Chapter 1: Stochastic Processes 4

What are Stochastic Processes, and how do they fit in?

STATS 310
Statistics
STATS 210 Randomness in Pattern
Foundations of
Statistics and Probability
Tools for understanding randomness
(random variables, distributions) STATS 325
Probability

Randomness in Process

Stats 210: laid the foundations of both Statistics and Probability: the tools for
understanding randomness.
Stats 310: develops the theory for understanding randomness in pattern: tools
for estimating parameters (maximum likelihood), testing hypotheses, modelling
patterns in data (regression models).
Stats 325: develops the theory for understanding randomness in process. A
process is a sequence of events where each step follows from the last after a
random choice.

What sort of problems will we cover in Stats 325?

Here are some examples of the sorts of problems that we study in this course.

Gambler’s Ruin
You start with $30 and toss a fair coin
repeatedly. Every time you throw a Head, you
win $5. Every time you throw a Tail, you lose
$5. You will stop when you reach $100 or when
you lose everything. What is the probability that
you lose everything? Answer: 70%.
 5

Winning at tennis
What is your probability of winning a game of tennis,
starting from the even score Deuce (40-40), if your
probability of winning each point is 0.3 and your
q
opponent’s is 0.7?
p VENUS p VENUS
WINS (W)
Answer: 15%. DEUCE (D)
AHEAD (A)

q VENUS VENUS
BEHIND (B) q LOSES (L)

p
Winning a lottery

A million people have bought tickets for the weekly lottery

draw. Each person has a probability of one-in-a-million
of selecting the winning numbers. If more than one person
selects the winning numbers, the winner will be chosen
at random from all those with matching numbers.

You watch the lottery draw on TV and your numbers match the winning num-
bers!!! Only a one-in-a-million chance, and there were only a million players,
so surely you will win the prize?

Not quite. . . What is the probability you will win? Answer: only 63%.

Drunkard’s walk
A very drunk person staggers to left and right as he walks along. With each
step he takes, he staggers one pace to the left with probability 0.5, and one
pace to the right with probability 0.5. What is the expected number of paces
he must take before he ends up one pace to the left of his starting point?
Arrived!

Answer: the expectation is infinite!

 6

Pyramid selling schemes

Have you received a chain letter like this one? Just send $10 to the person
whose name comes at the top of the list, and add your own name to the bottom
of the list. Send the letter to as many people as you can. Within a few months,
the letter promises, you will have received $77,000 in $10 notes! Will you?

Answer: it depends upon the response rate. However, with a fairly realistic
assumption about response rate, we can calculate an expected return of $76
with a 64% chance of getting nothing!
Note: Pyramid selling schemes like this are prohibited under the Fair Trading Act,
and it is illegal to participate in them.

Spread of SARS
The figure to the right shows the spread
of the disease SARS (Severe Acute
Respiratory Syndrome) through Singapore
in 2003. With this pattern of infections,
what is the probability that the disease
eventually dies out of its own accord?
Answer: 0.997.
 7

Markov’s Marvellous Mystery Tours

Mr Markov’s Marvellous Mystery Tours promises an All-Stochastic Tourist Ex-
perience for the town of Rotorua. Mr Markov has eight tourist attractions, to
which he will take his clients completely at random with the probabilities shown
below. He promises at least three exciting attractions per tour, ending at either
the Lady Knox Geyser or the Tarawera Volcano. (Unfortunately he makes no
mention of how the hapless tourist might get home from these places.)

What is the expected number of activities for a tour starting from the museum?
1/3
2. Cruise 4. Flying Fox 1

1/3 6. Geyser 1
1/3 1/3
1/3 1/3

1/3
1. Museum 3. Buried Village 5. Hangi
1/3 1/3
1/3
1/3 1/3

1
7. Helicopter 8. Volcano
1 Answer: 4.2.

Structure of the course

• Probability. Probability and random variables, with special focus on

conditional probability. Finding hitting probabilities for stochastic pro-
cesses.
• Expectation. Expectation and variance. Introduction to conditional ex-
pectation, and its application in finding expected reaching times in stochas-
tic processes.
• Generating functions. Introduction to probability generating func-
tions, and their applications to stochastic processes, especially the Random
Walk.
• Branching process. This process is a simple model for reproduction.
Examples are the pyramid selling scheme and the spread of SARS above.
 8

• Markov chains. Almost all the examples we look at throughout the

course can be formulated as Markov chains. By developing a single unify-
ing theory, we can easily tackle complex problems with many states and
transitions like Markov’s Marvellous Mystery Tours above.

