Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

The typical assumption of linear regression, weak exogeneity, states, $$E(\bf{\epsilon_{i}})=0$$ when the regressors are fixed and $$E(\bf{\epsilon_{i}}|\bf{x_{i}})=0$$ when the regressors are random. I can't figure out for the life of me why you don't still need to condition upon your regressors when they are fixed. If we are going to use our model to extrapolate y values for x values other than the fixed ones we selected, won't we need to assume that the expected value of the epsilons is zero at those points as well?

share|cite|improve this question

You don't generally condition on things that are fixed. I can write down $E(\epsilon_i \mid 1+1)$, but the conditioning is meaningless.

If you are interested in predicting for an unknown future $x_i$, then you need to condition on that when you write down your prediction distribution. But fitting the linear regression model is not the same as predicting.

share|cite|improve this answer
So, in order to be sure that you have unbiased estimates for some $y(x_{j})$, where $x_{j}$ is not in your original data set, you need to further assume that $E(\epsilon_i | x_j)=0$? Thank you! I am studying this from an econometrics perspective and it seems that this doesn't appear in many texts. – Kevin Oct 31 '12 at 3:12
I guess I am thinking that since we modelling the random process that generated the data, we would be concerned about how it functions at points other than just the few fixed points that we used to estimate the parameters. – Kevin Oct 31 '12 at 3:37

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.