Does momentum work on Kalshi's 15-minute markets?
Last updated: October 6, 2026
Not in our test. In the first six days of a paper test, our momentum strategies made 781 bets, mostly on Kalshi's 15-minute crypto markets, and lost 3.35¢ a contract after fees (p = 0.035). The early winners did not last. The result has caveats, set out below.
How the markets settle
Each 15-minute market asks one question: will the price at the end of the window be at least the price at its start? Kalshi lists a new market for every window. This is the rule of the bitcoin market (series KXBTC15M) that was open when we read Kalshi's API on October 6, 2026. The times in it change with every window:
“If the simple average of the sixty seconds of CF Benchmarks' BRTI before 9:15 AM EDT on Oct 6, 2026 is at least the simple average of the sixty seconds of CF Benchmarks' BRTI before 9:00 AM EDT on October 6, 2026, then the market resolves to Yes.”
BRTI is CF Benchmarks' real-time bitcoin index. Both ends of the window are averages over a minute, not single prices, and a tie resolves Yes.
The fee at typical prices
On October 6, 2026, Kalshi's API listed the KXBTC15M series with fee type “quadratic” and fee multiplier 1. That means an order that takes a resting price pays 0.07 × P × (1 − P) dollars a contract, where P is the price in dollars, and an order that rests and is filled later pays nothing.
| Price | Fee a contract, before rounding | 1 contract, after rounding | 100 contracts, after rounding |
|---|---|---|---|
| 50¢ | 1.75¢ | 2¢ | $1.75 |
| 95¢ | 0.3325¢ | 1¢ | 34¢ |
The fee peaks at 50¢ and shrinks toward the ends of the price range. Rounding is per fill, as Kalshi's rounding rules set out: the fee is rounded up to a millionth of a dollar, then the fill's cost with the fee is rounded up to a whole cent (a sale's proceeds round down), and the difference counts as fee. So one contract bought at 95¢ costs 96¢ in all. When an order fills in several pieces, Kalshi tracks that rounding across the order and refunds it a cent at a time, so the order's total fee comes close to what one fill of the same size would pay.
Our Kalshi fee calculator works out the fee, the all-in cost and the breakeven win rate for any price and order size, with a series' live fee settings.
What we ran
Momentum here means betting that a move already under way carries on to the end of the window. Early in each window, an agent looked at the market's price over the last few minutes. If it had moved clearly one way, the agent bought the side of the move at a mid-range price, always the same small number of contracts, and held it to settlement. No clear move, no bet.
The book moves between reading it and placing an order, so the agents set their limit a few cents above the ask they saw. Their orders took the prices resting in the book and paid the fee above.
The family included agents on Kalshi's 15-minute bitcoin, ether, solana, dogecoin and gold markets and on its hourly S&P 500 and WTI crude oil markets. It was one family in a test of about 90 strategies that ran from August 27 to September 6, 2026, on paper against the live order book.
The result
Pooled across the family, momentum lost 3.35¢ a contract over 781 bets (p = 0.035), counted on the sixth day of the test. That is the pooled figure. A single agent, in this family or out of it, can be ahead on a few dozen bets, and the leaderboard on our home page shows agents that are. On samples that small, being ahead says little about an edge. A coin-flip control in the same test lost about 0.4¢ a contract over more than 250 bets, which is about what the fees cost.
What looked like a winner first
- Bitcoin momentum was up $16 on its first day.
- Gold momentum looked like a leader at p = 0.050 over 91 bets. Like every early leader in the test, it drifted back toward zero as its sample grew.
- A different strategy, buying late favorites on ether, showed ten points of edge after 16 bets. It did not last either.
Small samples flatter. A run of good windows in the first day says little about the next few hundred.
Why backtests break here
A backtest usually fills at the price it saw when the signal fired. On a 15-minute market the price can move before an order arrives, which is why our agents set their limits above the ask they saw, so their orders would fill rather than rest. A backtest that ignores this assumes trades that were not there. Paper trading against the live book catches part of it, though not all: paper orders don't move the market, and the paper trading page says where paper differs from live.
Caveats
- The p-value is weaker than it looks. Bets in 15-minute windows close together in time are not independent, and we did not cluster the test for that.
- It is a family, not one market. The 781 bets pool several coins, gold and the hourly markets. It says nothing certain about any one of them.
- It was exploratory. The wider test went in over three waves, and losing strategies were paused on day six. Only its “sure things” result survives a correction for testing many strategies at once.
- It is one set of rules over one stretch of time. Other momentum rules, sizes or periods could do differently.
Test your own version
If you have a version of momentum you think is different, describe it in plain English and run it on paper against the live order book before you put money on it.