How Prediction Market Prices and Odds Work
A technical deep-dive into the price-to-probability relationship, how order books, LMSR, and AMMs set prices, what spreads and slippage cost you, and how fees erode returns in prediction markets.
The price of a prediction market share is simultaneously a number you pay and a probability estimate. Understanding why — and what that dual nature implies for how you should trade — requires working through the mechanics of price formation, market structure, and cost accounting. This article goes deeper than the basics; it assumes you already understand what prediction markets are.
The Price–Probability Relationship: Formal Basis
A prediction market share is a binary option that pays $1 if condition C holds and $0 otherwise. In a frictionless, liquid, rational market, the no-arbitrage price of this share equals the market’s implied probability of C.
To see why: suppose the share trades at $p$. If you believe the true probability is $q > p$, buying the share has positive expected value: $(q \times $1) + ((1-q) \times $0) - p = q - p > 0$. Rational buyers will purchase until the price rises to $q$. Similarly, if $p > q$, rational sellers short the share until price falls to $q$. Equilibrium price equals true probability under these assumptions.
In practice:
- Friction exists. Bid-ask spreads, fees, and illiquidity mean you trade at prices that deviate from true probabilities. You effectively pay a tax on both sides of every trade.
- Market participants are not all rational. Emotional trading, herd behavior, and limited information skew prices.
- Thin markets can be moved. A participant with large capital can push prices away from true probabilities, creating temporary mispricing.
Despite these departures, prediction market prices are surprisingly well-calibrated empirically. Shares priced at 70 cents do resolve YES roughly 70% of the time across large samples. The calibration is imperfect but real.
Order Books: How Prices Are Discovered
Most major prediction markets — including Polymarket and Kalshi — use central limit order books (CLOBs). This is the same structure used by stock exchanges.
Bids are orders to buy at or below a specified price. Asks (or offers) are orders to sell at or above a specified price. The spread is the gap between the best (highest) bid and the best (lowest) ask.
Best ask: 0.63 (seller willing to accept $0.63)
-------- spread: $0.02 --------
Best bid: 0.61 (buyer willing to pay $0.61)
When you place a market order, you trade at the best available price immediately — buying at the ask or selling at the bid. When you place a limit order, you specify a price and wait for the market to come to you.
The spread has a critical implication for returns. If you buy at $0.63 and immediately sell, you receive $0.61 — a round-trip cost of $0.02 on a $0.63 position, or roughly 3.2%. To profit from trading, your information advantage must exceed the round-trip cost.
Depth and Slippage
The order book has depth — multiple layers of bids and asks at different prices. A small trade may execute entirely at the best available price. A large trade will “walk the book,” consuming multiple layers:
Ask layers:
0.63 — 1,000 shares available
0.64 — 2,500 shares available
0.65 — 5,000 shares available
0.67 — 10,000 shares available
If you buy 15,000 shares, you exhaust the first three layers and pay an average price above $0.65. The difference between the price you expected (best ask: $0.63) and the average you paid is slippage. In thin prediction markets, slippage on large orders can be severe — a few percent easily, sometimes much more.
Slippage is particularly important on prediction markets because large, informed traders are often the ones most motivated to trade large sizes. An institution that has a strong probabilistic view will want to take a large position, but doing so at acceptable prices may be impossible in a thin market.
Logarithmic Market Scoring Rules (LMSR)
Order books require matching buyers and sellers. What happens when no natural counterparty exists? This is the core problem LMSR was designed to solve.
LMSR is an automated market maker where a mathematical function defines the price at every quantity. The mechanism was formalized by economist Robin Hanson and is used in several academic and corporate internal prediction markets.
The key property: LMSR provides infinite liquidity (you can always trade at some price) but at increasing cost as you move the market. The price function is:
price(YES) = e^(q_yes / b) / (e^(q_yes / b) + e^(q_no / b))
Where q_yes and q_no are shares outstanding and b is a liquidity parameter. As you buy more YES shares, the price of YES rises continuously.
The practical effect: LMSR ensures a market never goes dry, but large trades push prices significantly. The market maker absorbs all this by running at a loss bounded by b × ln(2) — this is the guaranteed maximum loss the LMSR operator accepts, in exchange for providing always-on liquidity.
