Kalshi 15-Minute Bitcoin: How KXBTC15M Works & Strategy Data

Each Kalshi 15-minute Bitcoin market (series KXBTC15M) asks one question: will Bitcoin's CF Benchmarks index, averaged over the last minute of the window, finish at least where it averaged in the minute before the window opened? YES pays $1 if it does; NO pays $1 if it does not. A new window opens every quarter hour. In our data the side that is ahead has been slightly underpriced, by about two-thirds of a cent per contract at the price you can actually buy at. The fee takes that edge, and entries at 90¢ or more showed no reliable edge after fees.

How a KXBTC15M window works

KXBTC15M is the series. Each 15-minute window is its own market with its own ticker. Here is a real one, as Kalshi's API listed it when it opened on October 20, 2026:

• Market: KXBTC15M-26OCT201215-15, "BTC price up in next 15 mins?"
• Trading window: 12:00 PM to 12:15 PM EDT.
• Target price: $82,513.98.
• Resolves YES if the 60-second BRTI average before 12:15 PM EDT is at least the 60-second average before 12:00 PM EDT.
• Payout: $1 per winning contract; the other side gets nothing.
• Price steps: 1¢ from 10¢ to 90¢, and 0.1¢ below 10¢ and above 90¢.

The ticker reads left to right: the series, the date (26OCT20 is October 20, 2026), the window's closing time in Eastern time (1215) and a minute suffix. Nobody picks the target price. It is the previous window's final value. The window that closed at 12:00 PM settled at exactly $82,513.98, which became the 12:00–12:15 PM window's target. Across the 1,000 most recent settled windows (October 10 to 20, 2026), every window that directly followed another (998 of them) had a target equal to that window's final value.

"At least" matters. Kalshi's contract terms define "at least X" as X or greater, so a tie goes to YES. At two-decimal precision ties are rare; none occurred in those 1,000 windows. YES won 502 of them and NO won 498: a coin flip at the open, which is what you would expect from a contract that asks only "up or down from here".

A full day has 96 windows. Each window is one YES/NO contract rather than a ladder of strike prices, so there is no strike to choose. You choose a side and a price.

The settlement source is CF Benchmarks, not your exchange

KXBTC15M settles on the CF Benchmarks Bitcoin Real-Time Index (BRTI), which is calculated from several exchanges' prices. In the last minute before expiration Kalshi collects 60 index values, one per second, and averages them, rounded to two decimals. The market's own rule text warns that Google or Coinbase may help guide a decision but do not decide it. Two consequences follow:

• Your chart can disagree with the result. A Coinbase or Binance tick at 12:14:59 is neither the final value nor the target. The final value is a one-minute average across exchanges, so a last-second spike on one venue barely moves it.
• The moves are small. In those 1,000 October windows the median distance between target and final value was 0.068%, about $56 with Bitcoin at $82,500. Only 17 windows moved 0.5% or more. Many windows are decided by tens of dollars.

If index data is missing or incomplete at expiration, the CRYPTO contract terms say the affected strikes resolve to No, and Kalshi can open a market-outcome review before settling. Kalshi's API also offers an authenticated CF Benchmarks value feed that carries trailing 60-second averages, which is the closest you can get to watching the settlement number itself.

Close, settlement and payout

• Trading stops at the close. After the close Kalshi rejects new orders and cancellations, and it cancels resting orders shortly afterwards, so a limit order that never filled simply expires. These markets are also flagged so that Kalshi can move the close earlier.
• Settlement usually takes seconds. Across the 1,000 October windows the median market settled 3.8 seconds after its close, and 90% settled within 6.8 seconds.
• Not always. The window that closed at 3:00 AM EDT on October 15 took about two hours to settle, and Kalshi's API has no windows for the two hours after it. The contract terms allow a latest expiration a week out and a review before settlement, so fast settlement is the norm, not a guarantee.
• No settlement fee. Kalshi's settlement documentation says settlement fees are zero for simple yes/no outcomes. Holding to the close costs nothing extra; you pay the fee when you trade.

Fees: the number that decides most strategies

The taker fee on this series is 0.07 × contracts × price × (1 − price), with price in dollars. Kalshi's API reports KXBTC15M with the plain "quadratic" fee type, not the variant with maker fees, and multiplier 1, so an order that rests on the book and is filled later pays no fee. An order that matches a quote already on the book is a taker fill, and that includes a limit order priced at or through the ask.

The fee is largest at 50¢ and shrinks toward the extremes, but so does the upside. At 95¢ the fee is a third of a cent, yet that is still about 7% of the 5¢ you can win.

