The Halt Ladder: 21,971 Volatility Pauses, and the Reopen Is the Exit, Not the Entry

TL;DR - The most-taught low-float day trade is "halt and go": a small stock runs into its limit-up band, gets paused for five minutes, and you buy the reopen for the next leg. We tested it on every volatility pause in the official halt record from 2 January 2024 to 30 September 2026 - 21,971 pauses in 2,239 common stocks across 688 sessions - with the hypotheses written down and hashed before any return was tabulated. It loses. Buying one bar after a limit-up reopen returns -1.63% over fifteen minutes (t = -8.12, n = 10,167), the median trade is -3.19%, and it wins 38.2% of the time. It loses in both halves of the sample, in all three calendar years, on every rung of the ladder and in every share-count tier. The continuation traders are chasing does exist: the reopening print lands +3.22% above the last print before the pause (t = 38.5) and is higher on 69.4% of reopenings. But that gap is printed in a single auction while nobody can trade, and from the reopening print itself the next fifteen minutes return -0.76% (t = -3.76). The move belongs to whoever was long before the pause. A placebo says the pause itself is doing this: the same stocks making a 6-9.5% five-minute move without being paused return +0.25% over the same window, a difference of 1.86 points (t = -6.34). The ladder does not improve with height - after a first limit-up pause the next pause is limit-up 20.0% of the time and limit-down 20.6%; after a fourth it is 33.0% against 37.0%. And nothing here is a short either: at the quoted spread - a median 0.90% each way - shorting the reopen nets +0.40% (t = 0.48). One rule survives all of it: if you are long into a limit-up pause, the reopening auction is statistically the best print you will see.

There is a trade that every low-float momentum course teaches in the first week. A small stock catches a headline, runs twenty or forty percent, and hits its Limit Up-Limit Down band. Trading stops for five minutes. The chat room fills with rocket emojis. When it reopens, you buy, because a stock strong enough to get halted up is strong enough to keep going.

It is a clean, mechanical, testable claim, and the halt record is public. So we tested it.

The set-up

Every trading pause on a US exchange is published, with its halt time and resumption time to the second. We took the complete record for 2 January 2024 through 30 September 2026 and kept the rows coded as Limit Up-Limit Down pauses: 28,979 of them. Dropping ETFs, leveraged products, warrants, rights, units and preferreds leaves 25,573 in common stock and ADSs.

The record was pulled from two exchange sources independently and compared. Across the 457 sessions both cover, they agree on 15,171 of 15,233 Nasdaq-listed pauses, with identical resumption times on 99.8%.

Each pause was then joined to one-minute bars from the consolidated tape for that stock and session, pre-market included. A pause stays in the study only if the tape is alive around it: a print in the ten minutes before, a print within five minutes of the reopen, and a resumption before 16:00. That leaves 21,971 pauses, 2,239 symbols, 8,238 symbol-sessions.

Three definitions carry the rest of the article:

• Direction. The record does not say which band was hit. A pause is limit-up if the last print before it sits above the mean of the prior five one-minute closes, limit-down if below. The deviation sits where the bands are: on 89.5% of classified pauses it is 5% or more, so the label is rarely ambiguous. 10,320 up, 10,475 down, 1,176 unclassifiable and set aside.
• Entry. The close of the first one-minute bar that starts at or after the resumption time. That is a median 89 seconds after the reopen. No fill in the reopening auction is assumed, and no fill in the first seconds of a market that has just reopened.
• Exit. The close of the bar 5, 15 or 30 minutes later, or 15:59. If the stock is paused again at the exit time, the first print after that.

The hypotheses, the entry and exit rules, the inference method and the bar for calling anything tradable were written down and hashed before any return was computed. Pauses in one stock on one day are not independent, so every t-statistic below is clustered by symbol-session.

Here is what one of these sessions looks like.

Who gets paused

This is a low-float phenomenon, and the record says so plainly.

• 89.6% of the pauses are in Nasdaq-listed stocks. The median price at the pause is $5.18, and the median stock traded $0.91M a day over the prior twenty sessions.
• Where the issuer's own filings give a usable share count (5,979 pauses), 72.5% are in companies with fewer than 20 million shares outstanding and 44.1% in companies with fewer than 5 million. The median is 6.8 million shares and a $27.4M market cap.
• At the moment of a limit-up pause, the median stock is already +43.6% on the day.
• 27.1% of all pauses happen between 09:30 and 10:00.
• Pauses come in clusters. 45.7% of symbol-sessions have more than one, and 352 sessions have ten or more. 24.5% of reopenings are paused again within two minutes.

