▲ Work in progress. This site is built and researched in the open: pages, numbers and findings may change or be wrong, and the record has gaps. Not advice. More
The Lab
Ideas traders use, put to the record
← the Lab · tokenised assets · write-up
#54 Stock perps while the stock market is shut Passed, on a weak bar
Past prices; no real orders were placed. Every figure is a share, a ratio or a count.
Last checked: 21 October 2026.
Tokenised assets, part 1 of a series.
Where the idea comes from
- A perp (perpetual future) is a contract that follows a price and never expires. Crypto exchanges now list perps on single stocks and stock funds, such as Tesla, Nvidia or the S&P 500 fund. They trade all day, every day.
- The stock they follow trades only 6.5 hours a weekday, 09:30 to 16:00 New York time. So for most of the week the perp trades while its stock can't.
- How prices behave while markets are shut is an old question. Kenneth French (1980) measured the weekend in stock returns. Kenneth French and Richard Roll (1986) found prices move far less per hour while markets are shut than while they are open: most news reaches prices through trading.
- How far a perp strays from what it follows was measured for crypto by Songrun He, Asaf Manela, Omri Ross and Victor von Wachter (2022).
- Using the perps to watch the shut hours is our own framing.
The question
- While the market is shut, does the perp's move tell you how the stock will open?
- What does the perp cost and track while the market is shut?
- How rough is the reopen, and what does it do to a stop-loss?
How we tested
- The markets: stock and stock-fund perps, not the leveraged funds:
- Bybit: Apple, Amazon, Coinbase, Circle, Google, Robinhood, Meta, MicroStrategy, Nvidia, Tesla, the S&P 500 fund (SPY) and the Nasdaq-100 fund (QQQ), listed 21 April to 14 May 2026;
- Binance: Apple, Coinbase, MicroStrategy, Nvidia and Tesla, listed 28 January to 6 April 2026;
- Hyperliquid's xyz market (a market an outside builder runs on Hyperliquid): ten single stocks, from 26 March 2026, in hourly candles only.
- Each from its listing to the open on Monday 19 October 2026.
- The real stock: Yahoo's public prices, used only as the reference.
- The shut stretches: from each trading day's 16:00 close to the next trading day's 09:30 open. A stretch is:
- a night (to the next day);
- a weekend (Friday to Monday);
- or a long stretch (it holds a weekday holiday).
- Stretches holding a dividend or a split were left out: 24 across the three exchanges.
- The bar, fixed before the run: on Bybit's 12 markets, the link between the perp's weekend move and Monday's opening gap must beat 95% of shuffled weekends, in each half of the window on its own. The halves split on 20 July 2026.
The maths, step by step
1. The gap. Friday's close 200, Monday's open 204: a gap of +2%.
2. The perp's move. The perp at Friday 16:00 was 200.2, and at Monday 09:30 it was 204.1: +1.95%.
3. The share already moved. Over many weekends, how much of the gaps' spread is left once you subtract the perp's move:
share moved = 1 − spread(gap − perp's move) ÷ spread(gap).
- If the perp always moved exactly the gap, it is 100%.
- If it knew nothing, it is about 0%, or less.
4. The shuffle. Give each weekend's perp move to another weekend of the same stock, at random, 1,000 times. A real link should beat almost all of those.
What we found
1. By 09:30 the perp had moved nearly all of Monday's gap.
| weekends | match (correlation) | share already moved | |
|---|---|---|---|
| Bybit, by 09:30 | 243 | 0.997 | 99.5% |
| Binance, by 09:30 | 139 | 0.998 | 99.7% |
| Hyperliquid, by 09:00 | 245 | 0.953 | 91% |
- It beat every one of the 1,000 shuffles, in both halves. The bar was passed.
- But it was a weak bar. From 04:00 New York time the real stock trades before the open (the pre-market). By 09:30 the perp has simply followed it. So "the perp knew the gap" is mostly "the perp watched the pre-market".
2. On weekend hours alone, the perp moved about half of the gap. This reading was added after we saw the first result, so it was not part of the bar: the perp from Friday 16:00 to Monday 04:00, before the pre-market opens.
| weekends | match | share already moved | sign right | |
|---|---|---|---|---|
| Bybit, all | 243 | 0.74 | 54% | 76% |
| Bybit, first half (to July) | 102 | 0.85 | 71% | 84% |
| Bybit, second half | 141 | 0.56 | 27% | 70% |
| Binance, all | 139 | 0.73 | 52% | 74% |
| Hyperliquid, all | 245 | 0.75 | 57% | 80% |
By 04:00 the perp had moved about half of the gap, less in the second half.
- It beat every shuffle in both halves. The weekend price isn't noise: it carries real news about the open.
- But it carried much less in the second half (27% against 71%). It was about half, varying a lot, not a fixed share.
