Analysis · The record

Bear Markets and Long Declines

A crash is a few sessions; a bear market is a period. One produces every extreme reading in this reference at once, and the other is mostly made of sessions that look like nothing at all.

Depth is the wrong axis

The conventional definition of a bear market, a fall of twenty per cent from a peak, sorts declines by depth, and depth is the least informative thing about them. Nothing changes in a market at nineteen per cent, and the threshold survives because it is a convenient round number.

Sort the same episodes by duration and they separate into two genuinely different kinds of event, with different volume paths, different breadth behaviour, and different reasons for being hard to read at the time.

Major declines by depth, with their durations notedA bar chart of six major declines ranked by percentage fall from peak: 1929 to 1932 at 89 per cent, 2007 to 2009 at 57 per cent, 2000 to 2002 at 49 per cent, 1973 to 1974 at 45 per cent, and 1987 and 2020 both at 34 per cent. The notes beside each bar give the duration, which ranges from thirty-three days to nearly three years.% fall from the peak1929–1932−89 %Nearly three years. The deepest on record and the slowest, a sequence ofdeclines, not one event.2007–2009−57 %Seventeen months. Contained several crash-like weeks inside a much longerdecline.2000–2002−49 %Two and a half years, and almost no single dramatic session. The archetype of aslow bear market.1973–1974−45 %Twenty-one months, through an inflationary period rather than a credit event.1987−34 %Weeks, not months. A crash by any definition, and it did not lead to a longdecline.2020−34 %Thirty-three days, recovered within months. The fastest decline of this size onrecord.Major declines by depth, with their durations notedA bar chart of six major declines ranked by percentage fall from peak: 1929 to 1932 at 89 per cent, 2007 to 2009 at 57 per cent, 2000 to 2002 at 49 per cent, 1973 to 1974 at 45 per cent, and 1987 and 2020 both at 34 per cent. The notes beside each bar give the duration, which ranges from thirty-three days to nearly three years.% fall from the peak1929–1932−89 %Nearly three years. The deepest on record and the slowest, asequence of declines, not one event.2007–2009−57 %Seventeen months. Contained several crash-like weeks insidea much longer decline.2000–2002−49 %Two and a half years, and almost no single dramatic session.The archetype of a slow bear market.1973–1974−45 %Twenty-one months, through an inflationary period ratherthan a credit event.1987−34 %Weeks, not months. A crash by any definition, and it did notlead to a long decline.2020−34 %Thirty-three days, recovered within months. The fastestdecline of this size on record.
Fig. 1: approximate peak-to-trough falls, widely documentedRanked by depth, the ordering is not very useful, the two shallowest entries here are a crash that lasted weeks and the fastest decline of its size on record, while the deepest took nearly three years. Read the notes rather than the bars: what separates these episodes is how long they took and what caused them, and the percentage is the least distinguishing feature of any of them. Figures are approximate peak-to-trough falls on the headline indices and vary by a point or two depending on whether closing or intraday extremes are used.

What each kind does to the measures on this site

A crash and a long decline, measure by measure
MeasureIn a crashIn a long decline
VolumeSeveral times baseline within days. Unmistakable, and the clearest evidence of forced selling.Mostly ordinary. Heavy sessions appear in clusters, and long stretches look like nothing.
Advance/decline dataCollapses to near-unanimity. Every issue falls, regardless of merit.Deteriorates gradually, usually starting before the index peak and continuing for quarters.
New lowsExpand across the whole list within a few sessions.Expand in waves, and are suppressed between them by the twelve-month window itself.
Volatility measuresExtreme immediately, and mean-revert quickly afterwards.Elevated for months at a level that becomes the new normal, so a threshold set beforehand stops discriminating.
Identifiable at the time?Yes, the readings are unmistakable while it is happening. What is unknown is what follows.No. It is a period, and periods are named afterwards.

The last row is the honest summary of this whole page. A crash announces itself in the data and tells you nothing about what comes next. 1987 was severe and was followed by a recovery, 2008 was severe and sat inside a much longer decline. A bear market is the opposite: it is only ever identified in retrospect, and the measure that would have described it earliest, breadth, was also correct for so long that it could not have been acted on.

Why breadth is early and stays early

Participation narrowing before a peak is the best-documented lead in this field and the least usable. Advance/decline data typically weakens while an index is still rising, because the advance is being carried by fewer and fewer issues, and that is a real description of a fragile market.

