Você está na página 1de 46

7

Decision Theory

Operations Research

Introduction
Decision theory is concerned with how to assist people in decision making. The theory provides a meaningful framework for improved decision making.

Operations Research

Decision Analysis
Intelligence

Define the problem


Search for alternatives
Design

Abstraction

Decision theory Assess Consequences Apply decision criteria


Operations Research

Evaluate the alternative


Choice

Select an alternative

Decision Theory
Decision theory steps:
1. 2. 3.

4.

Decision making environment Deterministic Situation (Certainty) Stochastic situation (risk) Situation of uncertainty Objective of decision maker Alternative plans of action or strategies Decision payoff (indication of effectiveness of strategies)

Operations Research

Decision Theory

Deterministic Situation (Certainty)- Action leads only one outcome. Important techniques: linear programming, transportation and assignment problems. Stochastic situation (risk)- many state of and decision maker knows the probability of occurrence. Important techniques: Decision matrix, decision tree, PERT, Markov Chain, Bayesion analysis Situation of uncertainty- Probability associated with state of nature are unknown . Important techniques: Matrix games, minimax actions and strategies.

Operations Research

Decision Theory
Decision process calls for:
1. 2. 3.

Identification of states of nature, or events Identification of courses of action, or acts Determination of the pay-offs, depicting outcomes of various combinations of acts and events (payoffs resulting from various act-event combinations can be either gains/losses or costs) Choosing, on the basis of some criterion, from among different alternatives
Operations Research

4.

Introduction to Decision Analysis


Steps in Decision Theory Approach
Step 1: Determine the various alternative courses of action (or choices or strategies) from which the final decision is to be made. Step 2: Identify the possible outcomes, called the states of nature or events for the decision problem. The events are beyond the control of the decision-maker. Step 3: Construct a pay off table Step 4: The decision-maker will choose the criterion which results in largest pay off. The criterion may be economic, quantitative or qualitative (for example, market share, profit, fragrance of a perfume and so on)

Operations Research

Decision theory question


A book store sells a particular book of Tax Law for Rs 100. it purchases book for Rs. 80 per copy. Unsold copies becomes outdated by the end of year and can be disposed off for Rs. 30. According to past experience, the annual demand for this book is between18 and 23 copies. Assuming the order can be placed only once in an year , the problem before sales manager is to decide on how many number of copies should be purchased for the next year.

Operations Research

Payoff Table
Let Q is Quantity, D is demand, P is Profit If D>Q Then P= 100Q-80Q P = 20 Q. or

If D< Q then P= 100D +30(Q-D)-80Q or P = 70D-50Q

Operations Research

Payoff Table
Event . Ei Act Aj (Course of action) A1 :18 360 360 360 360 360 360 A2: 19 310 380 380 380 380 380 A3: 20 260 330 400 400 400 400 A4: 21 210 280 350 420 420 420 A5:22 160 230 300 370 440 440 A6: 23 110 180 250 320 390 460

Demanded copies
E1 :18 E2: 19 E3: 20 E4: 21 E5: 22 E6: 23

Operations Research

Opportunity Loss or Regret Table


Event . Ei Demanded copies E1 :18 E2: 19 E3: 20 E4: 21 E5: 22 E6: 23 A1 :18 0 20 40 60 80 100 Act Aj (Course of action) A2: 19 50 0 20 40 60 80 A3: 20 100 50 0 20 40 60 A4: 21 150 100 50 0 20 40 A5:22 200 150 100 50 0 20 A6: 23 250 200 150 100 50 0

Operations Research

Decision under uncertainity


Decisions under Uncertainty

Laplace Principle: assumes equal likelihood of various states of nature Maximin (or Minimax in case of pay-offs as cost) Principle: pessimists criterion considers best of the worst Maximax (or Minimin in case of pay-offs as cost) Principle: optimists criterion considers best of the best Hurwicz Principle: uses decision-makers degree of optimism

Savage Principle: uses minimum of the maximum regret values of various acts
Operations Research

Laplace Criterion or Equally Likely Decision Criterion


Step 1: Assign equal probabilities 1/(number of states of nature) to each pay off of a strategy.
Step 2: Determine the expected pay off value for each alternative. Step 3: Select that alternative which corresponds to the maximum (and minimum for cost) of the above expected pay offs.

Operations Research

Decision under uncertainity


Laplace Principle: highest mean value is adopted.

