Decision Analysis

Prepared for
Role of Decision Analysis

 The purpose of business analytic models is to provide decision-

makers with information needed to make decisions.
 Making good decisions requires an assessment of intangible
factors and risk attitudes.
 Decision making is the study of how people make decisions,
particularly when faced with imperfect or uncertain information,
as well as a collection of techniques to support decision choices.
Formulating Decision Problems
 Many decisions involve making a choice between a small set
of decisions with uncertain consequences.
 Decision problems involve:
1. decision alternatives
2. uncertain events that may occur after a decision is made along with
their possible outcomes (which are often called states of nature), and are
defined so that one and only one of them will occur.
3. consequences associated with each decision and outcome, which are
usually expressed as payoffs. Payoffs are often summarized in a payoff
table, a matrix whose rows correspond to decisions and whose columns
correspond to events.
 The decision maker first selects a decision alternative, after which one
of the outcomes of the uncertain event occurs, resulting in the payoff.
Example 16.1: Selecting a Mortgage Instrument

 A family is considering purchasing a new home and wants to finance

$150,000. Three mortgage options are available and the payoff table
for the outcomes is shown below. The payoffs represent total
interest paid under three future interest rate situations.
 The best decision depends on the outcome that may occur. Since you cannot predict the future
outcome with certainty, the question is how to choose the best decision, considering risk.

ARM – Adjusted Rate mortgage

Decision Strategies Without Outcome
Probabilities
 Minimize Objective (e.g. payoffs are costs)
 Aggressive (Optimistic) Strategy
 Choose the decision that minimizes the smallest payoff that can
occur among all outcomes for each decision (minimum strategy).
 Conservative (Pessimistic) Strategy
 Choose the decision that minimizes the largest payoff that can
occur among all outcomes for each decision (minimax strategy).
 Opportunity Loss Strategy
 Choose the decision that minimizes the largest opportunity loss
among all outcomes for each decision (minimax regret)
Example : Mortgage Decision with the
Aggressive Strategy
 Determine the lowest payoff (interest cost) for each type of mortgage, and then
choose the decision with the smallest value (minimin).
Example : Mortgage Decision with the
Conservative Strategy
 Determine the largest payoff (interest cost) for each type of mortgage, and then
choose the decision with the smallest value (minimax).
Understanding Opportunity Loss

 Opportunity loss represents the “regret” that people often feel after making a
nonoptimal decision.
 In general, the opportunity loss associated with any decision and event is the
difference between the best decision for that particular outcome and the payoff for
the decision that was chosen.
 Opportunity losses can be only nonnegative values.
Example : Mortgage Decision with the
Opportunity-Loss Strategy

 Compute the opportunity loss matrix.

Step 1:
Find the best
outcome
(minimum
cost) in each
column.

Step 2:
Subtract the best
column value
from each value
in the column.
Example Continued
 Find the “minimax regret” decision
Step 3: Determine the maximum opportunity loss for each decision,
and then choose the decision with the smallest of these.
 Using this strategy, we would choose the 1-year ARM. This ensures
that, no matter what outcome occurs, we will never be more than
$6,476 away from the least cost we could have incurred.
 Decision Strategies without Outcome Probabilities

 Maximize Objective (e.g. payoffs are profits)

 Aggressive (Optimistic) Strategy
 Choose the decision that maximizes the largest payoff that can occur among all
outcomes for each decision (maximax strategy).
 Conservative (Pessimistic) Strategy
 Choose the decision that maximizes the smallest payoff that can occur among all
outcomes for each decision (maximin strategy).
 Opportunity Loss Strategy
 Choose the decision that minimizes the maximum opportunity loss among all
outcomes for each decision (minimax regret).
 Note that this is the same as for a minimize objective; however, calculation of the
opportunity losses is different.
Decisions with Conflicting Objectives

 Many decisions require some type of tradeoff among conflicting objectives,

such as risk versus reward.
 A simple decision rule can be used whenever one wishes to make an optimal
tradeoff between any two conflicting objectives, one of which is good, and
one of which is bad, that maximizes the ratio of the good objective to the bad.
 First, display the tradeoffs on a chart with the “good” objective on the x-axis, and the “bad”
objective on the y-axis, making sure to scale the axes properly to display the origin (0,0).
 Then graph the tangent line to the tradeoff curve that goes through the origin.
 The point at which the tangent line touches the curve (which represents the smallest slope)
represents the best return to risk tradeoff.
Example : Risk-Reward Tradeoff Decision for
Innis Investments Example
 From Figure below, if we take the ratios of the weighted returns to the minimum risk
values in the table, we will find that the largest ratio occurs for the target return of 6%.

