MySQL累计产品分组依据

Question

我一直在使用WRDS/CRSP数据集(由UPenn维护的用于学术研究的股票价格数据库)。我一直在下载Python语言的数据并将其插入到我本地的MySQL数据库中。

数据如下所示，主键在(quote_date，security_id)上：

quote_date  security_id tr              accum_index
10-Jan-86   10002       null            1000
13-Jan-86   10002       -0.026595745    973.4042548
14-Jan-86   10002       0.005464481     978.7234036
15-Jan-86   10002       -0.016304348    962.7659569
16-Jan-86   10002       0               962.7659569
17-Jan-86   10002       0               962.7659569
20-Jan-86   10002       0               962.7659569
21-Jan-86   10002       0.005524862     968.0851061
22-Jan-86   10002       -0.005494506    962.765957
23-Jan-86   10002       0               962.765957
24-Jan-86   10002       -0.005524862    957.4468078
27-Jan-86   10002       0.005555556     962.7659569
28-Jan-86   10002       0               962.7659569
29-Jan-86   10002       0               962.7659569
30-Jan-86   10002       0               962.7659569
31-Jan-86   10002       0.027624309     989.3617013
3-Feb-86    10002       0.016129032     1005.319148
4-Feb-86    10002       0.042328041     1047.872338
5-Feb-86    10002       0.04568528      1095.744679

我需要计算accum_index列，它基本上是股票总回报的一个索引，计算方法如下：

accum_index_t = accum_index_{t-1} * (1 + tr_t)

该表有80m行。我已经编写了一些代码来迭代每个security_id并计算累积乘积，如下所示：

select @sid := min(security_id)
from stock_prices;

create temporary table prices (
    quote_date datetime,
    security_id int,
    tr double null,
    accum_index double null,
    PRIMARY KEY (quote_date, security_id)
);

while @sid is not null
do

    select 'security_id', @sid;
    select @accum := null;

    insert into prices
    select quote_date, security_id, tr, accum_index
    from stock_prices
    where security_id = @sid
    order by quote_date asc;

    update prices
    set accum_index = (@accum := ifnull(@accum * (1 + tr), 1000.0));

    update stock_prices p use index(PRIMARY), prices a use index(PRIMARY)
    set p.accum_index = a.accum_index
    where p.security_id = a.security_id
    and p.quote_date = a.quote_date;

    select @sid := min(security_id)
    from stock_prices
    where security_id > @sid;

    delete from prices;

end while;

drop table prices;

但这太慢了，在我的笔记本电脑上，每个安全性都需要大约一分钟的时间，并且需要数年时间来计算这个级数。有没有办法将其矢量化？

干杯，史蒂夫

Answer 1

如果您使用的是MySQL 8，则可以使用window functions创建累积产品。不幸的是，据我所知，在任何PROD()数据库中都没有SQL聚合/窗口函数，除了you can emulate it using EXP(SUM(LOG(factor)))

SELECT
  quote_date,
  security_id,
  tr,
  1000 * (EXP(SUM(LOG(1 + COALESCE(tr, 1)))
    OVER (PARTITION BY security_id ORDER BY quote_date)) - 1)
    AS accum_index
FROM stock_prices

dbfiddle here。

Answer 2

如果您使用的是MySQL 5，则可以逐行模拟此函数，将当前电流与上一个tr相乘。然后我们取最后一行的累加值。

tr是百分比价值，现在？因此，让我们将1加到每个tr。

第一个存储值将为中性1。

试试这个：

SET @variation = 1;
SET @row_number = 0;

SELECT accumulateTr
FROM
    (SELECT
        @row_number := (@row_number + 1) AS rowNumber,
        @variation := (1 + variation) * @variation AS accumulateTr
     FROM
        prices) accumulatedTrs
ORDER BY rowNumber DESC
LIMIT 1;