SQL: Deduce Effective Pricing from overlapping Dates - sql

Question

I have pricing record with overlapping dates. On few dates there are more than one overlapping prices. Please follow the example below:
Example on 2022/02/15 there are 2 prices 10 and 8 .
article
price
startdate
enddate
123
10
2022/02/02
2049/12/31
123
8
2022/02/14
2022/09/14
123
5
2022/03/14
2022/04/06
123
4
2022/04/11
2022/04/27
I want to apply the effective price for date ranges like below and avoid conflicting prices in the output.
article
price
startdate
enddate
123
10
2022/02/02
2022/02/13
123
8
2022/02/14
2022/03/13
123
5
2022/03/14
2022/04/06
123
8
2022/04/07
2022/04/10
123
4
2022/04/11
2022/04/27
123
8
2022/04/28
2022/09/14
123
10
2022/09/15
2049/12/31
I can think of window functions to adjust the end dates and prices, but I cannot wrap my head around the problem completely to get the complete solution. Any suggestion/solution is appreciated.
Database: Snowflake
Thank you

Answer 1

Using the logic of new starting price window wins for overlaps.
Discreate Date version:
with data(article,price,startdate,enddate) as (
select * FROM VALUES
(123, 10, '2022-02-02'::date, '2049-12-31'::date),
(123, 8, '2022-02-14'::date, '2022-09-14'::date),
(123, 5, '2022-03-14'::date, '2022-04-06'::date),
(123, 4, '2022-04-11'::date, '2022-04-27'::date)
), dis_times as (
select article,
date as startdate,
lead(date) over(partition by article order by date)-1 as enddate
from (
select distinct article, startdate as date from data
union
select distinct article, enddate+1 as date from data
)
qualify enddate is not null
)
select
d1.article,
d1.price,
d2.startdate,
d2.enddate
from data as d1
join dis_times as d2
on d1.article = d2.article
and d2.startdate between d1.startdate and d1.enddate qualify row_number() over (partition by d1.article, s_startdate order by d1.startdate desc) = 1
order by 1,3;
gives:
ARTICLE
PRICE
S_STARTDATE
S_ENDDATE
123
10
2022-02-02
2022-02-13
123
8
2022-02-14
2022-03-13
123
5
2022-03-14
2022-04-06
123
8
2022-04-07
2022-04-10
123
4
2022-04-11
2022-04-27
123
8
2022-04-28
2022-09-14
123
10
2022-09-15
2049-12-31
Continuous Timestamp version:
with data(article,price,startdate,enddate) as (
select * FROM VALUES
(123, 10, '2022-02-02'::date, '2049-12-31'::date),
(123, 8, '2022-02-14'::date, '2022-09-14'::date),
(123, 5, '2022-03-14'::date, '2022-04-06'::date),
(123, 4, '2022-04-11'::date, '2022-04-27'::date)
), dis_times as (
select article,
date as startdate,
lead(date) over(partition by article order by date) as enddate
from (
select distinct article, startdate as date from data
union
select distinct article, enddate as date from data
)
qualify enddate is not null
)
select
d1.article,
d1.price,
d2.startdate,
d2.enddate
from data as d1
join dis_times as d2
on d1.article = d2.article
and d2.startdate >= d1.startdate and d2.startdate < d1.enddate
qualify row_number() over (partition by d1.article, s_startdate order by d1.startdate desc) = 1
order by 1,3;
which gives:
ARTICLE
PRICE
S_STARTDATE
S_ENDDATE
123
10
2022-02-02
2022-02-14
123
8
2022-02-14
2022-03-14
123
5
2022-03-14
2022-04-06
123
8
2022-04-06
2022-04-11
123
4
2022-04-11
2022-04-27
123
8
2022-04-27
2022-09-14
123
10
2022-09-14
2049-12-31
Thanks to MatBailie for the tighter join suggestion.
join dis_times as d2
on d1.article = d2.article
and d2.startdate between d1.startdate and d1.enddate
the continuous range I would normally do in this for
and d2.startdate between d1.startdate and d1.enddate and d2.startdate < d1.enddate
instead of this form
and d2.startdate >= d1.startdate and d2.startdate < d1.enddate
because I in experience it performed better. always test your complexities.

Answer 2

First thing I did was --I turned your price-per-date range data into a price-per-date lookup table.
create or replace temporary table price_date_lookup as
select distinct
article,
dateadd('day',b.index-1,start_date) as dates,
first_value(price) over (partition by article, dates order by end_date) as price
from my_table,
lateral split_to_table(repeat('.',datediff(day,start_date,end_date)), '.') b;
Notes:
first_value handles overlaps by overriding prices based on their end dates.
lateral... basically helps create a date column with all the days in the range
As soon as I created that table, I figured the rest could be approached like a gaps and island problem.
with cte1 as
(select *, case when lag(price) over (partition by article order by dates)=price then 0 else 1 end as price_start --flag start of a new price island
from price_date_lookup),
cte2 as
(select *, sum(price_start) over (partition by article order by dates) as price_id --assign id to all the price islands
from cte1)
select article,
price,
min(dates) as start_date,
max(dates) as end_date
from cte2
group by article,price,price_id;