The Calmest Market Tops Cause the Deepest Crashes

There is a comforting idea that you can see a correction coming in the RSI. When the market gets "overbought," the story goes, RSI pushes above 70, then 80, the rally is stretched, and a fall is overdue. So an elevated RSI at a new high is supposed to be a warning sign. We tested it the long way. We took every time the S&P 500 fell 10% or more from an all-time high before recovering, all the way back to 1950, 26 corrections in total, and read the RSI on the exact day the market peaked. Then we lined each peak's RSI up against how deep the decline that followed got.

There is a real relationship. It just runs the opposite way to the folklore.

The more overbought the top, the shallower the fall

Every 10%+ S&P 500 correction since 1950 (close basis), grouped by the RSI on the peak day. "Median drop" is the peak-to-trough decline; "became a bear" is the share that fell at least 20%.

RSI at the peak	Corrections	Median drop	Became a bear (20%+)
60 to 70 (steady momentum)	11	-22%	55%
70 to 80 (overbought)	10	-17%	40%
Over 80 (extreme)	4	-13%	25%

Read it top to bottom. The hotter the RSI at the peak, the smaller the typical correction and the lower the chance it turned into a bear market. Tops with merely "steady" momentum (RSI in the 60s) fell a median 22% and became bear markets more than half the time. Tops with extreme, blow-off momentum (RSI above 80) fell a median of just 13% and only rarely turned serious. Across all 26 corrections, the rank correlation between peak RSI and drawdown depth is about minus 0.36, a real, if noisy, negative tilt. (An RSI below 60 at an all-time high is so rare it happened exactly once in 76 years.)

The deepest declines began from ordinary RSI

The pattern is starkest at the extremes. Here are the five worst declines in the sample, and the RSI they peaked on.

The deepest S&P 500 drawdowns since 1950 and the RSI on the day the market topped.

The worst declines	RSI at the peak	Drop
Financial crisis, 2007	67	-57%
Dot-com bust, 2000	70	-49%
The 1973-74 bear	68	-48%
The 1968-70 bear	84	-36%
COVID crash, 2020	66	-34%

Four of the five worst crashes in three-quarters of a century topped out with RSI in the high 60s, what a chartist would call healthy, unremarkable momentum. None of those four was screaming overbought. The lone exception, 1968, is the one extreme-RSI reading that became a deep bear. Meanwhile the most overbought tops in the entire sample, January 2018 (RSI 87), September 1955, January 1953, were quick 10 to 15% pullbacks the bull market shrugged off in weeks. Split the corrections by severity and the same thing shows up: the shallow ones (10 to 20%) topped on a median RSI of 72, the bear markets on a median of 69. The serious declines came from the calmer tape.

Why: a blow-off is not a break

The mechanism makes sense once you separate the two kinds of top. An extreme-RSI peak is a momentum blow-off. The rally has gone near-vertical, prices have overshot, and what follows is a fast snap-back, a mean reversion of the overshoot, that rarely impairs the underlying trend. It is loud and shallow.

A deep bear market is something else entirely. It is not momentum running out, it is a fundamental break: a credit crisis, a valuation reset, a pandemic, an inflation shock. Those tops form quietly. The index grinds to one more new high on moderate, often already-fading momentum, the rally narrowing under the surface, and then the shock arrives. The RSI at those peaks looks ordinary because nothing in the price action is screaming. The danger was never in the momentum. It was in the fundamentals, and RSI cannot see those.

The honest limit

Be clear about what this is and is not. A minus 0.36 correlation across 26 events is a tendency, not a tool. The RSI on a peak day cannot tell you whether the next 10% will stop at a 12% dip or roll into a 50% bear, because that is decided by the shock that follows, and the peak knows nothing about the shock. So the takeaway is not "use RSI to forecast the crash." It is narrower and more useful: unlearn the reflex that a high RSI is the dangerous signal. It is the loud, shallow one. If any reading at a new high deserves respect, it is the calm one.

How we tested it

• Data. Daily S&P 500 closes from 1950 to today, about 19,000 trading days.
• The corrections. Every time the index closed 10% or more below an all-time high and later recovered to that high. That produced 26 non-overlapping corrections.
• The peak RSI. The standard 14-day RSI (Wilder smoothing) on the exact day the index made its high, before the decline began.
• The depth. The peak-to-trough close-to-close decline, and whether it crossed 20% into a bear market.
• The analysis. Drawdown grouped by peak-RSI band, the rank correlation between peak RSI and depth, and a split of shallow corrections versus bear markets by their peak RSI.
• A note on the measure: using closing prices, a couple of marginal declines that only touched 10% intraday are excluded. The pattern is the same with or without them.