Pairwise return covariance as a construction gate

The two-asset covariance estimate

The archive estimates two-asset return covariance as the average of paired products of each asset's subperiod return minus its full-period mean return. Covariance is the raw measure of co-movement. The estimator uses a denominator of one less than the number of subperiods rather than the raw subperiod count.

A subperiod return is the observed total return of an asset in one sampling interval inside a longer estimation window. The pairwise covariance estimate is the two-asset building block of a larger covariance matrix used to compare how every held pair co-moves.

From covariance to a correlation coefficient

Correlation analysis scales that covariance by the product of the two assets' standard deviations so co-movement is bounded and comparable across pairs. The correlation coefficient equals the two-asset return covariance divided by the product of the two assets' standard deviations. Standard deviation is the dispersion of an asset's own returns and is used only to scale covariance into a correlation coefficient.

The correlation coefficient is a unit-free association measure. Its sign shows whether two return series move together or opposite, and its size shows how tightly they do so. The absolute value of the correlation coefficient ranges from 0, indicating no linear association, to 1, indicating perfect linear association. A correlation of 1 describes two return series that rise and fall together. A correlation of -1 describes two series that move in opposite directions.

Diversification before a new weight

Diversification is the construction practice of preferring pairs whose association is near zero when positive, or negative enough that the two return paths offset each other. When a pairwise correlation is positive, a value near 0 is the construction target used to treat the pair as a diversification input rather than a stacked exposure. A negative pairwise correlation can remain large in magnitude because the two return series offset each other inside the portfolio.