Week 3 Flashcards
What are the assumptions of the Linear model?
- Model is linear
- The n x k matrix X is nonrandom and has rank k
- The n x 1 vector u has mean zero and variance sigma^2 I.n
- The n x 1 vector u follows a normal distribution
- If we let Q.n = X’X/n then Q.n -> Q as n to inf, where Q is a finite and positive definite k x k matrix.
What is meant by the assumption “The model is linear”?
It does not mean that x is linear, as x can be any function, but it can be rewritten to B f(x).
What is meant by the assumption “The n x k matrix X is nonrandom and has rank k”?
This means that X’X is non-singular
What is meant by the assumption: “The n x 1 vector u has mean zero and variance sigma^2 I.n”?
That the Cov(x, y) is zero for all
How is proven that (X’X)^-1X’y is the BLUE?
We want to estimate theta (true param) using a linear function. So we can write theta hat = A’y
Now if we impose unbiasedness, then:
E(theta hat) = E(A’y) = A’X β = W’β
And we get the variance var(theta hat) = var(A’y) = A’ var(y)A
This is given by an assumption, so we get:
= A’(sigma^2 I.n)A = sigma^2 A’A
A’A is minimized when A = X(X’X)^-1W
Thus the OLS is the BLUE estimator.
What is the vector w?
The vector which makes clear which value of beta we want to estimate
How is shown that A’A is minimized?
If we set D = A - X(X’X)^-1W, and we have X’A = W iff. X’D = 0
A’A = (D’ + W’(X’X)^-1X’)(D+X(X’X)^-1W) ≥ W’(X’X)^-1W
QED
What is the formula for the sample variance?
s^2 = y’By = (X beta + u)’B(X beta + u) = u’Bu + 2 beta’ X’Bu + beta’X’BX beta
How is shown that the sample variance is unbiased?
E(s^2) = E(u'Bu) + 2 beta'X'BE(u) + beta'X'BX beta = sigma ^2 tr(B) + beta ' X' BX beta As tr(B) = 1, X'BX = 0, we get = sigma^2