Pricing pairs against live books
The gap you can see is almost never the gap you can keep. Every figure here is priced by walking both order books to 500 contracts and netting off taker fees on both legs plus settlement basis.
| Contract | Buy | Sell | Gross | Friction | Net edge | Size | Ann. | Closes |
|---|---|---|---|---|---|---|---|---|
Will J.D. Vance win the 2028 Republican presidential nomination? 21 minutes old | 0.4300 Kalshi | 0.4840 Polymarket | +5.40pp | −3.54pp | +1.86pp | 500 | 2% | 757.0d |
Will Alexandria Ocasio-Cortez win the 2028 Democratic presidential nomination? 21 minutes old | 0.1400 Kalshi | 0.1770 Polymarket | +3.70pp | −1.94pp | +1.76pp | 500 | 6% | 757.0d |
Will Bernie Sanders win the 2028 Democratic presidential nomination? 21 minutes old | 0.0050 Kalshi | 0.0220 Polymarket | +1.70pp | −0.27pp | +1.43pp | 500 | 138% | 757.0d |
These figures are 15 minutes old, and they are the 3 largest of 60 matched pairs. On Trader every pair is priced against live books, ranked by what you would actually keep.
57 pairs are not shown here, and a gap this old has usually either closed or moved.
What each plan includes →Swipe the table sideways to see every column.
Only pairs scoring 0.8 or above are priced. That gate is set from measurement, not caution. Pricing 24 pairs against live books produced a sharp boundary: every pair at 87% or above was genuine, and every pair at 75% or below compared structurally different questions: a Polymarket contract on winning the 2028 presidential election against a Kalshi contract on winning the Democratic nomination, or winning a Brazilian election against merely advancing in it.
Those false pairs showed roughly twice the apparent spread of the genuine ones, mean gross 1.22pp against 0.64pp, because comparing two different events manufactures edge out of nothing. In an arbitrage surface that is the dangerous failure, not the harmless one: it prints money that was never there. At most 24 pairs are priced per pass, ranked by match confidence.
Taker fee = ceil(0.07 x contracts x P x (1-P)) in cents. Peaks at P=0.50 (1.75c per contract) and falls to near zero at the tails. Secondary sources report a halved multiplier on S&P 500 and Nasdaq-100 markets and a higher one on crypto; unverified against the schedule itself, so not modelled.
Taker fee = contracts x rate x P x (1-P) in USDC, rate by category: crypto 0.07; sports, economics, culture, weather, other 0.05; finance, politics, mentions, tech 0.04; geopolitics 0. Makers are never charged. Settlement is USDC, which carries basis against USD.
Plus 0.08pp of USDC/USD settlement basis: Kalshi settles in dollars and Polymarket in USDC, so a cross-venue position carries peg exposure for its whole life. The quadratic in Kalshi's fee is what makes this surface counter-intuitive, it peaks at even money and collapses at the tails, so an identical gap is worth very different amounts depending on where the contract trades.