Prediction Market
Strategy Guide
Prediction markets are beatable, but not by gut instinct or news-following alone. This guide covers the frameworks that matter: Kelly Criterion bet sizing, where edge actually comes from, how to avoid the biases that bleed retail bankrolls, and how to build a sustainable trading process across Kalshi, Polymarket, and other platforms.
The three inputs that determine profitability
Long-run profitability in prediction markets comes from three things, in this order of importance:
Kelly Criterion for prediction markets
The Kelly Criterion tells you what fraction of your bankroll to risk on a trade to maximize long-run growth. For binary prediction markets (YES/NO contracts paying $1):
p = your probability estimate for YES
q = market price of YES (in decimal, e.g., 0.60 for 60¢)
b = net profit per dollar risked if YES wins = (1 − q) / q
Your estimate: 72% probability of YES
b = (1 − 0.60) / 0.60 = 0.667
f* = (0.72 − 0.60) / 0.667 = 18% of bankroll
Half-Kelly recommendation: 9% of bankroll
On a $10,000 bankroll: bet $900 of YES contracts at 60¢ = 1,500 contracts
Why use half-Kelly?
Full Kelly maximizes long-run growth in theory, but it requires your probability estimates to be perfectly calibrated. In practice, traders overestimate edge: their p estimates are too confident. Half-Kelly (betting half the Kelly fraction) reduces this risk dramatically: you give up roughly 25% of maximum growth rate, but you halve variance and give yourself much longer survival through losing streaks.
Where does prediction market edge come from?
Edge is a probability estimate systematically better than the market's. Here are the most reliable sources:
Mistakes that bleed retail bankrolls
Overbetting on correlated markets
Buying YES on multiple election markets (presidential race, Senate seats, gubernatorial races) is not diversification: they all resolve in the same direction. A single bad night can wipe out a month of gains. Keep total exposure to any single resolution event under 20% of bankroll.
Chasing liquidity to exit
Binary contracts close to resolution can become illiquid. If a YES contract is at 85¢ and you want to sell, there may not be buyers at a fair price. Factor in exit risk when entering: can you actually close this position if you change your mind?
Ignoring the fee impact on edge
At $0.02 per contract on Kalshi, each round-trip costs $0.04. If you're buying a 60¢ YES contract and selling at 65¢ ($0.05 profit per contract), fees eat 80% of your gain. Fees matter most on small moves: trade less often, with larger expected value per trade.
Updating too slowly (or too quickly)
When new information arrives, markets move. The optimal response is neither to panic-exit nor to stubbornly hold. Bayesian updating: revising your probability estimate in proportion to how much the new information matters: is the systematic approach. Practice it explicitly.
No written process
Profitable prediction market trading requires a documented process: how you estimate probabilities, what Kelly fraction you use, how you track performance, when you cut losses. Without documentation, you're trading on vibes. After a win, you'll think you're better than you are; after a loss, you'll change strategies randomly.
Insufficient sample size
You cannot assess edge from 10 resolved contracts. Even 100 contracts may not be statistically meaningful if you're in similar markets. Track your calibration (predicted probability vs actual outcome rate) over at least 200 resolved contracts before drawing conclusions about your edge.
Which markets offer the best opportunity?
Not all prediction markets are equal. Here's how to evaluate which markets to focus on:
| Market type | Edge potential | Competition level | Notes |
|---|---|---|---|
| Major US elections | Low | Very high | Polymarket attracts professional quants and sharp political traders. Hard to beat. |
| Sports game outcomes | Medium | High | Sharp sports bettors participate. Requires real sports modelling. |
| Economic data releases | Medium | Medium | Economists and quants compete. Good for macro traders with models. |
| Small political markets | High | Low | Lower liquidity, harder to size, but less competition. Best for domain experts. |
| Weather / environment | High | Low | Few specialists participate. Numerical weather models can provide strong edge. |
| Science / tech milestones | Variable | Low–Medium | Long resolution timelines increase uncertainty. Good for domain experts. |
Where to trade: platform selection as strategy
Different platforms attract different participant types, which affects the edges available.
Lowest competition. Play-money, but calibration training without financial stakes. Start here if you're new to prediction markets. Sharpen your probability estimation before putting real money at risk.
Most liquid US real-money platform. Diverse participant types: retail and institutional. Best combination of market breadth and liquidity. Fee structure (0–7% of profit) rewards finding edge.
Global liquidity, highest volume. Attracts sophisticated international traders, especially on macro/political markets. Waitlisted for US users via QCEX. Competitive but with some structural advantages for non-US political markets.
Politics-only, $850 max per contract. Retail-dominated, less sophisticated on average. Better edge opportunities than Kalshi for political specialists but severe position limits constrain return.
Strategy questions
What is the Kelly Criterion and how does it apply to prediction markets? +
The Kelly Criterion is a bet-sizing formula that maximizes long-run capital growth given an edge. For a YES contract at 60¢ where you estimate 70% true probability: edge = 0.70 − 0.60 = 0.10; net odds = 0.40/0.60 = 0.667; Kelly = 0.10 / 0.667 = 15% of bankroll. In practice use half-Kelly (7.5%) to account for estimation error.
How many contracts do I need to assess my edge? +
At minimum 100 resolved contracts to start seeing patterns; 200–500 for statistical confidence. Use a calibration chart: plot your predicted probabilities in bins (0–10%, 10–20%, etc.) against actual outcome frequencies. If you're well-calibrated, they should match.
Should I trade YES or NO contracts? +
Neither is inherently better: it depends on where the mispricing is. However, retail traders systematically overweight memorable YES outcomes (upsets, popular candidates), which can make NO contracts underpriced on high-profile events. Consider whether you have a YES bias and correct for it.
What is a good annualized ROI target for prediction markets? +
A well-calibrated amateur might achieve 15–40% annualized ROI on deployed capital. Professional prediction market traders achieve higher but deploy larger capital and have systematic processes. These figures exclude the time and cognitive cost of trading. Compare against your opportunity cost before scaling up.