[R] System exactly singular with pgmm (package plm)

[Luca Gagliardone]

his is my first post, I'll do my best to be clear and complete.

I am trying to run a pgmm regression (Arellano Bond estimator) following
the example online with the EmplUK dataset.

My dataset is unbalanced, with some missing values (that I also removed,
without any difference). This is the paste from R' dataframe.

 row.names ID   Year    p       I
    1   23  1   1992    NA      NA
    2   22  1   1993    17.01   NA
    3   21  1   1994    15.86   NA
    4   20  1   1995    17.02   7.512347
    5   19  1   1996    20.64   7.685104
    6   18  1   1997    19.11   12.730282
    7   17  1   1998    12.76   12.633871
    8   16  1   1999    17.90   7.416381
    9   15  1   2000    28.66   6.396114
    10  14  1   2001    24.46   9.213729
    11  13  1   2002    24.99   20.117159
    12  12  1   2003    28.85   11.117816
    13  11  1   2004    38.26   11.242638
    14  10  1   2005    54.57   13.015168
    15  9   1   2006    65.16   18.507212
    16  8   1   2007    72.44   18.875281
    17  7   1   2008    96.94   24.459170
    18  6   1   2009    61.74   21.332035
    19  5   1   2010    79.61   17.119038
    20  4   1   2011    111.26  16.941914
    21  3   1   2012    111.63  19.964875
    22  2   1   2013    108.56  28.863894
    23  1   1   2014    99.03   15.182615
    24  45  2   1993    17.01   NA
    25  44  2   1994    15.86   NA
    26  43  2   1995    17.02   NA
    27  42  2   1996    20.64   NA
    28  41  2   1997    19.11   NA
    29  40  2   1998    12.76   NA
    30  39  2   1999    17.90   11.428262
    31  38  2   2000    28.66   20.232613
    32  37  2   2001    24.46   25.811754
    33  36  2   2002    24.99   18.959958
    34  35  2   2003    28.85   20.767074
    35  34  2   2004    38.26   29.260406
    36  33  2   2005    54.57   25.837434
    37  32  2   2006    65.16   32.675618
    38  31  2   2007    72.44   48.415190
    39  30  2   2008    96.94   42.444435
    40  29  2   2009    61.74   40.047462
    41  28  2   2010    79.61   49.090816
    42  27  2   2011    111.26  53.828050
    43  26  2   2012    111.63  61.684020
    44  25  2   2013    108.56  68.394140
    45  24  2   2014    99.03   55.738584
    46  76  3   1984    NA      NA
    47  75  3   1985    NA      NA
    48  74  3   1986    NA      NA
    49  73  3   1987    18.53   NA
    50  72  3   1988    14.91   NA
    51  71  3   1989    18.23   NA
    52  70  3   1990    23.76   17.046268
    53  69  3   1991    20.04   30.191128
    54  68  3   1992    19.32   30.414108
    55  67  3   1993    17.01   27.916000
    56  66  3   1994    15.86   26.437651
    57  65  3   1995    17.02   25.895513
    58  64  3   1996    20.64   26.791996
    59  63  3   1997    19.11   30.074375
    60  62  3   1998    12.76   42.636103
    61  61  3   1999    17.90   46.862510
    62  60  3   2000    28.66   30.154079
    63  59  3   2001    24.46   30.297644
    64  58  3   2002    24.99   34.851205
    65  57  3   2003    28.85   38.854943
    66  56  3   2004    38.26   37.542447
    67  55  3   2005    54.57   38.456399
    68  54  3   2006    65.16   43.465535
    69  53  3   2007    72.44   41.749414
    70  52  3   2008    96.94   48.371262
    71  51  3   2009    61.74   54.914470
    72  50  3   2010    79.61   65.444964
    73  49  3   2011    111.26  76.888119
    74  48  3   2012    111.63  81.833602
    75  47  3   2013    108.56  83.800483
    76  46  3   2014    99.03   79.713947

my codes are the following:

data <- plm.data(Autoregression,index=c("ID","Year"))


Panel <- subset(data, !is.na(I) )

Are <- pgmm( I~p+lag( I , 0:1)
            | lag(I, 2:99),
            data = Panel, effect = "twoways", model = "onestep")

I have tried also many other versions, including every possible number of
the lags, shorter or longer.

The error is the following :

Errore in solve.default(crossprod(WX, t(crossprod(WX, A1)))) :
  Lapack routine dgesv: system is exactly singular: U[3,3] = 0
Inoltre: Warning message:
In pgmm(I ~ lag(I, 1) + p | lag(I, 2:10), Panel, effect = "twoways",  :
  the first-step matrix is singular, a general inverse is used

Can you please help me? Thanks for the attention, i'll wait for an answer

Regards, Luca.

	[[alternative HTML version deleted]]