One nuance worth making explicit in the worked example: Cboe's COR3M methodology subtracts the single-name variance term Σwᵢ²σᵢ² before inferring average pairwise correlation.
That means (σ_index / Σwᵢσᵢ)² is a useful approximation, but not the exact COR3M calculation—especially when constituent weights are concentrated.
I'd be interested in seeing the exact top-50 reconstruction alongside the proxy before assessing how cheap correlation really is.
Have you tested how much the implied-versus-realized correlation spread and dispersion P&L change under the exact weighting?
Good catch, and it explains most of the gap we flagged. With the Cboe's exact formula wiith top-50 weights, h is around 0.05 to 0.06 today, which takes our 0.158 down to about 0.11, right where COR3M printed.
So correlation was even cheaper than the proxy suggested
On your second question: if, realized correlation is built the same way, both sides move by roughly (1 − ρ)h, so the implied minus realized spread shifts much less than the levels.
Regarding P&L, it is a different story. it depends on the vega notional per name, and concentration means a correlation-neutral book needs more of the mega caps.
We have not published the exact top-50 reconstruction yet. it is a good follow-up.
Appreciate you running the top-50 weights—that’s a meaningful reconciliation. Moving the proxy estimate from 0.158 toward ~0.11 changes the level interpretation, but as you note, it doesn’t tell us the P&L of a dispersion book by itself. I’d separate three checks: (1) reconstruct COR3M with matched weights and tenors; (2) compare implied vs realized correlation on the same basket; (3) translate the gap into vega-weighted, hedged P&L after costs. That last step is where a statistical discrepancy becomes—or fails to become—an executable edge. I’d be keen to read your top-50 follow-up.
One nuance worth making explicit in the worked example: Cboe's COR3M methodology subtracts the single-name variance term Σwᵢ²σᵢ² before inferring average pairwise correlation.
That means (σ_index / Σwᵢσᵢ)² is a useful approximation, but not the exact COR3M calculation—especially when constituent weights are concentrated.
I'd be interested in seeing the exact top-50 reconstruction alongside the proxy before assessing how cheap correlation really is.
Have you tested how much the implied-versus-realized correlation spread and dispersion P&L change under the exact weighting?
Good catch, and it explains most of the gap we flagged. With the Cboe's exact formula wiith top-50 weights, h is around 0.05 to 0.06 today, which takes our 0.158 down to about 0.11, right where COR3M printed.
So correlation was even cheaper than the proxy suggested
On your second question: if, realized correlation is built the same way, both sides move by roughly (1 − ρ)h, so the implied minus realized spread shifts much less than the levels.
Regarding P&L, it is a different story. it depends on the vega notional per name, and concentration means a correlation-neutral book needs more of the mega caps.
We have not published the exact top-50 reconstruction yet. it is a good follow-up.
Appreciate you running the top-50 weights—that’s a meaningful reconciliation. Moving the proxy estimate from 0.158 toward ~0.11 changes the level interpretation, but as you note, it doesn’t tell us the P&L of a dispersion book by itself. I’d separate three checks: (1) reconstruct COR3M with matched weights and tenors; (2) compare implied vs realized correlation on the same basket; (3) translate the gap into vega-weighted, hedged P&L after costs. That last step is where a statistical discrepancy becomes—or fails to become—an executable edge. I’d be keen to read your top-50 follow-up.
About the best explanation on dispersion trading I've read with real example, new sub here
A topic close to my heart, a topic where theory is easier than reality when people try to execute such a trade.
We hope that this article, as well as our future articles and the discussions under this post will be helpful to you
Been managing a cross asset correlation book for quite a few years. So dispersion is part of my trading DNA.
Same here
🔥