IME625A

C9.
HW
Wednesday, February 2, 2022 7:17 PM
I purchased a stock in the Bombay Stock Exchange. Every week, its price
either increases by or decreases by with probabilities and .
I have set a rule that the stock will be sold as soon as its price doubles
(compared to my purchase price) or additional units will be purchased as
soon as its price reduces to half the purchase price, whichever happens
earlier. Determine the probability that the stock is sold. How long, on an
average, would it take for either of the actions to happen?
C9 Page 1
C9.P1
Tuesday, February 1, 2022 7:27 PM
A recurrence relation for extinction probability: Now, we make our first

attempt to calculate extinction probability. The intuition that we obtained
from the above table seems to suggest that does not play significant
role in the long-run behavior of the branching chain. So, we take in Implication Extinction Extinction seems Extinction
seems guaranteed, but it does not seem
the remainder of the analysis. Later, we will examine the effect of .
guaranteed. would take ‘time’. guaranteed.
Let denote the time till extinction. Let
denote the probability of
extinction by -th generation for . Then the probability of (eventual)
extinction, . If we can
obtain the values of , then we have .
is the distribution function value of at . We

know that distribution function value of any random variable is as we
approach . Then extinction probability
must be . However, this argument is flawed as
may not be a ‘proper’ random variable.
Let us apply the first-step analysis to . Earlier,

we used it to obtain absorption probability through the transition
probabilities, which are not available to us for the branching chain. Here,
we take an indirect route of obtaining extinction probability via
. First, note that . For ,
In the above derivation, we wrote for

. It is correct for and , as and
by the definition of . For , we can
consider the branching chain with as the sum of number of
independent and identical branching chains with , each emerging
th
from a member of the 0 generation in the original chain. Let
denote -th such chain for . Then
due to independence and due to identicalness.
With and for , we can iteratively

calculate for any branching chain and obtain .
Let us consider a simple example. Let , ,
and in a branching chain, i.e., each member produces at
most 1 offspring. Then for .
C9 Page 2
Let us consider a simple example. Let , ,
and in a branching chain, i.e., each member produces at
most 1 offspring. Then for .
Then . i.e., extinction is

guaranteed.
C9 Page 3
C9.P2
Expected time till extinction: We can obtain mass function of from

as:
Then . Note that this calculation

makes sense only when . For the above example for
. Then
There is an alternate way of obtaining expected time till extinction:

. With this formula for the example problem,
. This formula is based on the general formula:
for non-negative integer valued random variable . It has
a continuous analogue: for
non-negative continuous random variable . Verify these yourself.
Effect of : We have discussed about a way to calculate extinction

probability and the expected time till extinction when . If
, say , then we can consider the branching chain as the sum
of number of independent and identical branching chains with ,
each emerging from a member of the 0th generation in the original chain.
Let denote -th such chain for . Then
where denotes the extinction probability for . If , then

for all . Thus, greater initial population cannot alter
eventual extinction if that is the case with . If , then
decreases with , but it is positive for all , i.e., the
possibility of extinction cannot be eliminated by increasing initial
population size.
With the above characterization, time till extinction in the original chain is
maximum of the times till extinction for the branching chains emerging
from members of the 0th generation in the original chain, denoted by
. Then
are iid random variables with distribution function

values at , which we obtained recursively for .
Then for ,
C9 Page 4
values at , which we obtained recursively for .
Then for ,
Hence, . Since
for all (Why?), then decreases with , and thus,
increases with . Hence, larger initial population can
delay extinction, even though it cannot stop it.
Quiz-2: 14 Feb (Monday)
C9 Page 5
C9.P3
Going beyond the recurrence relation: for with

and is sufficient to calculate extinction probability
for any given branching chain. However, it is not sufficient to extract
insights. We need to go beyond the recurrence relation. Taking limits of
both sides of as , we get
is computationally much simpler than the original

recursive relation. In its derivation, we changed the order of limit and sum,
which is always permitted for finite sum. For infinite sum, which is the
case here, the change of order may not be permitted. Consider two cases:
and . In the first case,
So, change of order is permitted in the first case. In the second case,
There are some results, known as the convergence theorems, that lays out
conditions for the change of order of limit and sum to be possible. One
such theorem is bounded convergence theorem, which will be invoked
number of times in this course.
Let be a real valued function defined on unbounded such that

exists. Let for be a sequence of real valued
function on such that for all and and
exist for all . Let for some real
valued function on . In this set-up, is same
as .
In , , and
for . Here, is unbounded, , i.e., it
exists, (as
is probability) and
exists for all . Bounded
convergence theorem applies here, and thus, the order of limit and sum
can be changed. It shall be noted that the theorem does not say anything
when the underlying conditions are not met.
For a branching chain with and rest zero,

. For ,
; since , is the solution, i.e.,
extinction is guaranteed. For , , i.e.,
extinction is guaranteed. For ,
; both the solutions are valid probabilities values. It’s not
immediately clear which one is extinction probability. Our intuition from
C9 Page 6
extinction is guaranteed. For , , i.e.,
extinction is guaranteed. For ,
; both the solutions are valid probabilities values. It’s not
immediately clear which one is extinction probability. Our intuition from
the analysis of mean and variance suggests that is the extinction
probability. It can be confirmed from the recurrence relations. Observe
that is a candidate solution of in all three cases. This
observation is true in general, as: . So,
when we solve for a branching chain, we always get as
a solution. If there is no other solution in , then , i.e., extinction
is guaranteed. If we get a solution in , we are not sure about . In the
following sections, we show that the smallest solution in is the
extinction probability.
C9 Page 7

IME625A

Uploaded by

Copyright:

Available Formats

IME625A

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IME625A

Uploaded by

Copyright:

Available Formats

C9.

A recurrence relation for extinction probability: Now, we make our first

is the distribution function value of at . We

Let us apply the first-step analysis to . Earlier,

In the above derivation, we wrote for

due to independence and due to identicalness.

With and for , we can iteratively

Then . i.e., extinction is

Expected time till extinction: We can obtain mass function of from

Then . Note that this calculation

There is an alternate way of obtaining expected time till extinction:

Effect of : We have discussed about a way to calculate extinction

where denotes the extinction probability for . If , then

are iid random variables with distribution function

Quiz-2: 14 Feb (Monday)

Going beyond the recurrence relation: for with

is computationally much simpler than the original

Let be a real valued function defined on unbounded such that

For a branching chain with and rest zero,

You might also like