Event markets · Market design · 28 July 2026
Every parlay price is a statement about a joint distribution, and the difference between that price and the product of its legs is an implied correlation — printed thousands of times a day, quoted by nobody. Prediction markets have become the deepest instance layer in finance — tens of billions of dollars of annual volume across thousands of contracts — and there is still no benchmark anywhere in the complex: no index that concentrates liquidity, no reference the granular contracts quote against. That is not a commercial oversight. It is an algebraic constraint, and it has a name: the partition rule. This piece states the rule, maps it across every live category, and then applies it twice — forward, to the most-watched trade of this week, the Fed decision ladder against the bitcoin price ladders; and backward, to a World Cup that settled nine days ago and quietly validated the framework's strangest prediction.
This extends the five-tier hierarchy built in Hedging corporate event risk inside a hierarchy and applied to compute in Implied compute forward curves. Those pieces built markets that do not exist yet; this one turns the same machinery on the markets that already trade — and on the parlay, the instrument through which most of that volume now flows.
The corporate-event work built a hierarchy from nothing. Established prediction markets present the exact opposite problem — and it is the easier one.
Designing corporate event contracts means constructing every tier: the benchmark, the names, the instances, the regime layer above them. The live prediction-market categories — politics, sports, macro, crypto, culture — have already built the bottom of that hierarchy to a depth no designed market has ever reached. Thousands of instance contracts, real spreads, continuous price discovery. What none of them has is anything above the instance layer: no benchmark tier, no sub-index tier, nothing that aggregates the granular flow into a factor anyone can hold or hedge.
Nothing exists. The design problem is construction.
The bottom tiers are saturated. The design problem is aggregation — and the parlay, it will turn out, is where the aggregate already trades.
The commercial claim from the framework piece carries over intact: repackaging fragmented granular flow into a standardised wrapper is what makes the granular legs cheap to trade. Portfolio trading cut corporate-bond execution costs by more than 40%, with the largest benefit accruing to the least liquid bonds. The venue that lists the first valid benchmark tier does not cannibalise its instance flow; it subsidises it.
Before any index work, sort the contract set. The class determines the mathematics, the units, and whether a benchmark is possible at all.
| Class A — bracketed continuous | Class B — categorical | |
|---|---|---|
| Shape | A strike ladder over a real-valued underlying | Discrete outcomes, no underlying continuum |
| Examples | Fed funds, CPI, BTC price, GPU rental price, temperature | Election winner, game result, court ruling, award |
| What the ladder is | A discretised risk-neutral CDF. Adjacent differences are probability mass | Nothing — the prices are the object |
| Natural benchmark | A synthetic forward in the underlying's own units | An expected-count index, Σpi, in units of "outcomes" |
| Factor space | Log-returns of the forward. Standard commodity machinery applies | Log-odds. A common shift is additive in logit space, not price space |
| Term structure | Real. Interpolate to constant maturity (VIX-style) | Event-dated. Roll on the event calendar (CDX-style) |
You cannot build a level index on a partition. If a contract set is mutually exclusive and exhaustive — the candidates in one race, the two sides of one game, the brackets of one Fed meeting — then Σpi ≡ 1 by construction. The set has n−1 degrees of freedom and no level factor can exist inside it. Any "index" computed on it is a rotation, not an exposure.
Simulated: the first principal component of a 6-outcome partition explains 23.7% of daily variation — near the 1/(n−1) = 20% pure noise would produce. The same six contracts drawn from separate events sharing a driver give PC1 53.3%, and the expected-count index Σpi correlates 0.91 with the true latent factor. Benchmarks are built across events, never within one. That single sentence decides where an index can live in every category below — and where it cannot.
