# Historical drawdowns: crashes, corrections and pullbacks

Research snapshot: 15 September 2026. Requested observation cutoff: **2026-09-14**. This is an editorial selection of **28 benchmark episodes across 14 markets**, not a definitive popularity ranking, independent global events, or a representative sample of all market outcomes. Several crises are included in different countries to expose their different local recoveries.

## Selection coverage and qualifiers

The original 20 cases all lost more than 33%: geographically varied, but biased toward severe disasters. Eight complementary cases add brief political-event pullbacks (UK 2016, India 2024), 10–20% corrections (US February and Q4 2018, Australia 2018), sharp 20%+ reversals recovering within a year (India 2004, Japan 2013), and the slower 2022 US inflation/rate-hike bear market. Existing cases retain banking, currency, technology-bubble, pandemic, stagflation and prolonged/unrecovered outcomes.

The picker shows **Drop (%) | Calendar months to bottom | Country · benchmark | Event name**, grouped by retrospective depth: pullback below 10%, correction from 10% to below 20%, bear market at least 20%. These are descriptive display bands, not diagnoses or causal classifications. A severe bear market can be short; a correction can take months. The [Yardeni/S&P/Haver table](https://archive.yardeni.com/pub/sp500corrbeartables.pdf) documents the conventional 10% and 20% thresholds and corroborates the 2018 and 2022 US declines.

**Length means peak to ultimate bottom, not time to recovery.** Recovery from the bottom and the complete peak-to-recovery duration remain separate. Sub-month picker values use two decimal places so a one-day decline does not misleadingly read 0.0 months; selected details show days.

This improves the range of shapes but is still not statistically representative: the selection overweights severe, famous and data-available outcomes, underrepresents Africa and some other regions, and contains overlapping episodes. It omits small everyday fluctuations and non-equity markets. No probabilities, category frequencies or expected future returns should be inferred from it. All types are known only with hindsight; an early 5% loss can either recover or become much deeper.

Additional event-context URLs are retained per new record in contextSources (distinct from the daily data source). RBI discusses the 2004 election turbulence and distinguishes closing from intraday losses; BIS covers the 2013 market reversal; RBA covers Brexit, 2018 volatility and Australian equities; Akashvani reports the June 2024 Sensex closing loss. Exact peak, trough and recovery calculations always use each record's linked daily-close data, not an event headline or a different index.

## Results

**All duration columns below are calendar months**, calculated as elapsed calendar days ÷ 30.436875. Values below one month use two decimals; others use one. Exact days, source levels and normalized paths are in [crash-data.json](crash-data.json).

