' UNIT 11 APPLIED PROBABILITY I

Structure
1.0 Objectives
1.1 Introduction
1.2 Sets .
1.3 probability
1.4 Randdln Variables

1.6 Special Distributions

1.7 Convergence and Limit Laws
1 .S kt us Sum up
1.9' Key Words
I .lo some Useful B O O ~ H
. <

1.11 Answer or Mints to Check YO;^. progress

1.12 Exercises

After going through this unit, you will be able to:

@ appreciate the application of probability theory in uncertain scenario; and
formulate problems with different probabilistic distributions

1.1 INTRODUCTION
You have already seen tHe basic theory of prgbability in statistics course (i,e.,
EEC-13) as well as in qu&titative methods (MEC-03). The present discussion is
intended for a review of those themes. Since we will rely on probability-based
approaches most of the time in our discussion of financial and insurance related
themes, the topics covered below aim at providing you the basic inputs for
understanding the analysis being carried on in those fields. While developing the
study material in this u it, we have used the notations, examples and proofs given

useful. L
f'
in ,Bialas (2005) and mbens (2000). For your reference, these sources will be

1.2 SET ' ,

Bwic Definitions '

We nked to use s# theory for describidg possible outcomes and scenarios in our
model of uncertainty.
Remember that dset is a collection of elements.
1 __I/-

Notntioa: gets qre denoted by capital l.&ters (A,B,a,.. .) while their elements are
--
Qua*ti*atiueTechniclues denoted by small letters (a ,6, w,.. .). If x is an element of a set A, we w r i t e x ~ A .
for Risk Analysis .
If x is not an element of A, we write x 4.A.
There are two sets of special importance:
i) The universe is the set containing all points under consideration and is
denoted by R .
ii) The empty set (the set containing no elements) is denoted by 0. .
If every point of set A belongs to set B, then we say that A is a subset of B (B is a
superset of A). We write
AcB; b A
A=B if and only ifA_cB and B z A .
A is a proper' subset of B VAcB and A #B.
Two sets A ahd B are disjoint (or mutually exclusive) if A nB = 0
Cardinaliq of Sets
The cardinality of a set A (denoted by ( A ( )is the number of elements in A.
For some (but not all) experiments, we have
IR( is finite.
Each of (he outcomes of the experiment is equally likely.
In such cas$s, it is important to be able to enumerate (i.e., count) the number of
elements of ,subsets of i2 .
Sample spnces and events
Using set notation, we can introduce the first components of our probability
model. I

A sample Space, a ,is the set of all possible outcomes of an experiment.

An eveit is any subset of a sample space. For example,
when you roll a die once, !2 for this experiment is given by

The subiets of C? that represent the following events are:

1) The die turns up even: A= {2,4,6)
2) The die turns up less than 5: B={1,2,3,4)
3) The die turns up 6: C={6}
-
Sets of evenis
1
It is sometimes useful to be able to talk about the set of all possible events T .
The set F is very different than the set of all possible outcomes, 0 . Since an
a set of sets. Each element of F is a
event is a set, the set of all events is ~~eally
subset of SZ. .
For'example, suppose we toss a two-sided coin twice. Then the sample space is
a= (HH,HT,TH,ul,
If you were asked to list all of the possible events for this experiment, you would
i
 need to list ail possible s~~bsets
o f R , namely, Applied Probabilit?

Thus we have .F = {AI,A2,...,A16)

A
Power Set
For a given set A, the set of all subsets of A is called the power set of A , and is
A
denoted by 2 ,
If 0 is a sample space, then we often use the set of all possible events .F= 2".
When R contains a finite number of outcomes, this often worlcs well. However,
for some experiments, !2 has an infinite number of outcomes (either countable or
uncountable). As a result, F = 2'is much too large. In those cases, we select a
smaller collection of sets, F , to represent those events we actually need to
consider.

1.3 PROBABILITY
1.3.1 Probability Spaces
"
In the above, we have seen how to use sample spaces Q and events A c S-2 to
describe an experiment. In following, we will define what we mean by the
probability ofan event, P(A) and the mathematics necessary to compute P(A) for
all events A R .
For infinite sample spaces, for example when C!= R , it is not possible to
reasonably compute P(A) for all A c R . In such cases, we restrict our definition
of P(A) to some smaller subset of events, 3, that does not necessarily include
every possible subset of rR .
Measuring Sets
Let us start with some important definition's.
1) The sample space, Q, is the ,set of allpossible outcomes ofan experiment.
2) An event is any subset of the sample space.
3) A probability measure, P(.), is a function, defined on subsets of ST, that,
assigns to each event a real number P(E) such that the following three
probability axioms hold:
i) P(E) > 0 for all events E
ii) P(SZ)=I
Qua11k Live Techniques Ifthe events E a ~ Fd ure nlzltually excltlsive, then P(E u F)=P(E)+P(F)
-J
for Ri innlysis
The above three specifications are called the Kolmogorov axioms for probibility
measures. We will discuss Inore about the probability l~ieasuresin the next unit.
For the purposes of the present unit, you need to remember that the probability of
an event A for every A c 0 lies in the closed interval of [0, I ] i.e.. 0 5 P ( A )2 1.
Therefore you can take P(.)as a function which assigns to each element of the set
of events 3 ,a single number fiom the interval [o, I]. That is, P : T + [0,1].
In calculus, you worked with real-valued functions X.1. That is, f : R -,R .
Such runctions assign each number in R a single number from IR . For example,
f(x)=x3, maps x=2 into ,fl2)=8. In probability, the function, P(.), is a function
\vhose argument is a subset of Q (liarnely, an event) rather than a number. In
mathematics, we call firnctions that assign real numbers to subsets, measures.
FoI lowing such a logic. we call P(.) as a prohrrhilitll n7eavw.e.
Irzterpretutiolz rvProbnhili(y
Logical Probability. Take i-2 as a Ill~itesct
~ = ~ C O,...~,LO,,;
, C O ~
and each of the outcolnes in i2 is equall~.liliely. That is.
1
P ( { W ~ } ) = f-o r i = I . .... nand f o r a n y A c S 2 1A 1
assign P ( A ) = -
M 17

For example, consider the outcolne of rolling a fair six-sided die once, Then we
have !2={ol,w2,... ,m6) where w, = the die shows i number for ii=1,2 ,,.. ,6. So

}) = - for i= 1 ,
we can assign. for exa~nple,P ({o,
6
1
.6
.. .
3
P ({even number})= P ( { O J ~ , W ~ , ~ "=) ( -
, ) )and, in gcnerul,
G

