Shrinkage Tuning Parameter Selection in Precision Matrices Estimation

Recent literature provides many computational and modeling approaches for covariance matrices estimation in a penalized Gaussian graphical models but relatively little study has been carried out on the choice of the tuning parameter. This paper tries to fill this gap by focusing on the problem of shrinkage parameter selection when estimating sparse precision matrices using the penalized likelihood approach. Previous approaches typically used K-fold cross-validation in this regard. In this paper, we first derived the generalized approximate cross-validation for tuning parameter selection which is not only a more computationally efficient alternative, but also achieves smaller error rate for model fitting compared to leave-one-out cross-validation. For consistency in the selection of nonzero entries in the precision matrix, we employ a Bayesian information criterion which provably can identify the nonzero conditional correlations in the Gaussian model. Our simulations demonstrate the general superiority of the two proposed selectors in comparison with leave-one-out cross-validation, ten-fold cross-validation and Akaike information criterion.

💡 Research Summary

The paper addresses a critical yet under‑explored aspect of penalized Gaussian graphical models: the selection of the shrinkage (tuning) parameter λ when estimating sparse precision matrices. While many recent works have focused on computational algorithms and penalty functions (Lasso, adaptive Lasso, SCAD), they typically rely on K‑fold cross‑validation (K‑CV) or leave‑one‑out cross‑validation (LOOCV) to choose λ. These approaches become computationally prohibitive in high‑dimensional settings and may not guarantee consistent model selection.

The authors propose two alternative selectors. First, they derive a Generalized Approximate Cross‑Validation (GACV) criterion tailored to the multivariate precision‑matrix likelihood. By approximating the LOOCV log‑likelihood using influence‑matrix techniques, GACV can be computed from a single fit to the full data, avoiding the repeated optimizations required by LOOCV or K‑CV. The derivation, detailed in Appendix A, extends the classic GCV/GACV ideas from smoothing splines and univariate regression to the matrix‑valued setting.

Second, they adopt a Bayesian Information Criterion (BIC) specifically designed for penalized likelihood estimators. The BIC score is
\