Queue Position & Order-Book Dynamics

By now you can price a spread and decompose a P&L. But there’s a brutal fact the textbook spread model quietly ignores: most of the time, in most liquid markets, the spread is already one tick wide and you cannot make it narrower. You can’t undercut — a better price would cross the spread and turn you into a taker. So how do two makers compete for the same fill at the same price? They compete on time. Whoever got there first gets filled first. Half of market making, at this point, stops being about price and becomes about your position in the queue. This lesson is about that queue — how it forms, how it empties, when being first is a gift and when it’s a trap.

Price–time priority and the FIFO queue

Analogy. Walk into a busy deli. Everyone wants the same thing at the same posted price, so the price isn’t how you compete — you grab a paper ticket. Number 47 is served before number 48 no matter who is hungrier or richer. The order book at a single price level is exactly that ticket line: a FIFO (first-in, first-out) queue of resting orders, served in the order they arrived.

Definition. Matching engines almost universally use price–time priority. First, orders are ranked by price (better prices match first — higher bids, lower asks). Then, among orders at the same price, they are ranked by time (earlier arrivals match first). So the book at the best bid is an ordered list:

When a marketable sell order arrives and hits that bid level, it consumes shares from the front: $o_1$ first, then $o_2$, and so on. Your order sits somewhere in that list. Everything resting ahead of you must trade or cancel before a single share reaches you.

Worked micro-example. You join the best bid at 100.00 behind 30,000 resting shares. A market sell of 12,000 shares arrives. It eats the front 12,000 of the queue. Shares ahead of you fall from 30,000 to 18,000; you got nothing. A second sell of 25,000 arrives: it eats the remaining 18,000 ahead of you, then reaches you and fills 7,000 of your shares. Only the volume that trades past the 30,000 in front of you ever touches your order.

Pitfall. Newcomers think “I’m at the best price, so I’ll get filled.” Being at the best price only buys you a spot in line. Price priority gets you to the level; time priority decides whether you eat. A huge queue at the touch can mean you wait minutes — or never fill at all before the price moves.

When it matters

Price–time priority is the dominant model on most equity, futures, and FX venues. The main alternative, pro-rata (fills split proportionally to size, common in some short-term interest-rate futures), changes the strategy entirely — there, size buys priority, not time. Know which regime your venue runs before you optimize for it.

Why queue position matters at a one-tick spread

Analogy. Imagine an auction where the law forbids anyone from bidding a cent more than the current top bid. You literally cannot outbid the leader. The only way to win is to have raised your hand before them. Competition collapses from “who pays most” to “who was earliest.”

The mechanism. The spread is the gap between best bid and best ask. The smallest legal increment is one tick. When best bid and best ask are already one tick apart, the spread is as tight as the rules allow. To “improve” your bid you’d have to raise it by a tick — but that equals the ask, so your order crosses and executes immediately against the resting ask. You’ve paid the spread and become a taker, not a maker. Posting a better price is simply impossible without abandoning the maker role.

So at a one-tick spread, every maker on the bid is stuck at the same price. Differentiation by price is off the table. The only dimension left is time — your rank in the FIFO queue. And what determines whether you ever fill is the size resting ahead of you, call it $Q_{\text{ahead}}$. You don’t fill until cumulative volume traded at your level exceeds $Q_{\text{ahead}}$ (net of cancellations in front of you).

Worked example. Two makers, A and B, both want to buy 5,000 shares at the locked best bid of 50.00.

• A posts at 09:30:00.100 when the queue ahead is 8,000 shares.
• B posts at 09:30:00.140 — 40 ms later — when the queue ahead is now 13,000 shares (A’s 5,000 slotted in front of B).

Over the next minute, 11,000 shares sell into the bid. A fills completely (8,000 ahead consumed, then A’s 5,000 — wait: 8,000 + 5,000 = 13,000 > 11,000, so A fills 3,000). B fills zero — the trade never reached B’s slot. A 40-millisecond head start was the entire difference between a partial fill and nothing.