 We can explain this easily from the chart by noting that for any other return, the risk is
relatively larger (if all points fell on the tangent line, the risk would increase
proportionately with the return).
Summary of Decision Strategies Under Uncertainty
Decision Strategies with Outcome
Probabilities
 In many situations, we might have some assessment of
probabilities, either through some method of forecasting or
reliance on expert opinions.
 If we can assess a probability for each outcome, we can choose
the best decision based on the expected value.
 The simplest case is to assume that each outcome is equally likely to
occur; that is, the probability of each outcome is 1/N, where N is the
number of possible outcomes. This is called the average payoff strategy.
Example : Mortgage Decision with the Average
Payoff Strategy

 Estimates for the probabilities of each outcome are shown in the table below.
 For each loan type, compute the expected value of the interest cost and choose the
minimum.
Expected Value Strategy

 A more general case is when the probabilities of the outcomes

are not all the same. This is called the expected value strategy.
 We may use the expected value calculation using following
formula
Example : Mortgage Decision with the Expected Value
Strategy
 Estimates for the probabilities of each outcome are shown in the table below.
 For each loan type, compute the expected value of the interest cost and choose the
minimum.

Probabilities
Evaluating Risk
 An implicit assumption in using the average payoff or expected value
strategy is that the decision is repeated a large number of times.
However, for any one-time decision (with the trivial exception of equal
payoffs), the expected value outcome will never occur – only one the
actual outcomes will occur for the decision chosen.

 For a one-time decision, we must carefully weigh the risk associated

with the decision in lieu of blindly choosing the expected value
decision.
Example : Evaluating Risk in the Mortgage
Decision
 Standard deviation of each decision:

 Based solely on the standard deviation, the 30-year fixed

mortgage has no risk at all, whereas the 1-year ARM appears
to be the riskiest.
 While none of the previous decision strategies chose the 3-year ARM,
it may be attractive to the family due to its moderate risk level and
potential upside at stable and falling interest rates.
Decision Trees
 A decision tree is a graphical model used to structure a
decision problem involving uncertainty.
 Nodes are points in time at which events take place.
 Decision nodes are nodes in which a decision takes place by choosing
among several alternatives (typically denoted as squares).
 Event nodes are nodes in which an event occurs not controlled by the
decision-maker (typically denoted as circles).
 Branches are associated with decisions and events.
 Decision trees model sequences of decisions and outcomes
over time.
Creating Decision Trees in Analytic Solver Platform

 Click Decision Tree button

 To add a node, select Add Node from the Node dropdown list.
Creating Decision Trees in Analytic Solver
Platform
 Click on the radio button for the
type of node you wish to create
(decision or event). This displays
one of the dialogs shown.
 For a decision node, enter the name of
the node and names of the branches
that emanate from the node (you
may also add additional ones). The
Value field can be used to input cash
flows, costs, or revenues that result
from choosing a particular branch.
 For an event node, enter the name of
the node and branches. The Chance
field allows you to enter the
probabilities of the events.
Example : Creating a Decision Tree
 Mortgage selection problem

 To start the decision tree, add a node for selection of the loan type.
 Then, for each type of loan, add a node for selection of the
uncertain interest rate conditions.
 Finally, enter the payoffs of the outcomes associated with each
event in the cells immediately below the branches
Example Continued

 First partial decision tree

 Second partial decision tree

 payoffs
Example Continued

 Full decision tree

with rollback
Expected value
calculations

Best decision branch

(#2: 3 Year ARM)
Example for Exercise : A Pharmaceutical R&D Model
 Moore Pharmaceuticals needs to decide whether to conduct
clinical trials and seek FDA approval for a newly developed
drug.
 $300 million has already been spent on research.
 The next decision is whether to conduct clinical trials at a cost of $250
million.
 Probability of success following trials is 0.3.
 If the trials are successful, the next decision is whether to seek FDA
approval, costing $25 million.
 Likelihood of FDA approval is 60%.
 If released to the market, revenue potential and probabilities are:
Example for Exercise: Continued

If successful, seek
approval

Choose to
conduct trials

If approved,
expected revenue
Decision Trees and Risk
 Decision trees are an example of expected value decision
making and do not explicitly consider risk.
 For Moore Pharmaceutical’s decision tree, we can form a
classical decision table.