That last number is the ladder: pause, reopen, pause, reopen, each rung a few percent above the last. It is what the trade is built on.

Finding 1: buying the reopen loses

All 10,320 limit-up pauses, long from the entry bar:

Hold	n	Mean	t-stat	Median	Win rate
5 minutes	10,273	-0.99%	-8.69	-1.61%	40.9%
15 minutes	10,167	-1.63%	-8.12	-3.19%	38.2%
30 minutes	9,944	-2.21%	-7.24	-4.43%	36.5%
To the close	10,298	-1.29%	-0.59	-8.38%	32.5%

The hypothesis we registered was that this number would be positive. It is negative at every horizon, and at the three intraday horizons it is nowhere near zero. Held to the close, the mean is saved from significance by a handful of multi-hundred-percent runners; the median trade is down 8.4% and two in three lose.

It does not depend on how you cut it:

Cut	n	Mean, 15 min	t-stat
All limit-up pauses	10,167	-1.63%	-8.12
Unambiguous direction only (deviation ≥ 5%)	9,043	-1.76%	-8.40
First limit-up pause of each symbol-session only	5,506	-1.92%	-8.93
Entries not interrupted by a new pause	7,831	-1.74%	-8.91
Jan 2024 - May 2025	4,823	-1.92%	-7.56
Jun 2025 - Sep 2026	5,344	-1.36%	-4.49
2024 / 2025 / 2026	3,249 / 3,847 / 3,071	-2.00% / -1.33% / -1.60%	-6.48 / -4.00 / -4.03

We also looked, after the fact, for any condition that rescues it - time of day, how far the stock was up on the day, price, pause length, prior liquidity, dollars traded before the pause. Thirty-one buckets across six variables, and every one of them has a negative mean. Cutting by listing venue and by market cap adds eight more; the only two that are not negative are NYSE American listings (+0.08%, t = 0.10) and companies above $2B (+0.37%, t = 0.30), neither distinguishable from zero. The worst is a stock paused when it is 25-50% up on the day: -3.64% in fifteen minutes (t = -7.14), negative in both halves of the sample.

Finding 2: the continuation is real, and it is in the print you cannot buy

So is the lore simply wrong? No. That is what makes this worth writing up.

Measure the same pauses from the last print before the pause instead of from the buyer's entry, and the picture inverts:

Leg	Mean	t-stat	Median
Last print before the pause → reopening print	+3.22%	38.5	+2.06%
Last print before the pause → 15 min after the reopen	+2.37%	10.7	-0.62%
Reopening print → 15 min later	-0.76%	-3.76	-2.79%
Reopening print → 30 min later	-1.50%	-4.96	-4.64%

The stock reopens higher than it was paused 69.4% of the time. The entire continuation - all of it, and then some - is printed in one auction at the end of five minutes in which nobody could trade. A five-minute pause does not cool the move. It compresses the next leg into a single print.

Whoever was long before the pause owns that print. Whoever buys afterwards pays for it, and then pays again: the first minute or so after the reopen tends to be the local high, which is why entering 89 seconds later (-1.63%) does worse than entering at the reopening print itself (-0.76%).

It also means the holder's best average outcome is the first one available. Selling the reopening print is worth +3.22% on average with a positive median. Holding fifteen more minutes is worth +2.37% with a negative median, and thirty minutes +1.61% with a median of -2.40%. In the 4,496 sessions whose first pause was limit-up, 65.0% closed below the price at which the stock was first paused, with a median of -6.5%.

Finding 3: the ladder does not get better as it gets taller

The second registered hypothesis was about rungs: if the first limit-up pause is the cleanest, the edge should decay as the ladder climbs.

Rung (limit-up pauses so far)	n	Mean, 15 min	t-stat	Median	Win rate
1st	5,506	-1.92%	-8.93	-2.89%	37.5%
2nd	1,885	-1.93%	-4.83	-4.12%	37.0%
3rd	967	-1.21%	-1.91	-4.04%	37.4%
4th	555	-0.79%	-0.93	-3.88%	39.5%
5th and later	1,254	-0.57%	-0.85	-2.89%	42.7%

That is not what we predicted. The mean drifts toward zero as the ladder climbs while the median stays put at three to four percent under water. A stock on its fifth limit-up pause is a survivor of four, and survivors have fatter right tails - but the typical outcome does not improve and no rung is positive.

