Anda di halaman 1dari 26

Decision Tree and Influence Diagrams

a. Decision Tree
1. Definition
There are many technical definitions of a decision tree, you may also use decision
trees for many different types of processes or in different industries. The simplest
definition of a decision tree is an analysis diagram, which can help aid decision makers,
when deciding between different options, by projecting possible outcomes. The
decision tree, gives the decision maker an overview of the multiple stages that will
follow each possible decision. Each branch shows the probability of the outcome. Or
decision tree is a schematic tree-shaped diagram used to determine a course of action or
show a statistical probability. Each branch of the decision tree represents a possible
decision or occurrence. The tree structure shows how one choice leads to the next, and
the use of branches indicates that each option is mutually exclusive. The use of decision
trees as follow
Enable a business to !uantify decision making.
"seful when the outcomes are uncertain.
#laces a numerical value on likely or potential outcomes.
$llows comparison of different possible decisions to be made.
%t utili&es a network of two types of nodes decision 'choice( nodes, and states of
nature 'chance( nodes.
2. Interpretation of Decision Trees
The options represented by branches from a decision node must be such that the
decision maker can choose only one option.
Each chance node must have branches that correspond to a set of mutually exclusive
and collectively exclusive outcomes. ' only one of them can happen, no other
possibilities exit(
$ )ecision Tree must show all the possible paths that the decision maker might
follow through time, including all possible decision alternatives.
*ometimes the nodes might occur in a time se!uence.
The se!uence of decisions is shown in the tree from left to right.
*!uare represents decisions to be made.
+ircle represents chance events. +hance nodes are random variables and they
represent uncertain !uantities that are relevant to the decision problem.
,ranches from a s!uare correspond to the choices available to the decision maker.
,ranches from a circle represent the possible outcome of a chance event.
The conse!uence is specified at the ends of the branches.
3. Advantages and Disadvantages of Decision Tree
$dvantages
a. "seful for operational decision making.
b. Enables effective use of back data.
c. "se of probability allows flexibility.
d. Objective analysis to decision making.
e. Encourages clear thinking and planning.
)isadvantages
a. -eliant on the accuracy of the data used.
b. -e!uires !ualitative input to give complete picture.
c. #robabilities only estimated.
d. -eal time data problems.
b. Influence Diagrams
1. Definition
$n influence diagram is a graphical structure for modeling uncertain variables and
decisions and explicitly revealing probabilistic dependence and the flow of information.
%t is an intuitive framework in which to formulate problems as perceived by decision
makers and to incorporate the knowledge of experts. $t the same time, it is a precise
description of information that can be stored and manipulated by a computer. .e
develop an algorithm that can evaluate any well-formed influence diagram and
determine the optimal policy for its decisions. *ince the diagram can be analy&ed
directly, there is no need to construct other representations such as a decision tree. $s a
result, the analysis can be performed using the decision maker/s perspective on the
problem. 0uestions of sensitivity and the value of information are natural and easily
posed. 1odifications to the model suggested by such analysis can be made directly to
the problem formulation, and then evaluated directly.
%nfluence diagrams present a decision in a simple, graphical form.
)ecisions, chance events and payoffs 'values( are drawn as shapes 'called nodes(
and are connected by arrows 'called arcs( which define their relationship to each
other.
%n this way, a complex decision may be reduced to a few shapes and lines.
%nfluence diagrams are excellent for showing the relationship between events and
the general structure of a decision clearly and concisely.
The term influence refers to the dependency of a variable on the level of another
variable.
A decision is a variable that you (or your organization) as the decision
maker have the power to modify directly. It could be whether to invest in a
new project, how much to invest, how much to bid, where to locate a new
site, or, in this eample, what budget to allocate for marketing.
A chance variable is an uncertain !uantity, whose value you don"t (yet)
know, because you don"t have complete information ## maybe it"s in the
future ## and which (unlike a decision) you cannot control directly.