The rest of this chapter covers:

• quick revision of sample spaces and random variables;
• formal definition of stochastic processes.

1.1 Revision: Sample spaces and random variables

Definition: A random experiment is a physical situation whose outcome cannot

be predicted until it is observed.

Definition: A sample space, Ω, is a set of possible outcomes of a random experi-

ment.

Example:
Random experiment: Toss a coin once.
Sample space: Ω ={head, tail}

Definition: A random variable, X, is defined as a function from the sample space

to the real numbers: X : Ω → R.
That is, a random variable assigns a real number to every possible outcome of a
random experiment.

Example:
Random experiment: Toss a coin once.
Sample space: Ω = {head, tail}.
An example of a random variable: X : Ω → R maps “head” → 1, “tail” → 0.

Essential point: A random variable is a way of producing random real numbers.

 9

1.2 Stochastic Processes

Definition: A stochastic process is a family of random variables,

{X(t) : t ∈ T }, where t usually denotes time. That is, at every time t in the set
T , a random number X(t) is observed.

Definition: {X(t) : t ∈ T } is a discrete-time process if the set T is finite or

countable.

In practice, this generally means T = {0, 1, 2, 3, . . .}

Thus a discrete-time process is {X(0), X(1), X(2), X(3), . . .}: a random number
associated with every time 0, 1, 2, 3, . . .

Definition: {X(t) : t ∈ T } is a continuous-time process if T is not finite or

countable.

In practice, this generally means T = [0, ∞), or T = [0, K] for some K .

Thus a continuous-time process {X(t) : t ∈ T } has a random number X(t)

associated with every instant in time.
(Note that X(t) need not change at every instant in time, but it is allowed to
change at any time; i.e. not just at t = 0, 1, 2, . . . , like a discrete-time process.)

Definition: The state space, S, is the set of real values that X(t) can take.

Every X(t) takes a value in R, but S will often be a smaller set: S ⊆ R. For
example, if X(t) is the outcome of a coin tossed at time t, then the state space
is S = {0, 1}.

Definition: The state space S is discrete if it is finite or countable.

Otherwise it is continuous.

The state space S is the set of states that the stochastic process can be in.
 10

For Reference: Discrete Random Variables

1. Binomial distribution
Notation: X ∼ Binomial(n, p).
Description: number of successes in n independent trials, each with proba-
bility p of success.
Probability function:
 
n x
fX (x) = P(X = x) = p (1 − p)n−x for x = 0, 1, . . . , n.
x
Mean: E(X) = np.
Variance: Var(X) = np(1 − p) = npq, where q = 1 − p.
Sum: If X ∼ Binomial(n, p), Y ∼ Binomial(m, p), and X and Y are
independent, then

X + Y ∼ Bin(n + m, p).

2. Poisson distribution
Notation: X ∼ Poisson(λ).
Description: arises out of the Poisson process as the number of events in a
fixed time or space, when events occur at a constant average rate. Also
used in many other situations.
λx −λ
Probability function: fX (x) = P(X = x) = e for x = 0, 1, 2, . . .
x!
Mean: E(X) = λ.
Variance: Var(X) = λ.
Sum: If X ∼ Poisson(λ), Y ∼ Poisson(µ), and X and Y are independent,
then
X + Y ∼ Poisson(λ + µ).
 11

3. Geometric distribution

Notation: X ∼ Geometric(p).

Description: number of failures before the first success in a sequence of in-

dependent trials, each with P(success) = p.

Probability function: fX (x) = P(X = x) = (1 − p)x p for x = 0, 1, 2, . . .

1−p q
Mean: E(X) = = , where q = 1 − p.
p p
1−p q
Variance: Var(X) = = , where q = 1 − p.
p2 p2
Sum: if X1, . . . , Xk are independent, and each Xi ∼ Geometric(p), then
X1 + . . . + Xk ∼ Negative Binomial(k, p).

4. Negative Binomial distribution

Notation: X ∼ NegBin(k, p).

Description: number of failures before the kth success in a sequence of in-

dependent trials, each with P(success) = p.