| Episode | Market / source benchmark | Pre-crash peak close | Lowest close | Fall | Peak → bottom (months) | Bottom → recovery (months) | Peak → recovery (months) | First recovery / last available date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1929 · Wall Street crash | United States · [S&P composite predecessor](https://finance.yahoo.com/quote/%5EGSPC/history/) | 1929-09-07 | 1932-06-01 | −86.22% | 32.8 | 267.7 | 300.5 | 1954-09-22 |
| 1973 · Oil shock & stagflation | United States · [S&P 500](https://finance.yahoo.com/quote/%5EGSPC/history/) | 1973-01-11 | 1974-10-03 | −48.20% | 20.7 | 69.5 | 90.2 | 1980-07-17 |
| 1987 · Black Monday | United States · [S&P 500](https://finance.yahoo.com/quote/%5EGSPC/history/) | 1987-08-25 | 1987-12-04 | −33.51% | 3.3 | 19.7 | 23.0 | 1989-07-26 |
| 1987 · Hong Kong market crash | Hong Kong · [Hang Seng](https://finance.yahoo.com/quote/%5EHSI/history/) | 1987-10-01 | 1987-12-07 | −52.02% | 2.2 | 43.3 | 45.5 | 1991-07-16 |
| 1989 · Japanese asset bubble | Japan · [Nikkei 225](https://finance.yahoo.com/quote/%5EN225/history/) | 1989-12-29 | 2009-03-10 | −81.87% | 230.3 | 179.5 | 409.8 | 2024-02-22 |
| 1994 · Mexican peso crisis | Mexico · [IPC](https://finance.yahoo.com/quote/%5EMXX/history/) | 1994-02-08 | 1995-02-27 | −49.38% | 12.6 | 10.2 | 22.8 | 1996-01-02 |
| 1997 · Asian crisis in Korea | South Korea · [KOSPI](https://finance.yahoo.com/quote/%5EKS11/history/) | 1997-06-17 | 1998-06-16 | −64.66% | 12.0 | 10.3 | 22.3 | 1999-04-27 |
| 1997 · Asian crisis in Thailand | Thailand · [SET Index](https://finance.yahoo.com/quote/%5ESET.BK/history/) | 1997-01-22 | 1998-09-04 | −75.87% | 19.4 | 106.3 | 125.6 | 2007-07-13 |
| 1997 · Asian crisis in Indonesia | Indonesia · [Jakarta Composite](https://finance.yahoo.com/quote/%5EJKSE/history/) | 1997-07-08 | 1998-09-21 | −65.33% | 14.5 | 63.6 | 78.1 | 2004-01-09 |
| 1999 · Greek bubble & debt crisis | Greece · [ATHEX Composite](https://finance.yahoo.com/quote/GD.AT/history/) | 1999-09-17 | 2016-02-11 | −93.06% | 196.8 | Not recovered | ≥ 322.0 | Not recovered through 2026-07-17 |
| 2000 · Dot-com bubble | United States · [Nasdaq Composite](https://finance.yahoo.com/quote/%5EIXIC/history/) | 2000-03-10 | 2002-10-09 | −77.93% | 31.0 | 150.4 | 181.4 | 2015-04-23 |
| 1999 · UK millennium peak | United Kingdom · [FTSE 100](https://finance.yahoo.com/quote/%5EFTSE/history/) | 1999-12-30 | 2003-03-12 | −52.57% | 38.4 | 143.5 | 181.9 | 2015-02-24 |
| 2000 · French dot-com bubble | France · [CAC 40](https://finance.yahoo.com/quote/%5EFCHI/history/) | 2000-09-04 | 2003-03-12 | −65.29% | 30.2 | 223.7 | 253.9 | 2021-11-02 |
| 2007 · Chinese equity bubble | China · [Shanghai Composite](https://finance.yahoo.com/quote/000001.SS/history/) | 2007-10-16 | 2008-11-04 | −71.98% | 12.6 | Not recovered | ≥ 227.0 | Not recovered through 2026-09-14 |
| 2007 · Global financial crisis | United States · [S&P 500](https://finance.yahoo.com/quote/%5EGSPC/history/) | 2007-10-09 | 2009-03-09 | −56.78% | 17.0 | 48.6 | 65.6 | 2013-03-28 |
| 2008 · Global crisis in India | India · [BSE Sensex](https://finance.yahoo.com/quote/%5EBSESN/history/) | 2008-01-08 | 2009-03-09 | −60.91% | 14.0 | 19.9 | 33.9 | 2010-11-04 |
| 2008 · Global crisis in Brazil | Brazil · [Ibovespa](https://finance.yahoo.com/quote/%5EBVSP/history/) **TR** | 2008-05-20 | 2008-10-27 | −59.96% | 5.3 | 106.5 | 111.7 | 2017-09-11 |
| 2015 · Chinese bubble & long slump | China · [Shanghai Composite](https://finance.yahoo.com/quote/000001.SS/history/) | 2015-06-12 | 2019-01-03 | −52.30% | 42.7 | Not recovered | ≥ 135.1 | Not recovered through 2026-09-14 |
| 2020 · COVID shock in Australia | Australia · [All Ordinaries](https://finance.yahoo.com/quote/%5EAORD/history/) | 2020-02-20 | 2020-03-23 | −37.09% | 1.1 | 12.7 | 13.8 | 2021-04-14 |
| 2020 · COVID shock | United States · [S&P 500](https://finance.yahoo.com/quote/%5EGSPC/history/) | 2020-02-19 | 2020-03-23 | −33.92% | 1.1 | 4.9 | 5.9 | 2020-08-18 |
| 2004 · Election uncertainty | India · [BSE Sensex](https://finance.yahoo.com/quote/%5EBSESN/history/) | 2004-04-23 | 2004-05-17 | −23.97% | 0.79 | 5.7 | 6.5 | 2004-11-08 |
| 2013 · Nikkei spring sell-off | Japan · [Nikkei 225](https://finance.yahoo.com/quote/%5EN225/history/) | 2013-05-22 | 2013-06-13 | −20.36% | 0.72 | 5.5 | 6.2 | 2013-11-28 |
| 2016 · Brexit vote shock | United Kingdom · [FTSE 100](https://finance.yahoo.com/quote/%5EFTSE/history/) | 2016-06-23 | 2016-06-27 | −5.62% | 0.13 | 0.07 | 0.20 | 2016-06-29 |
| 2018 · February volatility correction | United States · [S&P 500](https://finance.yahoo.com/quote/%5EGSPC/history/) | 2018-01-26 | 2018-02-08 | −10.16% | 0.43 | 6.5 | 6.9 | 2018-08-24 |
| 2018 · Fourth-quarter correction | United States · [S&P 500](https://finance.yahoo.com/quote/%5EGSPC/history/) | 2018-09-20 | 2018-12-24 | −19.78% | 3.1 | 3.9 | 7.1 | 2019-04-23 |
| 2018 · Australian market correction | Australia · [All Ordinaries](https://finance.yahoo.com/quote/%5EAORD/history/) | 2018-08-30 | 2018-12-21 | −14.35% | 3.7 | 4.1 | 7.8 | 2019-04-24 |
| 2022 · Inflation & rate-hike bear market | United States · [S&P 500](https://finance.yahoo.com/quote/%5EGSPC/history/) | 2022-01-03 | 2022-10-12 | −25.43% | 9.3 | 15.2 | 24.5 | 2024-01-19 |
| 2024 · Election-result pullback | India · [BSE Sensex](https://finance.yahoo.com/quote/%5EBSESN/history/) | 2024-06-03 | 2024-06-04 | −5.74% | 0.03 | 0.10 | 0.13 | 2024-06-07 |

