How To Win On Kalshi: Four Measured Ways People Fool Themselves
If you searched how to win on Kalshi, you want someone to tell you what works. We are not going to do that, and not out of modesty: we trade these markets ourselves, we measure everything, and the clearest thing our measurements show is that the "systems" people discover on this venue are mostly artifacts of how they measured. So this page gives you the more useful thing instead. Four ways we fooled ourselves with our own numbers, each with the number attached. By the end you will have seen all four, each with the number that caught us.
The Quick Answer
There is no strategy we can honestly hand you, because the measurable answers to "how do I win on Kalshi" are mostly answers about how people get it wrong. What we can give you is the four failure modes we have hit in our own trading, with our own numbers: a ladder profit that is pure arithmetic yet accessible only 12.5% of the time, a 34.1% and a 61.3% that were both really a 12.5%, roughly 2 cents of invisible cost, and a ranking mistake that reliably selects noise. All four are below, worst first.
Why The Honest Answer Is Not A System
A beginner asking this question usually wants mechanics, and mechanics are teachable: what a contract is, how the price works, when to use a limit order. We keep a full beginner's guide to prediction market strategy for exactly that, and a standing hub on how these markets work where the live numbers stay current. This page is the level after that. Once you can trade, the question becomes whether the thing your spreadsheet says is working actually works, and that is where we have the scar tissue. Start with the number that tempted us most.
Failure 1: The Worked Ladder, From Gross To Net
Kalshi prices a city's daily temperature as a ladder of bands, and on a six-rung ladder exactly one rung settles yes. That makes selling all six rungs an identity, not a forecast: your return is the sum of the six prices minus the one dollar the winning rung pays out, no matter what the weather does. We measured it across 1,566 complete ladders and the sums came out 6.03 cents above the payout on average. The identity's residual was +0.0000 cents, exact to the fourth decimal.
That zero is worth a sentence, because it is the number that keeps the rest honest. An exact zero residual is only possible when no fee has been applied; any fee at all breaks the identity. So the +6.03c figure proves its own label: it is gross, before Kalshi takes its cut. Add the cut back in and the shine comes off fast. Maker fees across six rungs run 1.09 cents at Kalshi's published 0.0175 maker rate, so the net is +4.94 cents per ladder, and quoting the gross alone overstates the result by 22%. And the caveat that belongs in this same paragraph, not three sections later: all six rungs are simultaneously fillable only 12.5% of the time, which in our data measured out at roughly 6.5 baskets a day and about $1,822 a year at 20x size. The arithmetic is an identity. The access to it is not. "Sell all six rungs and collect six cents" is the sentence a reader screenshots, and without this paragraph attached it would be wrong.
| The Same Ladder, Four Ways | Per six-rung ladder |
|---|---|
| Gross (Price Sum Minus $1, N = 1,566) | +6.03c |
| Less Maker Fees Across Six Rungs (0.0175 Rate) | −1.09c |
| Net After Fees | +4.94c |
| How Often All Six Rungs Are Fillable At Once | 12.5% |
The row that matters is the last one, and it is the one every version of this idea we have seen published leaves out. Why the six prices sum above a dollar at all is not a house cut; it is the spread built into the resting offers you are selling into, which is what standing on all six rungs at once gets paid. Kalshi's own take is the fee schedule, which we break down in how Kalshi's fees work; whether gaps like this can be harvested across venues is its own question, answered in our guide to arbitrage on prediction markets. For sportsbook prices, where the books quote against each other all day, we automate the equivalent scan with an arbitrage finder. On a ladder, the scan is easy. The fills are the hard part, which brings us to how we got the fill number wrong twice before we got it right.
Failure 2: Two Marginals Cannot Make A Joint
A 12.5% like that one does not arrive cleanly; ours has a confession behind it. One of our measurements said orders like ours filled 34.1% of the time. Another, built a different way, said 61.3%. Both were computed honestly and both were useless, because each was a marginal: a rate for one rung, or one moment, averaged over everything else. Measured jointly, on the same rung at the same moment, the rate was 12.5%. Both marginals erred optimistically, and even the more conservative of the two readings, the 34.1%, was 2.7x the joint rate on its own. A median of X compared against a median of Y is not a rate, and a basket trade only exists when everything fills together.
