从自联接计算

Question

我有一个股票代码符号的值和日期列表，并想在 SQL 中计算季度回报。

CREATE TABLE `symbol_details` (
  `symbol_header_id` INT(11) DEFAULT NULL,
  `DATE` DATETIME DEFAULT NULL,
  `NAV` DOUBLE DEFAULT NULL,
  `ADJ_NAV` DOUBLE DEFAULT NULL)

对于可以正常工作的固定季度开始和结束日期：

set @quarterstart='2008-12-31';
set @quarterend='2009-3-31';

select sha, (100*(aend-abegin)/abegin) as q1_returns from
(select symbol_header_id as sha, ADJ_NAV as abegin from symbol_details
  where date=@quarterstart) as a,
(select  symbol_header_id as she, ADJ_NAV as aend from symbol_details
  where date=@quarterend) as b
where sha=she;

这会计算所有品种的所有季度回报。有时季度末是非交易日，或者股票停止营业，所以我想获得最接近季度开始和结束日期的开始和结束日期。

解决方案是通过某种 GROUP BY 语句以某种方式仅获取每个 symbol_header_id 的一个开始和一个结束季度值，例如 (1)

SET @quarterstart = '2009-03-01';
SET @quarterend = '2009-4-31';
SELECT  symbol_header_id, DATE, ADJ_NAV AS aend FROM symbol_details
WHERE
 DATE BETWEEN @quarterstart AND @quarterend
 AND symbol_header_id BETWEEN 18540 AND 18550
GROUP BY symbol_header_id asc;

这为每个 symbol_header_id 值提供了最接近季度开始日期的 ADJ_NAV 值。

最后，(2)

SET @quarterstart = '2008-12-31';
SET @quarterend = '2009-3-31';

SELECT sh1, a.date, b.date, aend, abegin, 
       (100*(aend-abegin)/abegin) AS quarter_returns FROM
(SELECT  symbol_header_id sh1, DATE, ADJ_NAV AS abegin FROM symbol_details
WHERE
 DATE BETWEEN @quarterstart AND @quarterend
 GROUP BY symbol_header_id DESC) a,

(SELECT  symbol_header_id sh2, DATE, ADJ_NAV AS aend FROM symbol_details
WHERE
 DATE BETWEEN @quarterstart AND @quarterend
 GROUP BY symbol_header_id ASC) b
WHERE sh1 = sh2;

应该计算每个品种的季度收益。

不幸的是，这不起作用。出于某种原因，当我像 (1) 那样限制 ID 时，使用了正确的开始日期和结束日期，但是当删除“AND symbol_header_id BETWEEN 18540 AND 18550”语句时，会出现相同的开始日期和结束日期。为什么？？？？

展开 JOIN 的答案是：

SET @quarterstart = '2008-12-31';
SET @quarterend = '2009-3-31';


SELECT tq.sym                                AS sym,
       (100*(alast.adj_nav - afirst.adj_nav)/afirst.adj_nav) AS quarterly_returns
FROM

  -- First, determine first traded days ("ftd") and last traded days
  -- ("ltd") in this quarter per symbol
  (SELECT symbol_header_id AS sym,
          MIN(DATE)      AS ftd,
          MAX(DATE)      AS ltd
   FROM symbol_details
   WHERE DATE BETWEEN @quarterstart AND @quarterend
   GROUP BY 1) tq

  JOIN symbol_details afirst
  -- Second, determine adjusted NAV for "ftd" per symbol (see WHERE)
   ON afirst.DATE BETWEEN @quarterstart AND @quarterend
   AND afirst.symbol_header_id = tq.sym

  JOIN
  -- Finally, determine adjusted NAV for "ltd" per symbol (see WHERE)
    symbol_details alast
    ON alast.DATE BETWEEN @quarterstart AND @quarterend
    AND alast.symbol_header_id = tq.sym

WHERE
  afirst.date = tq.ftd
  AND
  alast.date  = tq.ltd;

