fx,y)=x+y, 0sxs1, 0sysl

Find ECX]Y) and E(YX)

Solution:

EX|Y]= f * | yde=
0
h(y)
where

hy)= (x+ y)dt y+

0
=
0Sys

Hence
1X=l

EX - j t a ' T . 2+3
1 3(2y +1)
0 *
y*

EY|X]= ||fo|2dy=y/)d
8(x)

where

s)- a+d=*+5. 0<x<l

Hence
,2

px+)
E(Y| X)= -dy=. y- 2+3x)
X+ 3(2x +1)

9.5 MOMENT GENERATING FUNCTIONN

The moment of a distribution occupies a central position both in theoretical
and applied statistics. Because of its role and importance, it would seem
useful if a generalized function could be found that would give us a

Statistics-37
 AND PROBARI
INTRODUCTION
TO
STATISTI
ILITY
AN
a finas
562
of all the
moments.
Fortunately,
Function
such
nately, such
(MG
function does exist
representation
Generating
Moment
call
which we
of ent
mome
generating
n t generating funes
Tunction in details
Before we
introduce
the
with
concept
how we express
the oments
moment of eals,
distributie
a
bution
be acquainted as we will see
see, are essentiallyin
let us
value.
These expressions,

terms of expected expectations

and are xactly analogous to
exactly
mathematical
there the
in Chapter IV. AS
we saw
special discussed
can
terms of'any rbitrary value
value or
moments
ordinary in any arbitrary n terms
in terms of origin,
be presented we present these moments in as
mean. Here too,
arithmetic
of the
of expected values.
fashion through the use

9.5.1 Moment
about an Arbitrary Point
moment about an arbitrary point 'a'
If v, stands for the rth
v, = E(X - a)
if X is discrete
--)S),

= la-a)'fe if X is continuous

9.5.2 Moment about Mean

random
We similarly define the rth moment about the mean u of the
can
have
variable X setting a=u in the above equations. Denoting this by A. we

,E(X-H
=-)f(x), if Xis discrete

Cr-4 Sx)dr,
-0
if X is continuous
second

In
particular, the first moment about mean is zero, i.e. ad
central moment ph is the
variance. Thus, when =2
=E(X-)?
9.5.3 Moment about
E(X)-{E(X)}<
Origin rigin o fthe
IfX is random variable, then
a
random variable is 4, where the rth moment abou
,
 MATHEMATICAL EXPECTATION
563

=
E(X") =
Sx), ifXis discrete
If X is continuous
CO

since
have o =E(X") =1,
when0.
we

E ( 1 ) - r ) =1, if Xis discrete

-Sxdr =i, ifXis continuous

when =1, u E(X) =L, which is simply the mean of the random variable
=

generating function (MGF) of a random variable is such a

The moment
function that, if it exists, can be used to generate all the moments of that
variable. This is defined as follows:
Definition: Let X be a random variable with probability density function
ft) Then the function M), called the moment generating function
(MGF) of the random variable X is defined by

M)= E) =
>"S). if X is discrete.
if X is continuous

Moment generating functions will exist only if the sum or the integral
VEConverges and if they exist, it can be used to generate all the
moments of the random
variable.
oht
Let us nowN xamine how the function M() defined above can help us in
b1aining the moments of a
distribution.
Since

,x
e=1+ x+- + -f..
We can 2! 3!
expand M) as follows
Mx)= E(X) E 1+LX X) , (CX).= (LX)
+ 2! 3
T...

r!
 ODABILITY
BABILITY
=1+t)EX*)+X)+. 2
2/ 3/
EX)+

r=0
demonstrates that the coefficient of
The above expression clearly IPin
but the rth rawW moment. In other wor
words
the expansion of M) is nothing
Coefficient of t/1! =h
Coefficient of "/2!=
Coefficient oft*/3! =4"
Coefficient of */4!=u

Coefficient of t"/r!=4'
into corrected moments using
be converted
can n o w
These r a w moments

relations established
in Chapter IV.
the to differentiae
the MMt) is
moments from
alternative way of deriving
An t=0. Specificallyy
with respect to / and then setting
MA)
-M,(
0
M,() once, we get
differentiating
Thus,
.
M , ) i +tph + 2
=

dt
Setting =0,
M,(0) =
E(X) =
u
dt v e c t e d

expecteu
=0
at =
evaluated at
yields the
o yields expc
derivative
of the
of the mgf
m8
first
Thus the random
v a r i a b l e X.

o f the
value
4ain
 o m v a r i a b l k

e rälndom
t h er andom
variab/

since
tne
M,)=Ele" e*..ed
= E(e*) E(e**).E(e**)
Mx, ) Mx, 0)..Mx, t)
- |Mx, )
i=l

H e n c et h ep r o o f .

Property
of Moment Generating Function
U n i q u e n e s s

now state one

more important property of the moment generating
W es h a lIl
of a ction. The proof of this property is beyond the
in the form
is thus omitted.
function

ofthis book and

scope
Let X and Y be wo random
Theorem 9.17. (Uniqueness property)
with mgf M) and M{) respectively. f M) M{) for all
=

uariables
X and Y have the identical probability distribution.
wtues oft, then

X is a discrete random variable having the

Example 9.23: Suppose,
following probability function:

e m x = 0,1,2.,..
f)=- x!
and variance.
generating function of X and its
mean
Find the moment

Solution: By definition

M,)=E(")=" s
ee"m me)_e-m_me =me' -1)
=0

Hence the mean is

 meme-1)
m
=0
twice and setting =0
Similarly, differentiating

ECX ne'emd-+ mnt-»

Hence the variance is

d-E(X)-LE()]* =m+m-m=m
Example 9.24: Suppose X is a random variable for which the
follows:
ft)= for x>0
0 elsewhere
Determine the mgf and hence the variance of X.
Solution: For any real number t,

My)= Ee") =
| "e"dt= ["dr
The final integral in this equation
will be finite if and
only if K
Therefore, M0) exists only for l . For
any such value oft,
M0)= 1-
Since M) is finite for all
values of t in an interval about the point F,a
moments of X exist. The
first two derivatives of Mt) are

M0)
dt
1-1
and drt). 2

(1-1
Setting -0, E(X) =I and
E(X') =2
M E(X) =l
Hence
lt=0
M,=E(X")=
and
ad2
l=0

also beVCX)-ECX)-{E(X)P=2-1=1.
The moments can
obtained by expanding Mi) as follows:
ng MM) as

=1+t+ +++.
 =ka *120
Higher
lgebra by
Algebra by Berna d Child
see

equation
(*),

For
derivation
of the 9.8 CHARACTERISTIC FUNCTION

zenerating function dise

discussed in the
generating d i s t e e d in

limitation
of
moment

not
exist for all distributio This
all

The
chief
section
is that it
does
function, whi
ristic function,
c h a r a c t e r i s t i c
which define in
previous in the
present
deficiency
is not
this section. distribution of the v.

denote the probabiliy

Definition
9.5: Let fx) aenolea by Pr ¬) is def
characteristic jumcthon,
Then the
variable X.

as
)=El if X is discrete
2S)

- fx)dr, if X is continuous
-00

where i=-1, the imaginary unit.

The properties of or) may be summarized as follows:
(i) yt) is continuous in f.
Gi) oy() is defined in every finite t interval.
(ii) dA0) =1.
(iv) oAt) and o(-1) are conjugate
quantities.
(v)1x)s e"1f(x)dr 1 ¢x(0) =

-00

continuous
A general theorem by Cramer states that for a ounded and Sutfticient
function ø() to be a
characteristic function, it is necessary
condition that d(0) =
1 and that