Arbitrage: the free money that is mostly already gone

The first is within a single market. Yes and No must sum to one, so if you can buy both sides for less than a dollar in total, the combination pays a dollar at settlement regardless of the outcome. That is the cleanest form and the rarest, because it is the easiest to detect automatically.

The second is within a multi-outcome market. If a question has five mutually exclusive answers, their prices should sum to one. When they sum to more, the set is collectively overpriced; when they sum to less, buying all of them guarantees a profit. Long candidate lists are where this shows up, because the tail entries get stale.

The third is across venues. The same event listed on two exchanges can trade at different prices, and buying the cheap side on one while selling the expensive side on the other locks the difference. This is the most visible form and the one most often mistaken for free money.

Automated participants monitor these relationships continuously and act within milliseconds. A genuine mispricing of the first kind, on a liquid market, is closed faster than a person can read the two numbers, and by the time it appears on a screen refreshed once a second it has typically already been taken.

What remains visible is therefore filtered: the gaps you can see are disproportionately the ones the machines declined. That is worth taking seriously as evidence rather than as an invitation. If a persistent inconsistency is visible to you, the first hypothesis should be that it is not what it appears to be.

Most often the two contracts are not the same question. Two venues can list what looks like the same event with different resolution sources, different deadlines, or different treatment of edge cases - and the price difference is the market correctly pricing that difference. This is the single most common explanation, and the resolution rules guide covers how to check it.

Second, capital cannot move freely. An arbitrage across venues requires funded accounts on both, and money committed on one side is unavailable elsewhere until settlement. The gap is real and the cost of capturing it includes locking up capital for the full duration, which on a long-dated market can exceed the gap.

Third, the costs eat it. A three-cent gap against a two-cent spread on each side, plus fees on both, is not a three-cent profit. Multi-outcome arbitrage is especially exposed to this, because capturing it means paying the spread on every single leg.

Fourth, one side may not be executable at size. A quote of 40 with 50 dollars behind it and a quote of 45 with 5,000 behind it do not represent a tradable gap at any meaningful size, and the depth is not visible in a price comparison.

Even when the sum of a multi-outcome market cannot be traded profitably, it is diagnostic. A candidate list summing well above one tells you the market as a whole is overpriced relative to certainty, usually because the tail entries carry a persistent premium - people buy longshots.

That has a practical use beyond arbitrage: it tells you which side of that market is systematically expensive. If the sum is above one, the favourite is the relatively cheaper part of the set, and a view that concentrates there is fighting less of a headwind than one spread across the tail.

The library uses the same reading in the other direction. A ladder of thresholds on one series - measles case counts, tariff rates, one leader's departure at four deadlines - can be read across as an implied distribution, and inconsistencies between adjacent rungs are more informative than any single price.

Treat arbitrage as a diagnostic tool rather than a strategy. Checking whether a set of related contracts is internally consistent takes a minute, costs nothing, and regularly reveals that one of them is stale - which is a research finding even when the combination is not tradable.

If you do find something that survives the four checks above - same question, capital available, costs covered, depth on both sides - act quickly and size to the smaller side. And expect it to be rarer than the arithmetic suggests, because everything easier than that has already been taken.