Fill probability as a function of the queue

Analogy. Your queue ticket is a lottery whose odds depend on three flows: how fast the line moves forward (trades executing at your level), how fast people ahead of you give up and leave (cancellations help you), and whether the whole deli closes and reopens elsewhere (the price moves away and your level evaporates).

The model. Treat the level as a flow problem. Let:

• $Q_{\text{ahead}}$ = shares resting ahead of you,
• $\lambda_{\text{trade}}$ = rate (shares/sec) of marketable orders consuming the level from the front,
• $\mu_{\text{cancel}}$ = rate (shares/sec) at which orders ahead of you cancel (this also drains the queue in front of you),
• $p_{\text{move}}$ = probability per unit time the price moves away before you reach the front.

The front of the queue drains at roughly $\lambda_{\text{trade}} + \mu_{\text{cancel}}$ shares/sec. Crucially, only the trade part actually fills you; the cancel part just shortens the line ahead. Your expected time to reach the front is approximately:

and the probability you fill before the level disappears falls as $Q_{\text{ahead}}$, $p_{\text{move}}$ rise and as $\lambda_{\text{trade}}$ rises.

Worked example — show the arithmetic. You rest with $Q_{\text{ahead}} = 50{,}000$ shares. At your level, marketable flow executes at $\lambda_{\text{trade}} = 4{,}000$ shares/sec and orders ahead cancel at $\mu_{\text{cancel}} = 6{,}000$ shares/sec. The line in front drains at $4{,}000 + 6{,}000 = 10{,}000$ shares/sec, so:

But notice the split: of every 10,000 shares draining the line, only 4,000 (40%) are trades. The other 6,000 are cancellations that move you up without filling anyone. Now layer in the price move: if there’s roughly a 50% chance the price ticks away within those ~5 seconds, your unconditional fill probability is closer to $0.5 \times (\text{chance the trade flow reaches you in time})$ — well under half. A “good” 5-second wait can still be a coin flip.

Pitfall. Counting cancellations as progress toward your fill. They shorten the queue, yes — but a level emptied mostly by cancellations is a level nobody is trading, which often means the price is about to move. Fast queue drain via cancels is frequently a warning, not good news.

Queue position as an option

Analogy. A front-of-queue spot is like holding a short-dated option: it has real value right now (you’re close to a fill), and that value decays as conditions change and as competitors stack in. And just like an option is cheapest deep out-of-the-money, queue position is cheapest to acquire when a brand-new price level first forms — there’s no line yet, so the first order in is at the front for free.

The idea. When the price ticks to a new level, the queue there is empty. The first maker to post claims position #1 at zero cost. The second pays the cost of resting behind one order, and so on. As the level matures, getting a good spot becomes expensive (you must wait behind everyone). This is why HFT makers obsess over detecting new levels forming and racing an order in microseconds — the value of front position is highest exactly when the level is born and decays thereafter.

Why “option.” Your resting order is a free option for the market: you’ve given everyone else the right (not obligation) to trade against you at your price. You hold it hoping to capture the spread; you’d cancel if the world turns against you. Queue position determines the probability that option gets exercised in your favor before you’d want to pull it. Front position = high exercise probability soon = valuable; back position = the option may expire (level disappears) worthless.

Worked intuition. Two strategies on the same newly-forming level:

• Race-to-front: post within 50 µs of the level forming → queue position ≈ #1, near-certain to be early enough that trade flow reaches you. High fill rate.
• Lazy join: post 200 ms later → 40,000 shares already ahead → you’ve converted a free front spot into a 40,000-deep wait. The option you grabbed is already deep “out of the money.”

Pitfall / trade-off. The race to be early costs infrastructure (colocation, fast feeds, fast order entry) and risk: you commit to a price before you’ve seen how the level develops. You’re buying the cheap option, but you’re also the most exposed to whatever happens next — which the next section makes brutal.

