The Correlation Trap: Why Your 'Diversified' Book Is One Bet in Disguise

The Illusion of Diversification

You open your portfolio. You've got twelve positions across Kalshi and Polymarket. Fed rate cuts, a couple Senate races, Trump's next cabinet pick, a Supreme Court ruling, some Middle East geopolitics stuff. Feels balanced. Feels diversified.

It isn't.

Most of those markets share hidden wiring. When one moves, several move. And on a bad day, they all move against you at once. That's not twelve bets. That's one bet dressed up in twelve costumes.

What Correlation Actually Looks Like in Prediction Markets

In stocks, correlation is a Greek letter on a Bloomberg terminal. In prediction markets, it's messier and less obvious. Two contracts are correlated when the same underlying event drives both prices.

Examples that trip people up:

- Betting YES on multiple GOP Senate seats. These aren't independent. A national environment shift hits all of them the same day.

- Betting on "Fed cuts in December" AND "S&P above X by year-end." Same macro thesis, two tickers.

- Betting YES on Trump doing three different specific things in his first 100 days. You're not betting on three things. You're betting on one thing: that Trump governs the way you think he will.

- Recession markets, unemployment markets, and "will GDP print negative" markets. All the same trade.

If you'd get the same P&L outcome from a single news event, you have one position.

The Math That Should Scare You

Kelly sizing assumes independence. When you size six correlated bets at 2% of bankroll each, you're not risking 12%. You're risking closer to 12% on one outcome, because they'll cash or crash together.

The real number to track isn't position size. It's cluster exposure — total capital riding on a single underlying driver.

A rough rule: if you'd size a single high-conviction bet at 5% of bankroll, don't let any correlated cluster exceed roughly 7-8% total. The extra buffer accounts for the fact that you might be wrong about the correlation being partial when it's actually near-total.

How to Audit Your Book in 10 Minutes

Open your positions. For each one, write the one-sentence thesis — the underlying belief that has to be true for the bet to win.

Then group them. Anything with the same thesis is one cluster.

Example cluster you might find:

- YES on "Fed cuts 50bps by March"

- YES on "10Y yield below 4% by Q1"

- YES on "Powell replaced by 2026"

- NO on "Recession declared by Q2"

Four tickers, one thesis: rates come down and the economy holds up. If inflation prints hot next month, all four bleed.

The Sneaky Correlations

Some clusters aren't obvious until they blow up:

- Platform risk. Every Polymarket position correlates on "Polymarket has an operational issue." Rare, but real.

- Resolution source risk. Multiple markets settling off the same AP feed or same government report share tail risk if the source is late, ambiguous, or wrong.

- Attention correlation. Markets that all get liquid because of the same news cycle also all get illiquid together when attention moves on. Your exit prices are correlated even if your theses aren't.

The Fix

You don't need to eliminate correlation. You just need to price it in before you size.

Three habits:

1. Cluster your book weekly. Ten minutes. Write the theses. Group them. Look at total exposure per cluster, not per ticker.

2. Hunt for negatively correlated bets. If your book leans risk-on, find a spot where a risk-off outcome pays. Not for the hedge P&L — for the sleep.

3. When you add a position, ask: is this a new bet or more of an existing bet? If it's more of an existing bet, that's fine. Just size it against the cluster limit, not the single-position limit.

A book of twelve bets that are actually one bet isn't a portfolio. It's a coin flip with extra steps.