Experimental Probabillity. Think of an experiment that can be identically

reproduced over and over again. Eacli rep1icate is called an experitnentcrl IriaE: We
also assume that the outcome of any one of thc trials does not affect the
probabilities associated with the outcomes of any of the other trials. For example,
take motors being produced by an assernbly line, with each motor an experimenlal
I
trial. The outcon~eof one of o ~ i experimer.ltal
r trials is whether that motor is either
'
good or bad. We record the outcotne as
T , = 0 if motor i is bad
T = . I if motor i is good
Thus, S2 = {bad, good)., If A = {good), the logical ,probability P ( A ) is assigned
outcotnes TI,T2,... such that
from the infinite sequence of experi~l~ental
 Subjective Probability. In some experiments, it is not possible for replication. Applied Probability 1
For example, consider the experiment of taking MEC-03, with the outcome being
the grade that you get in the course. Let
A = {you get an "A" in MEC-03)
T o determine the subjective probability of A, you need to construct an equivalent
case that is amenable to the format of selecting the probability. For example, you
may consider a lottery consisting of an urn, and an unlimited quantity of red and
white balls. The decision maker is asked to place in the urn any number of red
balls and white balls he or she desires, with the goal of making the event of
drawing a red ball equivalent to getting an "Ay'in MEC-03.
Once it is decided how Inany balls of each type to place in the urn, we simply
count the balls in the urn and assign
number of red balls in the urn
P(A)=
total number of balls,in the urn
Some Important Results
i) For any finite number of painvise disjoint sets AlrA2,...,An we have
P ( A , U A ? v...uA,)= P(A,)+P(A,)+ ...+ P ( A , )
ii) For any two events A and B

If you have a countable collection of disjoint sets,

i.e., ifA~,Az,.. . is a countable sequence ofdisjoint sets, then
I

Example. Consider the experiment of counting the number of accidents in a

factory in a month. Let Ei be the event that exactly i accidents occur in a month
with i=0,1,2,. . .. Suppose that

and we want to find the probability that the factory has at least one accident in a
month.
That is, we want to compute P(u,, E ,).

a'
Since l=O
-i( = e' for any a, we have
Quantitative Techniques 1.3.2 Conditional Probability
for Risk Analysis
The probability of an event A given the event B, denoted P(A(B),is the value that
solves ;he equation p ( A n 8 )= P ( A / 8)P ( B ) provided that P(B)> 0.

Example. Find the probability of two heads given that you have a1 least one head
in the two tosses. '

Sample space E, = { HH,T H , HT) with probability 4 and E, = (HH} with

probability + . So, ~ r ( $ =)-= 4 = j
Pr(/:,nb-)

... ,An
Generalising the above result, we have for any events AA,A2,
P(A,Az ... An)=P(A,)P(A2 (A,)P(A31 A I A ~ ) .P(An
. . I A I A2 ... An.,)
-
Remember that P(.) is ;robability measure if and only if *
I ) P(@. 2 0 for all events E
2) P(SZ)=l
3) If E n F = 0, then P (E u F)=P(E)+P(F)
,1.3.3 Independence
Two events A and B are independent if and only if
P ( ~ B)n=P (A) P (B).
That is, if two events A and B are independent and P (B) >O, then P(AIB)=P(A).
I
, Essentially it says, knowing whether B occurs does not change the probability
' ofA

.
Moreover, ifA and B are independent events, then
A' ond B' are independent events,
a A' and B ore independent events, and
a
c
A and are independent events.
Bayes' Theorem
Take a complete set of alternatives to a collection of events
BI,B2,...,.....,B~ such that
1) B,I)BtU ......LJBn=S2,and
2) BinB,=0foranyi#j.
Law of Total Probability,
Given an event A and a complete set o f alternatives B,,B2,... ,Bn,

Ba~es'Theorem: Let B,, B,,,..B,be a complete set ojulternutiv~.~,

then
 Probability I
~pplitd

Independenceof more than two events

f

We use the definition of independence given above to get results on more than
two events.
i) A finite collection of events A,,A2, ... ,A, are independent if and only if
P(A,, n. n A~,)=P(AI,)...P(A~,)
for all 2 s; j In and all I r k, c... ..< k, 5 n.
ii) For n=3, the events A / , A2, A3 are independent if and only if
P(AI n AZ)=P(AI)P(AZ)
P(Al ~ A ~ ) = P ( A I ) P ( A J )
@ P(A2 nA3)'P(A2) P(A3)
P(AI fl A2 f l A ~ ) = P ( A I ) P ( A ~ ) P ( A ~ )
All of the above must hold for A , , A2,A3to be inde,pendent.If only the first three
conditions hold, then-we say that AI,A2,A3 are pair wise independent. We will
discuss this concept with greater details in the next unit.

Check Your Progress 1

1) Write the definitions of the terms: universe set, empty set, disjointed set and
power set.

If you toss a coin twice, what would be the sample space? List all the
possible events for this experiment.

3) For a given set A, what is set of all sets?

I
Quantitative Techniques 4) Show that for any two events A and B,
for Risk Analysis

1.4 DQM VARIABLES

/

Basic Concepts
Definition. A random variable (r.v.) is a function X (.), which assigns to each
element w of IR a real value X ( a ) .Thus, we write
P((w E Q : X(w)= a ) )= P((X(w)= a ) ) = P(X = a)
Examples.
1) Flip a coin n times. Here R = {x,T)'.The random variable
X E (0,1,2,,.., n] can be defined to be the number of heads
2) Let Q = R , define the two rev.

3) Packet arrival times in the interval (o,T] : Here 2

! is the set of all finite
.
length strings (t,,t,,....t,,) E (o,T]' define the random variable X to be the
length of the string n = 0,1,.,.
Cumulative Distribution Function
Definition. For any random variable X; we define the cumulative distributian
function (COP, F, ( a ) as
~,(a)=p(X<a)
Properties of any cumulative distribution function
1) lirn ~ , ( a =) I
a+m

2) limF,(a)=O
a+4

3) F, (a) is a non-decreasing function of a.

4) Fr(x)is a right-continuous function of a. In other words, lip F,(x)= <(a)
x a

5) For any random variable X and real values a < b ,

P ( a c X I b ) = F,(b)-I;,.(n)
Discrete Random Variables Applied Probability 1
A random variable is said to be discrete if for some countable set X c R , i.e.,
X={x,,x*,...}, P f X € x } = l .
Example. From above take
TI"
n coin flips: Here R = (H, ; define the random variable X E {O, 1,2,...,n] and
it can be defined to be the number of heads.
A discrete random variable is completely specified by the probability mass
function (pmf). Thus, pmf of variable X i s given by
P,(x,)& P ( X = x , ) .
Properties of any probabili@ mass function
1) p, (x) r 0 for every x

Examples of Discrete Random Variables

* Bernoulli r.v.: X - ~r (p), for 0 5 p 5 1 has pmf

p , ( l ) = p , andp,(O)=l-p
- , 0 5 p 5 1 has pmf
Geometric rev.:X ~ e o m ( p ) for
P x ( t ) = p ( l - p ) k - l , for k = l , 2,....
This r.v. represents, for example, the number of coin flips until the first
heads shows up (assuming independent coin flips)
-
Binomial r.v.: X B(n, p ) , for integer n > 0, and 0 2 p 5 1 has pmf

This r.v. represents, for example, the number of heads in n independent

coin flips.
-
Poisson r.v.: X Poisson (A), for1> 0, has prnf

px(t)=rl*e-ay fork, .=0,1,2,...

