The economist consensus number you see quoted before a payrolls or CPI print is older than it looks. Those surveys close days before release, sometimes a week out, and then they just sit there as a single frozen figure while the world keeps moving underneath them. A jobless claims report comes in soft on Thursday, a services survey surprises to the downside, some Fed speaker says something about the labor market, and none of that touches the consensus you keep seeing on the screen. It was locked before any of it happened.
Kalshi contracts on those same releases do not have that problem. They trade continuously, so the price is a live estimate that absorbs new information as it arrives. If the market thinks the odds of a hot CPI just went up because oil moved, you see it in the contract within minutes. That is the whole appeal for me. I am not trying to beat the economists at forecasting. I am trying to watch a consensus that actually updates, and read what it is telling me about the distribution of outcomes going into the number.
A point estimate hides most of the information
The survey gives you one figure, maybe a high and low if you dig. The event-contract market gives you the shape. On Kalshi these releases are usually listed as a ladder of buckets, so for CPI you might have contracts for month-over-month coming in under a certain level, in a middle band, or hot above it. Each bucket has a price, and if you read the whole ladder together you get an implied probability distribution rather than a single guess.
That shape matters more than the midpoint. Two setups can share the same expected value and behave completely differently. One where the market is tightly clustered around a consensus print, and one where it is split with real weight on a surprise in either direction. The first tells you the number is boring and the reaction will be small unless it misses. The second tells you the market itself is uncertain, which usually means the post-release move is going to be larger whichever way it breaks. You cannot see any of that from a point estimate. You can see it in the ladder.
A few things I look at when I read the ladder:
- Where the mode sits relative to the frozen survey consensus. If the live market has drifted away from the economists, the survey is the stale one, not the market.
- How fat the tails are. Meaningful odds on the extreme buckets mean the market is bracing for a real surprise, and options and funding usually agree if you check.
- Whether the distribution is skewed. A market that puts more weight on a hot print than a cold one is telling you where the pain trade is.
Reading the drift into release day
The single most useful thing is not the level of the odds, it is how they move as the release approaches. I care about the path. A contract that grinds steadily toward hot over the two or three sessions before the print is a different animal from one that jumps on a single data point and then holds. The steady grind usually means the market is repricing on a run of supporting data, and that repricing tends to be more durable. The jump on one headline is more fragile and more likely to fade.
The mechanism to keep in mind is what feeds a nowcast. Payrolls odds react to the weekly jobless claims trend, to private payroll estimates, to survey employment components. CPI odds react to energy prices, to used-car and rent proxies, to the pieces that people can observe before the official number. When you watch the contract drift, you are really watching the market aggregate those inputs in real time. Historically that aggregate is not magic, and it does not beat the print on average, but it is a cleaner running summary than any single forecaster gives you, and it updates.
Here is the failure mode I have to remind myself about. Thin liquidity, especially days out from the release, means a couple of orders can move the odds in a way that looks like real repricing and is not. So I do not treat every wiggle as signal. I want the drift to be persistent, to survive a few sessions, and ideally to line up with the underlying data flow. If the contract moved but claims and the surveys did not, I assume it was noise until proven otherwise.
A pre-release routine
What I actually do in the day or two before a big number is boring and repeatable, which is the point. Roughly in this order:
- Pull the full contract ladder and sketch the implied distribution. Note the mode, the width, and any skew.
- Compare that mode to the frozen survey consensus. Flag the gap. The gap is the market saying the economists are behind.
- Look at the drift over the prior few sessions. Is it grinding or did it jump? Grind gets more weight.
- Cross-check against the inputs that feed the nowcast, claims for payrolls, energy for CPI. If the contract and the inputs disagree, trust the inputs and suspect the contract is thin.
- Decide what a given bucket outcome means for my actual book in rates, equities, or crypto, and size for the surprise scenario, not the base case.
That last step is where people go wrong. They read the distribution correctly and then position for the most likely bucket, which is exactly the outcome that is already priced in and pays almost nothing. The money is in being right about the tail the market is underweighting, or in being flat and patient when the distribution is tight and there is no edge. Most releases are the second case. A number lands inside the expected band, the market shrugs, and the people who forced a trade give back their edge in slippage.
The other honest caveat is that the reaction function is not fixed. The same CPI surprise moves crypto hard in one regime and barely registers in another, depending on what the market is obsessing over that month, rate cuts, growth, positioning. So I use the contract distribution to tell me how surprising the number is likely to be, and I use whatever I already know about the current regime to guess how much that surprise will actually move price. Those are two separate questions and it is easy to blur them.
When I want the odds drift and the underlying macro feeds in one place instead of stitching them together by hand, I lean on Blockcircle, but the routine matters more than any tool. Watch the ladder, respect the drift only when it is persistent, and size for the surprise rather than the consensus. That is most of the edge, and it is available to anyone willing to read the whole distribution instead of one number.