warnings mean greater danger
- Robin Powell
- 48 minutes ago
- 10 min read
When everyone's worried about a bubble, the danger is usually overstated. When almost nobody is, that's when prices collapse. New research analysing over 5,000 stock market crashes reveals why widespread bubble warnings are paradoxically reassuring — and what this means for AI stocks.
Google searches for “AI bubble” went vertical in 2025, surging 950% year‑on‑year in the U.S. and hitting the maximum reading on Google’s 0–100 interest index in early November. By mid‑January 2026, that spike has largely evaporated, with search interest down to a small fraction of its peak and global volumes also well below their late‑summer highs.
Here's the uncomfortable implication: November was probably the safer moment. All that anxiety, all those searches, all those warnings — they represented widespread scepticism. And widespread scepticism is the opposite of what precedes a crash.
Now? Palantir still trades at roughly 450 times earnings. Nvidia's market cap remains above $4 trillion. Michael Burry — the investor who called the 2008 financial crisis — has taken a bearish bet against Palantir. The valuations haven't budged. But the conversation has moved on. People have stopped asking the question.
That's when the danger rises.
New research reveals a troubling paradox: the fewer bubble warnings there are, the greater the risk. When prices actually peak, almost nobody recognises it.
Harry Markowitz won the Nobel Prize for proving that diversification is the only free lunch in investing. His favourite book on the subject? Extraordinary Popular Delusions and the Madness of Crowds, Charles Mackay's 1841 chronicle of the South Sea Bubble and other speculative manias. When he told me this, I asked why a mathematician who'd transformed portfolio theory would choose a 200-year-old book about human folly. His answer: human nature doesn't change.
He was right. The same behaviours Mackay documented — the extrapolation, the collective optimism, the certainty that this time is different — show up in every bubble. What's changed is our ability to study them systematically.
Robin Greenwood of Harvard Business School and Christian Stolborg of Copenhagen Business School have done exactly that. They've analysed 5,306 boom-bust episodes spanning 43 years. Their findings challenge one of investing's most persistent assumptions: that bubbles become obvious before they burst.
They don't.
What the research actually found
A boom-bust stock meets four criteria: it doubled in price over the previous year, hit extreme valuations (price-to-sales or price-to-book above 5), posted a positive return in the current month, and subsequently crashed by at least 50% within 24 months.
That last criterion matters. By requiring the crash to actually happen, Greenwood and Stolborg aren't speculating about which stocks might be bubbles. They're studying stocks that definitively were.
The sample captures household names — Yahoo in 1999, GameStop in 2021, AMC during the meme-stock frenzy — alongside thousands of companies most investors have never heard of. The forgotten manias that never made the history books.
One crashed stock is an anecdote. Two is a coincidence. More than 5,000 is a pattern.
The researchers tracked what analysts, short-sellers, and journalists were saying as these stocks approached their peaks. What they found overturns a comfortable assumption: that smart money sees what's coming.
It doesn't.
The problem with bubble warnings
Professional analysts don't spot bubbles. At the very moment prices peak, they're forecasting exceptional returns.
The numbers are stark. For the average boom-bust stock, analysts predict a one-year return of 52% at the peak. The market average is 35%. These aren't cautious forecasts hedged with caveats. They're aggressive bets that the run-up will continue.
How many analysts predict negative returns at the peak? Seven percent.
93% of forecasts say the stock is still going up. At the exact moment — visible only in hindsight — when it's about to collapse.
Long-term earnings growth forecasts tell the same story. Analysts expect these stocks to grow earnings at more than 30% annually. The market average is below 15%. They're wildly optimistic precisely when they should be most sceptical.
You might assume this reflects disagreement — bulls versus bears, with the bulls winning out. It doesn't. While analyst disagreement is elevated for these stocks, the entire distribution shifts upward during the run-up. There are optimists and super-optimists. What there aren't, for the most part, are pessimists.
As Greenwood and Stolborg put it: the dispersion reflects "variation in the degree of optimism rather than the presence of genuine pessimists."
This isn't a retail investor problem. Brunnermeier and Nagel (2004) documented that hedge funds — supposedly the sharpest minds in finance — didn't bet against the dot-com bubble. They rode it. They bought the same overvalued tech stocks everyone else was buying, hoping to get out before the crash.
Most didn't.
