Correlated Positions: How Linked Bets Amplify Risk
Correlated positions can triple your actual risk exposure. Learn the 4-step portfolio audit process with worked examples and position sizing adjustments.
What Correlated Positions Are and Why They Matter
Correlated positions are bets that move in the same direction when the same real-world event occurs. You hold five prediction market contracts. Each one is +EV. You sized each at 5% of bankroll using Kelly. You feel diversified. But three of those contracts all lose if a recession hits. "Fed cuts rates," "unemployment above 4.5%," and "S&P below 4000" are not three independent bets. They are one 15% bet wearing three disguises.
The danger is not obvious from looking at individual positions. Each one passes the EV test on its own. The risk lives in the joint probability of losing multiple positions simultaneously. Correlation is the multiplier that turns a well-diversified portfolio into a concentrated one.
Understanding and measuring correlation is the step that separates portfolio construction from position stacking. The math is straightforward. The discipline to apply it is the hard part.
How Correlation Works in Prediction Markets
Correlation measures how strongly two outcomes move together. The scale runs from -1.0 to +1.0.
- +1.0 (perfect positive): Both positions always win or lose together. You have effectively doubled your bet.
- 0.0 (uncorrelated): The outcomes are independent. One winning tells you nothing about the other.
- -1.0 (perfect negative): When one wins, the other always loses. They hedge each other.
Most prediction market positions fall between 0.3 and 0.7 correlation. This is the danger zone. Correlated enough to amplify drawdowns, but not so obviously linked that you would notice without running the numbers.
Here is a concrete example. You hold three $500 positions:
- "Democrats win the presidency" at $0.55
- "Democrats win the Senate" at $0.48
- "Democrat wins Georgia" at $0.50
These are not the same bet. But they share a common driver: Democratic voter turnout. If turnout disappoints nationally, all three lose.
Estimated pairwise correlations: presidency-Senate ~0.70, presidency-Georgia ~0.60, Senate-Georgia ~0.65.
If the positions were independent, the probability of all three losing simultaneously is: 0.45 x 0.52 x 0.50 = 11.7%
With these correlations, the joint loss probability jumps to approximately 30%. That is nearly 3x the independent estimate. Your maximum loss is still $1,500, but the chance of actually hitting that maximum is far higher than your individual position sizes suggest.
The Portfolio Variance Formula
For two positions with correlation r, the combined variance is:
Var(portfolio) = w1^2 x v1 + w2^2 x v2 + 2 x r x w1 x w2 x sqrt(v1 x v2)
Where w is position size and v is the variance of each bet. The third term is the correlation effect. When r = 0, it disappears and you get simple additive variance. When r > 0, it inflates your total risk beyond what the individual positions would suggest.
Worked example with two positions:
Position A: $1,000 at $0.55 (variance = 0.55 x 0.45 = 0.2475) Position B: $1,000 at $0.48 (variance = 0.48 x 0.52 = 0.2496) Correlation: r = 0.70
Independent portfolio variance (r = 0): $1,000^2 x 0.2475 + $1,000^2 x 0.2496 = $497,100
Correlated portfolio variance (r = 0.70): $497,100 + 2 x 0.70 x $1,000 x $1,000 x sqrt(0.2475 x 0.2496) = $497,100 + 2 x 0.70 x $1,000,000 x 0.2486 = $497,100 + $348,000 = $845,100
The correlation term adds $348,000 of variance. That is a 70% increase in portfolio risk from what the individual positions implied. Run your actual position set through the correlation calculator to see the combined exposure.
Four Rules for Managing Correlated Positions
1. Size Down When Correlation Is Above 0.5
If two positions share correlation above 0.5, treat them as partially the same bet. A practical rule: multiply your Kelly-recommended size by (1 - r/2) for each correlated position.
At r = 0.60, the adjustment factor is 1 - 0.30 = 0.70. If Kelly says bet $500, you bet $350. At r = 0.80, the factor drops to 0.60. Kelly's $500 becomes $300.
This is conservative. But full Kelly is already aggressive for most traders. Reducing size on correlated positions is the equivalent of moving from full Kelly to fractional Kelly on the correlated portion of your portfolio. Read the Kelly Criterion guide for why fractional Kelly is almost always the right call.
2. Seek Negative Correlation as a Hedge
Adding a position that is negatively correlated with your existing portfolio reduces total variance even if the individual position has only marginal +EV.
In the Democratic sweep example above, adding a "Republicans win the House" position at the right price offsets some of the directional risk. If Democrat turnout disappoints and your three Democratic positions lose, the Republican House position wins and absorbs part of the drawdown.
The math: adding a position with r = -0.40 to a portfolio reduces variance by the same magnitude that adding a position with r = +0.40 increases it. Negative correlation is a portfolio tool, not just a standalone bet evaluation.
3. Identify the Single Event That Breaks Your Portfolio
This is the most important audit question. Look at every open position and ask: what single real-world development would cause three or more of these to lose simultaneously?
