Key Takeaways
- Peak-to-trough declines in this library range from 21.9 percent (Dow Jones Industrial Average, 2022) to 89 percent (Dow, 1929 to 1932). There is no representative number.
- Recovery to the prior peak ranged from 181 days (S&P 500, 2020) to just over twenty-five years (Dow, 1929 to 1954).
- Speed and depth are unrelated. The fastest decline (23 trading days in 2020) and the largest single day (22.6 percent in 1987) both produced drawdowns of roughly a third, the same order as episodes that took two years.
- The most-cited advance warning, an inverted Treasury yield curve, fired usefully before one of these six, fired far too early before another, fired after the top in a third, did not fire at all before a fourth, and correctly preceded a recession caused by something it had no relationship to in a fifth.
- In four of the six, the central bank was cutting rates during the decline. In 1929 and 2022 it was raising or had recently raised. Those two episodes look very different from the others, and the difference is inflation.
- The recurring lesson is not a pattern to recognize. It is that recovery time is a planning input, and that leverage and concentration decide how bad any given decline is for a specific holder.
How the Six Compare
The table below is the core of this page. Every figure comes from the sources listed at the bottom, and every one is stated at the same measurement basis: close to close, price only, no dividends.
Peak-to-trough decline and time to recover the prior peak. Index figures computed from daily closing values; 1929 figures from Federal Reserve records.
| Episode | Index | Peak | Trough | Decline | Recovered | Peak to recovery |
|---|---|---|---|---|---|---|
| 1929 crash | Dow Jones Industrial Average | 3 Sep 1929 | 8 Jul 1932 | 89% | 23 Nov 1954 | About 25 years |
| Black Monday 1987 | S&P 500 | 25 Aug 1987 | 4 Dec 1987 | 33.5% | 26 Jul 1989 | 1 year 11 months |
| Dot-com bubble | Nasdaq Composite | 10 Mar 2000 | 9 Oct 2002 | 77.9% | 23 Apr 2015 | 15 years 1 month |
| Dot-com bubble | S&P 500 | 24 Mar 2000 | 9 Oct 2002 | 49.1% | 30 May 2007 | 7 years 2 months |
| 2008 financial crisis | S&P 500 | 9 Oct 2007 | 9 Mar 2009 | 56.8% | 28 Mar 2013 | 5 years 5 months |
| 2020 COVID crash | S&P 500 | 19 Feb 2020 | 23 Mar 2020 | 33.9% | 18 Aug 2020 | 6 months |
| 2022 rate shock | S&P 500 | 3 Jan 2022 | 12 Oct 2022 | 25.4% | 19 Jan 2024 | 2 years |
| 2022 rate shock | Dow Jones Industrial Average | 4 Jan 2022 | 30 Sep 2022 | 21.9% | 13 Dec 2023 | 1 year 11 months |
Read down the decline column and then down the recovery column. A 33.5 percent decline in 1987 took under two years to recover. A 33.9 percent decline in 2020 took six months. A 25.4 percent decline in 2022 took two years. The depth of a drawdown is a weak predictor of how long recovery takes, because recovery depends on what caused the decline and on what the surrounding economy and policy environment did next.
Speed of decline
Trading sessions from peak close to trough close, and the worst single session of each episode. Computed from daily closing values; 1929 and 1987 single-day Dow figures from Federal Reserve records.
| Episode | Trading days peak to trough | Worst single session |
|---|---|---|
| 2020 COVID crash | 23 | S&P 500 down 11.98% on 16 Mar 2020 |
| Black Monday 1987 | 71 | Dow down 22.6% on 19 Oct 1987; S&P 500 down 20.47% |
| 2022 rate shock | 195 | No session worse than about 4% |
| 2008 financial crisis | 355 | S&P 500 down 9.03% on 15 Oct 2008 |
| Dot-com bubble (S&P 500) | 637 | Spread across three consecutive down years |
| 1929 crash | About 2 years 10 months | Dow down nearly 13% on 28 Oct 1929 |
Did the Warning Signals Work?
The single most useful thing this library can offer is an honest record of how the standard warning indicators performed. The yield curve is the most cited of them, so here is what it actually did before each episode, computed from the Treasury Department's published daily series.
Behavior of the 3-month to 10-year Treasury yield curve ahead of each modern episode. Computed from the Treasury Department's published daily par yield curve series.
| Episode | What the curve did | Verdict |
|---|---|---|
| Dot-com bubble | No inversion during 1999. First inverted close of 2000 was 7 April, nearly a month after the Nasdaq Composite peak. | Fired after the top. Not a warning. |
| 2008 financial crisis | Inverted on 124 of 250 trading days in 2006 and 119 of 251 in 2007, last inverted close 27 August 2007. | Correct direction, roughly two years early, and the inversion ended before the peak. |
| 2020 COVID crash | Inverted on 104 of 250 trading days in 2019, from 22 March to 10 October. | Preceded a recession caused by a pandemic it had no relationship to. |
| 2022 rate shock | No inversion at any point during 2021. | Silent throughout the setup. |
An indicator with that record is not useless, but it is context rather than a trigger. It tells you something about the shape of the rate environment. It does not tell you when to act, and in two of the four cases above it would have told you nothing at all. Treating it as a timing signal requires ignoring most of its own history. The mechanics are covered in the yield curve, term premium and recession signals.
