Merge branch 'main' of https://github.com/ycroissant/plm

# Conflicts:
#	R/est_gmm.R

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: plm
 Version: 2.6-9999
-Date: 2024-04-04
+Date: 2025-01-02
 Title: Linear Models for Panel Data
 Authors@R: c(person(given = "Yves",      family = "Croissant",    role = c("aut"), email = "[email protected]"),
              person(given = "Giovanni",  family = "Millo",        role = c("aut"), email = "[email protected]"),

NEWS.md

@@ -7,12 +7,15 @@ subtitle: plm - Linear Models for Panel Data - A set of estimators and tests for
 # 2.6-9999 development version, changes since 2.6-4
 
 ### Fixes:
+* `pgmm`:
+  * for one-step GMM models: usual (non-robust) covariance matrix/standard errors fixed.
+  * argument `fsm` can now be set to influence the first step's weighing matrix.
 * `vcovXX`: FD models with only one observation per group prior to 
-           first-differencing errored ([#58](https://github.com/ycroissant/plm/issues/58)).
-* `pggls`: FD models errored with the data constellation as described for `vcovXX`. 
+    first-differencing errored ([#58](https://github.com/ycroissant/plm/issues/58)).
+* `pggls`: FD models errored with the data constellation as described above for `vcovXX`.
 * `is.pdata.frame` (non-exported helper function): fix part of the check if object 
-                                                   does not have an index.
-* Vignette A: fixed description of `pgmm`'s `effect` argument.
+    does not have an index.
+* Vignette A: improved description of `pgmm`'s `effect` argument.
 
 
 ***

R/tool_argvalues.R

@@ -81,3 +81,9 @@ oneof <- function(x){
 
 plm.arg <- c("formula", "data", "subset", "weights", "na.action", "effect", "model",
              "instruments", "random.method", "inst.method", "index")
+
+pgmm.fsm.list <- c(I    = "I",
+                   G    = "G",
+                   GI   = "GI",
+                   full = "full")
+

R/tool_vcovG.R

@@ -1225,6 +1225,7 @@ vcovHC.pgmm <- function(x, ...) {
   transformation <- describe(x, "transformation")
   A1 <- x$A1
   A2 <- x$A2
+  B1 <- x$B1
 
   if(transformation == "ld") {
 ##     yX <- lapply(x$model,function(x) rbind(diff(x),x))
@@ -1237,51 +1238,44 @@ vcovHC.pgmm <- function(x, ...) {
     yX <- x$model
     residuals <- x$residuals
   }
+
   minevA2 <- min(abs(Re(eigen(A2)$values)))
   eps <- 1E-9
 
   SA2 <- if(minevA2 < eps){
     warning("a general inverse is used")
     ginv(A2)
   } else solve(A2)
-
+
+  WX <- Reduce("+", mapply(function(w, y) crossprod(w, y[ , -1L, drop = FALSE]), x$W, yX, SIMPLIFY = FALSE))
+
+  # robust vcov for one-step GMM, see Roodman (2009), formula (15)
+  vcovr1s <- B1 %*% (t(WX) %*% A1 %*% SA2 %*% A1 %*% WX) %*% B1
+
   if(model == "twosteps") {
     coef1s <- x$coefficients[[1L]]
     res1s <- lapply(yX, function(x) x[ , 1L] - crossprod(t(x[ , -1L, drop = FALSE]), coef1s))
     K <- ncol(yX[[1L]])
     D <- c()
-    WX <- Reduce("+",
-                 mapply(function(x, y) crossprod(x, y[ , -1L, drop = FALSE]), x$W, yX, SIMPLIFY = FALSE))
     We <- Reduce("+", mapply(function(x, y) crossprod(x, y), x$W, residuals, SIMPLIFY = FALSE))
-    B1 <- solve(t(WX) %*% A1 %*% WX)
-    vcov1s <- B1 %*% (t(WX) %*% A1 %*% SA2 %*% A1 %*% WX) %*% B1 # robust vcov for one-step model
     for (k in 2:K) {
-      exk <- mapply(
-                    function(x, y){
+      exk <- mapply(function(x, y){
                       z <- crossprod(t(x[ , k, drop = FALSE]), t(y))
                       return(- z - t(z))
                     },
                     yX, res1s, SIMPLIFY = FALSE)
-      wexkw <- Reduce("+",
-                      mapply(
-                             function(x, y)
-                             crossprod(x, crossprod(y, x)),
-                             x$W, exk, SIMPLIFY = FALSE))
+
+      wexkw <- Reduce("+", mapply(function(x, y) 
+                                    crossprod(x, crossprod(y, x)),
+                                  x$W, exk, SIMPLIFY = FALSE))
       B2 <- vcov(x)
       Dk <- -B2 %*% t(WX) %*% A2 %*% wexkw %*% A2 %*% We
       D <- cbind(D, Dk)
     }
-    vcovr <- B2 + crossprod(t(D), B2) + t(crossprod(t(D), B2)) + D %*% vcov1s %*% t(D) # robust vcov for twosteps model, Roodman (2019) form. (18)
-  }
-  else {
-    # model = "onestep"
-    res1s <- lapply(yX, function(z) z[ , 1L] - crossprod(t(z[ , -1L, drop = FALSE]), x$coefficients))
-    K <- ncol(yX[[1L]])
-    WX <- Reduce("+", mapply(function(z, y) crossprod(z[ , -1L, drop = FALSE], y), yX, x$W, SIMPLIFY = FALSE))
-    B1 <- vcov(x)
-    vcovr <- B1 %*% (WX %*% A1 %*% SA2 %*% A1 %*% t(WX)) %*% B1 # robust vcov onestep model, Roodman (2019) form. (15)
+    # robust vcov for twosteps GMM model, Roodman (2019) form. (18)
+    vcovr2s <- B2 + crossprod(t(D), B2) + t(crossprod(t(D), B2)) + D %*% vcovr1s %*% t(D)
   }
-  vcovr
+  if(model == "twosteps") vcovr2s else vcovr1s
 }
 
