Order-Flow Imbalance

Order-flow imbalance is the simplest good answer to the question "which way is the next tick going to go?" Look at how much size is resting to buy versus to sell at the top of the order book. If far more shares are waiting to buy than to sell, the thin sell side gets eaten first and the price ticks up. That lopsidedness — the imbalance — turns out to be one of the most reliable very-short-horizon signals in all of trading.

The snapshot version

The quickest measure just compares the size resting at the best bid, $Q^b$, with the size at the best ask, $Q^a$:

$$I = \frac{Q^b - Q^a}{Q^b + Q^a}.$$

Here $Q^b$ is the number of shares people are offering to buy at the highest bid price, $Q^a$ is the shares offered for sale at the lowest ask, and $I$ is the imbalance. It always sits between $-1$ and $+1$. A value near $+1$ means the book is almost all bids (heavy buying pressure, thin selling), and near $-1$ means the opposite. Zero means the two sides are balanced. The logic is mechanical: whichever side is thinner will be consumed first by incoming market orders, so a bid-heavy book ($I$ close to $+1$) tends to tick up.

The event version, and why price is linear in it

The snapshot imbalance is a state; the sharper tool tracks change. The event-based order-flow imbalance (Cont, Kukanov & Stoikov, 2014) adds up the signed size changes at the touch over a short window: arriving bids and cancelled asks push it up, arriving asks and cancelled bids push it down. The striking empirical result is that the price change over that window is close to a straight line in this quantity,

$$\Delta P \approx \beta \cdot \text{OFI},$$

where $\text{OFI}$ is the net order-flow imbalance over the window and $\beta$ is a stock-specific slope roughly equal to one tick divided by the average queue size. In plain terms: net a thousand shares of buying pressure into the book and the price moves up by a predictable amount. This single regression explains far more of short-horizon price movement than trade volume does, which is why it anchors the quoting logic in The Avellaneda-Stoikov Model and short-horizon Latency & High-Frequency Trading strategies.

Worked example

The top of a book reads $Q^b = 900$ shares bid, $Q^a = 600$ shares offered. The snapshot imbalance is

$$I = \frac{900 - 600}{900 + 600} = \frac{300}{1500} = +0.20,$$

a modest tilt toward buying. Now watch the next second of events: 400 new shares are added to the bid, 150 shares of the ask are cancelled, and a 100-share market buy lifts part of the offer. The net pressure into the book is strongly positive — more demand piling up, supply pulling away — so both the snapshot and the event measures agree the tick should be up. If $\beta$ for this name is one tick ($0.01$) per 1,500 shares of net imbalance, then a net $\text{OFI}$ of about $+600$ shares predicts a move of $0.01 \times 600/1500 = 0.004$, i.e. the mid drifting up roughly four-tenths of a tick before the next full uptick prints.

Where it misleads

• Spoofing. Displayed size can be fake. A trader who posts large bids purely to tilt the imbalance signal, intending to cancel before they fill, manufactures a false read. This is illegal, but it happens, and it is why raw imbalance is noisier than it looks.
• It decays in milliseconds. The signal is real but perishable; by the time a slow system reacts, the imbalance has already been arbitraged into the price. It is a latency game.
• Only the touch is shown. Snapshot imbalance ignores hidden and iceberg size (Iceberg and Hidden Orders) and depth beyond the best level, so the visible book is a censored view of true pressure.
• Regime-dependent. The signal is strongest in large-tick names where queues are long and price competition is frozen, and weakest where the spread floats freely.

Order-flow imbalance is where the raw plumbing of the book becomes a tradeable edge, feeding directly into queue-position decisions, market-maker quoting, and the broader question of how Price Discovery actually happens tick by tick.