在Netezza中按滚动日期间隔进行分组

Question

I have a table in Netezza that looks like this

Date         Stock    Return
2015-01-01   A        xxx
2015-01-02   A        xxx
2015-01-03   A        0
2015-01-04   A        0
2015-01-05   A        xxx
2015-01-06   A        xxx
2015-01-07   A        xxx
2015-01-08   A        xxx
2015-01-09   A        xxx
2015-01-10   A        0
2015-01-11   A        0
2015-01-12   A        xxx
2015-01-13   A        xxx
2015-01-14   A        xxx
2015-01-15   A        xxx
2015-01-16   A        xxx
2015-01-17   A        0
2015-01-18   A        0
2015-01-19   A        xxx
2015-01-20   A        xxx

The data represents stock returns for various stocks and dates. what I need to do is group the data by a given interval, and day of that interval. Another difficulty is that weekends the (0s) will have to be discounted (ignoring public holidays). And the start date of the first interval should be an arbitrary date.

For example my out put should look sth like this

Interval    Q01    Q02    Q03    Q04    Q05
1           xxx    xxx    xxx    xxx    xxx
2           xxx    xxx    xxx    xxx    xxx
3           xxx    xxx    xxx    xxx    xxx 
4           xxx    xxx    xxx    xxx    xxx

This output would represent an interval of the length 5 working days, with averaged returns as results, in terms of the raw data from above, start date 1st Jan, 1st Interval includes 1/2/5/6/7 (3 and 4 are weekends and are ignored) Q01 would be the 1st, Q02 the 2nd, Q03 the 5th etc. The second interval goes from 8/9/12/13/14.

What I tried unsuccessfully is using

CEIL(CAST(EXTRACT(DOY FROM DATE) AS FLOAT) / CAST (10 AS FLOAT)) AS interval
EXTRACT(DAY FROM DATE) % 10 AS DAYinInterval

I also tried playing around with rolling counters and for variable starting dates setting my DOY to zero with s.th like this

CEIL(CAST(EXTRACT(DOY FROM DATE) - EXTRACT(DOY FROM 'start-date' AS FLOAT) / CAST (10 AS FLOAT)) AS Interval

The one thing that came closest to what I would expect is this SUM(Number) OVER(PARTITION BY STOCK ORDER BY DATE ASC rows 10 preceding) AS Counter

Unfortunately it goes from 1 to 10 followed by 11s where it should start from 1 to 10 again.

I would love to see how this can get implemented in an elegant way. thanks

Answer 1

I'm not entirely sure I understand the question, but I think I might, so I'm going to take a swing at this with some windowed aggregates and subqueries.

Here's the sample data, plugging in some random non-zero data for weekdays.

    DATE    | STOCK | RETURN
------------+-------+--------
 2015-01-01 | A     |     16
 2015-01-02 | A     |     80
 2015-01-03 | A     |      0
 2015-01-04 | A     |      0
 2015-01-05 | A     |     60
 2015-01-06 | A     |     25
 2015-01-07 | A     |     12
 2015-01-08 | A     |      1
 2015-01-09 | A     |     81
 2015-01-10 | A     |      0
 2015-01-11 | A     |      0
 2015-01-12 | A     |     35
 2015-01-13 | A     |     20
 2015-01-14 | A     |     69
 2015-01-15 | A     |     72
 2015-01-16 | A     |     89
 2015-01-17 | A     |      0
 2015-01-18 | A     |      0
 2015-01-19 | A     |    100
 2015-01-20 | A     |     67
(20 rows)

Here's my swing at it, with embedded comments.

select avg(return),
   date_period,
   day_period
from (
        -- use row_number to generate a sequential value for each DOW,
        -- with a WHERE to filter out the weekends
      select date,
         stock,
         return,
         date_period ,
         row_number() over (partition by date_period order by date asc) day_period
      from (
            -- bin out the entries by date_period using the first_value of the entire set as the starting point
            -- modulo 7
            select date,
               stock,
               return,
               date + (first_value(date) over (order by date asc) - date) % 7 date_period
            from stocks
            where date >= '2015-01-01'
            -- setting the starting period date here
         )
         foo
      where extract (dow from date) not in (1,7)
   )
   foo
group by date_period, day_period
order by date_period asc;

The results:

    AVG     | DATE_PERIOD | DAY_PERIOD
------------+-------------+------------
  16.000000 | 2015-01-01  |          1
  80.000000 | 2015-01-01  |          2
  60.000000 | 2015-01-01  |          3
  25.000000 | 2015-01-01  |          4
  12.000000 | 2015-01-01  |          5
   1.000000 | 2015-01-08  |          1
  81.000000 | 2015-01-08  |          2
  35.000000 | 2015-01-08  |          3
  20.000000 | 2015-01-08  |          4
  69.000000 | 2015-01-08  |          5
  72.000000 | 2015-01-15  |          1
  89.000000 | 2015-01-15  |          2
 100.000000 | 2015-01-15  |          3
  67.000000 | 2015-01-15  |          4
(14 rows)

Changing the starting date to '2015-01-03' to see if it adjusts properly:

...
from stocks
where date >= '2015-01-03'
...

And the results:

   AVG     | DATE_PERIOD | DAY_PERIOD
------------+-------------+------------
  60.000000 | 2015-01-03  |          1
  25.000000 | 2015-01-03  |          2
  12.000000 | 2015-01-03  |          3
   1.000000 | 2015-01-03  |          4
  81.000000 | 2015-01-03  |          5
  35.000000 | 2015-01-10  |          1
  20.000000 | 2015-01-10  |          2
  69.000000 | 2015-01-10  |          3
  72.000000 | 2015-01-10  |          4
  89.000000 | 2015-01-10  |          5
 100.000000 | 2015-01-17  |          1
  67.000000 | 2015-01-17  |          2
(12 rows)