1-Minute Day Trading on QQQ: 586 Variants Tested, None Held Up After Costs

A simple idea has circulated among day traders for years: price moves around a rolling midline, stretches too far away from it, and then snaps back — catch that snap-back and you should have a reliable small edge. On the 1-minute chart, this can look strikingly clean in hindsight. We tested the idea completely: across 15-plus years, on the Nasdaq-100 ETF QQQ, in every shape we could think of.

The result in four numbers

• 0 of 586 fully computed rule variants (entry, exit, stop management, time window, position size) stayed reliably profitable after costs. Four variants did end up in the black, but on at most 36 trades — below the minimum of 100 fixed in advance. On top came 339 purpose-built filter conditions meant to rescue the rule — not one held up under cross-checking.
• 15.4 years of market history (23 March 2011 to 25 August 2026), 1,506,408 one-minute candles inside regular trading hours, with individual test runs covering up to 11,016 trades.
• +$5.15 per trade before costs was the best candidate the entire search turned up with an adequate sample size (a short sale with the freshness filter, 461 trades) — the fixed round trip of $8.00 on a 400-share position ate even that: net, −$2.85 per trade remained.
• 0 of 15 years shown ended with the main variant in the black.

In one sentence: across every variant, time window, and filter, we found no tradable edge for this rule on the 1-minute chart.

The idea: bands around a moving middle

The rule works with three lines. The midline is the 20-minute moving average — where price "normally" sits. The outer bands sit above and below it, as far away as the largest price swing of the last 200 minutes — marking how far price has recently stretched from its middle. The outer 40% of the distance between the midline and the outer band counts as the "overextended" zone. The entry: price must have come from the opposite side, dipped into the outer zone, and turned back toward the middle.

All of these rules were fixed before the first result appeared. The pre-registered configuration of this study, put into words:

"Trading rules (pre-registered): midline over 20 minutes, band over 200 minutes, overextended zone the outer 40% (sensitivity variant 50%), entry no earlier than 10:30 a.m. New York time, flat at the open of the 15:55 candle, spread 1 cent per share, commission 0.5 cents per share with a $1.00 minimum per order, position sizes $10,000 and $50,000."

On a chart, this can look like a dependable pattern — you notice the clean, textbook reversal waves that confirm the rule. What's much harder to notice just by eye is the far larger number of failed signals: prices that dip into the zone, appear to turn, and then keep going until the stop fires. The eye remembers the hits better than the misses — a classic selection effect. As a hard check, we ran a random-entry control: same entry times, same exit mechanics, but a randomly chosen trading day instead of a rule-based signal. After costs the result was practically identical to the real rule (typical trade −0.033% against −0.038%); before costs, chance actually did better. So the pattern delivered no measurable edge over a coin flip.

Everything we tested

We worked through the idea step by step to give it every realistic chance:

1. Base rule — 96 variants. Long and short, two trigger variants, two zone widths, two stop styles, three exit types, and two time windows.
2. Base-rule add-on — another 96 variants. The same set-up, but with a stop that is never trailed and an exit at the first candle running against the trade.
3. Refined entry — 324 variants. Instead of any touch of the zone, we required a "real" wave: two, four or five candles inside the zone covering 85%, 90% or the full distance to the outer band. Entry only on the way back, once price left the zone again.
4. Ten exit styles. Including fixed profit targets (+0.04%, +0.08%, +0.12%), partial sells after the first, second, and third candle in the trade's direction, and an exit that locks in breakeven.
5. A freshness filter — 36 variants. Price had to have been on the opposite side of the band shortly before entry; windows of 10, 20 and 30 minutes were tested. This is the only filter in the whole project that provably improved the signal itself — but only on short sales.
6. A fixed share count — 24 variants. The core recipes again with 400 shares per position and a cost model that charges spread and commission per share and per order instead of a flat percentage.

That comes to 586 fully computed variants. On top of them came a large filter search, which does not count towards the 586: 24 additional features (time of day, weekday, wave shape, band width, reward-to-risk ratio, and more), yielding 339 individual filter conditions, checked with a split-half test: every condition was searched and tuned only on the first half of the history, then checked, unadjusted, against the second, previously unseen half.

None of it pushed the rule into the black.

The core variant, year by year

The main variant fixed in advance (long, entry only from 10:30 a.m. New York time) was negative in every one of 15 years. We deliberately leave out 2020 here, because the COVID crash distorts that single year's numbers so heavily that it says nothing tradable about the other years.

The main variant fixed in advance (long, entry from 10:30 a.m. New York time), year by year, after costs. *2026 is a partial year, measured through August 25. The year 2020 is left out. Source: QQQ minute prices, own calculations.

The typical trade (median) is the middle case: half of all trades landed above it, half below. In 15 of 15 years shown, it was negative — not just on average, but in the typical single trade.

Variant ranking (shortened)

Even the "least bad" and the worst of the 96 base variants all landed in the red — the gap between the best and worst variant is small. The 0.0% hit rate shown for the best variants is not an error: those variants trail the stop candle by candle and set a distant target. Almost every trade therefore ends with a tiny loss and practically none in the black — "least bad" here means smallest losses, not most wins.

The three best and three worst of the 96 base variants, ranked by the median per trade after costs. Source: QQQ minute prices, own calculations.

The average (everything added up and spread evenly, with individual outliers pulling it sharply up or down) departs from the median in both directions here: for the three best variants it sits between −0.033% and −0.035%, below the typical trade of −0.027%; for the three worst it sits between −0.025% and −0.029%, well above the typical trade of −0.045% to −0.046%. There, a few large winners lift the average without rescuing the typical trade — both measures stay in the red everywhere.

Four lessons from a clean failed attempt

Costs act like a wall at the one-minute level. Even 2 cents per share from spread and commission is enough to make a realistic-looking profit target practically unreachable on paper. Example: at a 400-share position, one round trip of buy and sell always costs exactly $8.00 — commission and spread combined. A fixed sell target of 0.04% was hit gross (before costs) on nearly every second trade. Net, out of 5,858 trades, it was hit exactly zero times. For 0.04% to actually arrive net, the target would have had to rise to roughly 0.066% gross on a $10,000 position — on a $50,000 position, around 0.052% would have sufficed, thanks to lower costs per trade. What looks like a small, frequently achievable gain on paper is often already a mathematical dead end after costs.

Tight stops die on noise. 98.7% of all trades exited at the stop, the remaining 1.3% at the time limit — and 64.1% of all trades were stopped out within the very first minute after entry (3,513 out of 5,483). A stop glued to the low of the trigger candle sits so close to price that ordinary one-minute noise sets it off before the underlying idea even has a chance to play out.

Partial exits cost small positions twice over. Every individual sell order pays the minimum commission again. For a variant with three partial sells (a third each after the first, second, and third candle in the trade's direction), that added roughly 0.016 percentage points per trade on a $10,000 position — enough to drop it to the bottom of the ranking. At $50,000, the same markup shrank to roughly 0.002 percentage points, because the commission barely registered against the larger position. Proof that fees, not the market, drove the difference: before costs the partial-sell variant was actually about 4 percentage points ahead of the variant that sold everything at once — after costs it was about 60 percentage points behind. Those 4 and 60 percentage points are summed across all trades, not per trade.

Try enough filters and one will always look great in hindsight. We tested 339 filter conditions built from 24 features. With that many attempts, roughly half can be expected to look good on any given stretch of history purely by chance — the way that in a large enough group, someone almost always flips heads several times in a row, if enough people are flipping. That's why we searched and tuned every condition exclusively on the first half of the history (2011–2018), then checked it, unadjusted, against the second, unseen half (2019–2026). Only two of the 339 conditions survived that split-half test. Both failed the next check: for both, a single trade carried most of the apparent profit — 92% and 240% (more than the entire profit) respectively rested on one single trade in June 2026. Not a pattern — a one-off that happened to fall inside the tested sample.

A cost example: why the same rule looks "cheaper" over time

Because QQQ's share price rose over the years, the same fixed cost per trade weighs less and less in percentage terms — without the underlying result changing at all.

The same fixed $8.00 round trip, measured against the average value of a 400-share position in that year; position value rounded to ten dollars. Source: QQQ minute prices, own calculations.

Even with this steadily cheaper-looking cost base, none of the tested variants stayed profitable across the full history. Individual years of individual variants did end up in the black after costs — the best case returned $3,144 in one year — but no variant did so across all the years.

Limits of this study

A single security was tested, the Nasdaq-100 ETF QQQ — the finding applies to this instrument first of all, not automatically to every other stock. Execution was assumed deliberately conservatively: entry at the open of the candle following the signal, and whenever a stop and a target fell inside the same candle, the stop was assumed to trigger first. The calculation contains no taxes, no financing costs and no market impact from one's own orders. The year 2020 is kept out of descriptions and averages; 2026, as the current year, is captured only through August 25.

The calculation itself passed 47 automated checks, including hand computations of the band formula, the wave logic and the cost arithmetic against the raw candles, plus a reproducible random-entry control.

What remains

On the 1-minute chart, this mean-reversion idea showed no tradable edge in any of the variants we tested. This isn't an isolated case: it matches the finding from our intraday study of five classic day-trading recipes, which reached the same picture for other short-term patterns on a minute-by-minute basis — after costs, what was left was usually nothing beyond chance. Further pre-registered backtests of short-term trading patterns — among them the Velez open-trade study, Gajjala day trading and episodic pivots — are collected in our study archive.

Methodology note: Testing used QQQ minute data from March 2011 onward, strictly within regular trading hours (9:30 a.m.–4:00 p.m. New York time), with no overnight positions. Costs were modeled as 0.5 cents of spread per side plus commission.

Note: This article is a retrospective statistical analysis of historical price data, not investment advice and not a recommendation to buy or sell. Past results say nothing about the future. Trading costs, taxes and the market impact of one's own orders can materially worsen real-world results.