Sie sind auf Seite 1von 8

' \$

Covariances

Definition
Cov(X, Y ) = E{[X − E(X)][Y − E(Y )]} (1)
..
.
= E[XY − Y E(X) − XE(Y ) + E(X)E(Y )]
= E(XY ) − E(X)E(Y ) − E(Y )E(X) + E(X)E(Y )
..
.
∴ also = E(XY ) − E(X)E(Y ) (2)

Remark Easy to see from definition (1) that Cov(X, X) =

E{[X − E(X)]2 } = Var(X), and from convenient computational

& %
formula (2) that Cov(X, X) = E(X 2 ) − [E(X)]2 .

' \$

Example 1

Let (X, Y, Z) ∼ trinomial(n; p1 , p2 , p3 ), where p1 + p2 + p3 = 1,

X + Y + Z = n, and the joint distribution of (X, Y ) is

f (x, y) = P(X = x, Y = y)
n!
= px1 py2 (1 − p1 − p2 )n−x−y .
x!y!(n − x − y)!
What’s Cov(X, Y )?

& %
ActSc 613 | Lecture 8 © 2015-17 by M. Zhu, PhD 2 of 8
' \$

Example For a,b non-random,

Var(aX + bY ) = E[(aX + bY )2 ] − [E(aX + bY )]2
 2 2 2 2

= a E(X ) + b E(Y ) + 2abE(XY ) −
 2
a [E(X)]2 + b2 [E(Y )]2 + 2abE(X)E(Y )

Rules E(a1 X1 + ... + an Xn ) = a1 E(X1 ) + ... + an E(Xn )

" n
# n
X X X
Var ai X i = a2i Var(Xi ) + ai aj Cov(Xi , Xj )
i=1 i=1 i6=j
 
n
X m
X n X
X m
Cov  ai X i , bj Yj  = ai bj Cov(Xi , Yj )
i=1 j=1 i=1 j=1

& %
ActSc 613 | Lecture 8 © 2015-17 by M. Zhu, PhD 3 of 8
' \$

Definition X and Y are independent if f (x, y) = fX (x)fY (y).

Implication Then,
Z Z Z Z
E(XY ) = xyf (x, y)dxdy = xyf (x)f (y)dxdy = ...
Z Z 
... = yf (y) xf (x)dx dy = E(X)E(Y )
| {z }
E(X)

so Cov(X, Y ) = E(XY ) − E(X)E(Y ) = 0.

& %
ActSc 613 | Lecture 8 © 2015-17 by M. Zhu, PhD 4 of 8
' \$

Example 2

Suppose X1 , X2 , ... are i.i.d. random variables with E(Xi ) = µ and

Var(Xi ) = σ 2 . Let

Sn = X1 + X2 + ... + Xn .

(a) What’s E(Sn ) and Var(Sn )?

(b) What’s E(SN ) and Var(SN ), if N is a random variable as well?
(c) Suppose X1 , X2 , ... are insurance claims, with µ = \$500,
σ = \$100, and N ∼ Poisson(100). Then, SN represents the
total liability. Compare Var(S100 ) and Var(SN ).

& %
ActSc 613 | Lecture 8 © 2015-17 by M. Zhu, PhD 5 of 8
' \$

Counter Example
f (x, y)

x\y −1 0 +1
−1 0 0.1 0
0 0.1 0.6 0.1
+1 0 0.1 0

independent.

Puzzle What kind of dependence is captured by the covariance?

& %
ActSc 613 | Lecture 8 © 2015-17 by M. Zhu, PhD 6 of 8
' \$

Correlation

Definition
Cov(X, Y )
Corr(X, Y ) = p
Var(X)Var(Y )

Theorem −1 ≤ Corr(X, Y ) ≤ +1, with equality if and only X

and Y are almost surely linear functions of each other.
(⇒): If Y = tX + c, then Cov(X, Y ) = tCov(X, X) + Cov(X, c) =
tVar(X), so
tVar(X) t
Corr(X, Y ) = p = = ±1.
2
Var(X)[t Var(X)] |t|

Exercise Complete the proof. [Hint: Notice Var(Y + tX + c), a

quadratic function of t, is ≥ 0 ∀ t ∈ R and consider when it is = 0.]
& %
ActSc 613 | Lecture 8 © 2015-17 by M. Zhu, PhD 7 of 8
' \$

Example 3

What’s the solution to the following optimization problem?

2
 
min E (Y − g(X)) , (1)
g

Remark A Recall that, if the constraint (2) is removed, then

the solution to (1) is simply g(X) = E(Y |X).

Remark B The constrained optimization problem (1)-(2) above

is often referred to as the simple linear regression problem.

& %
ActSc 613 | Lecture 8 © 2015-17 by M. Zhu, PhD 8 of 8