Lecture 58 - Coverage (Decision Tree Analysis)

IIT Kanpur July 201
20 followers
Jun 11, 2024
[Music]
uh welcome back my dear friends a very
good morning good afternoon good evening
to all those who are taking this course
in multicriterial decision
making and this is the last last week of
the classes uh which are there for this
NP M course which is spread over 12
weeks which is 60 lectures and each week
we have five lectures each lecture being
for half an
hour and after five lectures we have an
assignment so this is being the 12th
week which means you will take the 12th
assignment and then they will be the
final examination my good name is
ragunan singup from the IM department at
khur in
India now as you can see the slides are
in Word document the things which we
will be discussing today and there's a
reason for that because the diagrams and
all these things which I'll show as we
solve the problems it was difficult to
make an PP file that's why I converted
into I made it into Word document and
we'll discussing that I'll will share it
share it along with the slides uh for
the last two lectures so including this
we have another two lectures to go
before we wrap up this
course now till now we have discussed
the concepts of utility theory in
details univariate case consider
different
properties stochastic dominance first
order second order Safety First
principle what are the different type
of utility functions then we consider
different type of multi-objective
formulation problems linear programming
nonlinear programming integer
programming then we went into
multi attribute concept multivariate
utility theory in under the
multiattribute theory we
considered uh in definitely in some
detail uh the concepts of electr method
the Epsilon Electra in Electra method
and Epsilon specifically I mentioned
that how you can bring a one to one
correspondence considering the concept
of risk aversion risk indifferent person
and the property of risk loving and
there we had the concordance set
discordance set and the indifference set
in the Epsilon elector then we
considered yor method topsis method the
concept of NIS pis
different distance measure using the lp
Norm P being from 1 to infinity and how
you can analyze different ranking system
of the Alternatives considering both
positive and negative CR set of
criteria then in the last class we
formulated the concept of Da data
development analysis and I did mention
that da was a little bit different in
the sense that it was
based on optimization concept using
ideas of efficiency we formulated the
problem for the output oriented model
input oriented model and how you can
utilize simple linear programming to
solve the
problem the idea which you're going to
discuss here decision tree analysis is
also in a different
line and why we are considering that
under multicriterial decision making
that will become clear to you now
whenever we um make a decision tree and
we talk about decision tree decision
tree is a
process mathematical logical process
based on which we take the decision that
which objective needs to optimize and
even if we optimize how we rank them or
which objective function needs to be
minimized how we rank them and take that
decision and we have seen that when we
considering the multiobjective Case
multiattribute Case the semblance and or
the one toone correspondence in the
qualitative sense when we are talking
about different objective functions and
the Alternatives criterias constraints
decision
variables all these things were
discussed as we
proceeded now for all the
examples in the non-parametric method
the analysis or the idea was always is
to consider three different examples one
was about how to buy a house a
apartment then second was to buying an
vehicle a car and third one's choosing
your University where you want to study
the problem which you'll consider here
the two different problems are about
different projects and they would be
given in
detail now before we go into decision
Tre analysis the key steps in trying to
delate or Draw
a decision tree are as follows so is
basically first you have to identify the
problem and what are the Alternatives
different Alternatives we can utilize
to achieve the goal you want
to have set in mind goal can be trying
to maximize your profit can we minimize
the loss can be throughput has to be
maximized loss has to loss means not
monit loss loss can be per day loss in
in electricity if you're using
electricity has to be minimized or
number of trucks you want to basically
use to send forth goods from City 1 to
City 2 that has to be maximized so there
can be different ways of trying to
analyze decision tree problems so first
as let me read it identify the problem
what is to be achieved and what are the
Alternatives you draw the decision tree
and the drawing of the decision tree is
done from left to right so if I'm
looking at the problem looking the
problem in the sense when I draw the
problem is basically starting from the
initial stage which will
very simply mention as the starting
point and we'll have sets of ending
point onto the right now the decision
tree diagram is not exactly equal but it
basically resembles the P CPM diagram
which is a big part which is used in
project management and project
analysis we'll see that how the similar
con similar Concepts in the sense how
the logical
sequence are kept in mind and how the
diagram is drawn from the left to the
right you specify the probabilities and
monetary
outcomes now in P CPM in one of the meth
method it was deterministic time and in
another case stochastic time here we'll
consider the set of probabilities are
there depending on which decision is to
be taken and for our case we will
consider the monitary outcome based on
which the probabilities are assigned and
