CHAPTER 15: DECISION ANALYSIS 15.2-1. Phillips Petroleum Company developed a decision analysis tool named DISCOVERY to evaluate available investment opportunities and decide on the participation levels, The need for a systematic decision analysis tool arose from the uncertainty associated with various alternatives, the lack of a consistent risk measure across the organization and the scarcity of capital resources. The notion of risk is incorporated in the model with the use of risk-averse exponential utility function. The objective is to maximize expected utility rather than expected retum. DISCOVERY provides a decision-tree display of available alternatives at various participation levels. A simple version of the problem is one where Phillips needs to decide first on the participation level and second on whether to drill or not. The exploration of petroleum when drilled is uncertain, The analysis is performed for different levels of risk-aversion and the sensitivity of the decisions to the risk-aversion level is observed. When additional seismic information is available at a cost, the value of information is computed. This study "has increased management's awareness of risk and risk tolerance, provided insight into the financial risks associated with its set of investment opportunities, and provided the company a formalized decision model for allocating scarce capital" [p. 55] The software package developed has been a valuable aid in decision making. It provided a systematic treatment of risk and uncertainty. Other petroleum exploration firms started to use DISCOVERY in analyzing decisions, too. 15.2.2, @ State of Nature ‘Alternative Sell 10, 000 | Sell 100, 000 Build Computers 0 54 Sell Rights 5 5 1) 60 50 | Bull computers 240 = Crossover Paint 5 a0 é : B20) sellRights a 10 0 a 02 04 08 o8 1 Prior Probabiity of Selling 10,000 15-1(©) Let p be the prior probability of selling 10,000 computers. Build: EP = p(0) + (1 — p)(54) = —54p +54 Sell: EP = p(15) + (1 ~ p)(15) = 15 The expected profit for Build and Sell is the same when —54p + 54 = 15 => p = 0.722, They should build when p < 0.722 and sell if p > 0.722. @) 55 60 45 40 5 Et 5 20 16 Expected Profit ($ million) a 02 04 08 08 1 Prior Probability of Selling 10,000 © 05 Sell 10,000 Build Computers 6 a 6 a as Sell 100,000 54 —aK 60 5h a \ cat Rights 18 15 15 Building computers should be chosen, since it has an expected payoff of $27 million.15.2-3. fa) State of Nature ‘Alternative ‘Sell 12 Cases | Sell 13 Cases | Sell 14 Cases | Sell 15 Cases Buy 12 Cases 132, 132, 132 132 Buy 13 Cases 125 as 143 143 Buy 14 Cases Ts 136 154 15d Buy 15 Cases iT 129) 147. 165, Prior Probability | _0.1 03 04 02 (b) According to the maximin payoff criterion, Jean should purchase 12 cases, State of Nature ‘Altemative Sell 12 Cases | Se Sell 14 Cases Min Buy 12 Cas 13, 132, 132 Buy 13 Cas 125, 125 Buy 14 Cases Ts 1s Buy 15 Cases TT HI Prior Probability | __0.1 (©) She will be able to sell 14 cases with highest probability and the maximum possible profit from selling 14 cases is earned when she buys 14 cases. Hence, according to the maximum likelihood criterion, Jean should purchase 14 cases. (d) According to Bayes' decision rule, Jean should purchase 14 cases. State of Nature Exp. ‘Altemative: Sel 12 Cases | Sell 13 Cases | Sell 14 Cases | Sell 15 Cases | Profit Buy 12 Cases 132 132 132 132 132 Buy 13 Cases 125 143 143 143 1412 Buy 14 Cases 118 136 154 154 145 Buy 15 Cases Ti 129 aT 165, 1416 Prior Probability | __0.1 03 04 02 (©) 0.2 and 0.5: Jean should purchase 14 cases. State of Nature Exp. ‘Altemative Sel 12 Cases | Sell 13 Cases | Sell 14 Cases | Sell 15 Cases | Profit Buy 12 Cases 132 132 132 132 Buy 13 Cas 125 13 13 1412 Buy 14 Cases 118 136 Td 146.8 Buy 15 Cases Ty 129 165, 143.4 Prior Probability | __0.1 02, 02, 0.4 and 0.3: Jean should purchase 14 cases. State of Nature Exp. Altemative Sel 12 Cases | Sell 13 Cases | Sell 14 Cases | Sell 15 Cases | Profit Buy 12 Cases 132 132 132 132 Buy 13 Cases 125 143 as TAL Buy 14 Cases Ts 154 Tod 143.2 Buy 15 Cases TH 1a 165 139.8 Prior Probability [| _0.1 03 02and 0.2: Jean should purchase 14 eases. State of Nature ‘Alternative Sell 12 Cases | Sell 13 Cases | Sell 14 Cases | Sell 15 Cases Buy 12 Cases 132, 132, 132 132 Buy 13 Cases 125 13 143 13 Buy 14 Cas Tis 136 154 Tot Buy 15 Cases Tr 129) a7 165, Prior Probability | __0.1 05 02 02 15.2-4, (a) The optimal (maximin) actions are conservative and countercyclical investments, both incur a loss of $10 million in the worst case. (b) The economy is most likely to be stable and the alternative with the highest profit in this state of nature is to make a speculative investment. According to the maximum likelihood criterion, Warren should choose speculative investment. (©) To maximize his expected payoff, Warren should make a countercyclical investment. State of Nature Exp. ‘Alternative Improving | Stable [ Worsening | Profit Conservative 30 5_ | =10 15 Speculative a0 10 30 3 Countereyelical 10 0 15 5 Prior Probability | _0.1 05 04 15.2.5, (a) Warren should make a countercyclical investment. State of Nature Exp. Alternative Improving | Stable | Worsening | Profit Conservative 30 5 | —10 =15 Speculative 40 10 30 ca [Countereyelieal_[~—10 0 15 8 Prior Probability | _0.1 03 06 (b) Warren should make a speculative investment. State of Nature Exp. ‘Alemative Improving | Stable | Worsening | Profit Conservative 30 5 10 45 Speculative 0) to | 30 5 Countereyelical_|_—10 0 15 2 Prior Probability | 0.1 a7 02(©) The expected profit from countercyclical and conservative investments is the same when p = 0.62. The expected profit lines for conservative and speculative investments cross at p= 0.68. Those for countercyclical and speculative investments cross at p ~ 0.65; however, this crossover point does not result in a decision shift. Expected Profit (Smillions) 12.5} Speculative Conservative Prior Probabilit 06 08 TO or Stable Economy -23, (d) Let p be the prior probability of stable economy. Conservative EP = (0.1)(80) + p(5) + (1 ~ 0.1 ~ p)(~-10) = 15p—6 Speculative EP = (0.1)(40) + p(10) + (1 — 0.1 — p)(—30) = 40p — 23 Countereyclical: EP = (0.1)(—10) + p(0) + (1 — 0.1 — p)(15) = —15p + 12.5 Countercyclical and conservative cross when —15p + 12.5 Conservative and speculative cross when 15p — 6 = 40p Sp—6 => p=0617. 23 > p = 0.68. Accordingly, Warren should choose countercyclical investment when p < 0.617, conservative investment when 0.617

