Correlation: the silent killer in your strategy mix.

Most systematic strategy correlation matrices are computed on raw returns. Strip the vol exposure out and the diversification claim collapses.

"We don't have correlated bets" is the LP claim that doesn't get past a sharp allocator. Either your factor model is wrong or your strategy is. Pick the one you can defend.

I've sat across from enough traders now to know the move. The manager opens their deck. Slide six is the strategy mix. "Twelve sub-strategies, low pair-wise correlation, diversified across instruments and timeframes." The pair-wise correlation matrix is rendered in green and yellow. No red. Looks great.

A sharp allocator does one thing with that matrix and the manager has usually never thought about it. They strip out vol.

The argument runs like this:

Most systematic strategy correlation matrices are computed on raw returns. Raw returns embed volatility. Two strategies that both lever up in low-vol regimes and lever down in high-vol regimes will look uncorrelated on raw returns because their timing of position size is similar but their direction is uncorrelated. Strip the vol exposure out (compute residual returns after a vol-regime factor) and the green-and-yellow matrix turns red.

Strip the vol exposure out and the green-and-yellow matrix turns red.

That's not a marginal technical point. That's the heart of the diversification claim collapsing.

A lot of systematic managers genuinely don't know this is what's happening inside their book. They built twelve strategies, the pair-wise stats came out clean, they shipped the deck. The allocator finds it in the residuals, the conversation gets awkward, the allocation goes to the manager next door.

What choice does this force on you?

Either your factor model isn't pricing the shared vol exposure (in which case your factor model is wrong), or your strategy mix is genuinely loading the same factor twelve different ways (in which case your strategy is wrong). The matrix you're showing isn't telling you which.

The honest version of the claim looks different.

"We have twelve sub-strategies. After controlling for shared vol-regime exposure, residual pair-wise correlations sit between -0.2 and +0.4, with the cluster around +0.2. Three pairs are above +0.5 and we know which ones and why. We size the book accounting for the cluster, not assuming independence."

This matters for capital raisers at any size. The pattern on the LP side is the same whether you're pitching a smaller allocation or a fund-of-funds cheque. Any serious allocator is going to do the work on your correlation claim.

For anyone running real systematic money: go and recompute your strategy-mix correlation matrix on residuals after a one-factor vol regression. If the residuals look the same as your raw correlations, you have a clean book. If they don't, you have homework before the next allocator meeting. For a deeper look at how correlation budgeting works in practice at the portfolio level, see why your strategies are less diversified than you think.