Rounding matters for the results below. Kalshi computes each fill's fee to a millionth of a dollar, then aligns balances to $0.0001 for direct members and to $0.01 for members cleared through a broker, with a running credit for rounding overpaid across an order's fills. Our study rounded every contract's fee up to a whole cent, which overstates what a direct member pays, so we report both versions. The fee guide and the fee calculator work through more examples.

Liquidity: deep, fast and one tick wide

This is a busy market. In the 1,000 October windows the median window traded about 2.9 million contracts; a tenth traded fewer than 1.7 million and a tenth more than 3.9 million. The window that closed at 12:00 PM on October 20 printed 43,600 separate trades. In 15 windows we measured in June 2026, trades arrived at a median of 16.5 per second and the book was one tick wide 98% of the time. Across the June–August archive the average quoted spread was about 1.1¢ near the open and roughly 0.6¢ in the last minutes, where the finer 0.1¢ steps apply.

For a small order, depth is rarely the problem. In June the best bid and best ask each showed roughly 2,000 contracts, so a one-contract order nearly always finds a counterparty at the best price. The constraints are the spread, the fee and speed. In that June sample the price traveled a median of 50¢ within a window, a traded price level typically lasted about a second, and second-by-second moves showed no momentum to follow and no reversal to fade. A new resting order at the best price waits behind those 2,000 contracts.

What our data says about strategy

We ran versions of the common KXBTC15M strategies with real money and in Paper, then measured them against Kalshi's settled history. The short version: the market prices this question almost exactly right, and the fee is about the size of whatever is left over.

1. Buying the side that is ahead

The most popular approach is to buy whichever side is leading and hold it to the close. We tested it at eight entry times across 6,298 settled windows from June 9 to August 15, 2026. At each time we bought the side priced above 50¢ at its ask, held it to settlement and subtracted the fee.

Direction was not the problem. Measured at the midpoint between bid and ask, the leading side won slightly more often than its price implied in every band from 50¢ to 99¢, by 0.3 to 1.8 percentage points (about one point on average). At the ask, which is what you pay, the edge shrinks to about two-thirds of a cent per contract. The leader is genuinely, if slightly, underpriced.

Fee-versus-edge disclosure (KXBTC15M): across 6,298 settled windows, buying the leading side at the ask was worth about +0.67¢ per contract before fees. With each contract's fee rounded up to a whole cent, the taker fee averaged about 1.55¢, leaving about −0.88¢ per contract (t = −3.08). In our October re-run of the same archive (6,303 windows), Kalshi's unrounded formula gives an average fee of about 1.09¢ and a net of about −0.42¢, a loss too small to separate from zero in this sample. Either way the taker fee is larger than the measured edge.

The table is our October 20 re-run of the same archive (6,303 windows, five more than the original study), which reproduces the study's published figures within 0.03¢. Late rows have fewer windows because a minute with no quotes leaves no price, and many windows are already priced at 100¢ by then. With the whole-cent fee every entry time lost. With the unrounded fee two came out slightly positive, at 8 minutes and 1 minute left, and neither is distinguishable from zero (t = +0.04 and +0.93). The earliest entry was the weakest under both fee models (t = −2.15 even with the unrounded fee): early, you pay the highest fee for the least information. With 16 numbers in the table, a t-value near 2 can appear by chance, so read the pattern rather than any one cell. No entry time showed a reliable profit.

Short samples mislead here. An earlier seven-day sample from our own price feed (662 windows) showed the 10-minute entry earning +1.89¢ per contract; at the full sample it was −0.54¢. One good night says almost nothing either. Our own live account made $10.42 across 40 windows one night in August while running this strategy, which loses on average.

2. The 90–99¢ "sure thing" near the close

The second common idea is to wait until one side looks decided and buy it at 90¢ or more. It wins most of the time, which is exactly what makes it convincing. To test it we took one entry per window: the first minute in which the leading side's ask was between 90¢ and 99¢.

The win rate matched the price almost exactly. At 95–99¢ the leading side won 96.75% of the time at an average price of 96.74¢: a fair price, not a bargain. Before fees, the 90–99¢ result of +0.26¢ sits inside its own noise (a standard error of 0.33¢); after any fee it is negative. Counted at every minute instead of the first touch, 90–94¢ entries did slightly better: about +1¢ per contract before fees, but +0.5¢ after the unrounded fee, which is within noise and within the archive's half-cent rounding. The losses are the hard part. The 385 losing windows in the 90–99¢ group cost about 92¢ each, so one loss erased the gain from about 13 wins. At 95–99¢, one loss erased about 30.

