1

Introduction to Decision Theory

Decision Making Using Posterior (or revised) Probabilities

 States of nature: A set of potential future conditions that affects decision results
Alternatives: A set of alternative actions for the decision maker to choose from Payoffs: A set of payoffs for each alternative under each potential state of nature

Maximin: Identify the minimum (or worst) possible payoff for each alternative and select the alternative that maximizes the worst possible payoff (Pessimistic) Maximax: Identify the maximum (or best) possible payoff for each alternative and select the alternative that maximizes the best possible payoff (Optimistic)

Expected monetary value (EMV) criterion: Using prior probabilities for the states of nature, compute the expected payoff for each alternative and select the alternative with the largest expected payoff

A Payoff Table for the Condominium Complex Situation

1.

If a small complex is built, the worst possible payoff is $8 million

2.

If a medium complex is built, the worst possible payoff is $5 million

If a large complex is built, the worst possible payoff is -$11 million

3.

Since the maximum of these worst possible payoffs is $8 million, the developer should choose to build a small complex

1.

If a small complex is built, the best possible payoff is $8 million If a medium complex is built, the best possible payoff is $15 million If a large complex is built, the best possible payoff is $22 million

2.

3.

Since the maximum of these best possible payoffs is $22 million, the developer should choose to build a large complex.

If prior probabilities are 0.3 for low demand and 0.7 for high demand there is risk of probability errors 1. Small complex:
Expected value = 0.3($8 million) + 0.7($8 million) = $8 million
2.

Medium complex:
Expected value = 0.3($5 million) + 0.7($15 million) = $12 million

3.

Large complex:
Expected value = 0.3(-$11 million)+.7($22 million)=$12.1 million

Choose large complex

Sometimes it is possible to determine exactly which state of nature will occur in the future, e.g., by obtaining perfect information EVPI = Expected value under certainty expected value under risk To calculate EVcertainty : find the highest payoff for each state of nature ($8 mil for low and $22 mil for high demand) and then: EVcertainty EVPI = 0.3 * $8 + 0.7 * $22 = $17.8 million = $17.8 - $12.1 = $5.7 million

A Payoff Table for the capacity expansion plan for a company is given below
Alternatives Small Facilities Medium Facilities Large Facilities Possible Future Demand Low Moderate High $10 $10 $10 $7 $12 $12 -$4 $2 $16

1. Find the best alternative (and the resulting payoff) if it is known with certainty that demand will be moderate. 2. Find the alternative using maximin and maximax criterion 3. Construct a decision tree assuming prior probability of low, moderate, and high demand are 0.3, 0.5, and 0.2 4. Find the expected monetary value (EMV) for each alternative (small, medium, large). What is the best alternative? Find EVPI.

1. Medium facility, $12 million. 2. Minimum payoffs:

Large = $4 M, Medium = $7M, Small = $10M

Maximum payoff of the minimums (maximin): Small facility at $10M Maximum payoff :

Small = $10M, Medium = $12M, Large = $16M

Maximax payoff: Large facility at $16M

3. Dec. Tree

4. EMV (Small) = (10)(.3) + (10)(.5) + (10)(.2) = $10M EMV(Medium) = (7)(.3) + (12)(.5) + (12)(.2) = $10.5M EMV(Large) = (4)(.3) + (2)(.5) + (16)(.2) = $3M Best alternative: Medium facility

4.

EVcertainty
EVPI

= (10)(.3) + (12)(.5) + (16)(.2) = $12.2M

= $12.2M - $10.5M = $1.7M
Possible Future Demand Low Moderate High $10 $10 $10 $7 $12 $12 -$4 $2 $16

Alternatives Small Facilities Medium Facilities Large Facilities

When we use expected value to choose the best alternative, we call this prior decision analysis Often, sample information can be obtained to help us make a better decision In this case, we compute expected values by using posterior probabilities We call this posterior decision analysis

Prior Probability New Information

Revised Probability

The probability of an event A, given that the event B has occurred:

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

An oil company trying to decide about do drilling or do not.

There are three states of nature

1. No oil (S1) P(None) = 0.7

2. Some oil (S2)

3. Much oil (S3)

P(Some) = 0.2
P(Much) = 0.1

State of Nature Alternatives Drill No Oil (0.7) -$700,000 Some Oil (0.2) $500,000 Much Oil (0.1) $2,000,000

Do not drill

$0

$0

$0

EV(drill) = 0.7x(-$700,000) + 0.2x$500,000 + 0.1x$2,000,000 = -$190,000 EV(Do not drill) = $0 Prior Analysis tells us Do not drill.

The Oil Drilling Case - Summary

Prior Analysis tells us Do not drill. However, there is possibility for Seismic Experiment . Seismic Experiment has three readings: low, medium and high Info of posterior probability (e.g. from historical data):
P(low|no oil)=0.91 P(medium|no oil)=0.05 P(high|no oil)=0.04 P(low|some oil)=0.04 P(medium|some oil)=0.94 P(high|some oil)=0.02 P(low|much oil)=0.01 P(medium|much oil)=0.03 P(high|much oil)=0.96

The questions:
1) 2)

Do seismic survey (cost $100,000) or not If the seismic result=low, Drill or Do not drill? If the seismic result=medium, Drill or Do not drill? If the seismic result=high, Drill or Do not drill?