based on the probability based on the
monitary outcome we'll take the root
which will maximize or minimize the
objective function which is there in
mind as I said maximize The Profit
minimize the
loss so we'll evaluate the various
decision Alternatives and take the one
which basically best
suits the set of assumptions based on
which the problem has been
formulated remember the decision points
in decision tree I'll because as I would
not be able to utilize the marker here
let me check so I'll basically try to
highlight it using
my uh Mouse so remember the decision pre
also has decision
Forks so decision Forks are basically as
we proceed they will be denoted by
D and the and from the decision Forks
you will have alternative actions which
are available for experimentation or
moving from the direction from left to
right and actions are possible at this
points at the decision
folks while the chance points called
chance
Forks which I'm just now highlighted
will be denoted by C so do not confuse
this D and this C which you see in
decision Tre analysis as the discordance
set and the concordance set no or not as
the criteria
also so this chance folks are the points
and there would be many points in the
decision tree as you go from the left to
the right where the outcomes are
dependent on a chance process and the
likely outcomes at these points would be
decided accordingly
now let us see a problem so rather than
explaining for maximum of the cases what
we have done is that we have solved the
problems in order to understand how
somebody can apply these methodologies
in some of the theoretical problems and
in practical sense this basically gives
you an idea so ex so we'll consider two
examples first example will be where
probability would not be as involved as
the second one and when I mean by
probability it will be conditional
probability and we'll come to that so if
all of you have done the concept of
condition probability based theorem so
it will be easy for you to understand
problem two and that's a very
fundamental Concept in probability and
statistics the problem says as follows
the scientist at a company Spectrum have
come up with an electric moped the firm
is ready for pilot production and test
marketing so obviously
before the product is launched so the
company will made the product test
Market it get the customers feedback
about the quality the if as it a moped
they will check the
quality uh factor from the customers the
safety features from the customers what
is the maintenance cost how are the
braking system how is the power of the
engine whether the car or the moped can
be navigated in a big city crowded City
where more than two persons can sit and
so on and so
forth so they will get the feedback from
the customers and obviously the price is
also factor and then make changes
accordingly as it is required for the
pilot production and test marketing it
will cost 20 millions and take 6 months
now let me tell you few important facts
there would be some very simplistic
assumption we can
definitely go ahead by making the
Assumption more realistic
only thing which will be involved in the
problem the problem would get much more
indepth so we have to do much more
step-wise calculation but the idea would
remain the
same so management believes there is a
70% chance that the pilot
production which has been made for test
marketing it will be successful and
obviously 30% chance it would not be be
successful so I am saying that 30%
chance is not successful so there can be
more than two arms also so as it is
mentioned 70% successful it can be 20%
not successful 10 is lukewarm medium so
those issues I'll keep mentioning them
but we'll consider the simplest form of
trying to solve the problem which I just
mentioned few minutes back in the case
of a success Spectrum can build a PL
costing about rupes 150 million the
plant will generate an annual cash flow
of 30 million for 20 years if the demand
is high or an annual cash flow of rupees
20 million if the demand is
low now after the T pilot production
test marketing success is there the
company builds the plant at
150 and there are two ways of trying to
handle hand Le
the demand one is highly successful
world for 20 years annual cash for 30
and another case is where demand is low
annual cash flow 20 millions and I'm
considering simplistically again the
time frame is 20 years also now the
question may immediately come up that
cannot there be more than two outcomes
like high demand medium demand low
demand yes they can be I have not
considered here the next question can be
that let us consider for high demand 30
million cash flow is 20 years but for
low demand 20 million cash flow per year
but the time duration is not 20 years is
much smaller like say for example 15
years can it be yes it can be again I'm
saying I'm not considering that I'm
considering the simplistic
case high demand has a probability of 6
hence low demand has a probability of 04
again the question question can there be
three types of demand high demand medium
demand low demand here we have
considered again I'm mentioning high and
low if you consider high medium low it
can be say for
example5 High 3 medium and point to low
so obviously if you note down the
probabilities would all add up to
one now the question is that what is the
optimum course of action using the
decision tree
analysis so let us
first make the diagram and this is what
the diagram would look
like details for the diagram will come
up later but first the schematic
sketch now if you see the problem from
the company for the company what it
wants is this and I'll keep highlighting
that so these D's are the decision folks
and C are the chance Forks D are the
decision Forks C are the chance Forks if
you remember as I had mentioned when the
nomenclature was being
