Crypto maker adverse selection is the risk that a resting order fills disproportionately when the market is moving against that order. Measure price changes after actual fills, but retain unfilled orders in a separate decision ledger. A lower maker fee and an exact limit-price fill do not establish better execution, because the favorable opportunities may be the ones that never trade.
This AI-assisted educational article develops an original hypothetical worksheet for evaluating that selection effect. It concerns passive orders on a central limit order book, not liquidity provision in an automated market maker pool.
A fill changes which market paths you observe
Imagine leaving a buy at 100 while the market can rise or fall. If buyers lift the offers and prices move upward, your bid may remain unfilled. If sellers consume the bids and prices move downward, your order may buy just before the next lower quote. The completed purchases can therefore have a different price-path distribution from all the moments when you wanted to buy.
The mechanism does not require identifying an informed counterparty. Stale quotes, competing traders reacting faster, ordinary order flow and changes in available liquidity can all contribute. A negative observation alone cannot prove why someone traded with you.
In a live experiment on the Binance Bitcoin perpetual, researchers documented a negative relationship between maker fill likelihood and subsequent returns conditional on filling. Their orders, cancellation policy and short observation horizons define the evidence; the result is not a universal forecast for every crypto pair or passive strategy. See the primary paper, The Market Maker’s Dilemma.
There is also a queue between touching a price and executing your order. Coinbase Exchange, for example, specifies price-time priority at its matching engine. An observed trade at your limit does not establish that sufficient eligible quantity reached your place in line. Coinbase Exchange’s matching-engine documentation describes that venue’s matching rules. Other products require their own rule check.
For the placement permissions themselves, use the post-only and time-in-force guide. Here the question starts after a valid instruction: which opportunities actually become holdings?
Define the markout before calculating it
A markout compares an execution price with a later reference price. For this worksheet, use the midpoint of the best bid and ask on the same venue and instrument, observed 30 seconds after each fill.
Define side = +1 for a buy and side = −1 for a sell. Then:
gross markout, basis points = side × (later midpoint − fill price) ÷ fill price × 10,000
A positive result is favorable to the executed side. A buy followed by a lower midpoint is negative; a sell followed by a lower midpoint is positive. This stated convention avoids accidentally reversing half the sample.
The measure includes the fill’s position relative to the later midpoint. To inspect subsequent movement separately, also record the midpoint immediately around execution and its change to the later observation. Do not label every difference from the fill price as new information arriving afterward.
Midpoint marking is not a liquidation quote. Selling an acquired position may require crossing a spread, consuming depth and paying another fee. The slippage and order-book depth worksheet covers those additional execution questions. A markout is a diagnostic, not realized round-trip profit.
Four intentions, two fills
Assume four independent, invented ABC/USDT opportunities. Each begins with a best bid of 100.00 and best ask of 100.02. Each intention is a one-ABC passive buy at 100.00 with a five-second deadline. A and B fill exactly five seconds after the decision; C and D are confirmed canceled without fills at that deadline.
Observe every episode’s midpoint 35 seconds after its original decision. That gives a 30-second post-fill horizon for A and B, and a common decision horizon for all four episodes. Assume a maker entry fee of one basis point, paid in USDT. All rates, prices, quantities and outcomes are hypothetical.
| Opportunity | Result at five seconds | Midpoint at 35 seconds | Gross fill markout | After entry fee |
|---|---|---|---|---|
| A | Bought 1 ABC at 100.00 | 99.90 | −10 bp | −11 bp |
| B | Bought 1 ABC at 100.00 | 100.02 | +2 bp | +1 bp |
| C | No fill; canceled | 100.30 | Not applicable | Not applicable |
| D | No fill; canceled | 100.20 | Not applicable | Not applicable |
The fill rate is 2 ÷ 4 = 50%. Since the filled orders have equal notionals, their average gross markout is (−10 + 2) ÷ 2 = −4 bp. Subtracting the assumed one-basis-point entry fee gives −5 bp.
The dollar check agrees: the two purchases cost 200 USDT before fees, their separate later marks total 99.90 + 100.02 = 199.92, and entry fees total 0.02. The aggregate marked difference after entry fees is −0.10 USDT, or −5 bp of the 200 USDT filled notional. These are separate episodes combined for an accounting check, not a simultaneously tradable portfolio.
C and D have no fill markout. Recording them as zero would make an average look better by mixing observed executions with nonexecutions. Dropping them from the decision ledger would conceal that the rising episodes were missed.
Make the unfilled alternative explicit
For C, an imagined immediate purchase at the initial 100.02 ask would be worth 0.28 USDT more at the later midpoint; D’s corresponding difference would be 0.18. Those are hypothetical gross marked outcomes. They are not profits the passive policy earned or losses its account realized.
This alternative assumes sufficient ask quantity, timely arrival, a small order with no effect on the subsequent path and an immediate fill at the displayed price. A real taker comparison must model arrival, depth, fees and the same original intention. The alternative could also have bought during A’s decline. Selecting only the missed winners would introduce another bias.
Write the fallback rule before evaluation: cancel and abandon, cross immediately, or reconsider under a fresh decision. Each rule produces different exposure. Preserve the original benchmark when several replacement orders pursue one intention, and do not count retries as independent opportunities.
Changing visible size is another policy change that can alter the observed sample. The iceberg visible-size guide separates the displayed slice from total exposure and replenishment rules. Neither a smaller display nor a maker label establishes protection from adverse selection.
Keep the failure cases in the report
Partial fills need fill-level horizons and an explicit weighting rule. Multiply per-unit dollar outcomes by filled quantity; weight an aggregate basis-point markout by filled notional when execution prices differ. A one-unit favorable fill should not automatically receive the same economic weight as a ten-unit adverse fill. Retain the unfilled remainder separately, following the partial-fill inventory ledger.
A market-data outage at the target observation time is a missing mark, not a zero return. Record coverage and the rule for selecting the first valid observation after the horizon, including its actual delay. Otherwise a nominal 30-second comparison can silently mix very different intervals.
Cancellation timing can also change the sample. A quote withdrawn before a decline may look successful operationally while lowering the measured fill rate. Record the withdrawal decision, confirmed terminal outcome and any intervening fills. A backtest that fills every touched limit omits precisely the selection and queue questions being investigated; see the backtest evidence guide.
Finally, entry quality and exit protection are different obligations. A later conditional exit may itself be deferred under a venue’s guard conditions, as explained in stop-order price protection. An unfavorable entry markout does not establish that an exit was triggered or completed.
A reusable maker-policy worksheet
For each original intention, retain the venue, instrument, side, size, decision timestamp, initial bid and ask, submission and cancellation policy, distinct fills, fees and terminal remainder. For every fill, attach the reference source, horizon, actual observation timestamp and signed markout. For unfilled intentions, attach the predefined alternative and its assumptions without inventing an executed holding.
Report completion, fill-conditioned markouts, missing observations and the alternative-policy result separately. Repeat at predefined horizons appropriate to the strategy, rather than choosing whichever later timestamp makes the fills look best. A small sample or one market regime cannot establish a stable edge. The useful conclusion is which observed costs and selection effects the policy must overcome, with the remaining uncertainty still visible.