What is a stock market correction, and how long does it last?

A stock market correction is a fall of 10% to 20% from a recent high, measured on closing prices. A fall of 20% or more is a bear market. Corrections are common: US500, the large-cap US index, has had 13 of them since 1995, plus four bear markets. The typical correction took about three months to reach its low and about three more to get back to the old high. Bear markets usually took years.

The question came back last week. On Monday 2 November 2026 the tech-heavy US100 closed 10.36% below its 6 October record, a correction. On Thursday 5 November it fell 6.36%, its worst day since March 2020, and US500 fell 4.36%, its worst since April 2025. On Friday 6 November JP225, Japan’s main index, closed 20.63% below its June record: a bear market.

What is a stock market correction?

A correction is measured from the highest close before the fall, usually a record, to a later close. Traders use closes rather than intraday lows because the close is the price most funds value at and most data sets agree on. The thresholds are round numbers, not laws of nature, but they are the ones the market, the press and most research use:

• Pullback: 5% to 10%A dip that stops short of 10%. With the swing rule this page uses set at 5%, US500 had 48 from 1995 to 2026, counting the one under way in November: about three every two years.
• Correction: 10% to 20%The usual name for a real setback. US500 had 13 from 1995 to 2026, with a median fall of 13.3%.
• Bear market: 20% or moreA deep fall, often tied to a recession or a crisis. US500 had four: 2000 to 2002, 2007 to 2009, 2020 and 2022.

The arithmetic is lopsided on the way back. A 13.56% fall needs a 15.69% rise to get back to the high, a 20% fall needs 25% and a 50% fall needs 100%. That is why bear markets take so much longer to repair than they take to happen.

Correction vs bear market

Where markets stood on 6 November 2026

The same week put three big indices at three different stages. Here is each one’s distance from its record close at the end of every week since September:

The selling was concentrated in the US tech index and in Japan, where chip and AI stocks carry heavy weights. US100 crossed the 10% line on 2 November, three days before the worst of the fall, and stood 14.71% down at the 5 November close. JP225 had spent much of September in a correction from its June record and came back to 2.32% below it on 6 October, then fell again; its 5.96% drop on 6 November, the worst since April 2025, took it past 20%. European indices, US30 and AU200 stayed within 7% of their records.

How often do stock market corrections happen?

Over 31.6 years of US500 daily closes there were 17 falls of 10% or more, about one every 1.9 years, and a new one crossed the 10% line in 15 of the 32 calendar years from 1995 to 2026. They do not arrive on a schedule: 1998 had two, while almost five years passed between the crossings of January 2003 and November 2007. Every decline in the data:

To count them, a decline starts at a closing high and counts once the close is 10% below it. It ends at the lowest close before the index rises 10% from that low. A fall of 20% or more from a record is a bear market, and the smaller swings inside it, such as the rallies of 2001 and 2008, are part of it rather than events of their own. Other rules give slightly different lists; this one is in the script below.

The tech index corrects far more often. By the same rule US100 had 48 declines of 10% or more over the same years, about one every eight months, and 9 of them went past 20%. US100 stayed below its 2000 record until 2015, so the rule counts its 2007 to 2009 fall as six separate swings rather than one bear market. Three have crossed 10% in 2026 alone, in March, July and now November. Concentrated indices swing harder, which is also why their historical volatility runs higher.

How long does a stock market correction last?

A correction usually runs its course in months. The median US500 correction took 88 calendar days from the high to the lowest close and 84 days from there back to the old high; from high to recovery the median was 166 days, about five and a half months. The quickest fall took 13 days (January to February 2018) and the longest 157 (April to October 2011). Bear markets are a different scale:

The recovery took longer than the fall in 8 of the 13 corrections. The January 2018 correction bottomed in 13 days but took 197 more to regain the high, and the May 2015 correction took 321. These are price-index figures: dividends are left out, so an investor who reinvested them got back to even somewhat sooner.

What happened after the first close 10% down

The day an index first closes 10% below its high is the day headlines call it a correction. At that point nobody knows whether it will stop at 11% or run to 50%. Measured from that first close, here is what followed in the 17 US500 declines:

Four of the 13 corrections bottomed on the very day they crossed 10% (October 1997, October 1998, February 2018 and October 2023). The other side of the record is the 2000 and 2007 bear markets, where the index was still 12.1% and 39.5% lower a year after its first 10% close. Three months on, the picture was already mixed: the median gain after a first 10% close was 8.5%, but two of the 17 cases were lower.

Correction calculator

Find corrections in any index with Python

corrections.py pulls US500 daily bars from 1995 in three requests (a daily request can span at most 5,000 days), applies the rule above and prints every decline with its depth and duration. Change SYMBOL to US100 or DE40 to run it on another index; for JP225 set START to date(1990, 1, 1), since its 1990 high stood until 2024. It needs Python 3.9 or newer, the standard library and a TICKERLAYER_API_KEY environment variable.

