Decision Theory JAYANT SHARMAIntroduction ™ Decision theory or decision analysis is an analytical and systematic approach to decision making where the decision maker has several feasible and viable decision alternatives from which he or she has to select the best alternative on the basis of some standards decided in advance = The degree of certainty provides a foundation in developing decision models to arrive at the best possible decisions, The degree of certainty has two extreme points—complete certainty and complete uncertainty. The region under these two extreme points correspond to decision making under risk. Problems based on the phenomenon of decision making under risk are referred to as probabilistic problems.Elements of Decision Analysis ™ The state of nature is a future state of affairs that may result from the choice of an alternative from the list of available alternatives with the decision maker. In a decision problem, occurrences are chance occurrences, All the chance occurrences are governed by probabilities. " In probabilistic problems, it is assumed that duration is finite. After any combination of an act and an event, there is a final outcome. An outcome may be viewed in two ways: payoff (reward) or loss, A tabular arrangement of payoffs is referred to as payoff matrix. The values in the payoff matrix are conditional because of the uncertain state of natureCurd Pack Example ™ Suppose a leading dairy products company produces fresh curd packs weighing 200 grams every day. The expiry period for the curd pack is 24 hours. The demand (number of customers) for the curd pack is uncertain, ™ The decision maker has to take decisions about the number of curd packs to be produced every day because of the uncertain environment. In other words, he has to take an action. Action is in the control of a decision maker. In our example (decision problem), a decision maker can take any action (producing 100, 200, 300, or 400 curd packs). In this example, the number of customers are uncertain, as are possible events. ™ Suppose the cost of producing each curd pack is Rs 10 and it is sold for Rs 20. As one can understand very easily, Payoff = Selling price — Cost. The number of customers and the act and the payoff is exhibited in Table, which is known as a payoff table.Payoff table for the curd pack example ‘Mander of customers (even) —__—_Nanber.of urd pec preduved (Art) = O14) 10014) 20014) 30/4) 40014) og) or 1000-2000 3000 4000 100) 0 1000" 0 1000-2000 200) 0 1000 2000" = 1000 30012) 0 10002000 3000 2000 400) o 1000 2000 3000 4000" * represents the maximum payoff for the concemed act and event combination. apie Partial payoff table for the curd pack example when 100 curd packs are produced Number of customers (Events) Number of eurd packs produced (Act) 100 sees° (0-10 = 100 =— 1000) (100 = 20 — 10 « 100 = 1000) (100 x 20 = 10 = 100 = 1000) (100 = 20 - 10 « 100 = 1000) (100 x 20 = 10 = 100 = 1000)Opportunity Loss Table for the Curd Pack Example @ payoff table can also be constructed in the form of an opportunity loss table. This table is also referred to as regret table. The opportunity loss or regret can be defined as the amount of payoff not realized by not selecting the optimum gourse of action. Sa O_O 0-(—2000) 0 — 3000) ‘0— (4000) = 2000 = 3000 = 4000 1000-0 — 1000- (1000) 1000~ (-2000 = 1000 = 2000 = 3000 2000-2000 2000-1000 2000—0= 2000 =o = 1000 3000-— 2000 3000-3000=0 3000~2000 = 1000 = 1000 4000 - 2000 4000-3000 4000-4000 = 0 =a57. = ie Opportunity loss = The relative pay off which a decision maker could have realized - the Pay off which he has actually realizedDecision Making Under Uncertainty Asituation where the decision maker is unable to assess the probability of any state of nature is referred to as decision making under uncertainty. Laplace (Equally likely decision) criterion Maximin or minimax criterion Maximax criterion Hurwicz criterion Regret crieterionLaplace (Equally likely Decision) Criterion Laplace criterion is based on the simple principle that since probabilities of the state of nature are unknown, various events can be treated as equally likely. Under this assumption, the expected payoff for each act is computed first, followed by the mean of these expected payoff values. TABLE Mean (expected) payoff for the different acts in the