0.68. (@) WAMONDOSINd Expected Profit O41 02 08 04 06 06 O07 0.8 09 Prior Probability of Stable Economy15.2-6. (a) Ag should be chosen. State of Nature ‘Alternative Si] S2 | Ss | Min Al 220 [170 [110 | 110 A 200 [180 | 150 | 150 Prior Probability [0.6 [0.3 [0.1 (b) The most likely state of nature is $ and the alternative with highest profit in this state is Ay (©) Ai should be chosen. State of Nature | Exp ‘Altemative S| Ss |S | Payoit AL 220 | 170 | 110 | 194 Ap 200 | 180 | 150 | 189 Prior Probability [0.6 [0.3 [0.1 @) Crossover point Expected Profit (Sthousands) 17 164) 0.2 04 06 08 1.0 Prior Probability of $, Let p be the prior probability of Sy Aj: EP = p(220) + (1 — 0.1 — p)(170) + (0.1)(110) = 50p + 164 Az: EP = p(200) + (1 — 0.1 — p)(180) + (0.1)(150) = 20p + 177 Ay and Az cross when 50p +164 = 20p + 177 = p = 0.433. They should choose Az when p < 0.433 and A; if p > 0.433.fe) Crossover Expected A Profit a (Sthousands) 159 128 02 04 06 08 10 Prior Probability of S; Let p be the prior probability of S Ar: EP = p(220) + (0.3)(170) + (1 — 0.3 — p)(110) = 110p + 128, Agi EP = p(200) + (0.3)(180) + (1 — 0.3 ~ p)(150) = 50p + 159 ‘A; and Az cross when 110p + 128 = 50p + 159 = p = 0.517. They should choose Ay when p < 0.517 and Ay ifp > 0.517. @ Expected Profit (Sthousands) 180 176 02 04 06 O08 LO Prior Probability of S, Let p be the prior probability of Si Au: EP = (0.6)(220) + p(170) + (1 ~ 0.6 ~ p)(110) = 60p + 176 Ax: EP = (0.6)(200) + p(180) + (1 ~ 0.6 ~ p)(150) = 30p + 180 ‘A; and Ap cross when 60p + 176 = 30p + 180 = p = 0.133. They should choose Az when p < 0.133 and Ay if p > 0.133. (2) Aj should be chosen.15.27. fa) State of Nature ‘Alternative Dry_| Moderate | Damp Crop 1 90 150__| 180 Grop2 T25{ 135 180 Crop 3 120 105 | 105 Crop 4 00 0 90. Prior Probability | 0.2 05 03 (b) Grow Crop 1 State of Nature Exp. ‘Alternative Dry_| Moderate | Damp | Payoff Crop I 90 150 | 180 | 147 Crop, T25 [135 | 180_| 144 Crop 3 120 105 | 105 | 108 Crop 4 90 90 90_| 90 Prior Probability | 0.2 05 03 (©) Prior probability of moderate weather = 0.2: Grow Crop 2. State of Nature Exp. ‘Alternative Dry_| Moderate | Damp | Payoit Crop 1 90 150 | 180 | 156 Crop 2, m25{ 135 | 180_| 157.5 Crop 3 120 105 [ 105__| 108 Crop 4 90) 90) ‘90 | 90 Prior Probability | 0.2 0.2 06 | Prior probability of mo 0.3: Grow Crop 1 or 2 State of Nature Exp. ‘Alternative Dry_| Moderate [Damp | Payoff Crop 1 90 150 | 180] 153 Crop2 25] 135 | 180 | 153 Crop 3 120 105 | 105_| 108 Crop 4 90 90 90_| 90 Prior Probability | 0.2 03 05 | Pi babil = 0.4: Grow Crop 1 Exp. ‘Alternative Dry_| Moderate [ Damp | Payoff Crop T 90 150 | 180 | 150 Crop T25 | 135__| 180 | 1485 Crop 3, 120 105 | 105__| 108 Crop 4 90 90 90] 90 Prior Probability [0.2 04 04Prior probability of moderate weather = 0.6: Grow Crop 1 State of Nature Exp. Alternative Dry_| Moderate [ Damp | Payoff Crop T 90) 150 | 180 144 Crop 2 125 | _135___| 180 | 139.5 Crop 120 105 | 105__| 108 Crop 4 90) 90. 90 | 90 Prior Probability | 0.2 06 02 15.2.8, The prior distribution is P{9 = 0} = 2/3, P{@ = 0} = 1/3. Order 15: EP = 2/3(1.155 - 107) + 1/3(1.414- 10") = 1.241- 107 Order 20: —EP = 2/3(1.012- 107) + 1/3(1.207 - 107) = 1.077 - 107 Order 25: EP = 2/3(1.047- 107) + 1/3(1.135 - 107) = 1.076 - 107 The maximum expected profit, or equivalently the minimum expected cost is that of ordering 25, so the optimal decision under Bayes’ decision rule is to order 25. 153-1. This article describes the use of decision analysis at the Workers' Compensation Board of British Columbia (WCB), which is "responsible for the occupational health and safety, rehabilitation, and compensation interests of British Columbia's workers and employers” [p. 15]. The focus of the study is on the short-term disability claims that can later turn into long-term disability claims and can be very costly for the WCB. First, logistic regression is employed to estimate the probability of conversion for each claim. Then using decision analysis, a threshold is determined to classify the claims as high- and low- risk claims. For any fixed conversion probability, the problem consists of a simple decision tree. First the WCB chooses between classifying the claim as high risk or low risk and then whether the claim converts or not determines the actual cost. If the claim is identified as a high-risk claim, the WCB intervenes. The early intervention lowers the costs and ensures faster rehabilitation. The expected total cost is computed for various cutoff points and the point with minimum expected cost is identified as the optimal threshold. The new policy offers accurate predictions of high-risk claims. As a result, future costs are reduced and injured workers start working sooner. This study is expected to save the WCB $4.7 per year. The scorecard system developed to implement the new policy improved the efficiency of claim management and the productivity of staff. Overall, the benefits accrued from this study paved the way for the WCB's adoption of operations research in other aspects of the organization.15,3-2. @ State of Nature ‘Alternative Sell 10,000 | Sell 100, 000 Build Computers 0 54 Sell Rights Te 1 Prior Probability 05 05 | Maximum Payoff 15 34 Expected Payoff with Perfect Information: 0.5(15) +0.5(34) = 34.5, Expected Payoff without Information: 0.5(0) + 0.5(54) = 27 EVPI = 34.5 — 27 = $7.5 million (b) Since the market research will cost $1 million, it might be worthwhile to perform it (©) Prior Conditional Joint Posterior Probabilities. Probabilities Probabilities Probabilities P (state) P(finding |state) P(state and finding) (state | finding) 0.667 ‘ell 10k and sell 10k sell 10k given sell 10k Sell 10k given sell 10k Il 100k given sell 10k sell 10k and sell 100k sell 10k given sell LOOk SEP 100k and sell 10K sell 100k given sell 10k Sell 10k given sell 100k 1 100k given sell 100k 0.667 sell 100k and sell 100k sell 100k given sell 100k 15-10@ Natuie’~| Probability [Sell (0 Bde 10H Sell 10,000, OS. O.B666567 : 0.333333. will 100,00¢ O5. 0.3333333 | 0.6566667 Data: Pinging [Sta), State of_[__ Pron Finding Probabi Finding | PlFindna) [ell TO SCa [Sell 100,000 Sell 10,000 05 Ob666667 | 0.5333353 well 100,00¢ O5 0.3993333 | O.6665667 Posterior P(State [ Finding) Siate of Nature (©) EVE = [0.5(1800) +0.5(3600)] — 2700 = 0, so performing the market research is not worthwhile. 15.3-3. (a) Choose A; with expected payoff $2, 500. State of Nature | Exp. Alternative Si_[S2 [Ss _| Payot Al 6 fifi [25 Ay 1 [3 [0 [is As 7 [1 [2 [22 Prior Probability [0.3 [0.4 [0.3 (b) State of Nature Alternative Si_[S [Ss Ai 6 [ii AD 1 [3 [0 As Tay Prior Probability [0.3 [0.4 | 0.3 Maximum Payoff [6 [3 | 2 Expected Payoff with Perfect Information: 0.3(6) + 0.4(3) + 0.3(2) = 3.6 Expected Payoff without Information: 2.5 EVPI = 3.6 — 2.5 = $1.1 thousand (©) Since the information will cost $1, 00 and the value is $1, 100, it might be worthwhile to spend the money. 15-1115.3-4. (a) Choose A; with expected payoff $35. State of Nature _| Exp. ‘Alternative Si_[S2 [Ss _| Payoit Ai 50 [100 [ —100 | 35 Ad oO 10 lo {1 As 2 | [a [aa Prior Probability | 0.5 | 0.3 | 0.2 (b) fate of Nature ‘Altemative S| Ss Al Too | —100 A 0 | 10] ~10 As 20 | 40 | —40 Prior Probability [0.5 | 0.3 | 0.2 Maximum Payoff | 50 [100 | —10 Expected Payoff with Perfect Information: 0.5(50) + 0.3(100) + 0.2(-10) = 53 Expected Payoff without Information: 35 EVPI = 53 — 35 = $18 (©) Betsy should consider spending up to $18 to obtain more information. 153.5. (a) Choose Ag with expected payoff $7, 800. State of Nature | Exp. Alternative Si_[S2 [Ss _| Payoff Ai 3 [25 [49 AD 3 [a0 [46 As 2 [ie [78 Prior Probability [ 0.3 [0.3 [0.4 (b) If S; occurs for certain, then choose Ay with expected payoff $4, 000. If Sy does not occur for certain, then the probability that S> will occur is 3/7 and the probability that Ss will oceur is 4/7. Ay: (3/7)(8) + (4/7)(25) = 15.57 Aa: (3/7)(5) + (4/7)(10) = 7.86 As: (3/7)(2) + (4/7)(15) = 9.43 Hence, choose A; which offers an expected payoff of $15, 570. Expected Payoff with Information: 0.3(4) + 0.7(15.57) = 12.01 Expected Payoff without Information: 7.8 EVI = 12.01 — 7.8 = $4.21 thousand ‘The maximum amount that should be paid for the information is $4,210. The decision with this information will be to choose Ag if the state of nature is S; and A; otherwise. 15-12(©) If Sz occurs for certain, then choose Az with expected payoff $5, 000. If $2 does not oceur for certain, then the probability that S; will occur is 3/7 and the probability that Ss will occur is 4/7. Ay: (3/7)(—20) + (4/7)(25) 5.71 Ar: (3/7)(—3) + (4/7)(0) = 4.43 As: (3/7)(4) + (4/7)(15) 10,29 Hence, choose Ay which offers an expected payoff of $10, 290. Expected Payoff with Information: 0.3(5) + 0.7(10.29) = 9.91 Expected Payoff without Information: 7.8 EVI = 9.91 — 7.8 = $2.11 thousand ‘The maximum amount that should be paid for the information is $2,110. The decision with this information will be to choose Ag if the state of nature is S2 and Ay otherwise. (d) IFS; occurs for certain, then choose A; with expected payoff $25, 000. If Ss does not occur for certain, then $; and S, occur with equal probability. Ay: (1/2)(—20) + (1/2)(3) Ag: (1/2)(—3) + (1/2)(5) As (1/2)(4)+(1/2)(2) = 3 Hence, choose Ay which offers an expected payoff of $3, 000, Expected Payoff with Information: 0.6(3) + 0.4(25) = 11.8 Expected Payoff without Information: 7.8 EVI = 11.8 — 7.8 = $4 thousand ‘The maximum amount that should be paid for the information is $4,000. The decision with this information will be to choose Ay if the state of nature is Sy and Ay otherwise. (©) Expected Payoff with Perfect Information: 0.3(4) + 0.3(5) + 0.4(25) = 12.7 Expected Payoff without Information: 7.8 EVPI = 12.7 — 7.8 = $4.9 thousand ‘The maximum amount that should be paid for the information is $4,900. The decision with this information will be to choose As if the state of nature is $i, Az if the state of nature is Sy and A; otherwise. (®) The maximum amount that should be paid for testing is $4,900, since any additional information cannot add more value than perfect information. 15-13(b) Joint Posterior Probabilities Probabilities P (state) P(finding | state) (state and finding) (state | finding) Oil and FSS 08 Oil, given FSS FSS, given Oil ISS, given Oil ° Oil, given USS Oil and USS Dry and FSS FSS, given Dry Dry, given FSS ISS, given Dry 0.6 Dry and USS Dry, given USS Data: Pinging | State) State of |_Prior ” “Finding Neture | Probability | FSS) USS oul 18 a8 02 iy ter ce ft Posterior, P(State [ Finding) Probabilities? State of Nature Finding | PFindngy |i by 7 FSS a5 rr uss ag areas (©) The optimal policy is to do a seismic survey, to drill if favorable seismic surroundings are obtained, and to sell if unfavorable surroundings are obtained. 15. 7. (a) Choose Ay with expected payoff $100. State of Nature | Exp ‘Alternative Si Si__| Payoff Al 400 [—100__| 100 A 0 | 100 | 60 Prior Probability [| 0.4 | 0.6 15-14(b) State of Nature Alternative St Sy AL 400 100 Ar 0 100 Prior Probability _| 0.4 06 | Maximum Payoff [400 [100 Expected Payoff with Perfect Information: 0.4(400) + 0.6(100) = 220 Expected Payoff without Information: 100 EVPI = 220 — 100 = $120, so it might be worthwhile to do the research. (©) Let X denote the state of nature and Y denote the prediction. From Bayes’ Rule, P(X = wand ¥ = y) = P(X = 2)P(Y = y|X = 2). @ P(X =S) and ¥ =) = (0.4)(0.6) = 0.24 (i) P(X = $; and ¥ = Sy) = (0.4)(0.4) = 0.16 (iii) PCX = S; and ¥ = $)) = (0.6)(0.2) = 0.12 (iv) P(X = Sz and ¥ = $;) = (0.6)(0.8) = 0.48 (@) P(S1) = 0.24 + 0.12 = 0.36, P(S2) = 0.16 + 0.48 = 0.64 (©) Bayes’ Rule: P(X = 2|Y = y) = PAGemtyen P(S;|S1) = 0.24/0.36 = 0.667 P(S;|S2) = 0.16/0.64 = 0.25 P(S2|S)) = 0.12/0.36 = 0.333 P(S2|S2) = 0.48/0.64 = 0.75 o PE nang | Stat9) State of |___ Prior - Finding Natue "| "Brobabilliy |" Si St a4 a i 82 a8 02 oe f Posteri Piste] Finding) Probabilities: State of Nature Finding | P(Finding) | t = SI 0% [0eee67 030333 52 cet 5 ore (g) IFS) is predicted, then choose A, with expected payoff $233.33, State of