How accurate are Kalshi’s CPI markets? The day before, they named the exact inflation figure 48% of the time, and missed high twice as often as low
Kalshi lists markets on every US CPI report: the monthly change, core inflation and the annual rate. We checked 72 reports from late 2021 to September 2026 and took each market’s middle estimate 24 hours before the release. It hit the exact figure 48% of the time and was within 0.1 point 87% of the time. When it missed, inflation came in lower than the market expected 25 times and higher 12 times.
In this article
Kalshi’s inflation markets are a good guide to the CPI report, but not a precise one, and they have leaned too high. We looked at 72 Consumer Price Index reports with Kalshi price history, from October 2021 to September 2026, and took each market’s middle estimate 24 hours before the release. It matched the published figure exactly 34 times in 71, or 48%, and was within 0.1 percentage point 87% of the time.
When the market missed, the actual figure came in below its estimate 25 times and above it only 12 times. Over these reports, Kalshi traders expected a little more inflation than they got.
The short version
Source: Kalshi public market data and price history; PredictDesks analysis
- The middle estimate hit the exact figure 48% of the time, and was within 0.1 point 87% of the time.
- The market spread its bets sensibly. On average it put 34% on the figure that was published. It rarely put more than 60% on any single figure, which matches how often it was right.
- It leaned high. Of the 37 misses, 25 came in below the market’s estimate.
- Core inflation was easiest. The monthly core figure was hit exactly 53% of the time and within 0.1 point 95% of the time.
How Kalshi’s CPI markets work
For each monthly CPI report, Kalshi lists a ladder of contracts such as “Above 0.2%”, “Above 0.3%” and so on, for three measures:
- Headline CPI, the monthly percentage change in all prices (series KXCPI).
- Core CPI, the monthly change excluding food and energy (KXCPICORE).
- CPI year over year, the annual inflation rate (KXCPIYOY).
Each contract pays $1 if the published figure is above its strike. Reading across the ladder gives a probability for every figure: if “Above 0.2%” trades at 60¢ and “Above 0.3%” at 23¢, the market puts about 37% on a reading of exactly 0.3%. The contracts settle on the Bureau of Labor Statistics release at 8:30 a.m. ET (how to bet on economic data on Kalshi).
How we measured it
For each report we took the hourly price history of every contract in the ladder and read the price 24 hours before the market closed for the release. From the ladder we worked out:
- The middle estimate: the figure at which the market’s chance of a higher reading drops below 50%, its median forecast.
- The chance on the actual figure: the probability the ladder gave to the reading that was published.
The published figure comes from the contracts that settled Yes. We used every report for which Kalshi’s price history was available: 29 headline, 19 core and 24 annual readings, 72 in all. Most are from November 2024 onwards, when trading on these markets grew. One headline report had no usable middle estimate, so the hit rates use 71 reports. Volumes are counted in contracts, Kalshi’s unit, not dollars, and include trading on every strike in the ladder.
Results by measure
Source: Kalshi price history; PredictDesks analysis
Show the dataHide the data
| Value | |
|---|---|
| Core CPI (monthly): exact | 52.6% |
| Core CPI (monthly): within 0.1 | 94.7% |
| Headline CPI (monthly): exact | 46.4% |
| Headline CPI (monthly): within 0.1 | 82.1% |
| Annual rate: exact | 45.8% |
| Annual rate: within 0.1 | 87.5% |
| Measure | Reports | Exact hit | Within 0.1 | Lower than estimate | Higher |
|---|---|---|---|---|---|
| Headline CPI, monthly | 29 | 13 of 28 | 23 of 28 | 9 | 6 |
| Core CPI, monthly | 19 | 10 of 19 | 18 of 19 | 6 | 3 |
| CPI, year over year | 24 | 11 of 24 | 21 of 24 | 10 | 3 |
Core inflation has been the most predictable: the monthly core figure moves less than headline inflation, which swings with fuel prices. Headline misses were more evenly split in direction, while annual-rate misses came in lower than expected 10 times out of 13.
Better in 2026
Source: Kalshi price history; PredictDesks analysis
Show the dataHide the data
| Value | |
|---|---|
| 2022 | 16.7% |
| 2023 | 50.0% |
| 2024 | 66.7% |
| 2025 | 33.3% |
| 2026 | 66.7% |
Most of our sample comes from the last two years, and the difference between them is striking. Of the 27 reports released in 2025, the market’s middle estimate hit the exact figure only 9 times, 33%, though it was within 0.1 point 24 times. Of the 30 released in 2026 up to October, it hit 20, 67%, and was within 0.1 point 27 times.
Part of that may be the mix of months: the 2025 reports include the period of the government shutdown in October and November, when official data were delayed. Part may reflect a larger market with more trading. With samples this size we cannot separate the two, but the 2026 record is clearly better.
Does the last day help?