The contrarian who sees through the madness, sells at the top, and watches smugly as prices collapse? That person exists in financial folklore. In the data, they're almost impossible to find.
When the media sounds the alarm — and when it doesn't
The financial press doesn't warn investors about bubbles either. Media coverage increases as prices rise — but it almost never suggests anything is wrong.
Greenwood and Stolborg found that fewer than 10% of articles use words like "bubble" or "overvalued." Even as prices reach absurd levels. Even as valuations stretch beyond anything the fundamentals could justify.
Here's the part that should unsettle you: that fraction doesn't increase after the crash begins.
The press writes more about these stocks as they surge. Column inches multiply. But the coverage is breathless, not sceptical. The narrative is momentum, opportunity, the fear of missing out.
Consider Yahoo in 1999. The poster child of dot-com excess. Media attention roughly doubled in the three months before its peak. Yet there's little evidence the press recognised a bubble forming. Coverage framed Yahoo as a juggernaut, a new paradigm, a company rewriting the rules of business. The scepticism came later — much later — after the damage was done.
GameStop in 2021 was different. Terms like "bubble" and "mania" appeared in real time, as the stock surged. Journalists called it what it was while it was happening.
But GameStop is the exception. Most boom-bust episodes look like Yahoo: optimism unchecked until well after the collapse.
The investment strategist Joachim Klement draws an uncomfortable conclusion from this pattern. If widespread bubble talk actually preceded crashes, we'd expect the media frenzy to peak just before prices fall. It doesn't. The alarm comes too late, or not at all.
All that "AI bubble" chatter you've been reading? It might be a sign that the real danger is still ahead. Or it might mean nothing. What it almost certainly doesn't mean is that a crash is imminent.
Why being right won't make you rich
Even the few investors who recognise a bubble can't profit from it. The mechanics of the market work against them.
Start with short-selling — the classic way to bet against an overvalued stock. If sceptics were acting on their beliefs, you'd expect short interest to climb as prices reach dangerous heights. It doesn't.
Three months before the average price peak, short interest sits at around 5% of shares outstanding. Lower than the market average. The sceptics aren't piling in. They're largely absent.
Short interest only rises after crashes have already begun. By then, the opportunity has passed.
Why don't more investors bet against obvious excess? Because it's expensive. Share lending fees — what you pay to borrow stock for shorting — are elevated throughout the run-up and spike near the peak. These stocks are, in the researchers' words, "difficult and costly to bet against."
High fees signal something important. It's not that nobody wants to short these stocks. It's that the shares aren't available. Limited supply from lenders, not limited demand from sceptics, keeps short interest low. The few investors who do recognise overvaluation often can't access enough shares to make the trade worthwhile.
And then there's timing.
About 75% of boom-bust stocks crash within a year of their peak. That sounds like a reasonable window. But 25% take longer — sometimes much longer. Prices can linger near their highs for 12, 15, even 18 months before the collapse comes.
If you're short, you're bleeding money every day you wait. Borrowing costs accumulate. Margin calls loom. And the stock keeps drifting sideways or even climbing, making you look foolish.
As Klement observes: "Even if you know you are in a bubble, share prices can linger near the bubble peak for quite some time."
Being early and being wrong feel identical when you're haemorrhaging cash.
The bubble warnings paradox
Here's the strangest finding: extreme optimism predicts crashes. The most bullish analyst forecasts are associated with the highest probability of collapse.
Greenwood and Stolborg didn't just describe what happens at peaks. They tested whether beliefs could predict which run-ups would end badly. The answer is yes — but in reverse.
When analysts forecast aggressive long-term earnings growth, crash probability rises. A one-standard-deviation increase in growth expectations pushes the likelihood of a 50% crash from 67% to 75%. The more confident the professionals, the more likely disaster follows.
High share lending fees predict crashes too. Not because they signal scepticism — the sceptics are largely absent — but because they indicate constraints. When it's expensive to short a stock, the few doubters who exist can't correct the mispricing. Prices drift further from reality.
What doesn't predict crashes? Short interest itself. The presence of bears betting against a stock tells you almost nothing about whether it will collapse.
This is the paradox. Danger doesn't announce itself through disagreement or caution. It hides in consensus. The warning sign isn't that people are worried. It's that they're not.
"Optimism portends crashes," the researchers write. "The most bullish forecasts predict the highest crash risk."
The very confidence that drives prices up ensures nobody gets out in time.
