Stock Market Crash Warning Signs: What History Actually Says

TL;DR

• Almost every "crash warning sign" you will read has no predictive record. They are patterns found after the fact in five or six events, which is far too small a sample to learn from.
• The two indicators with a genuine record, the yield curve and valuation, are both useless for timing. The yield curve has led recessions by anywhere from 6 to 24 months. Valuation works over ten-year horizons, not one-year ones.
• Crashes do not share a cause. 1987 had no recession attached to it. 2020 was a virus. 2008 was credit. 2000 was capital spending. The only thing they share is that most people did not see them coming.
• Speed is not severity. The 2020 crash was the fastest 30% fall in history and was fully recovered within five months. The 2000-2002 decline was slow and took far longer to repair.
• The useful response to crash risk is not prediction, it is position sizing. That is a boring answer and it is the only one the evidence supports.

Why "Crash Warning Signs" Almost Never Work

Start with the problem nobody selling a crash indicator mentions: the sample size is tiny.

In roughly a century of American market history there have been about five events most people would call a crash. Five. If you go looking for what they had in common, you will find dozens of coincidences, because with five data points and hundreds of candidate indicators, something will always line up.

This is why the genre is so reliably wrong. An indicator that "correctly predicted the last four crashes" has, at best, four successes and an unknown number of false alarms nobody counted. The false alarms are the whole story, and they are almost never published alongside the hit rate.

There is a second problem, and it is worse. Markets are not a natural system with fixed laws. If a genuine, reliable crash signal existed and became widely known, people would sell on the signal rather than on the crash, which would move the crash forward and destroy the signal. Any indicator that works gets arbitraged into uselessness by the people using it.

So the honest starting position is scepticism, and the useful question is not "what predicts a crash" but "what has any record at all, and what can I actually do with it".

What The Five Big Ones Actually Had In Common

Very little, and that is the point.

October 1929 produced a 12.8% single-day drop and the deepest bear market on record, and it was followed by a depression, bank failures and a decade of damage.

Black Monday, October 19, 1987, saw the Dow fall 22.6% in a single day, the largest one-day percentage loss in its modern history, wiping an estimated $1.71 trillion off global markets. There was no recession. The economy carried on. Within two trading sessions the Dow had recovered 57% of the fall. A generation of investors who sold into it spent years regretting it.

March 2000 to October 2002 was the opposite shape: slow, grinding and enormous. The S&P 500 fell more than 47%, and the Nasdaq Composite fell roughly 78% peak to trough. This one was a capital spending bust. Companies had built for demand that did not arrive.

October 2007 to March 2009 took the S&P 500 down by roughly half, the deepest decline since the Depression. The cause was credit and leverage in housing, which almost nobody in equities was watching.

February to March 2020 produced the fastest 30% decline in S&P 500 history, 33 calendar days peak to trough, a fall of about 34%, caused by a virus that no financial indicator could have flagged. It made a new high within five months.

Read that list again and notice what is missing: a common cause. A monetary panic, a portfolio insurance feedback loop, a capital spending bust, a credit crisis and a pandemic. Anybody claiming a single indicator anticipates all five is describing a coincidence.

Notice also that speed and severity are unrelated. The fastest crash was among the shallowest and quickest to repair. The slowest was among the most damaging. "How fast is it falling" tells you nothing about how bad it will get.

The Two Signals With A Real Record, And Their Fatal Flaw

Two indicators genuinely have something behind them. Both are useless for the thing people want them for.

The yield curve

When short-term interest rates rise above long-term rates, the curve inverts, and that has a genuinely impressive record. The 10-year against 3-month spread has inverted before every US recession since 1969. The 2-year against 10-year spread has inverted before seven of the last eight recessions since 1968.

Now the flaw. The average lead time is around 11 months, and the range runs from 6 months to 24 months. Some measures put the median nearer 15 months.