Nature | Exp. Alternative Si Sy_| Payot Al 400 [100 | 233.5 A 0 100 | 33.3 Prior Probability [0.667 [0.333 | 15-15(h) If Sa is predicted, then choose Az with expected payoff $75. State of Nature | Exp. ‘Alematve Si_ [Se _| Payoit Ai 400 —100 | 25 Ad 0 100 | 75 Prior Probability [0.25 | 0.75 (i) Given that the research is done, the expected payoff is (0.36)(233.33) + (0.64)(75) — 100 = $32 {) The optimal policy is to not do research and to choose At. 153-8, (a) EVPI = [(2/3)(-1.012 - 107) + (1/3)(-1.135 - 107)] — (—1,076 - 107) = $230, 000. (b) (30 spares required @=20)P(0=21) P@ = 21] 30 spares required) = perp spree rquted = IFOSTFSPLSU spares rqured P=TTSTD, 0132/3) 3-0 419 77 PCO = 24] 30 spares required) = 1 — 0.419 = 0.581 Order 15: EP Order 20: EP 1,305 + 107 1.125 107 19(—1.155 - 10") + 0.581(—1.414 - 10") 19(—1.012 - 107) + 0.581(—1.207 - 10”) Order 25: EP = 0.419(—1.047 - 107) + 0.581(—1.135 - 107) = -1.098 - 107 The optimal alternative is to order 25. 18. (a) State of Nature Alternative Poor Risk | Average Risk | Good Risk Extend Credit___[—15,000 | 10,000 20, 000 Not Extend Credit 0 0 a Prior Probability 02 05 03 (b) Choose to extend credit with expected payoff $8, 000. Siate of Nature Exp. Alternative Poor Risk | Average Risk | Good Risk | Payoff tend Credit =15,000 [10,000 20,000 | 8,000 | Not Extend Credit 0 0 a 0 Prior Probability 02 05. 03 15-16©) State of Nature Alternative Poor Risk | Average Risk | Good Risk Extend Credit 15,000 | 10,000 20,000 Not Extend Credit a 0 0 Prior Probability a2 as 03 ‘Maximum Payoff. 0 | _10,000 20, 000 Expected Payoff with Perfect Information: 0.2(0) + 0.3(10, 000) + 0.4(20, 000) = 11,000 Expected Payoff without Information: 8, 000 EVPI = 11,000 — 8,000 = $3,000 Hence, the eredit-rating organization should not be used. d) PF = Poor Finding AF = Average Finding GF = Good Finding PS = Poor State AS = Average State GS = Good State Prior Conditional Joint Posterior Probabilities Probabilities Probabilities Probabilities P (state) P(finding | state) P(state and finding) (state | finding) PF, given PS PS, given PF AF, given PS” pS and AF PS, given AF GF, given PS PS, given GF PF, given AS, AS. given PF © AF, given AS AS and AF AS, given AF GF, given AS ASandGF "AS, given GF PF, given GS, GS, given PF AF, given GS' GS and AF GS and GF GS. given AF GF, given GS GS, given GF 15-17Data: P(Einging | State), State of [__ Prior : Finding Nature “| Brobabilty’ |" Poor” YAv erage” “Goad Poor 02 75 04 ot ‘Avarege [08 ae eo Good a3 02° 04 os Posterior PiSlate | rinding) Probabilities: State of Nature Finding | P(Finding) | Poor |Averaye — Good Foor 0.38 0.27778 0.55556 0.16667 _ ‘Bvarage’ Ua PUT 7778" 1 Bbb86 0 26067 Good ais" [010836 025318 si 8e (f) Vincent should not get the credit rating and extend credit. 153-10. Prior | Conditional Joint Posterior Probabilities Probabilities Probabilities Probabilities P (state) P(finding I state) (state and finding) (state | finding) 0.1455 0.8509 0.97 Positive, given user egative, given user 0.0045 User and positive yser, given posi 015 user and negative“ yser, given negative ‘Honuser and positive nonuser, given positive Positive, given nonuser jegative given nonuser O97 nonuser and negative nonuser, given negative (a) Given that the testis positive, the athlete is a drug user with probability 0.8509. (b) Given that the test is positive, the athlete is not a drug user with probability 0.1491. (c) Given that the test is negative, the athlete is a drug user with probability 0.0054. (d) Given that the test is negative, the athlete is not a drug user with probability 0.9946, 15-18(e) The answers in Excel agree with those found in parts (a), (b), (€), and (4). Data: FEE nding | State) State of |_ Prior - Finding Neture | Probability | Positive [Newative Positive | _ .15 age | 003 Negaiwe [7 0C8 87 087 y Posterio PGiate | Finding) Probabilities: State of Nature Finding | P(Finding) [Posie Necative Positive 0.171. | OBS088, 0.14972 | Negatwe 0.829 | 000843 0.99487 State of Nature Alternative Successful | Unsuccessful Develop New Product 1, 500, 000 | —1, 800, 000 Not Develop New Product 0 0 Prior Probability 0.667 0.333 (b) Choose to develop new product with expected payoff $400, 000. State of Nature Exp. Altemative Successful | Unsuccessful | Payoff Develop New Product | 1,500,000 | —1,800,000_| 400, 000 Not Develop New Product 0 0 0 Prior Probability 0.667 O85 © State of Nature ‘Alternative Successful | Unsuccessful Develop New Product T, 500,000 | —1, 800,000 Not Develop New Product oO oO Prior Probability 0.667 333 Maximum Payoff T, 500, 000 0 Expected Payoff with Perfect Information: 0.667(1, 500, 000) + 0.333(0) = 1,000,000 Expected Payoff without Information: 400, 000 EVPI = 1, 000,000 — 400, 000 = $600, 000 This indicates that consideration should be given to conducting the market survey. 15-19@ Data: P(Fincing | State) State of Prior - Finding Nature” | Brebanilty’| Sicce ssful Uneuccesstul Successful | ep6ee6s7 | 08 02 i Unsuccessful| 033323333 |" 0.3 ar i Posteri P@State | Finding) Probabilities: State of Nature, Finding | P(Findina) | Site ful Unsuccessful T Successful 0.69339333 | 0.842103 | 0.1578947 4 Unsuccessful; 0.366 6666/7 | (03636364 | 0.63636354 ©) ‘ation Prediction __| Expected Payoit Develop product | Successful] [0.8421(1.5) ~ 0.1579(—1.8)] - 10 — $979,000 Not develop product | Successful | 0 Develop product | Unsuccessful | (0.3636(1.5) + 0.6364(—1.8)] 10° = — $600, 000 ‘Not develop product | Unsuccessful | 0 It is optimal to develop the product if itis predicted to be successful and to not develop otherwise. Let S be the event that the product is predicted to be successful. Then, P(S) = P(S|0,)P(0,) + P(S|92)P(O2) = 0.8(2/3) +.0.2(1/3) The expected payoff given the information is 0.6(979, 000) +.0.4(0) = $587, 000, so 6, EVE = 587,000 ~ 400, 000 = $187, 000 < $300, 000 = Cost of survey. Hence, the optimal strategy is to not conduct the market survey, and to market the product, State of Nature Alternative p= 0.05 | p= 0.25 Screen T,500_| =1, 500 Not Sereen =3, 750 Prior Probability | 0.8 0.2 (b) Choose to not screen with expected loss $1, 350. State of Nature | Exp. ‘Altemative