The Real Economics of Prediction Market Arbitrage
Prediction market arbitrage looks like free money and mostly is not, because the fixed costs are charged per entry while the edge is charged per dollar. Here is the arithmetic that decides whether the trade is worth taking.
On a binary prediction market, YES and NO should sum to $1.00. Whoever wins, the pair pays out exactly one dollar. When the pair trades at $0.97, buying both locks in three cents of profit regardless of the outcome. No directional view, no timing, no thesis about the event.
That is the theory, and the theory is correct. The reason this is not free money is not the edge. It is everything charged before you get to it.
The edge is real and it is small
Mispricings appear constantly, usually on markets that are thin, newly listed, or resolving on something nobody is watching. They also close quickly, because this is the most obvious trade in the venue and plenty of participants are looking for it.
What survives long enough to be taken tends to be small. A scanner set to require a 2.5% edge on a typical venue will find very few entries in a month. Lower the requirement to 1.5% and the opportunity count rises substantially — but so does exposure to every cost below, because those costs do not shrink with the edge.
That threshold is the single most consequential number in the strategy, and it is a genuine tradeoff rather than a setting to optimise. Too high and you never trade. Too low and you trade into costs.
The cost nobody models: your capital is on the wrong chain
This is the part that turns a clean arithmetic problem into an operational one.
Most crypto trading capital sits on Solana. The large prediction market venues settle in USDC on Polygon. To take a position you have to move capital across chains, and a bridge transfer costs on the order of $5 in fees and takes minutes — during which the mispricing you identified may well have closed.
The fee is fixed per transfer. The edge is proportional to position size. Those two facts decide everything:
| Position | Gross edge at 1.5% | Bridge cost | Net |
|---|---|---|---|
| $20 | $0.30 | $5.00 | −$4.70 |
| $100 | $1.50 | $5.00 | −$3.50 |
| $333 | $5.00 | $5.00 | $0.00 |
| $500 | $7.50 | $5.00 | +$2.50 |
At a 1.5% edge, roughly $333 is break-even on the bridge alone — before venue fees, before slippage, before gas on the destination chain. A $20 position with a genuine, correctly identified arbitrage is a losing trade by a factor of fifteen.
This is why prediction market arbitrage is not a small-account strategy. The edge is real at every size. The economics only work at some of them.
Gas on the destination chain
Having bridged, you now need the destination chain's native token to transact — a small amount of MATIC on Polygon, in this case. It is not much, but an agent that bridges USDC and holds no MATIC is stranded: it owns the capital and cannot spend it.
Any strategy that operates across chains needs a gas reserve on each of them, not just the one it started on.
Capital is locked until resolution
An arbitrage pair is not a position you can exit at will. The profit is realised when the market resolves. Until then your capital is committed, and a strategy holding positions for up to 72 hours is not generating a 1.5% return — it is generating 1.5% over three days, which is a materially different number once you annualise it and account for the capital you could not deploy elsewhere.
Capital efficiency, not edge size, is usually what makes or breaks this strategy in practice.
What a reasonable scanner filters out
Not every market with a visible edge is worth taking:
- Low volume. A wide spread on a market with almost no 24-hour volume is not an opportunity. It is an absence of counterparties, and you will discover that when you try to fill.
- Categories you cannot price. Some markets resolve on criteria that are ambiguous, disputed, or subject to rules you have not read. A mispricing you do not understand is not an edge.
- Markets near resolution. Late in a market's life, apparent mispricings are often correct pricing of a settlement detail.
Filtering is most of the work. A scan that examines fifty markets and rejects most of them on category and volume before evaluating any prices is behaving correctly.
The obvious optimisation
If the bridge fee is fixed per transfer and the edge is proportional to size, then batching several entries into a single bridge transfer divides that fixed cost across all of them. Four entries bridged together pay $5 instead of $20.
This is the clearest available improvement to the strategy's economics, and it is worth knowing that most implementations — including, currently, ours — bridge per entry rather than in batches. The strategy is viable without it at sufficient size. It is meaningfully better with it.
What to take from this
- Fixed costs beat percentage edges at small size. This is the general lesson and it applies far beyond prediction markets. Any strategy with a per-transaction cost and a proportional return has a position size below which it cannot work, and that size is calculable before you deploy a cent.
- Work out your break-even position size first. Divide your fixed cost per entry by your expected edge. That number, not the edge, tells you whether you should run the strategy at all.
- A scanner that finds nothing most cycles is working. The edge in this strategy is in the rejections.
If you are evaluating any arbitrage product, the question worth asking is not what edge it targets. It is what the all-in cost per entry is, and what position size that implies. A product that will not answer the second question has usually not done the arithmetic.