Markout vs rebate: what your fills actually pay
On a tick-floored book, post-fill markout dominates the maker rebate by an order of magnitude. A multi-hour measurement on a major CEX perp.
Every passive maker order has the same two-line economics on the way in: you collect the rebate (or pay the maker fee), and you accept the adverse-selection cost embedded in the fill. Whether the strategy works depends entirely on which line is bigger. This is a short note on what those two lines actually look like at retail scale on a single-tick-spread venue.
TL;DR
- Over a multi-hour session on a major CEX perp, a few hundred fills: mean per-fill markout landed in the −1 to −2 bp band against a maker rebate around +0.5 bp. Net per-fill economics before any other cost: roughly −1 bp.
- The session spent essentially all of its minutes at the 1-tick spread floor. Spread capture as a PnL term was structurally zero. Whatever the strategy was supposed to earn, it had to earn from inventory-skew alpha, not gross spread.
- Markout did not reduce meaningfully when filtering for “benign” regimes (low volatility, flat inventory, neutral basis). The adverse-selection cost looks venue-level, not regime-level.
- Reaching break-even on this surface requires the rebate to grow, the markout to shrink, or the operator to stop competing for fills they shouldn't want.
What “markout” means here
Markout is the mid-price move from the moment of fill to a fixed horizon afterwards, signed against the position. For a bid that fills, markout is (mid_at_t+Δ − fill_price) / fill_price; for an ask, the negative of that. Positive markout means the position moved in your favour after the fill; negative means it moved against you. We use a 5-second horizon for the numbers below — short enough that the measurement isn't dominated by ambient drift, long enough that the tick-by-tick noise has averaged out.
Markout is not theoretical PnL. It is the realised cost the market charges you for being the resting side at the moment a trade happens. A negative markout band means informed flow is hitting your quotes and the price is moving away from you immediately afterwards. A positive markout band means your fills sit in front of mean-reverting moves.
The measurement
Single major-perp symbol, single CEX venue, a multi-hour continuous session, a few hundred fills. Maker rebate at the retail tier this account quoted sat around +0.5 bp per fill (positive — venue pays you). Cancel-race-flagged fills (those that filled within milliseconds of an outbound cancel request) are included in the sample; pure market-moves fills are not separately classified.
| Quantity | Value | Notes |
|---|---|---|
| Session length | ~20 h | Continuous |
| Fills | several hundred | Both sides |
| 1-tick spread exposure | ~all session | Effectively all the time |
| Maker rebate | ~+0.5 bp | Per fill, retail tier |
| Mean per-fill markout (5 s) | −1 to −2 bp | Signed against position |
| Net per-fill before spread capture | ~−1 bp | Rebate − |markout| |
Spread capture is structurally zero
SUI-USDT-SWAP on OKX has a sufficiently tight tick floor that the quoted spread is one tick essentially all of the time. A passive maker who quotes both sides at the inside is not capturing a spread in the textbook sense — they are accepting whatever fills the book sends them at the tick. There is no “sell at ask, buy at bid, pocket the difference” on this surface, because the difference between ask and bid is a single tick and your two-sided quotes rarely fill in close succession.
This is the same observation the earlier spread-feasibility note made about Aster, GRVT, and Bluefin — restated for a major CEX perp. Tick-floored books are economically structured against spread-capture strategies. Whatever the maker earns has to come from somewhere else: skew alpha, queue position, cross-venue lead. None of those are free.
Markout is venue-level, not regime-level
The instinct on a losing session is to look for “the bad regime” — high vol, stretched basis, full inventory — and attribute the markout to it. We bucketed the fills by basis state, inventory state, volatility quintile, and quote-engine mode, and computed per-bucket markout. The result was anticlimactic: the per-bucket markout numbers cluster within a few tenths of a bp of the session mean. There is no benign regime hiding inside the average.
That matches the observation in the retail MM economics note: on a small cross-symbol study at another venue, markouts also clustered in a tight band around −1 to −2 bp at the venue level, with symbol-level variation essentially noise. Adverse selection appears to be a property of the flow you face, not of the timing you face it at.
What this means operationally
A few cheap reframings:
- Treat the maker rebate as the budget for adverse selection, not the prize. If markout sits in the −1 to −2 bp band and rebate is sub-bp, every fill is a roughly 1 bp loss in expectation before any inventory move. The strategy then has to either avoid the bad fills or find compensating alpha elsewhere.
- “Quote tighter to get more fills” is the wrong direction. More fills at a negative net is more loss. The right direction is fewer, more-selected fills.
- Selection is the lever. Skipping quotes in regimes you can identify as bad (basis stretch with rising vol, one-sided inventory after a string of same-side fills, momentary cross-venue lead) is the only retail-scale degree of freedom that moves the markout number. Tuning quote width does not.
What this does not measure
One symbol, one venue, one session. We did not run a multi-symbol comparison on OKX, nor compare maker rebates across venues at scale. The 5 s markout horizon is a choice — at 1 s the number is noisier but typically larger in magnitude; at 30 s it converges toward ambient drift and stops being informative as a per-fill cost. The rebate figure is the as-quoted retail tier; designated-maker rebates at venue-program scale will obviously change the equation, which is the whole point of the earlier economics note.
The narrow claim is: on this surface, at this scale, the rebate does not cover the markout. The wider claim — that this holds for retail-scale passive market making on tick-floored major perps generally — is consistent with everything else we have measured but rests on a small number of sessions. Treat it as directional, not proven.