Question

I have the following problem to solve and I can’t seem to be able to come up with an algorithm yet, nevermind an actual solution.

I have a table of similar structure/data as the following, where IDs are not always in sequence for the same Ticker/QuouteType:

ID      Ticker PriceDateTime    QuoteType OpenPrice HighPrice LowPrice ClosePrice
------- ------ ---------------- --------- --------- --------- -------- ----------
2036430 ^COMP  2012-02-10 20:50 95/Minute 2901.57   2905.04   2895.37  2901.71
2036429 ^COMP  2012-02-10 19:15 95/Minute 2909.63   2910.98   2899.95  2901.67
2036428 ^COMP  2012-02-10 17:40 95/Minute 2905.9    2910.27   2904.29  2909.64
2036427 ^COMP  2012-02-10 16:05 95/Minute 2902      2908.29   2895.1   2905.89
2036426 ^COMP  2012-02-09 21:00 95/Minute 2926.12   2928.01   2925.53  2927.21

The information I need to extract from this data is the following:

• How many consecutive rows are there? Counting downwards from the most recent (as recorded in PriceDateTime), looking at ClosePrice?

IE: For the current example the answer should be 2. ClosePrice (row 1) = 2901.71 which is greater than ClosePrice (row 2) = 2901.67 but lower than ClosePrice (row 3) = 2909.64. As such, looking back from the most recent price, we have 2 rows that “go in the same direction”.

Of course I have to do this across a lot of other names, so speed is quite important.

PS: Thank you all for your help, I’ve drawn inspiration from all your answers when building the final procedure. You’re all very kind!

Answer 1

Try this: (I have simplified the test data I’m using as it only requires 2 columns to demonstrate the logic).

CREATE TABLE #Test (PriceDateTime DATETIME, ClosePrice DECIMAL(6, 2))
INSERT #Test VALUES 
('20120210 20:50:00.000', 2901.71),
('20120210 19:15:00.000', 2901.67),
('20120210 17:40:00.000', 2900.64),
('20120210 16:05:00.000', 2905.89),
('20120209 21:00:00.000', 2927.21)

-- FIRST CTE, JUST DEFINES A VIEW GIVING EACH ENTRY A ROW NUMBER
;WITH CTE AS
(   SELECT  *,
            ROW_NUMBER() OVER(ORDER BY PriceDateTime DESC) [RowNumber]
    FROM    #Test
), 
-- SECOND CTE, ASSIGNES EACH ENTRY +1 OR -1 DEPENDING ON HOW THE VALUE HAS CHANGED COMPARED TO THE PREVIOUS RECORD
CTE2 AS
(   SELECT  a.*, SIGN(a.ClosePrice - b.ClosePrice) [Movement]
    FROM    CTE a
            LEFT JOIN CTE b
                ON a.RowNumber = b.RowNumber - 1
), 
-- THIRD CTE, WILL LOOP THROUGH THE DATA AS MANY TIMES AS POSSIBLE WHILE THE PREVIOUS ENTRY HAS THE SAME "MOVEMENT"
CTE3 AS
(   SELECT  *, 1 [Recursion]
    FROM    CTE2
    UNION ALL
    SELECT  a.PriceDateTime, a.ClosePrice, a.RowNumber, a.Movement, b.Recursion + 1
    FROM    CTE2 a
            INNER JOIN CTE3 b
                ON a.RowNumber = b.RowNumber - 1
                AND a.Movement = b.Movement
)

SELECT  MAX(Recursion) + 1 -- ADD 1 TO THE RECORD BECAUSE THERE WILL ALWAYS BE AT LEAST TWO ROWS
FROM    CTE3
WHERE   RowNumber = 1 -- LATEST ENTRY

DROP TABLE #Test

I’ve tried to comment the answer to explain as I go. If anything is not clear from the comments let me know and I will try and explain further