The front-of-queue paradox: priority is adversely selected

Analogy. Being first in line at a store sounds great — until the only reason the doors burst open is a fire alarm, and the front of the line is first into the smoke. Front of queue means you fill first, including first when the people trading against you know something you don’t.

The paradox. Everything above says front position is valuable — it fills fastest. But what fills you matters as much as whether you fill. When informed flow decides to sweep a level (because it knows the price is about to move through you), it consumes the queue from the front. So the front-of-queue order is the first to be picked off by the toxic trade. Priority maximizes both your fill rate and your adverse selection.

Tie to adverse selection. Recall from the P&L lesson: your realized edge is spread captured minus adverse selection. Front position raises fills (good) but those fills are disproportionately the informed ones (bad). A back-of-queue order, by contrast, only fills when so much volume trades that even uninformed flow reaches it — its fills are less toxic on average, but it fills far less often.

Worked example. Suppose at the front of the queue, 70% of your fills come from informed sweeps that move 2 ticks against you, and 30% are benign. Average markout per fill might be: $0.30 \times (+1 \text{ tick}) + 0.70 \times (-2 \text{ ticks}) = +0.30 - 1.40 = -1.10$ ticks — you fill a lot and lose on average. A back-of-queue order fills a tenth as often, but only on heavy, benign-dominated flow: say $0.70 \times (+1) + 0.30 \times (-2) = +0.10$ ticks per fill — positive, but rare. Fill rate and toxicity trade off.

Pitfall. Optimizing purely for fill rate / queue position and ignoring markouts. A strategy can have a beautiful fill rate and bleed money, because the fills it wins are the ones it should have lost. The pros measure post-fill markout (price drift 1 ms / 100 ms / 1 s after the fill) precisely to detect this.

Order-book dynamics: add, cancel, replace, flicker

Analogy. The book looks like a stable ladder in a screenshot, but in motion it’s a strobe light — orders blink in and out hundreds of times a second. Most “quotes” you see were never meant to trade; they’re the market constantly re-optioning itself.

The churn. A modern book is dominated by three message types: add (post a new order), cancel (pull a resting order), and replace/modify (cancel-and-repost, often to chase price or reset queue intentions). The headline statistic: in liquid markets, the vast majority of submitted orders are cancelled before they ever execute — order-to-trade ratios of 20:1, 50:1, even higher are routine. Quoting is mostly cancelling.

Why so many cancels?

• Repricing. When the fair value moves, a maker must cancel its stale quote and repost at the new level — or get picked off at the old one.
• Fleeting / flickering quotes. Orders posted and pulled within milliseconds — sometimes to test the book, sometimes a side effect of fast requoting, sometimes (abusively) to mislead. They make the displayed book a noisy, partly-phantom signal.
• Risk management. A maker pulls all quotes instantly when its inventory or risk limits trip.

The cancel/replace race. Here’s the cruel part. When fair value jumps, you need to cancel your now-stale quote before a taker hits it. That is a race: your cancel message versus the incoming marketable order, both flying to the matching engine. Lose the race and you’re filled at a price you no longer want — classic adverse selection, and the dominant cost of being slow. This is why latency matters: not (mainly) to fill faster, but to cancel faster when you’re wrong.

Worked intuition. Fair value ticks up by 1 tick. Your resting offer is now 1 tick cheap. Two things race to the engine: (a) your cancel, (b) every fast taker’s buy order lifting your stale offer. If your cancel arrives in 5 µs and the fastest taker’s order in 8 µs, you escape. If the taker is at 3 µs, you’re run over and sell a tick too cheap. Microseconds, repeated millions of times, are the entire margin.

Order-book imbalance as a short-horizon predictor

Analogy. Stand at a tug-of-war. If one side has visibly more people straining on the rope, you can guess which way the flag is about to lurch — before it actually moves. Order-book imbalance is that headcount: far more resting size on the bid than the ask means buyers are crowding, and the next tick is more likely up.

