Multiple Regression Least Squares Criterion Algebra

254 Views Asked by At

I've been trying to understand the algebra behind:

enter image description here

Is there an equivalent of PEMDAS for matrices?

Even if I assume $(y^T-X^TB^T)(y-XB)$, I get $y^Ty-y^TXB-X^TB^Ty+X^TB^TXB$ which does not equal the above.

To get the above result, I used the following properties:

$(A+B)^T = A^T + B^T$ and assumed there is still a parenthesis wrapped around it. (PEMDAS Question)

$(A+B)(C+D)=(AC+AD+BC+BD)$ by multiplication and distributive property.

What above simplifications am I missing?

1

There are 1 best solutions below

6
On BEST ANSWER

It is $(y-XB)'=(y'-B'X')$

The order of $XB$ has to be interchanged when you transpose it.

Multiplying out the brackets of $(y'-B'X')(y-XB)$

$yy'-y'XB-B'X'y+B'X'XB$

$y'XB$ is a scalar. Thus $y'XB=B'X'y$

$yy'-2y'XB+B'X'XB$

Remark

Let $X$ be a $m\times n$ matrix and $B$ a $n\times 1$ vector. Then $X'B'$ wouldn´t work because $X'$ is a $n\times m$ matrix and $B'$ a $1\times n$ vector.