Stock A1 (18) A2 (19) A3 (20) A4 (21) A5 (22) A6 (23)

Mean (Expected) Pay-off (360+360+360+360+360+360)/ 6= Rs 360.0 (310+380+380+380+380+380)/6 = Rs 368.3 (260+330+400+400+400+400)/6 = Rs 365.0 (210+280+350+420+420+420)/6 = Rs 350.0 (160+230+300+370+440+440)/6 = Rs 323.3 (110+180+250+320+390+460)/6 = Rs 285.0

Operations Research

(Wald decision)(Minimax or Maximin)


Step 1: Find the minimum assured pay off for each alternative (course of action). Step 2: Choose that alternative which corresponds to the maximum of the above minimum pay off. FOR COSTS Step 1: Determine the maximum possible cost for each alternative. Step 2: Choose that alternative which corresponds to the minimum of the above costs.
Operations Research

Criterion of Pessimism

Criterion of Pessimism (Maximin or Minimax)


Event Ei Demanded copies E1 :18 E2: 19 E3: 20 E4: 21 E5: 22 E6: 23 Act Aj (Course of action) A1 :18 360 360 360 360 360 360 A2: 19 310 380 380 380 380 380 A3: 20 260 330 400 400 400 400 A4: 21 210 280 350 420 420 420 A5:22 160 230 300 370 440 440 A6: 23 110 180 250 320 390 460

Operations Research

Criterion of Optimism (Maximax or Minimin)


Step 1: Determine the maximum possible pay off for each alternative. Step 2: Select that alternative which corresponds to the maximum of the above maximum pay offs. In decision problems dealing with costs, the minimum for each alternative is considered and then the alternative which minimises the above minimum cost is selected. This is termed as minimin principle.

Operations Research

Criterion of Optimism (Maximax or Minimin)


Event Ei Demanded copies E1 :18 E2: 19 E3: 20 E4: 21 E5: 22 E6: 23 Act Aj (Course of action) A1 :18 360 360 360 360 360 360 A2: 19 310 380 380 380 380 380 A3: 20 260 330 400 400 400 400 A4: 21 210 280 350 420 420 420 A5:22 160 230 300 370 440 440 A6: 23 110 180 250 320 390 460

Operations Research

Criterion of Realism (Hurwicz Criterion)


1. Decide the coefficient of optimism and then the coefficient of pessimism 1 - . 2. Determine the maximum as well as minimum pay off for each alternative and obtain the quantities. h = maximum for each alternative + (1 - ) minimum for each alternative. 3. Select an alternative with maximum value.

Operations Research

Criterion of Realism (Hurwicz Criterion)


Act A1 A2 A3 A4 A5 A6 Max 360 380 400 420 440 460 Min 360 310 260 210 160 110 Criterion Value=(Max Value) +(1-) 0.6 X 360 + 0.4 X 360 0.6 X 380 + 0.4 X 310 0.6 X 400 + 0.4 X 260 0.6 X 420 + 0.4 X 210 0.6 X 440 + 0.4 X 160 0.6 X 460 + 0.4 X 110 360 352 344 336 328 320

Operations Research

Criterion of Regret(savage Criterion) or Minimax Regret Criterion


Step 1: From the given pay off matrix, develop an opportunity-loss (or regret) matrix.
(i) Find the best pay off corresponding to each state of nature (maximum for profit and minimum for cost) (ii) ith regret = (maximum pay off-ith pay off) for the jth event if the pay offs represent profits (minimum pay off-ith pay off) for the jth event if the pay offs represent costs.

Step 2: Determine the maximum regret amount for each alternative. Step 3: Choose that alternative which corresponds to the minimum regrets.
Operations Research

Criterion of Regret(savage Criterion) or Minimax Regret Criterion


Event Ei Demanded copies E1 :18 E2: 19 E3: 20 E4: 21 E5: 22 E6: 23 Act Aj (Course of action) A1 :18 0 20 40 60 80 100 A2: 19 50 0 20 40 60 80 A3: 20 100 50 0 20 40 60 A4: 21 150 100 50 0 20 40 A5:22 200 150 100 50 0 20 A6: 23 250 200 150 100 50 0

Operations Research

Decision Under Certainty


Since under this environment, only one state of nature exists, the decision-maker simply picks up the best pay off in that one column and chooses the associated alternative. Under conditions of certainty, the particular state of nature is associated with probability
Though the state of nature is only one, possible alternatives could be numerous. Linear programming, transportation and assignment techniques, input out analysis, activity analysis and economic order quantity models are used for such situations.

Operations Research

Decision-making Under Risk


The decision situations where the decision maker chooses to consider several possible outcomes and the probabilities of occurrence can be stated under risk. Probability of occurrence can be determined from past records. Suppose the book seller observes from the past sales that for the number of copies sold 18,19,20,21,22,23 copies the the probability 0.05, 0.10, 0.30, 0.40, 0.10, 0.05

Operations Research

Decision-making Under Risk


Under condition of risk, there are two criteria:

Maximum likelihood principle: for event with highest probability, choose the act with best pay-off Expectation principle: choose the act with best expected pay-off (highest in case of gains and lowest in case of costs)

