Min Correlation

Also seen as: minimum-correlation algorithm, MCA, minimum correlation weighting

Why does low correlation matter?

A portfolio's volatility depends on how its holdings move relative to each other, not only on how volatile each one is. Two instruments with identical individual volatility give a less volatile combination the less correlated they are — that is the whole mechanism behind diversification, and it is the correlation term in the mean-variance volatility formula.

Concretely, for two instruments held at equal weight and with the same volatility σ, the combination's volatility is σ × √((1 + ρ) ÷ 2), where ρ is their correlation. At ρ = 1 the pair is no calmer than one instrument alone; the lower ρ falls, the more the pair's volatility drops below σ.

Which settings does Min Correlation have?

None — the method's card carries no settings, because the weights fall out of the correlations themselves. The only input the method reads is the calculation window — the field labelled In-sample in the strategy editor — and that window is where the correlations are measured.

Where can you use Min Correlation?

Min Correlation is a single-strategy method only. It is one of the eleven methods offered inside a strategy and is not one of the six offered inside a Combined, so you cannot use it to split capital across the strategies a Combined contains.

Defaults in Fincanva

• Min Correlation exposes no settings of its own — selecting the method is the whole configuration.
• The calculation window (In-sample) defaults to 12 months and accepts any whole number of months from 1 upward. Correlations are measured over that window.
• Correlations are re-measured and weights recomputed at every rebalance, so a min-correlation weighting shifts over the life of a backtest — and it can shift sharply, because correlations between the same instruments change over time — a rolling correlation is how that movement is read.
• The method needs at least two instruments to have anything to correlate: with a single instrument there are no relationships to minimise.

Worked example

A strategy holds three instruments, each with 15% volatility. Their pairwise correlations over the window are: A with B, 0.85; A with C, 0.20; B with C, 0.35.

Held at equal weight, a pair's volatility is 15% × √((1 + ρ) ÷ 2):

Every one of these pairs is built from instruments with exactly the same 15% volatility, yet the calmest pair is nearly three percentage points calmer than the most correlated one. Nothing about the individual instruments explains that difference — only the correlations do, and picking on that difference is what Min Correlation is for. Fincanva reports the same pair-by-pair reading for the strategies a Combined holds in its correlation matrix.

How is Min Correlation different from Risk Parity?

Both methods take correlations into account, but they aim at different things. Risk Parity targets the split of risk — every instrument should supply an equal share of the portfolio's total risk. Min Correlation targets the co-movement itself, favouring the instruments that behave least like the rest of the strategy. A highly correlated instrument can still receive a substantial risk-parity weight; under Min Correlation it is the very thing being weighted down.

Fincanva describes how these methods work; it does not recommend one. See Is this financial advice?.