Blending Correlation Models Estimated at Different Frequencies

Summary

The document considers whether it is reasonable to average correlation matrices estimated from the same underlying prices but sampled at different return frequencies. Its answer indicates that this practice has been observed in risk modeling, including blending models built for shorter and longer horizons. It points to risk-model vendors that offer such blending as evidence that the approach is used in industry.

The response is deliberately limited: it does not claim that frequency-blended matrices are universally standard or appropriate, and it gives no empirical comparison, weighting method, or conditions for combining estimates. The example establishes precedent, not that a blend improves forecasts or portfolio decisions. Practitioners would still need to check that the matrices are on compatible scales and represent horizons relevant to their risk use; those implementation details are not addressed in the document.

Key ideas

• Correlation matrices estimated at different return frequencies can be blended in some risk-modeling settings.
• The response cites vendor support for combining shorter- and longer-horizon risk models as evidence of use.
• The document does not establish how common the practice is or whether it improves risk estimates.
• It provides no blending formula, validation results, or guidance on choosing weights.

Tags

# Averaging Correlation Matrices based on different Periodicity


# Averaging Correlation Matrices based on different Periodicity












Averaging correlation matrices based on different models, but the same data, is commonly done. If the correlation matrices are derived from return series, is it proper/common to also average the correlation matrices based on return series of the same underlying price series, but different periodicity?

## Answer by J-F (score 2, accepted)

https://quant.stackexchange.com/a/45513

I don't know how common this is, but I've seen it done. Many risk model vendors (Northfield, Axioma) allow the blending of different risk models with different periodicity (e.g. a shorter horizon risk model blended with a longer horizon risk model). Here's a Northfield deck about this:

https://www.northinfo.com/documents/779.pdf

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.