啥叫standard errors of the (log) differences series

正在读一片文章，它描述如何查找时间序列数据中的极端值。说：

extreme values are defined as those whose (log) change compared with both adjacent months is greater than 3.5 standard errors of the (log) differenced series.

比如，我现有一年的股市日指数数据，如果根据他的描述寻找极端值？多谢。

这是我目前的做法：

1. proc sort data = have out=A1; run;
   
2. proc expand data=A1 out=A1 method = none;
   
3.   id date;
   
4.   convert P = P_lag1   / transformout=(lag 1);
   
5.   convert P;
   
6.   convert P = P_lead1  / transformout=(lead 1);
   
7. run;
   
8. data A1; set A1;
   
9.   ret      = log(P) - log(P_lag1);
   
10.   ret_lead = log(P) - log(P_lead1);
    
11. run;
    
12. 
    
13. * Calculate standard error, which is used to refer to an estimate of that standard deviation,
    
14.   derived from a particular sample used to compute the estimate;
    
15. proc means data=A1 noprint;
    
16.   var ret;
    
17.   output out=std std=std;
    
18. run;
    
19. data A2  (drop= _type_ _freq_); merge A1 std;
    
20.   if abs(ret)> std and abs(ret_lead) > std then do; p_sm = (p_lag1+p_lead1)/2; extreme =1; end;
    
21.   else do; p_sm=p; extreme=.; end;

复制代码