1

fI mats pgfs continued

examples applications

T.iiiiiiii.m.name M t x̅ NEIN

Let's go through some examples

using this formula to compute moments
Examples if 2
Let x̅ have pmf Px x

if
Its mgt is
easily calculated
as My t eat test test
And its moments are

IE x̅ Mx o 21.2 f 3
31.5
IE X E Mx 0
1.4 f 9 5.25 71
x̅ Exp X We saw
previously that My f
Thus x̅ Mx
to t to

E X2
EM E
1 0 no

Var X IE x2 E x̅

We've seen these values before

An important property of the mgt Mx t is that it can be inverted
to determine the probability law of v.v x̅

Them mensionloperty let v.v I have mgt Mx t associated

with it and assume M t is finite teR in same interval 8,5
for 870 Then M t
uniquely determines the cdf E of x̅

While there exist explicit formulas for the inverse to let us recover
the pmf pdf they can be difficult to use in practice Instead mgts
are typically inverted via matching terms

Examples
t
Given My t e
It felt test Compare against general form

M t e
py x

Note that the coefficients match at values 1 0 4,5 Thus

if 1

Px x if 0

if 4,5

Given My t E t for apes

Recall the formula for the geometric series 1 α α't

for any 1 1 2 Try using this formula with αE 1 p et and

for t sufficiently close to 0 so that 1 p et 1

Then Mx t pet p et i ppe

I 1 1 ppest
Observing the coefficients of each term we see
1
IP x̅ k p 1 p
which is a Geometric distribution with parameter p
 of mgfs especially for continuous r.v.is is that

the integrals in their definitions sometimes do not converge Instead

there is another class of generating functions that are used to
bypass this problem

Dfe The characteristicfunction of continuous riv x̅ is 0 IR

defined by
LY I EE

Here ie is the camp

Like how the mgt is related to the Laplace transform the characteristic
function is related to the Fourier transform
t
f e f x dx

Some of the properties of characteristic functions are beyond the scope

of this course We will talk more about applications and properties
that will directly help us with calculations instead

F
iem p Then e EG laptett
x̅ Bin n
p In similar calculation to the
mgt
A t Ci p I
tetp
x̅ Exp x Then t eitx.se xxax
This which some knowledge of contour
is a complex integral requires
Be careful not to treat like real number
integrals over a

It tums out
t.FI
 It is common to have a table of characteristic functions as reference

Def For two v.v s x̅ Y the jointcharacteristicfunction is defined

by Oxy R sit

Oxy s t I e's e s t eR
Like the joint pgf this definition can be extended to an
arbitrary
number of v.v s

Theory R.V s X Y are independent if 0x s t s

0,4 s t

Theorem LinearTranformations Let 0 t be the characteristic

function associated with v.v x̅ Then the characteristic function

tb
associated with YE a b is
given by t e at a beR

Theorem Sums If X Y are independent with respective

characteristic functions 0 x Oy then y t 0 t t

spends his weekend playing the following game

involving a biased coin and some unfair dice On Saturday he
tosses the coin which lands H w.p.pt
On Sunday
 he rolls a 4 faced die with faces 1,2 3,4
if coin H
The probability of it landing on face is
12

ly pyggygagyygypgngp.gg.gs.gg go.mg

You can check that 2,125 1

After Sunday he collects the amount of money as shown on the

die's face If X is the riv representing his earnings what is
the pgf of x̅

If the gambler got a H on Saturday the pgf is

s 5 t 53 54

otherwise it is
53 54 1555 56

Thus the full pgf is

G s
p s 5 t 54 t
p sts't 56

Note that we can

easily recover the pmf of from this expression
Look at the exponents of s and match coefficients

IP 2
Ep P x̅ 2 Esp P x̅ 3 Esp t p

P X 4 Ept F l
p P X 5 t
p IPC 6 l
p

Exercise for You check that it properly sums to 1

One of the biggest class of applications of mgfs is when the randomness
occurs
sequentially We've seen examples of this before e g sequence
of coin tosses

Examples A drunk person tries to walk down a straight path At each

step his probability of stepping left or right are each while his

probability of staying straight is Each step is taken independently

of all previous steps
Let X be the direction of the
step the drunkard takes at time
i
i
i Note that

i X
If If it

The pgf of is

Gyls s
Is 5

The drunkard's position after ne IN steps is Sne

É x̅ Its pgf is
determined using the pgf property discussed before

Gsis s 5
1
Suppose we were interested in computing the drunkard's expected position
after n steps Take the derivative

E Sin Gs.to n
ks s
1 52

n
k E t t 101
Thus in
expectation the drunkard manages to stay on his path
This is because each individual step has expectation 0 E X3 0

We will see later that different behaviors arise when IEFX 0 ie

the drunkard starts to drift away from his path

To compute Var Sn let's get the mgt first

Ms t Elets ettfe.tt
Take second derivative

E S Ms.tt t

ln ett e
f ket te
r
non 1 fett e.tt et fet
n
fett e
f et fet to
L
Thus Var Sn IE x̅ E x 172