Common shared drivers in prediction markets:
- Economic conditions: recession, inflation, employment data
- Political sentiment: turnout, approval ratings, party momentum
- Regulatory action: CFTC rulings, state legislation, platform shutdowns
- Market-wide panic: crypto crash, geopolitical shock
If the answer comes easily, you are overexposed to that single driver. Reduce size or add a hedge.
4. Audit After Every New Position
Do not wait for a drawdown to discover your portfolio was correlated. Before adding any new position, estimate its correlation with every existing position. If the average correlation with your current book exceeds 0.4, you need to size down or pass.
Use the PM EV calculator to verify that each position is individually +EV. Then use the correlation calculator to check that the portfolio-level risk is acceptable. Individual +EV is necessary but not sufficient. The portfolio must also be sound.
Correlated Positions Across Platforms
Correlation risk gets more complex when you hold positions on multiple platforms. You might have a "Democrats win" contract on Polymarket, a "Biden approval above 45%" contract on Kalshi, and a Democratic Senate bet on a sportsbook. These are on different platforms but driven by the same underlying factor.
Cross-platform example: crypto regulation cluster. You hold "Bitcoin above $100K" on Polymarket at $0.62, "SEC approves spot ETH ETF" on Kalshi at $0.45, and "Coinbase stock above $300" on Robinhood at $0.38. These look like three different bets. They are all driven by one factor: regulatory sentiment toward crypto. If the SEC announces a crackdown, all three lose simultaneously. Estimated pairwise correlations: BTC-ETF ~0.55, BTC-Coinbase ~0.65, ETF-Coinbase ~0.50. Your $1,500 across three platforms is really one $1,500 bet on "crypto stays friendly."
Cross-platform example: sports conference futures. You hold "Chiefs win AFC" on Kalshi, "Mahomes MVP" on DraftKings, and "Chiefs over 12.5 wins" on a sportsbook. A Mahomes injury tanks all three. The correlation between team wins and player MVP is typically 0.50 to 0.70. Between conference title and regular season wins, it runs 0.60 to 0.80. Three platforms, one risk driver.
Cross-platform positions also add settlement risk on top of correlation risk. If you are hedging on one platform and the hedge is on another, settlement timing mismatches can create temporary exposure. A Polymarket contract might settle in hours while a Kalshi contract on a related event settles in days. The prediction market fees comparison breaks down how fee structures across platforms affect your net exposure on correlated positions.
For strategies on exploiting price differences across platforms, read cross-platform arbitrage. For in-play opportunities where correlations shift rapidly, the live arbitrage betting guide covers how to identify and capture real-time pricing gaps. But when building a portfolio rather than arbing, the goal is to diversify across drivers, not across platforms.
The Real Lesson: Diversification Is About Drivers, Not Positions
Holding 10 positions does not make you diversified. Holding 10 positions with low pairwise correlation makes you diversified. The number of bets is meaningless. The number of independent risk drivers is what matters.
A portfolio of 3 uncorrelated positions (r = 0.0) has less risk than a portfolio of 10 highly correlated positions (r = 0.7). The math proves this unambiguously through the portfolio variance formula above.
Before placing your next trade, map every open position to its primary risk driver. Count the unique drivers. If more than half your capital is exposed to a single driver, you do not have a portfolio. You have a leveraged bet.
If you are holding positions across sportsbooks and prediction markets, the comparison of sportsbooks vs prediction markets explains how the different platform structures affect your risk profile beyond just correlation.
Frequently asked questions
- What are correlated positions in prediction markets?
- Correlated positions are bets that tend to win or lose together because they share a common underlying driver. For example, contracts on Democratic presidential, Senate, and state-level wins are all correlated because they depend on the same voter turnout and political sentiment.
- How do correlated positions affect my bankroll risk?
- Correlation increases the probability of multiple simultaneous losses. Three positions with 0.65 average correlation have roughly 3x the joint loss probability of three independent positions. Your effective risk exposure is much higher than the sum of individual position sizes would suggest.
- How should I adjust position sizing for correlated bets?
- Multiply your Kelly-recommended size by (1 - r/2) for each correlated position, where r is the pairwise correlation. At r = 0.60, you reduce each position by 30%. This keeps your portfolio-level risk consistent with your intended Kelly fraction.
- Is a correlated parlay calculator the same as a correlation calculator?
- They address related problems. A correlated parlay calculator adjusts parlay odds for non-independent outcomes. A correlation calculator measures how multiple positions in your portfolio interact to amplify or reduce total risk. Both account for the fact that outcomes are not independent.
- How many uncorrelated positions do I need for proper diversification?
- There is no fixed number. What matters is that positions are driven by different underlying factors. Three truly uncorrelated positions (r near 0) provide more diversification than ten positions with r = 0.7. Focus on independent risk drivers rather than position count.