mm {regtools}

Method of Moments, Including Possible Regression Terms

Description

Method of Moments computation for almost any statistical problem that has derivatives with respect to theta. Capable of handling models that include parametric regression terms, but not need be a regression problem. (This is not Generalized Method of Moments; see the package gmm for the latter.)

Usage

mm(m,g,x,init=rep(0.5,length(m)),eps=0.0001,maxiters=1000) 

Arguments

m

Vector of sample moments, "left-hand sides" of moment equations.

g

Function of parameter estimates, forming the "right-hand sides." This is a multivariate-valued function, of dimensionality equal to that of m

.

init

Vector of initial guesses for parameter estimates. If components are named, these will be used as labels in the output.

eps

Convergence criterion.

maxiters

Maximum number of iterations.

x

Input data.

Details

Standard Newton-Raphson methods are used to solve for the parameter estimates, with numericDeriv being used to find the approximate derivatives.

Value

R list consisting of components tht, the vector of parameter estimates, and numiters, the number of iterations performed.

Author(s)

Norm Matloff

Examples

x <- rgamma(1000,2)
m <- c(mean(x),var(x))
g <- function(x,theta) {  # from theoretical properties of gamma distr.
   g1 <-  theta[1] / theta[2]
   g2 <-  theta[1] / theta[2]^2
   c(g1,g2)
}
# should output about 2 and 1
mm(m,g,x)

## Not run: 
library(mfp)
data(bodyfat)
# model as a beta distribution 
g <- function(x,theta) {
   t1 <- theta[1]
   t2 <- theta[2]
   t12 <- t1 + t2
   meanb <- t1 / t12
   m1 <- meanb 
   m2 <- t1*t2 / (t12^2 * (t12+1)) 
   c(m1,m2)
}
x <- bodyfat$brozek/100
m <- c(mean(x),var(x))
# about 4.65 and 19.89
mm(m,g,x)

## End(Not run)