moment generating functions - university of saskatchewanlaverty/s241/s241 lectures pdf/10... ·...
TRANSCRIPT
Moment Generating Functions
0
0.1
0.2
0.3
0.4
0 5 10 15
1
b a
a b
f x
x0
0.1
0.2
0.3
0.4
0 5 10 15
1
b a
a b
f x
x
1
b a
a b
f x
x
Continuous Distributions
The Uniform distribution from a to b
1
0 otherwise
a x bf x b a
The Normal distribution
(mean m, standard deviation s)
2
221
2
x
f x em
s
s
0
0.1
0.2
-2 0 2 4 6 8 10
The Exponential distribution
0
0 0
xe xf x
x
Weibull distribution with parameters a and b.
Thus 1x
F x eba
b
1and 0x
f x F x x e xba
b ba
The Weibull density, f(x)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 1 2 3 4 5
(a = 0.5, b = 2)
(a = 0.7, b = 2)
(a = 0.9, b = 2)
The Gamma distribution
Let the continuous random variable X have
density function:
1 0
0 0
xx e xf x
x
aa
a
Then X is said to have a Gamma distribution
with parameters a and .
Expectation of functions of
Random Variables
X is discrete
i i
x i
E g X g x p x g x p x
E g X g x f x dx
X is continuous
Moments of Random Variables
k
k E Xm
-
if is discrete
if is continuous
k
x
k
x p x X
x f x dx X
The kth moment of X.
the kth central moment of X
0 k
k E Xm m
-
if is discrete
if is continuous
k
x
k
x p x X
x f x dx X
m
m
where m = m1 = E(X) = the first moment of X .
Rules for expectation
Rules:
1. where is a constantE c c c
2. where , are constantsE aX b aE X b a b
20
23. var X E Xm m
22 2
2 1E X E X m m
24. var varaX b a X
Moment generating functions
Moment Generating function of a R.V. X
if is discrete
if is continuous
tx
xtX
Xtx
e p x X
m t E e
e f x dx X
Examples
1. The Binomial distribution (parameters p, n)
tX tx
X
x
m t E e e p x
0
1n
n xtx x
x
ne p p
x
0 0
1n n
x n xt x n x
x x
n ne p p a b
x x
1nn ta b e p p
0,1,2,!
x
p x e xx
The moment generating function of X , mX(t) is:
2. The Poisson distribution (parameter )
tX tx
X
x
m t E e e p x 0 !
xtx
x
e ex
0 0
using ! !
t
xt x
e u
x x
e ue e e e
x x
1tee
0
0 0
xe xf x
x
The moment generating function of X , mX(t) is:
3. The Exponential distribution (parameter )
0
tX tx tx x
Xm t E e e f x dx e e dx
0 0
t xt x e
e dxt
undefined
tt
t
2
21
2
x
f x e
The moment generating function of X , mX(t) is:
4. The Standard Normal distribution (m = 0, s = 1)
tX tx
Xm t E e e f x dx
2 22
1
2
x tx
e dx
2
21
2
xtxe e dx
2 2 2 22 2
2 2 21 1
2 2
x tx t x tx t
Xm t e dx e e dx
We will now use the fact that 2
221
1 for all 0,2
x b
ae dx a ba
We have
completed
the square
22 2
2 2 21
2
x tt t
e e dx e
This is 1
1 0
0 0
xx e xf x
x
aa
a
The moment generating function of X , mX(t) is:
4. The Gamma distribution (parameters a, )
tX tx
Xm t E e e f x dx
1
0
tx xe x e dxa
a
a
1
0
t xx e dx
aa
a
We use the fact
1
0
1 for all 0, 0a
a bxbx e dx a b
a
1
0
t x
Xm t x e dxa
a
a
1
0
t xtx e dx
tt
a aaa
a
a
Equal to 1
Properties of
Moment Generating Functions
1. mX(0) = 1
0, hence 0 1 1tX X
X Xm t E e m E e E
v) Gamma Dist'n Xm tt
a
2
2iv) Std Normal Dist'n t
Xm t e
iii) Exponential Dist'n Xm tt
1
ii) Poisson Dist'n te
Xm t e
i) Binomial Dist'n 1n
t
Xm t e p p
Note: the moment generating functions of the following
distributions satisfy the property mX(0) = 1
2 33212. 1
2! 3! !
