The correlation matrix shows how your strategies move together — the basis of real diversification. It is computed as the Pearson correlation of daily P/L across the visible strategies.
On a large book, read it by group
The matrix grows quadratically: 15 strategies mean 105 pairs to scan, which is more noise than signal. Four sleeves mean six — and they are the pairs you actually care about ("are my Iron Condors correlated with my Verticals?"). A group's series is the real P/L of its visible members, with weights, the group's size knob, dynamic sizing and weekday filters already applied, so what you correlate is what the portfolio actually traded. Because correlation is scale-invariant, turning a group's size knob changes its exposure but not its correlations.
Days where a strategy has no trade are treated as zero P/L so series align over the chosen window.
Correlation is measured on common trading days
Those filled zeros are used to align equity, drawdowns and the co-movement tallies — but the correlation itself is computed only on the days both strategies actually traded (pairwise complete). Otherwise two strategies that rarely trade on the same days would show a spurious correlation (usually strongly negative — when one is in the market the other sits at a filled zero). When a pair shares fewer than ~10 common trading days (≈4 on the weekly basis) there aren't enough points to measure, so the cell reads undefined (“—”) rather than a misleading number, and the pair panel reports how many common days it found. A strategy's correlation with itself is always 1, which carries no information: the diagonal is left blank so only the pairs worth reading are shown.
The tabs above the matrix pick which composition you are measuring: Current, or any variant. Both are computed the same way and with the same parameters — period, weekly resampling and the by-strategy / by-group selector all apply to the tab you are on — so the two matrices are comparable cell by cell. Edit a variant from the strategies rail and its matrix follows the change, including a strategy renamed or moved into a sleeve.
The pair detail below is available on Current only: it reads the live session's series.
Correlation runs from −1 to +1:
| Range | Meaning |
|---|---|
| ≥ +0.4 | strategies move together → concentration risk |
| +0.15 to +0.4 | partial overlap |
| −0.15 to +0.15 | largely independent |
| −0.4 to −0.15 | offset each other → diversifying |
| ≤ −0.4 | strong hedge |
A diversification score summarises each strategy's average absolute correlation to the others — low means it genuinely diversifies the book; high means it largely duplicates exposure you already have.
Diversification is a portfolio decision
Two strong strategies that are highly correlated add risk without much smoothing. A modest strategy with low or negative correlation can improve the portfolio's drawdown profile more than another copy of your best performer.
Click any off-diagonal cell to open the detail panel for that pair. Everything in it is computed on the same window, weekly setting, and weekday filters as the matrix, so the headline ρ matches the cell exactly.
Equal-volatility weighting
The combined metrics (risk reduction, combined Sharpe) normalise each leg to the same volatility before blending, so they describe the intrinsic relationship of the pair — independent of the weights you give the strategies in the rest of the book.
Two different day sets, on purpose
Strategies rarely trade the same calendar, so the pair is aligned on the union of their dates with
missing days filled as 0. Which numbers use that padded series is not a detail:
Var(a + b) is about the real contributions to a book, and a
day when only one leg is on the market is a genuine day of risk.When the two legs share fewer than 10 trading days, the pair figures are hidden rather than shown: below that threshold they are driven by an overlap that barely exists, and ρ is already suppressed for the same reason.