BU275 W23 - Lecture 3 - Decision Analysis (Part 3)

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 24

D AT E | M O N T H | Y E A R BU275: Business Decision

Presented by
First Name Last Name Models
Instructor – Dr Jason Hurley
WINTER 2023
Decision Analysis (part 3)
D AT E | M O N T H | Y E A R

Presented by
First Name Last Name
Obtaining More Information

• Is it possible to spend money to acquire more information (i.e.,


better estimates of the true state of nature/outcome that will occur)?

• If so, is the expenditure worthwhile?


D AT E | M O N T H | Y E A R

• Check by calculating
Presented by
First Name Last Name
EVPI and comparing to the cost of obtaining the
information
• If EVPI > Cost, then it may be worthwhile to obtain more info
• If EVPI < Cost, then it is not worthwhile to obtain more info

• Recall: EV of PI (EVPI) = EV with PI – EV without PI


Decision Tree with Perfect Information
• Obtain “perfect information” about the state of nature before making a
decision – reverse the chronological order of outcome and alternatives

D AT E | M O N T H | Y E A R

Presented by
EV WITH perfect information = Expected Payoff =
(0.25*700,000)+(0.75*90,000)Last
First Name = Name Max(700,000,90,000)

Expected Payoff =
Max(-100,000,90,000)
Decision Making with Sample Information
Obtaining perfect information about the future is rare, but what if we could gain some
amount of additional (imperfect) information that will improve decisions?

• New knowledge of sample/survey information can be used to revise


probability estimates for potential outcomes
D AT E | M O N T H | Y E A R

Presented by
• The prior
First probabilities
of the possible states of nature often are quite
Name Last Name
subjective in nature; potentially rough estimates

• It is often possible to do additional research, testing or surveying (at


some expense) to improve these estimates
• The improved estimates are called posterior probabilities
Calculating Posterior Probabilities

• With knowledge of conditional probabilities, P(finding|state), for


findings (indicator) of the sample or survey, prior probabilities, P(state),
can be revised
• For each state of nature, multiply prior probability by conditional
probability
D AT E | 
M O Ngives
T H | Y E Ajoint
R probabilities, P(finding and state)
• Sum joint probabilities
Presented by over all states  gives marginal (prior)
First Name Last Name
probability for findings (indicator)
• For each state, divide joint probability by marginal (prior) probability for
findings (indicator)  gives posterior probabilities
Bayesian Updating

The formula for calculating posterior probability is known as


Bayes’ theorem:

D AT E | M O N T H | Y E A R

where: Presented by
P(B|A)First
is Name Last Name that event B will occur given that event A has occurred
the probability
P(A|B) is the probability that event A will occur given that event B has occurred
P(A|B’) is the probability that event A will occur given that event B’ has occurred
P(A) is the probability that event A will occur
P(B) is the probability that event B will occur
P(B’) is the probability that event B’ will occur
Bayesian Updating

The formula to compute posterior probability is known as


Bayes’ theorem

D AT E | M O N T H | Y E A R

Presented by
First Name Last Name
Goferbroke Example continued…

The Goferbroke Co. can obtain improved estimates of the


chance of oil by conducting a detailed seismic survey of the
land (at a cost of $30,000).
D AT E | M O N T H | Y E A R

Presented by
Possible findings from a seismic survey:
First Name Last Name

1. FSS: Favorable Seismic Soundings – oil is likely.


2. USS: Unfavorable Seismic Soundings – oil is unlikely.
Goferbroke Example continued…

Conditional findings, P(finding|state), from a seismic survey:

• If the land has oil, 60% of the time the test result is Favorable
•D AT EP(FSS|Oil)
| M O N T H | Y E A=
R 0.6

•Presented
Therefore,
by P(USS|Oil) = 0.4
First Name Last Name

• If the land is dry, 80% of the time the test result is Unfavorable
• P(USS|Dry) = 0.8
• Therefore, P(FSS|Dry) = 0.2
Goferbroke Example continued…
Using Bayes’ Theorem to calculate Posterior Probabilities

Example:

D AT E | M O N T H | Y E A R

Presented by
First Name Last Name
Goferbroke Example continued…

D AT E | M O N T H | Y E A R

Presented by
First Name Last Name
Goferbroke Example continued…
Event Prior Conditional Joint Posterior
Bi P(Bi) x P(FSS|Bi) = P(Bi and FSS) P(Bi|FSS)

Oil 0.25 0.6 0.25(0.6) = 0.15 0.15/0.3 = 0.5

Dry 0.75 0.2 0.75(0.2) = 0.15 0.15/0.3 = 0.5


D AT E | M O N T H | Y E A R

Presented by P(FSS) = 0.3


Event First Prior
Name Last NameConditional Joint Posterior
Bi P(Bi) P(USS|Bi) P(Bi and USS) P(Bi|USS)

Oil 0.25 0.4 0.25(0.4) = 0.1 0.1/0.7 = 0.14


Dry 0.75 0.8 0.75(0.8) = 0.6 0.6/0.7 = 0.86

P(USS) = 0.7
Goferbroke Example continued…

D AT E | M O N T H | Y E A R

Presented by
First Name Last Name
Goferbroke Example continued…
If FSS, choose to
drill for oil.

