Professional Documents
Culture Documents
DECISION ANALYSIS
1
Week 3
Chapter 4
Making Choices
2
In Last lecture we covered:
Fundamental steps of creating a decision model
Influence Diagrams
3
Lecture 3- Objectives
A decision tree, a structured problem, presents
many alternatives. Which alternative is the best?
Learn how to use the details in a structured
problem to find a preferred alternative.
Decision Trees and EMV
Risk Profiles and Dominance: help make decisions
considering attitude towards riskiness of alternatives
Deterministic and Stochastic Dominance
4
Making Choices
Once we have structured our decision problem then
the next logical question is:
Which alternative is the preferred alternative?
Answer: Two Methods help us find the preferred
alternative:
EMV approach
Risk Profiles and Dominance consideration
5
Decision Trees and Expected Monetary Value
One way to choose among risky alternatives is to
pick the alternative with the best (highest for
maximization / lowest for minimization) expected
value (EV) --- (Calculation involved)
When the decision consequences involve only
money, we can calculate the expected monetary
value (EMV)
6
Solving Decision Trees
Backward Approach--Folding Back
At each chance node (circle or oval), calculate the EV (sum of
payoff times probabilities) and write it above (or below) the
node
At each decision node (square), find the best (maximum or
minimum depending on the objective) EV and write it above
the node, and then cut off (cross it out on your decision tree as
will be demonstrated in future slides) all decision branches
except the optimal one (the one with the preferred EV)
Repeat above steps until the left-most node is reached
Solving Decision Trees : An Example
Suppose we have two alternatives:
A1 : Flip a not-fair coin
Head (H) with probability of 30% -- Win $100
Tail (T) with probability of ____ -- Lose $20
Why A1 ?
Solving Decision Trees : An Example Revised
Complication.....
Sequential Events and/or Decisions
Suppose that we have further a third alternative as follows
A3 : If heads on the first flip (H1) with a fair coin, then flip again
where H2 : Win $200
T2 : Lose $50
If tails on the first flip (T1), then flip again. This time use a
weighted coin with probability of heads equal to 0.4 and of tails
equal to 0.6, where:
H2 : Win $50
T2 : Lose $100
Why?
Trading Tickets Example
You have a ticket that will let you participate in
a lottery that will pay off $10 with a 45% chance,
and nothing with a 55% chance. Your friend has
ticket to a different lottery that has a 20% chance
of paying $25 and an 80% chance of paying
nothing. Your friend has offered to let you have
his ticket if you will give him your ticket plus
one dollar. Should you agree to the trade and
play to win $25, or should you keep your ticket
and have a better chance of winning $10?
17
Trading Tickets Example
18
Trading Tickets Example
19
The Texaco – Pennzoil Case
Read Text Book Page 118– 122 carefully,
having summary of the case
20
The Texaco – Pennzoil Case
In early 1984, Pennzoil and Getty Oil agreed to the terms of
a merger. But before any formal documents could be signed,
Texaco offered Getty Oil a substantially better price, and
Gordon Getty, who controlled most of the Getty stock,
reneged on the Pennzoil deal and sold to Texaco. Naturally,
Pennzoil felt as if it had been dealt with unfairly and
immediately filed a lawsuit against Texaco alleging that
Texaco had interfered illegally in the Pennzoil-Getty
negotiations. Pennzoil won the case; in late 1985, it was
awarded $11.1 billion, the largest judgment ever in the
United States. A Texas appeals court reduced the judgment
by $2 billion, but interest and penalties drove the total back
up to $10.3 billion.