What was actually available in advance, in each case
The most usable piece of advance information in each episode, and what it did and did not tell you.
| Episode | Genuinely available in advance | Genuinely unavailable |
|---|---|---|
| 1929 | Margin buying at about 10 percent down was public and was the stated reason the Federal Reserve was tightening. | That the decline would run for another 32 months after the famous October days. |
| 1987 | Portfolio insurance existed and was widely sold. | The aggregate notional running the same trigger, which is the number that determined the size of the cascade. |
| Dot-com | How concentrated the advance was, which was measurable at any time. | Which specific companies would survive the decade. |
| 2008 | Subprime lender failures from April 2007, falling house prices from mid-2006. | The size and interconnection of leveraged exposures to securitized credit. |
| 2020 | Reports of a novel virus in January 2020. | Both the epidemiological path and, separately and harder, the policy response that drove the recovery. |
| 2022 | The duration of your own bond fund, published, and inflation already at 7.0 percent for December 2021. | That the Federal Reserve would raise 425 basis points in twelve months. |
Look at the left-hand column. In every case, what was genuinely available was structural rather than predictive: how much leverage, how much concentration, how much duration. None of it says when. All of it says how bad, for you specifically, if it happens. That distinction is the most portable thing in this library.
What the Central Bank Did, and Why It Varied
Federal Reserve policy direction during each decline, from the Federal Reserve Board's records of open market operations.
| Episode | Policy direction during the decline | Inflation constraint |
|---|---|---|
| 1929 | Had been tightening into the peak; failed to act as lender of last resort during the banking panics that followed | Deflation, not inflation. The money supply fell nearly 30 percent from autumn 1930 to winter 1933. |
| 1987 | A statement of readiness to supply liquidity, issued the next morning. No rate cut on the day. | Not binding |
| Dot-com | Eleven cuts during 2001, from 6.50 percent to 1.75 percent, and the market fell throughout and for ten months afterwards | Not binding |
| 2008 | Ten cuts from 5.25 percent to a 0 to 0.25 percent range, then forward guidance | Not binding |
| 2020 | Two cuts in twelve days to the zero bound, plus at least 700 billion dollars of announced asset purchases | Not binding. Twelve-month inflation was running below 2 percent. |
| 2022 | Seven increases, from 0 to 0.25 percent to 4.25 to 4.50 percent | Binding. Twelve-month inflation peaked at 9.1 percent in June 2022. |
The pattern most investors carry, that a central bank will eventually support falling markets, holds in four of these six and fails in the two where inflation or deflation removed the option. It is a conditional expectation, not a rule, and the condition is stated in the right-hand column. Note also that even where the support arrived, it did not mark the bottom: in 2008 the funds rate reached zero three months before the equity trough, and in the dot-com episode eleven cuts preceded ten more months of decline.
The Six Case Studies
The 2008 Financial Crisis
A credit and funding crisis that began in United States housing finance. The S&P 500 fell 56.8 percent over 355 trading days and took until March 2013 to recover. The yield curve warned two years early and stopped warning before the peak. The genuine information gap was the interconnection of leveraged exposures, which no public disclosure allowed an outsider to size. Read this one for how leverage converts a loss into a forced liquidation.
The 2020 COVID Crash
The fastest fall of its depth in the daily record: 33.9 percent in 23 trading sessions, with recovery in six months. Unemployment went from 3.5 percent to 14.8 percent in two months and was still 6.7 percent in December, after the index had made new highs. Read this one for what a compressed decline does to decision-making, and for why its speed is the least generalizable thing in this library.
The 2022 Rate Shock
The shallowest equity decline here, and for many portfolios the most uncomfortable, because bonds fell at the same time. On calendar-year total return a long-dated Treasury fund lost more than the S&P 500. Read this one for the regime dependence of diversification, and for why the duration of your own bond holdings is arithmetic you can run today.
The 1929 Crash
The reference case for the tail. An 89 percent decline over almost three years, and a return to the prior peak only in November 1954. What turned a crash into a depression was a banking collapse and a nearly 30 percent contraction in the money supply, not the equity market. Read this one for the recovery arithmetic and for how much of the transmission chain has since been deliberately dismantled.
Black Monday 1987
The largest single-day decline on record, 22.6 percent in the Dow, followed by no recession at all. Portfolio insurance created a mechanical selling cascade; circuit breakers exist because of it. Read this one for why a large price move is not automatically information about the economy, and for what crowded identical rules do in aggregate.
The Dot-Com Bubble
A 77.9 percent fall in the Nasdaq Composite and a fifteen-year recovery, against 49.1 percent and seven years for the S&P 500 over the same window. The internet thesis was correct. Read this one for concentration risk, and for the gap between being right about a technology and being right about a price.
How to Read This Library Without Fooling Yourself
Historical market episodes are unusually easy to misuse, because the outcome is known and the narrative writes itself backwards. Four habits make the difference between learning something and manufacturing confidence.
Separate what was knowable from what is known now. Every case study here has an explicit section on this, because it is the failure mode. If a warning sign only becomes legible once you know the ending, it was not a warning sign. It was a detail.
Do not average the six. There is no useful mean of 21.9 percent and 89 percent, and no useful mean of six months and twenty-five years. The range is the finding. Any single number extracted from this set and presented as typical is discarding the only real information in it.
Ask what each episode says about your holdings, not about markets. The generalizable content is almost entirely about position characteristics: leverage, concentration, duration, and the horizon over which you can afford to wait. Those are measurable today, without any forecast.
Notice that the mechanism was different every time. Housing credit, a pandemic, an inflation shock, a banking collapse, a trading product, a valuation unwind. Pattern-matching a new decline to the most recent one has been wrong six times in a row. Cognitive biases in trading covers why the instinct is so persistent.
The questions each episode leaves you with
One answerable question per episode, requiring no forecast.
| Episode | Question it poses about your own position |
|---|---|
| 2008 | If these assets fell by half and stayed down five years, what would you be forced to do? |
| 2020 | Which decisions have you already made in advance, and which would you still be making mid-decline? |
| 2022 | What is the effective duration of your bond holdings, multiplied by a two percentage point yield rise? |
| 1929 | What is the deepest and longest decline your plan can absorb without forcing a decision you would regret? |
| 1987 | Which of your holdings would be sold by someone else's automatic rule during a fast session? |
| Dot-com | How much of your equity exposure sits in the largest handful of holdings, including inside every fund? |
None of those six questions requires predicting anything. All six are answerable from documents you already have. That is deliberate: it is the only part of market history that reliably transfers.
References
Every figure on this page was verified against the following sources, each retrieved on 23 August 2026. Individual case studies cite additional sources specific to their episode.
- Federal Reserve History: Stock Market Crash of 1929: the 381.17 peak on 3 September 1929, the 41.22 close on 8 July 1932 at 89 percent below the peak, the nearly 13 percent decline on 28 October 1929, the 23 November 1954 recovery, and the description of margin buying at about 10 percent down.
- Federal Reserve History: Stock Market Crash of 1987: the 508 point, 22.6 percent decline on 19 October 1987, the role of portfolio insurance, the 20 October Greenspan liquidity statement, and the subsequent introduction of circuit breakers.
- Federal Reserve History: The Great Recession: the December 2007 to June 2009 recession dates and the description of the funds rate path from 5.25 percent to a 0 to 0.25 percent range.
- Federal Reserve History: The Great Depression: the nearly 30 percent contraction in the money supply from the autumn of 1930 through the winter of 1933 and the characterization of the failure to act as lender of last resort.
- National Bureau of Economic Research: US Business Cycle Expansions and Contractions: every recession peak and trough date referenced here, including August 1929 to March 1933, March 2001 to November 2001, December 2007 to June 2009, and February 2020 to April 2020.
- Federal Reserve Board: Open Market Operations and Federal Reserve Board: Open Market Operations Archive: every federal funds target change referenced in the policy table.
- US Bureau of Labor Statistics: Consumer Price Index for All Urban Consumers, Series CUUR0000SA0: the published index levels from which the 9.1 percent June 2022 and 7.0 percent December 2021 twelve-month rates were computed.
- US Bureau of Labor Statistics: Unemployment Rate, Series LNS14000000: the unemployment readings quoted for 2020.
- US Department of the Treasury: Daily Treasury Par Yield Curve Rates: the published daily series from which every yield curve inversion count in the signals table was computed.
Figures deliberately not stated. This library does not state total-return recovery dates, inflation-adjusted recovery dates, or index valuation ratios at any peak, because no source verified in this session supplied them. Where the shape of an event is known but the number is not, the shape is described and the number is left out rather than estimated.
Frequently Asked Questions
What is the worst stock market crash in history?
By depth, the 1929 to 1932 decline. The Federal Reserve records the Dow Jones Industrial Average falling from 381.17 on 3 September 1929 to 41.22 on 8 July 1932, 89 percent below the peak, and not returning to its pre-crash level until 23 November 1954. By single-day magnitude, Black Monday on 19 October 1987, when the Dow fell 22.6 percent, which the Federal Reserve records as the largest one-day decline in history.
