Order-Book Imbalance and the Microprice Explained

A limit order book records eligible buy and sell orders at prices for one instrument and trading venue. The best bid is the highest displayed buying price; the best ask is the lowest displayed selling price. The difference is the quoted spread. The displayed size beside each quote is the quantity currently shown at that price in the data feed being viewed.

That last clause matters. A screen may show only the best quotes, several price levels, or an exchange-specific order book. The displayed size is not every possible source of interest, a promise that the quantity will remain, or a record of shares already traded. Nasdaq's rulebook, for example, defines its depth feeds in terms of orders and quotes displayed in the Nasdaq Market Center. SEC material also distinguishes order-based feeds from level-book feeds and explains that reconstructing posted liquidity from an order-based feed requires applying additions, cancellations, modifications, and executions. See the Nasdaq depth-of-book definitions and the SEC's order-book reporting methods.

Before calculating anything, write down the instrument or contract month, venue or feed coverage, quote timestamp, and whether the screen reports shares, contracts, or another unit. A value built from one venue's queue should not be described as a complete market-wide balance. The futures order-book guide covers the separate question of how venue rules allocate an incoming order among resting orders.

Let Qb be displayed size at the best bid and Qa displayed size at the best ask. A common normalized measure is:

I = (Qb - Qa) / (Qb + Qa)

With nonnegative sizes and a positive total, I ranges from -1 to +1. Zero means equal displayed size on the two sides. A positive value means more size is displayed at the bid; a negative value means more is displayed at the ask. If the bid shows 90 shares and the ask 30, then I = (90 - 30) / (90 + 30) = 0.50. The result says that the bid queue is larger in this snapshot. It does not say that buyers just traded more shares, that 90 shares will remain, or that the next price change must be up.

The guide to futures bid size versus ask size discusses a similar comparison for futures quotes. Keep the normalized value separate from the raw sizes: a 90-to-30 imbalance and a 9,000-to-3,000 imbalance both equal 0.50, but their displayed depth and potential execution context are very different. If one side is absent, the total is zero, the feed is stale, or the quote is locked or crossed, the ordinary two-sided calculation may not be meaningful; treat that observation as a data-quality or market-state case rather than forcing a signal from it.

Indicator conventions vary. Some sources report the bid share Qb / (Qb + Qa) on a 0-to-1 scale. For the 90-to-30 example, that is 0.75, while the centered formula above is +0.50. Gould and Bonart distinguish the two normalizations in their study; match the formula before comparing values.

The ordinary midpoint ignores quote size: m = (Pb + Pa) / 2, where Pb is the best bid and Pa is the best ask. A size-weighted midpoint gives more weight to the price opposite the larger queue:

w = (Pa × Qb + Pb × Qa) / (Qb + Qa)

Using s = Pa - Pb and the imbalance above, the same calculation can be written as w = m + (s / 2) × I. This form makes the direction and scale of the adjustment visible. If both queues are equal, I is zero and the weighted midpoint equals the midpoint. If bid size is larger, the estimate moves above the midpoint, toward the ask; if ask size is larger, it moves below the midpoint, toward the bid. The adjustment is bounded by half the quoted spread when both displayed sizes are nonnegative.

The weighting can look counterintuitive at first. A larger bid queue does not pull the estimate toward the bid price. Instead, it increases the weight on the ask-side price in this estimator. This is a convention intended to reflect how queue imbalance relates to subsequent midpoint changes in a limit-order-book model. It is an estimate constructed from quotes, not a new quote submitted by a buyer or seller.

Assume a hypothetical best bid of $99.98 with 90 shares displayed and a best ask of $100.02 with 30 shares displayed. The midpoint is $100.00, the spread is $0.04, and the queue imbalance is +0.50. The size-weighted midpoint is:

($100.02 × 90 + $99.98 × 30) / 120 = $100.01

The shortcut gives the same result: $100.00 + ($0.04 / 2 × 0.50) = $100.01. If the displayed sizes reverse to 30 at the bid and 90 at the ask while prices stay fixed, the midpoint remains $100.00 but the weighted midpoint becomes $99.99. The two-cent swing in the weighted value comes entirely from the size inputs; no midpoint trade or quote change is required.

This is why a useful chart or log should show the ordinary midpoint, spread, both queue sizes, imbalance, weighted midpoint, and timestamp together. Recording only the weighted value hides whether it moved because prices changed, size changed, the spread widened, or the feed updated. The illustration after this section is conceptual: the larger bid-side queue shifts the weighted marker while the center midpoint stays in place. The markers are not data or a forecast.

In casual use, “microprice” is sometimes used for the size-weighted midpoint just calculated. Research uses the term more specifically for estimators of a future or efficient midpoint conditional on information in the book. Sasha Stoikov's paper defines a microprice from a sequence of expected future midprices and estimates its midpoint adjustment using observed spread and imbalance states. That requires historical high-frequency observations and a fitted relationship; the two-quote arithmetic above is a simpler weighted-mid benchmark, not the full estimator. Read the author's microprice paper before equating one with the other.

The distinction matters in code, backtests, and reporting. Label a formula by what it actually uses: “top-of-book weighted midpoint” if it uses only current best prices and sizes; “microprice estimate” only if it implements the stated model and estimation procedure. A value that looks precise to a cent may still be a noisy reference. It should not be called fair value without saying which data and model produced it, what horizon it targets, and how it was validated.

Queue imbalance is a state: it compares the displayed sizes that are present at one instant. Order-flow imbalance (OFI) is a sequence measure. It tracks how supply and demand at the best quotes change as orders arrive, trades consume liquidity, and resting orders are canceled or changed. If 90 shares are displayed at the bid and 30 at the ask, the queue ratio alone cannot tell whether the bid size was just added, has persisted for a while, or was repeatedly replenished after trades.

Cont, Kukanov, and Stoikov studied order-book events for 50 U.S. stocks and reported a short-interval relationship between OFI and price changes whose slope varied inversely with market depth. That is a result about event flows and a particular dataset; it is not the same as proving that a single size snapshot predicts a profitable trade. Gould and Bonart studied queue imbalance and next-midpoint direction for 10 Nasdaq stocks. Their classifier results differed between large-tick and small-tick stocks. Neither study turns a positive imbalance into a universal rule, and directional classification does not by itself account for spread crossing, fees, queue position, latency, or failed fills. See Cont, Kukanov, and Stoikov and Gould and Bonart.

Two book screens can disagree without either being “the market.” One may be a single venue's depth, another a consolidated best bid and offer, and another a delayed or sampled display. A full-depth product can add price levels while still covering only its named execution venue. Hidden or reserve quantities may not appear as displayed size. A queue can also change between a snapshot and an order arriving at the venue. Nasdaq's depth-feed definitions and the SEC's feed-methodology note make these boundaries explicit; check the exact data product rather than assuming every platform uses the same book.

Compare values only when they refer to the same instrument, side convention, venue coverage, session, timestamp, and unit. If one feed reports shares and another reports round lots, contracts, or aggregated depth, normalize before comparing. Do not combine a best bid from one instant with an ask size from a later update. For a futures contract, note the precise month and matching market; for an option, contract counts do not directly equal underlying share exposure. A metric with mismatched inputs may produce a mathematically valid number that has no coherent market meaning.

For a discretionary check, first confirm a valid two-sided quote and the spread. Calculate the queue imbalance and weighted midpoint from a single synchronized snapshot. Then watch whether the observation persists or vanishes as quotes and trades update. Compare it with recent midpoint changes and executed order flow; do not interpret the color or size of one ladder as proof of buyer or seller intent. The futures spread and slippage guide explains why a reference value and the price available for an actual order are different things.

For a model, define the forecast target before looking at results: next midpoint move, movement over a fixed horizon, probability of a fill, or execution cost are different outcomes. Keep an out-of-sample period, account for tick size and spread, and include fees, rebates, queue priority, latency, partial fills, cancellations, and the cost of crossing the market. Test by symbol, venue, session, and market condition; compare against a baseline that uses only the midpoint or spread. If imbalance only classifies a tiny next-tick move but cannot cover its execution costs, statistical association is not a usable edge.

The practical role is narrow but useful: a queue-weighted reference can add context to a quote snapshot and help structure a research question about short-horizon price pressure. It does not select a trade direction, guarantee an order fill, or replace a clear execution benchmark. Record which formula and feed you used so another person can reproduce the result.