Question

我有一些相当简单的SQL应该每个资产每季度提供1行1。相反，我每组获得多行。

下面是SQL，SAS数据步骤和一些输出数据。重复行的数量（在下面的数据中，227708）等于Num_borrowers，这是asset1的行数。

proc sql outobs=max;

create table table1 as
select 
    case 
        when period_dt ='01DEC2003'd then '2003Q4'
        when period_dt ='01DEC2004'd then '2004Q4'
        when period_dt ='01DEC2005'd then '2005Q4'
        when period_dt ='01DEC2006'd then '2006Q4'
        when period_dt ='01DEC2007'd then '2007Q4' 
        when period_dt ='01DEC2008'd then '2008Q4'
        when period_dt ='01DEC2009'd then '2009Q4'
        when period_dt ='01DEC2010'd then '2010Q4'
        when period_dt ='01DEC2011'd then '2011Q4'
        when period_dt ='01DEC2012'd then '2012Q4'
        when period_dt ='01DEC2013'd then '2013Q4'
        when period_dt ='01JUN2014'd then '2014Q2'   
    end as QTR,
    case 
        when MM_ASSET in ('C&I', 'Foreign', 'Leasing','Scored-WF','Scored-WB')  THEN 'C&I'
        when MM_ASSET='Construction' THEN 'Construction RE'
        when MM_ASSET='Mortgage-IP' THEN 'Income Producing RE'
        when MM_ASSET='Mortgage-OO' THEN 'Owner Occupied RE'
        when MM_ASSET='Mortgage-SF' THEN 'Mortgage-SF'
        when MM_ASSET='Unknown' THEN 'Other'
    end as asset1,
    count (period_dt) as Num_Borrowers, 
    exposure,
    co_itd,
    MM_NINEQTR_LOSS,
    MM_LIFE_LOSS
  from td_prod.OBLIGOR_COMBINED
  where period_dt in ('01DEC2003'd,'01DEC2004'd,'01DEC2005'd,'01DEC2006'd,'01DEC2007'd,'01DEC2008'd, '01DEC2009'd,'01DEC2010'd,'01DEC2011'd,'01DEC2012'd,'01DEC2013'd,'01JUN2014'd)
  and mm_asset in ('C&I','Foreign','Leasing','Construction','Mortgage-IP','Scored-WF','Scored-WB'
               'Mortgage-OO','Mortgage-SF','Unknown')
  group by 1,2
  order by 1,2;

quit;



data table2; set table1;

  Total_Exposure = exposure/1000000;
  if total_exposure = 0 then total_exposure=.;
  Total_Charge_Offs =co_itd/1000000;
  Total_9Q_Losses = MM_NINEQTR_LOSS/1000000;
  Total_Life_Losses = MM_LIFE_LOSS/1000000;
  avg_borrower_exp = total_exposure/num_borrowers;
  co_rate = total_charge_offs/total_exposure;
  life_lossR = Total_life_losses/total_exposure;
  nineQtr_lossR = total_9q_losses/total_exposure;

run;



*** sample of output data set ***;
qtr             asset1      num_borrowers
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708
2003Q4          C&I             227708

Answer 1

认识到我上面的评论更像是一个答案。

在SAS SQL中，在具有group by子句的查询中，该子句包含select语句中的无关列（即列不是组的一部分，而不是从聚合函数派生的列），SAS将“汇总”统计信息重新“重新合并”到原始数据（附有相关说明）。大多数SQL只会抛出错误。以下是一个例子：

data have;
  input gender $ age score;
  cards;
M 10 100
M 20 200
F 30 300
F 40 400
;
run;

proc sql;
  select gender, mean(age) as AvgAge, SCore
    from have
    group by gender
  ;
quit;

返回：

 gender      AvgAge     score
 F               35       300
 F               35       400
 M               15       100
 M               15       200

在你的代码中，曝光，co_itd，MM_NINEQTR_LOSS和MM_LIFE_LOSS都是无关的列，导致SAS重新合并。

每当重新出现时，您将在SAS日志中看到以下消息：

注意：查询需要重新汇总摘要统计与原来的统计数据。

有关详细信息，请参阅SAS documentation on summary-function中的重新合并数据部分

SAS proc sql按变量返回group by / order的重复值

1 个答案: