gmmFit-methods {momentfit}

~~ Methods for Function gmmFit in Package momentfit ~~

Description

Method to fit a model using GMM, from an object of class "momentModel" or "sysModel".

Usage

## S4 method for signature 'momentModel'
gmmFit(model, type=c("twostep", "iter","cue",
              "onestep"), itertol=1e-7, initW=c("ident", "tsls"),
              weights="optimal", itermaxit=100,
              efficientWeights=FALSE, theta0=NULL, ...)

## S4 method for signature 'formulaModel'
gmmFit(model, type=c("twostep", "iter","cue",
              "onestep"), itertol=1e-7, initW=c("ident", "tsls"),
              weights="optimal", itermaxit=100,
              efficientWeights=FALSE, theta0=NULL, ...)

## S4 method for signature 'sysModel'
gmmFit(model, type=c("twostep", "iter","cue",
              "onestep"), itertol=1e-7, initW=c("ident", "tsls", "EbyE"),
              weights="optimal", itermaxit=100,
              efficientWeights=FALSE, theta0=NULL, EbyE=FALSE, ...)

## S4 method for signature 'rnonlinearModel'
gmmFit(model, type=c("twostep", "iter","cue",
              "onestep"), itertol=1e-7, initW=c("ident", "tsls"),
              weights="optimal", itermaxit=100,
              efficientWeights=FALSE, theta0=NULL, ...)

## S4 method for signature 'rlinearModel'
gmmFit(model, type=c("twostep", "iter","cue",
              "onestep"), itertol=1e-7, initW=c("ident", "tsls"),
              weights="optimal", itermaxit=100,
              efficientWeights=FALSE, ...)

## S4 method for signature 'rformulaModel'
gmmFit(model, type=c("twostep", "iter","cue",
              "onestep"), itertol=1e-7, initW=c("ident", "tsls"),
              weights="optimal", itermaxit=100,
              efficientWeights=FALSE, theta0=NULL, ...)

## S4 method for signature 'rslinearModel'
gmmFit(model, type=c("twostep", "iter","cue",
              "onestep"), itertol=1e-7, initW=c("ident", "tsls", "EbyE"),
              weights="optimal", itermaxit=100,
              efficientWeights=FALSE, theta0=NULL, EbyE=FALSE, ...)

Arguments

model

A model class object.

type

What GMM methods should we use? for type=="onestep", if "weights" is not a matrix, the model will be estimated with the weights equals to the identity matrix. For restricted

itertol

Tolance for the stopping rule in iterative GMM

initW

How should be compute the initial coefficient vector in the first. For single equation GMM, it only makes a difference for linear models for which the choice is GMM with identity matrix or two-stage least quares. For system of equations, "tsls", refers to equation by equation two-stage least squares. It is also possible to start at the equation by equation estimate using the same GMM type as specified by "type".

weights

What weighting matrix to use? The choices are "optimal", in which case it is the inverse of the moment vovariance matrix, "ident" for the identity matrix, or a fixed matrix. It is also possible for weights to be an object of class gmmWeights.

itermaxit

Maximum iterations for iterative GMM

EbyE

Should the system be estimated equation by equation?

efficientWeights

If weights is a matrix or a gmmWeights class object, setting efficientWeights to TRUE implies that the resulting one-step GMM is efficient. As a result, the default covariance matrix for the coefficient estimates will not be a sandwich type.

theta0

An optional initial vector for optim when the model is nonlinear. By default, the theta0 argument of the model is used

...

Arguments to pass to other methods (mostly the optimization algorithm)

Methods

signature(model = "momentModel")

The main method for all moment-based models.

signature(model = "rnonlinearModel")

It makes a difference only if the number of contraints is equal to the number of coefficients, in which case, the method evalGmm is called at the contrained vector. If not, the next method is called.

signature(model = "rformulaModel")

It makes a difference only if the number of contraints is equal to the number of coefficients, in which case, the method evalGmm is called at the contrained vector. If not, the next method is called.

signature(model = "rlinearModel")

It makes a difference only if the number of contraints is equal to the number of coefficients, in which case, the method evalGmm is called at the contrained vector. If not, the next method is called.

signature(model = "sysModel")

Method to estimate system of equations using GMM methods.

Examples

data(simData)

theta <- c(beta0=1,beta1=2)
model1 <- momentModel(y~x1, ~z1+z2, data=simData)

## Efficient GMM with HAC vcov and tsls as first step.
res1 <- gmmFit(model1, init="tsls")

## GMM with identity. Two ways.
res2 <- gmmFit(model1, type="onestep")
res3 <- gmmFit(model1, weights=diag(3))

## nonlinear regression with iterative GMM.
g <- y~beta0+x1^beta1
h <- ~z1+z2
model2 <- momentModel(g, h, c(beta0=1, beta1=2), data=simData)
res4 <- gmmFit(model2, type="iter")

## GMM for with no endogenous vaiables is
## OLS with Robust standard error

library(lmtest)
model3 <- momentModel(y~x1, ~x1, data=simData, vcov="MDS")
resGmm <- gmmFit(model3)
resLm <- lm(y~x1, simData)
summary(resGmm)
coeftest(resLm, vcov=vcovHC(resLm, "HC0"))
summary(resGmm, df.adj=TRUE)
coeftest(resLm, vcov=vcovHC(resLm, "HC1"))

### All constrained
R <- diag(2)
q <- c(1,2)
rmodel1 <- restModel(model1, R, q)
gmmFit(rmodel1)

## Only one constraint
R <- matrix(c(0,1), ncol=2)
q <- 2
rmodel1 <- restModel(model1, R, q)
gmmFit(rmodel1)