- Full bookDepth captured
- ~20 HzSnapshot rate
- Bids +
asksFields per side - BTC /
ETH / SOL / XRPCoverage
Order book imbalance, the relative weight of resting bids versus asks, is one of the most studied short-horizon signals in traditional markets. It is barely studied on prediction markets for one boring reason: nobody kept the historical book. We do, and this is what it tells you.
Imbalance asks one question: at this instant, is there more size resting on the bid or on the ask? When buyers crowd the bid, the next move tends to be up; when sellers stack the ask, down. On a probability-priced contract, that "next move" is a move in the implied odds.
Two ways to measure it
Top-of-book imbalance
Compare only the size resting at the best bid versus the best ask. Fast and reactive, but noisy on thin books.
- Most responsive
- Best for liquid contracts
- Sensitive to single orders
Depth-weighted imbalance
Use the total resting size across the whole book. Steadier, and harder for one large order to distort.
- Uses full-depth totals
- Smoother signal
- Robust to spoof-sized orders
Both land in the same intuitive range, fully bid-heavy at one extreme, fully ask-heavy at the other, balanced in the middle. Every snapshot already carries pre-computed depth totals, so you do not have to reconstruct the book to get the depth-weighted version.
The decay horizon
An imbalance reading is not a forecast for all time, it is a forecast for a short window. The question that separates a usable signal from a curiosity is how long the edge lasts before it fades into noise. Measure the relationship at one second, five seconds, thirty seconds, a minute, and watch it decay. The horizon where it goes flat is the horizon you have to act inside.
- Sub-secondSharpest, but dominated by quote noise
- SecondsThe usual sweet spot on liquid books
- Tens of secEdge fades as fresh flow arrives
- MinutesTypically indistinguishable from noise
The exact numbers depend on the contract and the regime, which is the entire reason to keep the history: you measure your own decay curve rather than borrowing someone else’s and hoping it transfers.
Does it actually lead price?
One claim, one number, one source
The honest way to validate a signal is to measure whether imbalance at this instant predicts the price change over the next few seconds, and to report the number, including the horizon where the edge decays to nothing. A signal that only works in a chart with no axis is not a signal. The point of keeping the history is that you can falsify the claim yourself.
Turning a reading into a trade
- 1Smooth the raw reading, single snapshots at 20 Hz are jumpy, so average over a short window.
- 2Require persistence, act on imbalance that holds for several snapshots, not a one-tick spike.
- 3Fill at the touch, buy the ask, sell the bid, and charge the spread, exactly like a real order.
- 4Respect the depth, cap your size so your own order would not move the book you just measured.
Imbalance is most informative on liquid contracts, the tight 5- and 15-minute BTC books, and least reliable on thin markets where one resting order swings the ratio. The depth data is what lets you filter to the regime where the signal is real.
Liquid versus thin regimes
The same ratio means very different things depending on how much size is behind it. On a deep book, a tilt toward the bid is dozens of orders agreeing; on a thin one, it can be a single resting order that vanishes the moment you lean on it. Treating those two cases identically is how a signal that backtests well falls apart in production.
Liquid regime
Deep, two-sided books where the imbalance reflects genuine crowding. This is where the signal earns its keep.
- Many orders per level
- Ratio is hard to fake
- Trust shorter horizons
Thin regime
Sparse books where one order dominates. Filter these out using the depth totals before you act.
- Single-order distortion
- Treat readings with caution
- Widen or stand aside
A signal is only as trustworthy as the book underneath it, measure the depth first, then decide whether the imbalance is telling you anything at all.
Test imbalance against the archive
Read how the historical snapshots expose the full bid/ask arrays behind every reading, or open the dashboard to watch the current book on an active market as a secondary check.
Frequently asked questions
Does Polymarket provide historical order book depth?
No. Polymarket exposes the live book through its public API, but it does not publish a history of how that book looked over time. Resolved Markets captures point-in-time snapshots with the full bids and asks arrays, for crypto markets at roughly 20 Hz, so you can study microstructure signals like imbalance after the fact.
What is order book imbalance?
It is the relative weight of resting buy orders versus sell orders at a given instant, more size on the bid leans bullish, more on the ask leans bearish. On prediction markets it is a candidate leading indicator for short-horizon moves in the implied probability.
Which markets is imbalance most useful on?
Liquid, frequently quoted contracts, the 5-minute and 15-minute BTC Up/Down books are the densest. On thin markets a single large resting order can dominate the ratio and produce false signals, so use the depth totals to filter to liquid regimes.
How long does an imbalance signal stay predictive?
Only for a short window, and you have to measure it rather than assume it. Evaluate the relationship between imbalance now and the price change at several horizons, a second, a few seconds, tens of seconds, and watch where the edge decays to nothing. On liquid BTC books the useful horizon is usually measured in seconds; beyond that, fresh order flow tends to wash it out. Because the full snapshot history is preserved, you can compute that decay curve directly for the exact contract you care about.