Combining Univariate Variances with Multivariate Correlations

Summary

The document raises a portfolio statistics question about constructing a covariance matrix in two stages. The proposed approach estimates each asset's variance from its own data, estimates pairwise correlations separately, and combines the two sets of estimates into covariances. The author asks whether estimating variances jointly in a multivariate model offers advantages and whether relevant literature compares the approaches.

No answer, method comparison, references, or empirical evidence is included, so the document does not establish that either approach is preferable. It identifies a practical modeling choice: separate estimation may be convenient, while a joint approach could impose shared structure across assets. Any comparison would depend on the data, model assumptions, and intended use of the covariance matrix, such as portfolio construction. Those considerations are not developed here, and no specific estimator or asset class is identified.

Key ideas

• The document considers estimating asset variances and correlations separately before combining them into covariances.
• It asks whether joint multivariate variance estimation has practical or statistical benefits.
• No answer, literature citations, or evidence is provided to resolve the comparison.
• The appropriate estimation choice depends on model assumptions and the intended use of the covariance matrix.

Tags

# Disjoint covariance matrix estimation


# Disjoint covariance matrix estimation












I have always estimated correlations and variances disjointly and later combine them to construct covariance matrices. Specifically, variances are estimated in a univariate setting (only using the data for each particular asset).

Is there any downside in this approach? Wondering if there is a benefit to estimate variances in a multivariate setting / or if there is any literature on this?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.