 We can then apply aggressive, conservative, and opportunity

loss decision strategies.
Aggressive Strategy (Maximax)

 Developing the new drug maximizes the maximum payoff.

Conservative Strategy (Maximin)

 Stopping development of the new drug maximizes the

minimum payoff.
Opportunity Loss Strategy

Opportunity Losses

 Developing the new drug minimizes the maximum opportunity

loss.
Example : Constructing a Risk Profile
 Each decision strategy has an associated payoff distribution,
called a risk profile.
 Risk profiles show the possible payoff values that can occur and their
probabilities.
 Outcomes and probabilities:

 The probabilities are computed by multiplying the probabilities on the

18-
event branches along the path to the terminal outcome.
Example Continued

 Computing the probability of “Market large”

Probability = 0.3 × 0.6 × 0.6
Sensitivity Analysis in Decision Trees

 We may use Excel data tables to investigate the sensitivity of the optimal
decision to changes in probabilities or payoff values.

 Airline Revenue Management example Full and discount airfares are available for a flight.
 Full-fare ticket costs $560
 Discount ticket costs $400
 X = selling price of a ticket
 p = 0.75 (the probability of selling a full-fare ticket)
 E[X] = 0.75($560) + 0.25(0) = $420
 Breakeven point: $400 = p($560) or p = 0.714
Example : Sensitivity Analysis for Airline
Revenue Management Decisions
 Decision tree and data table for varying the probability of success with
two output columns, one providing the expected value from cell A10 in
the tree and the second providing the best decision.
 The formula in cell N3 is =A10
 The formula in cell O3 is =IF(B9=1, “Full”, “Discount”).
 The formula in cell H6 is =1-H1. Use H1 as column input cell in the data tables.
The Value of Information
 The value of information is the improvement in the expected
return if the decision maker can acquire additional information
about the future event that will take place.
 Perfect information tell us, with certainty, which outcome will
occur.
 Expected value of perfect information (EVPI) is expected value
with perfect information minus the expected value without it.
 Expected opportunity loss is the average additional amount the
decision maker would have achieved if the correct decision had
been made.
 Minimizing expected opportunity loss always results in the same decision as
maximizing expected value.
Example : Finding EVPI for the Mortgage-
Selection Decision
 Find the minimum expected opportunity loss

Opportunity Losses

= EVPI
Example Continued
 Alternate interpretation
 For each outcome (perfect information), find the best decision; then compute
the expected value

 Compute expected payoff of the best decisions:

0.6 × $54,658 + 0.3 × $46,443 + 0.1 × $40,161=$50,743.80
 Without perfect information, the best decision is the 3-year ARM with an
expected cost of $54,135.20. EVPI is the difference (amount saved by having
perfect information): $54,135.20 - $50,743.80 = $3,391.40.
Decisions with Sample Information

 Sample information is the result of conducting some type of

experiment, such as a market research study or interviewing an
expert.
 The expected value of sample information (EVSI) is the expected
value with sample information (assumed at no cost) minus the
expected value without sample information; it represents the
most you should be willing to pay for the sample information.
Example : Decisions with Sample Information
 A company is developing a new cell phone and currently has
two models under consideration.
 Historically, 70% of their new phones have had high consumer
demand and 30% have had low consumer demand.
 Model 1 requires $200,000 investment.
 If demand is high, revenue = $500,000
 If demand is low, revenue = $160,000
 Model 2 requires $175,000 investment.
 If demand is high, revenue = $450,000
 If demand is low, revenue = $160,000
Example Continued
 Decision tree (values in thousands)

Best decision is to
select model 1
Example Continued

 A market research study is conducted to obtain sample

information about consumer demand.
 Similar studies have found:
 90% of all products that had high consumer demand had previously
received high market survey responses.
 20% of all products that had low consumer demand had previously
received high market survey responses.
 We should expect that a high survey response would increase the
historical probability of high demand, whereas a low survey response
would increase the historical probability of a low demand.
 We need to compute conditional probabilities: P(demand |
survey response)
Bayes’s Rule
 Bayes’s rule allows revising historical probabilities based on
sample information.
Example : Applying Bayes’s Rule to Compute Conditional
Probabilities
 Define
 A1 = High consumer demand P(A1) = 0.70
 A2 = Low consumer demand P(A2) = 0.30
 B1 = High survey response
 B2 = Low survey response
 P(B1 |A1) = 0.90; therefore, P(ML |DH) = 1 − 0.90 = 0.10
 P(B1 |A2) = 0.20; therefore, P(ML |DL) = 1 − 0.20 = 0.80