kkXm t t t t t
k
m mmm
We use the expansion of the exponential function:
2 3
12! 3! !
ku u u u
e uk
tX
Xm t E e
2 32 31
2! 3! !
kkt t t
E tX X X Xk
2 3
2 312! 3! !
kkt t t
tE X E X E X E Xk
2 3
1 2 312! 3! !
k
k
t t tt
km m m m
0
3. 0k
k
X X kk
t
dm m t
dtm
Now
2 33211
2! 3! !
kkXm t t t t t
k
m mmm
2 1321 2 3
2! 3! !
kkXm t t t kt
k
m mmm
2 13
1 22! 1 !
kkt t tk
m mm m
1and 0Xm m
242 3
2! 2 !
kkXm t t t t
k
mmm m
2and 0Xm m
continuing we find 0k
X km m
i) Binomial Dist'n 1n
t
Xm t e p p
Property 3 is very useful in determining the moments of a
random variable X.
Examples
1
1n
t t
Xm t n e p p pe
1
0 0
10 1n
Xm n e p p pe np m m
2 1
1 1 1n n
t t t t t
Xm t np n e p p e p e e p p e
2
2
1 1 1
1 1
nt t t t
nt t t
npe e p p n e p e p p
npe e p p ne p p
2 2
21np np p np np q n p npq m
1
ii) Poisson Dist'n te
Xm t e
1 1t te e tt
Xm t e e e
1 1 2 121
t t te t e t e tt
Xm t e e e e
1 2 12 2 1
t te t e tt t
Xm t e e e e
1 2 12 3t te t e tte e e
1 3 1 2 13 23t t te t e t e t
e e e
0 1 0
1 0e
Xm e
m
0 01 0 1 02 2
2 0e e
Xm e e
m
3 0 2 0 0 3 2
3 0 3 3t
Xm e e em
To find the moments we set t = 0.
iii) Exponential Dist'n Xm tt
1
X
d tdm t
dt t dt
2 2
1 1t t
3 3
2 1 2Xm t t t
4 4
2 3 1 2 3Xm t t t
5 54
2 3 4 1 4!Xm t t t
1
!kk
Xm t k t
Thus
2
1
10Xmm m
3
2 2
20 2Xmm
1 !
0 !kk
k X k
km km
2 33211
2! 3! !
kkXm t t t t t
k
m mmm
The moments for the exponential distribution can be calculated in an
alternative way. This is note by expanding mX(t) in powers of t and
equating the coefficients of tk to the coefficients in:
2 31 11
11Xm t u u u
tt u
2 3
2 31
t t t
Equating the coefficients of tk we get:
1 ! or
!
kkk k
k
k
mm
2
2t
Xm t e
The moments for the standard normal distribution
We use the expansion of eu. 2 3
0
1! 2! 3! !
k ku
k
u u u ue u
k k
2 2 2
22
2
2 3
2 2 2
21
2! 3! !
t
kt t t
tXm t e
k
2 4 6 212 2 3
1 1 11
2 2! 2 3! 2 !
k
kt t t t
k
We now equate the coefficients tk in:
2 22211
2! ! 2 !
k kk kXm t t t t t
k k
m mmm
If k is odd: mk = 0.
2 1
2 ! 2 !
k
kk k
mFor even 2k:
2
2 !or
2 !k k
k
km
1 2 3 4 2
2! 4!Thus 0, 1, 0, 3
2 2 2!m m m m
Summary
Moments
Moment generating functions
Moments of Random Variables
k
k E Xm
-
if is discrete
if is continuous
k
x
k
x p x X
x f x dx X
The moment generating function
if is discrete
if is continuous
tx
xtX
Xtx
e p x X
m t E e
e f x dx X
Examples
1 0,1,2, ,n xx
np x p p x n
x
1. The Binomial distribution (parameters p, n)
1n n
t t
Xm t e p p e p q
0,1,2,!
x
p x e xx
2. The Poisson distribution (parameter )
1te
Xm t e
0
0 0
xe xf x
x
3. The Exponential distribution (parameter )
undefined
X
tm t t
t
2
21
2
x
f x e
4. The Standard Normal distribution (m = 0, s = 1)
2
2t
Xm t e
1 0
0 0
xx e xf x
x
aa
a
5. The Gamma distribution (parameters a, )
Xm tt
a
6. The Chi-square distribution (degrees of freedom n)
(a n/2, 1/2
21
2 2
112
2
0
0 0
xx e x
f x
x
n
n
n
2
2
12
12
1 2Xm t tt
n
n
1. mX(0) = 1
2 33212. 1
2! 3! !