We also read each ladder two hours before the market closed for the release, to see whether the last day of trading improved the forecast. On the 63 reports with prices at both points, the middle estimate two hours out hit the exact figure 31 times, against 33 for the estimate a day out. The last day of trading did not make the forecast better on average. Most of the information the market uses arrives well before the release morning.
The direction of the misses
Source: Kalshi price history; PredictDesks analysis
Show the dataHide the data
| Value | |
|---|---|
| −0.2 or more | 7 reports |
| −0.1 | 18 reports |
| Exact | 34 reports |
| +0.1 | 10 reports |
| +0.2 or more | 2 reports |
The tilt is clear. The published figure was 0.1 point below the market’s estimate 18 times and 0.2 or more below 7 times. It was above by 0.1 ten times and by 0.2 or more only twice.
Two of the biggest misses were in the annual rate. For November 2025, the market’s middle estimate was 3.0% and the published rate was 2.7%; a day before, the market had put almost nothing on 2.7%. For June 2026, it expected 3.8% and the figure was 3.5%.
A bias of this size, about one report in four coming in 0.1 point low, is small. But it is consistent across headline, core and annual figures, and it suggests that in this period traders were slightly too worried about inflation.
How confident the market was
The ladder rarely put a big chance on a single figure. On the reports where we could compute it, the market put:
- under 20% on the actual figure 9 times,
- 20% to 40% 33 times,
- 40% to 60% 20 times,
- more than 60% only once.
That spread fits the results. A market that puts about a third on the published figure, and is exactly right about half the time with its middle estimate, is describing real uncertainty rather than hiding it. For anyone trading these ladders, it means a single strike near the middle is rarely a sure thing, even the day before.
The busiest reports
Trading is concentrated in a few months. The single most traded event in our sample was the annual rate for April 2025, with about 11.4 million contracts. Other busy reports were the annual rates for June and July 2025 and the monthly headline readings for April and July 2025, each with more than 2 million contracts. In total, the 72 reports traded about 59.8 million contracts.
The October 2026 report
The latest report, for October 2026, was published on 10 November. The evening before, Kalshi’s ladder put the headline monthly figure most likely at 0.3%, with about a 37% chance, and core inflation most likely at 0.2%, with about 48%. Both came in one notch higher: headline prices rose 0.4% and core prices 0.3%, and the ladders settled accordingly.
So the newest report went against the pattern in our data: this time, the market expected too little inflation, not too much. One report does not overturn a tilt seen across 71, but it is a reminder that the direction of the misses can change, especially when fuel prices swing as they did in September and October.
The October report also had no annual rate. Because of the 2025 government shutdown, the Bureau of Labor Statistics did not collect enough October 2025 prices and did not publish a 12-month change for October 2026.
How to read a CPI ladder
The October 2026 ladder shows how to turn Kalshi’s prices into a forecast. The evening before the release, the headline contracts traded at about:
- “Above 0.1%”: 93¢
- “Above 0.2%”: 60¢
- “Above 0.3%”: 23¢
- “Above 0.4%”: 4¢
Subtracting each price from the one below gives the chance of each reading: 0.2% about 33% (93 − 60), 0.3% about 37% (60 − 23), 0.4% about 19% (23 − 4). The middle estimate is 0.3%, the first strike where the chance of a higher reading falls below 50%. The published figure was 0.4%, which the ladder had put at about one in five.
The same arithmetic works for any report. It also shows why a single contract is a blunt tool: buying “Above 0.3%” at 23¢ paid out on a 0.4% reading, but it would also have paid on 0.5% or more, so the contract is a bet on a range, not a number.
Limits of this study
- Sample. 72 reports with usable price history, most from late 2024 onwards. Older markets had less trading and patchier data.
- One snapshot. We used the price 24 hours before the close. Prices move during the last day, though in our data the forecast two hours out was no better.
- Rounding. CPI is published to one decimal place, so a “miss” of 0.1 point can be a difference of a few hundredths in the unrounded figure. The market and the economists face the same rounding.
- Settlement. Our published figures are the readings the contracts settled on. Later BLS revisions are not reflected.
FAQ
How accurate are Kalshi’s CPI markets? In our sample, the middle estimate a day before the release matched the published figure 48% of the time and was within 0.1 point 87% of the time.
Do the markets lean high or low? Over these reports, the published figure came in below the market’s estimate 25 times and above it 12 times. October 2026 was one of the exceptions.
Which CPI measure is easiest to predict? Monthly core CPI: the middle estimate hit it exactly 53% of the time and was within 0.1 point in 18 of 19 reports.
When is the next CPI report? The November report is due on 10 December 2026, with the annual rate published again. Kalshi lists its ladders for that report already, and the same arithmetic applies.
Data
All 72 reports, with the published figure, Kalshi’s middle estimate, the error, the chance it put on the actual figure and the volume, are free to download: kalshi-cpi-markets-2024-2026.csv. For the same test on rate decisions, see how well Fed markets predict decisions.
Reports from October 2021 to September 2026 with Kalshi price history. Nothing here is financial advice.
