SQL Server Rolling Rank - sql

Question

I have a table of stock prices for different dates.
DECLARE #table TABLE (MyDate DATE, Ticker VarChar(6), Price Decimal (6,2))
INSERT INTO #Table
VALUES ('1/1/13' , 'ABC' , '100.00')
,('1/2/13' , 'ABC' , '101.50')
,('1/3/13' , 'ABC' , '99.80')
,('1/4/13' , 'ABC' , '95.50')
,('1/5/13' , 'ABC' , '78.00')
,('1/1/13' , 'JKL' , '34.57')
,('1/2/13' , 'JKL' , '33.99')
,('1/3/13' , 'JKL' , '31.85')
,('1/4/13' , 'JKL' , '30.11')
,('1/5/13' , 'JKL' , '45.00')
,('1/1/13' , 'XYZ' , '11.50')
,('1/2/13' , 'XYZ' , '12.10')
,('1/3/13' , 'XYZ' , '17.15')
,('1/4/13' , 'XYZ' , '14.10')
,('1/5/13' , 'XYZ' , '15.55')
I want to calculate the running 3-day rank (based on today and the two prior days) for each Ticker and for each date.
I can create a static rank, but cannot produce a dynamic rolling rank.
SELECT
*,
RANK() OVER (PARTITION BY Ticker ORDER BY Price DESC) AS Rank_3d
FROM
#table
ORDER BY
Ticker ASC, MyDate DESC
Output:
MyDate Ticker PRICE Rank_3d
-------------------------------
2013-01-05 ABC 78.00 5
2013-01-04 ABC 95.50 4
2013-01-03 ABC 99.80 3
2013-01-02 ABC 101.50 1
2013-01-01 ABC 100.00 2
2013-01-05 JKL 45.00 1
2013-01-04 JKL 30.11 5
2013-01-03 JKL 31.85 4
2013-01-02 JKL 33.99 3
2013-01-01 JKL 34.57 2
2013-01-05 XYZ 15.55 2
2013-01-04 XYZ 14.10 3
2013-01-03 XYZ 17.15 1
2013-01-02 XYZ 12.10 4
2013-01-01 XYZ 11.50 5
My output should look like this:
MyDate Ticker Price Rank_3d
-------------------------------------
1/3/13 ABC 99.80 3
1/4/13 ABC 95.50 3
1/5/13 ABC 78.00 3
1/3/13 JKL 31.85 3
1/4/13 JKL 30.11 3
1/5/13 JKL 45.00 1
1/3/13 XYZ 17.15 1
1/4/13 XYZ 14.10 2
1/5/13 XYZ 15.55 2
What am I missing?

Answer 1

There is no "simple" way to do this with the ranking functions.
For your particular problem, it is not that hard to do with lag(). The logic is a bit tricky for the first two days (the missing values) and when there are ties. But here is one approach:
select t.*,
(case when price = max(price) over (partition by ticker order by date rows between 2 preceding and current row)
then 1
when price = min(price) over (partition by ticker order by date rows between 2 preceding and current row)
then 3
else 2
end) as rank_3day
from #table t;
Even with apply the logic is tricky:
select t.*, t2.rank
from #table t cross apply
(select count(*) as rank
from (select top (3) t2.*
from #table t2
where t2.date <= t.date
order by t2
) t2
where t2.price >= t.price
) t2