可以用加权估计

如果是异方差的话，就用WLS估计，而工具变量估计只有在X与U相关时才使用。

谢谢楼上两位。
那么如果异方差是和x相关的（如x增加，方差增大），在这种情况下，X和U不就相关了么？这时候是不是也可以用工具变量解决呢？
谢谢！！！

In this case, the variance is a function of x, OLS is still unbiased. But it is NOT efficient.  If you use SAS, then you can estimate the variance function with a regression equation. The weight OLS will work, but many OLS package needs to know the weights upfront. This is not true in your case.