Coskewness and Cokurtosis: Tail Co-Movement

Key Takeaways

• Coskewness cokurtosis are higher-order co-moments that measure joint tail behavior between assets, beyond ordinary covariance.
• Negative coskewness means assets fall hard together, which adds dangerous downside skew to the whole portfolio.
• The common mistake is diversifying on correlation alone, which can leave assets that still crash together.
• These co-moments shape a portfolio's true tail risk, so they belong in any serious risk model.

What It Is

Covariance measures how two assets move together on average, in the second moment. Coskewness cokurtosis carry that idea into higher moments. Coskewness is a third-moment co-movement and cokurtosis is a fourth-moment co-movement.

Coskewness measures whether two assets tend to make extreme moves in the same direction relative to a third variable, often the portfolio or market. Cokurtosis measures whether the tails of assets move together, that is, whether they produce simultaneous extreme outcomes.

Where skewness and kurtosis describe one asset, these cross moments describe the joint distribution. They are the inputs that explain why a portfolio's own skewness and kurtosis can differ so much from the simple average of its parts.

The Intuition

Diversification works because assets do not all move the same way at the same time. Correlation captures this for ordinary moves. But correlation can be deceptively low in calm markets and then spike toward one during a crisis, when everything sells off together.

Coskewness cokurtosis capture that crisis behavior directly. Two assets might have modest correlation day to day yet have strong positive cokurtosis, meaning they tend to crash on the same days. A portfolio built to look diversified by correlation alone can be dangerously concentrated in the tails.

The practical lesson is that real diversification means low co-movement in the extremes, not just on average. An asset that zigs when others zag in normal times but crashes alongside them in panics offers far less protection than its correlation suggests.

How It Works

Coskewness between two assets, measured against the second one squared, takes the expected product of the first asset's deviation and the square of the second's deviation:

Coskew(X, Y) = E[ (X - Xmean) * (Y - Ymean)^2 ] / (sigma_X * sigma_Y^2)

Cokurtosis extends this to the fourth moment, here measuring how X co-moves with the cube of Y's deviation:

Cokurt(X, Y) = E[ (X - Xmean) * (Y - Ymean)^3 ] / (sigma_X * sigma_Y^3)

Where:

X, Y     = returns of two assets
Xmean    = mean return of X
sigma_X  = standard deviation of X
E[ ]     = expected value, estimated as the sample average

For a full portfolio, these terms are organized into a coskewness tensor and a cokurtosis tensor, the higher-moment analogues of the covariance matrix. They grow quickly with the number of assets, which is why estimating them reliably is hard and why shrinkage methods are common.

Worked Example

Consider three assets. By covariance, asset C looks like a fine diversifier: its correlation with A and B is near zero in normal months.

Now examine cokurtosis. In the three worst months of the sample, C fell sharply at the same time as A and B. Its cokurtosis with the portfolio is strongly positive, meaning it joins the tail events even though it decouples in calm periods.

Adding C lowers measured variance, so a mean-variance optimizer happily includes it. But because of the positive cokurtosis, the combined portfolio's kurtosis rises and its left tail thickens. The optimizer, blind to higher moments, has quietly increased crash risk while reporting lower volatility. Including the co-moments would have flagged C as a poor diversifier where it counts.

Common Mistakes

1. Diversifying on correlation alone. Low average correlation does not mean low tail co-movement. Positive cokurtosis can leave assets that crash together regardless of their correlation.

2. Trusting mean-variance optimizers. Standard optimizers use only the covariance matrix. They ignore coskewness cokurtosis entirely and can build portfolios with hidden tail concentration.

3. Underestimating estimation error. Higher co-moments need long, clean samples. With few assets and short histories, the tensors are noisy and easy to overfit.

4. Forgetting the reference variable. Coskewness is defined relative to a third variable, often the market or portfolio. Reporting it without saying against what makes the number meaningless.

5. Treating them as exotic curiosities. These co-moments drive real drawdowns. Skipping them is why some diversified portfolios still suffered severe losses in 2008.

Frequently Asked Questions

What is coskewness cokurtosis in simple terms? Coskewness cokurtosis measure whether different assets tend to make big moves at the same time. They tell you if your holdings are likely to crash together, which ordinary correlation can miss.

How does coskewness cokurtosis affect investment decisions? They reveal whether your diversification holds up in a crisis. If holdings have high cokurtosis, you may reduce them or add a hedge, because the portfolio will take a heavier tail loss than its volatility implies.

What is a real-world example of coskewness cokurtosis? In the 2008 crisis, many assets that looked uncorrelated in calm years fell together. That simultaneous tail collapse is exactly what high cokurtosis captures and what plain correlation failed to warn about.

How can investors use these co-moments effectively? Look beyond the covariance matrix when building portfolios. Estimate co-moments over samples that include crises, prefer assets with negative or low coskewness to the portfolio, and use shrinkage to control estimation noise.

How is coskewness different from covariance? Covariance measures average joint movement, the second moment. Coskewness measures joint movement in the extremes, the third moment. Two assets can have low covariance yet strong coskewness, crashing together while drifting apart in calm times.

Sources

1. BreakingDownFinance. "Coskewness." https://breakingdownfinance.com/finance-topics/finance-basics/coskewness/
2. FinanceTrain. "Interpretation of Skewness, Kurtosis, Coskewness, Cokurtosis." https://financetrain.com/interpretation-of-skewness-kurtosis-coskewness-cokurtosis
3. QuantAtRisk. "Coskewness and Cokurtosis Computation for Portfolio Managers." http://www.quantatrisk.com/2013/01/20/coskewness-and-cokurtosis/
4. "Some connections between higher moments portfolio optimization methods." arXiv:2201.00205. https://arxiv.org/pdf/2201.00205

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.