Linear regression estimate $\hat{\beta}$

98 Views Asked by At

Follows the enunciated of the question and after my reasoning. In short, I didn't arrived to the answer of the book but I know the reasoning. Follows the question:

enter image description here

Follows the hint and the answer of the book:

enter image description here

My reasoning:

Let A = \begin{pmatrix}1_{m} & 0\\1_{m}&1_{m}\\1_{n}&21_{n}\end{pmatrix}, then

$A^{t}A = \begin{pmatrix}1_{m} & 1_{m}& 1_{n}\\0&1_{m}&-21_{n}\end{pmatrix} \times \begin{pmatrix}1_{m} & 0 \\1_{m}&1_{m}\\1_{n}&-21_{n}\end{pmatrix}= \begin{pmatrix}2m+n & m-2n \\m-2n&m-4n\end{pmatrix}$

Then

$(A^{t}A)^{-1}=\frac{1}{(2m+n)(m-4n)-(m-2n)^{2}}\begin{pmatrix}m-4n & 2n-m \\2n-m&2m+n\end{pmatrix}$

After:

$\hat{\beta} = (A^{t}A)^{-1}A^{t}Y$

Then, using the inverse matrix calculated above

$\hat{\beta} = \frac{1}{(2m+n)(m-4n)-(m-2n)^{2}}\begin{pmatrix}m-4n & 2n-m \\2n-m&2m+n\end{pmatrix}\begin{pmatrix}1_{m} & 1_{m} & 1_{n} \\0&1_{m}&-21_{n}\end{pmatrix}\begin{pmatrix}U \\V\\W\end{pmatrix}$

I don't getting much advanced when I develop the expression above. I know the way is this, but i need some help to arrive to the answer.

1

There are 1 best solutions below

6
On BEST ANSWER

Don't have enough rep to comment. It seems that you are using $AY$ instead of $A^tY$ in your expression for $\hat{\beta}$. Additionally the bottom right entry of $A^tA$ should be $m+4n$.