R!
This represents the number of random events in interval (O,I], e.g., arrivals of
packets, photons, customers, etc.
We will discuss these distributions elaborately along with some others
subsequently.
Contiauous Random Variables
A random variable X is said to have a continuous distribution if there exists a
nonnegative fundion f such that P(o i x r b) = If
( x ) h for every a and b .
To specify a random variable it is sufficient to specify its probability density
function (pdt). The pdf of a continuous random variable Xis given by
Quantitative Techniques
for Risk Analysls

Properties of any probability density function

1) fi(x) 2 0 for every x

Note that ifX is a discrete' random variable, then

P(X < a ) = P(X < a) and
ifX is a continuous random variable, then

Examples of contittuous distributions

Uniform distribution
The random variable is said to have a uniform distribution if it has the
probability density funption given by

.
fx(x)=
, :[b- a

Exponential distrib,ution
fora~xlb
otherwise

Tlhe random variable X is said to have an exponential distribution with

parameter A >0 if it has the probability density function given by
~e-" forx 2 0
0 otherwise
- .
Gaussian ~isiribution
Gaussian r.v. : X - N ( ~a',) has pdf

where p is the mean and a is the standard deviation

Transformation of Distribution Functions
Take a random variable X with distribution function F, ( x ) and pmf or pdf
/,(x), where f, (x) > 0 for x e X . Suppose that we are interested in the
distribution of a functi0.n of Y =g(X). The distribution of Y is defined by the
equation,

14
 P ~ ( YE A ) = P ~ ( ~ ( ExA)) = PI(@ EQ( C J ( X ( . ~E)A) ) . Applied Probability It

This is not very helpful, as we have not specified the actual experiment and the -
random variable. The question is how we can go from these functions for X to
the corresponding functions for Y. Let us discuss the general case for such
transformation:
Result 1. Define the mapping g-' ( A ) as:

( A )= ( xl g ( x ) E A ] .
9-'
Then

However, we need more useful results to rely on transformations that satisfy

particular conditions:
Result 2. Suppose ( x ) is a strictly monotone functions with inverse g-' (.). Lel
be ~ b e { ~ ~ ~y (f ox r)s =o m e x ~ ~ ) . '

( If X is a discrete random variable with pmf f, ( x ) on X,then Y is a

discrete random variable with pmf:

(ii) If X is a continuous random variable with pdf fy ( x ) on X , then Y is a

continuous random variable with pdf

on Y. I

In case of continuous variables, we proceed as follows: assume g(.) is increasing,

so that

*
= &(x 5: g-I ( Y ) )= F, (g-' ( y)). . .

Then use the, chain rule to take the derivative with respect to y you get the result. * :?I

Example 1. Suppose X has a Poisson distribution vvith paranleteiil . That is, the
pmf of X is

fx(x)=
,/ x!
x = o , I,,, ...
.
.,I
.

lo otherwise,
for some positive 2 . Usually we use Poisson distribution for counting events such
as arrival of insurance claims.
-, (-
Qu'ntitaave Techniques Take the distribution of Y = g ( ~= )2. X. Then inverse of this transformation is
for Risk Analysis
: X.=g-'( Y ) = Y / 2 and the pmf of Y is'given as

lo otherwise.
Example 2. Let X be an exponential distribution with pdf

jX (1' = {S x>O
otherwise. +

Then cdf its distribution is F, (x) z 1 - e-' . Units this distribution, when we write
f, ( x ) = ;lexp(-AX), for positive A , we can use it is widely used for modelling
mortalities.
Uniform Distribution
Consider the distribution of Y = c ~ ( x ) = l - e - ~This . is a monoione
transformation with inverse X = g-I(Y) = In (1 - Y ) . Then,

The range of Y is y = (0,l) and the pdf of Y is

Note that its pdf is constant over its range. ,

Example 3. Consider a non-monotone transformation. Let X be a random
variable with pdf

(0 otherwise,
It has a uniform distribution on the interval [-I, I]. Consider the distribution of
non-monotonic formal action Y = g ( ~=)x2. Because of the non-monotonicity 1
i
we have to do this on a more ad hoc basis. In this example we work directly
through the cumulative distribution functions. For its elements we can split things
, I
up into intervals where the transformation is monotone. First, we calculate the cdf
for X : .

Then the cdf for Y = X 2is

 Applied Probability 1

Then the pdf is

O<y<l
otherwise.

Check Your Progress 2

I) Write the definitions of CDF and PMF.

..........................................................................................
2) How do you write the PMF of a ~ernoulli'random variable?

..........................................................................................
. 3) What is PDF? Write down its properties.

.................................
+,,...d ..............................................
4) How do you write the PDF of
(i) exponential distribution
(ii) Gaussian distribution

..........................................................................................
5) What is meaning of transformation of distribution function?
Quantitative Techniques
for Risk Analysis

'5) What is meaning of transformation of distribution function?

1.5 EXPECTATIONS - - -

The ekpectation of a random variable X is

(g, X is o discrete random variable with pmf fx (x), equal to:

provided the sum Z, I x ( f, (x) exists, .

, .

(iii,if X is a continuous random variable with pdf fx (x), equal to

provided the integral 1x1fx ( x ) dr exists,

Example 1. Suppose you tow a single die. Let X , be the number (x) on top of
. the die, has a distribution withpmj
116 x = 1,2,3,4,5,6,
otherwise.
Then the expected value is

Example 2. Suppose X has an exponential distribution with parameter A and pdf

A exp(- Ax) x >0

0 o therw ise:
As we have seen above the exponential distribution'with X = 1 can be used to fit
distributions of durations such a9 waiting times of insurance claim arrivals. Its
expected,valueis
 Applied Probability I

-1
A exp ( - A X )
=-

If you are interested in finding expectations of transformations of X , then the

expected values for transformations have to be defined implicitly:
E [ ~ (=xE)[ ]Y ] ,where Y = r ( ~ ) .
That means we do not need to define expectations o f ,transformations. To
calciilate these expectations we take the help of the following results:
Result 1. The e-rpectarion of a function r (.) of a random variable X is
fi), fi X is a discrete random variable wifhpt-rfJY (x),

(ii), if x is a continuous random variable wilhpqY S;, ( x ) ,

For the discrete case with a monotqne kansformation:

For the continuous case with r ( + ) ,a monotone (increasing) transformation, is

given below:
Let Y = r (X). Then

Transform the integrand from y to z = r - ' ( y ) with inverse transformation

j)= r ( z ) to get

For the last step, we have taken the following: 'since z = r - ' ( r ( z ) ), by the ohain
rule for differentiation we differentiate both sides with respect to
Quantitative Techniques dr-' dr ( z )
for Risk Analysis zz 1=-
(4
( z ) )= I--
ar
az r ( 2 ) .

