A Kalshi price is a claimed probability. This page checks the claim: we record market prices daily, watch those markets settle, and measure how often each price level actually came true. No model, no opinion, just settled outcomes against the price roughly 24 hours before close.
Based on 1,757 settled markets matched to our recorded price history. This dataset grows every day.
Measured this week
Kalshi markets priced 0–9¢ settled YES 13.7% of the time.
That's 8.7 points MORE often than the price implied, measured across 146 settled markets at our recorded pre-event prices. Source: contracttax.com/calibration
All markets
PriceActual YES rateActualImpliedn
0–9¢
13.7%5%146
10–19¢
12.2%15%123
20–29¢
27.1%25%144
30–39¢
32.1%35%187
40–49¢
49.7%45%312
50–59¢
57.6%55%309
60–69¢
67.8%65%214
70–79¢
75.4%75%118
80–89¢
76.0%85%104
90–99¢
80.0%95%100
White tick = what the price implies. Bar = what actually happened. Green bars beat their price; red bars fell short.
Sports · 1,086 settled
PriceActual YES rateActualImpliedn
0–9¢
10.4%5%67
10–19¢
6.2%15%65
20–29¢
27.9%25%86
30–39¢
33.3%35%129
40–49¢
44.8%45%232
50–59¢
57.5%55%219
60–69¢
65.2%65%138
70–79¢
75.5%75%53
80–89¢
69.6%85%56
90–99¢
85.4%95%41
Economics · 156 settled
PriceActual YES rateActualImpliedn
0–9¢
44.4%5%9
10–19¢
33.3%15%9
20–29¢
26.7%25%15
30–39¢
53.3%35%15
40–49¢
85.0%45%20
50–59¢
90.9%55%22
60–69¢
86.4%65%22
70–79¢
94.7%75%19
80–89¢
86.7%85%15
90–99¢
50.0%95%10
Esports · 108 settled
PriceActual YES rateActualImpliedn
10–19¢
25.0%15%8
30–39¢
37.5%35%8
40–49¢
59.1%45%22
50–59¢
38.1%55%21
60–69¢
61.9%65%21
70–79¢
53.8%75%13
Crypto · 80 settled
PriceActual YES rateActualImpliedn
0–9¢
18.8%5%16
10–19¢
8.3%15%12
90–99¢
66.7%95%12
Methodology, honestly
For each settled market we take our recorded price nearest to 24 hours before the event, anchored on the earlier of the market’s close time and its expected expiration, because Kalshi frequently extends close times days past the actual event, and sampling after the outcome is known would fake perfect calibration. We accept snapshots from 2 to 48 hours out, bucket by decile, and compare each bucket’s implied probability to the share that settled YES. Our snapshots cover the higher-volume end of the board, so results describe the markets people actually trade, and thin buckets are hidden rather than shown with false confidence.
Why it matters: betting research has long documented a favorite-longshot bias, favorites slightly underpriced, longshots overpriced. Whether and where that holds on Kalshi is an empirical question, and this page is the running answer, category by category. Past calibration is not a guarantee about any single market. Its sibling, the Momentum Machine, measures what changing prices mean the way this page measures standing ones. Not financial advice.
Does the crowd learn anything before the event?
The curve above is anchored on a price recorded about a day before each event. We also hold prices recorded up to two weeks out, which is a much larger sample (1,761 observations against 1,757) and genuinely weaker evidence: a forecast made two weeks early is a harder one. Comparing them answers a question nobody has measured on this exchange: does a Kalshi price actually get better as the event approaches?
Across the bands where both samples are real, the two curves settle within 0.1 points of each other. A price two weeks out is about as good as a price a day out.
Band
A day out
Up to 2 weeks out
Difference
0–9¢
13.7% n=146
13.4% n=149
+0.3
10–19¢
12.2% n=123
12.2% n=123
0.0
20–29¢
27.1% n=144
27.1% n=144
0.0
30–39¢
32.1% n=187
32.1% n=187
0.0
40–49¢
49.7% n=312
49.7% n=312
0.0
50–59¢
57.6% n=309
57.6% n=309
0.0
60–69¢
67.8% n=214
67.8% n=214
0.0
70–79¢
75.4% n=118
75.4% n=118
0.0
80–89¢
76.0% n=104
76.2% n=105
-0.2
90–99¢
80.0% n=100
80.0% n=100
0.0
Only bands where both anchors have at least 30 settled markets are compared. The headline curve above, and every number we publish as fact, uses the day-out anchor. The wider one exists so that categories with a thin day-out record can still say something, clearly labelled as weaker evidence, rather than saying nothing.
Accuracy by category
The curve above is every market averaged together, and that hides the most useful thing in this dataset: Kalshi is a different forecaster depending on what it is forecasting. A crowd pricing a Fed decision and a crowd pricing an awards show are not the same crowd, and they are not equally good. Here is each category measured on its own.