 Using Bayes’s rule

P(A1 |B1) = (.9)(.7) / [(.9)(.7)+(.2)(.3)] = 0.913

P(A2 |B1) = 1 − 0.913 = 0.087

P(A1 |B1) = (.1)(.7) / [(.1)(.7)+(.8)(.3)] = 0.226

P(A2 |B2) = 1 − 0.226 = 0.774
Example Continued

 Compute marginal probabilities

 P(B1) = P(B1 |A1)*P(A1) + P(B1 |A2)*P(A2)
= (.9)(.7) + (.2)(.3)
= 0.69
 P(B2) = P(B2 |A1)*P(A1) + P(B2 |A2)*P(A2)
= (.1)(.7) + (.8)(.3)
= 0.31
Decision Tree with Market Survey
Information
 Select model 1 if the
survey response is
high; and if the
response is low, then
select model 2.
 EVSI = $202,257 -
$198,000 = $4,257.
Utility and Decision Making
 Utility theory is an approach for assessing risk attitudes
quantitatively.
 This approach quantifies a decision maker’s relative
preferences for particular outcomes.
 We can determine an individual’s utility function by posing a
series of decision scenarios.
Example : A Personal Investment Decision

 Suppose you have $10,000 to invest short-term.

 You are considering 3 options:
1. Bank CD paying 4% return

2. Bond fund with uncertain return

3. Stock fund with uncertain return
 Bond and stock funds are sensitive to interest rates
Constructing a Utility Function
 Sort the payoffs from highest to lowest.
 Assign a utility to the highest payoff of U(X) = 1.
 Assign a utility to the lowest payoff of U(X) = 0.
 For each payoff between the highest and lowest, consider the
following situation:
 Suppose you have the opportunity of achieving a guaranteed return of
x or taking a chance of receiving the highest payoff with probability p
or the lowest payoff with probability 1 - p .
 The term certainty equivalent represents the amount that a decision
maker feels is equivalent to an uncertain gamble.
 What value of p would make you indifferent to these two choices?
 Then repeat this process for each payoff.
Example : Constructing a Utility Function for
the Personal Investment Decision
U(1700) = 1
U(1000) = the probability you would give up
a certain $1000 to possibly win a
$1700 payoff. Suppose this is 0.9.

U(−900) = 0

Decision tree characterization:

Example Continued

 Final utility function

Risk Premium
 The risk premium is the amount an individual is willing to
forgo to avoid risk.
 For the payoff of $1000, the expected value of taking the gamble is
0.9($1,700) + 0.1(- $900) = $1,440. You require a risk premium of
$1,440 - $1,000 = $440 to feel comfortable enough to risk losing $900
if you take the gamble. Such an individual is risk-averse.
Example Utility Function for Risk-Takers

 For the payoff of $1,000, this individual

would be indifferent between receiving
$1,000 and taking a chance at $1,700 with
probability 0.6 and losing $900 with
probability 0.4.
 The expected value of this gamble is
0.6($1,700) + 0.4(-$900) = $660
 Because this is considerably less than $1,000,
the individual is taking a larger risk to try to
receive $1,700.
Using Utility Functions in Decision Analysis
 Replace payoffs with utilities.
 Example using average payoff strategy:

 If probabilities are known, find the expected utility.

Exponential Utility Functions
 An exponential utility function approximates those of risk-averse individuals:

 R is a shape parameter indicative of risk tolerance.

 Smaller values of R result in a more concave shape and are more risk averse.
Estimating the Value of R
 Find the maximum payoff $R for which the decision maker believes that taking a
chance to win $R is equivalent to losing $R/2.
 Would you take on a bet of possibly winning $10 versus losing $5?
 How about risking $5,000 to win $10,000?
Example : Using an Exponential Utility
Function
 For the personal investment example, suppose that R = $400.
 U(X) = 1 – e-X/400

 Use these utilities in the payoff table

Thank You !