kkXm t t t t t
k
m mmm
0
3. 0k
k
X X kk
t
dm m t
dtm
Properties of Moment Generating Functions
1i.e. 0Xm m
20Xm m
30 , etcXm m
Let lX (t) = ln mX(t) = the log of the moment generating function
Then 0 ln 0 ln1 0X Xl m
The log of Moment Generating Functions
1
X
X X
X X
m tl t m t
m t m t
2
2
X X X
X
X
m t m t m tl t
m t
1
0 0
0
X
X
X
ml
mm m
2
2 2
2 12
0 0 0 0
0
X X X
X
X
m m ml
mm m s
Thus lX (t) = ln mX(t) is very useful for calculating the mean and variance of a random variable
1. 0Xl m
2 2. 0Xl s
Examples
1. The Binomial distribution (parameters p, n)
1n n
t t
Xm t e p p e p q
ln ln t
X Xl t m t n e p q
1 t
X tl t n e p
e p q
10Xl n p np
p qm
2
t t t t
Xt
e p e p q e p e pl t n
e p q
2
20X
p p q p pl n npq
p qs
2. The Poisson distribution (parameter )
1te
Xm t e
ln 1t
X Xl t m t e
t
Xl t e
t
Xl t e
0Xlm
2 0Xls
3. The Exponential distribution (parameter )
undefined
X
tm t t
t
ln ln ln if X Xl t m t t t
11
Xl t tt
2
2
11 1Xl t t
t
2
2
1 1Thus 0 and 0X Xl lm s
4. The Standard Normal distribution (m = 0, s = 1)
2
2t
Xm t e
2
2ln t
X Xl t m t
, 1X Xl t t l t
2Thus 0 0 and 0 1X Xl lm s
5. The Gamma distribution (parameters a, )
Xm tt
a
ln ln lnX Xl t m t ta
1
Xl tt t
aa
2
21 1Xl t t
t
aa
2
2Hence 0 and 0X Xl l
a am s
6. The Chi-square distribution (degrees of freedom n)
21 2Xm t tn
ln ln 1 22
X Xl t m t tn
1
22 1 2 1 2
Xl tt t
n n
2
2
21 1 2 2
1 2Xl t t
t
nn
2Hence 0 and 0 2X Xl lm n s n
Summary of Discrete Distributions
Name
probability function p(x)
Mean
Variance
Moment
generating
function MX(t)
Discrete
Uniform Nxp
1)( x=1,2,...,N
2
1N
12
12 N
1
1
t
tNt
e
e
N
e
Bernoulli
0
1)(
xq
xpxp
p pq q + pet
Binomial xNxqp
x
Nxp
)(
Np Npq (q + pet)N
Geometric p(x) =pqx-1 x=1,2,...
p
1
2p
q
t
t
qe
pe
1
Negative
Binomial kxk qp
k
xxp
1
1)(
x = k, k+1,...
p
k
2p
kq
k
t
t
qe
pe
1
Poisson p(x) =
x
x! e- x=1,2,...
e(et-1)
Hypergeometric
p(x) =
A
x
N-A
n-x
N
n
n
A
N n
A
N
1-A
N
N-n
N-1
not useful
Name
probability
density function f(x)
Mean
Varianc
e
Moment generating
function MX(t)
Continuous
Uniform
otherwise
bxaabxf
0
1
)(
a+b
2
(b-a)2
12
tab
ee atbt
][
Exponential
00
0
x
xexf
x
1
1
2
t
for t
<
Gamma
00
0)(
1
x
xexxf
xaa
a
a
a
2
t
a
for t <
c2
n d.f. f(x) =
(1/2)n/2
(n/2) xn/21e-(1/2)x x ≥ 0
0 x < 0
n 2n
1
1-2t
n/2
for t < 1/2
Normal f(x) =
2
2
2
1
2
1s
m
s
x
exf e-(x-
m)2/2s2
m s2 etm+(1/2)t2s2
Weibull
f(x) =
x1 e-x/ x ≥ 0
0 x < 0
2/( )+1
2/
( )+2
-[ ]( )+1
2
not
avail.
Continuous distributions