Ten Ways Cross-Market Analysis Goes Wrong

Every one of these produces a number that looks like a finding. Collected so you can check your own work against them.

1. Comparing prices observed at different times

You fetched one venue at 14:00:05 and the other at 14:01:20. The market moved in between. The difference you measured includes seventy-five seconds of drift.

Fix: record the venue's own timestamp on every observation and reject pairs where the two sides are more than a short interval apart. Stale quotes and timing.

2. Using last trade instead of the current quote

A last trade from yesterday next to a live quote today produces an enormous gap that does not exist. Fix: compare mids computed from live bids and offers, and record the age of the last trade as a separate field. Mid price versus last trade.

3. Ignoring the spread

Comparing mid to mid overstates what is reachable by half the spread on each side. Fix: compute the gap you could actually transact — buy at the offer, sell at the bid — and report that alongside the headline. What a spread costs you.

4. Ignoring depth

A gap on a book with thirty contracts is not a gap you can use. Fix: require a minimum size at the top of book on both sides before a pair is eligible at all. Depth and slippage.

5. Matching on title similarity

Two questions whose titles read alike can have different sources, thresholds and deadlines. Text similarity is a starting point for a matcher and a terrible basis for a conclusion. Fix: match on structure — publisher, series, threshold, direction, timestamp — and refuse to pair when structure cannot be established. Matching markets across venues.

6. Ignoring the comparison direction

"Above 3.0" against "below 3.0" is a complement, not a match. Comparing them directly produces a gap of about a hundred cents. Fix: normalise direction explicitly before comparing. Normalising across venues.

7. Comparing a binary to one leg of a multi-outcome set

If one venue has a two-way market and the other has a six-way one, the correct comparison sums a subset of the six. Comparing to a single leg understates it by whatever the other members are worth. Multi-outcome markets.

8. Forgetting the capital cost

A three-cent gap on a contract settling in eight months is not a three per cent return. It is three cents against a dollar tied up for two thirds of a year, minus everything else. Fix: always express a gap as a return over the actual holding period, not as a raw difference. Capital lockup.

9. Sorting by gap size

The largest gaps in any table are dominated by staleness, thin books and mismatched questions. Reading down a list sorted by gap size is a machine for finding artefacts. Fix: filter by equivalence grade and liquidity first, then look at what remains. How to read a gap table.

10. Calling the result profit

Even a pair that survives every filter above is a theoretical cross-market edge, not a return. It has not paid fees, not crossed spreads, not paid for the capital, and not survived the possibility that the two contracts resolve differently.

The language matters because the language is what people remember. This site reports a theoretical cross-market edge and never uses the word profit, because the distance between those two things is exactly the content of what a gap is worth.

The meta-mistake

Building the system before recording the data. Almost every impressive-looking backtest of a cross-venue strategy is built on last-trade prices with no depth filter and no timestamp discipline, and the edge it finds is made entirely of the artefacts above. Backtesting honestly.

Next

Stale quotes and timing, the single largest source of phantom gaps.