21
The Texaco – Pennzoil Case
James Kinnear, Texaco's chief executive officer, had said that
Texaco would file for bankruptcy if Pennzoil obtained court
permission to secure the judgment by filing liens against
Texaco's assets. Furthermore, Kinnear had promised to fight
the case all the way to the U.S. Supreme Court if necessary,
arguing in part that Pennzoil had not followed Security and
Exchange Commission regulations in its negotiations with
Getty. In April 1987, just before Pennzoil began to file the
liens, Texaco offered to pay Pennzoil $2 billion to settle the
entire case. Hugh Liedtke, chairman of Pennzoil, indicated that
his advisors were telling him that a settlement of between $3
billion and $5 billion would be fair. In order to maximize the
settlement amount, Liedtke plans on some negotiations with
Kinnear. 22
The Texaco – Pennzoil Case
He can either accept the $2B offer or counter Texaco with
$5B. Liedtke believes that to his counteroffer, Texaco might
either refuse to negotiate and go to a final court w.p. 0.5 or
accept the offer w.p. 0.17 or counter him back with a $3B offer
w.p. 0.33. Liedtke does not want to counter offer Texaco if
Texaco counters them back with $3B. His team believes that in
case the case goes back to the court, the chances that the court
awards them with the full $10.3B is 0.2, or it may award them
$5B with 0.5 probability, or nothing w.p. 0.3.
23
Pennzoil Influence Diagram
Outcomes Prob
Accept 5 0.17
Accept $2 Texaco Refuse
Counter 3
0.50
0.33
Billion? Reaction
Choices:
Accept 2 Final Court
Counter 5
Decision
Outcomes Prob
Outcomes Prob
26
Pennzoil Case – (Payoff Table
Cont’d..)
27
Pennzoil Decision Tree
$ Billion
4 1
28
Pennzoil Decision Tree
Solve tree using EMV by folding back the Tree
Calculations invloved:
EMV(1) = 0.2 x 10.3 + 0.5 x 5 + 0.3 x 0 = $4.56
EMV(2) = 0.2 x 10.3 + 0.5 x 5 + 0.3 x 0 = $4.56
Decision @ Node 3 :
Compare $4.56 with $3 (Objective Maximizing, $4.56 is
selected.
Decision @ Node 5 :
Compare $2 with $4.63(Objective Maximizing, 4.63 is
selected. 29
Pennzoil Decision Tree
30
Risk
Sometimes you need alternatives to the Decision
Criterion: Expected Value (always choose highest
expected value for maximizing problem)
A common concern is that, risk, is not Considered
In any decision problem, the actual payoff is not the
expected value of all possible payoffs; it may be
much higher or lower depending upon the decision
outcome space
Decision makers are often concerned about the risk
of bad outcomes
31
Risk
We often fail to use the Expected Value decision criterion in
our daily lives! Think about the following examples:
If you buy any insurance, the expected value is always in
favor of the insurance company
• We buy insurance because the risk of loss is so high
Gambling?
• Casinos always have a better chance of winning, so that they
can stay in business. Then, why do you gamble? Because of
that little chance of winning that big jackpot!
Explicit risk consideration is a way to model this situation
Main study of this is after Midterm, where we introduce the
Utilities (Chapter 13). However, here we can still do
something!!
32
Explicit Risk Modeling
Sometimes we show the Decision Maker not only the
preferred decision based on expected value, but also the
risk associated with different decision strategies
We normally present decision risk by showing the
distributions of the possible outcomes for each decision
strategy – risk profiles
33
Decision Path:
A path starting add the left most node up to the values at the end
of a branch by selecting one alternative from a decision node or
by following one outcome from uncertainty nodes. Represents a
possible future scenario.
Decision Strategy:
The collection of decision paths connected to one branch of the
left most nodes by selecting one alternative from each decision
node along these paths. At every decision node in the decision
problem it specifies what we would do, if we get to that decision
Or, a series of decisions that fully define the behaviour of the
player.
Optimal Decision Strategy:
A Decision Strategy which results in the highest EMV if we
maximize profit and the lowest EMV if we minimize cost.
34
Recall thePennzoil Decision Tree
$ Billion
35
Decision Strategies: Texaco-Pennzoil Case
36
Decision Strategies: Texaco-Pennzoil Case
37
Decision Strategies: Texaco-Pennzoil Case
THIRD STRATEGY: "Counter 5 Billion, Accept $3 Billion”.
38
Optimal Decision Strategy: Texaco-Pennzoil Case
Counteroffer $5 Billion. Next,
if Texaco Counteroffers $3 Billion,
Refuse the counteroffer.
