The Nasdaq fell 5.5% in a day. What happened after the 48 earlier drops like it?

The Nasdaq Composite fell 5.5% on Thursday 5 November 2026, from 25,736.51 to 24,315.00. On Friday it closed at 24,579.42, up 1.09%.

We looked at the 48 earlier days since 1971 when the index fell 5% or more in one day, and at what happened over the next week, month, quarter and year. The short answer: it was not the same every time. The first few weeks looked a bit better than normal, the full year did not, and most of the signal comes from just two crashes.

The rule we tested

An event is any day the Nasdaq Composite closed 5% or more below the day before. We fixed that line before we looked at any result. For each event we measured the change from that day's close to the close 1, 5, 20, 60 and 250 trading days later. A trading day is a day the market is open, and 250 of them is about a year. We did not try any other thresholds or time spans.

The data is the daily close of the Nasdaq Composite (^IXIC) from the Yahoo Finance chart API, downloaded on 7 November 2026. It starts on 5 February 1971. It is price only and leaves out dividends. There are 49 such days in the file, including 5 November, so 48 earlier ones. The latest earlier ones were 11 June 2020, 13 September 2022, and 3 and 4 April 2025. The file, our script and the download command are in the Tickfloor repository under scripts/news/nasdaq-1107/.

What happened

• After 1 day: up 26 of 48 times (54%). The average was +1.28% and the middle result was +0.29%. The worst was -9.0% (the day after 19 October 1987) and the best was +14.2% (3 January 2001).
• After 5 days: up 26 of 48 (54%), average +1.22%, worst -17.0%, best +16.7%.
• After 20 days: up 28 of 48 (58%), average +3.10%, worst -24.1%, best +32.7%.
• After 60 days: up 31 of 48 (65%), average +5.95%, worst -30.5%, best +45.1%.
• After 250 days: up 29 of 48 (60%), average +12.13%, worst -58.9%, best +92.9%.

All 19 of the one-year results that were down came from events between January 2000 and September 2001, the dot-com crash. The 2008 and 2020 events were all up a year later.

Compared with a normal day

To find out whether that is different from a random day, we took every trading day since 1971 as a start day, 14,057 sessions, and measured the same spans.

A permutation test is a luck check: we pick 48 random start days, take their average, and repeat that 200,000 times to see how often chance alone beats the real average. The p column is the share of draws that matched or beat the drop days (0.0000 means none of the 200,000 did). The BH q column corrects for running five tests at once, using the Benjamini-Hochberg method.

Two things stand out. The average gain over the next 1 to 60 days was bigger than for a random day, but the share of results that were up was not: it was the same or lower. So a few large rebounds pull the average up. And over a full year, the drop days did no better than any other day, with 60% up against 76%.

The de-clustered check

Big drops come in bunches. Thirty-one of the 48 events fall in two crashes: 20 in 2000 and 2001, and 11 from September 2008 to January 2009. If one crash counts 20 times, it can swamp everything else.

So we kept only the first event in any 20-trading-day stretch (an event within 20 sessions of the last one we kept was dropped). That leaves 21 events. We reran the same tests.

• After 20 days: up 11 of 21, average -0.03%, against +0.97% for any day. Permutation p = 0.7869, BH q = 0.7869.
• After 250 days: up 13 of 21, average +11.82%, against +13.03% for any day. Permutation p = 0.5950, BH q = 0.7438.
• After 1 day the average was still +0.99% (p = 0.0009, q = 0.0044). After 5 days it was +0.30% (q = 0.7438) and after 60 days +6.06% (q = 0.2320).

Once each crash counts once, the one-month and one-year results look like ordinary days. Only the next-day bounce is still clearly better than chance.

What this does and does not show

• Past drops of 5% or more were followed, on average, by a better next day and a better next month than a random day. That gap mostly comes from clustered events in a few crashes, and it shrinks to nothing at 20 days once they are counted once.
• It does not show that a drop is a signal for anything. The results ranged from -58.9% to +92.9% a year later. The average hides that spread.
• The windows overlap each other, and events sit next to each other in time, so the p-values are rough guides and not exact.
• The test does not say why any of this happens. We have not stated a cause for the 5 November fall, and nothing here explains it.
• Price only. Dividends would add a little to every window, event and ordinary alike. There are no costs in these figures.
• Forty-eight events is a small sample, and 21 is smaller. A real but modest effect could be missed.

We apply the same standard to every claim we test: write the rule down first, compare it with a fair baseline, then correct for how many things we tried. Across all of Tickfloor's backtests, 1,346 strategies in 77 families were tested, 266 finished ahead of the benchmark before correction, and none passed after correcting across 1,778 tests (in-sample, not a promise). Our method is on the luck check page and every result is on the evidence page. To see how Tickfloor scores a stock out of 10 from ten chart checks, you can score a stock free.

General information, not personal advice.

Source: Yahoo Finance chart API, ^IXIC daily close, downloaded 7 November 2026. Figures are our own arithmetic on that file. Past results, not a forecast: they describe 48 earlier drops and say nothing certain about what follows this one.