avg sale of quarter with previous quarter avg sale

Question

I have a table one in which there are various attribute like region product,year,qtr,month,sale. I have to calculate the avg_qtr sale of each product having same region and

Answer 1

You can use the windowing clause of an analytic function if you have a single ordered value to sort by, so first create a DENSE_RANKing of year and qtr, then use that ranking in your analytic functions:

with t1 as ( 
  select one.*
       , dense_rank() over (order by year, qtr) qord
    from one
)
select product
     , year
     , qtr
     , month
     , sales
     , round(avg(sales) over (partition by qord),2) qtr_avg
     , round(avg(sales) over (order by qord
                              range between 1 preceding
                                        and 1 preceding),2) prev_qtr_avg
  from t1

The above solution assumes dense quarterly data as provided in the sample data set, if however, the data is sparse along the quarter dimension you can first densify the data as in this query:

with qtrs as (select level qtr from dual connect by level <=4)
, t1 as ( 
  select product
       , year
       , qtrs.qtr
       , month
       , sales
       , dense_rank() over (order by year, qtrs.qtr) qord
    from qtrs
    left outer join one partition by (year)
      on one.qtr = qtrs.qtr
)
select product
     , year
     , qtr
     , month
     , sales
     , round(avg(sales) over (partition by qord),2) qtr_avg
     , round(avg(sales) over (order by qord
                              range between 1 preceding
                                        and 1 preceding),2) prev_qtr_avg
  from t1

This ensures that for every year represented in the data at least one row will exist for each quarter, and consequently QORD will enumerate every quarter, and gaps in the data will result in gaps in the calculated quarterly averages.

You can also achieve a similar effect by altering the way QORD is calculated by exploiting the numeric natures of YEAR and QTR as in this example:

with t1 as (select one.*, year*4+qtr qord from one)
select product
     , year
     , qtr
     , month
     , sales
     , round(avg(sales) over (partition by qord),2) qtr_avg
     , round(avg(sales) over (order by qord
                              range between 1 preceding
                                        and 1 preceding),2) prev_qtr_avg
  from t1

Here no densification was required, and yet it still correctly leaves gaps in the prev_qtr_avg, but it does leave out records for missing quarters which the densified data includes.

Combining the last two examples, and adding in your new requirement for regions a at least one row of data per quarter will be returned or generated if required for every distinct region, product and year. Both averages are partitioned by region and product and calculated per current or previous quarter as the case may be:

with qtrs(qtr) as (select level from dual connect by level <= 4)
, t1 as (
select region, product, year, q.qtr, month, sales, year*4+q.qtr qord
  from qtrs q
  left join one partition by (region, product, year)
    on q.qtr = one.qtr
)
select region
     , product
     , year
     , qtr
     , month
     , sales
     , round(avg(sales) over (partition by region, product, qord),2) avg_sale
     , round(avg(sales) over (partition by region, product
                              order by qord
                              range between 1 preceding
                                        and 1 preceding),2) prev_avg_sale
  from t1
 order by year, region, qtr, product;