[Limdep Nlogit List] ABB DPD model with unbalanced panles

[Barry Williams]

Dear All,

I am trying to run the Arellano Bond and Bover Dynamic Panel model with an unbalanced panel.

So far I have been able to run the example code from section E11.15 quite successfully on the Cornwall and Rupert data and reproduce the results from the manual.

I have then applied the model to a sub-sample of my data that could be balanced quite easily without any real biases.

My code is:

NAMELIST; x1 = HHI2, CHGHHI2, TREND$
NAMELIST; x2 =LNTOTY, LNTOTYSQ$
NAMELIST; f1 = one$

NAMELIST; x = x1,x2 
	; f = f1
	; exog = x1,f1 
	; all = x1,x2,f1$
CREATE  ; y = SD_PROF$
CALC	; kx1=Col(x1) ; kx2 = Col(x2) ; kf1 = Col(f1) ; kf2 = 0$


REGRESS	; Lhs = y 
	; Rhs = all  
	; Start = kx1,kx2,kf1,kf2;
	; pds=25
	; Panel
	;DPD
	;Pattern=0$

My questions are:

1) Does any one have any suggestions on how to run the ABB model with an unbalanced data set? (The manuals indicate that is an intricate problem)
2) At this stage I am only able to use Pattern =0, as otherwise I receive the error message: Error   636: You have too many instruments (moment conditions).  I assume that this is because I have 25 quarters of data.

As a follow-up I have also run my model using the Kiviet (1995) approach from Greene (2008) page 246, with yit-1 being approximated by yit-2. Is there a mechanism to use yit-2 as an alternative to yit-1 within the ABB framework?

Many thanks for any help,

Barry