Definition. The classic top-of-book imbalance is:

where $Q_{\text{bid}}$ and $Q_{\text{ask}}$ are the resting sizes at (or near) the best bid and ask. $I \to +1$ means the bid dwarfs the ask (pressure up); $I \to -1$ means the ask dwarfs the bid (pressure down); $I \approx 0$ is balanced. Empirically, the sign and size of $I$ predict the direction of the next mid-price move over very short horizons — one of the most robust micro-signals in all of microstructure.

Worked example — show the arithmetic. Best bid shows $Q_{\text{bid}} = 18{,}000$ shares; best ask shows $Q_{\text{ask}} = 4{,}000$:

A strongly positive imbalance: thin ask, heavy bid. The thin ask is easy to clear, the heavy bid is hard to clear, so the mid is more likely to tick up next. A maker reads this and reacts: it might skew — quote a slightly more aggressive (or larger) bid to grab fills it expects to be benign, and cancel or thin its ask to avoid being run over as the price rises. If instead $Q_{\text{bid}} = 4{,}000$ and $Q_{\text{ask}} = 18{,}000$, then $I \approx -0.64$ and the maker does the mirror image: pull/thin the bid, lean on the ask.

How makers use it. Imbalance feeds (1) micro-price estimation — a size-weighted fair value $\text{mid} + \tfrac{1}{2}\,\text{spread}\cdot I$ that leans toward the heavier side; (2) quote skewing — shading quotes in the predicted direction; and (3) defensive cancellation — pulling the side about to be swept.

Pitfall. Imbalance is a short-horizon signal and is itself gameable — a large resting order can be spoof-like or simply get cancelled the instant before it would trade. Naively chasing imbalance can mean leaning into flow that vanishes. And displayed size hides icebergs (hidden quantity), so the visible $Q$ is an estimate, not ground truth.

Tick size and the queue regime

Analogy. Tick size is the width of the stairs in a stairwell. Wide, tall steps (a large relative tick) force everyone to bunch up on the same few steps — long lines form, and being early on a step is everything. Shallow, narrow steps (a small relative tick) let people slip onto a half-step just below you — so lines stay short and you compete by stepping down a hair, not by waiting.

The two regimes. The key quantity is the tick relative to price (and to volatility), often summarized by whether the spread is frequently locked at one tick:

• Large relative tick → “queue regime.” One tick is economically meaningful, so the spread is pinned at one tick almost always. You can’t profitably undercut (a tick is too much to give up), so makers pile onto the same price and queue position dominates — time priority is the whole game. Long queues, high order-to-trade ratios, fierce races to the front. Typical of many liquid ETFs, large-cap stocks under a coarse tick, and many futures.

• Small relative tick → “spread/undercutting regime.” One tick is economically trivial, so it’s cheap to undercut: post one tick better and jump the entire queue on price priority. Makers compete on price, the spread floats above one tick and flickers, and queues are short (no point stacking when someone will undercut you in a millisecond). Typical of high-priced stocks, crypto with fine ticks, and tick-constrained-but-not regimes.

Comparison table.

Worked contrast. A 25-dollar ETF with a 1-cent tick: a 1-cent spread is 4 basis points — meaningful, so it locks at a penny and a 200,000-share queue forms. Here you live or die by queue position. Now a 2,000-dollar stock with a 1-cent tick: a penny is 0.05 bp — trivial, so makers freely undercut, the spread sits at several cents and wobbles, and queues are a few hundred shares deep. Same tick in dollars, opposite regime, because what matters is the tick relative to price.

Pitfall. Porting a queue-regime strategy (optimize position, race to front) into an undercutting regime, where someone simply posts a tick better and your hard-won front position is now behind a better price. Time priority only protects you within a price level — it’s worthless the instant a competitor improves the price. Read the regime first.

Putting it together

Key Takeaways