2. Definitions used in influence diagrams
The advantage of starting with influence diagrams is that their graphic
representation is more appealing to the intuition of decision makers who may be
unfamiliar with decision technologies. %n addition, influence diagrams are more easily
revised and altered as the decision maker iterates with the decision analyst. )ecision
trees, because of their strict temporal ordering of acts and events, need completely
respecifying when additional acts and events are inserted into preliminary
representations. .e shall illustrate the applicability of influence diagrams through a
worked example. 2irst, however, we will present the basic concepts and representations
underlying the approach.
An objective is a measure of your satisfaction with possible outcomes. It
might be net present value, lives saved, or $%I&'A, or more generally,
(utility(. )sually, the decision maker is trying to find decisions to maimize
(or minimize) the objective. *ften an objective combines multiple sub
objectives or attributes, which may be in conflict # such as energy costs,
and environmental and health risks. )sually, when the objective is
uncertain, decision analysts suggest maimizing the epected value, or
more generally epected utility, based on risk preference.
2igure 3.44 presents the key concepts. $s with the decision tree, event nodes are
represented by circles and decision nodes by s!uares. $rrowed lines between nodes
indicate the influence of one node to another. 2or example, an arrow pointing to an
event node indicates that either the likelihood of events 'contained in the node( are
influenced by a prior decision or on the occurrence 'or not( of prior events.
$lternatively, an arrow pointing to a decision node indicates that either the decision is
influenced by a prior decision or on the occurrence 'or not( of prior events. The whole
set of interconnected decisions and event is called an influence diagram.
Obviously, a decision tree representation must preserve the ordering represented by
the arrows in an influence diagram and the tree must not have an event node as a
predecessor of a decision node for which it is not directly linked by an arrow in the
influence diagram. %f the tree did, it would imply that the decision depends on the event
node which, from the influence diagram, is not the case. One step-by-step procedure for
turning an influence diagram into a decision tree is as follows
4. %dentify a node with no arrows pointing into it 'since there can be no loops at least
one node will be such(.
5. %f there is a choice between a decision node and an event node, choose the decision
node.
6. #lace the node at the beginning of the tree and 7remove8 the node from the influence
diagram.
9. 2or the now-reduced diagram, choose another node with no arrows pointing into it.
%f there is a choice a decision node should be chosen.
:. #lace this node next in the tree and 7remove8 it from the influence diagram.
;. -epeat the above procedure until all the nodes have been removed from the
influence diagram.
To complete the tree, the possible choices at each decision node and the possible
events at each event node must now be specified. 2inally, subjective probabilities must
be assessed for the events and utilities must be assigned to the end points in the decision
tree. <ery complex decision trees can be represented as one-page influence diagrams.
=owever, the use of influence diagrams to construct decision trees where subse!uent
events and acts depend on the initial decision.
)ecision tree derived from influence diagram
Example 0uestion
4. $ company has to decide whether to invest money in the development of a
microbiological product. The company8s research director has estimated that there is a
;>? chance that a successful development could be achieved in two years. =owever, if
the product had not been successfully developed at the end of this period, the company
would abandon the project, which would lead to a loss in present value terms of @6
million. '#resent value is designed to take the company8s time preference for money
into account( %n the event of a successful development a decision would have to be
made on the scale of production. The returns generated would depend on the level of
sales which could be achieved over the period of the product8s life. 2or simplicity, these
have been categori&ed as either high or low. %f the company opted for large volume
production and high sales were achieved then net returns with a present value of @;
million would be obtained. =owever, large-scale production followed by low sales
would lead to net returns with a present value of only @4 million. On the other hand, if
the company decided to invest only in small-scale production facilities then high sales
would generate net returns with a present value of @9 million and low sales would
generate net returns with a present value of @5 million. The company8s marketing
manager estimates that there is a 3:? chance that high sales could be achieved.
a. +onstruct a decision tree to represent the company8s decision problem.
b. $ssuming that the company8s objective is to maximi&e its expected returns,
determine the policy that it should adopt.
c. There is some debate in the company about the probability that was estimated by
the research director. $ssuming that all other elements of the problem remain the
same, determine how low this probability would have to be before the option of not
developing the product should be chosen.
d. ,efore the final decision is made the company is taken over by a new owner, who
has the utilities shown below for the sums of money involved in the decision. 'The
owner has no interest in other attributes which may be associated with the decision
such as developing a prestige product or maintaining employment.( .hat
implications does this have for the policy that you identified in and whyA
$nswer
a.
b. To get expected return value the company have to decide largeBscale production.
