Question

I have a table something like this.

    
    count | date
    ------------------
     1     2012-01-01   
     4     2012-01-01   
     5     2012-01-02
    12     2012-01-03
     7     2012-01-04
     4     2012-01-05
    19     2012-01-06
     1     2012-01-07
     etc...

I’m looking for a way to calculate the average count per week over the previous 4 week period for each week.

The results should be something like…

    
    avg | yearweek
    ------------------
     3     201201   
     5     201202   
     6     201203
     1     201204
     11    201205
     3     201206
    18     201207
    12     201208
     etc...

…where each yearweek is the weekly average over the past 4 yearweeks.

Getting the weekly averages is simple enough but how do I then get that over the past 4 yearweeks? And then how to do I do that as a rolling average? Am I better off just doing this in code?

Answer 1

While you could certainly do this in the code of your application, if you really need to do it in SQL, you could first create a table of results aggregated by week and then join it to itself to get the 4-week moving average.

In doing so, instead of storing the averages, I would store the sums and the number of days (1st or last week of year might not have 7 days – thinking of the edge cases). That way, you would avoid calculating unweighted averages when the denominators of averages are different.

So let’s say you have a table “weekly_results”, which has fields: yearweek, sumcount, numdays. You can now self-join to the last 4 weeks and get the sums and counts, and then calculate the averages from that:

SELECT yearweek, sum_cnt/sum_dys as avg_moving_4wk
FROM (
    SELECT a.yearweek, sum(b.sumcount) as sum_cnt, sum(b.numdays) as sum_dys
    FROM weekly_results a
        join weekly_results b
        on a.yearweek - b.yearweek <4 and a.yearweek - b.yearweek >=0
    GROUP BY a.yearweek
) t1
GROUP BY yearweek