QuarcHome
Model catalogueBack to the model index

Ledoit–Wolf covariance shrinkage

Regularises a large sample covariance matrix by shrinking it toward a structured target.

How it works

Shrinkage blends a sample covariance matrix with a more structured target. The Ledoit–Wolf estimator chooses a shrinkage intensity intended to control estimation error. The blend often gives more stable matrices when the number of assets is large relative to the return history, which can help optimisation or inversion.

Worked example

With a sample covariance entry of 0.012 and a diagonal target whose corresponding entry is zero, 25% shrinkage gives 0.75 × 0.012 = 0.009. The diagonal entries are blended too, depending on the choice of target. The coefficient acts on a covariance matrix rather than separately on each correlation.

Validation and limitations

Compare realised portfolio risk and turnover out of sample across target choices. Check data alignment and conversion between correlations and covariance. Shrinkage trades sample fit for stability. It can reduce estimates of genuine dependence, especially in stressed markets.

Research sources