#ick the highest probability that give higher return from answer $. '4.;: 1(
+ ,-
c.
The probability to be unsuccess must be higher than success.
d.
+ ,-
../0
..1
0
,.20
,.20
.
..3
..4
..1
..5
,
,.3
#4 #3 . 3 4 1 5
Y-Values
6#7alues
Therefore, the new owner is risk averse, because the higher the result ';( the more
he wanted to invests '4(, and the lower the result '-6( the less he wanted to invests
'>(.
Applying Simulation to Decision Problems
a. Monte arlo Simulation
1. Definition
$ 1onte +arlo method is a computational algorithm that relies on repeated random
sampling to compute its results. %t uses random number generation, rather than analytic
calculations. 1onte +arlo $nalysis as a tool for managing uncertainty and depends on
the number of repeating simulation. The more simulation you do the more accurate data
you will get.
*tages of simulation 1onte +arlo $nalysis
a. %dentify the factors that will affect the payoffs of each course of action.
b. 2ormulate a model to show how the factors are related.
c. +arry out a preliminary sensitivity analysis to establish the factors for which
probability distributions should be assessed.
d. $ssess probability distributions for the factors which were identified in stage 6.
e. #erform the simulation.
f. $pply sensitivity analysis to the results of the simulation.
g. +ompare the simulation results for the alternative courses of action and use these to
identify the preferred course of action.
2. omparison t!e Probability Distribution
2or the seven stages of 1onte +arlo $nalysis we have to compare the result, this
comparison can be made in a numbers of ways, plotting the two distributions,
determining the option with the highest expected utility, *tochastic )ominance and
1ean-standard )eviation.
a. *tochastic )ominance
$pply when the expected utility of one option is greater than another for an entire
class of utility function. *tochastic dominance can be recogni&ed plotting the
cumulative probability distribution functions '+)2*(.
i. 2irst )egree *tochastic )ominance
This concept re!uires some very unrestrictive assumptions the nature of the
decision maker8s utility function.
2or example we are comparing #roduct # and #roduct 0, as you can see below
Product P
Profit Probability umulative Probability
> to under : >.5 >.5
: to under 4> >.6 >.:
4> to under 4: >.9 >.C
4: to under 5> >.4 4
Product "
Profit Probability umulative Probability
> to under : > >
: to under 4> >.4 >.4
4> to under 4: >.: >.;
4: to under 5> >.6 >.C
5> to under 5: >.4 4
%t can be seen that the cumulative probability distribution functions '+)2*( for
product 0 is always to the right of that for product #. This means that, for any
level of profit, 0 offers the smallest probability of falling below that profit. *o
we can say that 0 is superior than #.
ii. *econd )egree *tochastic )ominance
.hen the cumulative probability distribution functions '+)2*( for the option
intersect each other at least ones, it may still be possible to identify the preferred
option if, in addition to the weak assumptions we made for first degree
stochastic dominance we can also assume that the decision maker is risk averse
for the range of values under considerations.
2or example we are comparing #roduct - and #roduct *, as you can see below
Product #
Profit Probability umulative Probability
> to under : >.4 >.4
: to under 4> >.6 >.9
4> to under 4: >.9 >.D
4: to under 5> >.5 4
5> to under 5: > 4
Product S
Profit Probability umulative Probability
> to under : >.6 >.6
: to under 4> >.6 >.;
4> to under 4: >.5 >.D
4: to under 5> >.4 >.C
5> to under 5: >.4 4
%t can be seen that the profit between @>- @4: millions, - is dominant product.
.hile * dominates for the range @4:- @5: millions, this comparison can be
made by comparing area x, which show the extent to which - dominates * with
area E, the extent to which * dominates -.