P= 0.05 [p=0.25 | Payoit Screen —1,500 [—1,500 | —1,500 Not Screen’ 750 3,750 1,350 Prior Probability | __0.8 a2 15-20(o) State of Nature ‘Alternative p= 0.05 | p= 0.25 Screen T,500_[ 1,500 Not Sereen =750_|=3, 750 Prior Probability | 0.8 02 | Maximum Payoff 750 | —1,500 Expected Payoff with Perfect Information: 0.8(—750) + 0.2(—1, 500) = —900 Expected Payoff without Information: —1, 350 EVPI = —900 $450 This indicates that consideration should be given to inspecting the single item. @ Data: P(Fincing | State) State of Prior ” Fincing Nature "| Probability | Defective INondefecive: O08, 08 O08 me 7 O35 a2 ae a rf Posterior P(Siate | Finding) Probab State of Nature Finding | P(Finding 6 a6 Defective 0.09 | U444dedd (S5555656 Nondefecive 091) Dbsb Teas TO ieaesETe (©) P(defective) = (0.05)(0.8) + (0.25)(0;2) = 0.09 and P(nondefective) = 0.91 EVE = {(0.09)(—1500) + (0.91)(—1245)] — (—1350) = 82.05 Since the cost of the inspection is $125 > $82.05, inspecting the single item is not worth- while. (O If defective EP(screen, O|defective) = 0.444(—1500) + 0.556(—1500) = ~1500 EP(no screen, 6|defective) = 0.444(—750) + 0.556(—3750) = —2418 Ifnondefective: EP(screen, 6|defective) = —1500 EP(no screen, 6|defective) = 0.835(—750) + 0.165(—3750) = —1245 Hence, the optimal policy with experimentation is to screen if defective is found and not sereen if nondefective is found. On the other hand, from part (e), inspecting a single item, in other words experimenting is not worthwhile. Using part (b), the overall optimal policy is to not inspect the single items, to not screen each item in the lot, instead, rework each item that is ultimately found to be defective. 15-2115.3-13, (a) Say coin I tossed: EI Say coin 2 tossed: EP +0.4(-1) = -0.4 1) +0.4(0) = -0.6 The optimal alternative is to say coin 1 is tossed. (b) If the outcome is heads (H): Paton 1)Peoin 1) 0.310. P(coin 1/H) = pazeamDPeoin yr Meain DP CoM) — THUG) SO. P(coin 2\H) = 4 Saycoin}: EP = $(0) + $(-1) = —$ Say coin2: EP The optimal alternative is to say coin 2. If the outcome is tails (T): P(coin 1|T) = Pctjcoin 1)P(coin 1) 0.7241 P(coin 2\T) = 0.2759 Say coin]: EP = 0.7241(0) + 0.2759(—1) = —0.2759 Say coin2: EP = 0.7241(—1) + 0.2759(0) = -0.7241 The optimal alternative is to say coin 1 153-14, f@) State of Nature ‘Alternative in T | Coin 2 Predict OH 22.5 | 122.5 Predict 1H 105 | 105 Predict 2H 1225 | 225 Prior probabilities [0.5 | 0.5 Predict OH: EP = 0.5(22.5) + 0.5(122.5) = 72.5 Predict 1H: EP = 0.5(105) +0.5(105) 10! Predict 2H: EP = 0.5(122.5) +0.5(22.5) = 72. ‘The optimal alternative is to predict one heads with expected payoff $105. (b) Data P(Finging [ State) State oF Prior ” Fincing Nature” |"Brobapilty’ |” Heads ‘fais Coin 1 05 OF, 03 i Coin 2 i a3 ar f Posterior P@State [Finding) Probabilities: State of Nature Finding P(Findingy | Ceint Coin 2 Heads 05 o7 03 Tails 0 03 or 15-22(6) Ifthe outcome is heads (H): Predict 0H: EP = 0.7(22.5) + 0.3(122.5) = 52.5 Predict 1H: EP = 0.7(105) +0.3(105) = 105 Predict2H: EP = 0.7(122.5) + 0,3(22.5) 92.5 The optimal altemative is to predict one heads If the outcome is tails (T): PredictOH: EP 3(22.5) + 0.7(122.5) = 92.5 Predict 1H: EP = 0.3(105) +0.7(105) = 105 Predict 2H; EP = 0.3(122.5) + 0.7(22.5) = 52.5 The optimal alternative is to predict one heads. Since EP(H) = EP(T) = 105, the expected payoff is $105. (@) EVE = 105 — 105 = $0 < $30, so it is better to not pay for the experiment and choose to predict either one or two heads, 15.4-1, Driven by "the pressure to reduce costs and deliver high-impact technology quickly while justifying investments" [p. 57], Westinghouse initiated this study to evaluate R and D efforts effectively. At any point in time, the firm chooses between launching, delaying and abandoning an innovation, When the launch is delayed, there is a chance of losing the opportunity. R and D is hence treated as a call option with flexibility. The value of the innovation and the optimal decision rule in subsequent stages are found by using dynamic programming. This value is then used in the analysis of the decision tree constructed to find the present value of the project. In this tree, decisions consist of whether to fund or not at different stages and each decision node is followed by a chance node that represents either a technical milestone or strategic fit. Sensitivity analysis is performed to ensure robustness of the model, As a result of this study, explicit decision rules for funding R and D projects are obtained. Including flexibility in the model yields a more realistic model. The new system helps identifying cost-effective research portfolios with simplified data acquisition and easy implementation.15.4.2. Sell 10K 23 Build Computers w |e Sell 100K Predict 3 13 a5 [ Sell ink Sell Righs Market ee Research 26 Sell 0K 4 -——- 05 Build 5 Computers 35 35 Sell 100K Predict 3 Sell 100K Sell Rights a —___.@ 4 sell 10K 0 Os Build 27 Campers Sell 100K 2 at Ne Market 05 Research 27 Sell Rights | SelRights 15 sto build the computers without doing market research. on pe 15-2415.4.4, @ State of Nature ‘Alternative w L Hold Campaign 3 2 Not Hold Campaign [ 0 0 Prior Probability 0.6 O4 (b) Choose to hold the campaign with expected payoff $1 million. State of Nature | Exp. ‘Alternative Ww L_| Payoff Hold Campaign 3 [2 [1 ‘Not Hold Campaign [0 0 [0 Prior Probability [0.6 | 04 © State of Nature ‘Alternative Ww L Hold Campaign 3 [2 Not Hold Campaign | 0 0 Prior Probability [0.6 | 0 Maximum Payoff | 3 0 Expected Payoff with Perfect Information: 0.6(3) + 0.4(0) = Expected Payoff without Information: 1 EVPI = 1.8 ~1 = $0.8 million (d) Prior Conditional Joint Posterior Probabilities Probabilities Probabilities Probabilities P (state) P(finding Istate) P(state and finding) _P(state I finding) win and win 0.75 win. given win ‘win, given win Jose, given win win and lose "win, given lose lose and win lose, given win Win, given lose Jose, given lose 0.75 lose and lose lose, given lose 15.25Dat Pringing [State) State of,,|__ Prior Finding Natuie "| Brobabilfy |Win” toss Win 06 O76 0.25 Tose ie oe oe : Posterior F(State |r nding) Probabilities: State of Nature Finding | P(Finging) |Win