How Many Bear Markets Have There Been Since 1929?

Key takeaways

• Counting a fall of 20% or more from a cycle peak, the monthly S&P Composite had 14 bear markets between January 1929 and September 2026, roughly one every 7 years.
• Change one rule and the same file gives 8 instead. If an episode isn't over until the old high returns, the average gap stretches from about 7 years to 12.2 years.
• The median bear market fell 31.6% and took 15.5 months to reach its low. The mean fall was 36.6%, pulled deeper by the 81.3% collapse into June 1932.
• Gaps between one peak and the next ran from 0.6 years to 15.6 years, with a median of 7.0. For peaks since 1946 the median gap is 10.2 years.
• The March 2020 crash is absent from the monthly record. On monthly averages that fall measures 19.09%, short of the 20% line, while daily closes put it at 33.9%.

How many bear markets there have been, and why the answer is a range

How many bear markets have there been since 1929? On Robert Shiller's monthly S&P Composite series, the answer is 14. That count follows the definition in the SEC's own investor glossary, which says that "generally, a bear market occurs when a broad market index falls by 20% or more over at least a two-month period". It treats each episode as finished once the index has climbed 20% back off its low.

Those 14 episodes are spread across the 97.7 years from January 1929 to September 2026. That's roughly one every 7 years. The median one took the index down 31.6%, and 15.5 months to reach the bottom. The mean fall was deeper at 36.6%, because a single episode in the set lost 81.3%.

Now change one rule. Say a bear market isn't over when the index has bounced 20%, but only when it has regained its old high. The same file then yields 8 episodes rather than 14, and the average interval stretches from about 7 years to 12.2 years. No price changed. Only the counting rule did.

That's the honest answer, and it's why a list of bear markets is a weaker object than it looks. The prices are not in dispute. The boundaries are. What follows is the full table, what sits behind each row, and the four places where the count quietly breaks.

Every bear market since 1929, with its depth, its length and the macro reading at the low

Each row is one episode on monthly data from Shiller's S&P Composite series, which starts in January 1871 and carries its own consumer price index, so one file gives both the nominal and the real figures. The inflation column is the annual CPI-U rate in the month the index bottomed. The last column shows any NBER business cycle contraction overlapping that window, by its dated peak and trough months. The NBER counts a recession as beginning in the month after its dated peak.

The chart above plots the fall in each of the 14, in the same order. Two things stand out before any interpretation. Six of the 14 rows begin between 1929 and 1938, and the inflation column swings from -10.0% to 12.6%, which is most of the macro history of the twentieth century in one column.

Six of the 14 began between 1929 and 1938, and they bend every average

Between September 1929 and November 1938 the series produced six separate 20% falls. The gaps between the first five peaks were 0.6 years, 2.4 years, 1.4 years and 3.0 years. Nothing in the rest of the record comes close to that density.

Read as one continuous decline instead, the September 1929 peak to the June 1932 low is a fall of 84.8% in price and 80.6% after inflation. The two figures are close because consumer prices themselves fell 21.4% over those 33 months, and the annual rate at the low was -9.9%. Deflation is the reason the real loss was no worse than the nominal one, which is the opposite of the 1970s pattern.

The monthly index did not exceed its September 1929 reading until September 1954. That single episode is why the mean fall across the 14, at 36.6%, is 5 points deeper than the median of 31.6%. One row is doing the work. Averages built on 14 observations, six of them from the ten years after 1929, are descriptive statistics rather than frequencies you can lean on.

The March 2020 crash does not appear in the monthly record at all

Here is the first place the count breaks. Shiller's price series is built from "monthly averages of daily closing prices", so the monthly high of January 2020 reads 3278.20 and March 2020 reads 2652.39. That's a fall of 19.09%, which is short of the 20% line, so the sharpest fall of the past decade is not in the table above.

On daily closes it's unmissable. The S&P 500 closed at 3386.15 on 19 February 2020 and at 2237.40 on 23 March 2020, a fall of 33.9% in 33 days. It was back above the February high 181 days after the peak. The NBER dates the contraction from February 2020 to April 2020, 2 months, against an average of 10.3 months for US contractions from 1945 to 2020.

Notice what that does to the SEC's wording. A bear market there needs 20% or more "over at least a two-month period". The 2020 fall cleared the depth in 33 days and arguably fails the duration test. Any count that applies a 20% threshold to monthly averages inherits that blind spot, which is worth knowing before setting two lists side by side. Measured without a threshold, as a bear market recovery time table does, the same monthly series keeps 2020 in at a 19.1% fall.

Two recent falls stopped within 1.1 points of qualifying

The second break is the threshold itself. The S&P 500 fell 19.8% on daily closes between 20 September 2018 and 24 December 2018, and 18.9% between 19 February 2025 and 8 April 2025. Neither is a bear market on a 20% rule. Both would be on a 15% one.

So part of any bear market history is a record of rounding. The 20% figure is a convention, and the SEC's glossary hedges it with "generally" rather than stating it as a test. The 2018 and 2025 episodes are what that hedge is for: the experience of holding through them ran continuous with 2022, while the label is binary.

How often do bear markets happen? Not every four years, and not every six

The gaps between one bear market's peak and the next, on the 14-episode count, run from 0.6 years to 15.6 years. The median is 7.0 years. Measured to each peak from May 1946 onward, the median gap is 10.2 years, and on the 8-episode count the average interval is 12.2 years. There's no four-year or six-year rhythm anywhere in that, and the spread is wider than any average can carry.

