st: rollreg, gaps

[Fabian Schönenberger <fabian.schoenenberger@gmail.com>]

Dear all
I have a problem conducting rollreg. It says my sample has gaps.

rollreg lnebit lnsales, move(3) stub(sales)

Number of gaps in sample:  1415

Observations with preceding time gaps
----------------------------------
   Record |      cusip       fyear
----------+-----------------------
       90 |  00089J102        2005
etc.

However, doing 
tsset cusip fyear
       panel variable:  cusip (unbalanced)
        time variable:  fyear, 1979 to 2009
                delta:  1 unit

gives no evidence about gaps. The same holds fo xtset.

Before I conducted rollreg I checked my data and adjusted the following way:
sort fyear cusip
. by fyear: generate time2= _n
. tsset time2
. tset fyear
. tsfill

My data on record 90 looks that way:
	gvkey	datadate	fyear	lnebit	lnsales	ebitgrowth
salesgrowth	cusip
88 	63106	2.0e+07	2003	-.730925	-3.91202	-.622899
-.93007	00089J102
89				2004
00089J102
90	63106	2.0e+07	2005	-.561899	-.524249	-.299907
28.6		00089J102

I do not have data points for 2004 (and for other 1414 data points). The
reason why I would like to conduct rollreg is, that I am interest in the
coefficients of rollreg lnebit lnsales for firms at different point in times
in order to use these coefficients for further investigations. Also, the
number of fyears available for cusips is differently. 
So, do I have to drop all cusips with gaps or is there another way to adjust
my data?
Many thanks in advance, 

FS 

*
*   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/