Key takeaways
- Across the eight big S&P Composite drawdowns from the December 1968 peak onward, the median gap between a peak and the next new high was 31.5 months on nominal prices. Adjusted for inflation, the median was 63 months — twice as long.
- The December 1968 peak was back in nominal terms by March 1972, after 39 months. In real terms it wasn't regained until January 1992 — 277 months, or 23 years and 1 month.
- Dividends do most of the repair work. That same 1968 peak was recovered in 47 months on a real total-return basis against 277 months on price alone, at a dividend yield that averaged 3.99% over the stretch. In June 2026 the yield on the same series was 1.09%.
- The index's recovery time is not a person's. Someone who kept adding 6% of their starting balance each year through the 2000 bear market was whole again in 78 months against 153 for a lump sum; someone drawing 4% a year took 206 months.
- The worst case in the data isn't American. Japan's OECD monthly share-price index peaked in December 1989 and did not regain that level until July 2025 — 427 months, or 35 years and 7 months.
The median wait was two and a half years, or five and a quarter — the difference is inflation
If the market crashes, how long until you're back where you started? On the eight big US drawdowns from the December 1968 peak onward, the median answer is 31.5 months if you count in dollars and 63 months if you count in purchasing power. The worst case is far worse than either: the peak of December 1968 wasn't exceeded in real, inflation-adjusted price terms until January 1992, twenty-three years later.
Those two medians describe the same eight episodes. The only thing that changes is the yardstick. That is the argument of this piece, and it's why a recovery table that ignores inflation is a comfortable table rather than an accurate one.
How these numbers were built, because the method changes the answer
Everything below comes from one file: Robert Shiller's monthly US stock market dataset, which carries the S&P Composite price, dividends and the US consumer price index from 1871 to the present. Three choices in that file shape every figure here, and each matters more than it sounds.
Monthly, not daily. Shiller's price for a month is an average of that month's daily closing prices — except the final month in the file, which the file notes is a single day's close. Averaging clips the tops off peaks and fills in the bottoms of troughs, so drawdowns measured this way look shallower than the headline numbers you remember. The 2022 fall is a clean test. On Shiller's monthly averages it was 20.3%. The daily S&P 500 series at the St. Louis Fed fell 25.4%, from a close of 4,796.56 on 3 January 2022 to 3,577.03 on 12 October 2022. Depth changed by five percentage points. The recovery date barely moved: daily data has the index back above its old close on 19 January 2024, monthly data in December 2023.
Price only, unless stated. The main table tracks the price index, which is what people mean when they say the market went up, and what almost every chart shows. It excludes dividends, and excluding dividends makes recoveries look far longer than an investor's actually were. The last column corrects for that using Shiller's real total-return series, which reinvests dividends back into the index.
Real means CPI-adjusted. The real series is the nominal price multiplied by the ratio of the latest CPI to that month's CPI. Recomputing it by hand from the price and CPI columns reproduces Shiller's real column exactly, to four decimal places. Because every comparison here is a ratio between two months, the choice of base month cancels out — the recovery dates would be identical whichever month you deflated to.
A drawdown here means a fall from an all-time high in the monthly-average nominal price. Recovery means the first month the index — on whichever of the three yardsticks is being measured — got back to the level it held in that same nominal peak month. That anchoring choice does real work: measuring the real column against the prior real all-time high instead would push the median from 63 months to 109.5. Eight episodes from the December 1968 peak onward are in the neighbourhood of 20% or worse, and those eight are all of them. Two, 1980 and 2020, fall a little short of that line on monthly averages while being unambiguous bear markets on daily prices, so they are included and flagged rather than quietly dropped. The start date is a choice too: a 22.5% fall from a December 1961 peak sits just outside the window, and it was back to a new high in 21 months nominally and 24 in real terms, so including it would pull both medians down.
