robustMultExpSmoothing

Optimal smoothing matrix. If missing it is estimated.
The procedure maybe very slow for high-dimensional data. Also,
the objective function being very noisy, optimization across
multiple runs may lead to different smoothing matrices. #'

smoothMat

length of samples required to calculate initial values

startup_period

length of samples required to calculate forecast errors
for evalualating the objective if smoothing matrix is estimated

training_period

random seed to replicate the starting values for optimization

seed

number of strarting values to try for any optimization.
Large number of trials for high dimensions can be time consuming

trials

optimization method to use to evaluate an estimate of
smoothing matrix. Default is L-BFGS-B

method

known constant as described in the paper. Defaults to 0.2

lambda


Robust Multivariate Exponential Smoothing


We implement a collection of techniques for estimating covariance matrices.
Covariance matrices can be built using missing data. Stambaugh Estimation and
FMMC methods can be used to construct such matrices. Covariance matrices can
be built by denoising or shrinking the eigenvalues of a sample covariance
matrix. Such techniques work by exploiting the tools in Random Matrix Theory
to analyse the distribution of eigenvalues. Covariance matrices can also
be built assuming that data has many underlying regimes. Each regime is
allowed to follow a Dynamic Conditional Correlation model. Robust covariance
matrices can be constructed by multivariate cleaning and smoothing of noisy data.

robustMultExpSmoothing: Robust Multivariate Exponential Smoothing

Description

Usage

Arguments

Details