GMEselection: A variable selection procedure for regression models based on...

In jpedroan/GMEselection: Variable Selection with Generalized Maximum Entropy

Description

A variable selection procedure for regression models based on Generalized Maximum Entropy estimation (A. Golan, J. Judge and D. Miller, Maximum Entropy Econometrics, Wiley, 1996; Chapter 10). The procedure is useful for well- and ill-posed regression models, namely in models exhibiting small sample sizes, collinearity and non-normal errors.

A variable selection procedure for regression models based on Generalized Maximum Entropy estimation (A. Golan, J. Judge and D. Miller, Maximum Entropy Econometrics, Wiley, 1996; Chapter 10). The procedure is useful for well- and ill-posed regression models, namely in models exhibiting small sample sizes, collinearity and non-normal errors.

Usage

GMEselection(y, x, ssi = TRUE, si = c(-1, 1), intervals = NULL, m = 5,
  es = TRUE, error.csi = c(-1, 1), j = 3)

Arguments

y	– vector – of size k.
x	– matrix of size (n,k) – n samples of size k; The supports for all unknown parameters should be symmetric and') uniformly distributed around zero.
ssi	(resp) – TRUE/FALSE – use the same support interval for all the unknown.
si	(int1) – c(float,float) – support interval, in the form c(-|lower|,upper), common for all unknown parameters.
intervals	– list of lists – ex. list(c(-l1,s1), c(-l2,s2),...), list of pairs (.,.) constituting a different support for each parameter. Note that, in this model, you must specify the same number of pairs as size of vector y.
m	– integer – the number of points in each parameter support. Usually the estimation is performed with five points in the parameter supports. Naturally, you can define a higher value.
es	(resp1) – TRUE/FALSE – error supports using an estimate of the error standard deviation from the OLS residuals.
error.csi	(int2) – c(-e,e) – the support interval, [-e,e], for the error component.
j	– integer – number of points in each error support. Usually the estimation is performed with three points in the error supports. Naturally, you can define a higher value.

Value

b – vector – estimate of the unknown parameters; nepk – float – normalized entropy for the intercept (if it exists) and for each variable; nep – vector – normalized entropy for the signal.

jpedroan/GMEselection documentation built on May 19, 2019, 10:44 p.m.