Calculate forecast average using recursive CTE

问题

I was trying to answer a question here, where I need to calculate a forecast of sales based on the 3 previous months which either can be actuals or forecast.

Month   Actuals Forecast  
1       10    
2       15    
3       17    
4                14.00 
5                15.33  
6                15.44 
7                14.93

Month 4 = (10+15+17)/3 
Month 5 = (15+17+14)/3 
Month 6 = (17+14+15.33)/3
Month 7 = (14+15.33+15.44)/3

I've been trying to do this using a recursive CTE:

;WITH cte([month],forecast) AS (
    SELECT 1,CAST(10 AS DECIMAL(28,2))
    UNION ALL
    SELECT 2,CAST(15 AS DECIMAL(28,2))
    UNION ALL 
    SELECT 3,CAST(17 AS DECIMAL(28,2))
    UNION ALL
    SELECT
        [month]=[month]+1,
        forecast=CAST(AVG(forecast) OVER (ORDER BY [month] ROWS BETWEEN 3 PRECEDING AND 1 PRECEDING) AS DECIMAL(28,2))
    FROM
        cte
    WHERE
        [month]<=12
)
SELECT * FROM cte WHERE month<=12;

Fiddle: http://sqlfiddle.com/#!6/9ac4a/3

But it doesn't work as expected, as It returns the following result:

| month | forecast |
|-------|----------|
|     1 |       10 |
|     2 |       15 |
|     3 |       17 |
|     4 |   (null) |
|     5 |   (null) |
|     6 |   (null) |
|     7 |   (null) |
|     8 |   (null) |
|     9 |   (null) |
|    10 |   (null) |
|    11 |   (null) |
|    12 |   (null) |
|     3 |   (null) |
|     4 |   (null) |
|     5 |   (null) |
|     6 |   (null) |
|     7 |   (null) |
|     8 |   (null) |
|     9 |   (null) |
|    10 |   (null) |
|    11 |   (null) |
|    12 |   (null) |
|     2 |   (null) |
|     3 |   (null) |
|     4 |   (null) |
|     5 |   (null) |
|     6 |   (null) |
|     7 |   (null) |
|     8 |   (null) |
|     9 |   (null) |
|    10 |   (null) |
|    11 |   (null) |
|    12 |   (null) |

Expected output:

| month | forecast |
|-------|----------|
|     1 |       10 |
|     2 |       15 |
|     3 |       17 |
|     4 |    14.00 |
|     5 |    15.33 |
|     6 |    15.44 |
|     7 |    14.93 |
|     8 |    15.23 |
|     9 |    15.20 |
|    10 |    15.12 |
|    11 |    15.18 |
|    12 |    15.17 |

Can someone tell me what's wrong with this query?

回答1:

I propose something like this:

WITH T AS
(
    SELECT 1 AS [month], CAST(10 AS DECIMAL(28,2)) AS [forecast], CAST(-5 AS DECIMAL(28,2)) AS three_months_ago_forecast, CAST(9 AS decimal(28,2)) AS two_months_ago_forecast, CAST(26 AS decimal(28,2)) as one_month_ago_forecast
    UNION ALL
    SELECT 2,CAST(15 AS DECIMAL(28,2)), CAST(9 AS decimal(28,2)), CAST(26 AS decimal(28,2)), CAST(10 AS DECIMAL(28,2))
    UNION ALL 
    SELECT 3,CAST(17 AS DECIMAL(28,2)), CAST(26 AS decimal(28,2)), CAST(10 AS DECIMAL(28,2)), CAST(15 AS DECIMAL(28,2))
),
LT AS -- LastForecast
(
    SELECT *
    FROM T
    WHERE [month] = 3
),
FF AS -- Future Forecast
(
    SELECT *
    FROM LT

    UNION ALL

    SELECT 
        FF.[month] + 1 AS [month], 
        CAST( (FF.forecast * 4 - FF.three_months_ago_forecast) / 3 AS decimal(28,2)) AS forecast,
        FF.two_months_ago_forecast as three_months_ago_forecast,
        FF.one_month_ago_forecast as two_months_ago_forecast,
        FF.forecast as one_month_ago_forecast
    FROM FF
    WHERE
        FF.[month] < 12

)
SELECT * FROM T
WHERE [month] < 3
UNION ALL
SELECT * FROM FF

Output:

+-------+----------+---------------------------+-------------------------+------------------------+
| month | forecast | three_months_ago_forecast | two_months_ago_forecast | one_month_ago_forecast |
+-------+----------+---------------------------+-------------------------+------------------------+
|     1 | 10.00    | -5.00                     | 9.00                    | 26.00                  |
|     2 | 15.00    | 9.00                      | 26.00                   | 10.00                  |
|     3 | 17.00    | 26.00                     | 10.00                   | 15.00                  |
|     4 | 14.00    | 10.00                     | 15.00                   | 17.00                  |
|     5 | 15.33    | 15.00                     | 17.00                   | 14.00                  |
|     6 | 15.44    | 17.00                     | 14.00                   | 15.33                  |
|     7 | 14.92    | 14.00                     | 15.33                   | 15.44                  |
|     8 | 15.23    | 15.33                     | 15.44                   | 14.92                  |
|     9 | 15.20    | 15.44                     | 14.92                   | 15.23                  |
|    10 | 15.12    | 14.92                     | 15.23                   | 15.20                  |
|    11 | 15.19    | 15.23                     | 15.20                   | 15.12                  |
|    12 | 15.18    | 15.20                     | 15.12                   | 15.19                  |
+-------+----------+---------------------------+-------------------------+------------------------+

回答2:

Try this

WITH cte
     AS (SELECT *
         FROM   (VALUES (1,10,NULL),
                        (2,15,NULL),
                        (3,17,NULL),
                        (4,NULL,14.00),
                        (5,NULL,15.33),
                        (6,NULL,15.44),
                        (7,NULL,14.93)) tc (month, act, fore))
SELECT mon,avg(res) 
FROM   cte a
       CROSS apply (SELECT TOP 3 ( COALESCE(a.act, a.fore) ) AS res,
                                 b.month                     AS mon
                    FROM   cte b
                    WHERE  a.month < b.month
                    ORDER  BY a.month DESC) cs
GROUP  BY mon
ORDER  BY mon

or in Sql Server 2012+ use this

SELECT
    [month]=[month]+1,
    forecast=CAST(AVG(COALESCE(act,fore)) OVER (ORDER BY [month] ROWS BETWEEN 3 PRECEDING  AND CURRENT row  ) AS DECIMAL(28,2))
FROM
    cte

来源：https://stackoverflow.com/questions/35149373/calculate-forecast-average-using-recursive-cte

标签

sql-server

sql-server-2008-r2