.. index::
   single: G7 Command; sur
   single: Seemingly Unrelated Regressions
   single: Generalized Least Squares
   single: G7 Command; stack

Seemingly Unrelated Regressions
===============================

The format for the sur command is shown by this example::

   sur 					
   r y1 = x11, x12, ...
   ...
   r yn = xn1, xn2, ...
   con 10 1 = a2 + 4b5 + etc
   sma 100 a6 a12 1
   do

After the *sur* command come all of the individual regressions.  Any *constraint* or *sma* commands must come after the regressions.  In the *con* and *sma* lines, a's denote coefficients in the first regression; b's coefficients in the second, and so on.  Thus, "a2" denotes the second coefficient of the first regression and "b5" denotes the fifth coefficient of the second regression.

The predicted values, dependent variables, and regression coefficients of the successive equations appear in the workspace file as "predic1", "depvar1", "rcoef1", "predic2", "depvar2", "rcoef2", etc.  The results of the individual equations can be plotted by "gr \*1", "gr \*2", etc.  The *graph* commands should follow the *do* command and each should be preceeded by an appropriate *title* command.

**stack**
   The format for *stack* is exactly like that for *sur* except that the word "stack" replaces "sur".  With *stack*, no attention is paid to contemporaneous covariances.  The point of *stack* is solely to impose soft constraints across regressions.

Limitation:  Earlier editions of *G7* imposed a limit of 30 regression variables, counting dependent, intercepts, and independent variables, that quickly could become binding in *sur* and *stack*, since it applied to the total variables in all equations involved in the *sur* or *stack*.  In later versions, this constraint was relaxed so that *G7* now can employ up to 500 regression variables.