CODEWITHSUNDEEP
sqlsql-server-2008
Sreerag
Sreerag

Reputation: 1401

SQL: use WHERE clause in OVER()?

How can I use WHERE clause to filter in the OVER clause?

i.e. from the following data 

LoanID | Principal | Tenor | AmortizingPrincipal 
----------------------------------------
1         20000       1       5000
1         20000       2       5000
1         20000       3       5000
1         20000       4       5000

I need a fourth virtual column with the Balance Principal in each Tenor like the following: 

LoanID | Principal | Tenor | AmortizingPrincipal | BalancePrinicpal 
-----------------------------------------------------------
1        20000       1       5000                  20000  
1        20000       2       5000                  15000  
1        20000       3       5000                  10000 
1        20000       4       5000                  5000

Something like this:

SELECT 
    BalancePrincipal = Principal - SUM(AmortizingPrincipal) OVER(PARTITION BY LoanID WHERE Tenor < this row's tenor)

UPDATE:

The following query gives me the desired result:

SELECT 
    L1.*    
    ,BalancePrincipal = AL1.Principal - ISNULL(Cumulative.AmortizingSum,0) 
FROM
    Loan L1
CROSS APPLY 
    (
        SELECT 
            AmortizingSum = SUM(AmortizingPrincipal)
        FROM 
            Loan L2
        WHERE 
            L1.LoanID = L2.LoanID
            AND 
            L1.Tenor > L2.Tenor 
    ) Cumulative

Can it be bettered?

Upvotes: 15

Views: 38288

Answers (2)

Glenn
Glenn

Reputation: 9170

For the sample posted, it doesn't look like a filter is needed:

SELECT LoanID, Principal, Tenor, AmortizingPrincipal
      ,SUM(AmortizingPrincipal) OVER(PARTITION BY LoanID ORDER BY Tenor Desc) AS BalancePrincipal
  FROM loan
  ORDER BY LoanID, Principal, Tenor

UPDATE:

Seems Sql Server 2008 does not have the windowing clause? I didn't even think you could create an analytic function without a windowing clause. The above sql was run on Oracle and Postgres without issue. By default the window clause is UNBOUNDED PRECEDING AND CURRENT ROW (from - to). But you could change the order and go from CURRENT ROW to UNBOUNDED FOLLOWING.

UPDATE2:

So I puzzled: what meaning would a (cumulative) SUM have in an analytic function if you are unable to order the rows within the partition? Is there an implicit ordering? I can change the window (below) and get the same result, but must provide the ORDER BY (in Oracle and Postgres). I can't see how the analytic SUM would have any meaning without the ORDER BY.

SELECT LoanID, Principal, Tenor, AmortizingPrincipal
      ,SUM(AmortizingPrincipal) OVER(PARTITION BY LoanID ORDER BY tenor
                                     RANGE BETWEEN CURRENT ROW
                                       AND UNBOUNDED FOLLOWING) AS BalancePrincipal
  FROM loan
  ORDER BY LoanID, Principal, Tenor

Upvotes: 0

Damien_The_Unbeliever
Damien_The_Unbeliever

Reputation: 239824

If you're using SQL Server 2012, you'd be looking to specify ROWS/RANGE in your OVER:

Further limits the rows within the partition by specifying start and end points within the partition. This is done by specifying a range of rows with respect to the current row either by logical association or physical association. Physical association is achieved by using the ROWS clause.

Other database systems may have similar features. This feature is new in the 2012 version of SQL Server.

Upvotes: 6

Related Questions