$s area x is larger, we can say that product - has second degree stochastic
dominance over product *.
b. The 1ean *tandard )eviation $pproach
$pply when a decision problem involved a large number of alternative course of
action, it is helpful if inferior option can be screened out at an early stage. 2or
example we are comparing #roduct $, #roduct ,, #roduct +, #roduct ) and #roduct
E, as you can see below
%f we compare product $ and product ,, we see that, while they offer the same
expected return product , is much more risky, product $ is there for said to
dominate ,.
%f we compare product , and product +, which for the same level of risk offers
higher expected profits, we can say that , is dominated by +. 2or the same reason,
) dominates E.
The non-dominated products '$, + and )(, are there for said to lie on the
efficient frontier, and only these product would survive the screening process and be
considered further. The choice between these options will depend on the decision-
maker.
Fote that, for using this screening process, first, the probability distributions for
profit should be fairly close to the normal distribution shape, and secondly, the
decision-maker should have a utility function that not only indicates risk aversion
but also has a !uadratic form.
U 'x( G c H bx H ax
5
U 'x( G > H 0.4x B >.>9x
5
U 'x( G > H 0.25x B >.>9x
5
3. Applying Simulation to Investment Decision
$s for example, we compare these 5 machines, $lpha 1achine and ,eta 1achine, the
decision-maker wants to know which machine can give them more value, first we can
calculate the value of those machine with Fet #resent <alue method 'F<#(
Fet #resent <alue 1ethod
Alp!a Mac!ine
Time of cas! flo$ Eear 4 Eear 5 Eear 6
as! inflo$s 49>>> 4:>>> 4:>>>
as! outflo$s 5>>> 9>>> ;>>>
Time of cas! flo$ Eear 4 Eear 5 Eear 6
%eta Mac!ine
Time of cas! flo$ Eear 4 Eear 5 Eear 6
as! inflo$s D>>> 46>>> 4:>>>
as! outflo$s 9>>> 9>>> :>>>
Time of cas! flo$ Eear 4 Eear 5 Eear 6
$ssuming that the #resent <alue -ate is 4>? for both machine, we can get the result as
we can see below
Time of cas!
flo$
as!
Inflo$
as!
&utflo$
'et as!
(lo$
Present )alue
(actor
Discounted as!
(lo$
Alp!a
Mac!ine
'o$ > 6>>>> -6>>>> 4 -6>>>>
*ear 1 49>>> 5>>> 45>>> >.C>C4 4>C>C
*ear 2 4:>>> 9>>> 44>>> >.D5;9 C>C>
*ear 3 4:>>> ;>>> C>>> >.3:46 ;3;5
*ear + 49>>> 3>>> 3>>> >.;D6 93D4
Fet #resent <alue 4:95
Time of cas!
flo$
as!
Inflo$
as!
&utflo$
'et as!
(lo$
Present )alue
(actor
Discounted as!
(lo$
%eta
Mac!ine
'o$ > 6>>>> -6>>>> 4 -6>>>>
*ear 1 D>>> 9>>> 9>>> >.C>C4 6;6;
*ear 2 46>>> 9>>> C>>> >.D5;9 396D
*ear 3 4:>>> :>>> 4>>>> >.3:46 3:46
*ear + 54:>> :>>> 4;:>> >.;D6 4453>
Fet #resent <alue -496
"sing *imulation
Fow we can use the 1onte-+arlo *imulation, below you can see an example for
probability distributions, and we can generate the random number from each
distribution, and repeat the process for at least :>> simulation, remember that the
more you do the simulation, the more accurate the data.