CHOOSE TO DO
THE SURVEY AS
IT YIELDS
HIGHEST EV

Expected Payoff = D AT E | M O N T H | Y E A R
Max(237,000,100,00)
Presented by
First Name Last Name
If USS, choose to
sell the land.
Expected Value of Sample Information (EVSI)

• EVSI is the maximum amount a decision maker would be


willing to pay for the sample/survey information

D AT E | M O N T H | Y E A R

HowPresented
muchby better off does the info leave the decision maker?
First Name Last Name

EV of SI (EVSI)= EV with Sample Information - EV without Sample Information

EVSI = 237,000 – 100,000 = $137,000


Goferbroke Example continued…

What if we include the cost of


the information/survey (i.e.,
reduce payoffs by cost of info)?

D AT E | M O N T H | Y E A R

Presented by
First Name Last Name

Expected Payoff =
Max(207,000,100,000)
CHOOSE TO DO
THE SURVEY AS
IT YIELDS
HIGHEST EV
EVSI for the Goferbroke Company

• To calculate the EV with sample information (when payoffs are


reduced by the cost of the test/survey prior to rolling back
decision tree), you need to add back the cost of the info/test:

EV with Sample Information = 207,000 + 30,000 = $237,000


D AT E | M O N T H | Y E A R

Presented by
First Name LastEV without Sample Information = $100,000
Name
EVSI = 237,000 – 100,000 = $137,000

Is the $30,000 cost to perform the survey justifiable?

EVSI > Cost, therefore it is justifiable.


Efficiency of Sample Information

• The efficiency of sample information is the ratio of the EV of


sample information to the EV of perfect information:
𝑬𝑽𝑺𝑰
𝑬𝒇𝒇𝒊𝒄𝒊𝒆𝒏𝒄𝒚 𝒐𝒇 𝑺𝒂𝒎𝒑𝒍𝒆 𝑰𝒏𝒇𝒐𝒓𝒎𝒂𝒕𝒊𝒐𝒏=
D AT E | MONTH | YEAR
𝑬𝑽𝑷𝑰
Presented by
Efficiency for Goferbroke’s Seismic survey:
First Name Last Name

Efficiency = 137,000/142,500 = 96.14%

A high efficiency rating indicates that the information is very


good, or close to being perfect
Thanks for your attention!
D AT E | M O N T H | Y E A R

Presented by
First Name Last Name
Next class:
• Decision Analysis problem solving session
Review: Prior (Marginal) Probabilities
Sample space

D AT E | M O N T H | Y E A R

Presented by
First Name Last Name

𝑨
P(A) = 5/18 P(B) = 10/18
P() = 13/18 P() = 8/18
Review: Conditional Probabilities
What is the probability of A occurring, given that B has
occurred?
P(A|B) = P(A and B)/P(B)
D AT E | M O N T H | Y E A R
P(A and B) = 3/18
Presented by P(A|B) = 3/10
First Name Last Name
P(B) = 10/18

P(A and B)/ P(B) = = = P(A|B)


Review: Bayesian Updating (Bayes’ Theorem)
𝟑 𝟏𝟎
( )( )
𝑷 ( 𝑨∨ 𝑩) × 𝑷 ( 𝑩) 𝟏𝟎 𝟏𝟖 𝟑
𝑷 ( 𝑩| 𝑨 )= = =
( )( ) ( )( )
𝑷 ( 𝑨∨ 𝑩) × 𝑷 ( 𝑩 ) + 𝑷 ( 𝑨∨ 𝑩) × 𝑷 ( 𝑩) 𝟑 𝟏𝟎
𝟏𝟎 𝟏𝟖
+
𝟐
𝟖 𝟏𝟖
𝟖 𝟓

P(B|A) = 3/5
D AT E | M O N T H | Y E A R

Presented by
First Name Last Name

𝑨
Review: Bayesian Updating using Tables
Event Prior Conditional Joint Posterior
Bi P(Bi) x P(A|Bi) = P(Bi and A) P(Bi|A)

B 10/18 = 0.556 3/10 = 0.3 3/18 = 0.167 (3/18)/(5/18) = 3/5 = 0.6

B’ 8/18 = 0.444 2/8 = 0.25 2/18 = 0.111 (2/18)/(5/18) = 2/5 = 0.4


D AT E | M O N T H | Y E A R

Presented by P(A) = 5/18 = 0.278


Event First Prior
Name Last NameConditional Joint Posterior
Bi P(Bi) P(A’|Bi) P(Bi and A’) P(Bi|)

B 10/18 = 0.556 7/10 = 0.7 7/18 = 0.389 (7/18)/(13/18) = 7/13 = 0.538


B’ 8/18 = 0.444 6/8 = 0.75 6/18 = 0.333 (6/18)/(13/18) = 6/13 = 0.462

P() = 13/18 = 0.722

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy