Can stock market crashes be predicted? What 50 years of macro shocks actually show

The short answer

No one has reliably predicted the date of a crash, and the record says no one should claim to. But the same record shows two things that can be measured. How fragile conditions are, which told you how lasting the damage would be when a shock arrived. And what type of shock is underway, which told you how long recovery would take. Those two measurements are useful. Date-calling is not.

We examined fourteen US market stress events from 1979 to 2026, including Black Monday 1987, the dotcom crash, the 2008 financial crisis, the COVID crash, the 2022 bear market, and the 2026 US strikes on Iran. For each one we asked a strict question: what was in the public record before the event, on what date, and would acting on it have paid? Retrospective folklore (“everyone knew”) did not count. Only sources dated before the fact.

The famous crashes, and what was actually knowable

Black Monday, October 1987. The closest any event came to being called. The Wall Street Journal named the exact crash mechanism (portfolio insurance selling feeding on itself) seven days before it happened, and the market fell 10% in the week before the crash. Even then, the specific day depended on a weekend policy dispute. A reader of Friday's paper had every reason to reduce risk, and no way to know Monday would fall 20.5% in one session.

The dotcom crash, 2000. Every fragility gauge stood at record extremes, in public, for over a year: the highest CAPE ever recorded, 476 IPOs in 1999 with an average 71% first-day pop, margin debt going vertical. And every dated attempt to time it was wrong by one to three years. Alan Greenspan raised “irrational exuberance” in December 1996; the NASDAQ then roughly quadrupled. The fund managers most right about the bubble (Tiger, GMO) lost roughly half their investors before being vindicated.

The 2008 financial crisis. The credit market called the event eight to twelve months before stocks noticed. Housing prices rolled over on public data in 2006, HSBC warned on subprime in February 2007, and interbank lending froze in August 2007. The S&P 500 then made its all-time high in October 2007, after all of that. Fragility was public. Timing was not, and the Federal Reserve chairman's “contained” assessment was the median expert view, not an outlier.

The COVID crash, 2020. China locked down a city of 11 million people on January 23. The S&P 500 made a record high on February 19, four weeks later, with exponential case growth published daily in between. The information was free and public. The market lacked a reaction function, not data. This one was trackable, and almost nobody tracked it into a trade.

The 2022 bear market. Commonly remembered as the Ukraine war selloff, but the timeline says otherwise. The S&P peaked on January 3, 2022, 52 days before the invasion. Inflation had printed above 5% in seven consecutive monthly reports, and the Fed's hawkish turn was dated and public. When Russia invaded, the S&P closed up on the day and stood about 13% above its invasion-day low a month later. The driver was monetary policy. The war was the aggravator.

The 2026 US strikes on Iran. The most publicly telegraphed shock in the sample: a snapback of UN sanctions in September 2025, a liquid prediction market, and an announced 10 to 15 day deadline from the President nine days before the strikes. Even Iran signaled expectation, surging oil exports while tanker insurance repriced. Equities still entered the war at record highs, fell about 9%, and made a new all-time high 50 days after the bombs fell.

The part crash predictors never publish: the misses

Every famous “this indicator called the crash” story has a sibling story where the same indicator fired and nothing happened. We counted them.

• High-yield credit spreads: doubled off their lows before the 2008 crisis, a genuine early warning. They also doubled in 2011 and in 2015–16, both times passing the levels they showed at the market's 2007 peak, with no bear market following either.
• The yield curve: inverted before the 1980, 1990, 2000 and 2008 recessions. It also inverted in 2019 before a recession caused by a virus that did not exist yet, and stayed inverted from 2022 to 2024, the deepest inversion since 1981, while stocks returned 26% in 2023.
• Valuation: the CAPE ratio crossed its 1929 record around 1997, three years and a doubling before the top. It has spent most of the last decade above its 90th percentile.
• Market breadth: the NYSE advance/decline line peaked in April 1998 and correctly flagged trouble. It then stayed bearish through 1999, when the NASDAQ gained 86%.
• The VIX: anticipated nothing in fourteen events. It was low at the 1987, 2000, 2018, 2020 and 2026 tops. It is a confirmation gauge, and anyone selling it as an early warning is mistaken.

A single indicator that fires two or three false alarms per real event cannot be called a prediction tool. Treat it as a conditions gauge, and present it as one.

War headlines are the wrong sell signal

Since 2015, we count ten episodes of serious geopolitical escalation with a visible run-up, from North Korea in 2017 through the Soleimani strike in 2020, the Syria strike rounds, the Taiwan strait episodes, the direct Iran-Israel exchanges of 2024, the June 2025 strikes and Russia's invasion of Ukraine, plus the 2026 war itself as an eleventh. Not one of them drove a 15% decline in the S&P 500. The deepest event-driven selloff in the set was the 2026 war's: with the Strait of Hormuz closed and oil at $126, the index bottomed about 9% down and round-tripped to a record in about 50 days. The invasion of Ukraine marked a local bottom: the S&P closed higher on invasion day and stood well above its invasion-day low a month later. The index did trade 16% below its invasion-day level eight months on, but that bear began seven weeks before the invasion and tracked the Fed's tightening, which is why the 2022 section above treats it as a policy bear rather than a war selloff.

The moves were real. They just happened somewhere else: in oil, in volatility, in defense and energy stocks, in Treasuries. Selling stock indexes on war headlines has been a consistently losing reflex for a decade, and the study found the pattern goes back to 1990, when equities shrugged off 100,000 Iraqi troops massing on Kuwait's border for a week. The honest caveat: eleven episodes is a small sample from an era of fast de-escalation, and it contains no repeat of 1973.

What can be measured

Two things held up across fifty years.

Fragility predicted how bad it would get, and for how long. The bear markets that needed years to heal (2000 took seven, 2008 took five and a half, 2022 took two) were exactly the ones where valuation, leverage and credit conditions stood at measurable public extremes beforehand. When that fragility was absent, even enormous shocks did brief damage: the 2026 war hit a market with strong earnings momentum, bottomed about 9% down and healed in 50 days. The COVID crash is the exception that clarifies the rule: a 34% collapse without classic fragility, and also a five-month recovery, because the fuel was fear rather than leverage. Fragility gauges could not date anything, but they consistently separated drawdowns that healed in weeks from ones that consumed years.

Shock type predicted recovery. External shocks met by fast policy response recovered in weeks to months: 28 days in 1997, 50 days in 2026, about five months for COVID. Bursting bubbles built on leverage and valuation took years: five and a half for 2008, seven for 2000 (fifteen for the NASDAQ). Policy-driven bears ended when the policy reversed, on a dated announcement, and not at any particular price level. Knowing which type of drawdown you are in has historically been worth more than any attempt to know when the next one starts.

What this means for how you invest

• Treat elevated fragility as a position-sizing and hedging-cost input, not a sell signal. The investors who were right early in 2000 and 2008 nearly went out of business waiting.
• When a drawdown starts, ask what type it is before deciding how patient to be. Falling bond yields alongside falling stocks suggests a demand shock that policy can meet. Rising yields alongside falling stocks (2022, 2026) means bonds are not your hedge this time.
• Do not sell broad indexes on escalation headlines. If an event has a real transmission channel into the economy, it shows up first in commodities, credit and the affected sector.
• Distrust any product or pundit claiming to predict crash timing. Fourteen events, fifty years, zero dated public calls that held up.

How tickerseer applies this

Our platform tracks, for every covered stock, its current valuation against that company's own long-run norm. Aggregated across the 1,145 companies with a usable valuation basis, that gives a market fragility read no single index number can: as of August 2026, the median stock in our universe traded about 7% below its own historical valuation norm, while the same reading weighted by company size sat about 13% above, and the fifty largest companies about 19% above. A handful of very large companies carry most of the stretch. That gap between the median stock and the index is the same narrow-leadership signature the record shows in 1999 and 2021, and per-stock measurement makes it visible where a single blended ratio hides it.

Those figures are a dated snapshot rather than a live feed, recomputed each month by the job behind our valuation temperature report. Two limits are worth stating. The size-weighted readings cover the 493 companies whose market value we can place with confidence: a fault in how we collected that field dropped the magnitude from the rest, and a billion cannot be told from a million after the fact. And these are medians with the extremes trimmed rather than averages, because a company whose earnings have collapsed shows a price-to-earnings ratio many times its own norm, and a few of those are large enough to swing an average on their own.

We publish the full study, including every indicator's false alarms and the events these gauges would have missed, because a conditions gauge is only trustworthy if its misses are on the table. What we will not publish is a crash date. Fifty years of evidence says that number does not exist.