Think about what a two-year error bar means in practice. You sell on the inversion. The market then rises for another eighteen months, because late-cycle markets are frequently the strongest part of the cycle. You miss that, then you miss the recovery too, because nobody who sold early ever manages to buy back at the bottom. A signal that is right about direction and wrong by up to two years about timing will cost you more than the crash it warned about.

It is also worth understanding why the curve matters, which is about the price of money rather than magic. We covered the mechanics in why stocks fall when long yields rise.

Valuation

The cyclically adjusted price-to-earnings ratio, usually called CAPE or the Shiller P/E, does contain real information. High valuations have historically been associated with lower average returns over the following decade.

The flaw is the horizon. That relationship operates over ten-year periods, not one-year or two-year ones. An elevated CAPE can stay elevated for years, occasionally decades, before returns normalise. As a tool for setting long-run expectations it is excellent. As a tool for deciding whether to be in the market next quarter it is close to worthless, and the research is unusually blunt about that.

So the two best indicators available both say the same thing: they can tell you something about the next decade and nothing reliable about the next year.

The Signals With No Record At All

Treat these with the suspicion they deserve, because they circulate constantly:

• Chart patterns named after objects. Death crosses, head and shoulders, Hindenburg omens. When tested across long histories they generate far more false signals than true ones.
• Anniversary effects. "October is dangerous" survives because two famous crashes happened in October. September has historically been the weaker month, and neither fact is tradeable.
• Single sentiment readings. Extremes in sentiment surveys are more often contrarian buy signals than sell signals, which is the opposite of how they get quoted.
• "Everybody is talking about stocks at parties." There is no data behind this, only survivorship in the retelling.
• Any indicator with a precise date attached. Nobody who could genuinely time a crash to a month would tell you about it.

There is one soft signal worth respecting, not because it is measurable but because it recurs: a widely held belief that the old valuation rules no longer apply to a particular new thing. It was railways once, and radio, and dot-coms, and it recurs today around the trillion dollars of market value that moved on AI capital spending assumptions. This is not a timing tool. It is a reason to check your position sizes.

What To Do Instead

If prediction does not work, the only lever left is preparation, and preparation is mostly arithmetic rather than insight.

1. Size positions so a 50% fall is survivable. That is the real historical worst case for a diversified equity portfolio, twice in the last century. If a 50% decline in your equity allocation would force you to sell, your allocation is wrong regardless of what any indicator says.

2. Know your actual time horizon, honestly. Money needed inside five years does not belong in equities. Every crash on the list above was fully recovered eventually, but "eventually" ranged from five months to several years, and you do not get to choose which one you get.

3. Keep enough cash that a crash is an opportunity rather than an emergency. The people who did well out of 2009 and 2020 were not the ones who predicted them. They were the ones who did not have to sell and had something left to buy with.

4. Do not try to step out and back in. The market's best days cluster inside its worst periods, which is precisely when someone who sold on a warning sign is sitting in cash. We ran the arithmetic on that in time in the market versus timing the market, and it is brutal.

5. If you are adding money, keep adding it on a schedule. A fixed contribution schedule buys more shares when prices fall, which converts a crash from a disaster into a discount. The trade-offs are set out in dollar cost averaging versus lump sum.

6. Be careful with leverage, and understand what it does in a fall. Leveraged products decay in exactly the choppy, violent conditions a crash produces, which is covered in the leveraged S&P 500 ETF catch.

The Honest Summary

Nobody knows when the next crash is coming, including the people who called the last one. The indicators with real evidence behind them operate on horizons measured in years, which makes them useful for setting expectations and useless for setting trades. The indicators that promise precision have no record worth the name.

What history does support is narrower and duller: crashes happen every decade or so, they have unrelated causes, they are all eventually recovered, and the investors who come through them are the ones who were positioned to survive rather than positioned to predict. That is not a satisfying answer to "is the market about to crash". It is the only one the evidence will carry.