curd pack example. Act Mean (expected) payoff (4) (+0+0+0+0/5=0 4.) (1000 + 1000 + 1000 + 1000 + 1000)/5 = 600 (4,) (2000 + 0 + 2000 + 2000 + 2000)/5 = 800 G4.) (—3000 — 1000 + 1000 + 3000 + 3000)/5 = 600 (4) (4000 — 2000 + 0 + 2000 + 4000)/5 = 0Maximin Criterion Maximin criterion is a conservative approach to decision making. The decision maker tries to avoid the worst choice, In this approach, the minimum payoff over the various events or possible states of nature is determined by the decision maker and an act is selected for which the Maximin payoff is the highest. In the curd pack example, minimum profits associated with various act are given as below: 4:0 pl) :-200- 4-100 4-10 According to the maximin criterion, act Al is selected, which generates the maximum profits when different acts are compared.Maximax Criterion ™ Maximax criterion is an optimistic approach where a decision maker determines the maximum payoff for each act and then an act is selected which provides the highest returns. ™ In the curd pack example, maximum profits associated with various acts are given as below: AsO 421000 42000 4,300 44000 ®™ Applying the maximax criterion, a decision maker will select the strategy A5 which gives the overall maximum of the different maximum payoffs generated from different acts.Hurwicz Criterion Hurwicz who coined the Hurwicz approach, has introduced a coefficient of optimism, generally denoted by pas Alpha varies on a scale ranging from 0 to i.In this scale-0, indicates an extremely pessimistic approach to the re and 1 indicates an extremely optimi approach to the future. Hence, alpha represents the coefficient of optimism and (1- alpha) represent the coefficient pessimism. Hurwicz criterion value= @ (Maximum value) + (1 —@) (Minimum value) TABLE Hurwicz criterion values for different acts (for the curd pack example) Act Maximo value Minimum value —Hurwicz criterion vaiues (4) 0 0 0.7 «040.3 x0=0 (4) 1000 —1000 0.7 « 1000 + 0.3 « (1000) = 400 (4) 2000 —2000 0.7 * 2000 + 0.3 * (-2000) = 800 (4) 3000 —3000 0.7 * 3000+ 0.3 x (-3000) = 1200 () 4000 -4000 0.7 « 4000 + 0.3 * (4000) = 1600Regret Criterion ™ In regret criterion, a decision maker selects the course of action that minimizes the maximum regret. ™ In the curd pack example, from the opportunity loss table maximum regret values can be selected as below: A.:4000 4,:3000 4,:2000 4,:3000 4.4000 =" The regret value is minimum for act A3. Hence, the decision maker will select act A3 by applying regret criterion.Example A company is faced with the problem of a decline in its sales turnover. To overcome this problem, it has decided to opt for any of the four strategies: heavy advertisement ($1); increase in number of sales executives (52); adding new features to products ($3), and increasing the price of the product ($4). Out of these four acts, there may be four possible states of nature or events which are a 40% increase in sales (£1); a 30% increase in sales (£2); a 25% increase in sales (£3); and a 22% increase in sales (£4). The company executives have worked out the yearly net profit (in thousand rupees) that would result if any of the four strategies are selected. ALPHA VALUE CAN BE TAKEN AS .6Table : Payoff matrix for Example State of nature GS) G) G) 6, &) 100 250 «850 © 500 a) 200 300 300 700 (E) 400 600 600 200 ) 600 800 350 500 On the basis of the five criteria for decision making under uncertainty, suggest which act should be adopted by the decision maker.Solution (i) Laplace (equally likely decision) criterion «ict Mean (expected) payoff (sp (100 + 200 + 400 + GO00)/4 = 325 (S$) (250 + 500 + 600 + 800)/4 = 537.5 (S) (850 + 300 + 600 + 350)/4 = 525. (S) (600 + 700 + 200 + 500)/4 = 475 ‘Considering Laplace criterion, a decision maker can select strategy S2. (ii) Maximin or minimax criterion S100 S$,:250 S,:300 5S, 200 According to the maximin criterion, strategy $3 is selected which generates the maximum minimum net profit among the different strategies.Solution (iii) Maximax or minimin criterion $600 5:800 5:850 5700 Applying the maximax criterion, a decision maker will select the strategy $3, which provides the maximum payoff among the various strategies. (iv) Hurwicz criterion In order to use the Hurwicz criterion, the decision maker selects alpha as 0.6, that is, (1 — alpha = 0.4), Act Minimum