How does your order placement change on a venue that allocates pro rata within a price level instead of first in, first out?

What each rule makes valuable

Under price-time priority, better prices trade first and, within a price, earlier arrival trades first. The consequence is that a resting order holds a position in a queue, and that position has economic value: if you are at the front you receive the whole of a small aggressor, and if you are behind you receive nothing while quoting the same price and carrying the same risk. Everything about latency in equity markets follows from that single asymmetry, and so does the reluctance to touch a resting order, because amending its price or increasing its size sends it to the back.

Pro rata allocation removes that. An incoming aggressor is divided among the resting orders at the price in proportion to their size, so who arrived first is irrelevant, or nearly so. The asset is no longer your place in the queue; it is your share of the size at the level. Every behaviour that made sense under the first rule needs re-deriving, and the two that change most are how much you show and how freely you move it.

Level holds four resting bids at 99.50, total 2,000 lots.
An aggressor sells 500 lots.

  Order   Size    Arrived    FIFO fill        Pro rata fill
  A         100   09:30:01         100        25    = 500 x 100/2000
  B       1,200   09:31:40         400        300   = 500 x 1200/2000
  C         200   09:32:15           0        50
  D         500   09:33:02           0        125

Under FIFO the earliest two orders take everything.
Under pro rata all four are filled, in proportion to size, and the
ninety-nine-second head start A holds over B is worth nothing at all.

Read the two right-hand columns as two different businesses. FIFO gives lumpy, all-or-nothing fills that reward being fast. Pro rata gives everybody a smooth fraction of every aggressor, which rewards being large.

The rational response is to show more than you want

If you receive your share of the level, your expected fill is your size divided by the total size at that price, multiplied by the volume that arrives. Resting 200 lots on a level holding 2,000 gets you a tenth of every aggressor. If you want a materially larger fill, the only lever available is a larger displayed size, because you cannot buy your way forward in a queue that does not exist.

So participants quote sizes above their genuine interest, and the displayed book systematically overstates real intention. This is not manipulation and it is not irrational; it is what the allocation rule pays for. But it converts a modelling problem into a risk problem, and the risk is specific: an aggressor large enough to consume the whole level fills your displayed size in full, not your intended size. Over-display works exactly until the day it does not.

Two controls follow. First, a cap on total displayed quantity per instrument and per side that is measured against the position you are actually willing to hold, not against the position you expect from an average day's flow. Second, awareness that over-display compounds across price levels: showing 500 at three prices when you wanted 200 in total means a sweep through all three leaves you with 1,500. The quantity you can be given is the sum of everything you are showing, and a system that reasons about each level independently has no view of that total.

Repricing becomes cheap, and message rates rise

Under price-time, a cancel and replace is expensive because you surrender queue position you may have waited minutes to acquire, which is why a repegging algorithm on an equity venue is quietly paying a real cost. Under pro rata there is no position to surrender. Cancelling and reposting the same size costs you nothing beyond the round trip, so quotes chase the market far more readily.

The engineering consequences are direct. Message rates on pro-rata markets are higher relative to trade counts, and the ratio of cancels and replaces to fills is higher, so feed handling and any per-message capacity planning must be sized from captured data on those venues rather than by analogy with an equity feed. It also means that a reduction in your size takes effect on your allocation share immediately, which makes size the parameter your quoting logic modulates continuously rather than something set when the order is placed.

Allocation is not exactly proportional, and the details bite

A candidate who models pure proportional allocation will build a fill simulator that disagrees with the venue, and that disagreement propagates into every backtest and every child-order size the algorithm chooses.

Proportions do not divide evenly. Allocating 500 lots across four orders produces fractions, so venues define a rounding convention, a minimum allocation below which an order receives nothing, and a residual pass that distributes the lots left over after rounding, often by size or by time. When aggressor sizes are small relative to the number of resting orders, those rules dominate the outcome rather than qualifying it: a two-lot aggressor cannot be split proportionally across forty resting orders at all.

Many venues also run hybrids rather than a pure rule. A common shape allocates a defined portion of each aggressor by time priority, or reserves an allocation for the first order at the level, and distributes the remainder pro rata; some assign a share to designated liquidity providers under their obligations. A hybrid restores part of the value of being early, so the optimal behaviour sits between the two extremes, and which part it restores is a per-instrument fact from the venue's rulebook. Pro-rata allocation is most common in interest-rate futures and similar contracts, and the same venue can run different rules on different products, which is why the allocation rule belongs in your instrument reference data rather than in an assumption in the code.

What this does to fill modelling and to measurement

Execution modelling under price-time is largely queue modelling: where am I in the queue, how much has traded ahead of me, what is my probability of reaching the front before the price moves. None of that transfers. Under pro rata the model is a share-of-flow model, so the quantity to estimate is your proportion of the level and the arrival rate of aggressors, and the uncertainty is concentrated in what everyone else chooses to display rather than in the sequence of arrivals.

Measurement changes too. Because every resting order is filled a little at a time, one parent order produces many more small executions, which raises the count of fills to be processed, reported and cleared, and makes per-fill costs and post-trade volumes materially higher for the same traded quantity. Any comparison of two venues that looks at slippage while ignoring the number of fills is comparing partial costs.

Where a price-time intuition breaks silently

The dangerous part is that most of a price-time intuition still appears to work. Better prices still trade first, the book still looks the same on a screen, and an algorithm written for equities will place orders and receive fills without erroring. What has changed is invisible: the algorithm is paying for speed it is not being rewarded for, treating displayed size as intention when it is a bid for allocation, and modelling its own fill probability with a queue that has no bearing on the outcome. The strong answer names that the allocation rule is a per-product configuration item which changes placement, sizing, cancellation economics and the risk cap, and treats an assumption of price-time priority as a bug to be found rather than a default to be relied on.