Buying at All-Time Highs: What the S&P 500 Did Next
An all-time high is a close above every earlier close in a market's history. For the S&P 500 from 1950 to 2026, a year after a record close the index was higher 73.36% of the time, almost the same as the 74.63% for any day.
An all-time high is a close above every close that came before it. People hesitate to buy one because it feels like the top. The S&P 500’s own history since 1950 is a direct test of that feeling, and the result is plainer than either the fear or the cheerleading suggests.
How it works
The test is strict: higher than every earlier close. Equal does not count, and intraday highs do not count. A record close resets the reference, so the next record has to beat that one.
The history you count from matters. The S&P 500 closed at 31.86 on 16 September 1929 and did not close above that until 22 September 1954, at 32.00. Any count that starts its record-keeping in 1950 calls days in 1950-54 “records” while the index was still under its 1929 peak. This page tracks every close in Yahoo Finance’s index history from 30 December 1927 and counts record days from 3 January 1950, so a record here is a true all-time high. Counting from a 1950 start instead gives 1,510 record days, or 7.82%, which is 162 days too many.
It is a price-only measure. Dividends are not in the index level, so a record close means the price is at its highest, not that a holder’s total return is.
The latest record before this page was checked on 3 October 2026 was 13 August 2026, at 7,798.99, and no close through 2 October went above it. On 25 September 2026 the index closed at 7,743.41, 0.71% below it.
A worked example
The worst case first. Buy $10,000 of the index at the record close of 1,565.15 on 9 October 2007. One year later, on 8 October 2008, the index closed at 984.94 and the $10,000 was worth $6,292.94, a fall of 37.07%. The low came on 9 March 2009 at 676.53, 56.78% below the purchase level. The index first closed above 1,565.15 on 28 March 2013, at 1,569.19, and five years after the purchase it was still 6.98% below it.
Now the middle case. Across all 1,310 record days with a full year of history after them, the median one-year change was +10.25%. On $10,000 that is $11,024.86, against $11,063.88 at the +10.64% median of every day, a difference of $39.02.
The point of putting them side by side is that the 2007 example is real and so is the median. A record close was followed by the worst one-year result in this sample of record days, and also, far more often, by a year that looked like any other year.
The original data
The sample. S&P 500 daily closes from Yahoo Finance’s ^GSPC history, 30 December 1927 to 25 September 2026, price only. Record days are counted from 3 January 1950: 19,305 sessions, of which 1,348 closed at a record, or 6.98%. The other 93.02% of sessions closed below an earlier peak.
What followed a record, against what followed any day. “Higher” means the close N sessions later was above the starting close. A year is 252 sessions.
| Horizon | Record days | Higher after a record | All days | Higher after any day | Median after a record | Median after any day |
|---|---|---|---|---|---|---|
| 1 month | 1,348 | 59.12% | 19,284 | 61.84% | +0.65% | +1.12% |
| 1 year | 1,310 | 73.36% | 19,053 | 74.63% | +10.25% | +10.64% |
| 3 years | 1,225 | 88.41% | 18,549 | 85.64% | +24.83% | +28.30% |
| 5 years | 1,208 | 85.10% | 18,045 | 84.16% | +48.08% | +52.39% |
The differences are small and point both ways. A month after a record, the index was higher a little less often than after any day. Three years after, it was higher a little more often. The median change after a record was slightly lower at every horizon.
How sure is that? An exact binomial test of the record days against the all-day share gives p = 0.04 at one month, 0.29 at one year, 0.005 at three years and 0.39 at five years. Those p-values assume every day is independent, and these are not: overlapping windows share most of their days, and records arrive in clusters. The honest reading is that the gaps are small and the evidence for any of them is weaker than the p-values suggest.
Worst outcomes ran the other way from the fear. The worst one-year change after a record was -37.07%; after any day it was -48.82%. The worst five-year change after a record was -24.62%, against -41.52% for any day. The deepest losses in this sample started below a record, partway down a fall that was already under way: the worst one-year change after any day began on 5 March 2008, when the index was already 14.79% below its record. Every record day and its one-year change is in the record closes file, and the table above is in the forward changes file.
Records come in clusters
Of the 1,348 record days, 732 were followed by another record the very next session, 54.30% of them. A record is usually part of a run, not a lone spike that the market immediately gives back.
They also bunch by year. 46 of the 77 calendar years from 1950 to 2026 had at least one record close, so 31 had none. The recent run has been dense: 57 records in 2024, 39 in 2025 and 27 in 2026 up to 25 September.
That clustering is why “it is at a high, so it must be near the top” fails as a rule. In a rising stretch the next record is often days away, and the long gaps between records come after a fall has begun, not on the record day itself.
Distance from the record
The cleaner question is where the index sat against its peak when you bought. Each day from 1950 is grouped by how far its close sat below the highest close before it, and then the one-year change from that day is measured.
| Starting day | Days | Higher a year later | Median one-year change |
|---|---|---|---|
| At a record | 1,310 | 73.36% | +10.25% |
| 0-5% below the record | 7,030 | 71.48% | +8.48% |
| 5-10% below | 2,868 | 68.79% | +9.40% |
| 10-20% below | 3,837 | 72.17% | +12.18% |
| 20% or more below | 4,008 | 87.13% | +12.78% |
Only the deep group stands apart. Days at a record and days up to 20% below it all landed between 68.79% and 73.36%. Days 20% or more below landed at 87.13%, but those are days inside a bear market, and catching them meant buying while the fall was still running. Those 4,008 days come from just 12 falls, and 924 of them are days from 1950 to January 1954, when the index was still below its 1929 peak. They are far fewer independent episodes than the count suggests.
So the record day was not the expensive day. By this measure it sat with the ordinary days, slightly above some of the groups just below it. The rows are in the distance-from-record file.
When it fails
The first failure is reading this as a reason to buy at highs. It is not. It says the record day was close to ordinary for what followed, not that it was good. An ordinary year still included 2008.
The second is waiting for a pullback with no rule for when the wait ends. From 1950 to 2026 the index closed at a record on 6.98% of days and below one on the rest, so a pullback is almost always available somewhere; the question is whether the price you eventually pay is lower than the one you passed up, and nothing here says it will be.
A third is treating the index as a stock. These figures are for the S&P 500 as a whole. A single company at a record can fall and never return, which an index that replaces its losers does not show.
A fourth is forgetting what is excluded. Price only means no dividends and no inflation, and no costs or taxes. Dividends would add to every change in both columns; they are not measured here.
And a fifth is the size of the deep-fall sample. The 87.13% for days 20% or more below a record rests on 12 falls, the longest of them counted more than a thousand times. A future fall can be longer than any of them.
Related
The S&P 500 page covers what the index is and how it is built. Market timing is the decision this page informs, and dollar-cost averaging is the schedule that removes it. A correction is the 10% fall from a record that the second table measures from.
When a market is at a record, ask what your plan says to do on an ordinary day and do that. The history here says the record day itself was close to an ordinary day for what came next.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.