value Maximum value Hhawicz criterion values (8) 100 600 0.6 « 600 + 0.4 x 100=400 (8) 250 800 0.6 « 800 + 0.4 « 250=580 (8) 300 850 0.6 * 850 + 0.4 * 300 = 630 (S) 200 700 0.6 * 700 + 0.4 200 = 500Solution So, according to the Hurwicz criterion, a decision maker will select strategy 53 which gives the maximum Hurwicz criterion value of 630.Decision Making Under Risk Decision making under risk is a situation where more than one state of nature exists and the deci: maker has sufficient information to assign probability values to the likelihood of occurrence of each of these states. The three approaches a decision maker uses to evaluate various courses of action and select the best course of action are as follows: Expected monetary value (EMV) Expected opportunity loss (EOL) Expected value of perfect information (EVPI)Expected Monetary Value (EMV) Expected monetary value (EMV) is the sum of the payoffs for each course of action multiplied by the probabilities associated with each state of nature. From Table ‘expected monetary value (EMV) for the act 5. can be computed as ‘ EMV, = ¥ oN ‘where ks the number of possible states of nature, rthe numberof possible acts, x, the payoff associ= ated with ith act (S}) and jth state of nature (E), and p the probability of occurrence of the state of nature j.Example Suppose that in Previous Example, the probability of occurrence of various states of nature are also provided as indicated in Table below . State of Nature Probability ‘S, ) 015 100 «) 0.30 200 & 0.35 400 &) 0.20 600 On the basis of the expected monetary value (EMV) criterion, what decision should be taken by the decision maker? tS. 250 Acts (S, 850 300 600 350 tS, 500 700 200Expected monetary values (EMV) for selecting the best act are computed in Table Solution es ‘State of nanere Expected monetary value (EMV) @) @ @) @&) Probability als 030 O35 020 O15 « 100 + 0.30 = 200 + 035 « 400+ & 100 200400 0020 600335 O15 = 250 + 0.30 = 500 + 0.35 « 600+ % 20 00x 800-5575 O15 « 850 + 0.30 « 300 + 0.35 « 600+ & a a oe 30: 0.20 « 350 = 497.5 O.15 = 500 + 6.30 « 700 + 0.35 « 200+ % Soo 700204 500-455 From Table, it can be observed that the maximum expected monetary value is obtained for strategy (act) S2. Hence, a decision maker will select the strategy S2.Expected Opportunity Loss (EOL) Expected opportunity loss (EOL) criterion is another approach based on which a decision can be taken., the expected opportunity loss (EOL) can be computed as below: t FOL, = Vl, sl where isthe mabe of pestle state of uate, rhe number ofpssble acs, |. the opportu Toss ssvinted wth th at (5) and th st of mtr (E) and P the probity of occurence ofthe state of nature,Example Suppose in, the probability of occurrence of various states of nature are also provided as indicated in Table. State of nature Probability G) G) G6) (S) &) os 750 «600 «0350 (E) 030 500 200 «4000 () 035 200 0 0 400 &) 0.20 200 0 «= 450-300 On the basis of the expected opportunity loss (EOL) criterion, what decision should be taken by the decision maker?Solution ® For different acts (strategies), the expected opportunity loss (EOL) can be computed as below: For act 5,, expected opportunity loss (EOL) is [(0.15 « 750 + 0.30 = 500 + 0.35 x 200+ 0.20 = 200) = 372.5] For act S,, expected opportunity loss (EOL) is [(0.15 * 600 + 0.30 = 200 + 0.35 x 0+ 0.20 x 0) = 150] For act $,, expected opportunity loss (EOL) is [(0.15 « 0 +.0.30 = 400 + 0.35 x 0+ 0.20 « 450) =210] For act S,, expected opportunity loss (EOL) is [0.15 « 350 + 0.30 = 0 x 0.35 = 400 + 0.20 x 300) = 252.5] ™ Adecision maker will select an act (strategy) which will minimize the expected opportunity loss or expected regret. = It can be noticed that for act (strategy) S2, the expected regret value is minimum (150). Hence, on the basis of the expected opportunity loss (EOL) criterion, a decision maker will select strategy S2.Expected Value of Perfect Information (EVPI) Expected value of perfect information (EVPI) is referred to as the difference between the expected payoff with perfect information (EPPI) and the maximum expected payoff (EP) computed under uncertainty. Example : The payoffs with some probabilities associated with events (states of nature) for the curd pack example is exhibited in Table 19,18. Calculate the expected value of perfect information (EVPI). Curd