Maska321,
Можно примерно так...но чтобы именно как вы хотите надо из тогда пронумеровать еще, последовательность непонятно чем задана

Maska321,
Вот результат как вы хотите, но пришлось добавить последовательность записей
with
w_data as
(
select 1 as nr, to_date('01.02.2005', 'dd.mm.yyyy') as dt, 20 as val from dual union all
select 2 as nr, to_date('01.03.2005', 'dd.mm.yyyy') as dt, 20 as val from dual union all
select 3 as nr, to_date('01.04.2005', 'dd.mm.yyyy') as dt, 21 as val from dual union all
select 4 as nr, to_date('01.05.2005', 'dd.mm.yyyy') as dt, 21 as val from dual union all
select 5 as nr, to_date('01.06.2005', 'dd.mm.yyyy') as dt, 20 as val from dual union all
select 6 as nr, to_date('01.08.2005', 'dd.mm.yyyy') as dt, 26 as val from dual union all
select 7 as nr, to_date('01.07.2005', 'dd.mm.yyyy') as dt, 26 as val from dual
)
select max(dt) keep (dense_rank first order by nr) as dt,
max(val) as val
from (
select dt,
val,
nr,
sum(flg) over (order by nr) as grp
from (
select d.*,
case when lag(val) over (order by nr) = val then null else 1 end as flg
from w_data d
)
)
group by grp
order by dt