A better question than how many bear markets land in a decade is how much of the time the market is below where it has already been. Between January 1929 and September 2026 the monthly index sat below a previous peak in 907 of 1,173 months, or 77.3% of them. Being underwater is the normal state of a share index, not the exception, and the gap between that and how bear markets get discussed is most of the problem.

As of the close on 21 October 2026 the S&P 500 stood at 7757.17, which is 116.9% above its 12 October 2022 low. That's a four-year run without a 20% fall. The record says nothing about whether the fifth year contains one.

Change the index and the count changes by four times

The third break is which index you are counting. Run the same 20% rule over daily closes of the NASDAQ Composite from February 1971 to October 2026 and it produces 20 bear markets. The monthly S&P Composite produces 5 over the same stretch. Same country, same decades, four times the count.

Depth diverges too. The NASDAQ Composite closed at 5048.62 on 10 March 2000 and at 1114.11 on 9 October 2002, a fall of 77.9%. It did not close above the March 2000 level again until 23 April 2015. That's a worse drawdown than anything in the S&P table except the 1929 to 1932 collapse. The same sensitivity to index choice shows up in bitcoin drawdown history, where the thresholds that define an equity bear market are a routine monthly move.

For a UK holder the worst episode was 1972 to 1974, not 1929 or 2008

The fourth break is which country. On the OECD's monthly UK share price index, which starts in December 1957, the same rule gives 8 bear markets. The deepest ran from August 1972 to December 1974 and took the index down 70.2%. The US series fell 39.6% over those same months.

The UK index did not regain its August 1972 level until September 1977, in price terms and before any adjustment for the inflation of that decade. A reader whose mental image of the worst case is 1929 or 2008 has the wrong decade and the wrong country. Nothing in a US table surfaces that, so a US list answers a different question from the one a UK holder is asking.

Nine of the 14 overlapped a recession, and five did not

Matching the table against the NBER's dated contractions, 9 of the 14 peak-to-trough windows contained one and 5 did not. The falls of February 1934 to March 1935, December 1961 to June 1962, August 1987 to December 1987 and December 2021 to October 2022 all happened with no dated US contraction in the window. The 40.0% fall from November 1938 to April 1942 sat entirely inside a recorded expansion.

The 2021 to 2022 episode is the cleanest example of a bear market without a recession attached. The federal funds effective rate went from 0.08% in December 2021 to 3.08% in October 2022 and on to 5.12% by July 2023, with CPI inflation running at 9.1% in the year to June 2022. For scale, that rate peaked at 19.10% in June 1981. Prices fell 20.3% in nominal terms and 25.4% after inflation, and the NBER never dated a contraction. If you want the sequencing between the two, yield curve inversion and recession lead times are a separate record with its own lags.

This is why the table carries no cause column. A single defining cause per episode is an interpretation, and the data underneath can only support the weaker statement: whether a contraction and an inflation or deflation shock were present. On that test, five episodes had neither.

The strongest objection to counting this way

There's a fair objection to counting this way, and it lands on the 1930s. A rule that ends a bear market after a 20% rally splits the collapse from September 1929 to March 1933 into three separate rows, because the rebounds inside it cleared 20% twice. Anyone who thinks of the Great Crash as one event is right about the economics and is only disagreeing with an arithmetic convention.

The opposite rule has the mirror flaw. Require the old high before an episode ends, and the whole of 1937 vanishes: a 45.4% fall over 14 months stops being a bear market at all, because it happened on the way back up from 1932. Both counts hide something real. That's the case for publishing both numbers rather than picking the one that makes a cleaner headline, and it's also the reason 8 and 14 are both defensible.

What this table cannot tell you

Start with the measurement. Monthly averages of daily closes smooth away short, violent falls, and 2020 proves the effect is large enough to delete an episode. Daily data for the whole span would raise the count, and nobody publishing a long list of bear markets from monthly data can tell you by how much.

These are price figures with no dividends in them. That understates the real experience in both directions: the loss recovery arithmetic is harsher before dividends and gentler after them. After inflation the 1973 to 1974 fall was 53.5% rather than 43.4%, and the May 1946 to June 1949 fall was 42.2% rather than 25.3%, so the price column flatters the quiet inflationary episodes most.

The sample is 14 observations from one country, and one national history is not a distribution. The inflation column records what inflation was doing, not what caused the fall. The NBER's last dated contraction in the file ends in April 2020, so nothing after that is classified either way. And how many bear markets you count depends on three choices made before you look at a single price: the threshold, the end rule, and the data frequency.

What would change the conclusion

If the threshold moved. At 15% the 2018 and 2025 falls join the list and the count rises. At 25% the December 2021 episode, at 20.3%, drops out of it. The answer moves by several episodes on a choice nobody can justify from first principles.

If the frequency moved. A full daily series back to 1929 would almost certainly show more and shorter episodes than monthly averages do, on the evidence of what happened to 2020. Any comparison between two published lists has to start by asking which frequency each one used.

If inflation stays above the 2010s norm. The gap between the nominal and real columns widens, and a price-only list understates the loss by more. That gap was 10 points in 1973 to 1974 and 17 points in 1946 to 1949, and it is the one adjustment that changes the ranking of the episodes rather than just their size.

The figure to watch isn't the count. It's the 77.3% of months since 1929 spent below a previous peak, because that's what holding an index actually feels like, and it doesn't move when someone revises a threshold. LedgerTouch reports that distance from peak continuously, and a monthly table reports it twelve times a year. Either one answers something a list of 14 dates cannot: where you are inside an episode nobody has named yet.