done number one number two the numbering
as I mentioned the diagram will start
from the left hand side so numbering has
been made accordingly so first you have
D1 then obviously you have C1 I have not
marked anything inside the circle
C1 then from C1
obviously you will have two arms why two
arms I'll come to that later then you
have D2 as you're going from left to
right along with that you have D3 then
you go to the right top Corner which is
C2 and so on and so forth now why there
are two arms from each of
them now if you remember the problem
says the company
would invest 20 million for the pilot
production and test marketing so this D
111 arm which I'm mentioning and which
I'm
highlighting has the negative value of
20 million that's why it's marked in
red so the first path d11 is invest in
the pilot plant and the path D12 is do
nothing that means do not invest
anything that can be also be an option
for the company once this 20 million is
invested the test marketing is done if
you remember I had said that there 70%
chance it is suc ESS 30% chance is not
so this C12 for which the probability
is3 means that the company has to stop
production in spite of the fact there
was a pilot production and test
marketing was there but as the demand
was very low it was stopped for the case
C11 where the probability is 7 there the
success was high the company will build
the
plant or else the company can also build
the plant and do
nothing now obviously the question will
will come up if the company has a
success of
70% then uh the obviously the question
would be build the plant that's true and
this building the plant if you see from
D21 you have invested 150 million over
and number the 20 million which you have
already spent for building of the pilot
uh production plant and test marketing
so this D21 is basically the overall
Factory investment which you need to do
to start the actual
production and in spite of the fact if
you see the arm d22 in spite of the fact
that it was a success of 70% the company
still didn't want to do it may be
possible to many circumstances like say
for example the company was bought over
by some other company or the 150
Millions it has planned to invest
suddenly didn't come up from from
different type of of lenders or it
couldn't basically make the money and
meet the demands there can be many other
practical situations and once the
production starts we know that there is
60% chance is the high demand 40% chance
there's a low demand so this c21 with a
probability of 6 gives you 30 million
per year and this is blue in color
because there's a cash flow coming back
for the low demand which is c22 with the
probability of 04 with the low demand we
know there 20 million will come back to
the company and the time duration you
remember was given by 20 years for both
the cases even though 20 million uh 20
years for case 1 and case two case 1
being 30 million per year high demand
and case to being 20 million per year
for low demand the time rtion of 20
years may not be true for both the
cases now if you see there are some
vertical lines also green one one which
I'm now pointing out so these green
lines are the stage of
evaluation you as a decision maker will
do and try to consider which path to
take now remember the path were or the
decision tree was drawn drawn from the
right left to the right but when you do
the analysis you will do will be doing
the analysis from the right to the left
so obviously your analysis would start
being taken from the right side which is
trying to find out the
overall net value or the
revenues or the cost whatever it is
starting from the rightmost part so
obviously you are seeing only one set of
arrows which is c21 and c22 we'll start
from there and proceed backwards till
the
left now this is what we are going to
do now if you see for the the part which
is I'll just I think I have to write so
I
will so if I
consider now there are different ways of
trying to analyze and this I'll spend
some time
here consider 30
million now if it is 20 years consider
so now to give a background I would not
be writing much I'll be talking because
these ideas definitely uh many of you
who have taken finance courses would be
aware of that generally we know there
are uh two types of rates and this rates
we have discussed in utility Theory
which was um capital r and small R total
rate and the value of uh rate of return
so one was capital r
and one was small R so we'll only
consider small
R and we also know that when you are
considering interest rate we can utilize
the concept of simple interest and the
compound
interest and the compounding Factor if
it is Run infinite number of times in a
particular cycle or in a particular time
frame of one year you have the
continuous compounding interest
rate ccim US the short form continuous
compounding interest
rate and based on that we can find out
so I will first utilize the concept of
continuous compounding interest rate and
then proceed backwards backwards means
try to basically highlight that in in
case of continuous compounding interest
rate was not there how we can use the
concept of continuous compounding uh
rate or the simple interest rate and do
the calculations but the the idea would
the same now there there were many
assumptions which I said so another
assumption is that for this 20 years for
the 30 million and for this uh 20
million for this 20 years so for high
demand low demand probability being 6.4
we are considering the continuous
compounding interest rate to be fixed
even though that's not true we'll
consider that interest rate continuous
compounding interest rate let say for
example
RC so if I
consider the values so consider the time
frame and this time frame is basically
I'm considering from the value when
c22 has the production has started has
