Group timestamped record by 5, 10, 15 minutes block

Question

I have minute to minute financial records stored in similar format in my table,

         dt          |   open   |   high   |   low    |  close   |  vol  
---------------------+----------+----------+----------+----------+-------
 2018-05-04 15:30:00 | 171.0000 | 171.3000 | 170.9000 | 171.0000 | 42817
 2018-05-04 15:29:00 | 170.8000 | 171.0000 | 170.8000 | 170.9500 | 32801
 2018-05-04 15:28:00 | 170.8500 | 171.0000 | 170.8000 | 170.8000 | 22991
 2018-05-04 15:27:00 | 170.8500 | 170.8500 | 170.7500 | 170.8000 | 40283
 2018-05-04 15:26:00 | 170.9500 | 171.0000 | 170.8000 | 170.8500 | 46636

and so on.

I want to group them into blocks of 5 minutes, 10 minutes, 60 minutes just like candlesticks. Using date_trunc('hour', dt) is not possible as I want to group them as block of last 60 minutes, last 15 minutes etc.

I am using PostgreSQL.

Answer 1

You should use a GROUP BY with :

floor(extract('epoch' from dt) / 300)

to have your data grouped in 5 minutes intervals. 300 is the number of seconds in 5 minutes. Thus if you want 10 minutes, you'd divide by 600. If you want 1 hour, by 3600.

If you want your interval to begin at 00 05 10, use floor() . If you want them to finish at 00, 05, 10, use ceil()

In the SELECT clause, you should re-transform the Unix epoch used in the GROUP BY into a timestamp using

to_timestamp(floor((extract('epoch' from dt) / 300)) * 300)  as ts

Its not clear if you want all the "block" results in the same query, I assumed yes if you want a candlestick graph. I have also logically deduced the right aggregate function (MIN, MAX, AVG, SUM) for each column, following their names . You might have to adapt this.

Here we go :

 SELECT '5 minutes' as block,
        to_timestamp(floor((extract('epoch' from dt) / 300)) * 300)  as ts, 
        round(AVG(open),4) as avg_open,  
        round(MAX(high),4) as max_high, 
        round(MIN(low),4) as min_low, 
        round(AVG(close),4) as avg_close,  
        SUM(vol) as sum_vol  
 FROM mytable
 GROUP BY floor(extract('epoch' from dt) / 300)

 UNION ALL

  SELECT '10 minutes' as block,
        to_timestamp(floor((extract('epoch' from dt) / 600)) * 600)  as ts, 
        round(AVG(open),4) as avg_open,  
        round(MAX(high),4) as max_high, 
        round(MIN(low),4) as min_low, 
        round(AVG(close),4) as avg_close,  
        SUM(vol) as sum_vol  
 FROM mytable
 GROUP BY floor(extract('epoch' from dt) / 600)

  UNION ALL

  SELECT '1 hour' as block,
        to_timestamp(floor((extract('epoch' from dt) / 3600)) * 3600)  as ts, 
        round(AVG(open),4) as avg_open,  
        round(MAX(high),4) as max_high, 
        round(MIN(low),4) as min_low, 
        round(AVG(close),4) as avg_close,  
        SUM(vol) as sum_vol  
 FROM mytable
 GROUP BY floor(extract('epoch' from dt) / 3600)

Results:

    block       ts                  avg_open    max_high    min_low     avg_close   sum_vol
    5 minutes   04.05.2018 17:30:00 171         171,3       170,9       171         42817
    5 minutes   04.05.2018 17:25:00 170,8625    171         170,75      170,85      142711
    10 minutes  04.05.2018 17:20:00 170,8625    171         170,75      170,85      142711
    10 minutes  04.05.2018 17:30:00 171         171,3       170,9       171         42817
    1 hour      04.05.2018 17:00:00 170,89      171,3       170,75      170,88      185528

Test it on REXTESTER

Answer 2

You can use generate_series() to create any range you want

SQL DEMO:

SELECT dd as start_range, dd + '30 min'::interval as end_range
FROM generate_series
        ( '2018-05-05'::timestamp 
        , '2018-05-06'::timestamp
        , '30 min'::interval) dd
        ;

Then check if your record fall on that range.