Description
A package for estimating population eigenvalues and covariance
matrices, based on publications by Ledoit and Wolf (2004, 2012, 2015,
2016).
Details
A common assumption in statistics is that for a data matrix $X$
of dimension $n \times p$, the number of predictor variables (p)
vanishes relative to the number of datapoints (n) as $n \to \infty$.
However, in modern datasets, it is often the case that p is comparable to
or greater than n. In this scenario, a more appropriate asymptotic
framework is to assume that the ratio $c := p/n$ approaches a finite
positive value as $n,p \to \infty$. In this case, the sample covariance
matrix $S$ is no longer a consistent estimator of the population
covariance matrix $\Sigma$. Similarly, the sample eigenvalues deviate
substantially from the population eigenvalues. This package contains
implementations of Ledoit and Wolf's linear and non-linear shrinkage
population eigenvalue and covariance estimation methods, based on their
2016 publication and the accompanying MATLAB code. Theoretical and
implementation details of these methods can be found in the following
publications:
- Ledoit, O. and Wolf, M. (2004). A well-conditioned
estimator for large-dimensional covariance matrices. Journal of
Multivariate Analysis, 88(2)
- Ledoit, O. and Wolf, M. (2012).
Nonlinear shrinkage estimation of large-dimensional covariance matrices.
Annals of Statistics, 40(2).
- Ledoit, O. and Wolf, M. (2015). Spectrum
estimation: a unified framework for covariance matrix estimation and PCA in
large dimensions. Journal of Multivariate Analysis, 139(2).
- Ledoit,
O. and Wolf, M. (2016). Numerical Implementation of the QuEST function.
arXiv:1601.05870 [stat.CO].