Question

I have high frequency tick data which need to sort from microsecond to 6 seconds interval. The daily tick data start from 09:15:00.000 EST, and end at 15:15:00.000 EST.

This is my temp2 table:

 date1                       | close1 | volume1
-----------------------------+--------+---------
 2010-04-16 09:15:28.010 EST |  10001 |       4
 2010-04-16 09:16:00.020 EST |  10002 |       5
 2010-04-16 09:16:35.030 EST |  10003 |       6
 2010-04-16 09:16:35.040 EST |  10001 |       3
 2010-04-16 15:14:59.050 EST |  10007 |       3
 2010-04-19 09:15:05.050 EST |  10002 |       1
 ...                         |    ... |     ...

(date1 is a VARCHAR2(28); close1 and volume1 are both NUMBERs).

How do I get the following result?

 date2               | close2 | volume2
---------------------+--------+---------
 2010-04-16 09:15:30 |  10001 |       4
 2010-04-16 09:16:06 |  10002 |       5
 2010-04-16 09:16:36 |  10001 |       9
 2010-04-16 15:15:00 |  10007 |       3
 2010-04-19 09:15:06 |  10002 |       1
 ...                 |    ... |     ...

The close2 column use last_value of interval 00, 06, 12, 18, 24, 30, 36, 42, 48, 54. and if microseconds in range 00-05.999, just set close2 column value as last_vlue of the interval, set volume2 column value as sum of volume1 of the interval.

Answer 1

Rather than packing all the logic into a single query — which is possible, but incredibly ugly and painful — I think it makes sense to create a stored-procedure that translates your date1 to your date2:

CREATE OR REPLACE FUNCTION date1_to_date2 (date1 IN VARCHAR2) RETURN varchar2
IS

  ts TIMESTAMP;

BEGIN

  ts := TO_TIMESTAMP(SUBSTR(date1, 1, 19), 'YYYY-MM-DD HH24:MI:SS');

  IF SUBSTR(date1, 21, 3) > 0 THEN
    ts := ts + NUMTODSINTERVAL(1, 'SECOND');
  END IF;

  ts := ts + NUMTODSINTERVAL(MOD(60 - TO_CHAR(ts, 'SS'), 6), 'SECOND');

  RETURN TO_CHAR(ts, 'YYYY-MM-DD HH24:MI:SS');

END date1_to_date2;
/

SHOW ERRORS;

Then the rest, while not exactly straightforward, is nonetheless palatable:

SELECT date2,
       close1 AS close2,
       volume2
  FROM ( SELECT date2,
                close1,
                ROW_NUMBER() OVER (PARTITION BY date2 ORDER BY date1 DESC) AS rn,
                SUM(volume1) OVER (PARTITION BY date2) AS volume2
           FROM ( SELECT date1,
                         date1_to_date2(date1) AS date2,
                         close1,
                         volume1
                    FROM temp2
                )
       )
 WHERE rn = 1
 ORDER BY date2
;

(You can actually do it with just one subquery, by substituting the definition of date2 everywhere — and maybe even without any subquery at all, if the WHERE rn = 1 can be converted into a HAVING clause, which I didn’t try — but I think this is the clearest way.)