TL;DR
Order book imbalance measures how displayed size is distributed between the bid and offer sides. A common definition is bid size minus ask size divided by their sum at the top of the book, giving a number between minus one and one.
It is one of the most widely used short-horizon predictors in trading, because more resting demand than supply often precedes upward price pressure. It is also one of the easiest signals to overfit, because it is noisy, regime-dependent and heavily affected by cancellations.
Key points
- Book imbalance uses resting size; order flow imbalance uses actual events over time.
- Predictive power is short lived and varies by instrument, tick size and venue.
- Cancellations make displayed size an unreliable proxy for intent.
- Costs usually decide whether an imbalance signal is tradable.
Two different measures
| Measure | What it uses | Typical use |
|---|---|---|
| Order book imbalance | Resting size at the top levels, at a point in time | Immediate pressure estimate |
| Order flow imbalance | Net of arrivals, cancellations and trades over an interval | Directional flow estimate |
| Trade imbalance | Buyer versus seller initiated trade volume | Realised aggression |
| Queue-weighted imbalance | Size weighted by depth and distance from touch | More robust variants |
Researchers usually test several definitions, which immediately raises multiple-testing risk. Fix the definition before evaluating performance.
Why it works, when it works
Resting interest reflects willingness to trade at a price. When one side is much larger, aggressive orders on the thinner side can move the price more easily, and market makers adjust quotes to manage inventory. The effect is mechanical rather than mysterious, which is why it appears across markets.
It also decays fast. In liquid instruments the predictive horizon is often measured in milliseconds to seconds, which is why the signal is mostly useful to firms with low-latency infrastructure.
Where it breaks down
| Pitfall | What happens |
|---|---|
| Cancellation behaviour | Displayed size is withdrawn before it trades, so imbalance misleads |
| Hidden liquidity | Iceberg and hidden orders are not captured in displayed size |
| Tick size effects | In wide-tick instruments, queues are long and imbalance behaves differently |
| Regime dependence | Signal strength changes with volatility and time of day |
| Venue fragmentation | Single-venue imbalance ignores activity elsewhere |
| Cost sensitivity | The edge is often smaller than the spread plus fees |
How researchers evaluate it honestly
- Define the measure and horizon before testing.
- Test out of sample and across regimes, not just on a strong period.
- Include realistic costs and, for passive strategies, fill probability rather than assumed fills.
- Compare against a naive baseline such as recent returns.
- Check whether the effect survives in the instruments and venues you actually trade.
- Count how many variants you tried, and discount accordingly. See Sharpe and Sortino.
The fill probability trap
A backtest that assumes passive orders fill whenever the price touches their level will show excellent results from imbalance signals. In reality you are filled selectively, often when the market is about to move against you. Modelling queue position and adverse selection is what turns a plausible result into a realistic one.
Talent implications
Imbalance work is a standard first project for new quant researchers, and a standard interview topic. Interviewers use it to test whether candidates understand cancellations, fill probability and costs, rather than whether they can compute a ratio. Candidates who discuss the failure modes unprompted stand out. See the quant researcher guide.
Frequently asked questions
What is order book imbalance?
A measure of how resting size is distributed between the bid and offer sides, commonly bid size minus ask size divided by their total.
Is order book imbalance a reliable signal?
It has genuine short-horizon predictive power in many markets, but it decays quickly, varies by instrument and regime, and is easily overstated by unrealistic fill assumptions.
What is the difference between book imbalance and order flow imbalance?
Book imbalance is a snapshot of resting size. Order flow imbalance measures net activity, including arrivals, cancellations and trades, over an interval.
Why do backtests of imbalance signals look too good?
Usually because they assume passive orders always fill and ignore adverse selection, costs and cancellation behaviour.