Operations Research

Maximum likelihood principle


Event Ei Probability Demanded Pi copies E1 :18 E2: 19 E3: 20 E4: 21 E5: 22 E6: 23 0.05 0.10 0.30 0.40 0.10 0.05 Act Aj (Course of action) A1 :18 360 360 360 360 360 360 360 A2: 19 310 380 380 380 380 380 376.5 A3: 20 260 330 400 400 400 400 386 A4: 21 210 280 350 420 420 420 374.5 A5:22 160 230 300 370 440 440 335 A6: 23 110 180 250 320 390 460 288.5

Expected pay-off

Operations Research

Decision-making Under Risk


Expectation principle:
Expected Monetary Value (EMV) criterion Expected Opportunity Loss (EOL) criterion Expected Value of Perfect Information (EVPI)

Operations Research

Expected Monetary Value (EMV) Criterion


Step 1: List conditional profit for each act-event combinations, along with the corresponding event probabilities. Step 2: For each act, determine the expected conditional profits. Step 3: Determine EMV for each act where EMV (Aj) = pij Pi pij - pay off ; Pi- probability of occcurrence Step 4: Choose the act which corresponds to the optimal EMV.
Operations Research

Expected Payoff or Monetary value(EMV/ EPV)


EMV A3 = 0.05X 260 + 0.10X 330 + 0.30 X 400 + 0.40 X 400 + 0.10 X 400 +0.05 X 400 = Rs 386
Event Ei Demande d copies E1 :18

Probabilit y Pi
0.05

Act Aj (Course of action) A1 :18 360 360 360 360 360 360 360 A2: 19 310 380 380 380 380 380 376.5 A3: 20 260 330 400 400 400 400 386 A4: 21 210 280 350 420 420 420 374.5 A5:22 160 230 300 370 440 440 335 A6: 23 110 180 250 320 390 460 288.5

E2: 19
E3: 20 E4: 21

0.10
0.30 0.40

E5: 22
E6: 23

0.10
0.05

Expected pay-off

Operations Research

Expected Oppurtunity Loss (EOL) Criterion


Step 1: Construct the conditional profit table for each actevent combination, along with the corresponding event probabilities. Step 2: For each state of nature calculate the Conditional Opportunity Loss (COL) values by subtracting each pay off from the maximum pay off for that event. Step 3: For each act, determine the expected COL values and add these values to get the Expected Opportunity Loss (EOL) for that act where EOL(Ni) = lij pi lij - opppurtunity loss ; pi - probability Step 4: Choose that act which corresponds to the minimum EOL value.
Operations Research

Expected opportunity loss (EOL)


EOL A3 = 0.05X 100 + 0.10X 50 + 0.30 X 0 + 0.40 X 20 + 0.10 X 40 +0.05 X 60 = Rs 25
Act Aj (Course of action) A1 :18 0 20 40 60 80 100 51 A2: 19 50 0 20 40 60 80 34.5 A3: 20 100 50 0 20 40 60 25 A4: 21 150 100 50 0 20 40 36.5 A5:22 200 150 100 50 0 20 76 A6: 23 250 200 150 100 50 0 122.5

Event Ei Probability Demande Pi d copies E1 :18 E2: 19 E3: 20 0.05 0.10 0.30

E4: 21
E5: 22 E6: 23

0.40
0.10 0.05

Expected pay-off

Operations Research

Expected Pay-off of Perfect Information (EPPI)


When the bookstore manager knows that next year demand is 18 copies to obtain profit of 360 with probability 0.05 then Expected profit = 0.05X 360 =Rs. 18 With demand 19 on 2nd year with probability 0.10 then Expected profit = 0.10X 360 =Rs 38 Similarly for the each level of demand each year;

Operations Research

Expected Pay-off of Perfect Information (EPPI)


EPPI = 0.05X 360 + 0.10X 360 + 0.30 X 400 + 0.40 X 420 + 0.10 X 440 +0.05 X 460 = Rs 411
Event Ei Probability Demanded Pi copies E1 :18 E2: 19 E3: 20 0.05 0.10 0.30 Act Aj (Course of action) A1 :18 360 360 360 360 360 360 A2: 19 310 380 380 380 380 380 A3: 20 260 330 400 400 400 400 A4: 21 210 280 350 420 420 420 A5:22 160 230 300 370 440 440 A6: 23 110 180 250 320 390 460

E4: 21
E5: 22 E6: 23

.040
0.10 0.05

Operations Research

Expected Value of Perfect Information (EVPI)


Expected profit with perfect information (EPPI) represents the expected profit, in the long run, if we have perfect information before a decision is made. EVPI = EPPI - Expected profit without information =EPPI -EMV * where EMV* - maximum expected monetary value (386 in above case). EVPI = 411 386 = Rs. 25

Operations Research

Question
A news paper Boy has following probabilities of selling magazine: No. of Copies sold. 10 11 12 13 14 Probability 0.10 0.15 0.20 0.25 0.30

Cost of a copy is 30 paise and selling price is 50 paise. He can not return the unsold copies? How much should he order? Find EVPI.