Previously : P(none) = 0.7, P(some) = 0.2, P(much) = 0.1 Posterior probability of high reading: P(high|none) = 0.04, P(high|some) = 0.02, P(high|much) = 0.96
P(none) P(high | none) 0.7(0.04) 0.21875 P(high ) 0.128 P( some) P(high | some) 0.2(0.02) P( some | high ) 0.03125 P(high ) 0.128 P(much) P(high | much) 0.1(0.96) P(much | high ) 0.75000 P(high ) 0.128 P(none | high ) P(high ) P(none high ) P( some high ) P(much high ) P(none) P(high | none) P( some) P(high | some) P(high | much) (0.7)(0.04) (0.2)(0.02) (0.1)(0.96) 0.128

Expected payoff of no sampling

Expected payoff of sampling

 Expected

payoff of sampling EPS = .646($0) + 0.226($334,061) + 0.128($1,362,500) = $249,898 payoff of no sampling = 0

Expected Expected

value of sample information EVSI = EPS EPNS EVSI = $249,898 - $0 = $249,898 net gain of sampling ENGS = EVSI - $100,000 ENGS = $249,898 - $100,000 = $149,898

Expected

An investor wishes to choose between investing money in (1 ) a high-risk stock, (2) a low-risk stock, or (3) a savings account. The payoffs received depend on the behavior of the stock market - whether the market goes up, stays the same, or goes down over the investment period. The investor can hire an economist who will predict the future market behavior. The results of the consultation: (1) "economist says up (2) "economist says flat" (the same), or (3) economist says down. See the following table
Investment Decision Economist's Prediction Economist says up Economist says flat Economist says down True Market State Up Flat Down 0.80 0.15 0.20 0.10 0.70 0.20 0.10 0.15 0.60

In this table P(economist says up | market up) = 0.80. Prior probabilities of market state up, flat, and down = 0.5, 0.3, and 0.2.

The next figure gives an incomplete decision tree for the investor's situation.
1. a. Decision Alternatives? b. States of Nature? c. Sampling information? 2. Carry out a prior analysis of the investor's decision problem and determine the best investment choice 3. Set up probability revision tables to find: Probability economist says up, Posterior probabilities market up, market flat, and market down given that the "economist says up." Repeat for Probability economist says flat and economist says down Fill the decision tree

4. Carry out a posterior analysis of the investor's decision problem. That is, determine the best investment choice that should be made 5. Calculate the EPS, the EPNS and the EVSI

1. a. Alternatives: Savings Account, High-Risk stock, Low-Risk stock. b. States of Nature: Market up, Market flat, Market down c. Sampling info: says up, says down, says flat 2. EV(HighRisk) = (1500)(.5) + (100)(.3) + (1000)(.2) = 580 EV(LowRisk) = (1000)(.5) + (200)(.3) + (100)(.2) = 540 EV(SavingsAcct) = (500)(.5) + (500)(.3) + (500)(.2) = 500

Choose HighRisk stock; payoff = 580

3.

3.

a.

P(economist says up) = (.5)(.8) + (.15)(.3) + (.2)(.2) = .485 Posterior Probabilities: .8247, .0928, .0825

b.

P(economist says flat) = (.5)(.1) + (.7)(.3) + (.2)(.2) = .300 Posterior Probabilities: .1667, .7000, .1333

c.

P(economist says flat) = (.5)(.1) + (.15)(.3) + (.6)(.2) = .215 Posterior Probabilities: .2326, .2093, .5581

4.

a.

Economist says: Market Up

EMV(HighRisk) = (1500)(.8247) + (100)(.0928) + (1000)(.0825) = 1163.83 EMV(LowRisk) = (1000)(.8247) + (200)(.0928) + (100)(.0825) = 835.05

EMV(SavingsAcct) = (500)(.8247) + (500)(.0928) + (500)(.0825) = 500

Choose High-Risk stock b. Economist says: Market Flat

EMV(HighRisk) = (1500)(.1667) + (100)(.7) + (1000)(.1333) = 186.75
EMV(LowRisk) = (1000)(.1667) + (200)(.7) + (100)(.1333) = 293.37 EMV(SavingsAcct) = 500

Choose Savings Acct c. Economist says: Market Down

EMV(HighRisk) = (1500)(.2326) + (100)(.2093) + (1000)(.5581) = 188.27 EMV(LowRisk) = (1000)(.2326) + (200)(.2093) + (100)(.5581) = 218.65 EMV(SavingsAcct) = 500

Choose Savings Acct

4.

a.
b. c.

EPS = (.485)(1163.92) + (.3)(500) + (.215)(500) = 821.96

EPNS = 580 EVSI = 821.96 580 = 241.96

d.

Max amount to pay economist for advice: 242.00