43,185 halt reopens: the short side survives, the long side does not
When a US stock moves far enough and fast enough, the exchange stops trading it for five minutes. That pause is mechanical, not a judgement, and everyone can see it happen. We wanted to know what the reopen is worth on each side, measured properly.
The data
Every limit-up LULD halt in the NYSE's own consolidated halt file from January 2020 to August 2026 — 73,190 halt records, 43,185 of them reopens we could match to 1-minute bars. Delisted tickers included, so nothing is flattered by survivorship. 0.3% charged per leg. In-sample 2020-2024; 2025-2026 held back and looked at once, at the end.
The short side
Shorting the reopen of the day's first limit-up halt, with a 10% stop, held to the close:
It stays positive out of sample, and it survives having its three best trades removed. There is nothing clever in it: it is a fade of a forced, mechanical event.
The long side
The obvious-looking trade is the opposite one — buy the continuation after the second or third halt, because that is where the runners run. It loses in every form we could build:
| Version | Per trade | Win rate | Years positive |
|---|---|---|---|
| Long after halt #2+, simple stop | −$24 | 23% | 0 of 6 |
| Structural stop, volatility trail, 1/3 partial at +50% | −$24 | 23% | 0 of 6 |
| 72 further configurations (entry halt #1, inter-halt interval, reopen jump size, time of day × six exits) | 0 passed | ||
One slice looked alive: halt #2+ entered between 10:30 and 12:00 with a 25% stop, +$77 a trade, positive in all six years. Then we looked closer. Ten trades produce 92% of the profit and five of those ten are the same ticker on the same day. Remove the three best and half the total disappears. That is a lottery ticket wearing a strategy's clothes.
Two mistakes, ours
What decides whether any of this is tradeable
Borrow. These are exactly the names brokers will not lend, or will lend at 0.2-1% a day. A live desk refused two of our four locate requests in one week. A backtest that does not remove the names you could not borrow, and does not charge for the rest, is measuring a market you cannot trade in. Paper accounts hide this completely: on paper, everything is easy to borrow.
The data, free
The NYSE publishes its consolidated halt file but nobody packages it into something you can use. Here it is cleaned: every halt from February 2019 to September 2026, one row per halt, with the times parsed to minutes, the duration computed and LULD pauses flagged. No sign-up, no attribution required — take it and run your own tests.
⬇ nyse_halts_2019_2026.csv — 73,189 halts, 8.7 MB
Columns: halt_date, symbol, name, exchange, reason, halt_time,
resume_date, resume_time, halt_minute_utc, resume_minute_utc, is_luld, duration_min. Source:
the NYSE's own file at nyse.com/api/trade-halts/historical/download.