Categorical menus assign probability to labels. Scalar contracts assign it to where a number lands—headline CPI, monthly jobs, vote share, box office, a token price at a timestamp. The rulebook defines the metric, the units, the measuring authority, and whether payoff is a hard threshold, a bracket, or a sliding scale between bounds.
Three shapes you will meet
A threshold market is a binary in disguise: YES pays if inflation exceeds 3%, jobs reach 200k, or temperature beats a record. The price still reads as probability, but the event is inequality on a statistic, not a person’s name.
A bracket market partitions the real line into bands—CPI month-over-month below 0.2%, between 0.2% and 0.4%, above 0.4%—with one band winning $1 per share and the rest zeroing out. Brackets form a simplex like categoricals: the implied probabilities should sum to about 100%, with the same overround and Dutch-book logic.
A linear or partial payout design pays a fraction of face value inside a band—settlement might scale with how far the print lands between a floor and cap. Prices then move more smoothly with forecasts, and variance sits between a pure binary and a stock-like claim.
Always open the contract: some “scalar” listings settle as 0/1 on a range; others pay proportional cash. Expected value math follows the payoff table, not the headline label.
Mapping a forecast to brackets
Suppose you expect CPI month-over-month at 0.32% with uncertainty around 0.08%, and you assign roughly 7% to below 0.2%, 68% to the center band, and 25% to the hot tail. Compare that vector to market asks, not only to the bracket your point estimate sits in. The center can be the best trade even when your modal outcome is obvious.
If a binary “CPI ≥ 0.3%” trades YES at 62¢ while brackets covering the same domain sum to 58¢ on asks, a gap may signal arbitrage—or different index definitions (headline vs core, revision policy). Same release, two products, requires a rule diff before size.
Correlation on one print
Everything tied to a single data release moves together. Owning three brackets plus a threshold binary on the same CPI print is one cluster exposure, not three independent Kelly bets. After the number drops, the whole partition reprices in seconds; news-jump dynamics show up here in full force.
Jobs and CPI on the same week are related but not the same cluster; still, macro desks treat the whole calendar as correlated risk.
Revision and definition risk
Government statistics get revised. The contract may fix the first print or the final revised value. A market on “jobs” might use seasonally adjusted payrolls from a named report at a named time in Eastern Time. Scalar disputes often start with “which number counted,” not with who won an election.
Crypto “price on date” markets may use TWAP versus spot at one timestamp—identical headlines, different payoffs.
Execution realities
Marquee macro brackets on regulated books can be tight at the center and ugly at the tails. Obscure ranges on AMM venues punish size with curve slip. Retail caps on academic-style platforms turn bracket arb into signal discovery rather than scalable strategy.
Simulate the whole curve move on AMM entry; on CLOB, simulate lifting the ask ladder for your size.
Linking thresholds and brackets
The probability that CPI exceeds 0.3% should relate to the sum of bracket probabilities covering that region, in one coherent model. If the binary and the bracket bundle disagree persistently after fees, either traders are slow or the definitions differ. The second explanation is more common than free money.
A numeric bracket example
You buy the center bracket at 52¢ on a CPI slate whose asks sum to 104¢ overround. If the realized print lands in that band, you receive $1—profit 48¢ per share before fees. If a tail wins, you lose 52¢. Your ex-ante edge was the gap between your 68% band probability and the 52% implied cost, not the headline “0.32% forecast” alone.
Fallacy watch on numbers
Anchoring to last month’s print as if it were a prior breeds overconfidence. Treating 0.31% versus 0.29% as precision you can trade without a full distribution is fragile. Stacking “hot CPI AND hot jobs” without a joint story repeats conjunction errors from the probability module.
Threshold as digital option
“Jobs ≥ 200k” at YES 38¢ with your P(meet) = 46% gives +8¢ raw edge per share. The market is not guessing the exact payroll number—it prices the inequality under seasonal adjustment and revision rules named in the contract.
Forecast distribution, not point estimate
Serious macro trading carries a distribution: mean 0.32%, standard deviation 0.08%, then integrate over brackets. The bracket containing your mode is not automatically the best buy; edge lives where your p exceeds cost.
Platform listing habits
Regulated books list marquee CPI and jobs brackets with CLOB depth at the center. Crypto venues list crypto price thresholds with TWAP or spot nuances. Read the index definition before comparing a Kalshi bracket to a global platform threshold on “the same” release.
After the print
News jumps reprice the entire partition at once. If you hold one bracket, you are short the others in story if not in position. Exiting into a wide post-print book can cost more than the model edge you had pre-release.
Core concepts to remember
Scalars need metric, units, source, and timestamp. Threshold, bracket, and linear shapes differ. Brackets sum like categoricals. One release is one risk cluster. Link binaries and brackets only when definitions match.
Watching the whole slate on CPI night
When the print hits, every bracket reprices together. If you owned only the center band, your PnL is binary on that band—but your forecast should have been a full vector. Post-mortems that grade only the modal bucket miss whether your distribution was well calibrated.
Common mistakes
Anchoring to the prior print. Ignoring revision policy. Stacking correlated legs on one release. Comparing threshold and bracket prices without matching definitions.
What comes next
Scalars settle on levels. Index and spread contracts settle on differences—margins, relative returns, and paired statistics.
Next: Index and Spread Markets