1% Generated by roxygen2: do not edit by hand
2% Please edit documentation in R/est_gmm.R
3\name{pgmm}
4\alias{pgmm}
5\alias{coef.pgmm}
6\alias{summary.pgmm}
7\alias{print.summary.pgmm}
8\title{Generalized Method of Moments (GMM) Estimation for Panel Data}
9\usage{
10pgmm(
11  formula,
12  data,
13  subset,
14  na.action,
15  effect = c("twoways", "individual"),
16  model = c("onestep", "twosteps"),
17  collapse = FALSE,
18  lost.ts = NULL,
19  transformation = c("d", "ld"),
20  fsm = NULL,
21  index = NULL,
22  ...
23)
24
25\method{coef}{pgmm}(object, ...)
26
27\method{summary}{pgmm}(object, robust = TRUE, time.dummies = FALSE, ...)
28
29\method{print}{summary.pgmm}(
30  x,
31  digits = max(3, getOption("digits") - 2),
32  width = getOption("width"),
33  ...
34)
35}
36\arguments{
37\item{formula}{a symbolic description for the model to be
38estimated. The preferred interface is now to indicate a
39multi--part formula, the first two parts describing the
40covariates and the GMM instruments and, if any, the third part
41the 'normal' instruments,}
42
43\item{data}{a \code{data.frame} (neither factors nor character vectors
44will be accepted in \code{data.frame}),}
45
46\item{subset}{see \code{\link[=lm]{lm()}},}
47
48\item{na.action}{see \code{\link[=lm]{lm()}},}
49
50\item{effect}{the effects introduced in the model, one of
51\code{"twoways"} (the default) or \code{"individual"},}
52
53\item{model}{one of \code{"onestep"} (the default) or \code{"twosteps"},}
54
55\item{collapse}{if \code{TRUE}, the GMM instruments are collapsed (default is
56\code{FALSE}),}
57
58\item{lost.ts}{the number of lost time series: if \code{NULL}, this is
59automatically computed. Otherwise, it can be defined by the
60user as a numeric vector of length 1 or 2. The first element is
61the number of lost time series in the model in difference, the
62second one in the model in level. If the second element is
63missing, it is set to the first one minus one,}
64
65\item{transformation}{the kind of transformation to apply to the
66model: either \code{"d"} (the default value) for the
67"difference GMM" model or \code{"ld"} for the "system GMM" model,}
68
69\item{fsm}{the matrix for the one step estimator: one of \code{"I"}
70(identity matrix) or \code{"G"} (\eqn{=D'D} where \eqn{D} is the
71first--difference operator) if \code{transformation="d"}, one of
72\code{"GI"} or \code{"full"} if \code{transformation="ld"},}
73
74\item{index}{the indexes,}
75
76\item{\dots}{further arguments.}
77
78\item{object, x}{an object of class \code{"pgmm"},}
79
80\item{robust}{for pgmm's summary method: if \code{TRUE} (default), robust inference
81is performed in the summary,}
82
83\item{time.dummies}{for pgmm's summary method: if \code{TRUE}, the estimated
84coefficients of time dummies are present in the table of coefficients;
85default is \code{FALSE}, thus time dummies are dropped in summary's coefficient
86table (argument is only meaningful if there are time dummies in the model,
87i.e., only for \code{effect = "twoways"}),}
88
89\item{digits}{digits,}
90
91\item{width}{the maximum length of the lines in the print output.}
92}
93\value{
94An object of class \code{c("pgmm","panelmodel")}, which has the
95following elements:
96
97\item{coefficients}{the vector (or the list for fixed effects) of
98coefficients,}
99\item{residuals}{the list of residuals for each individual,}
100\item{vcov}{the covariance matrix of the coefficients,}
101\item{fitted.values}{the vector of fitted values,}
102\item{df.residual}{degrees of freedom of the residuals,}
103\item{model}{a list containing the variables used for the
104estimation for each individual,}
105\item{W}{a list containing the instruments for each individual (a matrix per
106list element) (two lists in case of system GMM,}
107\item{A1}{the weighting matrix for the one--step estimator,}
108\item{A2}{the weighting matrix for the two--steps estimator,}
109\item{call}{the call.}
110
111In addition, it has attribute \code{"pdim"} which contains the pdim object for
112model.
