i stopped trusting pre-market volume numbers

been doing this a long time. used to scalp opens, now i mostly sit on gap setups and wait. my whole process starts the night before. i pull a list, i mark levels, i go to bed. morning is just execution. no scanning at 9:28, that's how you end up chasing. the thing that changed for me was pre-market volume. i used to size off it. thin tape, wide spread, but the number said decent volume so i figured there was real interest. got burned enough times to look closer. turns out depending on where the numbers come from you're not seeing the same thing. some feeds show a lot more early prints than others. odd lots, dark prints reported late, timestamps that don't line up. so the same symbol at the same minute can look like two different stocks. now i don't size off the volume figure at all. i watch spread and how fast the bid refreshes. if the spread is a nickel wide on a five dollar stock i don't care what the volume says, i'm not there. curious what other people use as a liquidity read before the bell. i'm using spread plus how deep the book looks and that's basically it. feels crude. probably is. anyone tracking something better.
G
GapHunterMike
17 posts · 0 followers
+ Follow

Related reading

5 replies

CryptoKarl· May 2026 ago
spread + book depth isn't crude, it's the actual thing. volume is a lagging summary of what already happened, spread is what you're going to pay right now. in perps i basically live off the same read, if the book gets thin i cut size before i cut the idea.
MLcurious· May 2026 ago
ok this is helpful because i've been doing exactly the wrong thing lol. i've been feeding pre-market volume into a feature set for a little classifier i'm messing with and getting results that look great in backtest and awful live. never occurred to me the data itself might be inconsistent across sources. does that mean historical pre-market volume is basically unusable for training, or is it fine as long as i stay on one source the whole time? genuinely not sure how to check.
N
NightOwl_Yuki· May 2026 ago
overnight in crypto has the same problem in a different outfit, one venue looks busy and it's the same handful of bots passing size back and forth. i just watch how far the book moves when something real hits it.
BacktestBetty· May 2026 ago
@MLcurious that's a classic one, two things worth checking before you touch the model again: 1) is your backtest volume the same source as your live feed? if the vendor stitches in late-reported prints after the fact, your historical bars know things the 9:15 live bar can't. that alone will make a feature look predictive when it's just leakage. 2) even if the source matches, ask whether the feature survives if you replace it with a coarse bucket (thin/ok/heavy). if the edge only exists at fine resolution you're fitting noise in a number that isn't stable across days anyway. i'd rather have a crude feature that behaves the same in and out of sample than a precise one that doesn't.
Q
QuietVol· Jun 2026 ago· edited Aug 2026 ago
Agreed that spread isn't crude, it's just a lower-variance measurement than volume. Pre-market prints are a small, unevenly reported sample, so the count is noisy and the noise isn't even the same shape from feed to feed; quoted spread is observed directly and updates continuously. One thing I'd add to your read: it's not only the level of the spread but how much it moves. A nickel that stays a nickel is a different animal from one flickering between two and eight cents. Do you look at spread stability at all, or just the snapshot when you check?
Sign in to reply →
← All market data & research