Operations Research

Decision Trees
Decision tree is a schematic representation of a sequential and multi-dimensional decision problems. Decision tree is made of Nodes, Branches, probability estimates and Payoffs

Operations Research

Decision Trees
Node- indicated by square and represents the place of decision making Branches connects various nodes: has decision branches and chance branches. Branch that indicate end of decision tree is terminal branch Associated Probabilities likelihood that the chance outcome will assume the value assigned to the given branch. It is represented alongside of respective chance branch Payoff- positive (sales) or negative (cost) and associated with decision or chance branches

Operations Research

Steps In Decision Tree Analysis


Step 1: Identify the decision points and the alternative courses of action at each decision point systematically. Step 2: At each decision point determine the probability and the pay off associated with each course of action. Step 3: Commencing from the extreme right end, compute the expected pay offs (EMV) for each course of action. Step 4: Choose the course of action that yields the best pay off for each decisions. Step 5: Proceed backwards to the next stage of decision points. Step 6: Repeat above steps till the first decision point is reached. Step 7: Finally, identify the course of action to be adopted from the beginning to the end under different possible outcomes for the situation as a whole.

Operations Research

Decision Trees
Example: Raman Industries ltd. Has a new product which they expect a great potential. At the moment they have two course of action S1= test market the product and S2= Drop the product. Indicate the most preferred decisions.

Operations Research

Decision Trees
F
E G 2 A S2 D 3 C B H
-100,000

500,000

100,000

Operations Research

Decision Trees
Node E EMV(E) = 0.2x500,000+0.55x100,000+0.25x(-100,000) =Rs. 130,000 Selection of D = Rs 25,000 (certain payment) Node 2 Max. payment = 130,000 Node 3 = Rs 25,0000 (Certain payment) Node A EMV (A) = 0.7 x 130,000 + 0.3 x 25000 = Rs. 98,000 Value of Branch (1- A) = 98000 -50,000 = Rs 48,000 Value of Branch (1-B) = 25,000 (Certain payment)

Optimal Decision: Test market the product , if the result is positive , then market and if negative then drop.
Operations Research

Decision Trees
Example: A finance manager is considering drilling a well. In the past, only 70% of wells drilled were successful at 20 meters depth in that area. Moreover, on finding, no water at 20 meters, some persons in that area drilled it further up to 25 meters but only 20% stuck water at that level. The prevailing cost of drilling is Rs. 500 per meter. The finance manager estimated that in case he does not get water in his own well, he will have to pay Rs. 15000 to buy from outside for the same period of getting water from the well. The following decisions are considered: 1. Do not drill any well 2. Drill up to 20 meters 3. If no water is found at 20 meter, drill further up to 20 meters. Draw and appropriate decision tree and determine the optimal solution
Operations Research

Decision Trees
Water, 0.2

Decision Node 1 Decision Node 2


No Water Drill 20 m 0.3 Water 0.7 Buy Water Rs 15,000 Stop Drill Up to 25 m

500X25= Rs 12500 No Water Rs 27,500

Rs 25,000

Rs 10,000

Rs 15,000 + Rs 500 X 20 = Rs 25,000

Decision Tree for Example 13.19 (Vohra)


Operations Research

Analysis Table
Decision Node Options Expected Cost 0.8(15,000+1 2,500) + 0.212,500 = Rs 24,500 Rs 25,000 0.324,500+0.7 10,000 = Rs 14,350 Decision

Drill to 25 m

Stop 2 Drill to 20 m

Do not Drill

Rs 15,000

Optimal Decision: Drill up to 20 m, if no water, then drill up to 25 m

Operations Research

Advantages Of Decision Tree Approach


1. It structures the decision process and helps decision-making in an orderly, systematic and sequential manner. 2. It requires the decision-maker to examine all possible outcomes, whether desirable or undesir- able. 3. It communicates the decision-making process to others in an easy and clear manner, illustrating each assumption about the future. 4. It displays the logical relationship between the parts of a complex decision and identifies the time sequence in which various actions and subsequent events would occur. 5. It is especially useful in situations wherein the initial decision and its outcome affects the subsequent decisions. 6. It can be applied in various fields such as introduction of a new product, marketing, make or buy decisions, investment decisions and so on.

Operations Research

Limitations Of Decision Tree Approach


1. Decision tree diagrams become more complicated as the number of decision alternative increases and more variables are introduced. 2. It becomes highly complicated when inter-dependent alternatives and dependent variables are present in the problem. 3. It assumes that utility of money is linear with money. 4. It analyses the problem in terms of expected values and thus yields an average valued solution. 5. There is often inconsistency in assigning probabilities for different events.

Operations Research

Você também pode gostar