 

inst/tests/test_misc.R

@@ -288,7 +288,7 @@ z1 <- pgmm(log(emp) ~ log(wage) + log(capital) + log(output),
              gmm.inst = ~log(emp), lag.gmm = list(c(2,99)))
 summary(z1, robust = FALSE)
 
-## Blundell and Bond (1998) table 4 (cf DPD for OX p. 12 col. 4)
+## Blundell and Bond (1998) table 4 (cf. DPD for OX p. 12 col. 4)
 z2 <- pgmm(dynformula(log(emp) ~ log(wage) + log(capital), list(1,1,1)),
              data = EmplUK, effect = "twoways", model = "onestep",
              gmm.inst = ~log(emp) + log(wage) + log(capital),

inst/tests/test_misc.Rout.save

@@ -1,5 +1,5 @@
 
-R version 4.4.1 (2024-06-14 ucrt) -- "Race for Your Life"
+R version 4.4.2 (2024-10-31 ucrt) -- "Pile of Leaves"
 Copyright (C) 2024 The R Foundation for Statistical Computing
 Platform: x86_64-w64-mingw32/x64
 
@@ -1493,8 +1493,8 @@ lag(log(capital), 0:1)1 -0.424393   0.058479 -7.2572 3.952e-13 ***
 Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
 
 Sargan test: chisq(100) = 118.763 (p-value = 0.097096)
-Autocorrelation test (1): normal = -4.808434 (p-value = 1.5212e-06)
-Autocorrelation test (2): normal = -0.2800133 (p-value = 0.77947)
+Autocorrelation test (1): normal = -5.519159 (p-value = 3.4063e-08)
+Autocorrelation test (2): normal = -0.2524777 (p-value = 0.80067)
 Wald test for coefficients: chisq(5) = 11174.82 (p-value = < 2.22e-16)
 Wald test for time dummies: chisq(7) = 14.71138 (p-value = 0.039882)
 > 
@@ -1542,7 +1542,7 @@ Autocorrelation test (2): normal = -0.3325401 (p-value = 0.73948)
 Wald test for coefficients: chisq(7) = 371.9877 (p-value = < 2.22e-16)
 Wald test for time dummies: chisq(6) = 26.9045 (p-value = 0.0001509)
 > 
-> ## Blundell and Bond (1998) table 4 (cf DPD for OX p. 12 col. 4)
+> ## Blundell and Bond (1998) table 4 (cf. DPD for OX p. 12 col. 4)
 > z2 <- pgmm(dynformula(log(emp) ~ log(wage) + log(capital), list(1,1,1)),
 +              data = EmplUK, effect = "twoways", model = "onestep",
 +              gmm.inst = ~log(emp) + log(wage) + log(capital),
@@ -1579,8 +1579,8 @@ lag(log(capital), 1) -0.424393   0.058479 -7.2572 3.952e-13 ***
 Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
 
 Sargan test: chisq(100) = 118.763 (p-value = 0.097096)
-Autocorrelation test (1): normal = -4.808434 (p-value = 1.5212e-06)
-Autocorrelation test (2): normal = -0.2800133 (p-value = 0.77947)
+Autocorrelation test (1): normal = -5.519159 (p-value = 3.4063e-08)
+Autocorrelation test (2): normal = -0.2524777 (p-value = 0.80067)
 Wald test for coefficients: chisq(5) = 11174.82 (p-value = < 2.22e-16)
 Wald test for time dummies: chisq(7) = 14.71138 (p-value = 0.039882)
 > 
@@ -1891,4 +1891,4 @@ Coefficients:
 > 
 > proc.time()
    user  system elapsed 
-  10.04    0.75   10.76 
+   9.45    0.90   10.32