Risk Scoring: The Hidden Leverage
How to weigh tail-risk before you commit — Kelly, liquidity, correlation, and resolution traps.
Kelly Criterion, Explained Simply
The Kelly Criterion answers one question: given an edge and the payoff structure, what fraction of my bankroll should I bet to maximize long-run growth? For a binary market where you estimate probability p and the contract trades at price c, the full-Kelly fraction is roughly (p − c) / (1 − c) for a YES trade. If you estimate 65% and the price is 50¢, full-Kelly says 30% of bankroll.
That's way too much for almost everyone. Full Kelly assumes your edge estimate is perfect, which it never is. Most professional traders use quarter- or half-Kelly, which dramatically reduces drawdowns for a relatively small cost in expected growth.
Why Most People Over-Bet
Three cognitive traps. First, conviction inflation — when you're excited about a thesis, your brain rounds 62% probability up to 'obvious'. Second, ignoring variance — a 65% edge loses 35% of the time, which over ten bets means 3–4 losses, any of which can blow an oversized account. Third, the winner's curse — after a big win, traders size up out of emotional momentum rather than math.
The fix is mechanical: precommit to a maximum position size as a percentage of bankroll and never deviate. 3–5% per market for most traders. Write it down. The moment you find yourself justifying an exception is the moment the bankroll is at risk.
Liquidity Risk in Thin Markets
A contract with $5,000 of 24-hour volume and a $200 order book depth looks tradeable until you try to exit. On that size of market, a $500 sell order can move price 3–5¢ against you — a real 5–10% hit to your PnL that never shows up on the headline quote.
Check volume and book depth before you size in: a market with thin liquidity is expensive to exit, so treat it as hold-to-resolution or skip it. When the Analyzer sees an edge too small to be worth thin liquidity, it returns PASS rather than a recommendation — the absence of a call is itself information.
Correlation Risk
If you're YES on three political markets that all resolve the same direction under the same partisan wave, you don't have three positions — you have one bigger position. The risk scoring engine detects common resolution drivers across your open book and flags when cumulative correlated exposure crosses your comfort threshold.
This matters most in election seasons (presidential and Senate races move together), in macro (Fed cut markets across months are tightly linked), and in crypto (BTC-to-$X and ETH-to-$Y are not independent). The rule of thumb: total correlated exposure should be no more than 2–3x your max single-position size.
Resolution Risk and Arbitration
Every prediction market must eventually resolve, and resolution isn't always clean. Polymarket uses UMA's optimistic oracle, which occasionally produces contested outcomes on ambiguous markets. Kalshi has more rigid resolution rules but its sports and macro markets still rely on specific data sources that can be delayed or revised.
Read the market title and deadline carefully for ambiguity — things like 'Will X happen by Y' without a crisp data source, or multi-condition markets where only one leg is clearly verifiable. The Analyzer returns key risks alongside every call, and resolution ambiguity is one of the things it is asked to surface. Avoid these contracts or size them half as big as normal.
How Polykit's AI Surfaces Tail Risks
Every Analyzer run returns a risk score from 1 (low) to 10 (high) alongside the fair-value estimate. The score aggregates liquidity, correlation with your existing book, resolution ambiguity, and known issues with the specific market type. A 7+ doesn't mean 'don't trade' — it means 'size smaller and read the written warnings'.
The warnings are specific, not generic. 'Resolution depends on CDC final data release, historically revised twice after initial publication' is more useful than 'this is risky'. The goal is to turn invisible tail risk into a checklist you actually read before clicking buy.
Building a Max-Risk-Per-Market Rule
Combine everything: start with a base position size (say 3% of bankroll), reduce proportionally by risk score (multiply by 1 − risk/20), and cap correlated exposure at 3x a single position. That one formula, applied every trade, handles the majority of account-blowing scenarios.
Polykit's Sizing tool takes your bankroll, the Analyzer's edge and risk score, and outputs the suggested stake automatically. You can override it, but the default is the mathematically responsible answer. That's the hidden leverage in risk scoring: not picking better trades, but sizing the trades you'd pick anyway in a way that compounds instead of ruins.