有谁知道 如何用宏语言实现bootstrap resampling?

以下只是我个人的一些思路，希望高手能够指教一下！

Proc sql;

Create table newdata (drop=_1-_100) as

Select *, pre+_1 as y1,

pre+_2 as y2,

………

pre+_100 as y100,

From originaldata;

Quit;

怎么能把这个100步，用loop或宏语句表示呢？

Proc reg data=newdata;

Model y1=x1 x2 x3 x4

By co;

Run;

……….

Proc reg data=newdata;

Model y100=x1 x2 x3 x4;

By co;

Run;

这里也涉及到怎么用loop或宏语句表示.还有如何把我要的每家公司的每次sample的intercept, t value of intercept显示出来？用ods语句？

Data pvalue;

Count=0;

Numiters=100;

If t_new>t_obs then count=count+1;

End;

p_value=count/numiters;

Run;

用这个语句能够计算出我所要的p-value吗？

31# funwin
什么是pre？

I see, the PRE is the old Y without including intercept

yeah, pre is just a variable, equals to b*X1+c*X2+d*X3..., that's right, you got it,

funwin 发表于 2010-4-9 22:01
以下只是我个人的一些思路，希望高手能够指教一下！

Proc sql;

Create table newdata (drop=_1-_100) as

Select *, pre+_1 as y1,

pre+_2 as y2,

………

pre+_100 as y100,

From originaldata;

Quit;

怎么能把这个100步，用loop或宏语句表示呢？

Proc reg data=newdata;

Model y1=x1 x2 x3 x4

By co;

Run;

……….

Proc reg data=newdata;

Model y100=x1 x2 x3 x4;

By co;

Run;

这里也涉及到怎么用loop或宏语句表示.还有如何把我要的每家公司的每次sample的intercept, t value of intercept显示出来？用ods语句？

Data pvalue;

Count=0;

Numiters=100;

If t_new>t_obs then count=count+1;

End;

p_value=count/numiters;

Run;

用这个语句能够计算出我所要的p-value吗？

1. 
   
2. %macro RegMcr;
   
3. data new;
   
4.   set yours;
   
5.   array _{100}; array newY{100};
   
6.   do i = 1 to 100;
   
7.     newY[i] = pre+_[i];
   
8.   end;
   
9. run;
   
10.    proc reg data = new outest=est1(keep = sex _model_ _type_ _depvar_ Intercept where = (_type_ in ('PARMS', 'T'))) tableout noprint;
    
11.       by sex;
    
12.   %do i = 1 %to 100;
    
13.        newY%sysfunc(putn(&i, z5.)): model  newY&i = x1-x4;
    
14.      %end;
    
15. proc transpose data = est1 out = est2(rename = (PARMS = Intercept) drop = _name_ _label_);
    
16.   id _type_;
    
17.   var intercept;
    
18.   by co _model_;
    
19. run;
    
20. %mend RegMcr;
    
21. %RegMcr
    
22.   ;

复制代码

For each Y, you have 100 corresponding bootstrap-Y’s and thus 100 bootstrap-intercepts. Among those 100 intercepts' T statistics, how many exceed the T of the original intercept (only 1 for each company)? If that is what you are asking, so

1. data a;
   
2. input Co T; datalines;
   
3. 1 1
   
4. 1 2
   
5. 1 3
   
6. 2 1
   
7. ;
   
8. data b;
   
9. input Co T; datalines;
   
10. 1 2
    
11. 2 0
    
12. ;
    
13. proc sql;
    
14. select Co, count(*) as n 'N of Ta>Tb'  from
    
15.   (select a.*, b.T as Tb from a, b where a.Co = b.Co) where  T > Tb
    
16.   group by Co
    
17. ;
    
18. quit;

复制代码

34# jingju11

So useful. It's been done. I really appreciate your kindly help.
Have a good weekend.

35# jingju11

Yeah, that's what I need. You've totally grabbed my idea for data processing. Cheers.

上次多谢你指点，帮我解决了bootstrap中的残值再抽样问题，非常感谢。
现在想用另一种bootstrap方法(这simulation方法，我称之为jointly sampling.)：
1, 对每家公司，按公式：Yi,t=ai+bi*Bi,t+ci*Ci,t 做时间序列回归。得出每家公司的ai, 以及 Zi,t=Yi,t-ai;
2, 然后对每家公司的时间t进行有放回的再抽样（抽取300次，我的研究时间段一共300个月），不管每家公司原来有多少期，都抽出同样300期。比如有一家公司虽然只有70期，但也抽出300期来。
(上次进行的是残值再抽样，而且抽取的次数和公司原来的期数一致)
3;做 Zi,t =alphai+bi*Bi,t+ci*Ci,t 时间序列回归(每个公司都有抽取出来的300期了)，得出一个新的alphai and t-statistic of alpha ;
5. 步骤2－3，重复1000次。也就是说每家公司最后有1000个 new alpha and t-statitics of alpha.
不知如何能够实现？望jingju11能否再次指教一下！非常感激