Lege 7 Win O56 [Oeiet8 01318 loss tab" 0.33833 0.6667 (f Leland University should hire William, If he predicts a winning season, then they should hold the campaign and if he predicts a losing season, then they should not hold the campaign, ws Yaa fe yaa \ one al —a at — qd a0 |\ port tise comps \ Wn \ oom al area 15-2615.4.5, fa) 15.27| Zz. i = to . . fe (b) The comptroller should invest in stocks the first year. If growth occurs in the first year, then she should invest in stocks again the second year. If recession occurs in the first year, then she should invest in bonds the second year. 15-2815.46. The optimal policy is to wait until Wednesday to buy if the price is $9 on Tuesday. If the price is $10 or $11 on Tuesday, then buying on Tuesday is optimal. our 0 aa om » ry Brest toe 0 are a0 7 wat Bennet s 3 wa O\ 0 0s o wo 0 aur sim "a » 22 Bree iso tate 0 ‘0 0 22 vt Frnt si 3 eoO\ eo me tater si er ey <} 1100 Cio \aa / 7 Srna Mtoe es 1 ° se oe 7 "Hat san “a0 a0 15-2915.4.7. (b) Prior Distribution: a [he [% Po(k) | 0.2 [0.5 | 0.3 Qxnu(@) | & | & = Xi [0.5 [0.4 | 0.2 X, [04 [05 [04 Xs [01 [01 [04 Posterior Distribution: 0.267 0.632 (©) Itis optimal to not use credit rating, but to extend credit, see part (a).15.4.8, fa) (b) Prior Distribution: a] Polk) [ 0.667 | 0.333 Qxi-1(2) =z] o | & [08 [0.3 Xz [0.2 [0.7 Posterior Distribution: haix=e(h) zl | & Xr [0.842 | 0.158 X, [0.364 | 0.636 (©) It is optimal to not conduct a survey, but to market the new product, see part (a). 15.3115.4.9, =1500 -15 02 -750 -3F50 =| FFo “1750 “BFS “BERS wIF50 1950 -3F5 -38t5 (b) Prior Distribution: Oy | Oy Po(k) [0.8 | 0.2 Qxj-u(e) z[& | & X; | 0.95 | 0.75 Xp. | 0.05 | 0.25 Posterior Distribution: hoixma(h) zl a | & (Xr [0835 | 0.165 Xo | 0.4dd | 0.556 (©) Itis optimal to not test and to not sereen, see part (a) 15-3215.4-10, fa) (b) Prior Distribution: a | & Poth) [0.6 | 0.4 Qxjo-K(z) = [eo | & [03 | 0.6 x [07 [04 Posterior Distribution: hgix=z(h) z=] | & Xr | 0429 | 0.571 Ky [0.724 | 0.276 (c) It is optimal to choose coin 1 if the outcome is tails and coin 2 if the outcome is heads, see part (a).15.4-11, @ / \Wiseita sal “en = ve / ae 28 16 aay ra Sst Pecan util te Jo 200 fe a6 \ 026 / \ tice . ce et om a (b) Tatar Finding] State) State of Pring - Finding ature | Prob abiliy" "Pass Fai Successtul | 0.75 a8 at Unsuccessful] "0.25 a7 og : Pipiate [Finging) y State of Nature Finding | P(Finding) | Successful Unsuccessiul Pass, O7..[used2ee7 00a 743 Fat 03 O36 a (©) The optimal policy is to not pay for testing and to hire Matthew. (d) Even if the fee is zero, hiring Matthew without any further investigation is optimal, so Western Bank should not pay anything for the detailed report.Slate of Nature ‘Altemative Sel, 000 | Sell 100, 000 Build Computers| ___0 5a Sell Rights 15 15 os Sell 10.900 0 Build Computers /6 0 6 a \ os Sell 100,000 54 —aK 60 5h a \ cat Rights 18 15 15 They should build computers with an expected payoff of $27 million. (b) 55 60 45 40 5 0 5 20 15 0 02 04 08 08 1 Price Probability of Selling 10,000, —E é 3 e 5155-2. (a) The optimal policy is to not do market research and build the computers. The expected payoff is $27 million sercm aes |Past sea . ° Sansom . utecomues o* \ pc Maret Reseach / \ Ss 190m . (b) If the rights can be sold for $16.5 or $13.5 million, the optimal policy is still to build the computers with an expected payoff of $27 million. If the cost of setting up the assembly line is $5.4 million or $6.6 million, the optimal policy is still to build the computers with an expected payoff of $27.6 or $26.4 million respectively. If the difference between the selling price and the variable cost of each computer is $540 or $660, the optimal policy is still to build the computers with an expected payoff of $23.7 or $33.3 million respectively. For cach combination of financial data, the expected payoff is as shown in the following table Inall cases, the optimal policy is to build the computers without doing market research. Sell Righis_| Cost of Assembly Line | Selling Price — Variable Cost | Expected Payoff 313.5 million 35.4 million 3540 323.4 million $13.5 million $5.4 million, 3660 $30.9 million $13.5 million 36.6 million 3540 $23.1 million 313.5 million 36.6 million 3660 $29.7 million 316.5 million $5.4 million 3540 $243 million 316.5 million 35.4 million 3660 $30.9 million 316.5 million 36.6 million 3540 323.1 million 316.5 million 36.6 million 3660 329.7 million© Expected Payor! a 8 & Expected Parott 8 ee) Payoff from Selling Rights ams Expected Payot 5458 SB Assembly Line Cost a Ey 2 2” a 2 % m a c.o0ns4 ca0nge 0.00088 0.0008 o.000e2 0.co084 0.00068, Marginal Prost 15-37@ 2 a a a 2 B B Expected Payot B a zB 2 90.0% Marginal Pron Assernbly Line Cast Payoff from Selling Rigats, 2 94.0% ogo064 23 98.0% Input ue as % of Base Case 24 25 = Marana Pratt = Assembly Line Cost We Payoff from elling Rights} 102.0% 108.0% 110.0% 2% 27 2 2 30 St Expected Payofl 2155-4, 15.3915.5.5. our aa om » ry Brest toe are a0 7 wat Bennet 3 wa O\ 0 0s wo 0 aur "a » 22 Bree iso tate ‘0 0 22 vt Frnt 3 eoO\ eo me tater er | ew <} Cio \aa / 7 Srna Mtoe 0 stip \ 1100 “110 or 10% Hghor “1210 “100 310 “0m “1000 10 100 1100 120 The optimal policy is to wait until Wednesday to buy if the price is $9 on Tuesday. If the price is $10 or $11 on Tuesday, then buy on Tuesday. 15-41155-6. Data: P(Finding | State) Seale of Prior Finging Nature Brobabiliiy’ |” Excaliant Met Bvceliont SaisfactoryBox |__ 0.8 09 a1 i Unsatisfactory Box] 0.15, o4 i 06 Posterior : P(Siate Finding) Probabilities? Slate of Neture Finding P(Finding) | Satisfactory Bax (Uncai'cfactory Box Excellent 0805... 092027207... M0727 20273 Not Excelient’”"G.176 | O.aBe7 tazae DS 1aee 71a Salsas som ows NN er Eve / a fo aos =e Sa 4 = fo swe \\ asieee / \ tesa / amet dam ay _/ oo % pK oi The optimal policy is to sample the fruit and buy if it is excellent and reject if it is unsatisfactory. 