The "benchmark" column is the contract that does not exist in any of these categories today.
| Category | Class | The partition trap | The valid collection | Benchmark & units |
|---|---|---|---|---|
| Politics | B | Candidates within one race. Σp≡1 — no factor exists | The same party's win probability across all contested races | Expected seats, Σpi — a tradable generic ballot. Single races quote as spread to it |
| Sports | B / A | Sides of one game. Zero-sum by construction — a common factor is mathematically impossible | Totals across the slate; season win-totals across teams | Scoring-environment index — mean implied total across the slate. Pace, rules, officiating, weather are the shared drivers; team quality is idiosyncratic by design |
| Macro | A | Brackets within one CPI print or one FOMC meeting | The ladder strip across consecutive releases and meetings | Constant-maturity implied path — the event-market analogue of the OIS curve. Already the most credible case: Kalshi CPI beat Bloomberg consensus on MAE, 0.063 vs 0.081 |
| Crypto / rates | A | Strikes on one expiry | Strips across expiries and assets | Implied vol / drift surface. Competes with a real options market — lowest standalone value-add, highest cross-market value: see §04 |
| Awards & culture | B | Nominees in one category. Σp≡1 | A studio's or label's contracts across all categories | Expected wins per house. Thin, but structurally identical to expected seats |
The sports row is the sharpest test of the rule. Sides in a single game are exactly zero-sum, so no systematic factor can exist in game sides — not as an empirical finding but as an algebraic fact. Every dollar of game-side volume, which is the overwhelming majority of the industry's notional, sits in a structure that admits no benchmark. The factor structure in sports lives entirely in totals and props, where scoring environment moves every game on the slate at once. A tradable league scoring index is the sports benchmark, and game totals become basis trades against it. That instrument does not exist — which is why the largest event-market category by volume has produced no index despite years of trying.
The FOMC decides tomorrow. The decision ladder is a partition; nothing can be indexed inside it. But run the rule the other way and the interesting object appears: one factor, two ladders, and a correlation you can back out of a combo price.
As of this writing the July ladder prices roughly 79% hold, 20% +25bp, with fixed-income desks assigning up to a third to the hike on energy-led inflation at 3.7%. Within that meeting, the partition rule applies in full: the brackets sum to one, and any "Fed index" built on them is a rotation. The valid collection is the strip across meetings — July, September, October, December — from which a constant-maturity implied policy path falls out exactly as the compute forward curve fell out of the compute ladders:
The meeting strip and the implied path
Bars: bracket probabilities per meeting (illustrative, anchored to this week's July pricing and the "two hikes this year" fixed-income base case). Line: the cumulative expected policy rate those brackets imply — the event-market OIS curve.
Bitcoin trades near $63,905 — down from a $126,198 year high, below its 200-day — and its weekly and monthly price ladders are the venue's deepest Class A markets after macro. On their own, the framework grades them honestly: lowest standalone value-add of any category, because a real options market already publishes a better surface. The value is not in the marginal. It is in the joint.
A hawkish surprise tomorrow is a shock to the one factor both ladders load on. Take a deliberately minimal model: BTC's weekly move is normal with σ = 5%, and a hike delivers a −2.0% conditional drift (a 25bp move of which 80% is unpriced, at a documented FOMC-day beta). Then P(BTC < $62,000) is 27.3% conditional on hold and 41.9% conditional on a hike — a 14.6-point swing in a contract that quotes 29.9% unconditionally.
One event, two conditional distributions
The BTC weekly ladder implied by the model, conditional on each Fed outcome. The correlation between the markets is the horizontal displacement between these two curves.
Price the combo "Fed hikes AND BTC closes below $62,000". Leg multiplication — the sportsbook convention — gives 20% × 29.9% = 6.0¢. The one-factor model gives 20% × 41.9% = 8.4¢ — 2.4¢ over the product, an implied correlation of ρ ≈ 0.13 between the two events. Whoever sells the combo at the product is short that correlation for free; whoever quotes it wider is making a market in it. Either way, the combo price minus the product of the legs is a pure correlation print — the first tradable read on how tightly crypto is coupled to policy in any given week, published nowhere, implicit in every cross-market combo that clears.
The tournament settled on 19 July — Spain 1–0 Argentina, after extra time. Our June analysis is still in the repo, timestamped. So run the framework against what actually happened.
