自己顶一下，有没有人知道呀？

I did one a year ago. I give an outline here. Otherwise,  I need to dig up my archive.

1) Use sas proc nlmixed procedure to do ML estimation.
2) Ask for output gradient vector evalutaed at MLE
3) Calculate covariance matrix.

Hope this helps.

3# bobguy
谢谢，在试了，那个huber/white/sandwich 稳健回归又是怎么回事呢，用SAS怎么跑呀？

proc nlp data=t1;
  max logf;
parms const pri sec man her10 size tang gea exch ind1 ind2 ind3 ind4 ind5 ind6 ind7 ind8 ind9 ind10 ind11 ind12 ind13 ind14 ind15 ind16
ind17 ind18 ind19 ind20 ind21 ind22 y1 y2 y3 y4 y5;
   u=roa - (const + pri * prilarge + sec * second + man*manager+her10*h10+size*size_roa+tang*tangible+gea*gear+exch*exchange+ind1*B+ind2*C
+ind3*D+ind4*E+ind5*F+ind6*G+ind7*H+ind8*I+ind9*J+ind10*K+ind11*L+ind12*M+ind13*C0+ind14*C1+ind15*C2+ind16*C3+ind17*C4+ind18*C5+ind19*C6+ind20*C7
+ind21*C8+ind22*C99+y1*year99+y2*year00+y3*year01+y4*year02+y5*year03);
  sigmau2=0.0604154**2;
  sigmae2=0.0675433**2;
  no=_n_;
  f=-0.5*(((roa - const + pri * prilarge + sec * second + man*manager+her10*h10+size*size_roa+tang*tangible+gea*gear+exch*exchange+ind1*B+ind2*C
+ind3*D+ind4*E+ind5*F+ind6*G+ind7*H+ind8*I+ind9*J+ind10*K+ind11*L+ind12*M+ind13*C0+ind14*C1+ind15*C2+ind16*C3+ind17*C4+ind18*C5+ind19*C6+ind20*C7
+ind21*C8+ind22*C99+y1*year99+y2*year00+y3*year01+y4*year02+y5*year03)^2-sigmau2*u**2/(no*sigmau2+sigmae2))/sigmae2+log(no*
sigmau2/sigmae2+1)+no*log(2π*sigmae2));
  logf=log(f);
run;
询问，为什么我跑这个程序会出现提示：
           proc nlp data=t1;
              ERROR: 过程 NLP 没有找到。

可能是没有这个模块的授权。