Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision...

28
Decision Analysi s 1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis

Transcript of Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision...

Page 1: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 1

DSC 3120 Generalized Modeling

Techniques with Applications

Part III. Decision Analysis

Page 2: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 2

Decision Analysis A Rational and Systematic Approach to

Decision Making

Decision Making: choose the “best” from several available alternative courses of action

Key Element is Uncertainty of the outcome• We, as decision maker, control the decision• Outcome of the decision is uncertain to and

uncontrolled by decision maker (controlled by nature)

Page 3: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 3

Example of Decision Analysis

You have $10,000 for investing in one of the three options: Stock, Mutual Fund, and CD. What is the best choice?

Question: Do you know the choices?

Do you know the best choice?

What is the uncertainty?

How do you make your choice?

Page 4: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 4

Components of Decision Problem Alternative Actions -- Decisions

• There are several alternatives from which we want to choose the best

States of Nature -- Outcomes• There are several possible outcomes but which

one will occur is uncertain to us

Payoffs• Numerical (monetary) value representing the

consequence of a particular alternative action we choose and a state of nature that occurs later on

Page 5: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 5

Payoff TableState of Nature

Alternative S1 S2 Sm

A1 r11 r12 r1m

A2 r21 r22 r2m

An rn1 rn2 rnm

Page 6: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 6

An Example

S1

RainS2

No Rain

A1

UmbrellaOutcome (A1S1)Don’t get wet but alittle inconvenient.(80)

Outcome (A1S2)Cumbersome andinconvenient(-20)

A2

No UmbrellaOutcome (A2S1)Get wet (and possiblyget sick) (-40)

Outcome (A2S2)The best!(100)

State of Nature

Alt

ern

ativ

e

Page 7: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 7

Three Classes of Decision Models Decision Making Under Certainty

• Only one state of nature (or we know with 100% sure what will happen)

Decision Making Under Uncertainty (ignorance)

• Several possible states of nature, but we have no idea about the likelihood of each possible state

Decision Making Under Risk• Several possible states of nature, and we have an

estimate of the probability for each state

Page 8: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 8

Decision Making Under Uncertainty LaPlace (Assume Equal Likely States of Nature)

• Select alternative with best average payoff

Maximax (Assume The Best State of Nature)• Select alternative that will maximize the maximum payoff

(expect the best outcome--optimistic) Maximin (Assume The Worst State of Nature)

• Select alternative that will maximize the minimum payoff (expect the worst situation--pessimistic)

Minimax Regret (Don’t Want to Regret Too Much)• Select alternative that will minimize the maximum regret

Page 9: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 9

Payoff Table

Example: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3

0 0 -50 -100 -150

1 -40 35 -15 -65

2 -80 -5 70 20

3 -120 -45 30 105

Page 10: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 10

LaPlace Criterion State of Nature (Demand)

Alternative(Order) 0 1 2 3 Mean

0 0 -50 -100 -150 -75

1 -40 35 -15 -65 -21.25

2* -80 -5 70 20 1.25*

3 -120 -45 30 105 -7.5

Example: Newsboy Problem

Page 11: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 11

Maximax Criterion

Example: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3 Max

0 0 -50 -100 -150 0

1 -40 35 -15 -65 35

2 -80 -5 70 20 70

3* -120 -45 30 105 105*

Page 12: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 12

Maximin Criterion

Example: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3 Min

0 0 -50 -100 -150 -150

1* -40 35 -15 -65 -65*

2 -80 -5 70 20 -80

3 -120 -45 30 105 -120

Page 13: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 13

Minimax Regret Criterion: Step 1

Example: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3

0 0 -50 -100 -150

1 -40 35 -15 -65

2 -80 -5 70 20

3 -120 -45 30 105

Best 0 35 70 105

Page 14: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 14

Minimax Regret: Step 2 (Regret or Opportunity Loss Table)

Example: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3 Max

0 0 85 170 255 255

1 40 0 85 170 170

2* 80 40 0 85 85*

3 120 80 40 0 120

Page 15: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 15

Decision Making Under Risk

State of NatureAlternative S1 S2 Sm

A1 r11 r12 r1m

A2 r21 r22 r2m

An rn1 rn2 rnm

Probability p1 p2 pm

• In this situation, we have more information about the uncertainty--probability

Page 16: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 16

Decision Making Under Risk Maximize Expected Return (ER)

ERi = (pj rij) = p1ri1 + p2ri2 +…+ pmrim

Where ERi = Expected return if choosing the ith

alternative (Ai), (i = 1, 2, …, n)

pj = The probability of state j (Sj)

rij = The payoff if we choose alternative Ai

and Sj state of nature occurs

Page 17: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 17

Expected Return & VarianceExample: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3 ER Variance

