Kalshi In-Game Swing Tests: Every Variant Negative
We paper-traded a low-priced contract swing rule on Kalshi sports in six labelled variants. Every version came back negative, and the payoff math explains why.
Published 2026-11-03 · Data as of 2026-10-02 · Market & data intelligence · Educational, not advice.
Delta Arc paper-traded a simple rule on Kalshi sports: rest a buy order on a low-priced contract, then sell 15 cents higher, with no stop. Every variant averaged a loss per contract after fees. Wins were small and losses were close to the full entry price, so even a win rate above 50 percent could not cover the losses. The test harness had bugs, disclosed below, so treat the figures as rough.
Most strategy write-ups show the version that worked. This one shows a rule that did not. Why it failed teaches more than any winning backtest would.
The rule we tested
A prediction market is an exchange, not a house. Every contract has a buyer and a seller. The venue matches them and takes a fee. The price is just where the last buyer and seller agreed. That matters here, because this strategy relied on being the patient side of the order book.
In Delta Arc's paper test, the rule placed resting buy orders at the bid on Kalshi moneyline contracts priced 10 to 35 cents. Each one was paired with a sell order 15 cents higher, and there was no stop. It ran as forward paper trading across 2026 games, including MLB, NHL, WNBA and football.
The intuition is easy to follow. Low-priced contracts move around as momentum shifts. If a trailing team's contract climbs 15 cents for a while, you capture the move without needing that team to win.
Resting orders are maker orders, and that affects cost. Kalshi's fee schedule, effective July 7, 2026, charges makers 0.0175 × contracts × price × (1 − price) on markets that carry a maker fee, and only when a resting order actually executes. Canceling is free. Takers pay 0.07 under the same formula, rounded up. For a contract priced at 50 cents, the maker fee is at most about $0.0044 per contract, far smaller than the average losses reported below.
What the first results showed
As of July 2, 2026, Delta Arc's test had closed 2,429 paper positions: 1,085 won and 1,344 lost. That is a 44.7% win rate, with an average result of −$0.0718 per contract after Kalshi's maker fees.
The more telling split is by outcome. Winning positions averaged +$0.1276 per contract, and losing positions averaged −$0.2328. With no stop, a losing position usually gave up nearly its whole entry price. A capped gain set against an almost-total loss needs a very high hit rate to break even, and this rule did not reach it.
Six labelled variants, one sign
Delta Arc ran six labelled versions on the same games. Two of them, the stop variants, turned out to be duplicates of their no-stop counterparts because of a harness bug described below, so only four are distinct. Every one had a negative average result per contract:
- Baseline: −$0.0721 (n=870)
- With a stop (duplicate of the no-stop version): −$0.0754 (n=719)
- 10–20 cent entries only: −$0.0669 (n=430)
- 10-cent target: −$0.0747 (n=174)
- Require a dip before entry: −$0.0666 (n=143)
- 10–20 cent entries with a stop (duplicate of the no-stop version): −$0.066 (n=93)
Tweaking the entry band, the target or the trigger barely changed anything. When every adjustment lands in the same narrow negative range, the cause is probably the structure, not the settings.
Why a high win rate can still lose
Football makes the point most clearly. Through October 2, Delta Arc found that football swing positions opened 1 to 2 days before expiry won 54.5% of the time (n=606) yet still averaged −$0.051 per contract. Wins paid about +$0.13, while losses cost roughly the entry price, which put the breakeven win rate near 70%. Positions opened more than 2 days out won only 11.0% of the time (n=538) and averaged −$0.069. These football counts likely include duplicate stop-variant entries, and football losers were marked at a timeout bid rather than at settlement, so treat them as rough.
Win rate is the number people quote. Payoff shape is the number that decides the result.
The backtest that did not hold up
Before the forward test, a historical backtest on past trade prints had looked positive. That did not hold up in forward paper trading, and we did not isolate the reason.
Honest reporting also means reporting our own errors. On October 2, 2026, a Delta Arc review found three bugs in the paper-trading harness. The stop was never enforced, so the stop variant was really a copy of the no-stop variant. Entries were checked against the mid price but filled at the higher ask. And football losers were closed at the last observed bid after a 24-hour timeout instead of at settlement, which likely understates losses. A snapshot of the full record that day showed 12,332 closed positions across all sports, 5,020 won (40.7%), averaging −$0.0723 per contract with a fill rate of 1.000. That points in the same direction as the July result, though given the bugs it is not a clean confirmation.
The gate, and what it is for
Delta Arc's readiness gate sets the bar for treating any variant as real. It requires at least 100 closed positions, at least 10 game-days, a 95% confidence interval whose lower bound is above zero, a fill-supported rate of at least 0.9, and no single sport above 80% of the sample. Because every variant averaged a negative result, none could have a lower bound above zero, so none passed.
One negative result from one rule, run in a harness with known bugs, is not proof that these markets cannot be beaten. It does show that this simple low-price swing rule did not work as tested. Until a result passes the gate, nothing counts as a finding.
Source: Kalshi fee schedule.
Delta Arc Prediction Markets is Delta Arc's probability intelligence system covering Kalshi and Polymarket. Get early access.