Oracle SQL 或 PL/SQL：如何仅在上升趋势或下降趋势结束时识别烛台形态并在列中设置标志？

Question

此问题和相关答案仅用于教育或学习目的。

这个问题与我的其他帖子有很大不同，并且不重复。由于它造成了混乱并且按照@MT0的建议，我将其作为一个新问题发布在这里。

我有下表，我每天上传股票数据。

/* CREATE TABLE */
CREATE TABLE RAW_SOURCE(
  Stock  VARCHAR(100),
  Close_Date DATE,
  Open   NUMBER,
  High   NUMBER,
  Low    NUMBER,
  Close  NUMBER,
  Volume NUMBER
);

/* INSERT QUERY NO: 1 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '01/01/2021', 40, 40.5, 38.5, 38.8, 83057
);

/* INSERT QUERY NO: 2 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '02/01/2021', 39.2, 39.2, 37.2, 37.8, 181814
);

/* INSERT QUERY NO: 3 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '03/01/2021', 38, 38.5, 36.5, 37, 117378
);

/* INSERT QUERY NO: 4 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '04/01/2021', 36.5, 36.6, 35.6, 35.7, 93737
);

/* INSERT QUERY NO: 5 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '05/01/2021', 35.35, 36.8, 35.1, 36.7, 169106
);

/* INSERT QUERY NO: 6 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '06/01/2021', 36.5, 38.5, 36.5, 38, 123179
);

/* INSERT QUERY NO: 7 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '07/01/2021', 37.5, 39.5, 37.3, 39.4, 282986
);

/* INSERT QUERY NO: 8 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '08/01/2021', 39, 40.5, 38.5, 40, 117437
);

/* INSERT QUERY NO: 9 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '09/01/2021', 39.7, 39.8, 39.3, 39.4, 873009
);

/* INSERT QUERY NO: 10 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '10/01/2021', 39.2, 39.2, 37.2, 37.8, 62522
);

/* INSERT QUERY NO: 11 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '11/01/2021', 38, 38.5, 36.5, 37, 114826
);

/* INSERT QUERY NO: 12 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '12/01/2021', 36.5, 37.9, 36.3, 37.8, 281461
);

/* INSERT QUERY NO: 13 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '13/01/2021', 37.5, 39.5, 37.3, 39.4, 77334
);

/* INSERT QUERY NO: 14 */
INSERT INTO RAW_SOURCE(Stock, Close_Date, Open, High, Low, Close, Volume)
VALUES
(
'XYZ', '14/01/2021', 39, 40.5, 38.5, 40, 321684
);

以下是一只股票“XYZ”的样本数据：

+-------+------------+-------+------+------+-------+--------+
| Stock | Close Date | Open  | High | Low  | Close | Volume |
+-------+------------+-------+------+------+-------+--------+
| XYZ   | 01-01-2021 |    40 | 40.5 | 38.5 |  38.8 |  83057 |
| XYZ   | 02-01-2021 |  39.2 | 39.2 | 37.2 |  37.8 | 181814 |
| XYZ   | 03-01-2021 |    38 | 38.5 | 36.5 |    37 | 117378 |
| XYZ   | 04-01-2021 |  36.5 | 36.6 | 35.6 |  35.7 |  93737 |
| XYZ   | 05-01-2021 | 35.35 | 36.8 | 35.1 |  36.7 | 169106 |
| XYZ   | 06-01-2021 |  36.5 | 38.5 | 36.5 |    38 | 123179 |
| XYZ   | 07-01-2021 |  37.5 | 39.5 | 37.3 |  39.4 | 282986 |
| XYZ   | 08-01-2021 |    39 | 40.5 | 38.5 |    40 | 117437 |
| XYZ   | 09-01-2021 |  39.7 | 39.8 | 39.3 |  39.4 | 873009 |
| XYZ   | 10-01-2021 |  39.2 | 39.2 | 37.2 |  37.8 |  62522 |
| XYZ   | 11-01-2021 |    38 | 38.5 | 36.5 |    37 | 114826 |
| XYZ   | 12-01-2021 |  36.5 | 37.9 | 36.3 |  37.8 | 281461 |
| XYZ   | 13-01-2021 |  37.5 | 39.5 | 37.3 |  39.4 |  77334 |
| XYZ   | 14-01-2021 |    39 | 40.5 | 38.5 |    40 | 321684 |
+-------+------------+-------+------+------+-------+--------+

在一段时间内，每个股票代码将有超过数千条记录，我想仅在上涨/上涨趋势的顶部或在下跌/下跌趋势的底部识别烛台模式，但不在横向中（因为这将为假阳性）。以下是示例屏幕截图：

假设今天是2021 年 1 月 12 日，以下是预期输出：

+-------+-------------------+------------+------------+--------------+--------+---------------+
| Stock | Consecutive Count | Start Date |  End Date  | Latest Close | Volume |    Pattern    |
+-------+-------------------+------------+------------+--------------+--------+---------------+
| XYZ   |                 3 | 09-01-2021 | 12-01-2021 |         37.8 | 281461 | Piercing Line |
+-------+-------------------+------------+------------+--------------+--------+---------------+

由于源表将包含许多其他股票，因此如果发现任何模式，也希望显示其他股票在 2021 年 1 月 12 日的结果。我觉得这是一个非常具有挑战性和复杂的逻辑。因此在这里寻求帮助。提前致谢。

更新：谢谢@JustinCave
这是计算公式： 对于看涨吞没：

O1 > C1 and C > O and C > H1 and O < L1
where,

O1 = Previous day Open price  
C1 = Previous day Close price  
C  = Today's Close price  
O  = Today's Open price  
H1 = Previous day High price  
L1 = Previous day Low price 