39
Recall Solved Pennzoil Decision Tree
40
Risk Profiles
Risk profiles are simply bar charts that display the
chances associated with possible (discrete) outcomes
Continuous outcome distributions can also be plotted if
known (this is out of the scope of this course)
Each risk profile is associated with a decision strategy;
consider the following two strategies:
– Accept $2 Billion
– Counter offer $5 Billion: refuse Texaco counter offer
41
Risk Profiles
43
Collapsing Decision Tree
44
Collapsing Decision Tree
(Cont’d..)
45
Collapsing Decision Tree
(Cont’d..)
47
Risk Profiles
“Accept $2 Billion”
48
Risk Profiles
49
Risk Profiles
50
Cumulative Risk Profiles
Note:
Cumulative Risk Profile is a cumulative
distribution function for the discrete random
variable Y representing the outcomes for the given
decision strategy.
51
For any specific value along the horizontal axis, we can read off the chance
That the pay off will be less than or equal to that specific value. 52
Dominance : An Alternative to EMV
There are two types of dominance:
Deterministic Dominance: One Strategy is always better
than another, regardless of any chance outcomes
53
Deterministic Dominance
If the worst outcome of Alternative A is at least as good as the
best outcome of Alternative B, Alternative A deterministically
dominates Alternative B.
We may also draw conclusion on deterministic dominance
consideration by drawing CUMULATIVE RISK PROFILES.
Definition: Range of a Cumulative Risk Profile = [L,U].
L= Largest 0% point in Cumulative Risk Profile;
U= Smallest 100% point in Cumulative Risk Profile
Deterministic Dominance:
Step 1: Draw cumulative risk profiles in one graph
Step 2: Determine range for each risk profile
Step 3: Ranges are disjoint or their intersections contains a single
point
Deterministic Dominance is present
54
Deterministic Dominance
“What if the smallest return from court was $2.5
Billion rather than $0?” in the Pennzoil case.
If there is a value x such that the chance of payoff
<= x is 100% in one strategy, and the chance of the
payoff <= x is 0% in a second strategy, then the
second strategy deterministically dominates the first.
There is deterministic dominance which can be
observed by drawing the cumulative risk profiles.
55
Deterministic Dominance
56
Deterministic Dominance
57
StochasticDominance
If two cumulative risk profiles are such that no part of
Profile A lies to the left of B, and at least some part of
it lies to the right of B, then strategy A stochastically
dominates strategy B. For maximization objective;
what if we minimize cost?
58
Dominance: An alternative to EMV (cont’d)
Suppose that Liedtke is choosing between two different law firms
to
represent Pennzoil. He considers both law firms to be about the
same in terms of their abilities to deal with the case, but firm A
charges less in the event that the case goes to court.
59
Dominance: An alternative to EMV (cont’d)
10.5
10.5
60
Dominance: An alternative to EMV (cont’d)
61
Dominance: An alternative to EMV (cont’d)
62
Dominance: An alternative to EMV (cont’d)
Choosing Firm A is like choosing Firm B and possibly
getting a bonus as well.
The two cumulative risk profiles almost coincide: the
only difference is that Firm A’s profile is slightly to the
right of Firm B’s at $5 and $10.3, which represents the
possibilities of Pennzoil having to pay less in fees.
63
Dominance: An alternative to EMV (cont’d)
In this example decision tree comparing two law firms. Firm C has a
better chance of winning a higher award in court than does Firm D.
Carefully examine the probabilities in the branches for the final
Court decision.
64
Dominance: An alternative to EMV (cont’d)
65
Dominance: An alternative to EMV (cont’d)
66
Dominance: An alternative to EMV (cont’d)
67
Stochastic Dominance vs Expected Value
If one alternative dominates another, then the dominating
alternative must have the higher EV (Objective Maximizing)
Calculate the Expected payoff of Firm D (or in other words, the
amount you expect to make if you use Firm D)
Calculate the Expected payoff of Firm C from the Cumulative risk
profiles given in previous slide. Compare the two, which one is
bigger?
What about vice versa? That is, if one strategy has higher
expected value than another, does this mean that this
strategy stochastically dominates the other?
NO! Exercise: Find an example for this statement.
68
Summary
In today’s lecture we covered:
How to choose a preferred alternative?
EMV Approach;
Risk Profiles and Cumulative Risk Profiles
Dominance
Deterministic Dominance
Stochastic Dominance
69
Suggested Problems
4.3
4.6
4.8
4.14
70