time windows make sense for stocks but what do i do at 3am in crypto

hey all, night shift person here as usual. read that time-windowed skeleton thing in the strategy library, the one that's basically just an if on HhmmSs wrapped around whatever your actual entry is. really liked it, it's the kind of "shape not strategy" idea i can actually use. and the bit about forcing flat before the close, deciding to hold overnight on purpose instead of by accident, that hit a nerve because i have absolutely done the by-accident version more than once. but i mostly trade crypto and there is no close. that's the whole thing i'm chewing on. no bell, no forced flat moment, the tape just keeps going. so a couple of questions for people who've adapted this kind of thing: 1. do you still use a window at all, or do you just let it run 24/7? my honest experience is that the hours are not the same. the stretch when both asia and europe are quiet feels thinner to me, spreads feel worse, and my little mean reversion ideas behave differently there than they do when volume shows up. so a window seems like it should help. but i've never actually tested it properly, it's vibes. 2. if you do window it, what clock? the post uses local session times which makes sense when there's one exchange. for crypto i keep going back and forth between just hardcoding UTC and thinking in "my night" which drifts around the year anyway once daylight saving messes with everyone else. 3. what replaces the forced flat? i can't say "be flat by 15:45" because 15:45 is meaningless. the closest i've got is a hard hold time cap, like if this position has been open longer than N minutes it's not doing what i thought it was doing, get out. which is already in that skeleton as one of the exit conditions, just used as the main exit instead of the backstop. does that hold up for you or does it just chop you out of the trades that would eventually work? not looking for anyone's actual parameters, more curious how people think about it. cheers.
N
NightOwl_Yuki
27 posts · 0 followers
+ Follow

Related reading

11 replies

CryptoKarl· Jun 2026 ago
honestly i do window it and it's the dead hours i cut, same instinct you have. the flat check though, i'd say don't replace it with time, replace it with a state check on restart, because my worst nights were never "held too long" they were "script died and i woke up to a position i didn't know i had". and yeah just use UTC, thinking in your own night is how you end up with a window that quietly shifts an hour twice a year and you spend a weekend wondering why the behaviour changed.
SlowSwing_Sam· Jul 2026 ago
the fact that crypto never closes is exactly why i stopped trading it at 3am. no bell means no permission to go to bed, and you'll invent one eventually. your hold time cap is basically a bell you built yourself, i'd keep it.
GrandpaGrizzly· Jul 2026 ago· edited Aug 2026 ago
Back when I traded futures overnight the mistake wasn't the hours, it was that I trusted my judgment at 3am the same as I did at 10am. You don't need a window to protect the strategy so much as to protect you from yourself.
C
CoffeeAndCharts· Jul 2026 ago
UTC, no question. Hardcode it and never think about it again. The second you start thinking in "my night" you're going to be debugging why a rule fired an hour early every spring and autumn, and that's a miserable afternoon.
MLcurious· Jul 2026 ago
dumb question maybe but couldn't you just tag every trade you've already taken with its UTC hour and look at the results bucketed by hour? that's not really a backtest but it'd at least turn the vibes into a table? i did something similar for day-of-week and half of what i believed turned out to be like nine trades.
S
SpreadWatcher· Jul 2026 ago
the spreads thing isn't vibes, it's the one part of your post that's measurable. log the quoted spread and your actual fill vs intended at entry, bucket by hour, and you'll have your answer in a couple weeks. dead hours usually cost you on the way in and again on the way out, and mean reversion is exactly the style that can't absorb that.
G
GapHunterMike· Jul 2026 ago
hold cap is fine. yes it chops some winners. it also caps the tail where the thing goes against you for six hours and you're staring at it. karl's right about restart state btw. mine has to check position before it does anything else. learned that one the hard way.
Q
QuietVol· Jul 2026 ago
Worth adding a caveat to the hour-bucketing idea: with 24 buckets you'll fragment your sample fast, and the quiet hours are exactly the ones where you have fewest trades, so those buckets will look the noisiest whether or not there's a real effect. I'd start with three or four coarse blocks instead of 24, and if something shows up, split further. Also, the clock hour is a proxy for what you actually care about, which is volume and spread regime, so you could condition on those directly and let the hours fall out. Do you have enough logged trades yet to get a decent count per bucket, or would you be reading tea leaves for a while?
R
RiskFirstRita· Jul 2026 ago· edited Aug 2026 ago
Checklist version of what's already been said, because it's the part that saves you: position state check before any entry logic on restart, spread ceiling that blocks entry rather than just widens your expectations, and smaller size in the thin block until you've got the data. The hold cap is a risk control, not an alpha filter, judge it on how it shapes your worst outcomes, not your average one. And please don't size the 3am ideas the same as the ones you take awake.
ZenTrader_Ana· Jul 2026 ago
The line that stood out to me was "i have absolutely done the by-accident version more than once". How did you feel the morning after those? For me the accidental hold was never the money, it was the whole day spent slightly rattled afterwards, and that leaks into every decision you make while tired.
N
NightOwl_Yuki· Aug 2026 ago
all good stuff, thanks. the restart state check is going in first, that's clearly the actual hole and i was busy solving a different problem. and yeah ana, rattled is exactly the word, the position size was fine but i checked the phone about forty times that day. going to log spread and fill slippage with a utc timestamp for a few weeks before i touch the window at all, and coarse blocks like quietvol says rather than 24 buckets i'll misread. keeping the hold cap either way, sam framing it as a bell i built myself kind of settled it.
Sign in to reply →
← All automated trading