CGF
CGF
CGF
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
!
MX (t) = E(etX ) =
"r tr /r!, t T ,
r=0
where T = {t R : MX (t) < }. The "r are called the moments of X (around
the origin).
This has the following properties:
!
!
!
!
""
"
var(X) = MX
(0) {MX
(0)}2 ;
"n
Week 5 slide 5
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
X
tX
KX (t) = log MX (t) = log E(e ) =
r tr /r!, t T ,
r=1
""
var(X) = KX
(0) = 2 ;
P
if X1 , . . . , Xn are independent, then KX1 ++Xn (t) = n
j=1 KXj (t), so the cumulants
of a sum of independent variables are the sums of the cumulants;
Week 5 slide 6
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
)
,
exp
2 2
(2 2 )1/2
x R.
Properties:
! the standard normal variable Z = (X )/ has mean 0 and variance 1,
Z N (0, 1);
!
/2
0.8
0.84
0.9
1.28
0.95
1.64
0.975
1.96
0.99
2.33
0.995
2.58
Week 5 slide 7
Normal distribution II
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
MZ (t) = e
t R,
2 /2
, t R.
Lemma 5 (a) If X1 , . . . , Xn are independent random variables with MGFs MXj (t),
for t Tj , and a, b1 , . . . , bn R are constants, then
Ma+b1 X1 ++bn Xn (t) = eat
n
&
j=1
n
'
{b1
j Tj }.
j=1
ind
Week 5 slide 8
Chi-squared distribution
Introduction
Some Distributions
Moment generating
function
Cumulant generating
function
Reminder: Normal
distribution
Normal distribution
II
Chi-squared
distribution
Student t
distribution
F distribution
Computing
Normal Random
Sample
Multivariate
Distributions
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
iid
Week 5 slide 9
Density of 21
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
10
w
15
20
Week 5 slide 10
Density of 22
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
10
w
15
20
Week 5 slide 11
Density of 24
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
10
w
15
20
Week 5 slide 12
Density of 26
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
10
w
15
20
Week 5 slide 13
Student t distribution
Introduction
Some Distributions
Moment generating
function
Cumulant generating
function
Reminder: Normal
distribution
Normal distribution
II
Chi-squared
distribution
Student t
distribution
F distribution
Computing
Normal Random
Sample
Multivariate
Distributions
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
< t < , = 1, 2, . .
Week 5 slide 14
Density of Student t1
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
0
w
Week 5 slide 15
Density of Student t5
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
0
w
Week 5 slide 16
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
0
w
Week 5 slide 17
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
0.0
0.1
Density
0.2
0.3
0.4
0.5
0
w
Week 5 slide 18
F distribution
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
ind
Definition 9 If W1 , W2 21 22 , then
W1 /1
F =
W2 /2
has the F distribution on 1 and 2 degrees of freedom: we write
F F1 ,2 . Its density function is
fF (u) =
1 /2 2 /2
+
2
1
2 1)
)1 *
1
2 1 2 2
)1
1
2
2
*
1
u 2 1 1
(2 + 1 u)
(1 +2 )/2
u > 0, 1 , 2 = 1, 2, . . . ,
Week 5 slide 19
Computing
Introduction
Some Distributions
Moment generating
function
Cumulant generating
function
Reminder: Normal
distribution
Normal distribution
II
Chi-squared
distribution
Student t
distribution
F distribution
Computing
Normal Random
Sample
Multivariate
Distributions
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
Week 5 slide 20
iid
Theorem 10 If X1 , . . . , Xn N (, 2 ) and
X = n1
Multivariate
Distributions
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
n
!
Xj ,
j=1
then
X
(n 1)S 2
1 !
2
S =
(Xj X)2 ,
n1
j=1
N (, 2 /n)
2 2n1
independent;
equivalently we write
D
S2
+ n1/2 Z,
(n
1)1 2 W,
Z N (0, 1),
W
2n1 ,
Z, W independent.
Week 5 slide 21
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
Week 6 slide 4
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
1
(2)p/2 ||1/2
exp
12 (x
)
T
(x ) ,
x, Rp ;
Week 6 slide 5
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
Definition 4
Let X = (X1 , . . . , Xp )T be a random variable taking values in Rp . The joint moment
generating function of X is defined to be
tT X
MX (t) = E(e
Pp
) = E(e
j=1 tj Xj
),
t T Rp ,
if T is open, then
MX (t)
E(Xj ) =
,
tj t=0
2 MX (t)
MX (t)
MX (t)
cov(Xj , Xk ) =
;
tj tk t=0
tj t=0
tk t=0
the joint cumulant generating function of X may be defined as KX (t) = log MX (t);
Week 6 slide 6
MGF of Np
Introduction
Multivariate
Distributions
Covariance and
correlation
Multivariate normal
distribution
Joint
moment-generating
functions
MGF of Np
Reminder: Theorem
10, week 5
Marginal and
conditional
distributions
"
Likelihood
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
tT X
) = E(e
Pp
r=1 tr Xr
&T
'
1 T
) = exp t + 2 t t ,
t Rp .
(c) If X1 , . . . , Xn N (, 2 ), then
Xn1 = (X1 , . . . , Xn )T Nn (1n , 2 In ).
(d) Linear combinations of normal variables are normal:
ar1 + Brp X Nr (a + B, BB T );
this is non-degenerate iff |BB T | > 0.
Week 6 slide 7
C O L E P O L Y T E C H N I Q U E
F DR A L E D E L A U S A N N E
Week 6 slide 9