QQQ · MOVE FORECAST
How far ahead a move can be seen
A forecast of how much QQQ will move over a coming period — a session, a window inside the day, the night or the whole day — made only from what is known before the period starts. Checked on 01 Jan 2024 — 23 Sep 2026, days the model never saw while it was fitted.
In short
- Whether a period will be busy or quiet can be seen in advance quite well. Best of all for windows of 15 minutes to an hour inside the regular session: the forecast explains 72–75% of the differences between busy and quiet windows. Whole periods less so: the regular session 60%, the day 52%, the after-market 50%, the night 34%, the pre-market 29%.
- Quiet periods are recognised reliably: of the 30-minute windows the forecast put in the quietest fifth, 70% were indeed there, and only 0.2% turned out among the busiest fifth.
- Where the price ends at the close of a period is barely forecastable — and that is a limit, not a weak model: even the exact volatility of a 30-minute window would explain its final change by only 22%; the forecast gets 16%.
- For options: a period that is going to be quiet can be skipped in advance; a busy one is likely to bring a move to take profit on, but not necessarily a price far from the entry at a fixed exit time.
How to read the numbers
on 30-minute windowsVolatility — the typical size of QQQ’s move within the period, in percent of the price: the square root of the sum of squared one-minute changes. A typical 30-minute window has 0.19%; windows the forecast called the quietest usually have 0.07–0.15%, the busiest 0.25–0.61%.
“Explains 75%” does not mean “right in 75% of cases”. Picture two extremes: a forecast that always names the same average number knows nothing — that is 0%; a forecast that always names the exact coming volatility is 100%. 75% means the forecast knows in advance that much of what sets busy windows apart from quiet ones; the rest is chance it cannot see. The simple forecast “as on average over the last 22 sessions” explains 31%. (Statistically, R² of the logarithm.)
How close a single forecast is. On average the forecast is right, but each window misses a little: in 61% of windows the actual differed from the forecast by at most 1.25 times either way (with a forecast of 0.20%, between 0.16% and 0.25%), in 88% by at most 1.5 times (between 0.13% and 0.30%).
Busy or quiet. When the forecast put a window in the busiest fifth, it was indeed there in 73% of cases; in the quietest fifth, in 70%. Guessing at random would give 20%.
Range and the final change. Range is the high minus the low: was there a move to exit on. The final change is where the price ended against the start: what an option held to the end of the period is paid on. For both the page also gives the limit: how much would be explained if the period’s volatility were known exactly. With the same busyness the price may end back where it began or far away, and that is largely chance.
Forecast against the actual
15-minute windows · Volatility · 17472 periods in the checkThe periods of the check split into five equal groups by the forecast, from the quietest to the busiest. If the forecast is right on average, the line and the dot of each group coincide; the width of the bar is how far single periods stray.
80% of actualsforecastactual, median
| Forecast group | Forecast | Actual, median | 80% of actuals | Where the actual fellscale 0–0.44% |
|---|---|---|---|---|
| 1 · quietest | 0.08% | 0.07% | 0.04–0.11% | |
| 2 | 0.10% | 0.10% | 0.07–0.15% | |
| 3 | 0.13% | 0.13% | 0.09–0.19% | |
| 4 | 0.17% | 0.17% | 0.12–0.25% | |
| 5 · busiest | 0.25% | 0.26% | 0.17–0.44% |
Explains 72% (the 22-session average alone: 29%). The actual differed from the forecast by at most 1.25 times in 57% of periods, by at most 1.5 times in 84%. Put in the busiest fifth → was there: 72%; in the quietest → was there: 67%; in the quietest → was among the busiest: 0.4%.
Examples from the check: 29 Jan 2024 14:00 — forecast 0.07%, actual 0.07%; 07 Jul 2026 15:15 — forecast 0.13%, actual 0.19%; 15 Apr 2024 09:30 — forecast 0.31%, actual 0.23%.
