Reading a Prediction Market Order Book: Why a 4 Percent Market Can Cost More Than 4 Percent

# Reading a Prediction Market Order Book: Why a 4 Percent Market Can Cost More Than 4 Percent

Kresmion Research

You open a prediction market. A contract that pays $1 if the Federal Reserve raises interest rates by 25 basis points after its July 2026 meeting is showing 4 percent. It looks simple: put in $100 and, if you are right, get back about 25 times your money. You are almost certainly not buying at 4 cents.

Here is the whole idea in one sentence: the number on the screen is the market's probability estimate, not the price you pay. It is the midpoint between the best buy and best sell order, and when you buy you take the sell side of the book, climbing past several price levels on a thin market before your order fills. In one live Polymarket snapshot we captured for this piece, a contract with a 4.0 percent midpoint filled a hypothetical $100 buy at an average of 4.202 cents, a $1,000 buy at 4.446 cents, and a $10,000 buy at 5.342 cents, roughly 34 percent above the quote. Reading depth, spread, and slippage before you size an order is the whole skill, and it is why Kresmion shows live book depth and slippage estimates on every market detail page.

Key takeaways

What the price on the screen actually means

A prediction market contract is a share of one dollar: a YES share pays exactly $1 if the event happens and $0 if it does not. So price maps to probability. A 4 cent share implies 4 percent, your gross payoff multiple is 1 divided by the price (25 times at 4 cents), and your breakeven probability equals your average purchase price: buy at 4 cents and the event must occur 4 percent of the time to break even; at 5 cents you need 5 percent.

Three prices sit near that contract. The last price is the most recent completed trade, history that can sit well below the current best offer. The midpoint (mid) is the best buy plus best sell divided by two, the standard reference for the market's fair estimate. The ask is what a seller will currently accept. A buyer pays the ask side, so when a market reads 4 percent (usually the last or the mid), your first share costs the best ask, already at or above it.

The order book, in plain terms

An order book is the live list of unfilled limit orders. Bids are prices buyers wait to pay; asks (offers) are prices sellers wait to receive. The best bid is the highest bid, the best ask is the lowest ask, and depth is how many shares rest at each price. A market buy eats the cheapest asks first, then climbs; if only a few dollars sit at the best ask, a larger order walks up the book, paying worse at each step. That climb is slippage. (For the underlying idea, see our explainer on the bid-ask spread.)

One cost is visible before you trade a single share. The spread is best ask minus best bid, the cost of immediacy even for a tiny order. In our example the best bid was 3.9 cents and the best ask 4.1 cents, a spread of 0.2 cents, or 5.0 percent of the 4.0 cent mid. A buyer does not pay the whole spread on entry, though: buying one share at the 4.1 cent ask instead of the 4.0 cent mid costs about 2.5 percent, half the spread. The full 5 percent is the round-trip cost of buying and later selling, and this piece measures entry only.

A live example, timestamped

We pulled the YES order book for "Will the Fed increase interest rates by 25 bps after the July 2026 meeting?" from Polymarket's public, unauthenticated order-book endpoint at 2026-07-16 00:36:20 UTC. We chose it as the highest 24-hour-volume active market priced near 4 percent.

Now the slippage. We walk the visible asks: sort them cheapest first, spend a fixed dollar amount, and consume the displayed resting size at each level.

To place $10,000 into a market quoted at 4.0 percent, the buy consumed 29 price levels and averaged 5.342 cents, about 34 percent more per share than the mid and about 30 percent more than the best ask alone. Even the $100 buy paid 4.202 cents, up 2.48 percent versus the best ask, because only $18.28 of shares rested at the 4.1 cent best ask. The very top of the book was nearly empty.

Depth, not headline volume, sets your fill

This market was not obscure: about $13.8 million in lifetime volume and about $568,000 in the prior 24 hours. Volume tells you a market is active; it does not tell you what your order costs right now. What matters is the shape of the book. Only $18.28 of ask-side notional sat at or below 4.1 cents, and $18,701.48 by 10 cents, out of roughly $22.35 million of asks in total, almost all of it far above the quote and useless to a buyer near 4 percent. Cost is also not linear: a 10-times-larger order pays more than 10 times the dollars of slippage, because each deeper level of the book is priced worse than the one above it. The per-share average fill still climbs more slowly than the order grows, but the total dollars you lose to slippage grow faster than your order size.

The tick-size wrinkle, and an honest correction

A tick is the smallest price increment a market allows, and at very low prices a coarse tick can loom large: at 4 cents, a 1-cent tick would be 25 percent of the price. That is a real risk in principle, but for this market it did not bite. Polymarket assigns each market a tick from a set of allowed values (documented as 0.1, 0.01, 0.001, and 0.0001 in dollars), and the Fed market traded on a 0.001 tick, only about 2.4 percent of its 4.1 cent price. We sampled six active low-probability markets: five used the finer 0.001 tick and only one (a Bitcoin market near 9.5 cents) used the coarse 0.01 tick. Many low-probability markets sit on that finer grid, so treat coarse ticks as a conditional risk that bites markets still on a 1-cent grid, not the universal explanation. In our example the dominant costs were spread and slippage, not the tick.