One caveat: our candle archive stores whole cents, while Kalshi quotes 0.1¢ steps above 90¢. Rounding to the nearest cent can move an entry price by up to half a cent, which is about the size of the effects in this table. Read it as "no measurable edge", not as a precise loss.

3. Resting limit orders instead of paying the ask

Because makers pay no fee on this series, the obvious fix is a resting limit order at the bid: let someone else cross to you. In the backtest, buying the leader at its bid instead of its ask turns the result positive, at about +1.49¢ per contract. That number is a ceiling, not a forecast. It assumes every resting order fills, and real resting orders do not fill at random.

We tested the resting version in Paper and with real money. From July 19 to August 3, 2026, the same design (rest on the favorite at up to 57¢) ran on nine 15-minute crypto series, Bitcoin included. Settlements could be attributed cleanly for four Paper runs, on NEAR, DOGE, SOL and ZEC. Across their 815 contracts the fills averaged 49.5¢ and won only 45.3% of the time, against the 49.5% their prices implied. Paper lost about 5.87¢ per contract, about 1.4¢ of it fees, and 207 live contracts from the same design lost about 5.79¢, so the simulation was not flattering the result.

The likely cause is adverse selection. A resting order on the favorite fills only when someone sells down to it, which tends to happen as that side stops being the favorite. The orders were meant to buy the leader; the fills delivered coin flips that had just turned against them. The Paper sample was not on Bitcoin and was analyzed after the fact, so treat it as a warning rather than a measurement of KXBTC15M. It is also why we do not quote the +1.49¢ ceiling as an expected result.

4. Ideas that did not help

• Retiming the entry. The entry-time table above shows no reliably positive time.
• Skipping volatile windows. We expected favorites to flip more often after volatile stretches of Bitcoin trading. Across 4,020 windows, favorites held 2.6 percentage points more often in the most volatile quarter than in the calmest: the opposite of the prediction, and within noise.
• Fading the leader. Buying the trailing side gives up the leader's small underpricing and still pays a fee that is symmetric around 50¢. If the leader is worth slightly more than its price, the trailer is worth slightly less than its price. In our archive, buying the trailing side at its ask lost about 1.6¢ per contract before fees, then paid about 1.1¢ more in fees.
• Letting a language model decide. We let a large language model choose buy YES, buy NO or skip every minute, capped at $10 per window. Its first session skipped almost every window, so we rewrote the prompt to lean toward buying. Over 54 windows it filled 136 contracts and lost $10.71 after Kalshi's actual fees, within noise for a sample that size. It bought the current favorite on 166 of its 168 orders, so it repeated the leader strategy rather than testing whether a model can pick its spots.

If you trade it anyway

None of this makes the market unfair. It is a clean, liquid, well-specified contract that leaves very little for a simple rule to take. If you trade it, these habits keep the cost visible:

1. Read the rule on the market you are trading. Check the target price, the close time in your own time zone and the "at least" wording: ties go to YES.
2. Watch the index, not one exchange. The final value is a one-minute BRTI average, and one venue's last trade can mislead you near the close.
3. Price the fee before the trade. Work out the fee at your price and compare it with the edge you believe you have. If you cannot name the edge in cents, assume the fee is larger.
4. Use limit orders, and know which kind of fill you got. A limit order priced at the ask is still a taker fill. A resting order avoids the fee but changes which trades you get.
5. Size for the loss, not the win rate. At 95¢, one loss costs as much as 19 wins earn. Keep each position small enough that a full loss is boring.
6. Grade yourself on realized versus implied. Log every fill and compare how often you won with what your average entry price implied. That one comparison exposed both the fee problem and the resting-order problem in our own tests.

If you would rather run a rule like this on a schedule than by hand, the Kalshi Bitcoin bot walkthrough builds a KXBTC15M rule in Paper mode first, with a price ceiling and limit orders, so you can check its fills against these numbers before any real money is involved. For the wider context, see the other Bitcoin markets Kalshi lists and how to backtest a Kalshi strategy without fooling yourself.

Bottom line

Kalshi's 15-minute Bitcoin market is a simple contract: up or down from the last window's final value, settled on a one-minute index average, usually within seconds. The hard part is the arithmetic. In our data the leader was underpriced by about two-thirds of a cent at the ask, the taker fee consumed that edge, high-priced entries showed no reliable edge after fees, and resting orders were filled on the wrong trades in our Paper and live tests. Treat any 15-minute Bitcoin strategy as negative until your own logged fills, after fees, say otherwise.