对于看跌孕妇：

(O1 < C1) and (O > C) and (O < C1) and (C > O1) and (H < H1) and (L > L1)
where,
    O1 = Previous day Open price  
    C1 = Previous day Close price  
    C  = Today's Close price  
    O  = Today's Open price  
    H1 = Previous day High price  
    L1 = Previous day Low price 
    H  = Today's High price  
    L  = Today's Low price 

对于穿孔线：

(O < C) and (O1 > C1) and (C > (C1 + O1)/2) and (O < C1) and (C < O1)
    where,
        O1 = Previous day Open price  
        C1 = Previous day Close price  
        C  = Today's Close price  
        O  = Today's Open price 

Answer 1

模式的MATCH_RECOGNIZE工作方式与正则表达式类似；你想要类似的东西：

（注意：您的PIERCING_LINE公式没有给出预期的输出，所以我假设您想要C > (C1 + O1)/2而不是C > C1 + (O1/2)。）

SELECT *
FROM   raw_source
MATCH_RECOGNIZE (
  PARTITION BY stock
  ORDER BY Close_Date
  MEASURES
    CLASSIFIER() AS pttrn
  ALL ROWS PER MATCH
  PATTERN (
    ^initial_value
    |
    down+ (bullish_engulfing | piercing_line | $)
    |
    up+ (bearish_harami | $)
    |
    other
  )
  DEFINE
    down AS
          PREV(open) > open
      AND PREV(close) > close
      AND PREV(open) > PREV(close)
      AND open > close,
    up AS
          PREV(open) < open
      AND PREV(close) < close
      AND PREV(open) < PREV(close)
      AND open < close,
    bullish_engulfing AS
      -- O1 > C1 and C > O and C > H1 and O < L1
          PREV(open) > PREV(close)
      AND close > open
      AND close > PREV(high)
      AND open  < PREV(low),
    bearish_harami AS
      -- O1 < C1 and O > C and O < C1 and C > O1 and H < H1 and L > L1
          PREV(open) < PREV(close)
      AND open > close
      AND open < PREV(close)
      AND close > PREV(open)
      AND high < PREV(high)
      AND low > PREV(low),
    piercing_line AS
      -- O < C and O1 > C1 and C > (C1 + O1)/2 and O < C1 and C < O1
          open < close
      AND PREV(open) > PREV(close)
      AND close > (PREV(close) + PREV(open))/2
      AND open < PREV(close)
      AND close < PREV(open)
)

哪个输出：

库存	CLOSE_DATE	PTTRN	打开	高的	低的	关闭	体积
XYZ	01/01/2021	初始值	40	40.5	38.5	38.8	83057
XYZ	2021年2月1日	向下	39.2	39.2	37.2	37.8	181814
XYZ	2021年3月1日	向下	38	38.5	36.5	37	117378
XYZ	2021年4月1日	向下	36.5	36.6	35.6	35.7	93737
XYZ	2021年5月1日	牛市_吞噬	35.35	36.8	35.1	36.7	169106
XYZ	2021年6月1日	向上	36.5	38.5	36.5	38	123179
XYZ	2021年7月1日	向上	37.5	39.5	37.3	39.4	282986
XYZ	2021年8月1日	向上	39	40.5	38.5	40	117437
XYZ	2021年9月1日	BEARISH_HARAMI	39.7	39.8	39.3	39.4	873009
XYZ	10/01/2021	向下	39.2	39.2	37.2	37.8	62522
XYZ	11/01/2021	向下	38	38.5	36.5	37	114826
XYZ	12/01/2021	穿孔线	36.5	37.9	36.3	37.8	281461
XYZ	13/01/2021	向上	37.5	39.5	37.3	39.4	77334
XYZ	14/01/2021	向上	39	40.5	38.5	40	321684

db<>在这里摆弄

Answer 2

我已经赞成@MT0的答案，我match_recognize自己也会用它来处理这类事情，因为这正是它旨在处理的问题。然而，match_recognize这是一个非常复杂的构造，并且您正在寻找的模式非常简单。正如所表达的，您可以通过仅使用一些lag分析函数的更简单的查询来解决您的问题。随着您正在寻找的模式变得更加复杂，您会发现使用它们来表达它们会更容易，match_recognize而仅使用它们来处理它们lag会更困难，但当前的问题相对容易用这种方式表达。

请注意，我对“穿刺线”公式进行了与 @MT0 相同的更改

with data as (
select src.stock,
       src.close_date,
       src.open o,
       src.close c,
       src.high h,
       src.low l,
       lag(src.open) over (partition by src.stock order by src.close_date) o1,
       lag(src.close) over (partition by src.stock order by src.close_date) c1,
       lag(src.high) over (partition by src.stock order by src.close_date) h1,
       lag(src.low) over (partition by src.stock order by src.close_date) l1
  from raw_source src
)
select d.*,
       case when o1 > c1 and c > o and c > h1 and o < l1
            then 'Bullish Engulfing'
            when (O1 < C1) and (O > C) and (O < C1) and (C > O1) and (H < H1) and (L > L1)
            then 'Bearish Harami'
            when (O < C) and (O1 > C1) and (C > (C1 + O1)/2) and (O < C1) and (C < O1)
            then 'Piercing Line'
         end pattern
  from data d
     

在此 dbfiddle 的pattern列中产生相同的结果。不过，由于我们可以使用与表达公式相同的语法，因此遵循此查询中的逻辑可能比理解语法更容易。match_recognize