使用 Oracle SQL 进行带有一些逻辑网络的汇总查询

我有一个如下表“AuctionResults”

Auction  Action    Shares  ProfitperShare   
------------------------------------------- 
Round1   BUY        6      200    
Round2   BUY        5      100   
Round2   SELL      -2      50   
Round3   SELL      -5      80

现在，我需要按照“净值先到先得”的原则，在后续几轮中扣除卖出后，对每次拍卖的买入结果进行汇总

因此，在第 1 轮中，我买入 6 股，然后在第 2 轮中卖出 2 股，在第 3 轮中卖出“4”股，总净利润为 6 * 200-2 * 50-4 * 80 = 780

在第 2 轮中，我购买了 5 股，并在第 3 轮中卖出了“1”（因为之前的“4”属于第 1 轮），净利润为 5 * 100-1 * 80 = 420

...所以结果输出应该如下所示：

Auction   NetProfit
------------------
Round1    780    
Round2    420   

我们可以仅使用 Oracle SQL(10g) 而不是 PL-SQL 来完成此操作吗

提前致谢

我知道这是一个老问题，对原始海报没有用，但我想尝试一下，因为这是一个有趣的问题。我没有对其进行足够的测试，所以我认为这仍然需要纠正和调整。但我相信这种做法是合理的。我不建议在产品中使用这样的查询，因为它很难维护或理解（而且我不相信这真的是可扩展的）。创建一些替代数据结构会更好。话虽如此，这就是我在 Postgresql 9.1 中运行的：

    WITH x AS (
        SELECT round, action
              ,ABS(shares) AS shares
              ,profitpershare
              ,COALESCE( SUM(shares) OVER(ORDER BY round, action
                                          ROWS BETWEEN UNBOUNDED PRECEDING 
                                                   AND 1 PRECEDING)
                        , 0) AS previous_net_shares
              ,COALESCE( ABS( SUM(CASE WHEN action = 'SELL' THEN shares ELSE 0 END)
                            OVER(ORDER BY round, action
                                     ROWS BETWEEN UNBOUNDED PRECEDING 
                                              AND 1 PRECEDING) ), 0 ) AS previous_sells
          FROM AuctionResults
          ORDER BY 1,2
    )

    SELECT round, shares * profitpershare - deduction AS net
      FROM (

           SELECT buy.round, buy.shares, buy.profitpershare
                 ,SUM( LEAST( LEAST( sell.shares, GREATEST(buy.shares - (sell.previous_sells - buy.previous_sells), 0)
                                    ,GREATEST(sell.shares + (sell.previous_sells - buy.previous_sells) - buy.previous_net_shares, 0)
                                   )
                             ) * sell.profitpershare ) AS deduction
             FROM x buy
                 ,x sell
             WHERE sell.round > buy.round
               AND buy.action = 'BUY'
               AND sell.action = 'SELL'
             GROUP BY buy.round, buy.shares, buy.profitpershare

           ) AS y

结果：

     round | net
    -------+-----
         1 | 780
         2 | 420
    (2 rows)

为了将其分解为几部分，我从这个数据集开始：

    CREATE TABLE AuctionResults( round int, action varchar(4), shares int, profitpershare int);

    INSERT INTO AuctionResults VALUES(1, 'BUY', 6, 200);
    INSERT INTO AuctionResults VALUES(2, 'BUY', 5, 100);
    INSERT INTO AuctionResults VALUES(2, 'SELL',-2, 50);
    INSERT INTO AuctionResults VALUES(3, 'SELL',-5, 80);
    INSERT INTO AuctionResults VALUES(4, 'SELL', -4, 150);  

    select * from auctionresults;

     round | action | shares | profitpershare
    -------+--------+--------+----------------
         1 | BUY    |      6 |            200
         2 | BUY    |      5 |            100
         2 | SELL   |     -2 |             50
         3 | SELL   |     -5 |             80
         4 | SELL   |     -4 |            150
    (5 rows)

“WITH”子句中的查询将一些运行总计添加到表中。

• “previous_net_shares”表示在当前记录之前有多少股可供出售。这也告诉我在开始将其分配给“买入”之前需要跳过多少“卖出”股票。

• “previous_sells”是遇到的“SELL”股票数量的运行计数，因此两个“previous_sells”之间的差异表示当时使用的“SELL”股票数量。

   round | action | shares | profitpershare | previous_net_shares | previous_sells
  -------+--------+--------+----------------+---------------------+----------------
       1 | BUY    |      6 |            200 |                   0 |              0
       2 | BUY    |      5 |            100 |                   6 |              0
       2 | SELL   |      2 |             50 |                  11 |              0
       3 | SELL   |      5 |             80 |                   9 |              2
       4 | SELL   |      4 |            150 |                   4 |              7
  (5 rows)
  

通过此表，我们可以进行自连接，其中每个“买入”记录都与每个未来的“卖出”记录相关联。结果如下：

    SELECT buy.round, buy.shares, buy.profitpershare
          ,sell.round AS sellRound, sell.shares AS sellShares, sell.profitpershare AS sellProfitpershare
      FROM x buy
          ,x sell
      WHERE sell.round > buy.round
        AND buy.action = 'BUY'
        AND sell.action = 'SELL'

     round | shares | profitpershare | sellround | sellshares | sellprofitpershare
    -------+--------+----------------+-----------+------------+--------------------
         1 |      6 |            200 |         2 |          2 |                 50
         1 |      6 |            200 |         3 |          5 |                 80
         1 |      6 |            200 |         4 |          4 |                150
         2 |      5 |            100 |         3 |          5 |                 80
         2 |      5 |            100 |         4 |          4 |                150
    (5 rows)

然后是疯狂的部分，试图计算订单中可出售的股票数量与尚未出售的股票数量。以下是一些有助于遵循这一点的注释。带有“0”的“最大”电话只是说，如果我们处于负数，我们就无法分配任何股票。

   -- allocated sells 
   sell.previous_sells - buy.previous_sells

   -- shares yet to sell for this buy, if < 0 then 0
   GREATEST(buy.shares - (sell.previous_sells - buy.previous_sells), 0)

   -- number of sell shares that need to be skipped
   buy.previous_net_shares

感谢大卫的协助 https://stackoverflow.com/questions/8813743/can-multiple-rows-within-a-window-be-referenced-by-an-analytic-function