3. While shut, the perp stays close to the exchange's own index.
Each exchange prices its stock perps against an index, its own reference price. Is that index frozen at Friday's close while the market is shut?
| the index stood still in an hour | perp over its index | |
|---|---|---|
| Bybit, open hours | 0.3% of hours | +0.046% |
| Bybit, weekends | 5.6% of hours | +0.035% |
| Binance, open hours | 0% of hours | +0.045% |
| Binance, weekends | 24% of hours | +0.007% |
- Bybit's index keeps moving at weekends. It is quoted from somewhere while the stock can't trade, so its weekend premium is real.
- Binance's index stood still in about a quarter of weekend hours. For those hours, its "premium" is measured against a stale price.
- Hyperliquid's xyz market publishes no history of its index, so it can't be checked.
4. Holding a stock perp long costs funding, paid to the shorts:
| a year, on average | |
|---|---|
| Bybit | about 8% |
| Binance (to September) | about 3% |
| Hyperliquid xyz | about 8% |
- On Bybit it was higher at weekends (about 13% a year against 4.5% in open hours), but the bands overlap, so that difference isn't shown.
- Holding the real stock costs no funding. This cost comes back in a later part of the series.
5. The reopen hour is rough. In the hour after 09:30, the perp moved about 4.4 times as much as in an average hour (Bybit, after weekends; band 3.8 to 5.1 times). After a night it was 4.7 times.
6. But a weekend reached a stop less often than weekdays did. How often a stop at each distance from the Friday close was reached in the 66.5 hours to Monday 10:30, reopen included. Beside it, the same 66.5 hours from Monday 16:00 to Thursday 10:30. The stop is judged on the mark price, as our bots judge it:
| stop distance | Bybit, a weekend | Bybit, weekdays | Bitcoin, the same weekends | Ether, the same weekends |
|---|---|---|---|---|
| 1% | 87% | 97% | 95% | 100% |
| 2% | 60% | 84% | 64% | 91% |
| 3% | 39% | 70% | 27% | 55% |
| 5% | 14% | 42% | 14% | 23% |
A weekend, reopen included, reached a stop less often than the same hours on weekdays.
- Weekdays reached stops far more often, even with the reopen jump on the weekend side. Most of a stock's moving happens while its market is open, as French and Roll found in 1986.
- A stock perp's weekend was about as calm as Bitcoin's, and calmer than Ether's.
- Binance's five markets gave the same picture (3%: 46% at weekends, 81% on weekdays).
What it means
- A stock perp's weekend price carries real news, about half of Monday's gap on these months. It isn't a perfect forecast, and the share varied a lot.
- By the open, the perp has priced the pre-market. Anyone treating its 09:30 price as a forecast is reading the stock's own early trading.
- The real cost of holding a stock perp is funding: about 3–8% a year for a long, against nothing for the stock itself.
- For a stop or a grid, the weekend was not the danger. Weekdays moved more. The reopen hour was the roughest single hour.
- The index matters. On Binance, a quarter of weekend hours were priced against an index that wasn't moving.
Limits
- Months, not years: these perps are new. There are only about 25 separate weekends, because the stocks share them.
- The stocks move together, so the markets are not independent.
- The window mostly rose, which shows up as a positive premium to the last close.
- Hyperliquid is in hourly candles, read at 09:00 instead of 09:30.
- Binance's funding stops at September: its files are monthly.
- A fix before this write-up: the first run read Binance's mark and index files as its trade file (they share one file name). It was found and fixed, and everything was run again.
Update, 21 October 2026: wording
Added after this page was published. The text above is left as it was, except the wording this lists.
- Three conclusions rewritten as what the data showed, in a researcher's voice rather than as advice to the reader. The findings and figures are unchanged. Before:
- "Read it as "about half, varying a lot", not a fixed share."
- "For a stop or a grid, the weekend is not the danger. Weekdays move"
- "Check the index. On Binance, a quarter of weekend hours were priced"
Credits
- Kenneth R. French, "Stock Returns and the Weekend Effect", Journal of Financial Economics, 1980.
- Kenneth R. French and Richard Roll, "Stock Return Variances: The Arrival of Information and the Reaction of Traders", Journal of Financial Economics, 1986: prices move far less per hour while markets are shut.
- Songrun He, Asaf Manela, Omri Ross and Victor von Wachter, "Fundamentals of Perpetual Futures", working paper, 2022.
- Dimitris N. Politis and Joseph P. Romano, "The Stationary Bootstrap", Journal of the American Statistical Association, 1994: the bands.
- R. A. Fisher, The Design of Experiments, Oliver and Boyd, 1935: the shuffled weekends.
- Our own: using the perps to watch the shut hours, the stretches, the stop test and the reading at 04:00.
- Data: Bybit's public market data (trade, mark and index prices, funding); Binance's public archive; Hyperliquid's public data; Yahoo's public stock prices, as the reference only.
- PerpGremlin and Claude, Stock perps while the stock market is shut, TerminalGremlin, 2026: the tests, the code and the words.
Educational research from public market data; any trading discussed is paper, never real. Nothing here is advice. The full notice.