Correlated Positions: The Hidden Risk in Your Portfolio

The problem you don't see

You hold 5 positions across different prediction markets. Each one is +EV. You sized each at 5% of your bankroll using Kelly. You feel diversified. But three of those positions all lose if the same underlying event happens — say, a recession hits and your "Fed cuts rates," "unemployment above 4.5%," and "S&P below 4000" contracts all move against you simultaneously.

That's not three independent 5% bets. That's one 15% bet with extra steps.

What correlation means in practice

Two positions are correlated when the same real-world outcome affects both. The degree matters. Perfect correlation (1.0) means they always move together — you've effectively doubled your position. Zero correlation means they're independent. Negative correlation (-1.0) means they hedge each other.

Most prediction market positions fall somewhere between 0.3 and 0.7 correlation, which is the danger zone: correlated enough to amplify drawdowns, but not so obviously linked that you'd notice without running the numbers.

A concrete example

You hold these three positions, each sized at $500:

• "Democrats win the presidency" — bought at $0.55
• "Democrats win the Senate" — bought at $0.48
• "Democrat wins Georgia" — bought at $0.50

These aren't the same bet, but they're highly correlated. If Democrat turnout disappoints nationally, all three lose. Estimated pairwise correlations: presidency-Senate ~0.7, presidency-Georgia ~0.6, Senate-Georgia ~0.65.

Your maximum loss looks like $1,500 (3 x $500). But because of correlation, the probability of losing on all three simultaneously is much higher than if they were independent. With independent bets, the chance of all three losing is roughly 0.45 * 0.52 * 0.50 = 11.7%. With these correlations, it's closer to 30%.

How to measure it

Use the correlation calculator to estimate the joint risk of your positions. Input your estimated pairwise correlations and position sizes, and it'll show your portfolio's worst-case exposure.

The key metric is portfolio variance. For two positions with correlation r, combined variance is:

Var = w1^2 * v1 + w2^2 * v2 + 2 * r * w1 * w2 * sqrt(v1 * v2)

Where w is position size and v is the variance of each bet. Higher r means higher combined variance — more risk than you intended.

What to do about it

1. Size down correlated positions

If two positions have correlation above 0.5, treat them as partially the same bet. A simple rule: multiply your Kelly-recommended size by (1 - r/2) for each correlated position. At r = 0.6, that's a 30% reduction in size.

2. Seek negative correlation

Adding a "Republicans win the House" position at the right price offsets some of your Democratic-leaning portfolio. The negative correlation reduces your portfolio variance even if the individual position is only marginally +EV.

3. Audit your portfolio regularly

List every open position and ask: "What single event would make three or more of these lose?" If the answer is easy to find, you're overexposed. Use the PM EV calculator to check that each position justifies its risk on a standalone basis, then adjust for correlation.

Diversification in prediction markets isn't about holding many positions. It's about holding positions that don't fail together.

For more on how to size individual positions, read the Kelly Criterion guide. And if you're holding positions across platforms, bankroll turnover explains why locked capital in long-dated contracts is a hidden cost.