28.01.2015 Views

PDF of Lecture Notes - School of Mathematical Sciences

PDF of Lecture Notes - School of Mathematical Sciences

PDF of Lecture Notes - School of Mathematical Sciences

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

1. DISTRIBUTION THEORY<br />

Remark<br />

This is also a re-statement <strong>of</strong> previously established results. To see this, observe that<br />

if a T i is the i th row <strong>of</strong> A, we see that Y i = a T i X and, moreover, the (i, j) th element <strong>of</strong><br />

AΣA T = a T i Σa j = Cov ( a T i X, a T j X )<br />

= Cov(Y i , Y j ).<br />

1.9.4 Properties <strong>of</strong> variance matrices<br />

If Σ = Var(X) for some random vector X = (X 1 , X 2 , . . . , X r ) T , then it must satisfy<br />

certain properties.<br />

Since Cov(X i , X j ) = Cov(X j , X i ), it follows that Σ is a square (r × r) symmetric<br />

matrix.<br />

Definition. 1.9.4<br />

The square, symmetric matrix M is said to be positive definite [non-negative definite]<br />

if a T Ma > 0 [≥ 0] for every vector a ∈ R r s.t. a ≠ 0.<br />

=⇒ It is necessary and sufficient that Σ be non-negative definite in order that it can<br />

be a variance matrix.<br />

=⇒ To see the necessity <strong>of</strong> this condition, consider the linear combination a T X.<br />

By Theorem 1.9.8, we have<br />

=⇒ Σ must be non-negative definite.<br />

0 ≤ Var(a T X) = a T Σa for every a;<br />

Suppose λ is an eigenvalue <strong>of</strong> Σ, and let a be the corresponding eigenvector. Then<br />

Σa = λa<br />

=⇒ a T Σa = λa T a = λ||a|| 2 .<br />

Hence Σ is non-negative definite (positive definite) iff its eigenvalues are all nonnegative<br />

(positive).<br />

If Σ is non-negative definite but not positive definite, then there must be at least one<br />

zero eigenvalue. Let a be the corresponding eigenvector.<br />

=⇒ a ≠ 0 but a T Σa = 0. That is, Var(a T X) = 0 for that a.<br />

=⇒ the distribution <strong>of</strong> X is degenerate in the sense that either one <strong>of</strong> the X i ’s is<br />

constant or else a linear combination <strong>of</strong> the other components.<br />

56

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!