본 쿼리가 어떻게 되는지 모르겠으나 기술할 컬럼이 많다면

max keep 보다 row_number over를 쓰시는 게 나아 보여요..

with t as
(select 'a' gp, '2016-04-01' dt from dual union all
select 'a' gp, '2016-01-01' dt from dual union all
select 'a' gp, '2016-02-01' dt from dual union all
select 'b' gp, '2016-03-01' dt from dual union all
select 'b' gp, '2016-04-01' dt from dual
)
SELECT *
  FROM (SELECT gp
             , dt
             , ROW_NUMBER () OVER (PARTITION BY gp ORDER BY dt DESC) rn
          FROM t)
 WHERE rn = 1

참고해서 작성해보세요.

with t as
(select 'a' gp, '2016-04-01' dt, '2016-09-01' dt2 from dual union all
select 'a' gp, '2016-01-01' dt, '2016-03-01' dt2 from dual union all
select 'a' gp, '2016-02-01' dt, '2016-06-01' dt2 from dual union all
select 'b' gp, '2016-03-01' dt, '2016-07-01' dt2 from dual union all
select 'b' gp, '2016-04-01' dt, '2016-08-01' dt2 from dual
)
SELECT gp
     , max(dt)
     , max(dt2) keep(dense_rank last order by dt) dt2
  FROM t
 group by gp

max keep 으로는 이렇게 하면 될 것 같은데 컬럼이 많을 수록 좀 보기가 안좋아질 것 같아서요.

그리고 dt가 동일하고 dt2가 불일치하는 데이터가 있을 경우의 처리 문제도 있구요.

성능은 어느 게 좋을지 잘 모르겠네요.

--참고로 keep은

with t(gp, dt, data) as ( 
select 'a', '2016-04-01','a' from dual union all 
select 'a', '2016-01-01','b' from dual union all 
select 'a', '2016-02-01','c' from dual union all
select 'a', '2016-02-16','d' from dual union all 
select 'a', '2016-03-01','e' from dual union all 
select 'b', '2016-03-01','a' from dual union all 
select 'b', '2016-04-01','b' from dual union all 
select 'b', '2016-02-16','c' from dual union all
select 'b', '2016-02-01','d' from dual union all
select 'b', '2016-01-01','e' from dual 
)
select gp
      ,max(dt) dt
     , max(data) keep(DENSE_RANK LAST ORDER BY dt) data
from t
group by gp;