Skip to content
Foreseez. Get early access
Data story · reconstructed with sample data 8:29:52

Anatomy of a CPI morning.

One number comes out at 8:30. Three markets react. Here is how it unfolds, minute by minute and then second by second, and what it says about how prices really move.

FForeseez ResearchOct 15, 20266 min read
8:30:00 8:009:30
8:00

The quiet before

Half an hour before a CPI release, prediction markets go quiet in a particular way. Prices barely move, but the order books thin out. Market makers pull their resting orders rather than get caught on the wrong side of the number.

Below is a morning in the market for Fed cuts in December, across three venues. Press play, or drag the slider, and watch the gap between venues open before anything has happened.

FIGURE 1 · REPLAYNinety minutes, minute by minute
5055606570 123456 8:008:158:308:459:009:159:30
8:30
Consensus63%Kalshi65%Polymarket59%PredictIt61%
Spread6ppVolume4,000
8:30 · The printEvery venue reprices inside the minute. Volume is ten times normal.
White is the Foreseez consensus, weighted by liquidity. Colored lines are Kalshi, Polymarket and PredictIt. Red bars are contracts traded each minute. Numbered dots mark the moments described in the text.
8:30:00

The print

The release lands at 8:30 sharp. Within seconds the venues reprice, but not together. At one-minute resolution it looks like a single jump. Zoom in to seconds and it comes apart.

FIGURE 2Who moved first
spread now5pp
8:30:00 +15s+30s+45s+60s+75s+90s Kalshi60% → 65%Polymarket55% → 60%PredictIt57% → 62% +0s
Each lane is one venue's last trade, second by second. Blue ticks are our agent's steps. Reconstruction with sample data.

The fastest venue moved in six seconds. The slowest took thirty-eight. For about half a minute the same question carried a spread of up to 10 points, not because traders disagreed, but because some had not looked yet. If you compare venues during that window, you are comparing reaction times.

8:31

The overshoot

Then comes the overshoot. Consensus keeps climbing for another quarter of an hour, peaks two points above the first reaction, then drifts back below it over the next hour. First moves after data often go too far. The heatmap shows one reason why.

FIGURE 3 · ORDER BOOK DEPTHThe book drained, then flooded back
70%65%60%55%50% 8:158:208:258:308:358:408:45 ThinDeep

On Kalshi the book was already thin by 8:25. The first trade after the print walked straight through three empty price levels.

Brightness shows resting orders at each price level, each minute. White line is the venue's price. Sample data.

Before the print, depth on both sides of the price drained away. The first trades after the release hit a thin book and pushed the price further than the news alone justified. As makers came back, depth rebuilt around the new level and the price drifted back toward it.

8:30 + 88s

Meanwhile, the agent

While the markets were moving, our agent was writing. It does not race to be first. It waits until every venue has repriced, because a read that quotes a number about to change is worse than a read that arrives a minute later.

FIGURE 4Ninety seconds inside the agent
Pick a step
0s30s60s90s
+7s to +38sWait for every venueHolds until the slowest venue has repriced. Publishing before that would quote a spread that is about to vanish.
Times are seconds after the release. Blue steps run in parallel with the markets; amber is waiting on the slowest venue.

A release morning compresses everything this blog is about into ninety minutes: venues that disagree for reasons that have nothing to do with opinion, first moves that overshoot, and the value of knowing which gap is real.

Every number in this post is reconstructed from sample data to illustrate a typical release morning. It is not a record of a specific day and nothing here is investment advice.