Can the observed loss be decomposed cleanly into vig versus residual mispricing?

Why the question matters

Publication 001 measured a market that is calibrated and still loses money, and observed that the loss looks close to the fee. Looking close to something is not the same as being accounted for by it.

The distinction matters for anyone testing a model. If the loss is the cost of transacting, the bar a strategy has to clear is a property of the price it transacts at. If part of the loss is something else, that part is what a model would be trying to find.

Nobody had tested it. The publication said so plainly and left the question open, which is why it became the first Pro Note rather than a footnote.

Where it comes from

This note answers question 3 of Publication 001's own "What we would investigate next" section. It uses the same sealed fact layer, the same de-vig path and the same event-clustered interval method, so its numbers sit alongside the publication's rather than competing with them.

Read Publication 001, free and in full

What a Pro member can do with the answer

1. Benchmark against the right thing

   How to score a candidate model at the price it would actually transact at, and why beating the market's no-vig estimate is a diagnostic rather than a result.

2. Measure the hurdle where it applies

   Why a single pooled cost figure misstates the bar for any strategy that concentrates in part of the price range, and how to compute it per selection instead.

3. Deprioritise with evidence

   Which unconditional hypothesis about the closing moneyline this note makes a poorer use of research time, and which questions it leaves entirely open.

4. Set an evidence bar

   The three conditions an apparent subgroup effect has to meet before it counts as a finding, and why inspecting many cohorts without multiplicity control makes an apparent one likely.

5. Know where to look next

   Why the open-versus-close comparison is the vein this result points at, and what it would take to measure it.

6. Know what not to conclude

   The inferences this note does not support, including the unit confusion that makes Publication 001's margin figure look comparable to this one when it is not.

The note itself carries the measured result, its interval and the guidance above in full. None of those figures appear on this page.

The underlying data is public

Every closing price this note is built from is queryable in the Historical Explorer, free and without an account.

Explore the underlying closing-price data