st: lag length for newey west estimator

[Bulent Koksal <bkoksal@gmail.com>]

Hello All,

   I have an unbalanced panel data set consisting of stock returns. I
want to use newey-west errors for heteroskedasticity and serial
correlation. In addition, one of the RHS variables is endogenous. So I
will use lags of this variable as instruments.

my model is

y1=alpha0+alpha1 y2 + beta1 x1 + beta2 x2 + beta3 x3 +error

where y1 and y2 are endogenous.

If I am not mistaken, I can do this by xtivreg2. Something like

xtivreg2 y1 x1 x2 x3 (y2= l(1/2).y2), fe bw(1+L) robust

where L is the required lag length for Newey-West estimator.

Two questions:

Greene (econometric analysis) states that in practice the smallest
integer greater than equal to n^0.25 is used to determine L. For a
panel data set, should I take n to be the overall sample size , or the
size of the stock that has the longest time series? The latter seems
to be the correct choice but I am not sure.

While using lags of a variable as its instruments, how do we decide
for the correct number of lags, assuming that sample size is not a
problem?

Thanks in advance for any help.

-- 
Bülent Köksal

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/