What this means for AI stocks
Nobody knows whether AI stocks are in a bubble. But we can apply the framework.
Palantir trades at a price-to-sales ratio above 125. The research defines "extreme valuation" as a P/S above 5. Nvidia's ratio has fluctuated but recently sat around 25-30 — still stretched by historical standards. Many AI-adjacent stocks have doubled or more within the past year.
The structural conditions look familiar. Analyst optimism is elevated. Short-selling remains constrained by high borrowing costs. Media coverage is extensive but, as we've seen with the Google Trends data, cycles between alarm and complacency without resolving anything.
Does this mean a crash is coming? I don't know. The researchers don't claim to know either. Their paper isn't a prediction tool — it's an explanation of why prediction fails.
What the research does demolish is a common investor strategy: "I'll ride the wave and get out when it becomes obvious."
It never becomes obvious.
Yahoo's bubble wasn't obvious in December 1999, when analysts were forecasting 52% returns. "Most cases resemble Yahoo! in 1999," Greenwood and Stolborg write, "where optimism remained unchecked until well after the crash."
If you're waiting for a clear signal — a consensus that the party's over — you'll be waiting alongside everyone else. And by the time that consensus forms, prices will already have collapsed.
This doesn't mean AI stocks will crash. It means that if they do, you almost certainly won't see it coming.
The evidence-based alternative
The rational response to bubble uncertainty isn't better prediction. It's a portfolio that doesn't require you to predict.
Greenwood and Stolborg are blunt: "Bubble timing is extremely difficult. Even investors who recognise elevated crash risk may find it difficult to profit from or avoid these episodes."
This isn't a counsel of despair. It's a redirect. If timing doesn't work — and 43 years of data suggest it doesn't — the answer lies elsewhere.
Diversification. Broad exposure across sectors, geographies, and asset classes. Not because it eliminates risk, but because it eliminates the need for foresight. You don't have to know which stocks will crash if no single crash can sink you.
The instinct when bubble talk intensifies is to act. Sell the frothy positions. Raise cash. Wait for clarity. But clarity never comes — not before the crash, anyway. And the cost of stepping aside too early can exceed the cost of riding through.
Research consistently shows that investors who try to time bubbles destroy more wealth through mistimed exits than they'd have lost by staying put. Missing a handful of the market's best days — which often cluster around the worst days — devastates long-term returns.
The uncomfortable truth is that bubbles are only reliably identifiable in hindsight. Any strategy that depends on spotting them in real time is built on a foundation that doesn't exist.
Stay diversified. Stay invested. Accept that some of your holdings will occasionally look absurd — and that you won't know which ones until it's too late to matter.
Google searches for "AI bubble" spiked, then faded. The headlines moved on. The valuations didn't. Somewhere, an investor is still watching the rearview mirror, waiting for the road behind to reveal what lies ahead.
It won't.
Harry Markowitz kept a 200-year-old book about speculative manias on his shelf — not because it taught him how to avoid them, but because it taught him he couldn't. His life's work wasn't prediction. It was building portfolios robust enough that prediction became unnecessary.
The AI bubble may burst tomorrow. It may inflate for years. This research offers no forecast, only a hard truth: by the time consensus forms around a bubble, the damage is already done or the opportunity has already passed.
The paradox resolves simply. More warnings don't mean less danger, and less danger doesn't mean no danger. It means danger is unknowable — and the only sane response is to stop trying to know it.
You don't need to see the crash coming. You need a portfolio that can survive it either way.
Resources
Greenwood, R., & Stolborg, C. (2025). Bubble beliefs [Working paper]. Harvard Business School and Copenhagen Business School.
Brunnermeier, M. K., & Nagel, S. (2004). Hedge funds and the technology bubble. The Journal of Finance, 59(5), 2013–2040.
Recently on TEBI
Advisers, is your content working hard enough?
The best advisers aren't just good at what they do. They're visible. They're trusted before the first meeting even happens.
Regis Media, which produces The Evidence-Based Investor, helps financial advisers and planners build that trust through regular, professional content. Videos that showcase your expertise. Articles that demonstrate your thinking. All written, produced and edited for you.
Prices start from £200 a month.
Prefer watching to reading?
Subscribe to our YouTube channel for evidence-based investing insights explained visually. Complex concepts made clear in just a few minutes. New videos every week — hit subscribe so you never miss one.
