Order-book imbalance

Order-book imbalance compares the size resting on the bid with the size resting on the offer, often across the first few levels, and updates as orders are added, canceled, or traded. Cont, Kukanov, and Stoikov studied how this imbalance lines up with short-term price changes and how the impact scales with depth. It is a microstructure measurement. It lives on horizons of seconds, and it needs the book.

Not investment advice. Nothing here is a recommendation to buy or sell anything. Backtests and historical studies are not guarantees, and a rule that worked in one sample can fail in the next. Trading can lose money. Links are for education, not endorsements. Do your own research.

Rules

Define imbalance from a public book snapshot, for example bid size minus ask size over the top levels, normalized by total size. A research rule expects the next price change to lean in the direction of the imbalance, with impact smaller when the book is deeper. Any trading version needs a latency assumption, a queue model, and a cost per fill. If you cannot see the book, you cannot run the rule.

Where AI fits

AI can reconstruct the event types, align them to trades, and test the linear impact relationship out of sample. It can also show when a simple trade-sign rule captures the same thing. It cannot invent level-two data from candlesticks. A model trained on one exchange's book should not be assumed to transfer.

Markets

Very liquid electronic futures and equities where the full book is recorded. Retail candlestick data is not enough.

Timeframes

Seconds or a small number of events, not daily bars.

Risks

Adverse selection is the business model of the people who leave resting orders. Latency, fees, and queue position erase a paper relationship. Imbalance can be spoofed or pulled. Results from one market's tick data do not survive a slower feed.

← Back to all strategies