Every major drawdown since the December 1968 peak, measured three ways
| Peak | Trough | Fall, nominal | Fall, real | Recovery, nominal price | Recovery, real price | Recovery, real total return |
|---|---|---|---|---|---|---|
| Dec 1968 | Jun 1970 | -29.0% | -35.1% | Mar 1972 (39m) | Jan 1992 (277m) | Nov 1972 (47m) |
| Jan 1973 | Dec 1974 | -43.4% | -53.5% | Jul 1980 (90m) | Aug 1987 (175m) | Jan 1985 (144m) |
| Nov 1980 | Jul 1982 | -19.4% | -29.3% | Nov 1982 (24m) | Apr 1983 (29m) | Dec 1982 (25m) |
| Aug 1987 | Dec 1987 | -26.8% | -27.5% | Jul 1989 (23m) | Jan 1992 (53m) | Aug 1989 (24m) |
| Aug 2000 | Feb 2003 | -43.7% | -46.8% | May 2007 (81m) | Nov 2014 (171m) | May 2013 (153m) |
| Oct 2007 | Mar 2009 | -50.8% | -51.7% | Mar 2013 (65m) | Nov 2013 (73m) | Mar 2013 (65m) |
| Jan 2020 | Mar 2020 | -19.1% | -19.1% | Aug 2020 (7m) | Aug 2020 (7m) | Aug 2020 (7m) |
| Dec 2021 | Oct 2022 | -20.3% | -25.4% | Dec 2023 (24m) | Jun 2024 (30m) | Mar 2024 (27m) |
Read the last three columns across and the pattern is consistent. The nominal price column is the flattering one, the real price column is the punishing one, and real total return sits between them. Medians across the eight episodes: 31.5 months nominal price, 63 months real price, 37 months real total return.
The chart plots the sixth column — months to recover in real price terms — for each of the eight peaks. Two bars dominate it, and they are the two a nominal table hides completely.
In real terms, the market of August 1987 was still below the market of December 1968
The 1970s are the standing example of nominal recovery flattering, and the data is more extreme than the reputation. From the December 1968 peak, nominal prices took 39 months to recover. The real index kept falling, on and off, for another decade, bottoming in July 1982 at 62.6% below its 1968 level. Consumer prices had by then risen 2.75 times over. The index had gone essentially nowhere in dollars — 106.50 in December 1968, 109.40 in July 1982 — while the dollar had lost nearly two-thirds of its purchasing power.
One comparison makes it concrete. In August 1987, at the top of the bull market that ended with Black Monday, the S&P Composite stood at 329.40 against 106.50 in December 1968. That is a tripling. In inflation-adjusted terms it was still 4.0% below where it had been nineteen years earlier. The 1968 peak was only cleared in real terms in January 1992, by which point the price index had quadrupled.
That single episode is why both columns get computed. Anyone told in 1972 that the bear market was over was hearing something true and something useless at the same time.
Dividends did most of the repairing, and the yield that made that possible is gone
The 277-month figure is the harshest honest number in this dataset, and it is also incomplete, because it measures an index that throws dividends away. Reinvest them and the same December 1968 peak was recovered in real terms in November 1972, after 47 months.
The gap between 47 and 277 is not a rounding difference. It is the entire character of the episode. The mechanism is dull and powerful: over the span from December 1968 to January 1992 the dividend yield on Shiller's series averaged 3.99% and reached 6.24% at its highest. A holder who reinvested was buying more shares every quarter at prices that stayed depressed for years. A price index can't see any of that. A real account can.
The same effect shows up in the 2000 episode, but weaker: 171 months on real price against 153 months on real total return. The yield over that stretch averaged 1.89%, less than half the earlier period's. In June 2026 the yield on the same series was 1.09%. A drawdown starting from that level of income would have dividends doing considerably less of the work than they did in the 1970s, which is a reason to read the 47-month figure as the friendly end of the range rather than the typical case.
The index's recovery time is nobody's actual recovery time
Here is the strongest objection to everything above, and it is a good one. Time to break even is the wrong metric for most people. Almost nobody puts a lump sum in at the exact peak and then does nothing for a decade. People add money, or they take it out, and both change the answer enormously.
Running that through the same real total-return series, from the August 2000 peak: a lump sum was whole again in 153 months. Someone holding that same starting balance and adding a constant real amount equal to 6% of it each year got back to break-even — portfolio value above starting balance plus everything contributed since — in 78 months. Roughly half. At 12% a year of the starting balance it was 74 months. The mechanism is not clever. A bear market is a long sequence of purchases at low prices, and enough of them drag the average cost down faster than the index climbs back.
It cuts the other way just as hard. Withdrawing a constant real 4% a year of the starting balance from that same August 2000 peak pushed break-even out to 206 months, in October 2017 — four and a half years later than the lump-sum investor. Selling units into a falling market permanently removes the shares that would have participated in the recovery. This is why retirement research obsesses over the order in which returns arrive rather than their average, and it is close cousin to the arithmetic behind the gap between fund returns and investor returns.
