Instrument variable.

If some variables are correlated with error term, you can use instrument variables to substitute them so that you can obtain consistent estimators.

Greene's book has detailed information about IV estimator.

[此贴子已经被作者于2006-6-13 11:33:07编辑过]

The IV should be uncorrelated with the error term and highly correlated with the regressor.

when you run OLS, the regressors should be uncorrelated with the error terms, other wise you can't use OLS. The least squares estimator will not be consistent. For example if one of your regressor X is correlated with the error term. You should pick some varialbe Y to help you to do the estimation. Please also check GMM and 2SLS.

我超喜欢有人问楼主这样的问题
因为我也是entry level的菜鸟 需要大师们指导……

还是英文的句子助于理解呵呵