15-42155-7. (a) Choose to introduce the new product with expected payoff of $12.5 million. State of Nature Exp. ‘Alternative Successful_| Unsuccessful | Payoit Tntroduce New Product $40 million | —SI5 million | $12.5 million Don't Introduce New Product [0 0 a Prior Probabilities 05 05 (b) With perfect information, Morton Ward should introduce the product if it will be successful and not introduce it if it will not be successful Expected Payoff with Perfect Information: 0.5(40) + 0.5(0) = 20 Expected Payoff without Information: 12.5 EVPI = 20 — 12.5 = $7.5 million (©) The optimal policy is to not test but to introduce the new product, with expected payoff $12.5 million. Dat P(finging | State) State of Prior Finding Nature “| Probability | “Aspiev'ad”” Wot Approved! Successful [05 a8 0.2, Unsuccessful [05 35 78 Posterior P@ate | Finding, Probabilities: State of Nature Finding Peinding) | “Successul” Unsuccesstul | Approved 0.825 | 0.761904762_0.238095238 Not Appreved ~“t.a76 [020526316 0: 789473684 The associated decision tree is on the next page. (d) If the net profit if successful is only $30 million, then the optimal policy is to test and to introduce the product only if the test market approves. The expected payoff is $8.125 million. If the net profit if successful is $50 million, then the optimal policy is to skip the test and to introduce the product, with an expected payoff of $17.5 million. If the net loss if unsuccessful is only $11.25 million, then the optimal policy is to skip the test and to introduce the product, with an expected payoff of $14.375 million. If the net loss if unsuccessful is $18.75 million, then the optimal policy is to conduct the test and to introduce the product only if the test market approves. The expected payoff is $11.656 million. For each combination of financial data, the expected payoff and the optimal policy are as shown below. Successful_[ Unsuccessful | Optimal Policy Expected Profit $30 million | -S11.25 million | Skip Test, Introduce Product $9.375_million $30 million | -S18.75 million | Test, Introduce Product if Approved | _$7.656 million $50 million | -SI1.25 million | Skip Test, Introduce Product $19,375 million $50 million | -S18.75 million | Test, Introduce if Approved 15.656 million(©) Expected Proft 18 7 16 18 14 13 2 " 10 20 s fo moore some Doetinbede roa 40 Net Profit Successful 15-44 4 \ overs \ recone aExpacted Prott " B 16 na 19 Net Loss if Unsuccesetu © » 0 ” 6 6 im 8 Expestec Prof Net Profit Gurvesstal Je Net Lose runsuccessna| 15% 56 95% 105% 115% 125% Input value as % of Base Case Net Profit If Successful Net Loce If Uneuscessful 1875 11.25 708 9 DM M 12 1 4 18 Bo WB 1g Expected Prot Both charts indicate that the expected profit is sensitive to both parameters, but is somewhat more sensitive to changes in the profit if successful than to changes in the loss if unsuccessful. 15-45155-8, Chelsea should run in the NH primary. If she does well, then she should run in the ST primaries. If she does poorly in the NH primary, then should not run the ST primaries, The expected payoff is $666, 667. = Pe” / : ome \ Dow Run ins oo 45 \ ean “ Veet gp sence st 48 unin / ees ae * (b) If the payoff for doing well in ST is only $12 million, Chelsea should not run in either NH or ST, with expected payoff of $0. If the payoff for doing well in ST is $20 million, Chelsea should not run in NH, but run in ST, with expected payoff of $2 million. If the loss for doing poorly in ST is $7.5 million, Chelsea should not run in NH, but run in ST, with expected payoff of $1.9 million. If the loss for doing poorly in ST is only $12.5 million, Chelsea should run in NH and run in ST if she does well in NH, with expected payoff of $166, 667. For each combination of financial data, the expected payoff and the optimal policy is as shown below. Wellin ST | Poorly in ST _| Optimal Policy Expected Funds ‘S12 million | -87.5 million | Run in ST Only $300,000 ‘S12 million | -S12.5 million | Don't Run in Fither 30 ‘$20 million | -S7.5_million | Run in ST Only 33.5 million’ ‘$20 million | -S12.5 million | Run in NH, Run in ST if Well | _$1.233 million 15-46(o) Expected Payot Exnected Payot? 18 18 14 12 08 06 04 02 18 16 14 12 08 06 04 02 2 4 16 18 a Gainif Win ST r 8 Q 10 1" 2 13 Lossif Lase ST 15-47@ Saini win ST ig [= Lossit Lose St Expected Payatt om = 70% MOH: TOTS 14H Input Value as % of Base Case GaninwinsT 12 20 Loss ifLose ST 125 75 “02 0 02 Of O86 o8 1 12 44 16 18 2 22 expactea Payer Both charts indicate that the expected payoff is sensitive to both parameters, although it is slightly more sensitive to changes in the profit if she does well than to changes in the loss if she does poorly. 15-4815.6-1. (a)- (b) The optimal policy is to not conduct a survey and to sell the land. fo onaoess\, a0 oor / eg, / “—— : = oa 7 au 15. (a) Choose to not buy insurance with expected payoff $249, 840. State of Nature Exp. ‘Altemative Earthquake [No Earthquake | Payoff Buy Insurance 249, 820 249,820__| 249,820 ‘Not Buy Insurance [ 90, 000 250,000 | 249, 840 Prior Probability 0.001 0.999 (b) — U(insurance) = U (250, 000 — 180) = \/249, 820 = 499.82 U (no insurance) = 0.999U (250, 000) + 0.001L(90, 000) = 499.8 The optimal policy is to buy insurance. 15.6-3. Expected utility of $19,000: U(19) B=5 Expected utility of investment: 0.3U (10) + 0.7U'(30) = 0.3/6 + 0.7,/36 Choose the investment to maximize expected utility. 15-4915.6-4, Expected utility of Ar = Expected utility of Az pU (10) + (1 ~ p)U(30) = U(19) 0.3U (10) +.0.7(20) = 16.7 + U(10) =9 15.6-5. (a) Expected utility of Ay = Expected utility of Az pU (10) + (1 pJU(O) = U1) 0.125U (10) + 0.875(0) = 1 = U(10) =8 (b) Expected utility of Ay = Expected utility of Ay pU (10) + (1 — p)U(0) = UGS) 0.5625(8) + 0.4375(0) = U(5) + U(5) =4.5 (©) Answers will vary. 15.6-6. (a) Expected utility of Ay = pU(36) + (1 — p)U(49) = 6p + 7(1—p) =7—p Expected utility of Ay = pU (144) + (1 — p)U(0) = 12p + 0 = 12p Expected utility of A; = pU(0) + (1 — p)U(81) =0 + 9(1 Expected Viility ‘A; and Ag cross when 7 — p = 12p > p= 7/13. A; and Aj cross when 7 — p = 9— 9p => p= 1/4. ‘Thus, As is best when p < 1/4, Ay is best when 1/4 < p < 7/13, and Ap is best when p> 7/13, 15-50)-U(M) = 500 = P 025 2566239 at 2508209 2600239 0 essa \ 075 anus sigaeas 323485 0.25 2 19928, 4719308 | 47.19928 sours | 0 aire \ 0.75 0 0 40.10507 40.1050? 