Note some special expectations. In all such cases X is a random variable with
pdflpmf equal to f,y ( x ) :
Mean is another name for the expectation or expected value o f x ,
I).
p = E(x).

2) K"' rnomenr of X is /r, = ,5(xk)(i.e., p, = p ) .

4) Moment generating function (mgf), denoted by .I!,. (1). It is the espccted

value of exp(tax).We are interested in this function for values o f t around
zero. It has a number of interesting properties:
(a) It uniquely defines the distribution of a random variable: if two
random variables have the same mgf for all t in an interval around
zero, they have the same cdf and pmflpdf (up to sets of measure zero).
(b) K'" moment is equal to k"' derivative of the mgf evaluated at zero:

dkM , ~
In particular, using M,!, ( t ) as shorthand for -(r), (0) = I ,
we have M.,
at

5) Cumulative generating function is the log of the moment generating

function, K , (1) = In M, (f). Here / = K,:. (0) and (7' = K.:. (0).

6) Characteristic function is defined as

w,y ( t ) = E [exp (Xit)],
where i = fi.The advantage working with. the characteristic function
. rather than the moment generating function is that the fornler always exists,
while the latter does not exist.
7) Expectation of a linear function of a random variable is the linear function
of the expectation:

, 8) Variance of a linear function of a random variable is the variance of the

random variable multiplied by the square of the slope coefficient:
v(~+~*x)=~~.Y(x).
 ) x>0
Example 3. Consider the exponential distribution with pdf ~ e x ~ ( - ; l xfor Applied Probability 1
, gnd zero otherwise. We have already calculated its expected value above. Using

the moment generating function, we can derive its expectation and avoid the
procedure integration by parts.
Thus,

However, it only works for t K A . If you are interested only about values of t
around zero, you can use this result. The derivative of the mgfis

M,;.( t )k !
a
(A - '
so that

Its mean is 11A, the second moment 2 / A', and the variance 1 / A*.
Example 4. Suppose that an experiment can have one of two outcomes, success
or failure. Let p be the probability of success. Let Y be the indicator for success,
equal to one if the experiment is a success and zero otherwise. This ,is called as a
Bernoulli trial. When we repeat this experiment n times, and assulne that the
repetitions are independent, we have the following result:
Let for i = 1, ...,n denote the success in the it11 ' ~ e r n o lul i trial. Define
I
X= x" (=I
Y, as the total number of successes. Then X has a binomial
distribution. To find its pmJ; consider the probability of particular sequence of x
successes and n - x failures. For example, first x one's and then n - x zero's.
The probability of any one such a sequence is pl(1- p)"-'. To get the prlobability
of x successes and n - x failures, we need to count the number of such
sequences. This is equal to the number of ways you can select x objects out of a
set of n, which is
(3 . Hence thepmf of X is

for x = 0, 1,2, ..., n, and zero otherwise.

'
The mean of the binomial distribution is given by prw$
Quantitative Techniques Therefore,
for Risk Analysis

Let m = n - I and y = x - 1. Then instead of summing from x = 1 to x = n - 1 we

take the sum from y = 0 to y = m and get:

Another approach to get the above result is to write X as the sum of independent
zn
and identically distribution random variables, X = /=I I; , .with

and mean p. Then the mean of X is given as

Result 2. Chebyshev's Inequality. For any random variable Y with mean pand
variance a 2, andfir any k > 0

Proof. Let X be a nonnegative random variable. Then, for any positive c,

Taking the continuous case, we get the above result as,

E ( X ) =, C x f x ( x ) &
When X = ( Y - p ) 2 and c = k 2 d , we rewrite the inequality as Applied Prob?bility 1

P r (X2 C) 5 E [ x ] / c

~r( ( Y-,u) > k 2 o 2 ) s 1 / k 2 ,

1.6 SPECIAL DISTRIBUTIONS

1) Gamma Distributions. In contrast to an exponential distribution
representing the waiting time till the occurrence of first event, Gamma
distribution represents the waiting time till the rth event. If X is this
waiting time, then
P', (x) = P ~ ( 5X x ) = 1 - P r (less than r events in interval [0, x)).
'Thus, if we consider the probability that a Poisson random' variable with
arrival rate Ax is less than r, then

, The probability density filnction can be obtained by differentiating with

respect to x :

Note that the second summand is only from k = 1 tor - 1, not from k = O
to r. - l . Changing the second summation from k = 1 to k = r - 1 the
summation from k = 0 to r,- 2 , we can.write this as

So, we get a gamma distribution with parameters r'and A .

The gamma distribution is more general. We don't require r to be an
integer while formulating it. But when we do this, the interpretation o f
waiting for the rth 'event cannot be made.
The
. .
distributed random variable X is obtain
general''pdf'fpr 'a.,~arqrn+
. . I .. . ,. ,, ! .
5
#<,, , ;
' ' .,: .;
.;.
P=--&d a:;=,,r: .:':.:
' I ,' ,

when 1 ' . . .
a , . ., ! . .* . ._ ,
.I.

\ . ....
. .

for x r 0 . In this formulation the parameters a: a n d P are both positive

and the function is defined as

r(a)= p ta-ie-rdl.
Quantitative Tech j f PS It's properties are:
for Risk Analysi.,
r ( a + l ) = a . r ( a ) , ~(l)=$e-'dt=l,and ~(1/2)=&,

s o that
,r(rt)= (n- I)!, ,
/

-- -
______-
for integer n. Tdke a random variable Y with an exponential distribution
with arrival rate A . Then its moment generating function is
A I
M , (t) = E l e x p ( t ~ )=] -=--
A - t 1-pt '
where p=I/A is the mean of the exponential distribution.
Let Y;, ...,Y, be independent exponential random variables with mean 1.
Then the momentgenerating function for the gamma distribution is

The cumulant generating function is

K, (t) = -a In (1 .-pt) ,
with
alp
K; (t) -.-
1-pt ' ~ : ( t )= (1-pt)2
a@ *

The mean and variance are a;O and ap2respectively.