0 0 -50 -100 -150 -85 2025

1 -40 35 -15 -65 -12.5 1306.25

2* -80 -5 70 20 22.5* 2181.25

3 -120 -45 30 105 7.5 4556.25

Probability 0.1 0.3 0.4 0.2

Page 18: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 18

Decision Making Under Risk High return is good, but on the other hand,

low risk is also important Variance -- a measure of the risk

Variancei = pj (rij - ERi)2

Where pj = The probability of state j (Sj)

rij = The payoff if choose Ai and Sj occurs

ERi= Expected return for alternative Ai

Page 19: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 19

Expected Value of Perfect Information

EVPI measures the maximum worth (value) of the “Perfect Information” that we should pay for in order to improve our decisions

EVPI = ER w/ perfect info. - ER w/o perfect info.

• ER w/ perfect info. = pj max(rij)

• ER w/o perfect info. = max(ERi)

= max( pj rij)

Page 20: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 20

Calculate EVPIExample: Newsboy Problem

State of Nature (Demand)

Alternative(Order)

0 1 2 3 ER

0 0 -50 -100 -150 -85

1 -40 35 -15 -65 -12.5

2 -80 -5 70 20 22.5

3 -120 -45 30 105 7.5

Best 0 35 70 105 59.5

Probability 0.1 0.3 0.4 0.2 37.0

ER w/ PI

ER w/o PI

EVPI

Page 21: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 21

Expected Opportunity Loss (EOL) We can also use EOL to choose the best

alternative

Minimizing EOL = Maximizing ER

• both criteria yield the same best alternative

EOLi = pj OLij

where pj = The probability of state j (Sj)

OLij = The opportunity loss if choose Ai and Sj occurs

min(EOLi) = EVPI

Page 22: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 22

Expected Opportunity LossExample: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3 EOL

0 0 85 170 255 144.5

1 40 0 85 170 72

2* 80 40 0 85 37*

3 120 80 40 0 52

Probability 0.1 0.3 0.4 0.2

EVPI

Page 23: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 23

Decision Making with Utilities Problem with Monetary Payoffs

• People do not always just look at the highest expected monetary return to make decisions; they often evaluate the risk

• Example: A company wants to decide to develop a new product or not

Success Fail ER

Develop $1,000,000 -$400,000 $20,000

Don't develop 0 0 0

Probability 0.3 0.7

Page 24: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 24

Decision Making with Utilities Utility -- combines monetary return with people’s

attitude toward risk Utility Function -- a mathematical function that

transforms monetary values into utility values• Three general types of utility functions

0 MV

Utility

0 MV

Utility

0 MV

Utility

(1) Risk-Averse (2) Risk-Neutral (3) Risk-Seeking

Page 25: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 25

Risk-Averse Utility FunctionUtility

Properties of Risk-averse Utility Function • non-decreasing: more money is always better

• concave: utility increase for unit ($100, e.g.) increase of money is decreasing (extra money is less attractive)

100 200 300 400 500 Dollars0

0.524

0.680

0.7750.8500.910

Page 26: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 26

How to Create Utility Function Method I. Equivalent Lottery

Start with two endpoints A (the worst possible payoff) and B (the best possible payoff) and assign U(A) = 0 and U(B) = 1

Then to find the utility for a possible payoff z between A and B, select the probability p (=U(z)) such that you are indifferent between the following two alternatives

– receive a payoff of z for sure

– receive a payoff of B with probability p or a payoff of A with probability 1 - p

Page 27: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 27

How to Create Utility Function Method II. Exponential Utility Function

rxexU /1)( where x is the monetary value, r>0 is an adjustable

parameter called risk tolerance First, the value of r can be estimated such that we are

indifferent between the following choices a payoff of zero a payoff of r dollars or a loss of r/2 dollars with 50-50 chance

Then the utility for a particular monetary value x can be found using the above assumed exponential utility function

Page 28: Decision Analysis1 DSC 3120 Generalized Modeling Techniques with Applications Part III. Decision Analysis.

Decision Analysis 28

Expected UtilityExample: Newsboy Problem

State of Nature (Demand)

Alternative(Order) 0 1 2 3 EU

0 0 -0.65 -1.72 -3.48 -1.58

1 -0.49 0.30 -0.16 -0.92 -0.21

2* -1.23 -0.05 0.50 0.18 0.10*

3 -2.32 -0.57 0.26 0.65 -0.17

Probability 0.1 0.3 0.4 0.2