For the model $Y=X\beta$ wher have estimators $beta_0$ and $\beta_1$ that are linear functions of $y_i$, such that:

$\beta_0=\sum c_i y_i$ and $\beta_1=\sum d_i y_i$

$c_i=\frac{1}{n}- \frac{x_i-\bar x}{S_{xx}}\bar x$ and $d_i=\frac{x_i-\bar x}{S_{xx}}$

Now I want to find $\mathrm{cov}(\sum c_i y_i,\sum d_i y_i)$

This should be $c_id_i\sum \mathrm{var}Y_i$

I tried doing it together without doing a 1-case, whithout success.

I have $$\mathrm{cov}(\sum c_i y_i,\sum d_i y_i)=E(\sum c_i y_i \times \sum d_i y_i)-E(\sum c_i y_i)E(\sum d_i y_i)$$ which I don't know how to calculate

But for a single observation I have:

$$\mathrm{cov}(c_i y_i, d_i y_i)=E(c_id_i y_i^2)-c_id_iE(y_i)^2=c_id_i \mathrm{var}Y_i$$ , then I can sum over $i$

Can somebody tell me if I can do the whole summ in 1 go?


Hints: I hope all you need.

The fundamental rule you need here is that $Cov$ is linear in both its arguments. Particularly, $Cov(aX, bY) = ab\,Cov(X,Y).$ Then use $Cov(X,X) = Var(X).$

To evaluate the product of sums, you also need to use the independence of $X_i$ and $X_j$ for $i \ne j.$ Hence, $Cov(X_i, X_j) = 0.$ And all of the 'cross-products' disappear.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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