2) Chi-squared Distribution. Chi-squared distribution is a special case of
Gamma distribution if we take a = k / 2 , where k is a positive integer, and.
p = 2 , than we have a Chi-squad distribution with degrees offireedorn
equal to k. Itspdfis

for x positive. Its moment generating function is

and its mean and variance are k and 2k respectively,

3) Normal Distribution. The normal distributiofi is of fundamental
importance as an approximation to a large number of statistics through tht
central limit theorem. A'random variable X has a normal distribution
with parameters p and v 2 ,denoted by N ( , u , ~ if) itspdfis
,
 for --a < x < a,and cr2 > 0 . If we consider the molnent generating Applied Probability I
function, then the fact that thepdf andpmf add up to one; viz.,

- 1 ( ( x - p)2- 2axtl
-Ezexp(- 2a2 Jdr

exp f t' 02/2+ pi - ( x - p - 02 2 t 2 ) )

-E,=
- I
( 20 J "

1 As the curnulant generating function is K , ( t )= pt - 0 2 t ' / 2 and hence the

I
1
- mean is p and the variance cr2.
Properties of normal distribution: Linear transformations of normal
1
random variables are also nor~nallydistributed. Talce a random variable
I
i
X with a ~ ( p0 ',) distribution, and consider the transformation
I Y = a+ bX . Then,

= e'" M,y (bi) = exp( a + pbt + cr2b2t2/2)

I

) = exp( ( a+ bp)t - b2o2r2/2)

Hence Y has a normal distribution with mean G , = a+ b p and variance

6' = b2a\ When you take Y = ( X -,u)/o, then the transformation is
-
callqd Y N (0, I ) , the standard normal distribution with mean zero and
.unit variance.
See that the normal distribution and the chi-squared distribution are
closely connected. If X has a standard normal distributionN(~,l),then
Y = X 2 has a Chi-squared distribution with degrees of freedom equal to I

one. If we work with the moment generating function of Y,then

: Quantitative Techniques
for Risk Analysis

We get the mpJ for a Chi-squared distribution with degrees of freedom

equal to one.
4) Cauchy Distribution. A random variable X has a Cauchy distribution
centered around 8 if it has pdf . %

for -CQ < x < a.This distribution has no moments. The median and mode
are both equal to 8. The pdf is similar to the pdf for the normal distribution
with a thicker tail. We use it for modelling variables with high kurtosis,
that is, which infrequently take on extremely large values, such as stock
prices, event like loss in Tsunami. An interesting property of the Cauchy
distribution is that if the sequence of independent random variables
X,,X,, ...,X,, all have the Cauchy distr,$ution centered around 8, then the
Z''
average 1=. !=I X , / n also has that same Cauchy distribution centered
around 0, with the same quantile. In ,other words, laws of large numbers
, will not to apply to the ones like Cauchy distributions.
5) Beta Distribution. If two independent random variables Y; and Y, have
Gamma distributions with parameters a and 1 and O , and 1, then the ratio
X = Y;/(Y, t Y,) has a Beta distribution with parameters a arid/?. Itspdl
is

for 0 < x < 1 and zero elsewhere. The mean and variance are a/(a+ p)
and ap/((a+p)' ( a tp+ I)) respectively. For modelling variables that
take on values in the unit interval, we can apply this distribution.
6) Let X have a standard normal distribution (i.e., a normal distribution with
mean zero and unit variance). Let W have a Chi-squared distribution with
r degrees of freedom, and ,let X and W be independent. Then
Y= ~/,,@q has a t -distribution with degrees of freedom equal to r.
The probabili-Q density function of Y is:

for -a < y < a . The t -distribution .has thicker tails than the normal
distribution. As the degrees of freedom gets larger, then the t-distribution
gets closer to the normal distribution. (For E[W l r ]= I , and
, V ( WI r) = 2 l r , as r -+ a , the mean of the denominator is one, but the
variance is zero.)
7) If V and Ware independent Chi-squared randoin variables with degrees of Applied ~robabilitiI
freedom equal to r, and r2 respectively, then Y = ( V l r , ) / ( /~r 2 ) has an
F-distribution with degrees of freedom equal to r, and r, . The probability
density function is .

for positive values of y, and zero elsewhere.

1mportant.Exponential of Distributions.
Definition 1. A family of pdys (or pmj") is an exponential family f it can be
wrillen us

for -oo < x < a.

'The key feature of this representation is that we can separate the distribution into
functions of the parameters 0 and functions of the variablesx. If we parameterise
the above exponential distribution as

for -a3 < x < a,then to q as the natural parameters. Consequenlly we have cases
like:
1) Binomial distribution: .

for x = 0, I, 2, ..., n, and zero otherwise. This can be written as

so, k = l , a n d
 Quantitative Techniques 2) Poisson distribution:
for Risk Analysis
f x ( x ) = ~ ( ~ 2, ...}I.(
~ ( l , ~/x!).exp{-l)exp(xlnl).
3) Exponential distributio~:
~y(x)=~{x>~)~~-exp(-xl). '

4) Normal distribution:

I A distribution that does not fit in the above group is the uniform distribution:
I
.fX(x)=1{a<x<b].-.b - a

In case of indicator iirnction 1 {a ix < b} you cannot factor into a function of the
parameters and a function of the variable.
Joint Distributions Conditional Distributions,
Independence of Random Variables
Definition 1. Let any N - dimensional random variable be afirnction from the
sample space to R * .
Definition 2. Take (x,Y) be a discrete bivariate vector. Then the function
F,, (x, y ) from R 2 to R , defined by

is the joint probability mass function of (x,Y ) .

Example 1. Toss a coin three times. The sample space is
-0{ HHH, HHT, HTH, HTT, THH, THT, TTH, TTT] .

An example of a bivariate random variable ~ s ( x , Y ) ,where ~ ( ois )the number

of heads in the first two tosses, and Y(U) is ihe number of heads in the two
tosses. Assuming the coin is a fair one, and the probability of heads is 112 at
every toss, with different tosses independent events, the joint prnf is
118 if (x, y ) = (0,0) ( U E (TTT}),

Ills if (x,Y) = ( o , ~ ) ( U E (TTH)),

118 i f ( ~ , ~ ) = ( 1 , 0 ) (0E (HTT}),

2/8 if (x, y ) = (1, I) ( u E { HTH, THT}) ,

118 if (x, y ) = ( 1 , ~ ) ( u E ( TIYH}],
118 if (x, y) = (2,l) (0 E { HHT)),
118 if (x, y) = ( 2 , ~ ) (W E (HHH}),
0 otherwise
 Norr ihat even was we define two random variables, the definition of the Applied Probability I
;1*uoa4ili.y Tass function for any of the variable does not chhnge. Ignoring the
dcfiiilfion of Y ( w ) ,the padfor x when you do not consider x lo)is

h ( ~ )14, =
I 114
214

0
ifx-:O
ifx=l
ifx=2
elsewhere.
(0E(TTT,TTH}),
(~E{HTH,H~T;THH,THT)),
HHT, H H H ) ) ,
(o-;{

Similarly we can calculate the pmf for Y without x . In case of joint random
variables, we typically refer to these variable probability mass functions as
marginal probability mass functions. These functions general they can be
calculated from the jointpmf as

Similarly,

Given the joint pn$ we can calculate probabilities involving both random
variables. For example,

Alternatively we can have joint continuous random variables:

Definition 3, Let ( x Y, )
be a continuous bivariate vector. Then the function
xuv ( x , ~ from
) R' to R is the joint probability densiv function of ( x Y, ) if il'
satisJies

