我用的是F(K1,K2,L1,L2), 对资本和劳动进行了拆分。OLS时DW通不过，然后用AR(1)，居然无解。检查了一下发现是共线性。

In your case, if needed and allowed as well you can impose constant return to scale to eliminate the multicollinearity. But this is valid only when you want to impose some restriction but not test for them.

[此贴子已经被作者于2005-9-11 0:36:30编辑过]

probably you want to first check whether the colinearity andautocorrelation are indeed the case with the data, just as a fact check.
if indeed colinearity exists, then dependent on your analysis purpose,you can always use some linear transformation (i.e. coordinate change)to eliminate the colinearity and then proceed next.

出现多重共线性可以用很多的方法避免，譬如：增加样本量，变换模型形式，采用横截面数据和时间数列数据并用，等等。可以参考高级调查分析师用书——《经济计量分析》这本书中有讲到。

可以用广义差分法，科克伦—奥克特迭代法，一阶差分法，德斌两步法……

 

 

用差分最简单啦