TR = total return including distributions. All other entries use price indices excluding dividends. An unrecovered episode's elapsed time is a lower bound, **not an estimate of when recovery will happen**.

## Definitions and reproducibility

1. Each benchmark's main source is Yahoo Finance's chart service, using the unadjusted daily **close** field, not intraday high/low or adjusted close. The missing Saturday 7 September 1929 peak is supplied from Yardeni's S&P/Haver historical table, explicitly recorded in each affected record's sourceSupplements. Index return basis still depends on the index: Ibovespa itself includes distributions. The table links the main provider's historical-data page. JSON records the exact API URL, raw-response SHA-256 and any sourced supplements for every episode.
2. The script explicitly defines the pre-crisis search window for each episode, then finds its maximum closing level. These are selected local pre-crisis peaks, not necessarily all-time highs. It scans all later available closes for the first close at or above that level.
3. The bottom is the **lowest close over the entire interval until that first recovery**, or through the last available observation when not recovered. It is not necessarily the bottom of the initial sell-off. Japan includes the 2008–09 crisis; Greece includes the debt crisis; the 2015 China episode's lowest close is in 2019. These are full underwater cycles, not continuous declines or causal claims about a single event.
4. Recovery refers to the benchmark, not an investor's personal break-even. Contributions, selling, reinvestment, fees, taxes, distributions, inflation and FX exposure change investor outcomes. An index's constituents also change over time; it is not a fixed basket of the original companies.
5. Exchange-local dates are used. Daily values are validated as finite and positive. Nontrading days and missing quotes are not invented. The source's last valid date is retained separately from the requested cutoff. **Greece's source currently ends on 17 July 2026**, so its status is not asserted through September. Other unrecovered entries have observations through 14 September 2026.
6. Cycles of up to 366 calendar days retain every available daily close. Longer cycles retain the last close in each seven-day bin and the exact peak, ultimate trough and first recovery (or last observation). Intermediate values are linearly interpolated; this is not tick-by-tick or exact trade execution. Daily statistics are computed before sampling. Weekly sampling can miss other local highs/lows, so portfolio drawdown is approximate. The sampling field identifies each record's resolution.
7. Run `node scripts/research-crashes.mjs` to regenerate the JSON, browser module and this report. Raw daily responses are cached in ignored `tmp/crash-research/`; cached runs reuse the captured snapshot. Do not silently refresh a published snapshot: review provider revisions, update the cutoff deliberately and rerun tests. The saved normalized data allows the site to work without a live third-party API.

## Source cross-checks and apparent discrepancies

The calculations for every row come from its linked daily dataset. The following additional publications were used to check definitions, benchmark identity or selected landmarks; they do not independently verify every observation in all 28 episodes.