pack example (payott) with some probabilities associated with events (state of nature) Number of curd packs produced (Act) 100(4,) —-200(4,) 300(4,) 4004.) Number of customers(Events) Probability 0) 0.10 100(E,) 0.15 200(E,) 020 300(E,) 025 400(E) 030 oa) 0 0 0 0 0 -2000 3000 4000 0 —1000 2000 2000 1000 0 2000 3000 2000 2000 3000 4000Computaton of expected payott with perfect information (EPPO for the curd pack example Number of Probabil- Number of curd packs produced (Act) ‘Max payoff (Prob customers ity — ||| = for each = Max (Events) Of4) 10064) 20014) 300(4) 40004) event pao Ol) 0.10 o -1000 2000-3000 4000 0 o WOE) = OS 0 1000 o =1000_ =2000 1000 130 200 (Z,) 020 oO 1000 2000 1000 o 2000 400 3O(E,) 025 o 1000 2000 3000 2000 3000 730 400(E) 9.30 0 1000 2000 3000 4000 4000 1200 So, the expected payoff with perfect information (EPPI) is the sum of all the elements in the last column of Table. Expected payoff with perfect information (EPPI) = 2500 Maximum expected payoff (EP) is the expected monetary value (EMV), which is already computed as 1400. sees soapactia valon 6 tock lec alti (ENT) = Expected payoff with information (EPPI) — Maximum expected payoff (EP) = 2500 — 1400 =1100Decision Trees ™ A decision tree can be referred to as a graphic model of a decision process. In other words, a decision tree is a graphic representation of a sequence strategy—nature of state combination available to a decision maker. ™ Example : A consumer durables company wants to diversify into other sectors of business. The company can choose to diversify into four different fields— the fast moving consumer goods sector, the consumer electronics sector, the concept selling sector, and the print media. The company has sought advice from a reputed consultancy firm. The advice received from the consultancy in terms of probability statements are as below:Example Fast moving consumer goods sector: Chances are 20% that the net profit of the company will decline by 10% in first three years; chances are 45% that the company will breakeven (no profit no loss) in the first three years, and chances are 35% that the net profit of the company will increase by 20%. Consumer electronics sector: Chances are 15% that the net profit of the company will decline by 15% in the first three years; chances are 55% that the company will breakeven (no profit no foss) in the first three ears, and chances are 30% that the net profit of the company will Increase by 15%. Concept sel : Chances are 25% that the net profit of the company will lecli y 20% in the first three years; chances are 35% that the company breakeven (no profit no loss) in the first three years, and chances are 40% that the net profit of the company will increase by 15%. Print media: Chances are 35% that the net profit of the company will decline by 25% in the first three years; chances are 35% that the company will breakeven (no profit no loss) in the first three years, and chances are 30% that net the profit of the company will increase by 20%. Construct a decision tree and using the expected value criterion, select the alternative with the highest expected payoff.Solution Decision tree Net pri dectines ty 1095Solution For 4 nodes the expected payoff can be computed as Node 1: 0.20 x 90 + 0.45 x 100 + 0.35 x 120 = 105 Node 2: 0.15 x 85 +0.55 x 100 + 0.30 x 115 = 102.25 Node 3: 0.25 x 80 + 0.35 x 100 + 0.40 x 115=83 Node 4: 0.35 x 75 + 0.35 x 100 + 0.30 x 120= 97.25Ex A company is engaged in the process of launching a new product. The top mgt of the company has 3 options in terms of launching the product in 3 sales zones: North region (N), South region (S) and East region (E). The management has decided to take the final decision on the basis of the demand for the product, which is divided into 3 categories: High demand, medium demand and low demand. On the basis of past demand and management's view, the respective probab. Of 3 kinds of demand are estimated to be .45, .35 and .20. The table on next page indicates the estimated profit (in thousand Rs.) for various combination of events and actsbeteta rtf) Poteet Sm Colne) PveT la East Utell) Region Carel (el) High 100 -60 -120 Medium 35 80 -20 40 Low 20 60 150 135 Construct a Decision tree and using the expected value criterion, select the alternatives with the highest expected pay off