Cleaning Covariance Matrices with Temporal Dependence

Serial dependence changes the sampling noise in high-dimensional covariance matrices, potentially invalidating cleaning rules designed for independent observations. Under a Gaussian separable covariance model, we characterize the temporal adjustments needed for four classical procedures and compare them with the correlated-sample rotationally invariant estimator of Burda and Jarosz. Basic linear shrinkage requires a single temporal moment; advanced shrinkage requires joint sampling moments involving its estimated target; clipping requires a dependence-adjusted spectral edge; and population-eigenvalue substitution, like rotationally invariant cleaning, requires the temporal spectral law. Consequently, no single effective observation ratio corrects all of them. Simulations distinguish oracle from feasible temporal adjustment and covariance loss from global minimum-variance (GMV) portfolio risk. Oracle corrections can materially lower covariance loss without comparable GMV-risk gains. A pooled Bartlett-tapered Toeplitz estimator, supported by an operator-norm concentration result for Gaussian panels with heterogeneous row variances, preserves much of the oracle covariance-loss improvement for linear shrinkage and rotationally invariant cleaning; clipping is more sensitive to spectral-edge estimation and tapering bias. In rolling daily US, Japanese, and European portfolio panels with training windows of twice the number of assets, cleaning substantially reduces out-of-sample GMV risk relative to the sample covariance, while the incremental effect of feasible temporal adjustment is below one percentage point.

Authors

Institutions

Publication Details

Journal
Mathematics
Published
2026-11-18
DOI
https://doi.org/10.3390/math14224249
Primary Topic
Financial Risk and Volatility Modeling
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Cleaning Covariance Matrices with Temporal Dependence

Chihwa D. Kao, Jiaqi Wang
1 links
Mathematics
Financial Risk and Volatility Modeling
article

Cleaning Covariance Matrices with Temporal Dependence

Chihwa D. Kao, Jiaqi Wang
article en

Abstract

Serial dependence changes the sampling noise in high-dimensional covariance matrices, potentially invalidating cleaning rules designed for independent observations. Under a Gaussian separable covariance model, we characterize the temporal adjustments needed for four classical procedures and compare them with the correlated-sample rotationally invariant estimator of Burda and Jarosz. Basic linear shrinkage requires a single temporal moment; advanced shrinkage requires joint sampling moments involving its estimated target; clipping requires a dependence-adjusted spectral edge; and population-eigenvalue substitution, like rotationally invariant cleaning, requires the temporal spectral law. Consequently, no single effective observation ratio corrects all of them. Simulations distinguish oracle from feasible temporal adjustment and covariance loss from global minimum-variance (GMV) portfolio risk. Oracle corrections can materially lower covariance loss without comparable GMV-risk gains. A pooled Bartlett-tapered Toeplitz estimator, supported by an operator-norm concentration result for Gaussian panels with heterogeneous row variances, preserves much of the oracle covariance-loss improvement for linear shrinkage and rotationally invariant cleaning; clipping is more sensitive to spectral-edge estimation and tapering bias. In rolling daily US, Japanese, and European portfolio panels with training windows of twice the number of assets, cleaning substantially reduces out-of-sample GMV risk relative to the sample covariance, while the incremental effect of feasible temporal adjustment is below one percentage point.

MathematicsVol. 14(22)
University of Connecticut (US)
Financial Risk and Volatility Modeling
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon