Running Multiplication in T-SQL

Question

GTS Table

CCP months   QUART   YEARS  GTS
----  ------  -----    ----- ---
CCP1    1       1   2015    5
CCP1    2       1   2015    6
CCP1    3       1   2015         


        

Answer 1

Another method that uses the EXP(SUM(LOG())) trick and only window functions for the running total (no recursive CTEs or cursors).

Tested at dbfiddle.uk:

WITH 
  ct AS
  ( SELECT  
        ccp, years, quart, 
        q2 = round(exp(coalesce(sum(log(sum(gts))) 
                                OVER (PARTITION BY ccp 
                                      ORDER BY years, quart 
                                      ROWS BETWEEN UNBOUNDED PRECEDING 
                                               AND 1 PRECEDING)
                               , 0))
                  , 2)     -- round appropriately to your requirements
    FROM gts 
    GROUP BY ccp, years, quart
  )
SELECT  
    g.*, 
    result = g.gts * b.baseline * ct.q2,
    baseline = b.baseline * ct.q2
FROM ct 
  JOIN gts AS g
    ON  ct.ccp   = g.ccp
    AND ct.years = g.years 
    AND ct.quart = g.quart 
  CROSS APPLY
        ( SELECT TOP (1) b.baseline
          FROM baseline AS b
          WHERE b.ccp = ct.ccp
          ORDER BY b.years, b.quart
        ) AS b
  ;

How it works:

• (CREATE tables and INSERT skipped)

• 1, lets group by ccp, year and quart and calculate the sums:

    select 
        ccp, years, quart,
        q1 = sum(gts)
    from gts 
    group by ccp, years, quart ;
GO

ccp  | years | quart | q1       
:--- | ----: | ----: | :--------
CCP1 |  2015 |     1 | 18.000000
CCP1 |  2015 |     2 | 8.000000 
CCP1 |  2015 |     3 | 6.000000 
CCP1 |  2015 |     4 | 9.000000 
CCP1 |  2016 |     1 | 12.000000

• 2, we use the EXP(LOG(SUM()) trick to calculate the running multiplications of these sums. We use BETWEEEN .. AND -1 PRECEDING in the window to skip the current values, as these values are only used for the baselines of the next quart.
  The rounding is to avoid inaccuracies that come from using LOG() and EXP(). You can experiment with using either ROUND() or casting to NUMERIC:

with 
  ct as
  ( select 
        ccp, years, quart,
        q1 = sum(gts)
    from gts 
    group by ccp, years, quart
  )
select 
    ccp,  years, quart, -- months, gts, q1,
    q2 = round(exp(coalesce(sum(log(q1)) 
                            OVER (PARTITION BY ccp 
                                  ORDER BY Years, Quart 
                                  ROWS BETWEEN UNBOUNDED PRECEDING 
                                           AND 1 PRECEDING),0)),2)
from ct ;
GO

ccp  | years | quart |   q2
:--- | ----: | ----: | ---:
CCP1 |  2015 |     1 |    1
CCP1 |  2015 |     2 |   18
CCP1 |  2015 |     3 |  144
CCP1 |  2015 |     4 |  864
CCP1 |  2016 |     1 | 7776

• 3, we combine the two queries in one (no need for that, it just makes the query more compact, you could have 2 CTEs instead) and then join to gts so we can multiply each value with the calculated q2 (which gives us the baseline).
  The CROSS APPLY is merely to get the base baseline for each ccp.
  Note that I change this one slightly, to numeric(22,6) instead of rounding to 2 decimal places. The results are the same with the sample but they may differ if the numbers are bigger or not integer:

with 
  ct as
  ( select 
        ccp, years, quart, 
        q2 = cast(exp(coalesce(sum(log(sum(gts)))
                                OVER (PARTITION BY ccp 
                                      ORDER BY years, quart 
                                      ROWS BETWEEN UNBOUNDED PRECEDING 
                                               AND 1 PRECEDING)
                               , 0.0))
                   as numeric(22,6))           -- round appropriately to your requirements
    from gts 
    group by ccp, years, quart
  )
select 
    g.*, 
    result = g.gts * b.baseline * ct.q2,
    baseline = b.baseline * ct.q2
from ct 
  join gts as g
    on  ct.ccp   = g.ccp
    and ct.years = g.years 
    and ct.quart = g.quart 
  cross apply
        ( select top (1) baseline
          from baseline as b
          where b.ccp = ct.ccp
          order by years, quart
        ) as b
  ;
GO

CCP  | months | QUART | YEARS | GTS      | result        | baseline    
:--- | -----: | ----: | ----: | :------- | :------------ | :-----------
CCP1 |      1 |     1 |  2015 | 5.000000 | 25.000000     | 5.000000    
CCP1 |      2 |     1 |  2015 | 6.000000 | 30.000000     | 5.000000    
CCP1 |      3 |     1 |  2015 | 7.000000 | 35.000000     | 5.000000    
CCP1 |      4 |     2 |  2015 | 4.000000 | 360.000000    | 90.000000   
CCP1 |      5 |     2 |  2015 | 2.000000 | 180.000000    | 90.000000   
CCP1 |      6 |     2 |  2015 | 2.000000 | 180.000000    | 90.000000   
CCP1 |      7 |     3 |  2015 | 3.000000 | 2160.000000   | 720.000000  
CCP1 |      8 |     3 |  2015 | 2.000000 | 1440.000000   | 720.000000  
CCP1 |      9 |     3 |  2015 | 1.000000 | 720.000000    | 720.000000  
CCP1 |     10 |     4 |  2015 | 2.000000 | 8640.000000   | 4320.000000 
CCP1 |     11 |     4 |  2015 | 3.000000 | 12960.000000  | 4320.000000 
CCP1 |     12 |     4 |  2015 | 4.000000 | 17280.000000  | 4320.000000 
CCP1 |      1 |     1 |  2016 | 8.000000 | 311040.000000 | 38880.000000
CCP1 |      2 |     1 |  2016 | 1.000000 | 38880.000000  | 38880.000000
CCP1 |      3 |     1 |  2016 | 3.000000 | 116640.000000 | 38880.000000