Three market structures, three very different games
Modern prediction platforms ship three market structures out of the box, and choosing between them is one of the most important product decisions you will make. The shape of the market determines liquidity, slippage, the way traders perceive risk, and how clean the resolution will be. Pick the wrong topology and you will pay for it in confused users, fragmented order flow, and disputed payouts.
The three structures are:
- Single-Market Binary (SMB) — exactly two outcomes (Yes / No).
- Single-Market Multiple (SMM) — three or more mutually exclusive outcomes that share one liquidity pool.
- Multiple-Market Binary (MMB) — a bundle of related binary markets that together describe a multi-way question.
Each carries different liquidity dynamics, payout rules, and resolution semantics. The good news is that the choice is not arbitrary — for any well-defined question, one of the three is almost always best. The framework below makes that choice mechanical.
If you are new to the broader category, our introduction to prediction markets has the foundational definitions.
Single-Market Binary (SMB)
The classic Yes/No market. One question, two outcomes, prices that always sum to 1. The deepest, most liquid prediction markets on Polymarket, Kalshi, and the Iowa Electronic Markets are almost all SMBs.
Use SMB when:
- The question is genuinely binary (e.g. "Will the FOMC raise rates this meeting?").
- You want maximum liquidity concentrated on one decision.
- You need fast price discovery — binaries are the most efficient AMM topology because all incoming flow lands on the same pair.
- You want a single number that journalists can quote without footnotes.
Avoid SMB when:
- The natural question has more than two answers and you would be forcing a Yes/No (e.g., "Will GDP growth be 1–2%, 2–3%, or above 3%?" should not be three separate binaries).
- The resolution criteria collapse multiple distinct scenarios into one ("Yes" — but which path got us there?).
The math: why binaries are cheap to run
In an LSAMM with n = 2 outcomes, the cost function reduces to a single log-sum-exp over two quantities. The subsidy budget needed to support a given level of slippage scales linearly with n, so a binary market costs ~half what a four-outcome market costs to operate at the same depth. This is one of the underappreciated reasons why production operators default to binaries: they are simply cheaper to subsidise. We unpack the math in our LSAMM algorithm deep dive.
Single-Market Multiple (SMM)
The same AMM logic extended to n outcomes. Prices still sum to 1, but spread over Bitcoin / Ethereum / Solana / Cardano — or any other discrete, mutually exclusive set. SMMs are the natural fit for questions like "Who will be the Republican presidential nominee?" or "What rating will S&P assign to country X?" — anywhere the universe of outcomes is genuinely categorical.
Use SMM when:
- The question has clearly defined, mutually exclusive options.
- You want the market to reveal a probability distribution, not just a binary signal.
- The number of outcomes is small enough (typically 3–10) that liquidity does not get pulverised.
- Resolution is unambiguous — exactly one outcome must be the truth, and "none of the above" must either be impossible or explicitly included as an outcome.
Watch out for:
- Liquidity dilution. A 5-outcome market with 1,000 total shares behaves like five 200-share binary markets. Long-tail options can stall at near-zero prices because no one wants to take the other side.
- Resolution disputes. What if "none of the above" actually happens? Always include a
nulloutcome unless the universe is provably closed. - Spread asymmetry. The favourite outcome trades with tighter spreads than the long-tail outcomes, which can mislead casual traders into thinking the long-tail markets are mispriced.
A clean SMM example
"What will Brazil's Central Bank do at the next COPOM meeting?"
| Outcome | Price | Implied probability |
|---|---|---|
| Cut 25 bps | 0.18 | 18% |
| Hold | 0.65 | 65% |
| Cut 50 bps | 0.10 | 10% |
| Hike | 0.05 | 5% |
| Other (cut 75+, intermeeting) | 0.02 | 2% |
A trader looking at this market sees the full distribution, not just a yes/no. That is what makes SMMs uniquely useful for policy questions where the shape of the distribution matters as much as the point estimate.
Multiple-Market Binary (MMB)
A grouping of n independent binary markets that share a theme. Example: "Q1 2026 crypto performance":
- Bitcoin: Up 20%? Yes / No
- Ethereum: Up 30%? Yes / No
- Solana: Up 50%? Yes / No
- Cardano: Up 40%? Yes / No
Each pair is its own AMM. Prices sum to 1 within each pair, not across the bundle. The four contracts can resolve independently — Bitcoin can be up 20% (Yes) while Solana is down (No), without contradiction.
Use MMB when:
- You want comparable signals across multiple assets or scenarios.
- Outcomes are not mutually exclusive — both Bitcoin and Ethereum can rise.
- You want users to take directional positions on individual statements.
- The platform UX can group related markets cleanly so users do not get lost.