finished and then the demand is on the
highest side for 20 years so if I
consider the
fact that if I want to be basically find
out the value of
money if this is
p and the value is
P1 and in the continuous compounding
interest rate and consider this time
frame is
one then p e to the power because
continuous compounding interest rate
means continuous compounding it becomes
exponential e to the power
RC time frame is
one is equal to
P1 so if I want to find out the value P1
is is basically the perer rate which is
given so P can be found out as P1 e to
the power minus
rc1 so this is for one year now if it is
for two
years so obviously in this case it will
be p is equal
to
P2 E power minus
rc2 and that can be considered
accordingly now based on the fact that
we would have basically such 20 such
payments being made so obviously if I
have the time
frame so this is zeroth year 1 year 2
year then dot dot it goes on 18th year
year 19th year and 20th year so if I
consider the first year value at time
frame zero that is when c22 that chance
node has just been
made so this would be for a time frame
of one year similarly when I
consider the second payment of 30
million after the second year this will
be considered for time period of 2 years
similarly this one from the 18th year it
will be for 18 years 19 years and 20
years so basically if I have I need to
basically sum up all the values of p p
means the present value based on the
fact The Continuous compounding interest
rate is given RC which is fixed which is
an assumption and the time frame is
changing from one year then 2 years 3
years 4 years till the 20th year and add
up these values that would be the net
present value given at the case when I'm
considering the value for the high
demand now similarly I'll find out the
low Demand only in the low demand case
remember this value of P1 which is given
here for the high demand case was 30
million because for per year and this
low demand is 20 million per year so P1
basically changes from 30 to 20 that's
point one time frame is same point
number two when I want to find out the
expected value at this
stage as I'm going from the right to the
left I need to find out the expected
value of the chance node considering how
many arms are there and this type of
problem you have already done if you
remember in neutral Theory we found out
the expected value is exactly the same
now the question immediately which may
come up in many of the candidates and
the students mind is that can we use
variance also in place of expected value
yes you
can now the moment the answer is yes you
can immediately this should give you all
of you the
information that see here we can use
multiobjective case um functions in
order to analyze the problem and what
are the multiobjectives multiobjectives
the first case is try to take the
expected value only second case can be
try to take the variance only so
depending on the problem formulation you
can take different objectives and
formulate the problem and rank them
accordingly so that's why when I started
discussing the decision tree analysis I
did mention is a multi criteria problem
which can be seen it is not the
classical type of multiobjective
problems which you have already solved
but they can be fine-tuned as a
multiobjective problems and the problems
can be solved accordingly now this was
at stage c22 so if I consider the
problem as c22 is if you consider so
obviously the expected
value expected
value I am using
EV at
C2 would be 0.6 is the
probability
for the high
demand which would be multiplied by the
net present value
for 30 million plus
0.4 ultied by net present value for 20
million and where is this npv 20 and npv
30 coming from this npv 20 and npv 30
are coming up from trying to find out
the the net value for the high demand
separately multiplying it by 6 and
finding a net present value for the low
demand then multiplying by 6 so once you
have that the net value of c22 is given
and the net value of c22
is based
on the formulation of the case when you
have the simple interest rate now if you
remember I did mention that you can use
the concept of continuous comp ing
interest rate compounding interest rate
simple interest rate so if I consider
the simple interest rate only for the 30
million per year arm high demand so it
will be basically given by the case of
30 this is for the first
year 1
+ R1 R1 is the simple interest for one
year to the power one because you're
doing it for one year trying to find out
the net present value for one year and
what is this 30 30 is that per year
payback this is 30 for the second year
it is
R12 oh this is R1 sorry because we are
considering the interest rate as fixed
square and it goes to
30 1 + R1 to the^ 20 why 20 this 20 is
not 20 million 20 is the year so once
you find out this and multiply this by
6 Plus
0.4 multiplied for the case when you
have 20 million coming up per
year again interest rate is fixed R1
this is for the low demand that's why
it's
04
so this
value calculated comes out to be
approximately
194. now why there is a 150 minus
because if you go back to the slide or
the word document my my apologies 194 is
the income but 15 is the outcome So
based on that if I want to find out
44.2 is the net value which is positive
for the arm which we have here for for
that means for D2 arm total combined
input and output and obviously that
value of 194 minus 150 which is
basically
44.2 has to be compared with the arm
which is d22 which is you do not do
anything so obviously that being
positive that will be subsumed that you
should go through that
process so with this lecture I'll stop
here and continue the discussion in the
next class and try to wrap up the
decision TP problems accordingly thank
you very much and have a nice
[Music]
[Music]
day
[Music]