The difficulty is duration. A divergence can run for many months, and it looks exactly the same on the day it is right as on the day it is early. Every published example of breadth "calling" a top is chosen after the outcome was known, and the same configuration appears repeatedly in advances that continued for years. The market analysis section states this as the site’s general position: breadth supports a statement about how an advance is being made and never a date.

Why volume has no signature through a long decline

Volume responds to urgency, and most of a long decline is not urgent. Prices grind lower through sessions that would be unremarkable in any other month, and the participants who will eventually sell have not yet decided to.

So the heavy volume in a bear market arrives in bursts, the crash-like phases inside it, and those bursts are where the extreme readings in every measure on this site are found. Between them, a relative volume figure sits near one and adds nothing. Anyone waiting for volume to confirm that a decline is under way will be waiting through the largest part of it.

Recovery, and why depth predicts it poorly

The two shallowest declines in the figure recovered on completely different timescales, and the deepest took a generation. What separates them is the mechanism rather than the size.

A decline that is a repricing of expectations can reverse as fast as it arrived, because nothing needs to be repaired, the 2020 episode fell a third and was recovered within months. A decline that follows a credit event has to wait for balance sheets, which takes years. And the 1929–1932 decline was not one event at all but a sequence, which is why its recovery is measured in decades rather than years.

That is the argument for reading the cause rather than the number. A percentage tells you what has happened; the mechanism is the only thing that carries any information about what happens after, and it is knowable at the time in a way that a bottom is not.

Frequently asked questions

What is a bear market?

By the most widely used convention, a decline of twenty per cent or more from a peak. The threshold is a round number with no analytical basis — nothing changes at nineteen per cent — and it survives because it is convenient rather than because it identifies anything. What actually distinguishes the episodes on this page is duration: they unfold over quarters and years, which changes how every measure on this site behaves through them.

How is that different from a crash?

A crash is a decline compressed into days or weeks, and it is a volume event: forced selling arrives all at once, breadth collapses, and every extreme reading in this reference appears at the same time. A long decline is mostly made of ordinary sessions. It can fall further in total and never produce a single session that looks remarkable, which is exactly what makes it harder to recognise while it is happening.

Do the two overlap?

Frequently, and the 2007–2009 decline is the clearest case: a seventeen-month bear market containing several weeks that were crashes by any definition. The useful way to hold the distinction is that a crash is a description of a few sessions and a bear market is a description of a period, so one can contain the other. The 1987 crash is the counter-example, a severe crash that was not followed by a long decline at all.

What does breadth do through a long decline?

It deteriorates steadily rather than collapsing, which is both the useful part and the difficult part. Advance/decline data typically weakens well before the index peak and continues weakening through the decline, so the signal is early and stays true for a very long time. That is genuinely informative about the character of the market and close to useless for timing, a divergence that has been correct for eighteen months is indistinguishable at the time from one that will be wrong for another eighteen.

And volume?

It rises in the sharp phases and is unremarkable through the rest, which is why a long decline has no consistent volume signature. Some of the largest bear markets contain long stretches of perfectly ordinary volume, and any rule expecting heavy trade to confirm a decline will miss them. The heavy sessions cluster in the crash-like phases inside the decline, and often near its end, which is the pattern the volume-at-turns page sets out.

Is a twenty per cent decline a useful definition at all?

As a label for talking about the past, yes; as a threshold, no. It is measured from a peak that is only identifiable afterwards, so a market is never in a bear market at the time; it becomes one retrospectively once the peak is known. Two people using the same definition on the same data will also disagree about start dates depending on whether they use closing prices or intraday extremes, and on which index.

Why do the recovery times differ so much?

Because the causes differ. A decline driven by a repricing of expectations can reverse as quickly as it arrived, the 2020 episode recovered within months. A decline that follows a credit event has to wait for balance sheets to be repaired, which takes years, and the 1929–1932 decline required more than two decades to be fully recovered in index terms. Depth and duration of recovery are only loosely related, and the mechanism is what separates them.

What can be measured while one is happening?

The state of the market rather than its remaining duration: how many issues are participating, whether new lows are expanding, whether the sharp sessions are appearing in clusters. All of those are honest descriptions and none of them dates the end. This is the point at which this reference stops, because a claim about when a decline finishes is the kind that the data does not support and that a static page cannot responsibly carry.

Where to go next