Every Nasdaq-100 drawdown since 1999

Last updated: 2026-08-20

The short answer

Since QQQ began trading on 10 March 1999, the Nasdaq-100 has fallen 10% or more from an all-time high 19 times. That is roughly once every 17 months.

Most were forgettable. Thirteen of the nineteen were corrections of 10-15%, and the median one took 28 days to get back to its prior high after bottoming.

The tail is what matters. Only two drawdowns exceeded 30%, and both were career-defining:

• The dot-com crash (2000-2002) fell 83.0% and took 16 years and 5 months from peak to full recovery. QQQ did not close above its 27 March 2000 high of $117.80 until 6 September 2016.
• The 2022 bear market fell 35.6% over 13 months and recovered in a further 12. Just over two years, peak to peak.

The COVID crash, for all its violence, was the third-deepest at 28.6% and took just 107 days peak to recovery.

Full reference table

All 19 Nasdaq-100 (QQQ) drawdowns of 10% or more, March 1999 to August 2026.

Durations are calendar days. "Decline" is peak to trough. "Recovery" is trough to the first close above the prior peak.

By severity

The distribution is the finding. Drawdowns under 20% resolve in weeks to months with striking consistency. Past 30%, the recovery clock changes units, from months to years.

Notable episodes in context

Dot-com (2000-2002)

The outlier that anchors every long-run statistic. An 83.0% decline over 2 years and 7 months, followed by a 13.9-year climb back. Peak to full recovery: 6,007 calendar days. Any study of Nasdaq drawdowns that includes this episode will be dominated by it, and any study that excludes it is understating tail risk. It belongs in the table with a warning label, not in the average.

Q4 2018 (-23.2%)

A four-month decline ending on Christmas Eve, recovered in a near-symmetrical 114 days. The cleanest example in the record of a severe drawdown that was, in hindsight, a non-event.

COVID (-28.6%)

The fastest severe drawdown on record here: 26 calendar days from peak to closing trough. Note the split between the closing low (16 March 2020, $169.30) and the intraday low (23 March 2020, $164.93). Most published accounts cite 23 March, which is correct on an intraday basis but not on a closing basis.

2022 (-35.6%)

The only other bear market in the record. Notable for its shape: a slow 13-month grind down rather than a crash, then a 12-month recovery. Duration, not depth, was what made it hard to sit through.

Methodology

Reproducible in full. Every number above comes from one query against raw daily OHLCV data.

Instrument. Invesco QQQ Trust (QQQ), the longest continuously traded Nasdaq-100 proxy, from its inception on 10 March 1999 through 17 August 2026.

Drawdown definition. For each trading day, the running all-time-high closing price is computed from inception. An episode is the set of consecutive days sharing the same running peak. It begins the day after a new all-time high and ends when the next all-time high is set. Depth is the lowest closing price within the episode measured against that peak close.

Threshold. Episodes reaching -10% or deeper on a closing basis are reported. Shallower episodes are excluded.

Durations. Calendar days, not trading sessions. Decline is peak date to closing-trough date. Recovery is closing-trough date to the first subsequent close above the prior peak.

Intraday depth uses the lowest intraday low within the episode against the peak close. It can therefore fall on a different date than the closing trough, as it does for COVID.

The SQL

The full query. It runs in under a second and can be verified against any daily QQQ price series.

WITH px AS (
  SELECT date, close, low
  FROM shibui.stock_quotes
  WHERE ticker = 'QQQ' AND date >= DATE '1999-01-01'
),
runmax AS (
  SELECT date, close, low,
    MAX(close) OVER (ORDER BY date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS peak_close,
    arg_max(date, close) OVER (ORDER BY date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS peak_date
  FROM px
),
episodes AS (
  SELECT peak_date,
    MAX(peak_close)      AS peak_close,
    MIN(close)           AS trough_close,
    arg_min(date, close) AS trough_date,
    MIN(low)             AS intraday_low
  FROM runmax
  GROUP BY peak_date
),
seq AS (
  SELECT *, LEAD(peak_date) OVER (ORDER BY peak_date) AS recovery_date
  FROM episodes
)
SELECT
  peak_date, trough_date,
  ROUND((trough_close / NULLIF(peak_close, 0) - 1) * 100, 1) AS depth_close_pct,
  ROUND((intraday_low  / NULLIF(peak_close, 0) - 1) * 100, 1) AS depth_intraday_pct,
  (trough_date - peak_date)     AS decline_cal_days,
  (recovery_date - trough_date) AS recovery_cal_days,
  recovery_date
FROM seq
WHERE (trough_close / NULLIF(peak_close, 0) - 1) <= -0.10
ORDER BY peak_date;

Limitations

Stated plainly, because a reference table is only useful if you know where it breaks.

1. Price return, not total return. QQQ closing prices are split-adjusted but not dividend-adjusted. On a total-return basis every recovery is somewhat faster than shown. The effect is negligible for a three-week correction and material for the 13.9-year dot-com recovery. QQQ's distribution yield over the period was low but non-zero, so reinvestment would have pulled the recovery date meaningfully earlier. We have not computed the total-return version and do not quantify it here.
2. QQQ is a proxy, not the index. The fund tracks the Nasdaq-100 with small tracking error and a 0.20% expense ratio. Index-level figures will differ marginally. QQQ was chosen because it is what an investor could actually have held.
3. The record starts in 1999. QQQ did not exist for the 1987 crash, the 1990 recession, or the 1998 LTCM episode. A 27-year sample containing two bear markets is a small sample for tail estimation.
4. The 10% threshold is a convention, not a natural boundary. An episode bottoming at -9.8% is excluded and one at -10.1% is included, though they are the same event in substance.
5. Overlapping episodes are not separately counted. A decline that briefly recovers to a new high before falling again registers as two episodes, not one. Episodes 1 and 2 (April-May 1999) and 8 and 9 (early 2018) are examples. Readers who consider those single events should merge them.
6. Nothing here is predictive. Median recovery times describe 19 historical episodes. They are not an estimate of how long the current drawdown will last, and the sample is far too small to support that use.

Frequently asked questions

How many Nasdaq-100 drawdowns have there been?

Since QQQ began trading on 10 March 1999, the Nasdaq-100 has experienced 19 drawdowns of 10% or more on a closing basis. That is roughly once every 17 months.

How long did the dot-com crash take to recover?

The dot-com crash peaked on 27 March 2000 and fell 83.0% to a closing trough on 9 October 2002. QQQ did not close above its prior high of $117.80 until 6 September 2016, a total recovery time of 5,081 calendar days (about 13 years and 11 months from the trough).

How long does a typical Nasdaq-100 correction take to recover?

The median 10-15% correction recovered in 28 calendar days from trough to new all-time high. The worst took 91 days. Only drawdowns past 30% changed the recovery clock from months to years.

Can I reproduce this data?

Yes. Connect the free Shibui Finance MCP server to Claude and ask it to build a complete table of every Nasdaq-100 drawdown since 1999. The full SQL is published above and runs in under a second.

Is this financial advice?

No. This is a historical reference table built from raw market data. Past performance does not predict future results.

Cite this

Free to use with attribution.

Shibui Finance, Every Nasdaq-100 Drawdown Since 1999: Depth, Duration, and Recovery Time, August 2026. https://shibui.finance/nasdaq-100-drawdown-history

Charts and tables may be reproduced with a link to this page. If you need a different threshold, a different index, or the same analysis on total-return data, email chris@shibui.finance and we will run it.

Reproduce it yourself

Every figure on this page came from a single natural-language question asked of Claude with the Shibui Finance connector attached. No spreadsheet, no Python, no data download.