What You Miss When You Sell in a Crash: The S&P 500's '10 Best Days', Calculated

• Daily return = today's close ÷ yesterday's close − 1
• Buy-and-hold multiple = (1 + day 1 return) × (1 + day 2 return) × … × (1 + last day return)
• Missing the N best days = set the returns of the top N days to 0 (in cash that day) and do the same multiplication again

The window runs from the close of 31 December 2004 to the close of 31 December 2024 (5,033 trading days). The total-return index with dividends reinvested (^SP500TR) and the price-only index (^GSPC) were calculated separately.

In money: US$10,000 held grows to about US$71,750; miss only the 10 best days and it is about US$32,870. Per year that is 10.35% vs 6.14%. Ten days, 0.2% of the 5,033, decided more than half of the result.

When the market shakes hard, big down days and big up days arrive together. The worst days of the same 20 years sat inside the same crashes: 16 March 2020 (−12.0%), 12 March 2020 (−9.5%), 15 October 2008 (−9.0%).

• 7 of the 10 best days came within 14 calendar days of one of the 10 worst days.
• 13 March 2020 (+9.3%) came the day after the 12 March worst day (−9.5%) and three days before 16 March (−12.0%).
• On the 10 best days, 20-day volatility was 1.2 to 8.0 times usual (its median over the past year). Across all 20 years an ordinary day is 0.97 times.

So whoever steps out to avoid the scary days usually misses the best days that follow right after. No signal tells you when to come back.

• Drop = today's close ÷ highest close so far − 1
• Levels: 0 to −10% noise, −10 to −20% correction, −20% or more storm (the hold zone)
• Wind = standard deviation of the last 20 daily returns ÷ the median of that value over the past 252 days (1× = usual)
• Wind easing = the first day in a storm when the wind is 25% or more below that storm's strongest wind

2008 example (price index): the highest close was 1,565.15 on 9 October 2007. The close of 1,244.69 on 9 July 2008 gives 1,244.69 ÷ 1,565.15 − 1 = −20.5%, the first day in the storm.

The highest close changes with every new high, and the 20-day and 252-day windows move one day forward every day, so every value has to be recalculated daily.

• It is a what-if: nobody misses exactly the best days. Someone who really sold may also have avoided the worst days. That is why section 7 calculates a separate rule: sell at −20% and buy back at the old high.
• Dividends or not: price only, holding falls to 4.85× and missing the 10 best days to 2.23×. Compare the same index with itself.
• Which high: the drop depends on how far back the 'highest close' goes. Load too little history and the real high is missing, so the drop looks smaller.

On 9 July 2008 the S&P 500 closed at 1,244.69 and crossed the −20% storm line. We compared someone who sold that day with someone who kept holding, up to 28 March 2013, when the index regained its old high (1,569.19). If the seller bought back after seeing the old high, the gain over that time is 0%.

After the storm line the index fell another 45.6% to 676.53 on 9 March 2009. Holding was not easy, but with the same rule the S&P 500 since 1927 gives 6.17% per year for holding vs 2.62% for selling at the storm line (holding was higher in all 8 markets tested, 0.1% cost per trade included).

The index did regain its old high, but a single stock may never come back. Markets differ too: in gold only 2 of the 10 best days came in storms. All numbers are calculations on past data without taxes, fees or currency effects.

1. After the close, write down the same index's close and update the highest close so far.
2. Calculate the drop and compare it with −10% / −20% to decide noise, correction or storm.
3. Take the standard deviation of the last 20 daily returns and divide it by its median over the past 252 days to get the wind.
4. In a storm, keep a note of that storm's strongest wind and check whether today's wind is 25% or more below it.
5. Sort the daily gains of the period you are looking at, find the top 10 and check the drop on each of those days. Repeat whenever you change the period.