What the framework said in June, before a ball was kicked in the knockouts: game sides are partitions and carry no factor; the tradable structure is the milestone chain — group exit, round of 32, 16, quarters, semis, final, title — decomposed per team and checked for internal consistency across books. Two of those June numbers are worth revisiting now.
| Milestone | Argentina, June | Spain, June | Settled |
|---|---|---|---|
| Make final | 30% | 18% | Both did — joint 5.4% if independent |
| Win outright | 20.5% | 11.0% | Spain — the 11% leg, fourth favourite |
| Win given final | 68% | 61% | 68% + 61% > 100% — jointly impossible if they meet. They met. |
The third row is the lookback's finding. Each team's outright-to-final ratio implied a conditional finals-win probability; the two conditionals summed to 68% + 61% — more than certainty — which is only consistent if the books were pricing each team's final opponent as, on average, someone weaker than the other. The June engine flagged exactly this class of inconsistency (its book-consistency check), and the tournament then produced the one matchup that made the inconsistency undeniable. The mispricing was structural, visible in advance, and it lived across contracts — in the joint — precisely where the partition rule says the information is. No game-side index would ever have seen it.
The same logic grades the rest of the slate. The sides carried no factor, and settled like it. The factor lived where the framework put it: in the milestone chains and totals, where tournament-wide drivers — schedule congestion, heat protocols across the North American summer venues, extra-time frequency — moved every team's numbers together. Spain's own chain (94% → 78% → 52% → 30% → 18% → 11%) settled true at every rung, which is what an internally consistent decomposition looks like after the fact.
The framework's benchmark tier does not exist. Its correlation tier, it turns out, does — it just goes by another name and clears $100M a week.
Kalshi introduced same-game parlays through an RFQ facility last September and rolled out "combos" to all users in December; within weeks they cleared over $100M in a single week. Polymarket self-certified its own multi-leg contracts to chase the same flow. Structurally a combo is a digital on a joint outcome: it pays only if every leg settles yes. Which means every combo price is a statement about a joint distribution — and the difference between that price and the product of its leg prices is, exactly as in §04, an implied correlation.
| Legs (marginal prices) | Independent product | Correction at ρ=0.2 | at ρ=0.4 |
|---|---|---|---|
| 20¢ × 35¢ | 7.0¢ | +2.0¢ | +4.4¢ |
| 50¢ × 50¢ | 25.0¢ | +3.0¢ | +6.3¢ |
| 30¢ × 60¢ | 18.0¢ | +2.5¢ | +5.2¢ |
Combo fair-value engine
Two legs and a correlation. The readouts show the independent product, the copula fair value, and the correlation a quoted price implies.
Now put the two halves of this piece together. Same-game parlays are within-partition objects — conditionals inside one event, structurally the same animal as the World Cup's win-given-final ratios, informative about a single joint outcome and nothing else. Cross-event combos are the valid collection: legs drawn from separate events that share a driver. A cross-event combo book is therefore a standing survey of implied pairwise correlations across the whole complex — the raw material for exactly the factor extraction the benchmark tier needs. The framework's corporate piece put a tranche-and-correlation tier above its benchmark and treated it as the sophisticated end of the hierarchy. In established markets the order inverted: the correlation tier arrived first, at retail, mispriced by leg-multiplication, before any benchmark exists to price it against.
A venue clearing combos at or near leg products is short every positive correlation on its own book — the World Cup row above and the Fed×BTC construction are both cases where the product is the wrong price in a knowable direction. The fix is the same one this framework keeps arriving at: list the factor. A scoring-environment index makes slate-wide totals combos priceable. An implied policy path makes macro×crypto combos priceable. The benchmark tier is not a rival product to the parlay flow — it is the hedge for it, and the venue that lists it first gets to price everyone else's correlation book.
Every live category divides cleanly into partitions (no benchmark possible) and collections (benchmark waiting to be listed). The rule made one non-obvious call in advance — that the information in sports lives in joints and milestones, never in sides — and the World Cup settled it the framework's way.
A constant-maturity implied policy path from the meeting strip is the complex's most credible first index — the CPI accuracy result says event markets can carry the settlement weight. And the correlation that path shares with the crypto ladders is already implicitly priced every time a cross-market combo clears. The instruments exist; the quotation convention is what is missing.
$100M a week of joint-outcome flow, priced predominantly by leg-multiplication, is a standing inventory of free or mispriced correlation. That is simultaneously the strongest commercial argument for listing the benchmark tier and the cleanest dataset for calibrating it. The complex built the top of the hierarchy's risk stack before the middle. The middle is the opportunity.