JA f, (x,y)dxdy = p r ( 0 E Q I ( X ( @ ) , Y ( ~ EA),
))
j o r ail sets A.
,
Example 2. Pick a point randomly in the triangle (0,0),(0,0),(0,1), (40).Let X
be the x,-coordinate and y be the y -coordinate. A reasonable choice for the joint
pdf appears to be

Q
for ( ( x , y) I 0 < x < I , 0 < y < 1, x + y < 1) and zero otherwise (that is, constant on
the domain).. What is the appropriate value for c ? Since we know that the
probability of the entire sample space is equal to one, we can write
 j,6
1 I-y
Quantitative Techniques.
for Risk Analysis = ~(r.y,,o<r<l.o<"<l,I+Y<Il
fH ( X , Y ) ~ ~=Y C ~ X ~ Y

Hence c = 2.
Another way of writing such integrals differently is:
I I-y 1 I-x

~ i x . y U o ~ x ~ l . o . y ~ ~ .g ~y)dxdy
I + Y( ~ ~~, = 66 g (xlY ) dxdy = 6 9 (xyY) d y h ,

depending on the order of integration, ivithout changing the result. In such a case
, the marginalpdfs are defined as

f, (x) = p fn
-m
( x , y ) dy = f-''2yy = 2 - 2x,
for 0 < x < 1 and zero otherwise, Similarly,
fy(Y)=~-~Y, 0 < y .= 1, and zero otherwise.

In joint random variable case, expectations are calculated in the same way as
before, but using double sums or integrals at present:

,:<
E [ ~ ( x ,y ) ]= j J
J r(xl Y)/,
Y
(x,Y) dY&,

for continuous bivariate random variables, and

E[~(X,Y~]=CC~(~,Y)~~(X,Y),
x Y
, 6 ,
for discrete bivariate randon! variables.
Example 3. Consider the expected value of XY in the previous example:

"-:I
~ ( ~ ~ ) = ~ $ ~ ~ 2 sd x~= l xd( l ~- x )d' d xx ~ ~ x ~

1',.2j I
I

= l x 3 - z x z + x a = - x4 --x
3 +-.'(l
2 o =-!-12'

Recall how the conditional probability of events was defined:

~ r (n~~ 12 )
"(E, I 4)= pr(E2) * 3

provided Pr (E,) 0.
Case of bivariate discrete random variables.
Definition 4. Given two discrete random variables X and Y wilhjoint probability
mass function S,, (x, y ) , the conditional probability mass function for X given
Y = y is

A,,( x I Y )= fH ( X ~ Y )
fY (Y) '

30
provided fy ( y )> 0. Applied Probability I

Example 4. Consider the distribution of X given that Y = I . In that case we are

conditioning on w E {HTH,HH T , THT,TTH}

[ 0 elsewhere.
Note that this is the same as the marginal pmfof X and will not be' true if we
condition on Y = 2 . In that case we condition on w E {HHH,T H H ) such that.
I
112 x = l (0E ( T H X } ) , I
JYIY (XI Y = 2) = (O E { H H H } ),
0 elsewhere.
Definition 5. (Continuous r.v.) Given two continuous random variables X and
Y with jointoprobability density function f, ( x , y ) , the conditional probability
density,function for X given Y = y is

To see' this formulation we first take the conditional cumulative distribution

function where we condition pn Y ~ ( y , p +A ) , which' does have positive 1

probability for nonzero, A :

-00
It"fm ( u ,v ) dvdu .

!i
Now take the limit as A + 0 . In that case both numerator and denominator go tlo 1i

zero, so to get the limit we have to use. L' Hospital's rule and differentiate both 11is
numerator and denominator with respect to A . In that case we get: : If

Take the derivative with respect to x to get the conditional probability density
function:
fm (x9.J))
JYIY (xIY) =
fY (Y) '

31
 Quantitative Techniques However, this does not yield the conditional density of X given Y = y . Its
for Risk Analysis
inte~retation is that of the limit .of the conditional densities, conditioning on
. y < Y < y + A, and taking the limit as A goes to zero.

Definition 6. Let (x,Y) be a bivariate random vector with pmfpdf f, (x, y),
~ Then the random variables X and Y are independent if

where f , (x) and f , (y) are the marginalpdf/pmPs of X and Y .

Result 1. The two random variables X and Y are independent if and only fi the
joint pdflppnlfcan be written as
J;Y (xYY) = g(x),h(y),

forall -co<x<ca and -c.o<y<m.

Proof:
< ,

If X and Y are independent, then we can choose, and g(x) = f,(x), and
h(y) = fy(y)' and the result is trivial. N'ow suppose we can factor the joint
density of X and Y as f, (x, y) = g (x). h ( y ) . Then the marginal density of X

Similarly

r,0) Ifm ( x . Y ) ~=. h(y) la(?)&.

=

Multiply f ,(x) and f,(y) ,,we get, -

. >

Result 2. Let X and Y be two independent random variables, Then

I
I
'(9 fX, (X I Y) = f X (x) and
J
II

. Example 5. In the previous example we had I

f ~ ( x , y ) = 2 ~ 1 { 0 < x < 1 } ~ ( 0 < y ~ l } ~ l { x tandzeroelsewhere.

y<~],
It may appear that X and Y are independent by the previous results: Choose
g(x) = 2 and h(y) = I . The reason thii does not work is because of.the '
restrictions on the values of X and Y . To see this krite
 1
I

for all -co < x < co and -a iy < co . The indicator fuhctian 1 { A } is equal to one Applied Probability I
if the argument is true and zero if it is fails. This indicator function l { x + y < I}
cannot be separated into a function of x and a function of y and therefore X
and Y are not independent.
Let there be two random variables X , and Y with means p, and /I,and
variances 4. and ot respectively. Then the covariance of X and Y is defined
as
C(X,Y) = E[(x-P,Y).(Y-PY)],
and the correlation coefficient as

'The niean and variance of the sum X + Y are

E [ x + Y ] = ~ ,i p , . ,
and

If the two rando~nvariables are independent, the cohriance is zero (note that the
reverse is not necessarily true), and the variance simplifies to the slim of the
variances cr;, + 0;.
To see how tricky conditional probability density functions can be with
i continuous random variables, consider the Borel-Kolrnogorov paradox. The
joint density of two random variables X and Y is

for x , y > 0 . The marginal density function of X is

'The conditional density of X given Y = 1 is

( x I y = 1) = exp(-x) ,
'because X and Y are obviously independent.
Taking the transformation from (x,Y) to (x,z) where Z is ( Y -I)/x.The
inverse of the transformation i s X = g , ( x , z =) X and Y = ~ , ( x , z=)XZ + 1
with Jacobian (the correction factor)

the joint density af Z and X is

Quantitative Techniques for z , - I / ~ and > 0
for Risk Analysis
Let us calculate the margin4 density of z . For z > 0, the marginal density is

For z c 0, the marginal density is

. .
Take the limit as z 10 and z ? 0 to get in both cases f,(0) = e-' :
The conditional density of xCgivenZ = 0 is then

for x > 0, The parado? is that Z = 0 implies, and is implied by, Y = I , yet the
. conditional density of X given
Z = 0 differs from that of X given Y = 1.
The above results are due to the way the limit is taken. In one case we take the
limit for 1< Y < 1 + 8. &nthe other case we take limit 0 < Z < 6.The latter is the
same as the limit 1 < Y <1+ 6 x which kckarly different from the first limit. For
fixed 6 the latter [o&itiariihg set obviously includes more large values of X .

