QuarcHome
Model catalogueBack to the model index

Higham nearest correlation matrix

Finds a nearest positive semidefinite correlation matrix under a stated distance.

How it works

Higham formulates an explicit distance from the input and projects onto the set of positive semidefinite, unit-diagonal matrices. Corrected alternating projections can solve the equal-weight Frobenius problem. The minimum-distance objective supplies a reproducible criterion for trading off movements across entries.

Worked example

For the three-asset input in our report, the nearest solution under equal weights has off-diagonals approximately 0.826425, 0.623776 and 0.075424. Its Frobenius distance is about 0.256087. The spectral repair is slightly farther away, while its portfolio risk for the particular hedge is smaller.

Validation and limitations

Report the norm, weights, convergence criteria and minimum eigenvalue. Fixing entries or imposing a strictly positive eigenvalue floor changes the feasible set and can require a different solver. See the side-by-side comparison and report.

Research sources