One suggestion:
1407 PROC SQL;
1408 CREATE TABLE est2 AS SELECT
1409 est2.*, est1.gvkey1 AS gvkey_j, est1.a_i AS a_j, est1.b_i AS b_j
1410 FROM est2 LEFT JOIN est1
1411 ON est2.gvkey1 ~= est1.gvkey1
1412 AND est2.sic2 = est1.sic2;
1413 QUIT;
has at least a chance of being a problem.
When I run a simpler program I get this warning:
9 proc sql;
10 create table work.class as
11 select *
12 from work.class
13 ;
WARNING: This CREATE TABLE statement recursively references the
target table. A consequence of this is a possible data
integrity problem.
NOTE: Table WORK.CLASS created, with 19 rows and 5 columns.
14 quit;
General style notes when working with macro code: Always use a run, or quit, to end a procedure or data step.
The nature of the macro language makes it very possible to generate code that does not end steps at the time you expect them too otherwise.
Do not reuse data set names inside a macro. When you reuse the same name then you cannot tell where the contents have problems.
Always explicitly state the input data set by using data=somesetname on procedures. It is very easy to have a macro code branch execute and you forgot which set was intended to be sorted leaving 1) the proper set unsorted and 2) possibly place the wrong order on another set or 3) using some other proc on the wrong set.
Any time you use constructs like
1398 data temp1a; merge temp1a nfirms; by gvkey1;
or Set, Modify or Update as well as merge, you completely replace the input data set. So logic errors can make it impossible to tell whether a set has bad values before or after the merge (or set ).
In this block of code:
1415 data est2; set est2;
1416 a_dif = a_i - a_j;
1417 b_dif = b_i - b_j;
1418 drop a_i a_j b_i b_j;
1419
1420 proc sort; by gvkey1 datadate1 fqenddt1 datacqtr1 gvkey_j;
1421
1422 data est2; set est2;
1423
1424 proc sort; by gvkey1 datadate1 fqenddt1 datacqtr1 gvkey_j fqenddt;
1425
1426 data est2; set est2;
1427 error = ABS(a_dif + bhr * b_dif);
Lines 1420 through 1422 do, for practical purposes, NOTHING.
When you sort data without a need to use by processing, as the three data steps do, then you are wasting potentially lots of time for no purpose.
The whole bit can be accomplished with
data est2;
set est2;
a_dif = a_i - a_j;
b_dif = b_i - b_j;
drop a_i a_j b_i b_j;
error = ABS(a_dif + bhr * b_dif);
run;
General rule of thumb when you find that you need a new variable: go back to a previous data step and add the code there, don't run a data step to add one variable.
And the Log should have included:
The statement that calls the macro
Any messages, notes or errors that appear from running the macro.
The Option Mprint, in addition to providing details of the code generated the call will place the notes and messages in relation to the code that creates them.