The cleaner way to see the ladder is to ask what comes next.

After a first limit-up pause, the next pause inside thirty minutes is limit-up 20.0% of the time and limit-down 20.6%. After a fourth: 33.0% and 37.0%. The limit-down bar is at least as tall as the limit-up bar on every rung. A taller ladder is more likely to do something next, not more likely to do the thing you are long for.

And the payoffs are symmetric. With hindsight, the reopen-long made +8.57% in fifteen minutes when the next pause turned out to be limit-up, lost 9.47% when it turned out to be limit-down, and lost 2.46% when nothing followed. A coin that is slightly against you, paying about nine either way, with a drag in the middle.

Finding 4: it is the pause, not the move

An obvious objection: maybe any stock that has just run 10% in five minutes gives some back, and the pause is incidental.

So we built the placebo. Same stocks, same sessions, priced at $3 or more, between 09:45 and 15:35 when the band is a flat 10%. Find every moment the price closed 6% to 9.5% above its five-minute mean - most of the way to the band - and no pause followed. Enter two bars later, hold fifteen minutes.

	n	Mean, 15 min	t-stat	Median
Near-miss: the move, no pause	6,166	+0.25%	1.61	-1.27%
Limit-up pause, same price and time window	5,963	-1.62%	-6.25	-3.02%
Difference		-1.86 pts	-6.34

Without the pause the same kind of move in the same stocks is roughly flat. With it, the buyer loses. The pause is not a neutral timeout. It is the mechanism that hands the next leg to existing holders in one print and leaves the reopen as the local high.

What did not work, reported anyway

Two more hypotheses were registered. Neither found anything, and they are here because a study that only reports its hits is not a study.

Float rotation. The lore says a float that has "rotated" - traded more than its own size on the day - behaves differently. There is no point-in-time float for this population, so we used the stricter version: session volume to the pause divided by shares outstanding from the issuer's latest SEC cover page filed before the pause, restated for splits and no more than 135 days old. That covers only 2,851 of the 10,320 limit-up pauses. Below one turn, -2.16% (t = -5.50). Above one turn, -1.12% (t = -1.87), with a worse median (-4.19% against -2.58%). Across six rotation buckets the sign never turns reliably positive and the order is not monotone. No usable signal, on data that is frankly too noisy to say more: share counts in serial reverse-splitters go stale within weeks.

What the share counts do show is a gradient in the typical outcome. The median fifteen-minute result is -4.10% in companies under 5M shares, -3.31% at 5-20M, -1.86% at 20-100M and -1.18% above 100M, with the win rate rising from 36.4% to 42.4%. The lower the share count, the worse the reopen-long - the opposite of where the trade is taught.

The reopening gap. We expected a reopen above the last print to keep going and a reopen below it to fade. The correlation between the reopening gap and the next fifteen minutes is -0.006. All six gap buckets are negative, five of them significantly; a reopen more than 10% below the last print is the least bad at -0.22% (n = 290). The gap tells you nothing about what follows it.

The spread, and why this is not a short

If buying loses 1.63%, shorting makes 1.63%. On paper.

We pulled the NBBO at the entry print and at the fifteen-minute exit print for a random sample of 1,026 pauses, 487 of them limit-up.

• Median half-spread at entry: 0.90% of mid. At the exit: 0.79%.
• Round trip, crossing both times: 1.93% median, 3.93% mean. One entry quote in ten is more than 4.66% from its own mid.

Buy the offer one bar after the reopen and sell the bid fifteen minutes later, and the limit-up trade loses 4.97% (t = -6.51), wins 29.2% of the time, and is negative in both halves of the sample.

Reverse it - hit the bid, cover at the offer - and the short nets +0.40% (t = 0.48) with a median of exactly zero and a 49.5% win rate. The spread is the size of the edge. And that is before the two things the quote cannot show. These are hard-to-borrow names where a locate may not exist at any price. And on the limit-down side, a stock that has traded 10% below its prior close is under the short-sale price test for the rest of that day and the next, so the short cannot hit the bid at all. We have covered both in hard-to-borrow mechanics and why four locate vendors is not redundancy.

