Stock Market Crashes

There is no official threshold, which is worth knowing before reading any list including this one. In common use a correction is a decline of about ten per cent, a bear market twenty per cent or more, and a crash a decline that is both severe and sudden, days or weeks rather than months. The categories overlap badly: 1987 was a crash without a lengthy bear market, and 2000–02 was a long bear market containing no single day comparable to October 1987.

Because three different questions get the same name. Nominal price recovery asks when the index next closed at its old level. Total return adds reinvested dividends and is considerably faster. Real terms adjust for inflation and are considerably slower, after 1929 the Dow did not recover in inflation-adjusted terms for far longer than the nominal figure suggests. This page uses nominal price recovery throughout and says so, because a table that mixes the three is worse than no table.

Only loosely. The market of 1929 had no circuit breakers, far less disclosure, margin requirements that permitted leverage unavailable today, and a much narrower list of listed companies. The 89 per cent figure is real, but treating it as a scenario for a modern market ignores that most of the mechanisms that produced it have since been legislated away or automated out.

Very differently, and the onset is a better guide to what a decline feels like than its depth. 1987 and 2020 began abruptly from near a high, with most of the damage done in weeks. 1929, 1973 and 2000 began as ordinary-looking pullbacks that failed to recover and only became recognisable as something else months later. 2007–09 sits between: the index high was in October and the character of the market had changed well before, but the severe phase arrived nearly a year afterwards.

One structural condition recurs and no reliable signal does. In each case a large number of holders eventually needed to reduce exposure through a market that could not absorb them at once, leverage in 1929, programme selling in 1987, concentrated positioning in 2000, funding markets in 2008, and an abrupt reassessment in 2020. Narrowing breadth and advances on thinning volume preceded several of them, and also preceded many periods that resolved into nothing. The companion page on volume covers what that pattern does and does not support.

Nothing that generalises usefully, which is itself the finding. 1932 and 2009 ended with policy intervention on a scale not previously attempted; 1987 ended within days for reasons still argued over; 2002 ended after valuations had unwound rather than because of any single event; 2020 ended after fiscal and monetary support arrived within weeks. Anyone constructing a rule about how declines end from five episodes is working with a sample of five.

The honest answer is that the sample is far too small to say, and the framing invites a false answer either way. Two severe declines in the first two decades of this century looks like an acceleration against the second half of the twentieth; it looks like nothing against the 1929–1937 sequence. Market structure has also changed enough — circuit breakers, disclosure, the composition of who holds equities — that frequency across that span is not measuring a constant thing.

Longer, in every case, and usually by a wide margin. The 2020 decline took 33 sessions and the recovery about five months; 2007–09 fell for seventeen months and recovered over four years; 1929–32 fell for nearly three years and recovered over twenty-five. The asymmetry is the most consistent feature of the whole table (declines are fast and recoveries are not), and it is the single most useful thing in it for anyone deciding what a drawdown will actually cost.

It supports the observation that every decline in the table was eventually recovered in nominal terms, which is a weaker statement than it is often used to make. The recovery periods range up to twenty-five years, and an investor whose horizon or circumstances did not span that period would not have experienced the recovery at all. The record is compatible with several conclusions; anyone quoting it in support of one is choosing, and the choice belongs to the reader rather than to a reference page.

Because a list has to stop somewhere and the boundary is a judgement, which is worth admitting rather than hiding. The 1962 decline took the S&P 500 down roughly 28 per cent, 1990 about 20, and 2011 about 19, all bear markets or close to it, none producing a crash session comparable to October 1987. Including them would double the table and change none of its conclusions; excluding them is a choice about what "major" means, and a different author would draw the line elsewhere.

The first month in which the same index closed at or above its pre-decline peak, on a nominal price basis. That definition is stated because the alternatives give materially different answers: adding reinvested dividends shortens every figure, and adjusting for inflation lengthens them, after 1929 the inflation-adjusted recovery took substantially longer than the nominal one this table reports. A single consistent basis across all seven rows is what makes the comparison mean anything.

No. Every depth and recovery figure here is price only. For a long-horizon holder that understates what actually happened by a wide margin, because dividends compounded through the recovery periods are substantial, most of the twenty-five-year figure for 1929 closes considerably faster on a total-return basis. The price series is used because it is what index charts show and what most sources quote, not because it is the most economically meaningful.

It is the convention, and it is arbitrary. Nothing distinguishes 19.6 per cent from 20.4 per cent except which side of a round number it fell on, and yet the threshold determines whether an episode is recorded as a bear market at all. Several declines sit close enough to the line that different sources classify them differently. Treat the label as a filing category rather than as a description of severity.

The same one throughout, and it usually cannot be. The Dow is the only series long enough to cover 1929, the S&P 500 begins in its modern form much later, and the Nasdaq Composite did not exist until 1971, so any table spanning a century is necessarily stitched from different baskets. That is why each row here names its index. A chart that plots a single line labelled "the market" from 1929 to today is hiding at least two substitutions.

October has an outsized reputation (1929, 1987 and 2008 all had severe October sessions), and the reputation is mostly an artefact of a small sample. Two or three memorable episodes in one month across a century is exactly what randomness produces, and the many Octobers that passed without incident are not remembered. Seasonal explanations for crash timing consistently fail out of sample.