# st: Re: sureg is not xtreg

 From Kit Baum To statalist@hsphsun2.harvard.edu Subject st: Re: sureg is not xtreg Date Tue, 16 Mar 2004 13:46:59 -0500

On Mar 13, 2004, at 2:33 AM, Dimitris wrote:


I have a panel of firms with yearly data on a number of (log) variables.
Assume there are 5 variables. Without firm effects, I can use -sureg- to
detrend all variables by a single constrained growth rate as follows:

gen t1 = time
gen t2 = time
gen t3 = time
gen t4 = time
gen t5 = time

constraint define 1 t2=t1
constraint define 2 t3=t2
constraint define 3 t4=t3
constraint define 4 t5=t4

sureg (X1: lnx1 t1) (X2: lnx2 t2) (X3: lnx3 t3) (X4: lnx4 t4) (X5: lnx5
t5), constraint (1 2 3 4)

The model detrends each variable using the constrained slope across the 5
variables and allows the intercepts (i.e. estimated initial values) to
vary. The fitted values for any variable (e.g. lnx1), which vary with time
but not with firm, are given by:

predict fvx1, equation(X1) xb

To introduce firm effects in -sureg- I may use -by firm- . The SURs then
run successively for a large number of firms.

Using this approach, how do I
(i) obtain a coparable table of parameter estimates for all the estimated
- -by firm, sureg-
(ii) extract a vector of fitted values for each of the variables?

ALso, I want to run this as a single model, in order to then compare with
other models, including -xtreg-, -xtregar- and -xtabond-.

To set the data up as panel data, we use the commands

gen year = time + <base year>
tsset firm year, yearly

Therefore, how can I run the SUR model with panel data?
SURE (sureg) is an estimator for use with panel data in the "wide" format--e.g. firms' data in adjacent variables. It always is applied to panel data. If you want to use an estimator in the "long" format, such as xtreg, you must reshape long. A fixed-effects model (xtreg, fe) will estimate the same regression across units of the panel, with intercepts firm-specific; the only difference between that and the slope-constrained SURE is that the latter allows for \sigma^2 to vary across firms, which xtreg,fe does not.

Kit

*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/