Answer 1

更新：

完全结合@Hogan 的建议......并对其进行测试。 :) 这EXPLAINs 更简单，应该是更好的表现。

再次假设 ANSI_QUOTES 行为：

SELECT tq.sym AS sym,
       (100*(alast.adj_nav - afirst.adj_nav)/afirst.adj_nav) AS quarterly_returns
FROM
  (SELECT symbol_header_id AS sym, -- find first/last traded day ("ftd", "ltd")
          MIN("date") AS ftd,
          MAX("date") AS ltd
   FROM symbol_details
   WHERE "date" BETWEEN @quarterstart AND @quarterend
   GROUP BY 1) tq
JOIN symbol_details afirst         -- JOIN for ADJ_NAV on first traded day
  ON tq.sym = afirst.symbol_header_id
     AND
     tq.ftd = afirst."date"
JOIN symbol_details alast          -- JOIN for ADJ_NAV on last traded day
  ON tq.sym = alast.symbol_header_id
     AND
     tq.ltd = alast."date"

原文：

假设SET SESSION sql_mode = 'ANSI_QUOTES'，试试这个：

SELECT tq.sym                                AS sym,
       (100*(adj_end - adj_begin)/adj_begin) AS quarterly_returns
FROM
  -- First, determine first traded days ("ftd") and last traded days
  -- ("ltd") in this quarter per symbol
  (SELECT symbol_header_id AS sym,
          MIN("date")      AS ftd,
          MAX("date")      AS ltd
   FROM symbol_details
   WHERE "date" BETWEEN @quarterstart AND @quarterend
   GROUP BY 1) tq
  JOIN
  -- Second, determine adjusted NAV for "ftd" per symbol (see WHERE)
  (SELECT symbol_header_id AS sym,
          "date"           AS adate,
          adj_nav          AS adj_begin
   FROM symbol_details) afirst
  ON afirst.sym = tq.sym
  JOIN
  -- Finally, determine adjusted NAV for "ltd" per symbol (see WHERE)
  (SELECT symbol_header_id AS sym,
          "date"           AS adate,
          adj_nav          AS adj_end
   FROM symbol_details) alast
  ON alast.sym = tq.sym
WHERE
  afirst.adate = tq.ftd
  AND
  alast.adate  = tq.ltd;

Answer 2

从以下查询展开最后一个子查询的示例：

（注意——我没有测试）

SELECT tq.sym                                AS sym,
       (100*(alast.adj_nav - adj_begin)/adj_begin) AS quarterly_returns
FROM
  -- First, determine first traded days ("ftd") and last traded days
  -- ("ltd") in this quarter per symbol
  (SELECT symbol_header_id AS sym,
          MIN("date")      AS ftd,
          MAX("date")      AS ltd
   FROM symbol_details
   WHERE "date" BETWEEN @quarterstart AND @quarterend
   GROUP BY 1) tq
  JOIN
  -- Second, determine adjusted NAV for "ftd" per symbol (see WHERE)
  (SELECT symbol_header_id AS sym,
          "date"           AS adate,
          adj_nav          AS adj_begin
   FROM symbol_details) afirst
  ON afirst.sym = tq.sym
  -- Finally, determine adjusted NAV for "ltd" per symbol (see WHERE)
  LEFT JOIN symbol_details alast ON tq.sym = alast.symbol_header_id AND tg.ltd = alast."date"

WHERE
  afirst.adate = tq.ftd

Answer 3

您确定在查询中使用的是 GROUP BY ... ASC 吗？ GROUP BY 对满足某些条件的行进行分组，使用 ASC/DESC 没有意义。顺便说一句，在您的情况下，您的标准是 symbol_header_id 列的相等性，假设“id”表示它是一个键，有效地使组由单行组成。

无论如何，如果您的组是真正的组，您可能会遇到问题，因为您不仅要选择分组属性和聚合函数，还要选择其他一些属性。在这种情况下，它们的价值是不可预测的。