비슷한 유형의 문제를 풀어본적이 있습니다.
http://www.gurubee.net/lecture/2848
그런데 요구조건이 조금 다르네요.
같은 방식으로 풀기는 어려울것 같아,
날짜별로 분리 한뒤 다시 그룹바이 하는 방식으로 풀어봤습니다.
99991231 에 대한 부담으로 20201231 로 변경하여 처리했습니다.
 

WITH t AS
(
SELECT 1 no, 'A' cd, '20150916' sdt, '20150922' edt, 4000 amt FROM dual
UNION ALL SELECT 2, 'A', '20150918', '20150918', 8000 FROM dual
UNION ALL SELECT 3, 'A', '20150923', '99991231', 7000 FROM dual
UNION ALL SELECT 4, 'A', '20151014', '20151014', 3000 FROM dual
UNION ALL SELECT 5, 'A', '20151114', '20151116', 5000 FROM dual
)
SELECT cd
     , TO_CHAR(MIN(dt), 'yyyymmdd') sdt
     , REPLACE(TO_CHAR(MAX(dt), 'yyyymmdd'), '20201231', '99991231') edt
     , amt
  FROM (SELECT cd, dt, no, amt
             , dt - ROW_NUMBER() OVER(PARTITION BY cd, no, amt ORDER BY dt) grp
          FROM (SELECT cd
                     , sdt + lv - 1 dt
                     , MAX(no) no
                     , MAX(amt) KEEP(DENSE_RANK LAST ORDER BY no) amt
                  FROM (SELECT no, cd, amt
                             , TO_DATE(sdt, 'yyyymmdd') sdt
                             , TO_DATE(DECODE(edt, '99991231', '20201231', edt), 'yyyymmdd') edt
                          FROM t
                        )
                     , (SELECT LEVEL lv FROM dual CONNECT BY LEVEL <= 9999)
                 WHERE lv <= edt - sdt + 1
                 GROUP BY cd, sdt + lv - 1
                )
        )
 GROUP BY cd, grp, no, amt
 ORDER BY cd, sdt
;

http://www.gurubee.net/lecture/2848
링크의 방법(선분이력을 점이력으로 변경하여 처리하는 방법)을 적용해 봤습니다.
이 방법을 적용하기 위해 원본데이터를 적용가능한 형태(tmp)로 우선 가공했습니다.

WITH t AS
(
SELECT 1 no, 'A' cd, '20150916' sdt, '20150922' edt, 4000 amt FROM dual
UNION ALL SELECT 2, 'A', '20150918', '20150918', 8000 FROM dual
UNION ALL SELECT 3, 'A', '20150923', '99991231', 7000 FROM dual
UNION ALL SELECT 4, 'A', '20151014', '20151014', 3000 FROM dual
UNION ALL SELECT 5, 'A', '20151114', '20151119', 5000 FROM dual
UNION ALL SELECT 6, 'A', '20151115', '20151115', 6000 FROM dual
)
, tmp AS
(
SELECT a.no, a.cd
     , TO_DATE(a.sdt, 'yyyymmdd') sdt
     , TO_DATE(a.edt, 'yyyymmdd') edt
     , a.amt - NVL(MAX(b.amt) KEEP(DENSE_RANK LAST ORDER BY b.no), 0) amt
  FROM t a
     , t b
 WHERE a.cd = b.cd(+)
   AND a.sdt > b.sdt(+)
   AND a.sdt < b.edt(+)
 GROUP BY a.no, a.cd, a.sdt, a.edt, a.amt
)
SELECT cd
     , TO_CHAR(sdt, 'yyyymmdd') sdt
     , NVL(TO_CHAR(edt, 'yyyymmdd'), '99991231') edt
     , amt
  FROM (SELECT cd, sdt
             , LEAD(sdt - 1) OVER(ORDER BY sdt) edt
             , SUM(SUM(amt)) OVER(ORDER BY sdt) amt
          FROM (SELECT cd
                     , DECODE(lv, 1, sdt, edt + 1) sdt
                     , DECODE(lv, 1, amt, -amt) amt
                  FROM tmp
                     , (SELECT LEVEL lv FROM dual CONNECT BY LEVEL <= 2)
                 WHERE lv = 1 OR edt != '99991231'
                )
         GROUP BY cd, sdt
        )
 WHERE amt != 0
;