near_posdef: Nearest Positive-Definite Matrix

In antshi/CovEstim: (High-dimensional) Covariance Estimation

Description

Implements the algorithm of \insertCitehigham2002computing;textualCovEstim to compute the nearest positive-definite matrix to an approximate one, typically a correlation or variance-covariance matrix.

Usage

near_posdef(
  mat,
  corr = FALSE,
  keep_diag = FALSE,
  do_dykstra = TRUE,
  eig_tol = 1e-06,
  conv_tol = 1e-07,
  posd_tol = 1e-08,
  maxit = 100,
  trace = FALSE
)

Arguments

mat	a pxp matrix.
corr	a logical, indicating if the result should be a correlation matrix. Default value is FALSE.
keep_diag	a logical, indicating if the result should have the same diagonal as the original matrix mat.
do_dykstra	a logical, indicating if Dykstra's correlation is to be used. Default value is TRUE.
eig_tol	a double, defining the relative positiveness of eigenvalues compared to the largest eigenvalue. Default value is 1e-6.
conv_tol	a double, defining the convergence tolerance for the Higham algorithm. Default value is 1e-7.
posd_tol	a double, defining the tolerance for enforcing positive definiteness. Default value is 1e-8.
maxit	maximum number of iterations. Default value is 100.
trace	a logical, indicating whether iterations are to be traced (printed out). Default value is FALSE.

Details

Note that setting corr = TRUE just sets diag(.) <- 1 within the algorithm. The near_posdef() is originally found under \insertCitematrixpackage;textualCovEstim.

Value

a positive-definite matrix.

References

\insertAllCited

Examples

data(sp200)
sp_rets <- sp200[1:100,-1]
ml_sigma <- sigma_estim(sp_rets, "ML")
ml_sigma_near_posdef <- near_posdef(ml_sigma, eig_tol=1e-8)

antshi/CovEstim documentation built on Nov. 13, 2020, 2:25 p.m.