TL;DR — Prediction market prices are probabilities. A $0.65 Yes share = 65% implied probability. If you think the real probability is 75%, that’s a +EV trade. Finding these mispricings consistently — and trading them before the market corrects — is the entire game. AI agents do it at scale.
Price = Probability
Every prediction market contract is a binary bet. It pays $1.00 if the event happens, $0.00 if it doesn’t.
The price of a Yes share is the market’s consensus probability.
Price │ Market Says │ You Think │ Action
─────────┼────────────────┼───────────────┼────────────
$0.65 │ 65% likely │ >65% likely │ Buy Yes
$0.65 │ 65% likely │ <65% likely │ Buy No ($0.35)
$0.65 │ 65% likely │ =65% likely │ No edge, don't trade
That’s the entire framework. Everything else — odds conversion, expected value, strategy — builds on this single concept.
The Math of Binary Contracts
Every market has two sides. They always sum to $1.00.
┌───────────────────────────────────────────┐
│ "Will the Fed cut rates in June 2026?" │
│ │
│ Yes: $0.55 ←──── 55% probability │
│ No: $0.45 ←──── 45% probability │
│ Sum: $1.00 ←──── always │
│ │
│ Buy Yes at $0.55: │
│ Event occurs → receive $1.00 │
│ Profit: $1.00 - $0.55 = $0.45 │
│ │
│ Event doesn't occur → receive $0.00 │
│ Loss: $0.55 │
└───────────────────────────────────────────┘
Your maximum loss on any prediction market trade is the amount you paid for the shares. Your maximum gain is $1.00 minus the price you paid. This makes risk management straightforward — unlike leveraged products, you can’t lose more than your stake.
Converting Between Odds Formats
Prediction market prices, decimal odds, and American odds all express the same thing: probability and payout. You need to convert fluently between them.
Prediction Market Price → Decimal Odds
Decimal Odds = 1 / Price
$0.40 → 1/0.40 = 2.50
$0.65 → 1/0.65 = 1.538
$0.80 → 1/0.80 = 1.25
$0.10 → 1/0.10 = 10.00
Prediction Market Price → American Odds
If price ≥ $0.50 (favorite):
American = -(price / (1 - price)) × 100
$0.65 → -(0.65/0.35) × 100 = -186
$0.80 → -(0.80/0.20) × 100 = -400
If price < $0.50 (underdog):
American = ((1 - price) / price) × 100
$0.40 → (0.60/0.40) × 100 = +150
$0.20 → (0.80/0.20) × 100 = +400
Full Conversion Table
Price │ Implied Prob │ Decimal │ American │ Fractional
───────┼──────────────┼─────────┼──────────┼───────────
$0.10 │ 10% │ 10.00 │ +900 │ 9/1
$0.20 │ 20% │ 5.00 │ +400 │ 4/1
$0.25 │ 25% │ 4.00 │ +300 │ 3/1
$0.33 │ 33% │ 3.03 │ +200 │ 2/1
$0.40 │ 40% │ 2.50 │ +150 │ 3/2
$0.50 │ 50% │ 2.00 │ +100 │ 1/1
$0.60 │ 60% │ 1.67 │ -150 │ 2/3
$0.67 │ 67% │ 1.49 │ -200 │ 1/2
$0.75 │ 75% │ 1.33 │ -300 │ 1/3
$0.80 │ 80% │ 1.25 │ -400 │ 1/4
$0.90 │ 90% │ 1.11 │ -900 │ 1/9
───────────────────────────────────────────────────────
For how odds compare between sportsbooks and prediction markets, see sports betting vs prediction markets.
How the Order Book Sets Prices
Prediction markets don’t have a single “price” set by a central authority. The price emerges from supply and demand on the order book.
ORDER BOOK — "Will Bitcoin exceed $150K by Dec 2026?"
BID (Buyers) │ ASK (Sellers)
──────────────────────────┼──────────────────────────
$0.41 × 5,000 shares │ $0.43 × 3,200 shares
$0.40 × 12,000 shares │ $0.44 × 8,500 shares
$0.39 × 8,000 shares │ $0.45 × 15,000 shares
$0.38 × 20,000 shares │ $0.46 × 6,000 shares
Best Bid: $0.41 Best Ask: $0.43
Spread: $0.02 Midpoint: $0.42
The "price" is typically the midpoint ($0.42)
or the last traded price.
The bid-ask spread is the prediction market equivalent of the sportsbook’s vig. A spread of $0.02 on a $0.42 contract represents roughly a 4.8% round-trip cost. High-volume markets have tighter spreads (lower costs). Thin markets have wider spreads.
If you place a market order, you buy at the ask or sell at the bid — you pay the spread. If you place a limit order at the midpoint, you might get a better price but you wait for someone to fill it.
For the full technical breakdown of order types and API endpoints, see the prediction market API reference.
Expected Value: The Only Number That Matters
Expected value (EV) is the average profit or loss per trade, repeated infinitely. Positive EV (+EV) means the trade is profitable long-term. Negative EV (-EV) means it loses money long-term.
EV = (P_win × Profit) - (P_lose × Loss)
Where:
P_win = your estimated probability of winning
Profit = payout if correct ($1.00 - price)
P_lose = 1 - P_win
Loss = price paid per share
Example 1: Positive EV
Market says: 40% (Yes at $0.40) You estimate: 50%
EV = (0.50 × $0.60) - (0.50 × $0.40)
EV = $0.30 - $0.20
EV = +$0.10 per share
→ Over 100 trades like this, you expect to profit ~$10
Example 2: Negative EV
Market says: 40% (Yes at $0.40) You estimate: 35%
EV = (0.35 × $0.60) - (0.65 × $0.40)
EV = $0.21 - $0.26
EV = -$0.05 per share
→ Over 100 trades like this, you expect to lose ~$5
Example 3: Break Even
Market says: 40% (Yes at $0.40) You estimate: 40%
EV = (0.40 × $0.60) - (0.60 × $0.40)
EV = $0.24 - $0.24
EV = $0.00
→ No edge. Don't trade.
