Remove cursur SQL statement - sql-server

Question

I want to remove a cursor in SQL, to increase performance (and because I want to learn how to use best practice and best practice is supposed to be set based, without cursor).
Anyway, I have a temp table that looks like this:
+------------+--------+-------+----+
| Period | Change | Value | NR |
+------------+--------+-------+----+
| 201705 | 7 | 26055 | 1 |
| 201704 | 29 | 0 | 2 |
| 201703 | -92 | 0 | 3 |
| 201702 | -338 | 0 | 4 |
| 201701 | 81 | 0 | 5 |
| 201612 | 107 | 0 | 6 |
| 201611 | 72 | 0 | 7 |
| 201610 | 54 | 0 | 8 |
| 201609 | 64 | 0 | 9 |
| 201608 | 47 | 0 | 10 |
| 201607 | 23 | 0 | 11 |
| 201606 | 45 | 0 | 12 |
+------------+--------+-------+----+
Currently, the Cursor acts as follows:
DECLARE #Value INT
BEGIN
DECLARE c_Value CURSOR FOR
SELECT NR
FROM ##TMP
WHERE Value = 0
----
OPEN c_Value
FETCH NEXT FROM c_Value
INTO #Value
WHILE ##FETCH_STATUS = 0
BEGIN
SELECT #Value = Value - Change
FROM ##TMP
WHERE NR = (Select MAX(NR) From ##TMP WHERE Value <> 0)
BEGIN
UPDATE ##TMP
SET Value = #Value
WHERE NR = (Select MAX(NR)+1 From ##TMP WHERE Value <> 0)
END
FETCH NEXT FROM c_Value
INTO #Value
END
CLOSE c_Value
DEALLOCATE c_Value
END
Result:
+------------+--------+-------+----+
| Period | Change | Value | NR |
+------------+--------+-------+----+
| 201705 | 7 | 26055 | 1 |
| 201704 | 29 | 26048 | 2 |
| 201703 | -92 | 26019 | 3 |
| 201702 | -338 | 26111 | 4 |
| 201701 | 81 | 26449 | 5 |
| 201612 | 107 | 26368 | 6 |
| 201611 | 72 | 26261 | 7 |
| 201610 | 54 | 26189 | 8 |
| 201609 | 64 | 26135 | 9 |
| 201608 | 47 | 26071 | 10 |
| 201607 | 23 | 26024 | 11 |
| 201606 | 45 | 26001 | 12 |
+------------+--------+-------+----+
So, how can I achieve this result, without the use of a cursor? I tried it with a CTE, but I can not get this result.

Answer 1

First you need get the starting value.
SELECT [Value] as StartValue
FROM Table1
WHERE NR = 1
Then using cumulative SUM() you can modify the starting Value, notice you have to ignore the [Change] value for each row
SQL DEMO
WITH CTE as (
SELECT [Value] as StartValue
FROM Table1
WHERE NR = 1
)
SELECT T.*,
- SUM(CHANGE) OVER (ORDER BY [NR])
+ [CHANGE] as TotalChange, -- just for debug, dont need this.
CTE.StartValue
- SUM([CHANGE]) OVER (ORDER BY [NR])
+ [CHANGE] as NewValue
FROM Table1 T
CROSS JOIN CTE
OUTPUT

Answer 2

SQL Server 2012 or higher:
CREATE TABLE ##TMP (
Period int
,Change float
,Value float
,Nr int
);
INSERT INTO ##TMP VALUES
(201705, 7 , 26055, 1)
,(201704, 29 , 0, 2)
,(201703, -92 , 0, 3)
,(201702, -338 , 0, 4)
,(201701, 81 , 0, 5)
,(201612, 107 , 0, 6)
,(201611, 72 , 0, 7)
,(201610, 54 , 0, 8)
,(201609, 64 , 0, 9)
,(201608, 47 , 0,10)
,(201607, 23 , 0,11)
,(201606, 45 , 0,12)
;with cte as (
SELECT Period, Change, value as Value_Org, Nr, SUM(Value - Change) OVER (ORDER BY Nr ASC ) as Value
FROM ##TMP
)
select a.Period, a.Change, a.nr, a.value_org, a.value, b.value,
isnull(b.value, a.value_org)
from cte as a
left outer join cte as b
on a.nr = b.nr+1
order by a.Nr

Answer 3

This can be solved by using the windowing functions introduced in SQL Server 2014.
select period,
change,
NR = Row_Number() Over(Order by period),
Value = Sum(Change) Over(Order by period rows unbounded preceding)
This was freehand and may not parse, but should get you close enough.