Dear R-users,
I want to reproduce "a dynamic panel data model, autoregressive of order 1,
with fixed effects" that was originally estimated with 2-step GMM in STATA.
Reproducing the model in R shouldn't be a big deal, however, unfortunately I
don't have the original STATA code and the lagged dependent variable is not
"free-standing", but rather subtracted from another independent
variable like this:
Y(t) - Y(t-1) = b1 * [X(t) - X(t-1)] + b2 * [X(t-1)-Y(t-1)]
So my question is on how to specify the instruments. The model I am trying to
reproduce is instrumenting the endogenous lagged term by its 4th and the 5th
lag. Do I need dynformula() for that or does one of the following specifications
work as well?
form1 <- diff(value.y) ~ diff(value.x) + lag(I(value.x-value.y),1) |
lag(value.y,4:5)
form2 <- diff(value.y) ~ diff(value.x) + lag(I(value.x-value.y),4:5) |
lag(value.y,4:5)
form3 <- diff(value.y) ~ diff(value.x) + diff.xy | lag(diff.xy,4:5)
form4 <- diff(value.y) ~ diff(value.x) + lag(diff.xy,1) | lag(diff.xy,4:5)
form5 <- diff(value.y) ~ diff(value.x) + lag(diff.xy,1) |
lag(I(value.x-value.y),4:5)
library(plm)
value.x <- rnorm(336)
value.y <- rnorm(336)
diff.xy <- value.x - value.y
my.data <-
data.frame(expand.grid(ids=1:28,times=1:12),value.x,value.y,diff.xy)
mod1 <-
pgmm(formula=form1,data=my.data,index=c('ids','times'),
effect="individual", model="twosteps")
summary(mod1)
Kind regards,
Erich
[[alternative HTML version deleted]]