A maker quotes both sides and accumulates inventory. What pairs into sets is safe — merge it and the spread is locked. What does not pair carries all the risk. So the metric that matters is not spread, it is pair rate — and pair rate is set almost entirely by tenor.
| Paired into a set | Naked leg | |
|---|---|---|
| Worth | exactly $1, regardless of outcome | either $1 or $0 |
| Realising it | merge at any time | wait for settlement, or sell at market |
| Risk | none — price movement is irrelevant | all of the directional risk |
| You earn | the two legs' prices summed, minus $1 | — |
Which explains something counter-intuitive: the higher your pair rate, the less your fill prices matter. A completed set is $1 whatever the market does. Conversely, at a low pair rate you are not really market making — you are taking directional positions.
Simple reason: the longer you hold, the more chances the other leg has to arrive. Truncating one real market-making wallet's flow to different holding windows:
| Holding window | Pair rate | Naked exposure |
|---|---|---|
| 5 minutes | 47.0% | 53.0% |
| 15 minutes | 70.9% | 29.1% |
| 30 minutes | 87.4% | 12.6% |
| 60 minutes | 95.4% | 4.6% |
From five minutes to an hour, naked exposure falls from over half to under 5%. That is what "tenor decides everything" means — same approach, same person, different business.
Fifty-four times. One unpaired share in every 54 fills wipes out the rebate entirely — and on five-minute boards more than half go unpaired.
This is the part of the entry worth keeping, because it overturns the part above.
Those numbers came from truncated replay: take an hourly wallet's real flow, pretend it were a five- or fifteen-minute holding window, recompute. That is an approximation — and it assumes short-tenor flow looks like long-tenor flow, which it does not.
Later I ran a full replay against recorded real books plus complete tape:
| Tenor | Truncated replay said | Full simulation | Overstated by |
|---|---|---|---|
| 5 minutes | 47.0% | 36.3% | 1.3× |
| 15 minutes | 70.9% | 33.6% | 2.1× |
| 1 hour | 95.4% | 54.9% | 1.7× |
Fifteen minutes had been "undecided" — 70.9% looked high enough to be worth trying. The full simulation returned 33.6%, worse than five minutes. Killed outright.
The problem is not that the approximation was inaccurate — approximations are. It is that I had not decided in advance which way it would err. Thinking through "does simulating short tenors from long-tenor flow overstate or understate pairing" would have marked 70.9% as an optimistic upper bound, unusable for a decision.
This is now a house rule: every replay and approximation states its bias direction. Method in replays must state their bias.
In the full simulation the hourly row looks positive (+4.31¢/share of trading return). Broken out: one board out of six contributed +$667; the other five had a median of −$19.
A common trap: small sample plus fat tail means the mean is meaningless. When you see a set of backtest results, ask for the median first, then ask what is left after removing the single best case.
| Conclusion | Note |
|---|---|
| Pair rate is the first metric | it tells you whether you are making markets or taking direction |
| Tenor is its main driver | not quoting technique — tenor |
| Static laddering loses on short tenors | 5m and 15m both ruled out |
| Discount any modelled pair rate | measured overstatement of 1.3× to 2.1× |
| Read medians, not means | one lucky board can carry an entire set of results |
Paired inventory does not care what the price does; the unpaired part carries everything. So what a maker is really betting on is not the spread — it is whether the other leg shows up. And that is decided mostly by tenor, not by how well you quote.