$nd after the several simulations, we get the result as below, for $lpha
1achine, the F#< could be between @-5>>>> B @6;>>>, there is 5>? chance the
F#< would be negative, as for ,eta 1achine, the F#< could be between @-4>>>> -
@4:>>>, there is 6>? chance the F#< would be negative
#is, and -ncertainty Management
a. #is, Management
1. Definition
-isk 1anagement is the identification, assessment, and prioriti&ation of risks,
followed by coordinated and economical application of resources to minimi&e, monitor,
and control the probability andIor impact of unfortunate events or to maximi&e the
reali&ation of opportunities. -isks can come from uncertainty in financial markets,
threats from project failures 'at any phase in design, development, production, or
sustainment life-cycles(, legal liabilities, credit risk, accidents, natural causes and
disasters as well as deliberate attack from an adversary, or events of uncertain or
unpredictable root-cause. *everal risk management standards have been developed
including the #roject 1anagement %nstitute, the Fational %nstitute of *tandards and
Technology, actuarial societies, and %*O standards. 1ethods, definitions and goals vary
widely according to whether the risk management method is in the context of project
management, security engineering, industrial processes, financial portfolios, actuarial
assessments, or public health and safety.
The strategies to manage threats 'uncertainties with negative conse!uences(
typically include transferring the threat to another party, avoiding the threat, reducing
the negative effect or probability of the threat, or even accepting some or all of the
potential or actual conse!uences of a particular threat, and the opposites for
opportunities 'uncertain future states with benefits(.
2. -ncertainty
"ncertainty is a term used in subtly different ways in a number of fields, including
philosophy, physics, statistics, economics, finance, insurance, psychology, sociology,
engineering, and information science. %t applies to predictions of future events, to
physical measurements, that are already made, or the unknown. "ncertainty arises in
partially observable andIor stochastic environments, as well as due to ignorance andIor
indolence.
3. #is, and -ncertainty Management
,usiness practices need to be dialed in to uncertainty. 1any companies have risk
management programs which might sounds familiar, but really just part of the
challenge. The best practices in traditional risk management are vital, but they are not
enough to protect a company from economic risk.
,ased on Jary #atterson8s book, 1illion )ollar ,linds *pots 5>I5> <ision for
2inancial Jrowth, we should
4. %dentify risk events
5. $ssess the probability of each event
6. 1ake a cost-benefit analysis of response alternatives
9. +hoose a response
:. -e-assess probability and impact with company response
;. On-going monitoring of risk events
,usiness response to risky events can be categori&ed as
$voidance 'don8t do the act which brings forth the risk
-eduction 'reduce the probability of the event or the damage of the event(
*haring 'spread the risk, such as through insurance(, and
$cceptance 'live with it(
-isk is different from uncertainty, -isk is to describe cases of known probability.
2or example, 1c)onald8s knows how many times a person will slip and fall for every
million visits to their stores. 1eanwhile, "ncertainty is when you don8t know the
probabilities. Kike when a company considering an investment in a copper smelter
which could last years and years. The company has no good idea what the price of
+opper will be in 5> years, nor it is certain what is the probability of different possible
prices.
-isk and uncertainty are really two ends of a single spectrum. %n some cases we
have a <ery accurate idea of the odds of an event happening, such as the 1c)onalds
example above. Then we have some cases where we have a rough idea of the odds 'as
when your mechanic guesses that there8s a :>-:> chance your radiator will last for
another six months(.
+. T!e T$o )alleys ompany
The Two <alleys +ompany is diversifying into the production of an electronic
product and has to decide where to locate its manufacture. Two suitable factories are
available to be rented.
Exploring sources of uncertainty
$s a first step to managing this uncertainty, an exploratory tree was formulated
to represent the factors that contributed to the uncertainty of financial return.
'*ources of uncertainty at Two <alleys +ompany(
%dentifying possible areas for uncertainty management
$ structured method of risk management first allows the manager to identify
and evaluate the most promising areas where risk might be reduced or payoffs
enhanced. This can involve the following approach
a. +alculate the effect of perfect control
This can be achieved by first looking at the option that the decision model
suggests should be preferred and determining how the probability distribution of
the annual profit would change if the decision maker is able to exercise control
over the events that the model assumed were uncertain.
b. -epeat the above process for the next best option
This is only worth considering if the next best option has performed almost
as well as that of the currently favored option in the initial risk analysis. %f it has,
or if the choice between the options is not clear, then this option may actually
offer better opportunities if imaginative strategies can be designed to reduce
risks and enhance returns.