The limit-down side is no better. A limit-down pause reopens a further 2.86% lower on average (t = -33.8), then keeps drifting - another -0.68% in fifteen minutes from the entry bar (t = -3.76). After a pause in either direction the price falls. Shorting it at the quote nets -0.53% (t = -0.74).

What we would actually take from this

• The reopen is the exit. If you are long a low-float name going into a limit-up pause, the reopening print is on average the best price of the next half hour: +3.22% against the pause price, and above it 69.4% of the time. Holding through costs you the mean and the median. Both the Nasdaq halt cross and the NYSE reopening auction accept orders during the pause - which is the case for having direct access to the auction rather than a market order that fires a second after it.
• Do not buy the first two minutes. The reopening print is a local high on average and the bar after it is worse. If the thesis is a multi-day one, the pause has not handed you an entry; it has handed someone else an exit.
• The rung count is not a signal. Second pause, fourth pause, sixth - the next one is a coin flip between up and down at every height, tilted slightly down.
• Size for the path, not the average. Inside thirty minutes of the entry, the median best point of a limit-up trade was +10.3% and the median worst point -12.3%. One in twenty finished thirty minutes below -32%; one in twenty above +37%. A stop inside that range is not risk management, it is a coin toss on which side prints first - the same result we found in the time-stop premium.
• A loser is not a short. A negative-expectancy long does not become a positive-expectancy short when the spread is the size of the edge and the borrow is not there.
• The band is the level. If you want to trade the approach rather than the aftermath, the arithmetic of where the band sits is in the halt tripwire.

Method and caveats

Sources. Pauses are from the exchanges' published trade-halt records (the NYSE all-market historical file as primary, the Nasdaq Trader record as the cross-check). Prices are one-minute consolidated-tape (SIP) bars, raw and unadjusted, 04:00-20:00 ET. Quotes are the SIP NBBO. Share counts are XBRL cover-page facts from each issuer's own SEC filings, keyed on the filing date so nothing is used before it was public.

Pre-registration. Hypotheses, entry and exit rules, clustering and the tradability bar were fixed and hashed before any return was computed. Two addenda were hashed later and are disclosed in full: the switch of primary source (the first feed throttled mid-download), a carry-forward rule for the direction reference when no trade printed in the prior five minutes, and the choice of shares outstanding as the float proxy - all before outcomes - and, after the first tabulation, a robustness cut for re-pauses. The first tabulation and the final one give the same headline: -1.63%, t = -8.12.

Direction is inferred. It is the sign of the last print against the prior five-minute mean, not a field in the record. Restricting to deviations of 5% or more strengthens the result. The 1,176 pauses where the two were equal are excluded.

The entry is conservative, not optimal. It is 60-120 seconds after the reopen in most cases and later when the stock is re-paused immediately, which happens after 26.6% of limit-up reopenings. Results from the reopening print itself are given separately and are also negative.

Trade prints are not fills. The headline uses bar closes, which carry bid-ask bounce. The quote sample is the check on that: it is 1,026 pauses, not 21,971, and its own mid-to-mid mean is noisy (-0.53%, t = -0.46, for limit-up pauses) because the distribution is that fat. Read the spread from it, not a second estimate of the drift.

Share counts are the weakest data here. 604 of the 2,372 symbols no longer map to a current SEC registrant, so the tier and rotation tables lean toward companies that still exist. Counts under 200,000 shares, or implying a market cap under $1M or over $500B, were treated as filing errors and dropped - a filter added after the first tabulation, with the rotation result null either way. Nothing price-based depends on that mapping: the halt record and the tape include every delisted name.

What was dropped. 76 pauses with no bars at all, 1,038 with no print in the ten minutes before, 1,948 with none in the five minutes after the reopen, 531 outside the 09:30-15:50 window or resuming after the close. These are the least liquid events in an illiquid population; including them could only make the cost side worse.

The exploratory cuts are exploratory. Thirty-nine buckets were examined after the registered tests. That none of them is significantly positive is a weaker claim than a registered result - but it is the claim.

Nothing here is a recommendation to trade any security. It is a measurement of what happened to one widely taught entry over 33 months.