S&P 500 Crash Clusters

Thirteen drawdowns since 1990, grouped by how fast and how long they fall. The group decides which family of hedge trigger can arrive in time, and which will always be late.

No single rule fits every crash. A hedge tuned on extended bear markets loses money in fast corrections, and neither reaches a flash shock before it is over. The useful question is not which trigger is best, but which group you want to hedge, and what that costs each year.

Thirteen episodes in thirty-six years

Log scale. The solid band is the fall from peak to low; the pale band is the time it took to make a new high.

Month-end closes plus the peak, low and new high of each episode. From 1993 the level is the SPY closing price times ten, a close proxy for the S&P 500; 1990–92 uses index closes at quarter-ends.

The same episodes, without the time axis

Length of the decline against speed of the fall. Both axes are logarithmic, so the dashed diagonals are lines of equal depth, and marker size also shows depth. This is where the groups separate.

Speed is peak-to-low depth divided by sessions down, so it averages out any acceleration near the low. Shaded areas show the rule that defines each group; hollow markers are controls that stayed under the 15% threshold.

Three shapes, three different problems

Peak set to 100, months on the horizontal axis. One archetype per group: the shape of the decline and of the recovery is what makes the matching triggers incompatible.

Illustrative profiles drawn from monthly closes; the depth of each low is indicative.

Four groups

Time to a new high is measured from the low of each episode.

Flash shocks

Under 25 sessions, over 1% a day

Covid, February–March 2020

New high after 5 months

No price-based trigger arrives in time. The trend breaks its moving average when most of the move has already happened. Either the hedge is already in the portfolio, or you accept not hedging.

Fast corrections

About 30–65 sessions, 0.3–0.6% a day

Gulf War 1990 · LTCM 1998 · 2010 · Euro debt 2011 · Q4 2018 · Tariffs 2025

New high after 3–5 months

The hardest ground. The trigger enters mid-decline, and the sharp recovery hands back the gain on the hedge. What matters here is the exit rule, not the entry.

Extended bears

Over 190 sessions, under 0.2% a day

Dot-com 2000–02 · Financial crisis 2007–09 · Rates and inflation 2022

New high after 15–56 months

Trend following’s territory. Months below the moving average repay the false signals many times over. It is the only group where a full 100% hedge makes sense.

Controls

Under the 15% threshold, kept out of the sample

Volmageddon 2018 · Carry-trade unwind 2024 · China and oil 2015–16

New high after 2–6 months

They measure false positives. These are the cases where a hedge switches on, pays its cost and never finds the crisis it was waiting for.

The thirteen episodes

Recomputed on daily closes from 1993 onward. Depth is peak to low; speed is depth divided by sessions down; the last column runs from the low to a close above the old peak.

Episode	Group	Depth	Sessions down	Speed, % a day	Low to new high

† Before the price series starts in January 1993: public index closes, not recomputed. ‡ Measured from the July 2011 high; from the April high the decline was 19% over 108 sessions.

What the study does not show

The window starts in 1990. Black Monday, 1987, is left out. It is the fastest crash on record and would sit with the flash shocks.
The flash-shock group is one case. Covid is the only drawdown over 15% that falls that fast in the window, so read the group as an example, not a statistic.
Thirteen episodes is a small sample, and the group boundaries are set by hand on speed and length, not found by an algorithm.
The controls are hand-picked. Three sub-threshold episodes are shown; other declines of 10–12% (1997, 1999, 2023) are not included, so the false-positive count is a sample, not a census.
Numbers are recomputed on SPY closing prices, used as a proxy for the index from 1993. Depths can differ from the index itself by a few tenths of a point, mostly because of dividends. The 1990 Gulf War episode predates the series and was not recomputed.

The distance between the groups is the result that matters. A rule calibrated on the extended bears loses systematically in the fast corrections, and neither touches the flash shocks. The practical conclusion is not to pick the best trigger, but to decide which group to hedge and at what annual cost.

For information and research purposes only. This is not investment advice or a recommendation to buy or sell any security. Past drawdowns do not predict future ones.

S&P 500 crash clusters