Moving Average PER TICKER for each day

Question

I am trying to calculate the, say, 3 day moving average (in reality 30 day) volume for stocks. I'm trying to get the average of the last 3 date entries (rather than today-3 days). I've been trying to do something with rownumber in SQL server 2012 but with no success. Can anyone help out. Below is a template schema, and my rubbish attempt at SQL. I have various incarnations of the below SQL with the group by's but still not working. Many thanks!

select dt_eod, ticker, volume
from
(
  select  dt_eod, ticker, avg(volume)
    row_number() over(partition by dt_eod order by max_close desc) rn
  from mytable
) src
where rn >= 1 and rn <= 3
order by dt_eod

Sample Schema:

CREATE TABLE yourtable  
    ([dt_date] int, [ticker] varchar(1), [volume] int);

INSERT INTO yourtable  
    ([dt_date], [ticker], [volume])
VALUES  
    (20121201, 'A', 5),  
    (20121201, 'B', 7),  
    (20121201, 'C', 6),  
    (20121202, 'A', 10),  
    (20121202, 'B', 8),    
    (20121202, 'C', 7),  
    (20121203, 'A', 10),    
    (20121203, 'B', 87),    
    (20121203, 'C', 74),  
    (20121204, 'A', 10),    
    (20121204, 'B', 86),  
    (20121204, 'C', 67),  
    (20121205, 'A', 100),  
    (20121205, 'B', 84),  
    (20121205, 'C', 70),    
    (20121206, 'A', 258),  
    (20121206, 'B', 864),  
    (20121206, 'C', 740);

Answer 1

Three day average for each row:

with top3Values as
(
  select t.ticker, t.dt_date, top3.volume
  from yourtable t
    outer apply
    (
      select top 3 top3.volume
      from yourtable top3
      where t.ticker = top3.ticker
        and t.dt_date >= top3.dt_date
      order by top3.dt_date desc
    ) top3
)
select ticker, dt_date, ThreeDayVolume = avg(volume)
from top3Values
group by ticker, dt_date
order by ticker, dt_date

SQL Fiddle demo.

Latest value:

with tickers as
(
  select distinct ticker from yourtable
), top3Values as
(
  select t.ticker, top3.volume
  from tickers t
    outer apply
    (
      select top 3 top3.volume
      from yourtable top3
      where t.ticker = top3.ticker
      order by dt_date desc
    ) top3
)
select ticker, ThreeDayVolume = avg(volume)
from top3Values
group by ticker
order by ticker

SQL Fiddle demo.

Realistically you wouldn't need to create the tickers CTE for the second query, as you'd be basing this on a [ticker] table, and you'd probably have some sort of date parameter in the query, but hopefully this will get you on the right track.

Answer 2

You mentioned SQL 2012, which means that you can leverage a much simpler paradigm.

select dt_date, ticker, avg(1.0*volume) over (
    partition by ticker
    order by dt_date
    ROWS BETWEEN 2 preceding and current row
)
from yourtable

I find this much more transparent as to what is actually trying to be accomplished.

Answer 3

You may want to look at yet another technique that is presented here: SQL-Server Moving Averages set-based algorithm with flexible window-periods and no self-joins.

The algorithm is quite speedy (much faster than APPLY and does not degrade in performance like APPLY does as data-points-window expands), easily adaptable to your requirement, works with pre-SQL2012, and overcomes the limitations of SQL-2012's windowing functionality that requires hard-coding of window-width in the OVER/PARTITION-BY clause.

For a stock-market type application with moving price-averages, it is a common requirement to allow a user to vary the number of data-points included in the average (from a UI selection, like allowing a user to choose 7 day, 30 day, 60 day, etc), and SQL-2012's OVER clause cannot handle this variable partition-width requirement without dynamic SQL.