The entire game of prediction market trading is finding situations where your estimate of the true probability differs from the market price — and being right about that estimate more often than not.
Why Markets Misprice Events
If prediction markets are so accurate, why would there ever be an edge?
Markets are accurate on average and over time. In the short term, mispricings are everywhere. Here’s why:
1. Information Asymmetry
Not all traders have the same information. An FDA scientist knows more about drug approval odds than a college student trading on vibes. An election campaign insider knows about internal polling data. A weather modeler has better hurricane models than the average Polymarket trader.
When informed traders enter the market, the price adjusts. The window between when new information exists and when it’s fully reflected in the price is the opportunity.
2. Behavioral Biases
Traders are human (for now). Common biases that create mispricing:
- Recency bias — Overweighting the latest news. A candidate has one bad poll and the market overreacts.
- Anchoring — The price started at $0.50 and traders anchor to that level even as evidence shifts.
- Favorite-longshot bias — Longshots are consistently overpriced, favorites consistently underpriced. Traders overpay for the dream of a 10x payout.
- Herding — Traders follow the crowd rather than their own analysis.
3. Liquidity Gaps
Thin markets (few traders, low volume) don’t have enough information to find the efficient price. A $50,000-volume market is more likely to be mispriced than a $50,000,000-volume market. Smaller markets are where the edges are biggest — but also where it’s hardest to trade size.
4. News Lag
When breaking news drops, the market takes seconds to minutes to reprice. The first traders to process the information and execute a trade capture the repricing. This window is measured in seconds for major news on high-volume markets and can last minutes or hours for smaller markets.
5. Cross-Platform Discrepancies
The same event priced on Polymarket, Kalshi, and DraftKings can have different prices because each platform has different traders, different information flow, and different fee structures. See the sports betting vs prediction markets comparison for how these differences create arbitrage.
How to Find Mispriced Markets
Three approaches, ordered from simplest to most sophisticated:
1. Domain Expertise
You know more about a topic than the average trader. Maybe you’re a doctor evaluating FDA approval odds, a political operative with campaign intelligence, or an energy analyst with insights on commodity markets.
This is the simplest edge and requires no technology. But it doesn’t scale — you can only be an expert in a few domains.
2. Cross-Platform Comparison
Compare prices for the same event across Polymarket, Kalshi, and sportsbooks. When you see a meaningful discrepancy, buy the cheaper side and (if possible) sell the expensive side.
This is arbitrage — risk-free profit when the gap exceeds transaction costs. It’s the most reliable edge and the first thing most trading bots implement.
For the technical implementation, see the sports betting arbitrage bot guide and the prediction market API reference.
3. Model-Based Pricing
Build a quantitative model that estimates probabilities independently of the market. Feed it data — polls, fundamentals, historical patterns — and compare its output to market prices. When your model says 55% and the market says 45%, that’s a potential trade.
This is where AI agents become essential. No human can run a probability model across hundreds of markets simultaneously, update it in real time as new data arrives, and execute trades within seconds.
Where AI Agents Change the Game
Every mispricing source listed above is something an AI agent exploits better than a human:
| Mispricing Source | Human Trader | AI Agent |
|---|---|---|
| Information asymmetry | Knows 1-2 domains | Processes data across all domains simultaneously |
| Behavioral biases | Subject to all of them | Executes model output without emotion |
| Liquidity gaps | Monitors a few markets | Scans every market on every platform |
| News lag | Reads one article at a time | Processes hundreds of sources in parallel |
| Cross-platform gaps | Checks 2-3 platforms manually | Compares all platforms in real time |
An agent betting stack combines:
- Intelligence (Layer 4): LLMs and models that estimate probabilities — Claude, Polyseer, CrewAI
- Trading (Layer 3): APIs that execute trades — Polymarket CLOB, Kalshi API
- Wallet (Layer 2): Autonomous fund management — Coinbase Agentic Wallets
- Identity (Layer 1): Authentication and reputation — Moltbook, SIWE
The result: an agent that continuously scans markets, estimates true probabilities, identifies +EV opportunities, and executes trades — all without human intervention.
Browse current prediction market bots and agents to see what’s live today.
Putting It Together: A Trade Decision Framework
1. IDENTIFY
Find a market where you have a view different
from the current price.
2. ESTIMATE
What is your probability estimate?
Be specific: "I think this is 55%" not "I think
this is more likely than the market says."
3. CALCULATE EV
EV = (P_win × profit) - (P_lose × cost)
Is it positive? How positive?
4. SIZE
Use Kelly Criterion or flat betting.
Never bet more than your edge justifies.
5. EXECUTE
Buy at the best available price across platforms.
Use limit orders to reduce spread costs.
6. MONITOR
Track the market. Did the price move toward
your estimate (good sign) or away from it?
7. EVALUATE
After resolution: were you right?
Over many trades: is your CLV positive?
For beginners: do steps 1-7 manually for your first 20-50 trades. Track everything in a spreadsheet. This teaches you the mechanics and helps you calibrate your probability estimates.
For builders: automate steps 1-7 with an agent. Start with the agent betting stack guide.
What’s Next
- New to prediction markets? → Prediction Markets 101
- New to sports betting? → Sports Betting 101
- Compare the two ecosystems → Sports Betting vs Prediction Markets
- Advanced strategies → Sharp Betting Guides
- Build arbitrage bots → Sports Betting Arbitrage Bot Guide
- Full agent architecture → The Agent Betting Stack
- API reference → Prediction Market API Reference