"sing brainstorming to create actions to improve the preferred policy
+reative thinking can be used by the management team to produce actions that
will potentially reduce the risk and identify opportunities. ,rainstorming is one
techni!ue that can enhance creativity. "nderlying brainstorming is the idea that
people8s creativity is restricted because they tend to reject ideas at too early a stage.
,rainstorming has four basic rules
4. )o not critici&e ideas B the solution to the problem may turn out to lie in an idea
that, initially, may seem to be cra&y.
5. Encourage participants to put forward any idea that they can think of B
particularly unconventional or outlandish ideas.
6. $im to generate large !uantities of ideas B in that way there is a greater chance
that one or more of the ideas will lead to a solution to the problem.
9. Encourage people to combine or modify ideas that have already been put
forward.
Example 0uestion
4. The $, +harity is planning its annual campaign to raise money. This year three
alternative methods are being considered
a. street collections,
b. a television advertising campaign and,
c. a direct mail appeal.
$fter using simulation to assess the risk associated with the alternatives the charity8s
managers have opted for a direct-mail appeal. The direct-mail appeal will involve
sending out 696.>>> letters to selected people. To encourage donation these will include
a free ballpoint pen displaying the charity8s logo and people not replying after three
weeks will receive a reminder. .hile the fixed costs of the campaign and the cost of
sending out each letter and reminder are known for certain the charity8s managers have
had to estimate probability distributions for the following four factors
a. The percentage of people who will reply to the first letter in the Forth 'F(, +entral
'+( and *outh '*( regions of the country, respectively.
b. The average donation of those replying to the first letter in each of these regions.
c. The percentage of people who will reply to the reminder in each of the three
regions.
d. The average donation of those replying to the reminder in each of the regions.
#robability distributions have been estimated for the different regions because their
different economic conditions are likely to have a major effect on people8s propensity to
donate to the charity. 2igure 44.; shows the cumulative probability distribution of net
returns 'i.e. the total value of donations less the cost of running the direct-mail appeal(.
%t can be seen that there is approximately a 5>? probability that the net returns will be
negative, causing the charity to lose money. %n the simulation the possible losses
extended to nearly @4:> >>>. The managers of the charity are keen to take action to
reduce this risk, but are not sure where their actions should be directedA
2igure 44.3 shows a tornado diagram for the appeal. The numbers at the ends of the
bars show what are thought to be the highest and lowest possible values for each factor.
2or example, the possible average donation in the Forth is thought to range from @5 to
@43.
a. %dentify the areas where risk management is likely to be most effective.
b. +reate a set of possible risk management strategies that might reduce the risk of the
charity losing money and increase its expected return.
$nswer
a. 2irst, we need to seperate each area from 2igure 45.3
4. Forth $rea
5. *outh $rea
6. +entral $rea

2rom diagram above we can see that +entral area is the best way to use risk
management. ,ecause the probability in that area less risky than the other, if we see
on $verage donation 4st letter for each area 'Forth around 5:>>> to 5>>>>>, *outh
9>>>> to 4;>>>>, and +entral around 45D>>> to 465>>>( which can be considered
that +entral area is less risky than the other even maximum money we can get is less
than the other area.
These areas several risk management strategies that might reduce the risk the charity
losing money and increase its expected return.
=ow can we reduce the costA
4. -eminder letter sent not by direct mail but via telephone or email which can
reduce cost from printing the letter and postal services.
5. -e-assessment the area which have lower possibility to get more money. ' from
the people who didn8t join the charity or the cumulative people in that area
which have lower probability to get more money(
=ow can we raise the donationA
4. The reminder letter sent after a week from 4st mail sent.
5. 1ake a brochure about the charity event which include in the mail.
6. *ent more mail to the higher expectation area to raise the money.
9. *treet collection method is also highly recommended to be done in crowded
places.
:. ,uild a website for the donation to expand the coverage, not only for 696.>>>
people but can be more.

Anda mungkin juga menyukai