So the table is not a forecast of anyone's recovery. It describes a specific, artificial investor: one who bought everything at the worst possible moment and then neither added nor withdrew a cent. That experience brackets the range. It isn't the middle of it.
The recession ends years before the index does
A reliable disappointment in this data is the gap between the economy turning and the index recovering. The NBER dates the 2007-09 contraction from a peak in December 2007 to a trough in June 2009. The S&P Composite didn't regain its October 2007 nominal level until March 2013 — 45 months after the recession had officially ended, and 65 months after the market peak. The 2001 recession ran from March to November 2001; the August 2000 market peak wasn't recovered nominally until May 2007, and in real terms not until November 2014.
The pattern holds in the other direction. The February-to-April 2020 recession was the shortest on the NBER's record, and the drawdown around it was correspondingly the shortest here at seven months. Short slump, short recovery. But a recession ending is not, on this evidence, evidence that a drawdown has ended.
Eight episodes, one country, and that country is the survivor
The limits are severe and they all point the same way.
Eight is a very small sample. Two episodes, 1968 and 2000, supply almost all of the tail. Remove them and the real median collapses. A median drawn from eight observations describes eight things that happened; it is not a distribution.
The US is the wrong country to generalise from. The OECD's monthly share-price index for Japan peaked in December 1989, fell 74.5% to a trough in November 2011, and didn't regain the 1989 level until July 2025 — 427 months, or 35 years and 7 months, in nominal local-currency price terms. It came within 0.5% in July 2024 and then slipped away again for a year. That is a nominal, price-only figure, and a Japanese holder reinvesting dividends did better. The point survives: the US recovery record isn't a law of nature.
Survivorship is baked in. The UBS Global Investment Returns Yearbook 2026, compiled by Dimson, Marsh and Staunton across 35 markets since 1900, puts US real equity returns at 6.6% a year from 1900 to 2025 against 1.6% for bonds, and notes that the US now accounts for around 62% of world equity market value — a share built on precisely the returns being measured here. Anarkulova, Cederburg and O'Doherty, working with 39 developed countries from 1841 to 2019 in the Journal of Financial Economics, argue that a broad sample mitigates the survivor bias in US-only work, and estimate a 12% chance that a diversified investor with a 30-year horizon loses money relative to inflation. That is a direct challenge to the idea that any recovery table implies a guarantee.
The most recent CPI is partly estimated. Shiller flags three months of CPI in the file as estimates — the last two, and October 2025; the BLS series puts June 2026 at 333.952 against Shiller's 336.175, a difference of 0.7%. Because every real figure above is a ratio between two months, that gap doesn't move a single recovery date.
What would change the conclusion
A different inflation regime. Every long real recovery in this table is an inflation story, not a stock story. The 1968 peak needed 277 months in real terms because consumer prices nearly quadrupled over the span. In the 2007 episode, where inflation was mild, real recovery took 73 months against 65 nominal — a gap of eight months rather than 238. If inflation stays near target, the nominal and real columns converge and this whole distinction shrinks to a footnote.
A different yield. Dividends closed most of the 1968 gap because the yield averaged 3.99%. At the 1.09% of June 2026, and with more of the payout arriving as buybacks, real total-return recovery would sit far closer to the real price column than it did in the 1970s.
A wider sample. Eight US episodes is what a single monthly series back to 1871 supports. A dataset spanning many countries, of the kind Anarkulova and co-authors assemble, produces a fatter and less reassuring tail, and there's no strong argument for why the US distribution should be the right prior.
As of the latest data in that file — a July 2026 figure that the file's own footnote records as the 7 July close — the S&P Composite was at a nominal and real high, so none of these clocks was running on that date. The measurement that carries information isn't the index level but the distance from your own peak, and which of the three columns above you're measuring it in. LedgerTouch reports that drawdown continuously; a spreadsheet with a CPI column does the same job once a month. Either beats the nominal chart, which is the one that says you recovered in 1972.
Extending this from an index to a whole portfolio runs straight into the next question — whether everything else recovered on the same clock. Mostly it didn't, and gold's own 45-year real drawdown record is the sharpest illustration available.