Python implementation of Ledoit-Wolf covariance shrinkage towards a constant correlation target (Ledoit & Wolf, JPM 2004).
Most LW shrinkage implementations (e.g. sklearn's LedoitWolf) target a scaled identity. This one implements the constant-correlation target from the original paper, which preserves pairwise correlation structure while regularizing variance. The rho-factor is the key differentiator. Useful when you care about the off-diagonal structure of your covariance estimate in high-dimensional, low-sample-count settings.
Features:
- Shrinks sample covariance matrix towards constant correlation, not the more common constant variance target
- Returns both the shrunk covariance matrix sigma and the shrinkage intensity scalar
- Direct Matlab-to-Python translation of the authors' original covcor.m