When Do the Stock Market's Best and Worst Days Happen?

Key takeaways

• All 25 of the Nasdaq Composite's biggest one-day gains between February 1971 and October 2026 arrived while the index was already more than 20% below its highest close to date.
• Only 39.2% of the 14,047 trading days in that series were entered more than 20% below the peak, so all 25 of the best days fell inside a minority of the sample.
• 14 of those 25 best days fell within 10 trading days of one of the 25 worst. On 12, 13 and 16 March 2020 the index fell 9.43%, rose 9.35%, then fell 12.32%.
• Removing the 10 best days cuts the compound return from 10.49% a year to 8.61%. Removing the 10 worst raises it to 12.51%. Removing both leaves 10.60%.
• On the S&P 500, where just 2.9% of the last 2,511 trading days were entered 20% or more below peak, 6 of the 10 best days still landed there.

Where the stock market's best and worst days actually happened

You have met the chart that says missing the 10 best days ruins a long-run return. The question it never answers is when those days turn up.

On the Nasdaq Composite the answer leaves little room for argument. Every one of its 25 biggest one-day gains between 5 February 1971 and 23 October 2026 happened with the index already more than 20% below its highest close to that point. Not most of them. All 25.

The worst days keep the same company. 15 of the 25 steepest one-day falls came from more than 20% below the peak, and 14 of the 25 best days landed within 10 trading days after one of those 25 worst. The tightest sequence in the whole series is March 2020. The index fell 9.43% on the 12th, rose 9.35% on the 13th, and fell 12.32% on the 16th, which is the largest single-day fall in 55.7 years of data.

So the best and worst days of the stock market aren't two separate populations. They are the same few weeks, sorted by sign.

The series, and how the drawdown was measured

The data is the Nasdaq Composite daily close as published by the Federal Reserve Bank of St Louis: 14,048 observations from 5 February 1971, when the index stood at 100.000, to 23 October 2026, when it closed at 25,955.34. That gives 14,047 daily moves. The average move, ignoring direction, is 0.85%.

For every day we measured the drawdown entering it, meaning how far the previous close sat below the highest close the index had ever recorded up to that date. A day entered more than 20% below the record high is a day inside a bear market, whatever happens next. All 25 of the best days opened there, and so did all 50 of the 50 best. Widen it to the 100 best days and 87 of them still sit there.

The ten biggest up days, and how deep the hole was

The second column is the one that gets quoted. The third is the one that doesn't.

Two of those readings need a note. The 2008 entries look extreme because the Nasdaq Composite's record close was still 5,048.62, set on 10 March 2000, and the index did not close above it again until 23 April 2015. Being 67.3% down in October 2008 means 67.3% below the dot-com peak, not below a 2007 high. The April 2025 reading is measured against 20,173.89, the record close of 16 December 2024.

The pairing isn't new either. On 19 October 1987 the index fell 11.35%, the second-largest one-day fall in the whole series. 2 sessions later, on 21 October, it rose 7.34%, which ranks 16th among the best days. Both sit inside one week.

The dates are also concentrated. Across the 50 largest daily moves, up and down together, 36 fall in four calendar years: 2000, 2001, 2008 and 2020. Calendar year 2000 alone supplies 13 of them. A fund launched in 1989 and closed in 1996 wouldn't have met a single one of them.

What clusters is news, not luck

None of this is a coincidence of this particular index. It is the property Robert Engle built a model for, and described in his Nobel Lecture of 8 December 2003: "Returns are almost unpredictable, they have surprisingly large numbers of extreme values and both the extremes and quiet periods are clustered in time." Those three features, he writes, are "often described as unpredictability, fat tails and volatility clustering". His explanation of the mechanism is one sentence long: "Volatility clustering is simply clustering of information arrivals."

In a drawdown the information arrives constantly, and a good deal of it arrives from the authorities. Look at what the biggest up days were reacting to.

• The largest gain in the series, 14.17% on 3 January 2001, came the day after a 7.23% fall, and on the day the Federal Open Market Committee announced it had decided "to lower its target for the federal funds rate by 50 basis points to 6 percent".
• On 13 October 2008 the index rose 11.81%, the day the Bank of England, the European Central Bank and the Swiss National Bank announced dollar tenders at full allotment, in a joint statement with the Federal Reserve and the Bank of Japan, in which "Counterparties in these operations will be able to borrow any amount they wish against the appropriate collateral in each jurisdiction". 2 trading days later it fell 8.47%.
• On Sunday 15 March 2020 the FOMC "decided to lower the target range for the federal funds rate to 0 to 1/4 percent". The next trading day the index fell 12.32%, the worst day in the series. A week later, on 23 March, the FOMC said it would buy Treasuries and agency mortgage-backed securities "in the amounts needed to support smooth market functioning", having previously announced "at least $500 billion of Treasury securities and at least $200 billion of mortgage-backed securities". The index rose 8.12% the following day.
• On 9 April 2025, Executive Order 14266 suspended the country-specific tariff rates until July and set "an additional ad valorem rate of duty of 10 percent" instead. The index rose 12.16%, 4 trading days after a 5.97% fall.

Notice that the policy news didn't reliably produce a good day either. The cut to a 0 to 1/4 percent range on 15 March 2020 was followed by the worst day in the entire series. The moves around all four announcements were large in both directions, which is what volatility clustering looks like from the inside.

The options market agrees. Across 9,301 VIX closes from 2 January 1990, the median reading is 17.56. That series starts in 1990, so it covers 24 of the 25 best days and 21 of the 25 worst. On those best days the median reading is 28.95, and on those worst days 34.72. The lowest VIX close on any of them is 20.39, still above the median for the series as a whole.

Missing the best days and missing the worst days nearly cancel out

Take the same 14,047 daily moves and delete days by hand. Over 55.7 years the index multiplied by 259.6, which is 10.49% a year before dividends.

The first two rows are the familiar warning: 259.6 times your money becomes 99.7 times it. The third row is the half nobody prints, and it is larger, 712.0 times. The fourth row is the interesting one. Dodge both tails and you land at 10.60% a year, 10 basis points above having done nothing at all for the whole 55.7 years.

The stock market's best and worst days don't merely resemble each other in size. They share a calendar, and that's what the clustering does to the argument. Being out of the market for the worst days is worth more per day than being in for the best ones, because a 12.32% fall is bigger than a 9.35% rise. But the two sets overlap in time so tightly that any rule crude enough to miss one tends to miss the other, and the net is close to zero. The 25-day version lands at 10.62%, a 13 basis point difference over more than half a century. That isn't a strategy. It's a rounding error with half a century of nerve attached to it. Whether that overlap shows up in what real investors earn is a separate measurement, reported as the behavior gap between fund returns and investor returns.

The strongest objection, and what the S&P 500 says about it

The obvious criticism is a base-rate one. The Nasdaq spent much of its life underwater, so of course the fireworks happened there. The numbers don't contradict the premise: 5,500 of the 14,047 days, or 39.2%, were entered more than 20% below the peak, and 3,424 of those 5,500 fall between 10 March 2000 and 23 April 2015, the 15 years the index spent climbing back to one price. Finding all 25 best days in a state that covers 39.2% of the sample is striking, but it is not a miracle.

The test is an index that spent almost no time deeply underwater. The S&P 500 series published by the Federal Reserve Bank of St Louis goes back 10 years: 2,512 closes from 24 October 2016 to 21 October 2026. In that decade just 74 of 2,511 days, or 2.9%, were entered 20% or more below the peak. Even so, 6 of the 10 best days landed in that 2.9% sliver. Its three worst days were 9, 12 and 16 March 2020, and its second and third best days were 24 and 13 March 2020. Its single best day, 9.52% on 9 April 2025, came 3 trading days after a 5.97% fall. On that series, removing the 10 best days takes the return from 13.70% a year to 6.52%.

The NASDAQ 100, 10,284 closes from 2 January 1986, gives the same answer: 24 of its 25 best days were entered more than 20% below peak. Three indices, three different samples, one pattern that doesn't go away. The base-rate objection weakens the Nasdaq Composite result. It doesn't survive the S&P 500 version of it.

What this measures, and what it cannot tell you

These are price indices. Dividends are excluded, which understates every total return here, though it has no bearing on where the extreme days sit. The drawdown is measured against the highest close to date, so a long dry spell drags the measured depth along with it, which is why the 2008 rows read as 67.3% and 70.2%.

The 25-day cut-off is a choice. It happens to be robust in this sample, holding at 50 days and nearly holding at 100, but a different cut on a different index could read differently. The deletion exercise also charges nothing for trading, spreads, tax or the cost of holding cash in between, and it assumes a switch executes at the close. A real round trip is not free.

Most of all, this is history, not a forecast, and it is US history. Every series here is American, so the stock market's best and worst days described above are American ones, measured across three indices over one stretch of 55.7 years. A UK index could behave differently, and nothing above tests that. The sample contains four exceptional years, and if the coming decades contain a different sort of crisis, the pattern can change. Picking out the 25 best days needs the entire series in hand, which nobody has in advance. That is the whole difficulty, and it is a different problem from measuring volatility vs drawdown after the fact.

What would change the conclusion

Three things would. The first is a sample in which the biggest up days clustered near record highs rather than deep below them. Nothing in these three series looks like that, but three indices over one stretch of history can't settle a claim about markets in general.

The second is volatility that stops clustering. Engle's property is an empirical regularity, not a law. If the extremes spread out evenly across calm periods, the pairing of best and worst days dissolves, and avoiding the falls stops costing you the rebounds.

The third is a signal that separates them in advance. The pattern here says the best and worst days share a neighbourhood. It doesn't say which one comes next, and the 2025 episode is the newest test of that: a 24.3% drawdown, a 5.97% fall on 3 April, a 12.16% rise 4 sessions later. Anyone watching the next drawdown can check the same thing in real time, by asking how many trading days separated its deepest fall from its biggest rebound. In this series the median answer is 6, and the price of being wrong about the order is the loss recovery arithmetic working against you from a lower base.