113
114It has \code{print}, \code{summary} and \code{print.summary} methods.
115}
116\description{
117Generalized method of moments estimation for static or dynamic
118models with panel data.
119}
120\details{
121\code{pgmm} estimates a model for panel data with a generalized method
122of moments (GMM) estimator. The description of the model to
123estimate is provided with a multi--part formula which is (or which
124is coerced to) a \code{Formula} object. The first right--hand side part
125describes the covariates. The second one, which is mandatory,
126describes the GMM instruments. The third one, which is optional,
127describes the 'normal' instruments. By default, all the variables
128of the model which are not used as GMM instruments are used as
129normal instruments with the same lag structure as the one specified
130in the model.
131
132\code{y~lag(y, 1:2)+lag(x1, 0:1)+lag(x2, 0:2) | lag(y, 2:99)} is similar to
133
134\code{y~lag(y, 1:2)+lag(x1, 0:1)+lag(x2, 0:2) | lag(y, 2:99) | lag(x1, 0:1)+lag(x2, 0:2)}
135
136and indicates that all lags from 2 of \code{y} are used
137as GMM instruments.
138
139\code{transformation} indicates how the model should be transformed for
140the estimation. \code{"d"} gives the "difference GMM" model
141\insertCite{@see @AREL:BOND:91}{plm}, \code{"ld"} the "system GMM" model
142\insertCite{@see @BLUN:BOND:98}{plm}.
143
144\code{pgmm} is an attempt to adapt GMM estimators available within the
145DPD library for GAUSS \insertCite{@see @AREL:BOND:98}{plm} and Ox
146\insertCite{@see @DOOR:AREL:BOND:12}{plm} and within the xtabond2
147library for Stata \insertCite{@see @ROOD:09}{plm}.
148}
149\examples{
150
151data("EmplUK", package = "plm")
152
153## Arellano and Bond (1991), table 4 col. b
154z1 <- pgmm(log(emp) ~ lag(log(emp), 1:2) + lag(log(wage), 0:1)
155           + log(capital) + lag(log(output), 0:1) | lag(log(emp), 2:99),
156            data = EmplUK, effect = "twoways", model = "twosteps")
157summary(z1, robust = FALSE)
158
159## Blundell and Bond (1998) table 4 (cf. DPD for OX p. 12 col. 4)
160z2 <- pgmm(log(emp) ~ lag(log(emp), 1)+ lag(log(wage), 0:1) +
161           lag(log(capital), 0:1) | lag(log(emp), 2:99) +
162           lag(log(wage), 2:99) + lag(log(capital), 2:99),
163           data = EmplUK, effect = "twoways", model = "onestep",
164           transformation = "ld")
165summary(z2, robust = TRUE)
166
167\dontrun{
168## Same with the old formula or dynformula interface
169## Arellano and Bond (1991), table 4, col. b
170z1 <- pgmm(log(emp) ~ log(wage) + log(capital) + log(output),
171            lag.form = list(2,1,0,1), data = EmplUK,
172            effect = "twoways", model = "twosteps",
173            gmm.inst = ~log(emp), lag.gmm = list(c(2,99)))
174summary(z1, robust = FALSE)
175
176## Blundell and Bond (1998) table 4 (cf DPD for OX p. 12 col. 4)
177z2 <- pgmm(dynformula(log(emp) ~ log(wage) + log(capital), list(1,1,1)),
178            data = EmplUK, effect = "twoways", model = "onestep",
179            gmm.inst = ~log(emp) + log(wage) + log(capital),
180            lag.gmm = c(2,99), transformation = "ld")
181summary(z2, robust = TRUE)
182}
183
184}
185\references{
186\insertAllCited{}
187}
188\seealso{
189\code{\link[=sargan]{sargan()}} for the Hansen--Sargan test and \code{\link[=mtest]{mtest()}} for
190Arellano--Bond's test of serial correlation.  \code{\link[=dynformula]{dynformula()}} for
191dynamic formulas (deprecated).
192}
193\author{
194Yves Croissant
195}
196\keyword{regression}
197