CODEWITHSUNDEEP
mysql
Tom Grant
Tom Grant

Reputation: 428

MySQL calculate moving average of N rows

I'm trying to calculate the moving average of N rows, for all rows in a single query. In the example case, I am attempting to calculate the moving average of 50 rows.

SELECT 
    h1.date, 
    h1.security_id, 
    (   SELECT 
            AVG(last50.close)
        FROM (
            SELECT h.close
            FROM history as h
            WHERE h.date <= h1.date AND h.security_id = h1.security_id
            ORDER BY h.date DESC
            LIMIT 50
        ) as last50
    ) as avg50
FROM history as h1

However, MySQL gives me an error when running this query:

Unknown column 'h1.date' in 'where clause'

I'm trying this method because the other solutions listed don't really seem to work for my use case. There are solutions for a moving average of N days, but since all dates are not accounted for in my data set, I need the average of N rows.

This solution, shown below, doesn't work because AVG (also SUM and COUNT) doesn't account for LIMIT:

SELECT
  t1.data_date
  ( SELECT SUM(t2.price) / COUNT(t2.price)
    FROM t as t2
    WHERE t2.data_date <= t1.data_date
    ORDER BY t2.data_date DESC
    LIMIT 5
  ) AS 'five_row_moving_average_price'
FROM t AS t1
ORDER BY t1.data_date;

This question looks promising, but is somewhat indecipherable to me.

Any suggestions? Here's an SQLFiddle to play around in.

Upvotes: 2

Views: 6174

Answers (2)

vahbuna
vahbuna

Reputation: 151

In mysql 8 window function frame can be used to obtain the averages.

SELECT date, security_id, AVG(close) OVER (PARTITION BY security_id ORDER BY date ROWS 49 PRECEDING) as ma
FROM history
ORDER BY date DESC

This calculates the average of the current row and 49 preceding rows.

Upvotes: 2

amdixon
amdixon

Reputation: 3833

plan

  • self join history on last 50 days
  • take average grouping by date and security id ( of current )

query

select curr.date, curr.security_id, avg(prev.close)
from history curr
inner join history prev 
on prev.`date` between date_sub(curr.`date`, interval 49 day) and curr.`date`
and curr.security_id = prev.security_id
group by 1, 2
order by 2, 1
;

output

+---------------------------+-------------+--------------------+
|           date            | security_id |  avg(prev.close)   |
+---------------------------+-------------+--------------------+
| January, 04 2016 00:00:00 | 1           | 10.770000457763672 |
| January, 05 2016 00:00:00 | 1           | 10.800000190734863 |
| January, 06 2016 00:00:00 | 1           | 10.673333485921225 |
| January, 07 2016 00:00:00 | 1           | 10.59250020980835  |
| January, 08 2016 00:00:00 | 1           | 10.432000160217285 |
| January, 11 2016 00:00:00 | 1           | 10.40166680018107  |
| January, 12 2016 00:00:00 | 1           | 10.344285828726631 |
| January, 13 2016 00:00:00 | 1           | 10.297500133514404 |
| January, 14 2016 00:00:00 | 1           | 10.2877779006958   |
| January, 04 2016 00:00:00 | 2           | 56.15999984741211  |
| January, 05 2016 00:00:00 | 2           | 56.18499946594238  |
| ..                        | ..          | ..                 |
+---------------------------+-------------+--------------------+

sqlfiddle

reference

modified to use last 50 rows

select
rnk_curr.`date`, rnk_curr.security_id, avg(rnk_prev50.close)
from
(
select `date`, security_id,
@row_num := if(@lag = security_id, @row_num + 1,
               if(@lag := security_id, 1, 1)) as row_num
from history 
cross join ( select @row_num := 1, @lag := null ) params
order by security_id, `date`
) rnk_curr
inner join
(
select date, security_id, close,
@row_num := if(@lag = security_id, @row_num + 1,
               if(@lag := security_id, 1, 1)) as row_num
from history 
cross join ( select @row_num := 1, @lag := null ) params
order by security_id, `date`
) rnk_prev50
on  rnk_curr.security_id = rnk_prev50.security_id
and rnk_prev50.row_num between rnk_curr.row_num - 49 and rnk_curr.row_num
group by 1,2
order by 2,1
;

sqlfiddle

note

the if function is to force the correct order of evaluation of variables.

Upvotes: 4

Related Questions