LMSR is less common in high-volume retail markets (order books are more efficient when there are many natural buyers and sellers) but remains important in low-liquidity, specialized prediction markets.
Automated Market Makers (AMMs) in Prediction Markets
DeFi-native prediction markets sometimes use liquidity pool AMMs — the same structure used by Uniswap and similar exchanges. A pool holds both YES and NO tokens. A constant-product formula (x × y = k) determines prices from the ratio of tokens in the pool.
AMMs provide always-on liquidity but introduce several problems specific to prediction markets:
Impermanent loss becomes permanent loss at resolution. In a standard AMM for two assets that fluctuate, liquidity providers experience “impermanent loss” that can reverse if prices mean-revert. In a prediction market, prices converge to either 0 or 1 at resolution — the loss for liquidity providers who held the losing side is permanent and total.
Price impact is predictable. Since the pricing formula is public, sophisticated traders know exactly what a large trade will cost. This is transparent but means AMMs are not well-suited to markets where participants want to hide their order size.
Fees are built into every swap. Most AMMs charge a fixed percentage per trade, which compounds across multiple trades.
Fee Structures and Their Impact on Expected Returns
Different platforms and market structures charge fees in different ways. Understanding the true all-in cost is essential for evaluating whether trading makes economic sense.
| Cost type | Example | Impact |
|---|---|---|
| Maker/taker spread | 2¢ on a $0.50 share | 4% round-trip cost |
| Platform trading fee | Kalshi: ~7% of profit | Reduces net payout on wins |
| Gas fees (on-chain) | Polygon: fractions of a cent | Minimal at current rates |
| Bridging/withdrawal fees | Varies by network | 0.1–0.5% of amount moved |
| USDC conversion | Exchange spread | 0.1–0.5% each way |
Consider a concrete example. You believe an event has a 70% chance of occurring. The market prices it at 60 cents. Your edge is 10 percentage points.
- You buy at the ask: $0.61 (spread eats one cent)
- If YES, you receive $1.00 minus a 7% fee: $1.00 − ($1.00 − $0.61) × 0.07 = $0.973
- Your actual profit: $0.973 − $0.61 = $0.363 per share
- Your actual return given 70% win rate: (0.70 × $0.363) + (0.30 × −$0.61) = $0.254 − $0.183 = $0.071
Without fees, the same edge generates: (0.70 × $0.40) + (0.30 × −$0.60) = $0.28 − $0.18 = $0.10 per share. Fees cut expected profit by roughly 30% in this example.
This is why the question “what is my edge?” must be followed immediately by “is my edge larger than the all-in cost of trading?” For small edges, fees frequently eliminate expected profit entirely.
Calibration, Bias, and Market Inefficiencies
Prediction markets are generally well-calibrated at the aggregate level, but systematic biases exist that sophisticated traders attempt to exploit:
Longshot bias. Participants tend to overvalue low-probability outcomes. A 5% share may trade at 8 cents because people overweight small probabilities. This is a well-documented finding in both sports betting and prediction markets.
Favorite-longshot reversal near resolution. As an event approaches, prices sometimes overshoot the true probability as late information arrives with high volatility.
Liquidity concentration at round numbers. Bids and asks tend to cluster at 0.10, 0.25, 0.50, 0.75, 0.90. Prices between these levels may briefly offer small edges.
Sentiment-driven mispricing. Markets with high public interest — major elections, championship games — attract emotionally motivated traders whose behavior creates temporary inefficiencies. Exploiting these requires being on the other side of public sentiment, which is psychologically difficult.
Summary
Prediction market prices encode probability estimates, but the cost of accessing those prices — spreads, fees, slippage — means the break-even point is not the fair-value price but a price that accounts for all-in friction. Market structure (order books, LMSR, AMMs) determines how that friction is distributed between participants and market makers.
For participants, the practical lessons are: trade only in liquid markets where spreads are narrow; understand the fee structure before sizing positions; recognize that slippage scales with position size; and be skeptical of edges that look good before fees but marginal after.
For more context on where these mechanics can go wrong, see prediction market risks and manipulation, or read about the fundamentals of crypto and blockchain for background on the on-chain infrastructure these markets run on.