Query to filter out transactions with exact offset amount

The query below is a rough draft containing the relevant columns in the query I'd like to write, so don't look at it as a solution. Use it as a guide for the table and column names. I'm trying to remove any transactions that offset each other for the same ORDER_ID and ACCOUNT_ID. I don't think I can do an aggregate using SUM, since that would add together all TX_AMOUNT values for the grouping. See TX_ID 6 and 7. Those both need to show in the result set. How can I output the TX_ID from the table below, and filter out anything that doesn't say "SHOW THIS"?

SELECT 
T1.ACCOUNT_ID
T1.ORDER_ID,
T1.TX_ID
FROM TRANSACTION AS T1
WHERE
T1.ACCOUNT_ID IN (
SELECT T2.ACCOUNT_ID
FROM TRANSACTION AS T2
GROUP BY T2.ACCOUNT_ID, T2.ORDER_ID
HAVING SUM(T2.TX_AMOUNT) != 0 AND T2.ORDER_ID IS NOT NULL 
)
AND T1.ORDER_ID IN (
SELECT T3.ORDER_ID
FROM TRANSACTION AS T3
GROUP BY T3.ACCOUNT_ID, T3.ORDER_ID
HAVING SUM(T3.TX_AMOUNT) != 0 AND T3.ORDER_ID IS NOT NULL 
)

TX_ID   ORDER_ID ACCOUNT_ID  TX_AMOUNT
------------------------------------
1       A1       200         -3.00  <--------- DON'T SHOW THIS; OFFSET BY #2
2       A1       200         3.00   <--------- DON'T SHOW THIS; OFFSET BY #1
3       A1       200         3.00   <--------- SHOW THIS
4       A2       999         -10.01 <--------- DON'T SHOW THIS; OFFSET BY #5
5       A2       999         10.01  <--------- DON'T SHOW THIS; OFFSET BY #4
6       A2       999         10.01  <--------- SHOW THIS
7       A2       999         5.02   <--------- SHOW THIS

Solution

VERSION 2: MUCH cleaner... Working DEMO with comments (you may need to click Run it!) to see desired results (or maybe I have a caching problem)

The CTE (Common Table Expression) is just your data setup that you show
The CTE2 simply adds a row number partitioned by the tx_amount, order_Id, account_Id. Key here is the fact that we get a row_number for each order_ID, Account_ID and tax_Amount restarting when those 3 values change but incrementing when they stay the same. This later allows us to exclude like matches on opposite tx_amounts without eliminating those when one side has more than the other (your $3.00 example)
The Select simply pulls in records from the base set where the opposite value with the same row number order_id and account exist. if one doesn't then we know it's a value w/o a matching opposite tx_Amount and thus one we want to keep.
Ask if you have questions! Happy to help if something's not clear
Lastly if we change CTE2 so the rowNumber() is ordered by tx_ID asc instead of tx_Amount desc (which really serves no purpose other than I needed an order by on the row_nubmer), then we'd get rid of the lowest number matches first consistently following the FIFO approach)

With CTE (TX_ID,   ORDER_ID, ACCOUNT_ID,  TX_AMOUNT) as (

SELECT 1,       'A1',       200,         -3.00  UNION ALL
SELECT 2,       'A1',       200,         3.00   UNION ALL
SELECT 3,       'A1',       200,         3.00   UNION ALL
SELECT 4,       'A2',       999,         -10.01 UNION ALL
SELECT 5,       'A2',       999,         10.01  UNION ALL
SELECT 6,       'A2',       999,         10.01  UNION ALL
SELECT 7,       'A2',       999,         5.02 ),

cte2 as (
SELECT A.*, row_number() over (partition by order_ID, Account_ID, Tx_Amount order by tx_Amount desc) RN
FROM cte A)

SELECT * 
FROM cte2 A
WHERE NOT exists (SELECT * 
                  FROM cte2 B
                  WHERE A.Order_ID = B.Order_ID
                    and A.Account_ID = B.Account_Id
                    and A.tx_Amount*-1 = B.tx_Amount
                    and A.RN = B.RN)

Giving us: (note we should eliminate the RN by changing * to desired fields but i'm too lazy at this point)

+----+-------+----------+------------+-----------+----+
|    | TX_ID | ORDER_ID | ACCOUNT_ID | TX_AMOUNT | RN |
+----+-------+----------+------------+-----------+----+
|  1 |     2 | A1       |        200 |      3,00 |  2 |
|  2 |     7 | A2       |        999 |      5,02 |  1 |
|  3 |     5 | A2       |        999 |     10,01 |  2 |
+----+-------+----------+------------+-----------+----+

VERSION 1: (Scratch this ugly; I mean seriously; who thinks like this?) I do...

Do something. (version 1)
Laugh(an important and overlooked step)
Then do it right (See version 2 above)
Now make it better. (indexes, tweak joins spelling, layout comments, use the right order by on the row_number logic in CTE2)

DEMO

With CTE (TX_ID,   ORDER_ID, ACCOUNT_ID,  TX_AMOUNT) as (

SELECT 1,       'A1',       200,         -3.00  UNION ALL
SELECT 2,       'A1',       200,         3.00   UNION ALL
SELECT 3,       'A1',       200,         3.00   UNION ALL
SELECT 4,       'A2',       999,         -10.01 UNION ALL
SELECT 5,       'A2',       999,         10.01  UNION ALL
SELECT 6,       'A2',       999,         10.01  UNION ALL
SELECT 7,       'A2',       999,         5.02 ),
cte2 as (
SELECT * 
FROM (Select A.Tx_Id aTx_ID
           , A.order_ID as AOrderID
           , A.Account_ID as AAccount_ID
           , A.tx_Amount as ATx_Amount
           , Row_number() over (partition by Order_ID, Account_ID, tx_Amount order by tx_Amount asc) ARN

      from cte a 
      WHERE tx_Amount <=0) A
FULL OUTER JOIN (SELECT b.tx_Id
                      , b.order_Id
                      , b.Account_Id
                      , b.tx_Amount
                      ,  Row_number() over (partition by Order_ID, Account_ID, tx_Amount order by tx_Amount desc) BRN 
                 FROM  CTE B 
                 WHERE  tx_Amount>0) B
  on A.AOrderID = B.Order_ID
 and A.AAccount_ID = B.Account_ID
 and A.ATx_Amount*-1 = B.tx_Amount
 and A.ARN=B.BRN
Where a.Atx_ID is null
  or B.tx_ID is null)

  Select ATX_ID, AORDERID, AAccount_ID, ATX_AMOUNT from cte2 where ATX_ID is not null
  UNION ALL
  Select TX_ID, ORDER_ID, Account_ID, TX_AMOUNT from cte2 where TX_ID is not null