Summing and then getting min and max of top 70% in SQL - sql

Question

I have data for the purchases of a product formatted like this:
Item | Price | Quantity Bought
ABC 10.10 4
DEF 8.30 12
DEF 7.75 8
ABC 10.50 20
GHI 15.4 1
GHI 15.2 12
ABC 10.25 8
... ... ...
Where each row represents an individual purchasing a certain amount at a certain price. I would like to aggregate this data and eliminate the prices below the 30th percentile for total quantity bought from my table.
For example, in the above data set the total amount of product ABC bought was (4+20+8) = 32 units, with average price = (4*10.10 + 8*10.25 + 20*10.50)/32 = 10.39.
I would like to organize the above data set like this:
Item | VWP | Total Vol | 70th %ile min | 70th %ile max
ABC 10.39 32 ??? ???
DEF ... 20 ??? ???
GHI ... 13 ??? ???
Where VWP is the volume weighted price, and the 70th %ile min/max represent the minimum and maximum prices within the top 70% of volume.
In other words, I want to eliminate the prices with the lowest volumes until I have 70% of the total volume for the day contained in the remaining prices. I would then like to publish the min and max price for the ones that are left in the 70th %ile min/max columns.
I tried to be as clear as possible, but if this is tough to follow along with please let me know which parts need clarification.
Note: These are not the only columns contained in my dataset, and I will be selecting and calculating other values as well. I only included the columns that are relevant to this specific calculation.
EDIT:
Here is my code so far, and I need to incorporate my calculation into this (the variables with the '#' symbol before them are inputs that are given by the user:
SELECT Item,
SUM(quantity) AS Total_Vol,
DATEADD(day, -#DateOffset, CONVERT(date, GETDATE())) AS buyDate,
MIN(Price) AS MinPrice,
MAX(Price) AS MaxPrice,
MAX(Price) - MIN(Price) AS PriceRange,
ROUND(SUM(Price * quantity)/SUM(quantity), 6) AS VWP,
FROM TransactTracker..CustData
-- #DateOffset (Number of days data is offset by)
-- #StartTime (Time to start data in hours)
-- #EndTime (Time to stop data in hours)
WHERE DATEDIFF(day, TradeDateTime, GETDATE()) = (#DateOffset+1)
AND DATEPART(hh, TradeDateTime) >= #StartTime
AND HitTake = ''
OR DATEDIFF(day, TradeDateTime, GETDATE()) = #DateOffset
AND DATEPART(hh, TradeDateTime) < #EndTime
AND HitTake = ''
GROUP BY Item
EDIT 2:
FROM (SELECT p.*,
(SELECT SUM(quantity) from TransactTracker..CustData p2
where p2.Series = p.Series and p2.Size >= p.Size) as volCum
FROM TransactTracker..CustData p
) p
EDIT 3:
(case when CAST(qcum AS FLOAT) / SUM(quantity) <= 0.7 THEN MIN(Price) END) AS min70px,
(case when CAST(qcum AS FLOAT) / SUM(quantity) <= 0.7 THEN MAX(Price) END) AS max70px
FROM (select p.*,
(select SUM(quantity) from TransactTracker..CustData p2
where p2.Item = p.Item and p2.quantity >= p.quantity)
as qcum from TransactTracker..CustData p) cd

Answer 1

There is some ambiguity on how you define 70 % when something goes over the threshold. However, the challenge is two fold. After identifying the cumulative proportion, the query also needs to choose the appropriate row. This suggests using row_number() for selection.
This solution using SQL Server 2012 syntax calculates the cumulative sum. It then takes assigns a sequential value based on how close the ratio is to 70%.
select item,
SUM(price * quantity) / SUM(quantity) as vwp,
SUM(quantity) as total_vol,
min(case when seqnum = 1 then price end) as min70price,
max(case when seqnum = 1 then price end) as max70price
from (select p.*,
ROW_NUMBER() over (partition by item order by abs(0.7 - qcum/qtot) as seqnum
from (select p.*,
SUM(quantity) over (partition by item order by vol desc) as qcum,
SUM(quantity) over (partition by item) as qtot
from purchases p
) p
) p
group by item;
To get the largest value less than 70%, then you would use:
max(case when qcum < qtot*0.7 then qcum end) over (partition by item) as lastqcum
And then the case statements in the outer select would be:
min(case when lastqcum = qcum then price end) . .
In earlier versions of SQL Server, you can get the same effect with the correlated subquery:
select item,
SUM(price * quantity) / SUM(quantity) as vwp,
SUM(quantity) as total_vol,
min(case when seqnum = 1 then price end) as min70price,
max(case when seqnum = 1 then price end) as max70price
from (select p.*,
ROW_NUMBER() over (partition by item order by abs(0.7 - qcum/qtot) as seqnum
from (select p.*,
(select SUM(quantity) from purchases p2 where p2.item = p.item and p2.quantity >= p.quantity
) as qsum,
SUM(quantity) over (partition by item) as qtot
from purchases p
) p
) p
group by item
Here is the example with your code:
SELECT Item,
SUM(quantity) AS Total_Vol,
DATEADD(day, -#DateOffset, CONVERT(date, GETDATE())) AS buyDate,
MIN(Price) AS MinPrice,
MAX(Price) AS MaxPrice,
MAX(Price) - MIN(Price) AS PriceRange,
ROUND(SUM(Price * quantity)/SUM(quantity), 6) AS VWP,
min(case when seqnum = 1 then price end) as min70price,
max(case when seqnum = 1 then price end) as max70price
from (select p.*,
ROW_NUMBER() over (partition by item order by abs(0.7 - qcum/qtot) as seqnum
from (select p.*,
(select SUM(quantity) from TransactTracker..CustData p2 where p2.item = p.item and p2.quantity >= p.quantity
) as qsum,
SUM(quantity) over (partition by item) as qtot
from purchases TransactTracker..CustData
) p
) cd
-- #DateOffset (Number of days data is offset by)
-- #StartTime (Time to start data in hours)
-- #EndTime (Time to stop data in hours)
WHERE DATEDIFF(day, TradeDateTime, GETDATE()) = (#DateOffset+1)
AND DATEPART(hh, TradeDateTime) >= #StartTime
AND HitTake = ''
OR DATEDIFF(day, TradeDateTime, GETDATE()) = #DateOffset
AND DATEPART(hh, TradeDateTime) < #EndTime
AND HitTake = ''
GROUP BY Item