Bounces out of 18 weeks, if a losing week changes nothing (base rate 82.3%)
- 17 or 18 of 18: 14.6% of the time
- Every other count
Hover or tap a column.
October 17, 2026 · 3 min read
The 17-out-of-18 Bounce Stat, Checked Against Its Base Rate
A losing week in the Nasdaq 100 is followed by a bounce almost every time. So is every other week. The viral streak stat, checked against its base rate on 13 years of NQ and ES.
A stat went round this week: the Nasdaq 100 fell on every trading day of the week, one of fewer than twenty such weeks on record, and after the previous ones it was higher again within four trading days 17 times out of 18. The suggestion was that a bounce should follow. The number is probably right. The question it leaves out is how often the index is higher within four trading days after any week at all.
The base rate
We ran the same test on Nasdaq-100 futures (NQ) and S&P 500 futures (ES), using regular-session closes. For each week we asked whether any of the next four closes was above the Friday close, then compared weeks that fell every day against all weeks.
| Condition | Market | Weeks | Higher close within 4 sessions | 95% interval |
|---|---|---|---|---|
| Every week (base rate) | NQ | 660 | 82.3% | 79–85% |
| Fell every session | NQ | 9 | 9 of 9 | 70–100% |
| Every week (base rate) | ES | 753 | 79.2% | 76–82% |
| Fell every session | ES | 10 | 9 of 10 | 60–98% |
On NQ, 9 of 9 losing weeks were followed by a higher close within four sessions. After an ordinary week, the rate is 82%. Stock index futures rise more often than they fall, and over four sessions the price only has to close above one level once, so almost every week passes this test.
ES has one failure: the week ending 29 July 2011. Over the next four sessions ES fell another 7.1%, into the August 2011 downgrade selloff. A stat that holds 17 times in 18 says nothing about how large the 18th case can be.
How surprising is 17 out of 18?
Suppose a losing week tells you nothing, and the true rate afterwards is just the 82% base rate. Then 9 bounces in 9 tries happens 17% of the time, and 17 or more in 18 happens 15% of the time. Neither is rare. Roughly one in seven runs of 18 random weeks would produce the headline with no effect behind it. The chart at the top of this post shows the whole distribution: the most likely count is 15, and 17 or 18 together come up about once in seven tries.
A larger sample: any five-day losing streak
Calendar weeks are arbitrary, so we dropped them and counted every time NQ or ES closed down five sessions in a row, measured from the fifth day. That roughly triples the sample.
| Condition | Market | Cases | Higher close within 4 sessions | 95% interval | Average 4-session change |
|---|---|---|---|---|---|
| Every session (base rate) | NQ | 3,263 | 80.3% | 79–82% | +0.30% |
| Day 5 of a losing streak | NQ | 23 | 91.3% | 73–98% | +0.35% |
| Every session (base rate) | ES | 3,720 | 79.2% | 78–80% | +0.21% |
| Day 5 of a losing streak | ES | 31 | 90.3% | 75–97% | +0.73% |
Streaks do bounce a little more often than ordinary days: about 90% against 80%. With 23 and 31 cases, though, the intervals reach down to the base rate, and the exact chance of results this good with no effect at all is 14% for NQ and 9% for ES. On NQ the average size of the bounce, +0.35% over four sessions, is close to the +0.30% that follows any session. There may be a small tilt here. This sample is too small to tell.
Higher close within four sessions, with 95% intervals
- After the losing condition
- Base rate
Hover or tap a row.
The choices inside the stat
Every condition in a stat like this was picked by someone: four days rather than three or five, "higher at any close" rather than "higher at the end", calendar weeks rather than streaks, the ETF's history rather than the index's. Each choice is a chance to find a framing that looks strong. That does not make the stat false. It means you cannot judge it until you know how many framings were tried.
- Base rate: how often does the outcome happen without the condition? Here, about 80% of the time.
- Sample size: how many cases, and how wide is the interval? Nine or eighteen cases cannot separate 82% from 95%.
- Look count: how many versions of the condition were tried before this one was posted?
Data and method. E-mini Nasdaq-100 (NQ) and E-mini S&P 500 (ES) futures, regular-session closes (4:00 pm ET, 1:00 pm on early-close days), NQ from 2 January 2013 and ES from 3 January 2011, both to 8 October 2026. A week counts when it has at least four sessions and no contract roll inside it; a day-to-day change across a roll is left out. "Higher within four sessions" means any of the next four closes is above the reference close. Intervals are 95% Wilson intervals; the chance figures are exact binomial tails at the base rate.