with t as (
   select  1  id, null parent_id, 'Top' description,  0  amount from dual union all
   select  2  id,  1  parent_id, 'Top-One' description,  0  amount from dual union all
   select  3  id,  2  parent_id, 'One-one' description,  2000  amount from dual union all
   select  4  id,  2  parent_id, 'One-two' description,  3000  amount from dual union all
   select  5  id,  1  parent_id, 'Top-Two' description,  1000  amount from dual 
)
select *
from t
start with parent_id is null
connect by prior id = parent_id
model 
dimension by (id, parent_id, rownum r)
measures(amount)
rules (
   amount[any, any, any] order by r desc = nvl(amount[cv(), cv(), cv()], 0 ) + nvl(sum(amount)[any, cv(id), any], 0 )
)

with t as ( --id - уникальны
   select  1  id, null parent_id, 'Top' description,  10  amount from dual union all
   select  2  id,  1  parent_id, 'Top-One' description,  100  amount from dual union all
   select  3  id,  2  parent_id, 'One-one' description,  2000  amount from dual union all
   select  4  id,  2  parent_id, 'One-two' description,  3000  amount from dual union all
   select  5  id,  1  parent_id, 'Top-Two' description,  1000  amount from dual union all
   select  4  id,  2  parent_id, 'One-three' description,  300  amount from dual 
)
select id, description, amount, 
   case l 
      when  1  then sum(amount) over(partition by p1)
      when  2  then sum(amount) over(partition by p2)
      when  3  then sum(amount) over(partition by p3)
      when  4  then sum(amount) over(partition by p4)
      when  5  then sum(amount) over(partition by p5)
   end ha
from (
select level l, id, parent_id, description, amount, 
sys_connect_by_path(id,'/') p,
substR(sys_connect_by_path(id,'/'),  1 ,  1 * 2 ) p1,
substR(sys_connect_by_path(id,'/'),  1 ,  2 * 2 ) p2,
substR(sys_connect_by_path(id,'/'),  1 ,  3 * 2 ) p3,
substR(sys_connect_by_path(id,'/'),  1 ,  4 * 2 ) p4,
substR(sys_connect_by_path(id,'/'),  1 ,  5 * 2 ) p5
from t
start with parent_id is null
connect by prior id = parent_id
) tt;

with t as ( 
   select  1  id, null parent_id, 'Top' description,  10  amount from dual union all
   select  2  id,  1  parent_id, 'Top-One' description,  100  amount from dual union all
   select  3  id,  2  parent_id, 'One-one' description,  2000  amount from dual union all
   select  4  id,  2  parent_id, 'One-two' description,  3000  amount from dual union all
   select  5  id,  1  parent_id, 'Top-Two' description,  1000  amount from dual union all
   select  6  id,  2  parent_id, 'One-three' description,  300  amount from dual 
)
select 
   min(id) keep(dense_rank first order by l) id,
   min(description) keep(dense_rank first order by l) description,
   sum(amount)
from (
     select level l, id, parent_id, description, amount, 
     sys_connect_by_path(id,'/') p,
     std.subword(sys_connect_by_path(id,'/'),  1 ) p1,
     std.subword(sys_connect_by_path(id,'/'),  2 ) p2,
     std.subword(sys_connect_by_path(id,'/'),  3 ) p3,
     std.subword(sys_connect_by_path(id,'/'),  4 ) p4,
     std.subword(sys_connect_by_path(id,'/'),  5 ) p5 --максимальная глубина хиерархии
   from t
   start with parent_id is null
   connect by prior id = parent_id
) tt
group by p1, rollup(p2,p3,p4,p5)
having grouping_id(p2, p3, p4, p5) = power( 2 ,  5 -min(l))- 1 
;