Watch out for:
- More markets = more housekeeping, more KYC overhead, more resolution surface area.
- Users can get confused: "But I bet on Bitcoin Yes! Why didn't I win when Bitcoin was up?" — turns out the threshold was 25% and Bitcoin was only up 18%.
- Correlated risk. If three of your MMBs are highly correlated, the platform's net exposure can balloon during a single news event. You may need to manage that with cross-market position limits.
Correlation matrix awareness
A common MMB design mistake is launching markets that are nearly perfectly correlated without telling traders. If "BTC up 20%" and "ETH up 30%" tend to move together with 0.85 correlation, then a trader who buys Yes on both is taking nearly twice the risk they think they are. Operators should publish correlation matrices — at minimum, an asterisk noting that the markets in a bundle are correlated.
A practical decision tree
Use this checklist whenever you design a market.
| Question | Use SMB | Use SMM | Use MMB |
|---|---|---|---|
| Exactly two outcomes? | ✅ | ||
| 3+ mutually exclusive outcomes? | ✅ | ||
| Multiple parallel scenarios? | ✅ | ||
| Want a single probability distribution? | ✅ | ||
| Want to express opinion on each scenario? | ✅ | ||
| Maximum liquidity concentration needed? | ✅ | ||
| Need to reveal a full distribution? | ✅ | ||
| Outcomes are correlated but distinct? | ✅ |
When two columns get a check, lean toward the topology that is cheapest to subsidise at your expected volume. Below ~$50k of expected lifetime notional, that is almost always SMB.
How market structure affects price discovery
The shape of the market does not just affect liquidity — it affects how quickly truth gets baked into the price.
SMB: fast and clean
All flow lands on one pair, so a single news event triggers a price move on a single market. The signal-to-noise ratio is high, and journalists can quote a single number. The downside is the loss of distributional information — you cannot tell from an SMB whether the market is uncertain or unanimous.
SMM: rich but slow
The same news event has to update n prices simultaneously. Arbitrageurs do this by trading the most mispriced outcome first and then rebalancing the others. In practice, SMMs lag SMBs by 30–90 seconds on fast-moving news, especially when the headline outcome is also the largest position. The richer distributional information is worth the latency — but only if traders are pricing the long tail correctly.
MMB: parallel and fragmented
Each binary in the bundle is its own discovery process. The bundle as a whole offers comparison value (Bitcoin vs Ethereum), but no single price summarises the bundle. Operators sometimes publish a "bundle index" — a weighted average of the contained binaries — to give the press a single number to quote.
For deeper coverage of price-formation mechanics across these structures, see market microstructure for forecasting.
Liquidity allocation: how much depth to seed
A common operator mistake is splitting a fixed subsidy budget evenly across outcomes. The right allocation is proportional to expected volume, not to the number of outcomes.
If your model says 80% of the trading on an SMM will land on outcomes A and B, allocate 80% of your subsidy there. Leave the long-tail outcomes thinner — traders who actually have a thesis on those outcomes will be the ones to seed the price discovery, and your subsidy is better spent making the deep markets tighter.
For MMBs, allocate subsidies based on which binaries you expect to be most popular. The Bitcoin and Ethereum binaries in a crypto bundle will get 70% of the flow; the long-tail altcoin binaries will get the rest. Subsidise accordingly.
This dovetails with the broader question of how you organise your market catalog, which we cover in category design for market discovery.
Resolution criteria are non-negotiable
The single most important field in any prediction market — regardless of topology — is the resolution rule. Define it precisely before the market goes live, and never change it.
For SMBs, the resolution rule is usually a single sentence: "Resolves Yes if the Bureau of Labor Statistics reports headline CPI above 3.0% for August 2026, per the official BLS release."
For SMMs, you need an outcome rule and an explicit handling of edge cases. What happens if the data source delays the release? What if the resolution data is later revised? Most operators include a "Resolution timing" clause: "Market resolves at the close of business on the publication day; later revisions do not affect the resolution."
For MMBs, each binary needs its own resolution rule, and the operator should publish a master sheet that ties them together — especially if the bundle is marketed as a single product.
The CFTC's event contract guidance is a useful reference for the level of precision regulators expect. Vague resolution rules are the single fastest way to lose user trust.
Implementation lessons from production
After running hundreds of markets in production, three lessons stand out.
Most markets are SMB. Builders default to multi-outcome too early. Binaries trade better and produce cleaner data. If your team is debating "should this be SMB or SMM?" — the answer is almost always SMB.
Group MMBs by theme. Users explore by topic; show all related binaries together on the same page so traders can scan the comparison value. A bundle of crypto binaries scattered across the catalog is fragmented orphans; the same binaries on a single page become a coherent product.