Two things that are not slippage

Keep three ideas apart, because blurring them is where people go wrong. Slippage is not a fee. No platform charges it and no one banks it; it is the mechanical result of your order consuming several price levels, with the traders who posted the higher asks receiving their prices. Real fees are separate: Polymarket documents a taker-only fee (makers pay zero) that peaks, in dollar terms, when the price is near 50 cents, but is at most about 1.8 percent of the $1 payout, nothing like the price itself. As a share of what you actually stake, that fee is somewhat larger at low prices than at 50 cents, so a buyer near 4 percent should not read the 50 cent peak as a break. Polymarket also waives fees on certain world-events markets. This piece models no fees, so its numbers are entry-side slippage only and real all-in costs could be higher.

The favorite-longshot bias is a valuation effect, not an execution cost. Racetrack and sports-betting research (from Griffith in 1949 onward) finds longshots tend to be overbet and overpriced while favorites are underbet. In a large US horse-racing study, betting every favorite lost about 5.5 percent per dollar, random betting lost about 23 percent, and betting horses at odds of 100/1 or longer lost about 61 percent. This bias sits in the price estimate itself, separate from slippage. It is well-documented in betting markets but mixed and unquantified in prediction markets specifically, so we assert no particular discount here. Past patterns need not repeat.

A well-priced market is not a good fill

Prediction-market prices approximate real probabilities, are often reasonably accurate, and have at times forecast elections more closely than polls. But that describes the mid as a probability estimate, not the price you transact at: even a fair mid is negative expected value once spread, slippage, and any fees exceed your edge. Whether the number is right, and how much worse than it you trade, are two different questions, and this piece is about the second.

Methodology and sources

Method: one live example, one market, one moment, fully timestamped. The live quote and depth came from Polymarket's public, unauthenticated order-book endpoint (clob.polymarket.com/book) for the YES token of the Fed +25bps July 2026 market, with a book server timestamp of 2026-07-16 00:36:20 UTC. Slippage was computed by walking the displayed asks: sorting them cheapest first, spending $100, $1,000, and $10,000 against the posted resting size, then reporting the size-weighted average fill against the midpoint and the best ask. That one-way gap between your average fill and the mid is what market microstructure calls price impact, or the effective half-spread; doubling it to count a round trip would give the effective spread proper, but this piece measures entry only. Market selection and coverage figures came from the Kresmion production database of Polymarket markets; the tick size came from Polymarket's public market-metadata endpoint.

Data window: the live book snapshot is a single instant at 2026-07-16 00:36:20 UTC. Database coverage of Polymarket market snapshots begins 2026-04-30. Every figure here is illustrative and not durable: the database row for this same market already read about one tick stale within 28 minutes.

Limitations

• One market, one instant. The book moves in seconds. The database snapshot for the same market read 4.1 cents / 4.2 cents at 00:10:09 UTC while the live book read 3.9 cents / 4.1 cents at 00:36:20 UTC, about one tick apart within half an hour. Every number here is a photograph, not a rule.
• The walk assumes the visible book is real and static. It consumes posted asks at their posted sizes and does not model makers cancelling or refreshing. A real order can do better (hidden or replenishing size) or worse (quotes pulled the instant you reach for them).
• No fees, and entry only. We model no trading, gas, or settlement fees, and no cost of exiting across the same spread later. The figures are one-directional entry slippage.
• One venue, one database, with gaps. Only Polymarket is covered, and the database carries a live best bid for only about 61 percent of active markets, so a database-wide study silently drops many. The live endpoint, not the cached fields, is the authority.

Frequently asked questions

Why does a $100 order pay more than 4 cents when the market says 4 percent?

Because 4 percent is the midpoint, and you buy on the ask side. In the live example the best ask was 4.1 cents and only $18.28 of shares rested there, so even a $100 order climbed to a 4.202 cent average across three price levels. The mid is a reference point, not an offer to sell you shares.

Is slippage a hidden fee that Polymarket charges me?

No. Slippage is not charged by anyone; it is the arithmetic of your order consuming several price levels. Fees are separate: Polymarket documents a taker-only fee schedule, and this piece models no fees, so treat its slippage figures as a floor on total cost, not the whole of it.

Where can I see this on Kresmion?

Kresmion market detail pages show live order-book depth and slippage estimates for standard order sizes, read from the public order book. Browse Polymarket coverage at /intel/polymarket, see what is moving at /intel/markets/movers, scan public odds at /odds, and read how we source and caveat this data at /about/methodology#prediction-markets.

Sources

• Wolfers and Zitzewitz, "Prediction Markets," Journal of Economic Perspectives 18(2), 2004: market prices approximate probabilities and are broadly accurate.
• Snowberg and Wolfers, "Explaining the Favorite-Longshot Bias," NBER Working Paper 15923, 2010: low-probability longshots are systematically overpriced.
• Berg, Nelson and Rietz, "Prediction market accuracy in the long run," Iowa Electronic Markets: prediction markets have at times outperformed polls (qualitative claim).
• Polymarket documentation, Orders and CLOB overview: all orders are limit orders; a market order fills at the best available book prices; allowed tick sizes of 0.1, 0.01, 0.001, 0.0001.
• Polymarket documentation, Fees: a taker-only fee (makers zero) set by fee = C x feeRate x p x (1 - p), which in dollar terms is largest near a 50 cent price and small relative to the $1 payout, with certain world-events markets fee-free.
• The Block, "What is order book depth and why does it matter": a plain-language explainer on how deep and thin books absorb orders.

Live data: Polymarket public order-book endpoint (clob.polymarket.com/book) and market-metadata endpoint (clob.polymarket.com/markets), captured 2026-07-16 00:36:20 UTC.