This is the quietest failure on the page and probably the most common one off it. Any time a strategy needs several things to be true at once and you measured each one separately, the product of your optimism compounds. Ours compounded into believing an idea was somewhere between 2.7x and 4.9x more available than it was, depending on which of our two wrong numbers we trusted. The numbers above are from our own fill data, which is the only reason we caught it: liquidity decides whether you can trade at all, and it refuses to be estimated one column at a time.
Failure 3: The Two Cents No Backtest Can See
Fixing the fill rate still was not enough, because there is a cost that is structurally invisible to any backtest built on quote data. When your resting order gets lifted, your counterparty chose which rung to hit, and that choice is not in the archive. The quotes tell you where the market stood; they cannot tell you that you got filled precisely when someone knew something, or the sky was already changing, and passed on the rungs that would have paid you. From what we can measure of it, that adverse selection is worth roughly 2 cents of cost.
The corollary is uncomfortable. Every historical prediction-market result in existence that was computed from quote archives, including ours above, is optimistic by about that much. The +4.94c net from failure 1 is not a floor; it is an upper estimate produced by a method that cannot see one of its own costs. We would rather tell you that about our own work than have you discover it about yours, because a price on a screen is a claim about probability, and a fill is a claim about who wanted to trade with you. They are not the same fact.
Failure 4: Ranking On The Best Number Selects For Noise
The last failure is the one that cost us the most ideas. When we screened a set of candidate approaches and ranked them by their best number, the top of the list was reliably the noisiest candidate, not the best one. Nothing about that is bad luck; it is what ranking on a point estimate does. The candidate with the widest error bars gets the most chances to post an outlier, so the outlier is what you promote. We killed three of our own leads after re-ranking them on the estimate divided by its standard error, and every one of them had looked like a winner on the raw number.
That re-ranking is the one habit worth taking from this page, and it explains why we flinch when someone shows us a screenshot of their best week as evidence a Kalshi trading strategy works. A best week is a point estimate selected for being the best. The question is never what the top number is; it is how far the top number sits from zero once you account for how noisy the measurement was.
SO Can You Make Money On Kalshi?
Honestly: the parts we can measure with confidence are the costs. The fees are real and published, the fill constraint is real and severe, and the adverse selection is real and invisible. The venue itself operates with broad, state-specific availability under federal oversight. We trade these markets, we hold positions in them, and we do not publish performance figures; our own log is short, currently negative, and far too small to confirm or refute anything, and current figures live on the hub, which is rebuilt through the day. What we will say without numbers is the shape of the risk, because it never changes: selling an unlikely outcome collects a small premium and risks most of a dollar, so one loss erases the premiums from roughly 15 wins. That arithmetic, not the hit rate, is what makes position sizing the whole game.
Notice what this section is not. It is not "so here is what to do instead," and that is deliberate. The moment this page hands you a method, it becomes the thing it exists to warn you about: a system whose supporting numbers you have not seen measured, sold to you by someone with a reason to sell it. The thing you can actually control on this venue is not the weather, the fills, or the counterparty. It is whether you fool yourself, and the four numbers above are what fooling yourself looked like for us, caught in the act. The only figure on this page that was exactly right was the one arithmetic forced to be right, and even it needed a fee label to stay honest.
We publish how we check our own work because that checking habit is the product. Our free expert picks on sports betting boards show the reasoning next to every selection, free, every day; that is a different market from anything on this page, but the discipline is the same. When you want the full toolkit our analysts use to compare prices across books, OddsShopper Pro comes with a free week trial, and the code KALSHIWIN20 takes 20% off your first month if you stay past the week.
Disclosure and fine print. Stokastic trades event markets on Kalshi and holds positions in them; where this page describes measurements, they are measurements of our own trading. We have no affiliate or commercial relationship with Kalshi, though we do carry sign-up offers for some other prediction-market and betting platforms. Kalshi contracts are CFTC-regulated event derivatives traded on a designated contract market, not sportsbook wagers, and they can lose their full value. 18+, available where Kalshi operates; the risk of loss is real and, on the side we trade, individually large. This is part of an open research log of a strategy we have not yet proven, and nothing here is trading advice.
