When deriving the OLS estimators, we get to a point where we need to analyze asymptotically the following term:

$n^{-1}\sqrt{n} \sum_{i=1}^n (X_i - \mu_x)u_i$

We'll normally have the weaker assumption that $E(X_iu_i) = 0$, which will imply that $E((X_i - \mu_x)u_i) = 0$.

However, to apply the Central Limit Theorem, we need that the sequence $((X_i - \mu_x)u_i)_{i=1}^n$ be independent and identically distributed (i.i.d).

So my question is, given that the sequence $(u_i, X_i)_{i=1}^n$ is i.i.d by assumption, how can we grant $(u_i, X_i)_{i=1}^n$ is i.i.d?

$E(X_iu_i) = 0$ will only grant uncorrelatedness in this case.

  • it is the sampling procedure that grants you that ...You extract the pair $(Y_i,X_i)$ that come from a bivariate (or multivariate in general) distribution. The error is just a linear combination of $Y$ and $X$. If the sampling procedure is done correctly then the observations $(Y_1,X_1), (Y_2,X_2), ... , (Y_n,X_n)$ are i.i.d and also a function of those is formed by independent observation since they are defined on the same sample space (that is a product space) – Hard Core Mar 16 at 21:39
  • Thank you @HardCore. Do you have a source for that information? Also, the result that any function of those is i.i.d holds? I mean, for example, $(X_i)^8$ + $ln(Y_i)$ will be i.i.d across i? – jpugliese Mar 16 at 22:00
  • try this .. it explains my brief comment .. I have my lecture notes on measure theory as source, but I think @whuber explanation is more than enough stats.stackexchange.com/questions/94872/… – Hard Core Mar 16 at 22:32

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.