"""Every 10% decline in an index since 1995: corrections, bear markets, depth and duration. Python 3.9+, standard library."""
import json
import os
import statistics
import sys
import urllib.request
from datetime import date, datetime, timedelta, timezone

API = "https://api.tickerlayer.com"
KEY = os.environ.get("TICKERLAYER_API_KEY") or sys.exit("Set TICKERLAYER_API_KEY first.")
SYMBOL, START, END = "US500", date(1995, 1, 1), date(2026, 11, 6)


def daily_closes():
    """Daily closes, oldest first, in windows of 4,900 days (a request may span at most 5,000)."""
    out, start = [], START
    while start <= END:
        stop = min(END, start + timedelta(days=4899))
        url = f"{API}/indices/agg/{SYMBOL}/1/day/{start}/{stop}?sort=asc&limit=5000"
        req = urllib.request.Request(url, headers={"x-api-key": KEY})
        with urllib.request.urlopen(req, timeout=30) as resp:
            out += [(datetime.fromtimestamp(b["t"] / 1000, timezone.utc).date(), b["c"]) for b in json.load(resp)["results"]]
        start = stop + timedelta(days=1)
    return out


def swings(closes, size=0.10):
    """(peak, trough) index pairs: a fall of `size` from a high, ended by a rise of `size` from the low."""
    found, high, low, falling = [], 0, 0, False
    for i, (_, c) in enumerate(closes):
        if not falling:
            if c > closes[high][1]:
                high = i
            elif c <= closes[high][1] * (1 - size):
                falling, low = True, i
        elif c < closes[low][1]:
            low = i
        elif c >= closes[low][1] * (1 + size):
            found.append((high, low))
            falling, high = False, i
    return found + ([(high, low)] if falling else [])


def bear_markets(closes):
    """Falls of 20% from a record close, to the lowest close before the record is regained."""
    found, peak, i = [], 0, 0
    while i < len(closes):
        if closes[i][1] >= closes[peak][1]:
            peak = i
        elif closes[i][1] <= closes[peak][1] * 0.8:
            low = j = i
            while j < len(closes) and closes[j][1] < closes[peak][1]:
                low = j if closes[j][1] < closes[low][1] else low
                j += 1
            found.append((peak, low))
            peak, i = j, j
        i += 1
    return found


closes = daily_closes()
bears = bear_markets(closes)
# A swing inside a bear market is part of it, not a correction of its own.
events = bears + [s for s in swings(closes) if not any(p <= s[0] and s[1] <= t for p, t in bears)]
print(f"{SYMBOL} daily closes {closes[0][0]} to {closes[-1][0]}: {len(closes)} sessions")
print(f"{'peak':<11} {'trough':<11} {'fall':>6} {'days down':>9} {'back to peak':>13}")
rows = []
for peak, trough in sorted(events):
    back = next((j for j in range(trough, len(closes)) if closes[j][1] >= closes[peak][1]), None)
    fall = (1 - closes[trough][1] / closes[peak][1]) * 100
    down = (closes[trough][0] - closes[peak][0]).days
    up = (closes[back][0] - closes[trough][0]).days if back is not None else None
    rows.append((fall, down, up))
    print(f"{closes[peak][0]!s:<11} {closes[trough][0]!s:<11} {fall:>5.1f}% {down:>9} {f'{up} days' if up is not None else 'not yet':>13}")
for name, group in (("corrections (10-20%)", [r for r in rows if r[0] < 20]), ("bear markets (20%+)", [r for r in rows if r[0] >= 20])):
    print(f"{name}: {len(group)}, median fall {statistics.median(r[0] for r in group):.1f}%, "
          f"median {statistics.median(r[1] for r in group):.0f} days down, "
          f"median {statistics.median(r[2] for r in group if r[2] is not None):.0f} days back to the peak")
high = max(range(len(closes)), key=lambda i: closes[i][1])
print(f"{closes[-1][0]}: {closes[-1][1]:,.2f}, {(1 - closes[-1][1] / closes[high][1]) * 100:.2f}% below the {closes[high][0]} record of {closes[high][1]:,.2f}")

US500 daily closes 1995-03-24 to 2026-11-06: 7959 sessions
peak        trough        fall days down  back to peak
1997-10-07  1997-10-27   10.8%        20       39 days
1998-07-17  1998-08-31   19.3%        45       84 days
1998-09-23  1998-10-08   10.0%        15       13 days
1999-07-16  1999-10-15   12.1%        91       32 days
2000-03-24  2002-10-09   49.1%       929     1694 days
2002-11-27  2003-03-11   14.7%       104       62 days
2007-10-09  2009-03-09   56.8%       517     1480 days
2010-04-23  2010-07-02   16.0%        70      125 days
2011-04-29  2011-10-03   19.4%       157      144 days
2015-05-21  2015-08-25   12.4%        96      321 days
2015-11-03  2016-02-11   13.3%       100      117 days
2018-01-26  2018-02-08   10.2%        13      197 days
2018-09-20  2018-12-24   19.8%        95      120 days
2020-02-19  2020-03-23   33.9%        33      148 days
2022-01-03  2022-10-12   25.4%       282      464 days
2023-07-31  2023-10-27   10.3%        88       35 days
2025-02-19  2025-04-08   18.9%        48       80 days
corrections (10-20%): 13, median fall 13.3%, median 88 days down, median 84 days back to the peak
bear markets (20%+): 4, median fall 41.5%, median 400 days down, median 972 days back to the peak
2026-11-06: 7,376.86, 5.65% below the 2026-10-06 record of 7,818.93

An open decline prints not yet in the last column. Run on US100 with the same dates, the last line of the table is the November 2026 correction, 14.7% deep at its 5 November low and not yet recovered.

Reading a correction in market data