How long does the stock market usually take to recover from a crash?
There is no usual figure, and that is the finding rather than an evasion. Across the six episodes documented here, recovery to the prior peak ranged from 181 days for the S&P 500 after the 2020 low to just over twenty-five years for the Dow Jones Industrial Average after 1929. The dot-com decline took seven years for the S&P 500 and fifteen for the Nasdaq Composite, from the same period. Averaging that range destroys the only useful information in it.
Which market crash was the fastest?
The 2020 COVID crash. Computed from daily closes, the S&P 500 fell 33.9 percent in 23 trading sessions, from 3,386.15 on 19 February 2020 to 2,237.40 on 23 March 2020, and was already more than 20 percent below its peak after 16 sessions. Black Monday 1987 had a larger single day but its full drawdown took 71 sessions, and the 2008 decline took 355.
Does the yield curve reliably predict market crashes?
Its record across these episodes is mixed enough to make it context rather than a trigger. It inverted persistently through 2006 and 2007, roughly two years before the 2008 peak, and stopped inverting before that peak. It did not invert at all during 1999 and first inverted in 2000 nearly a month after the Nasdaq Composite top. It did not invert at any point during 2021 before the 2022 decline. It did invert in 2019 before a recession caused by a pandemic it had no relationship to.
Does the Federal Reserve always cut rates during a market crash?
No. In four of the six episodes here the Federal Reserve was easing or supplying liquidity during the decline. In 1929 it had been tightening into the peak and did not act as lender of last resort during the banking panics that followed. In 2022 it raised its target range seven times during the decline, because twelve-month inflation peaked at 9.1 percent in June and eased policy would have worked against that objective. Policy support is conditional on inflation, not automatic.
Do rate cuts stop a market decline?
Not reliably, and the dot-com episode is the clearest counterexample. The Federal Reserve cut its target rate eleven times during 2001, from 6.50 percent to 1.75 percent, and cut again in November 2002, while the Nasdaq Composite continued falling until 9 October 2002. In 2008 the funds rate reached its 0 to 0.25 percent range in December 2008 and the S&P 500 did not bottom until March 2009, about three months later. Policy action and price bottoms are separate events.
What do all market crashes have in common?
Less than most narratives suggest. The six documented here were caused by, respectively, margin leverage and a banking collapse, a mechanical trading product, a concentrated valuation unwind, a securitized credit failure, a pandemic, and an inflation shock. What repeats is not the mechanism but the consequences of position characteristics: leverage converts losses into forced selling, concentration determines the dispersion of outcomes, and the holder's horizon determines whether a drawdown is survivable.
Which crash was worst for a balanced portfolio?
Among these six, 2022 was distinctive because the bond allocation fell alongside equities rather than offsetting them. On calendar-year total return computed from dividend-adjusted prices, the S&P 500 returned negative 19.4 percent and a US aggregate bond index fund negative 13.0 percent, with a long-dated Treasury fund at negative 31.2 percent. A 60/40 blend with weights set at the start of the year returned about negative 16.9 percent, only about 2.5 percentage points better than equities alone.
Was the 2008 crash worse than 1929?
No. The S&P 500 fell 56.8 percent from October 2007 to March 2009 and recovered its peak in March 2013, about five and a half years. The Federal Reserve records the Dow falling 89 percent from September 1929 to July 1932 and not recovering until November 1954, about twenty-five years. The 1929 episode was also accompanied by a banking system collapse and a nearly 30 percent contraction in the money supply, which the Federal Reserve identifies as the decisive factor.
How much does a portfolio need to gain to recover from a big decline?
Recovery requirements rise disproportionately with depth. A 20 percent decline needs a 25 percent gain, a 33 percent decline needs 49 percent, a 50 percent decline needs 100 percent, a 57 percent decline needs 133 percent, a 78 percent decline needs 355 percent and an 89 percent decline needs about 809 percent. This arithmetic is why the Nasdaq Composite took fifteen years after its 77.9 percent fall while the S&P 500, after 49.1 percent, took seven.
Can I use market history to time the next crash?
The record in this library argues against it. Across six episodes the mechanism differed every time, the most-cited warning indicator worked usefully in at most one case, and the information that was genuinely available in advance was structural rather than predictive: how much leverage, how much concentration, how much duration. None of that says when. All of it says how bad, for a specific holder, if it happens. That is what these case studies are written to support.
What is hindsight bias in market history?
It is the tendency to see a past outcome as more predictable than it was, because knowing the ending makes the relevant details stand out from the irrelevant ones. In market history it usually appears as a warning sign that only becomes legible once the outcome is known, presented as though it were available at the time. Each case study in this library has an explicit section separating the two, because the alternative is a set of narratives that produce confidence rather than caution.