1.7 CONVERGENCE AND LIMIT LAWS

We have already seen that the mean and variance of two independent random
*,variablesX and Y :
E [ x + Y ] = E[x]+E[Y],'
and -- " 1

V ( X + Y )= v(x)+v(Y).
..., X, are N independent random varihbles, we have mean
More generally, if X,;
 Applied Probability I

and variance

Note that we only use the independence for the variance. Without independence
the mean is unchanged, but the variance becomks

Independence insures that all the covariance terms in the double sum are equal to
zero.
Supposing all random variables to be independent and have the same mean p,
variance 0 ' . Then the sum has mean
r~ i

and variance

The mean and variance of the average are

and

Examice the behavior of the average of the fi;st N random variables,

F, = cXl/N as N gets large. In that Ease the variance of the sample average
,=I
gets smaller and smaller. Using Chebyshev's inequality, this implies that the
. .
QuantitativeTechniques probability that the sample average is more than E away from p can be made
for Risk Anqlysis
arbitrarily small by taking N large enough, if we fix E , then

which can be made arbitrarily small for fixed E by choosing N large enough. It
would seem uncontroversial to say that in this case X, converges t o p .
In other 'cases this result is not so clear. For example, consider the following
sequence of random variables X,, X,, ... with the pdf of X, equal to
-Iln,x<lln,
n<.~<n+l,
elsewhere

The mean of xll is 1 + 1/2n. l'he variance increases with n and approaches infinity
as n goes to infinity. However, the probability that XI, is more than c away from
zero is at most l / n . Thus, XI, does not converge to its asymptotic mean of 1.
This example demonstrates the need for different concepts of convergence. We
consider three such concepts.
Definition 1. A. sequence of random variables X,,
converges to kt in probability
ifforall E > O ,

Definition 2. A sequence of random variables X,, converges to p in yzmdratic

mean if

Definition 3. A sequence of random variables XI, converges to ;L alrnost szlrely if

forall E > O ,

Note that in the first example the convergence is clearly in quadratic Inearl and
probability. The independence implies it is also convergence almost surely. The
following relations hold between the different convergence concepts:
(i) Con\;ergence in quadratic mean implies co!ivergence in probability.
(ii) Convergence almost surely implies convergence in probability.
(iii) Convergence in quadratic mean does not imply, and is not implied by,
convergence almost surely.
Difference between convergence in quadratic mean
and convergence almost surely. ,
Consider the following.sequence of random variables, defined as X,,( w ) , for
m € 0= [0,1], and with the probability of o in some interval a, b) equal to b - a
for O ~ a l b ~ l :
 X,(a)
= 1 for o E [o, I] and zero elsewhere, ' Applied Probability I
I
X 2 (o) = I for w E [0,1/2] and zero elsewhere,
1
X3 (w) = I for w E [1/2,1/2 + 1/31 and zero elsewhere,
:I
A', (o) = 1 for o E [1/2 + 1/3,1] 1) [o,11121 and zero elsewhere,
Xq(a)= 1 for o E [1/2 + 113,I ] u [1/12,1/12+1/5] and zero elsewhere,
For .~,,(o)the intervals where x,, (u) is equal to one have length l / n . That
mealis probability is l / n . 'They shift to the right till they hit 1 , and then start over
again at 0. Then the lnean is I/n, and the variance is l/n - l/n2. Both of these go
to zcro, so that we have convergence to zero in quadratic mean and probability.
Consider a particular value of w in the sequence of values X, ( w ) , X, ( w ). This
sequence does not converge, No matter liow large n , the sum I=n
Em
l/i diverges,
implying that the sequence is always going to return to 1. Hence the probability of
an w such that the lirnit X,,(a)even exists, let alone that it equals zero, is zero,
and not one as required by allnost sure convergence.
With these convergel-tce'concepts we can formulate laws of large numbers.
Result 1. Let X ,, X,, ... be a sequence of independent and identically distributed
random'vmiubles wit!, cornnmn mean tr and vdriance cr2 =
. Let XN 7/ = I X , / N
he the average up to [he AT'" r a ~ ~ d ovariable.
m The17
fill1
JN P.
The result implies convergence in probability as well, as we already showed using
Chebyshev's inequality.
A second result produces an even stronger.for almost sure convergence, That is;
convergence in quadratic mean does not necessarily hold because the variance .
does not necessarily exist,.
Result 2. Lel X,, X2,... (be a sequence ofindependent and identically distributed
rrr,do,n vnrirrbles with common mean p . Let zN= x-x,
N
be the average up to
'
/he N'"random variable. Then
.TNa'Li .
Definitioln 4. A sequence of random variables X , ,X,, .,, converges in distribution
to m rcndtom variable Y , f a t all continuitypoints of F, (y)l
nm F~~( y )= 6,
II-+m
(y).

The restriction.:to continuity points is for the following reason:

supposc
f,,,( x )= n, for 0 < x < l/n,
. Quantitative and zero elsewhere. The value of Fxn(0) is zero for all n , but X,converges in
for Risk Analysis
distribution to a random variable Y with Fy (0) = I + limn,. Fxn(0).
If a random variable converges in'distribution to a (degenerate) random variable
with all mass in a single point ,u, the random variable converges to ,u in
probability, and vice versa.
Central Limit Theorems:
Result 3. Let ;x,,
X,, ... be a sequence of independent and identically distributed
random variables with moment generatingfinetion M x( t ) in a neighborhood of
zero, and witi mean p and variance a'. Then

That is, the (ormalized sum converges to a standard normal distribution.

Slutsky's ~Georem:
1

Result 4. If; X, converges in distribzction to X , and X,, converges in probability

to a constnet c , then
x,,+q-,ctx.
and

/q,"-,x / c .
x,,
The last result is known as the delta method:
satisfies -
~ e s u l t ' 5I. f a sequence of random variables X,,

then for any function g ( - ) continuously differentiable in a neighborhood of p

with derivative c ~ ' ( p ) ,

Check Your Progress 3

. . /-
I) Write the rtieaning: '

I .
i) cumulative generating knction; ii) characteristic function

- ,
.......................................................................................... i
-
.................,.......................................,............,.,................
/

I . . .
 2) Explain the meaning of Chebyshev's inequality. Applied Probability I

'

3) Which distributions come under the family of exponential distribution?

4) Write the meaning of independent bivariate random variable.

5) Write themeaning of random variables X, conversing to be almost surely.

-
1.8 LET US SUM UP
.-
h This unit reviews the im'portant concepts of probability which find application in
actuarial problems . The basic set operations, union and intersections, are given
that help describe possible outcomes in model building where component of
uncertainty forms a part. The ideas like universe, empty set and set of events that
are required for probability space have been documented. The concepts like
'conditional probability and independence have also been included and their use in
the context of random events, discrete and continuous, have been highlighted. The
meaning of uniform, exponential and Gausian distributions have been used to
explain the transformation of distribution functions. Use of Poisson distribution to
count events has been pointed out. Expectation of a random variable formulated
for discrete and continuous cases has been examined. Use ~fpmfandpdf has. been
considered. Some special distributions viz., Gamma Chi-squared, normal and
Cauchy, along with the family of exponential distribition have been included.
QualltitativeTechniques The joint a*d conditional distributions of independent random variables and the
for Risk Analysis
convergence and limit laws constihrte the last part of discussion in the unit.

1.9 KEY WORDS

Borel s Paradox: A conditional probability distribution that runs contrary to
f

intuition: that conditional probability density functions are not invariant under
coordinate transformations.
Characteristic Function: Function that takes the value I for numbers in the set, . '

and value 0 for numbers not in the set.

Chebyshev s Inequality: A probability distribution where all the values are close
f

to the mean value

Conditional Probability: The probability of some event A , given the occurrence -
of some other event B.
Cumulative Distribution Function (cdf): The function that completely describes
the probability distribution of a real-valued random variable.
i

Delta Method: A method for deriving an approximate probability distribution for

a function of a statistical estimator from knowledge of the limiting distribution of
that estimator.
Discrete Uniform Distribution: Eele~nentsof a finite set that are equally likely.
Exponential Distributions: A class of continuous probability distribution often
used to model the time between independent events that happen at a constant
average rate.
Indicator Function: A function defined on a set X that indicates membership of
an element in a subset A of X.
. .k Joint probability: The probability of both events together.
Marginal probability: One event, regardless of the other event.
Moment-Generating Function: A function used to generate the moments of the
probability distribution..
. Poisson Distribution: A discrete probability distribution which is the limiting
I
form of the binomial distribution, provided 1) the number of trials is very large in
fact tending to infinity; 2) the probability of success in each trial is very small;
tending to zero.
Probability Density Function: A continuous random variable having a
probability distribution f(x) which is a continuous non-negative function that
I
gives the probability that the random variable will lie in a specified interval when
(1
integrated over the interval, i.e., P(c5 x~ d) = f (x) . The function f(x) is called
c

the probability density function (p.d.f.) provided it satisfies following two

conditions (i) f(x) 2 0; ( ii)the range of the continuous random variable is (a,
b

b1I.f ( x ) = 1 .
0

Probability Mass' Function: Often. the probability of the discrete random

variable X assuming the value x i s represented by f (x), f(x) = P (X = x) =
probability that X assumes value x. The function f(x) is known as the prabability
 I

mass function (p.m.f.) given that it satisfies following two conditions: 1) f(x) z 0; Applied Probability I
2) fI = ! f ( x , ) = l .

1.0 SOME USEFUL BOOKS

Bialas, Wayne F. (2005), Lecture Notes in Applied Probability, Department of
Industrial Engineering, University at Buffalo (see internet)
Freund J.E. (2001), Mathematical Statistics, Prentice Hall of India
Hoel, P (1962), Introduction to Mathematical Statistics, Wiley John & Sons, New
I
York
Imbens, Guido (2000), Probability and Statistics (Lecture Notes), UCLA (see
in ternel)
~ l k i nI.,
, L.J. Gleser; and C. Derrnan (1980), Probability Models and
Applications, Macmillan Publishing, New York

1.11 ANSWER OR HINTS TO CHECK YOUR

PROGRESS
Check Your Progress 1
I) See Section 1.2 and answer.
2) See Section 1.2 and answer.
3) See Section 1.2 and answer.
4) Read Section 1.3 and answer.
f

Check Your Progress 2

1) See Section 1.4 and answer.
2) See Section 1.4 and answer. . .

3) See &ection 1.4 and answer.

4). Section 1.4 and answer.
5) See Section 1.4 and answer.
Check Your Prqgress 3
1) See Section 1.5 and answer.
2) See Section 1.5 and answer.
3) See Section 1.6 and answer.
4) See Section 1.6 and answer.
5) See Section 1.7 and answer.
Qurrltitative Techniques
far Risk Analysis
I. 12 EXERCISES
1) Six fair dice are rolled one time. What is the probability that each face
appears?
2) If two far dice ,are tosses, what is the smallest number of throws for which
* the probability of getting at least one double six exceeds 0.5.
3) Two dice are thrown and three events are defined as follows: A means "odd
face with first die"; B means "add face with second die"; and C means "odd
sum" (one face is odd, one face is even). If each of the 36 sample points has
probability 1/36, then i-

a) Show that the events are pair wise independent.

b) Show that the events are not jointly independent.
4) Bow I contains 3 red chips and 7 blue chips. Bow! I1 contains 6 red chips
and 4 blue chips. A bowl is selected at random and then I chip is drawn
from this bowl. What is the probability that the chip drawn is red? Given
that this chip is red, what is the conditional probability that it came from
bowl II?
5) Bowl I contains 6 red chips and 4 blue chips. Fives of these 10 chips are
selected at random and without replacement and put in bowl 11, which was
originally empty. One chip is then drawn at random from bowl 11. Find the
conditional probability that 2 red chips and 3 blue chips were transferred
from bowl I to bowl 11 given that a blue chip is drawn from bowl 11.
6) Suppose that the random variable X has an exponential distribution with
pdf f, ( x ) = exp(-#) ,x > 0 and 0 elsewhere.

a) Find the pdf for l/X

b) Find the pdf for In (x)

c) Find the pdf for Y = 1 - F, (x).

7) Let f (x) = 113 for - 1 < x < 2 and zero elsewhere be the pdf for a random
variable X . Find the pdy and distribution function for the random variable
Y=X'.
8) Suppose that the random variable X has cdf

F' (x) =

Is X a discrete random variable? Is X a continuous random variable:

Calculate the mean and variance of X .
9) Let X be a random variable with moment generating function Probabilitrl
M , (t.),-h < t < h . Prove that

and
P(~~a)lexp(-at).M,(t),-h<t<0.
Hint: Use Chebyshev's inequality.
10) Let X be a discrete random variabl'e with
P ( X = - I ) = 118, P ( X = 0 ) = 6 / 8 and P(X = 1 ) = 118, and P ( x = c ) = O
for all other values of c . Calculate the bound on P( I X -pxl & k - o ; ) for
k = 2 using Chebyshev's inequality.