Hi to denoise the correlation matrix you can use the marcenko pastur distribution. Even without getting into its detail,. its easy, you just use t/n to get the lambda value under which you will discard eigenvalues. But how do you adapt T/N when the matrix is an expotentially weighted one?

this paper seems to have it, but its math is above my specialty: https://www.researchgate.net/publication/222925495_The_asymptotic_spectrum_of_the_EWMA_covariance_estimator

more on Marcenko https://en.wikipedia.org/wiki/Marchenko%E2%80%93Pastur_distribution


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.