Probability function:
 
k+x−1 k
fX (x) = P(X = x) = p (1 − p)x for x = 0, 1, 2, . . .
x
k(1 − p) kq
Mean: E(X) = = , where q = 1 − p.
p p
k(1 − p) kq
Variance: Var(X) = = 2 , where q = 1 − p.
p2 p
Sum: If X ∼ NegBin(k, p), Y ∼ NegBin(m, p), and X and Y are independent,
then
X + Y ∼ NegBin(k + m, p).
 12

5. Hypergeometric distribution
Notation: X ∼ Hypergeometric(N, M, n).
Description: Sampling without replacement from a finite population. Given
N objects, of which M are ‘special’. Draw n objects without replacement.
X is the number of the n objects that are ‘special’.
Probability function:
M N −M



x n−x x = max(0, n + M − N )
fX (x) = P(X = x) = N
for
to x = min(n, M).


n
M
Mean: E(X) = np, where p = .
N
N − n M
Variance: Var(X) = np(1 − p) , where p = .
N −1 N

6. Multinomial distribution
Notation: X = (X1, . . . , Xk ) ∼ Multinomial(n; p1, p2, . . . , pk ).
Description: there are n independent trials, each with k possible outcomes.
Let pi = P(outcome i) for i = 1, . . . k. Then X = (X1 , . . . , Xk ), where Xi
is the number of trials with outcome i, for i = 1, . . . , k.
Probability function:
n!
fX (x) = P(X1 = x1, . . . , Xk = xk ) = px1 1 px2 2 . . . pxk k
x1 ! . . . xk !
k
X X k
for xi ∈ {0, . . . , n} ∀i with xi = n, and where pi ≥ 0 ∀i , pi = 1.
i=1 i=1

Marginal distributions: Xi ∼ Binomial(n, pi) for i = 1, . . . , k.

Mean: E(Xi) = npi for i = 1, . . . , k.
Variance: Var(Xi ) = npi(1 − pi ), for i = 1, . . . , k.
Covariance: cov(Xi, Xj ) = −npipj , for all i 6= j.
 13

Continuous Random Variables

1. Uniform distribution

Notation: X ∼ Uniform(a, b).

1
Probability density function (pdf ): fX (x) = for a < x < b.
b−a
Cumulative distribution function:
x−a
FX (x) = P(X ≤ x) = for a < x < b.
b−a
FX (x) = 0 for x ≤ a, and FX (x) = 1 for x ≥ b.
a+b
Mean: E(X) = .
2
(b − a)2
Variance: Var(X) = .
12

2. Exponential distribution

Notation: X ∼ Exponential(λ).

Probability density function (pdf ): fX (x) = λe−λ x for 0 < x < ∞.

Cumulative distribution function:

FX (x) = P(X ≤ x) = 1 − e−λ x for 0 < x < ∞.
FX (x) = 0 for x ≤ 0.
1
Mean: E(X) = .
λ
1
Variance: Var(X) = .
λ2
Sum: if X1, . . . , Xk are independent, and each Xi ∼ Exponential(λ), then
X1 + . . . + Xk ∼ Gamma(k, λ).
 14

3. Gamma distribution

Notation: X ∼ Gamma(k, λ).

Probability density function (pdf ):

λk k−1 −λx
fX (x) = x e for 0 < x < ∞,
Γ(k)
R∞
where Γ(k) = 0 y k−1e−y dy (the Gamma function).

Cumulative distribution function: no closed form.

k
Mean: E(X) = .
λ
k
Variance: Var(X) = .
λ2
Sum: if X1, . . . , Xn are independent, and Xi ∼ Gamma(ki, λ), then
X1 + . . . + Xn ∼ Gamma(k1 + . . . + kn , λ).

4. Normal distribution

Notation: X ∼ Normal(µ, σ 2).

Probability density function (pdf ):

1 2
/2σ 2 }
fX (x) = √ e{−(x−µ) for − ∞ < x < ∞.
2πσ 2
Cumulative distribution function: no closed form.

Mean: E(X) = µ.

Variance: Var(X) = σ 2 .

Sum: if X1, . . . , Xn are independent, and Xi ∼ Normal(µi , σi2), then

X1 + . . . + Xn ∼ Normal(µ1 + . . . + µn , σ12 + . . . + σn2 ).
 15

Probability Density Functions

fX (x)
Uniform(a, b)
1 111111111111
0
1 00000000000
0
1
0
1
b−a 0
1
0
1
0
1
0
1
0
1
0
1 x
a b

Exponential(λ)

λ=2

λ=1

Gamma(k, λ)

k = 2, λ = 1

k = 2, λ = 0.3

Normal(µ, σ 2 )

σ=2

σ=4