Accuracy by period
click a period to see it against the actualShare of the differences the forecast explains. In brackets for volatility, the simple forecast “as on average over the last 22 sessions”; for the range and the final change, the limit: what would be explained if the period’s volatility were known exactly.
| Period | Periods in the check | Volatility (22-session average) | Actual within 1.5× of the forecast | Range (limit) | Final change (limit) |
|---|---|---|---|---|---|
| Pre-market 04:00–09:30 | 672 | 29% (8%) | 65% | 34% (56%) | 7% (18%) |
| Regular session 09:30–16:00 | 672 | 60% (20%) | 92% | 37% (64%) | 6% (10%) |
| After-market 16:00–20:00 | 672 | 50% (3%) | 75% | 31% (73%) | 14% (26%) |
| Night 16:00 → 09:30 | 665 | 34% (8%) | 70% | 30% (70%) | 6% (24%) |
| Day 04:00–20:00 | 672 | 52% (17%) | 88% | 37% (60%) | 2% (12%) |
| 5-minute windows | 52416 | 59% (20%) | 69% | 64% (81%) | 16% (28%) |
| 15-minute windows | 17472 | 72% (29%) | 84% | 62% (81%) | 16% (23%) |
| 30-minute windows | 8736 | 75% (31%) | 88% | 60% (79%) | 16% (22%) |
| 1-hour windows | 8064 | 75% (30%) | 89% | 56% (78%) | 14% (20%) |
| 2-hour windows | 6720 | 71% (26%) | 89% | 50% (76%) | 10% (19%) |
| 3-hour windows | 5376 | 68% (24%) | 88% | 44% (73%) | 10% (18%) |
Busy or quiet: how often the forecast is right
guessing at random gives 20%Periods of the check split into fifths by the actual volatility; the forecast picks its own busiest and quietest fifths. The last two columns are the costly mistake for a bought option: a period called quiet that turned out among the busiest.
| Period | Called busiest → was busiest | Called quietest → was quietest | Called quietest → was among the busiest | Same by the range |
|---|---|---|---|---|
| Pre-market 04:00–09:30 | 44% | 52% | 5.2% | 4.4% |
| Regular session 09:30–16:00 | 67% | 64% | 0.7% | 5.9% |
| After-market 16:00–20:00 | 69% | 53% | 1.5% | 3.0% |
| Night 16:00 → 09:30 | 44% | 58% | 3.8% | 2.3% |
| Day 04:00–20:00 | 58% | 62% | 2.2% | 5.2% |
| 5-minute windows | 65% | 58% | 0.6% | 0.7% |
| 15-minute windows | 72% | 67% | 0.4% | 0.8% |
| 30-minute windows | 73% | 70% | 0.2% | 0.6% |
| 1-hour windows | 74% | 72% | 0.1% | 1.2% |
| 2-hour windows | 72% | 68% | 0.3% | 1.8% |
| 3-hour windows | 70% | 66% | 0.4% | 2.0% |
Year by year
volatility, share explainedWindows inside the day hold up in all three years. The regular session, the after-market and the night are weaker in 2026 than in 2024–2025: they have one point a day, and 2026 in the check is under nine months, so a single year is noisier.
| Period | 2024 | 2025 | 2026to 23 Sep |
|---|---|---|---|
| Pre-market 04:00–09:30 | 16% | 37% | 20% |
| Regular session 09:30–16:00 | 59% | 66% | 45% |
| After-market 16:00–20:00 | 53% | 57% | 28% |
| Night 16:00 → 09:30 | 28% | 43% | 10% |
| Day 04:00–20:00 | 44% | 57% | 44% |
| 5-minute windows | 49% | 65% | 58% |
| 15-minute windows | 66% | 78% | 70% |
| 30-minute windows | 69% | 79% | 73% |
| 1-hour windows | 70% | 79% | 72% |
| 2-hour windows | 66% | 76% | 65% |
| 3-hour windows | 64% | 73% | 59% |
30-minute windows by time of day
share explained within one start timeThe first half hour after the open is the busiest and the hardest to forecast: there is no regular-session move yet, only the pre-market and the overnight gap. Within one start time the shares are lower than 75%: part of the overall figure is simply the known daily routine — the open is always busier than lunch.