- **1929:** [Federal Reserve History](https://www.federalreservehistory.org/essays/stock-market-crash-of-1929) describes the **Dow**, a different benchmark. Here we use the legacy **S&P composite predecessor** (the S&P 500 did not exist in 1929). Yahoo omits Saturday sessions and would select 16 September at 31.86. [Ed Yardeni's Appendix 15.3, sourced to S&P/Haver](https://www.yardenibook.com/pub/15.3appendix.pdf), documents the actual 7 September peak at **31.92** and the 1 June 1932 low at **4.40**. The script adds that missing peak observation, then recomputes the decline, recovery search and normalized path. Pre-launch series are historical reconstructions, not a then-investable S&P 500 fund. Other omitted Saturday fluctuations remain a limitation of the sampled replay.
- **Japan:** [Nikkei's own November 2012 report](https://indexes.nikkei.co.jp/en/nkave/archives/news/20121101E_1.pdf) gives the 1989 peak of 38,915.87 and March 2009 low of 7,054.98. [AP's 22 February 2024 report](https://apnews.com/article/93003abeedbc99cf8f3dab78de78c298) reports the return above the old record. The roughly 34-year duration is the full peak-to-recovery cycle, not 34 years from the bottom.
- **Korea:** [KEPCO's SEC-filed 2005 Form 20-F](https://www.sec.gov/Archives/edgar/data/887225/000119312506139957/d20f.htm) lists the KOSPI's 1997 high of 792.29 and 1998 low of 280.00. The episode starts at the 1997 local high; using 1994 instead would produce a different recovery duration. Thailand likewise uses its 1997 local peak. [IMF Asian-crisis research](https://www.imf.org/external/pubs/ft/wp/1999/wp9928.pdf) supplies regional context, not these exact return calculations.
- **Nasdaq:** The [Nasdaq Composite series distributed by FRED](https://fred.stlouisfed.org/data/NASDAQCOM) records 5,048.62 on 10 March 2000 and 5,056.06 on 23 April 2015, corroborating the recovered closing peak.
- **UK and France:** [Reuters' FTSE record report](https://www.lse.co.uk/news/ftse-100-sets-new-record-high-passing-december-1999-peak-t664dk4z7eweyc2.html) identifies 24 February 2015 as the FTSE's return past its 1999 closing record. [Reuters' French-market report](https://www.nippon.com/en/news/reu20211102KBN2HN05P/) records the CAC 40 closing record on 2 November 2021. Intraday records can be reached on different dates and are not used here.
- **China:** The [Shanghai statistical yearbook](https://tjj.sh.gov.cn/tjnj/2015tjnje/E1812.htm) lists the widely cited **intraday** 2008 low, 1,664.93 on 28 October. This report uses the lowest **closing** level, approximately 1,706.70 on 4 November; therefore its trough date and percentage fall differ. The 2007 and 2015 starting peaks define overlapping, not independent, underwater cycles.
- **Brazil:** [B3's Ibovespa methodology](https://www.b3.com.br/data/files/51/85/B6/30/7F6947102255C247AC094EA8/IBOV-Metodologia-en-us__Modelo_Novo_.pdf) explicitly defines a total-return index. [B3's 11 September 2017 record announcement](https://www.b3.com.br/pt_br/noticias/recorde-historico.htm) identifies the prior 20 May 2008 closing record. B3 reports 73,516 versus the provider's rounded 73,517; the app consistently uses its source series. Do not compare this as though dividends were excluded.
- **Australia:** The benchmark is **All Ordinaries**, not ASX 200. [ASX's historical statistics](https://www.asx.com.au/about/market-statistics/historical-market-statistics) distinguish the indices. Their peak and recovery dates need not match.

## How the simulator uses the research

**“A crash in 3 months” means onset exactly 3 months from now**, not the start of the third month, a window of possible dates, or a three-month decline. The historical peak is placed at that future time. The benchmark path then supplies the decline duration and the recovery duration without compression. Before onset, the custom growth assumption applies. Only daily closes are researched: intraday-only crashes and intraday extremes are excluded.

Monthly contributions and scheduled sales occur at month end. Cash re-entry is independent of crash onset and historical duration: the illustrative default waits **3 additional months after onset**, editable in assumptions. Thus a crash starting at month 3 initially has cash re-entry at month 6, whether or not the market has bottomed. This is not an estimate or recommendation of when to buy back. The same schedule applies in the no-crash world. Historical mode supplies the crash depth; switching back restores the user's custom one-off-drop depth. The old crash-window/unknown-month averaging controls have been removed because the premise now specifies an exact onset.

“Look ahead” can show a shorter interval or the entire observed cycle. The comparison never extends past the first historical recovery or the last available observation. The final interval can end between month ends: it earns proportional savings interest but no uncompleted monthly deposit. The no-crash comparison uses the user's assumed constant growth over the same horizon, not a second historical dataset. Long horizons can make this assumption dominate the comparison.

Portfolio values are displayed in euros by applying normalized benchmark percentage changes. **This is not a currency-converted historical euro backtest.** There are no taxes, fees, inflation adjustments or FX returns. No crash probability or investment recommendation is inferred from these 28 cases; they are stress tests with hindsight, not forecasts.

Moving out of stocks means moving into **interest-bearing savings**. The rate is constant by default. Where matching data exists, an optional mode applies historical savings/deposit-rate changes to the user's starting rate. See [savings research and retained source URLs](SAVINGS.md) for definitions, proxies, coverage and the post-coverage constant-rate assumption.

## Publication caveat

This local research snapshot does not establish commercial redistribution rights. Before publishing an advertising-funded site, confirm the data provider's and index owners' permitted display/redistribution terms and use an appropriately licensed dataset if required. Keep the research date and benchmark return-basis notices visible.