To quote:

It is well known that, if $W_1, ..., W_n, Z_1, ..., Z_m$ are random variables and $a_1, ..., a_n, b_1, ..., b_m$ are constants, then

$Cov ( \sum_{i=1}^n a_iW_i, \sum_{j=1}^m b_jZ_j) = \sum_{i=1}^n \sum_{j=1}^m a_i b_j Cov(W_i,Z_j)$.

Use this property (and other properties of covariance) to prove that for the model $Y_i = \alpha + \beta x + \epsilon$, with $\epsilon \sim N(0,\sigma^2)$, we have $Cov(\hat{\beta},\bar{Y}) = 0 $,

where $\hat{\beta} = S_{XY}/S_{XX}$.

Currently my most successful attempt was to try to put $S_{XY} \over S_{XX}$ as $\sum_i (x_i - \bar{x})(Y_i - \bar{Y}) \over \sum_i (x_i - \bar{x})(x_i - \bar{x})$ then substituting $Y = \alpha + \beta x + \epsilon$, but even in this best attempt I just end up with a mess and a stray epsilon I can't deal with (one reduces to 0 though the summing to get $\bar{Y}$, but the other ends up as a nasty coefficient).

I'm really not sure what to do.

Edit: Added a run down on some previous issues and the self-study tag.

  • $\begingroup$ If this is for homework or an assignment, could you please add the [self-study] tag and read its wiki? $\endgroup$
    – Andy
    May 25, 2016 at 10:36
  • $\begingroup$ Sorry about that - I'd assumed that the rules were the same as math.stackexchange for them. Added, also editing in some information about what I've already tried. $\endgroup$
    – liqiudilk
    May 25, 2016 at 10:42
  • $\begingroup$ No worries. Just wanted to highlight this given that you are new here. In this sense, welcome to CV :-) $\endgroup$
    – Andy
    May 25, 2016 at 10:43
  • $\begingroup$ Is the regressor $x$ treated as deterministic or as a random variable? $\endgroup$ May 25, 2016 at 13:01

1 Answer 1


Here are some tips. First, just like variances, covariance can ignore additive constants. You can use this to show: $$cov (\hat {\beta},\overline {Y})=cov (\hat {\beta},\overline {\epsilon}) $$

Then, you can show that

$$\hat{\beta}=\sum_iw_iY_i $$

Where $w_i=\frac {x_i-\overline {x}}{S_{XX}} $. You can ignore constants as well to get

$$cov (\hat {\beta},\overline {Y})=cov (\sum_iw_i\epsilon_i,\overline {\epsilon}) $$

Then you can apply your result at the start of your question.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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