| Window start | Typical volatility | Volatility: explained | Range: explained | Final change: explained |
|---|---|---|---|---|
| 09:30 | 0.35% | 54% | 35% | 5% |
| 10:00 | 0.29% | 60% | 46% | 8% |
| 10:30 | 0.24% | 67% | 53% | 13% |
| 11:00 | 0.20% | 68% | 53% | 13% |
| 11:30 | 0.19% | 72% | 52% | 12% |
| 12:00 | 0.17% | 72% | 52% | 10% |
| 12:30 | 0.16% | 70% | 56% | 13% |
| 13:00 | 0.16% | 66% | 49% | 10% |
| 13:30 | 0.15% | 70% | 55% | 15% |
| 14:00 | 0.15% | 65% | 46% | 14% |
| 14:30 | 0.14% | 70% | 55% | 19% |
| 15:00 | 0.14% | 70% | 57% | 18% |
| 15:30 | 0.18% | 70% | 50% | 8% |
What it means for options
- Quiet periods can be skipped. Windows inside the day that the forecast puts in the quietest fifth turn out among the busiest in 0.1–0.6% of cases. This is the most reliable use: an option bought for such a period is the one most likely to expire without a move.
- A busy forecast means a move to exit on, not a far final price. Strategies that take profit on the move (a target, a lock at k×) gain more from the forecast than those that exit at a fixed time.
- The market sees much of it too. VIX, one of the forecast’s main inputs, is itself computed from option prices: a period that looks busy is usually priced as busy. The next check is to compare the forecast with the move implied by option prices and see whether buying where the forecast is above the price, and selling where it is below, earns money year after year.
Method, data and reproduction
Data. QQQ minute bars 04:00–20:00 New York time from the project’s verified minute cache (the same as the volatility pages): 1263 sessions; forecast targets from 07 Mar 2022. Outside the regular session a minute without trades has no bar, and the next return spans the gap. Half days and sessions with missing regular-session minutes are left out.
Periods. Pre-market 04:00–09:30, regular session 09:30–16:00, after-market 16:00–20:00, the night from the 16:00 close to the next 09:30 open (the 20:00–04:00 gap included), the day 04:00–20:00; inside the regular session, 5-, 15- and 30-minute windows back to back and 1-, 2- and 3-hour windows starting every 30 minutes and ending by 16:00.
Measures, in percent of the period’s first price. Volatility: 100 × the square root of the sum of squared one-minute log returns (the first from the first minute’s open). Range: the highest high minus the lowest low. Final change: |last close − first open|.
What the forecast knows. Only what is known before the period starts: the same period’s measure 1, 5 and 22 sessions back; the volatility of the last 30 and 60 minutes and of the regular session so far; the pre-market and the overnight gap; the regular session and its last 30 minutes (for the after-market and the night); the share of yesterday’s variance from falling minutes; the previous VIX close; CPI and NFP days (08:30), FOMC (14:00) and after-close earnings in the project basket (AAPL, MSFT, NVDA, AMZN, META, Alphabet, TSLA, AVGO, AMD, MU); the start time for windows. The model is least squares on the logarithms of the measures.
Fit and check. Fitted on 07 Mar 2022 — 29 Dec 2023, checked on 01 Jan 2024 — 23 Sep 2026; every number on this page is from the check. “Explains” is R² of the logarithm; the limit is R² of the range or the final change given the period’s own exact volatility. Busy and quiet are the top and bottom fifths of the check.
Limits. One symbol, QQQ. Minute bars are trade aggregates, not quotes. The forecast says how much the price will move, not where; and it does not say whether options are cheap or expensive for that move.
Reproduction. research/qqq_volatility_forecast_2026-09-25/qqq_vol_forecast.py writes the results this page shows (results.json); REPORT.md in the same folder has the full write-up. CPI, NFP and FOMC dates: BLS and the Federal Reserve; VIX: Cboe; earnings: the project’s Nasdaq calendar.