Resolve in public. Define the data source before launch and publish the resolution explicitly. The market structure does not save you from a fuzzy resolution. The Iowa Electronic Markets resolution archive is a great template — every resolved market has a public record of which source produced the truth.
Default to SMB, justify SMM, design MMB carefully. Operators should treat SMB as the baseline. SMM should be justified by a clear distributional question. MMB should be designed with explicit attention to correlation and UX.
For trading discipline once these structures are live, see risk management for prediction market traders and the bankroll framework playbook.
Common anti-patterns to avoid
A few mistakes I have seen repeatedly.
Forcing a binary on a categorical question. "Will Trump or Biden win?" is a fine binary in a two-candidate race. "Will the Republican or Democratic nominee win?" with a possible third-party candidate should be an SMM with three outcomes, not a binary.
Splitting an SMM into many SMBs. "Will the S&P be up?" plus "Will the S&P be down 1–5%?" plus "Will the S&P crash >5%?" creates three near-duplicate markets that compete for liquidity. Use an SMM with four outcomes instead.
Bundling uncorrelated MMBs. A "Q1 2026 outlook" bundle that includes one CPI market, one crypto market, and one sports market is not a coherent product. Users cannot reason about it as a portfolio. Bundle MMBs around a single theme.
Late-stage topology changes. Once a market is live, the topology is fixed. Splitting an SMM into MMBs mid-flight strands existing positions and crushes trust. Plan the topology before launch.
Frequently Asked Questions
Why are most prediction markets binary?
Three reasons. First, binaries concentrate liquidity on a single AMM pair, which produces the tightest spreads and the fastest price discovery. Second, the subsidy cost to operate a binary is roughly half that of a four-outcome market at the same depth. Third, users find binaries easier to reason about — Yes or No is unambiguous, while categorical outcomes require more cognitive load.
Can a single market be both SMB and MMB?
No, those are mutually exclusive structures. A market is either a single Yes/No pair (SMB) or a bundle of related Yes/No pairs (MMB). However, a single platform can offer both structures side by side — the headline election market as an SMB, plus a bundle of state-level binaries as an MMB.
How does the AMM differ between SMB and SMM?
The math is the same — LSAMM or LMSR with n outcomes — but the subsidy budget needed to deliver a given level of slippage scales roughly linearly with n. An SMM with 5 outcomes needs roughly 5x the subsidy of an SMB with the same depth, all else equal. This is the dominant cost driver behind operator topology choices.
What is a "grouped binary" and how is it different from MMB?
Some platforms use "grouped binary" to mean exactly what we call MMB — multiple independent Yes/No markets shown together. Other platforms use "grouped binary" to mean an SMM that has been visually presented as a series of Yes/No questions. The math matters: if there is one liquidity pool, it is an SMM; if there are multiple independent pools, it is an MMB.
Can I arbitrage between an SMM and a related set of MMBs?
Yes, if the operator lists both. The classic arbitrage is buying all n outcomes of an SMM (which guarantees a $1.00 payout) for less than the sum of the equivalent positions in an MMB bundle. In practice the prices on the SMM and MMB drift apart by 1–3% in fast-moving markets, which creates arbitrage windows that close quickly. Sophisticated traders run these flows continuously.
Which structure is best for political election markets?
For a two-candidate race, SMB wins. For a multi-candidate primary or a "Who will win?" question with several plausible candidates, SMM produces the richest signal. For state-by-state electoral college modelling, MMB is the natural fit — each state is its own binary, and the bundle reveals the projected map.
How do I handle "Other" or "None of the above" outcomes?
Always include them explicitly in SMMs, and make the resolution rule clear about what triggers them. If your market is "Who will win the Republican primary?", the SMM outcomes should be [Trump, DeSantis, Haley, ..., Other], with the Other bucket explicitly resolved if any non-listed candidate wins. Failing to include Other is a common failure mode that leads to disputes when the unexpected happens.
Can I change market structure after launch?
Effectively no. Once positions are open, changing the topology strands existing traders. The right move is to launch a new market with the new structure and let the old market run to resolution. Some platforms allow "migration" via voluntary swaps, but it is administratively painful and not recommended for any market with non-trivial volume.
Where to go next
You have the toolkit to choose the right market structure. The next decisions in the design pipeline:
- For the deeper math behind why each structure prices the way it does, see the LSAMM algorithm explained and our comparison of AMM liquidity models.
- To organise these markets into a coherent catalog, read category design for market discovery.
- To understand how these structures behave under real trading conditions, see market microstructure for forecasting.
- For trader-side discipline across structures, our risk management playbook covers the full position-sizing framework.
Choosing the right market shape is a product decision before it is a math one. The AMM only optimises within the topology you give it. Pick the topology that matches the question, then trust the algorithm to do its job.
