You are mixing Watcom and TSQL dialects. You must remove the AS for the exception handler to be understood. Also, you should be using SET rather than SELECT variable = to assign values for Watcom dialect procedures. You may want to use EXECUTE IMMEDIATE rather than execute(variable) for that dialect also.

There are other issues with this procedure including the use of keywords such as "KEY" which may need to be quoted depending on your quoted_identifiers setting. Here is a rewrite that has no syntax errors. Other problems include missing statement separatators and a bad LOOP construct which mixed the LOOP ... END LOOP syntax with a bad WHILE contruct.

 CREATE PROCEDURE "YBWI_PROC"()
 BEGIN

DECLARE err_notfound
    EXCEPTION FOR SQLSTATE '02000';

DECLARE V_IK NUMERIC(9);
    DECLARE V_KEY NUMERIC(1,0);
    DECLARE V_DATEA NUMERIC(6,0);
    DECLARE V_WERTTYP VARCHAR(4);
    DECLARE @sqlcmd varchar(999);

DECLARE YBWIIRW CURSOR FOR
    SELECT IK, "KEY", DATEA
    FROM "PBS"."YBWIIRW"
    group by IK, "KEY", DATEA;

OPEN YBWIIRW;

YBWIIRWLoop:
    LOOP
           FETCH NEXT YBWIIRW
          INTO V_IK, V_KEY,V_DATE;
       IF SQLCODE <> 0 THEN
          LEAVE YBWIIRWLoop;
       END IF;

set  @sqlcmd =
    'select max(TYP) as max_typ from "PBS"."YBWIIRW" where ' +
    'IK = ' + V_IK + ' and KEY = ' + V_KEY + ' and DATEA = ' + V_DATEA;
    execute (@sqlcmd);

set @sqlcmd =
    'update "PBS"."YBWIIRW" a set a.PLANWERT= b.planwert from "DBA"."GDB_S_PLANWERTE" b where a."IK" = ' + V_IK + ' and a."KEY" = ' + V_KEY +
    ' and a."DATEA" = ' + V_DATEA + ' and a.TYP = ' + max_typ;
    execute (@sqlcmd);

END LOOP YBWIIRWLoop;

CLOSE YBWIIRW;
 END

Besides my comment on the question, some hints:

• You do not need dynamic SQL (aka an EXECUTE statement) to build a SQL statement based on SQL variables - they are known within the procedure's body, so a simple

  where IK = V_IK and KEY = V_KEY and DATEA = V_DATEA

should do, and the SELECT/UPDATE statement can be run "normally". (However, "KEY" may be a reserved word and might need quoting.)

• An UPDATE based on a max value can be done without a cursor loop - you could just put the "max selection" into the UPDATE's join condition. That would let you omit all the variable and exception handling stuff.

• If you do prefer a cursor loop, you may take a look at the FOR statement, which combines several cursor handling "steps" into a very handy form - again omitting the check for the "row not found" warning (which is not an error, BTW).