40.10507 p= 50% 05 5.56235 a 2556299 2566239 a massa” \ 05 3.23465 31.2348 31.2344 os evtwe ey 15-57= ae ak ‘SE'S") Competition E[SalesiCempettion} ces engin _ come ae secu Seerg, gh atmeman encom, ge sia \ sae mente L — rs cata cot ~pASI"MY) Gempettion L. EfSalosiCompetitien} seca |_| eamrona on wisn / wa ma seca eninge ease, meow eS se a. ven acer atc aaa 5 €fBelesiCompetition) — ee Ne SW competion Ee EpsateniCompetitiony * a ae ee ai Since the expected value for the decision tree with additional (imperfect) information is less than that without information, Charlotte should not purchase the services of the market research company. 15-58CASE 15.2 Smart Steering Support (a) The available data are summarized in the table. ‘Smaz sweoe Spon. CosTs AMD REVENUES On 0 sens: paves, fama ansketendes en fetranaason en gue "ota "Sine amenacsl a oe ROMA: cstoneanner ans entme ane pense: uk Pote: (b) The basic decision tree is shown, Rectangular nodes represent decision forks and oval nodes represent chance forks. sas oeisON TREE, Pie Red Bat Mia 7 4 (d) The best course of action is to do the research project. The expected payoff is $2.4888 million. (e) The decision tree with perfect information on research is displayed. The expected value in this case equals $2.5488 million. The difference between the expected values with and without information is $60,000, which is the value of perfect information on research 15-59ge ae ra (f) The decision tree with perfect information on development is displayed. The expected value in this case equals $2.7618 million. The difference between the expected values with and without information is $273, 000, which is the value of perfect information on development (g) - (h) - (i) The decision tree with expected utilities is displayed. The expected utilities are calculated in the following way: for each of the outcome branches of the decision tree (c.g., profit of $6, 700, 000), the corresponding utility is computed (¢.g., 12.45992). Once this is done, the expected utilities are calculated. The best course of action is to not do research (expected utility of 10.14469 vs, 9.846267 in the case of doing research), (j) The expected utility for perfect information on research equals 9.939397, which is still less than the expected utility of not doing research (10.14469). Therefore, the best course of action is to not do research, implying a value of zero for perfect information on the outcome of the research effort. 15-60(k) The expected utility for perfect information on development equals 10.321347, which is more than the expected utility without information (10.14469). The value of perfect information on development is the difference between the inverses of these two utility values, U~1(10.321347) — U-1(10.14469) 0.93274 — 20 = 0.93274. The value of perfect information on the outcome of the development effort is $93.274. 15-61CASE 15.3 Who Wants to be a Millionaire (a) The course of action that maximizes the expected payoff is to answer $500,000 question alone. If you get the question correct, then use the phone-a-friend lifeline to help answer $1 million question. The expected payotf is $440,980, Praconect os Swien(thone at tnd) — s1s00m0 "$5004 (alone) 085 reo sone / mm sano eumeiprene a ikea) 18 XK ee fo Ver a oe em - ‘e000 [ona cm 00 es s200 am 2000 one secur ‘siamo Proves? end “omones #100800 Pe ee sill eet is sean son exon ‘som s2en000 (b) Answers will vary depending on your level of risk aversion. One possible solution is obtained by setting U (Maximum) = U($1 million) = 1 and U (Minimum) = U ($32,000) = 0. If getting $250,000 for sure is equivalent to a 60% chance of getting $1 million vs. a 40% chance of getting $32,000, then U' ($250,000) = p = 0.6, If getting $500,000 for sure is equivalent to a 90% chance of getting $1 million vs. a 10% chance of getting $32,000, then U ($500,000) = p = 0.9. (©) With the utilities derived in part (b), the decision changes to using the phone-a-friend lifeline to help answer the $500,000 question, and then walk away. 15-62Sinn (aon 08 os 1 milon(PhonesFrien?) O85 ceerect ‘nk Glen) OS — 4 yomk(Prone-aFilens) 05 nse aoe ° 1 s1poomn USt illon= 4 005 \ os ‘ugerck)= 08 Incorrect ugzenn= 08 ° ugea}= 0 ° 3200 Phone a Friend Dont nse e oa os eoo0 02 Ireorect ar) s200 Conroe ase it a 4 Phone'-frena ° a s1poomn 0085” \ 0 ose / \ Incorrect a7 ° oo sa2a0) \se ane _ i ee - nt oe Ireorect | oo smn Den'tarewer 5 os oe oo 15-63CASE 15.4 University Toys and the Business Professor Action Figures fa) Template for Posterior Probabi ies. Data: Frndng! State) Slate of Prior Finding, Nature Brohabilty |" Wellin Test Poor it Test Wel in Full Marker” 0.5 08 i 02 i Poor inF ull Maiket| 0.5 4 f Oe, f Poster F(Siate [ Finding). Probabilities: State of Nature — Finding P(Finding) | Wellin Full Market Poorin Full Market Wellin Test 05 OBBSBB6067, 0333333333 Poor in Test a4 aa 076 15-64(b) The best course of action is to skip the test market, and immediately market the product fully. The expected payoff is $1750. Sila anit / se a peptorame ae rl Foran 6a sae ce ee BR oe otter” Sane = Tie [= = | coma (el to \ fom, Fly tstet mm eo PP eect on oy papa (o) If the probability that the LSPAFs enter the market before the test marketing would be completed increases this would make the test market even less desirable, so it would still not be worthwhile to do. However, if the probability decreases, this would make the test market more desirable. It might reach the point where the test market is worthwhile. 15-65(d) Let p denote the probability that the LSPAFs will enter and EP the expected payoff. p_|_EP_| Test Market? 31,750 | No 0.0 [$1,906] Yes OT [S755 | Yes 0.2 | $1,750 | No 0.3 | $1,750 | No 04 [$1,750 | No 0.5 | $1,750 | No 0.6 [$1,750] No 07 [$1,750] No 0.8 [$1,750 | No [0.9 [$1,750 | No 1.0 | $1,750 | __No (e) It is better to perform the test market if the probability that the LSPAFs will enter the market is 10% or less. It is better to skip the test market if this probability is greater than 10%. 15-66CHAPTER 17: QUEUEING THEORY 172-1. A typical barber shop is a queueing system with input source being the population having hair, customers being the people who want haircut and servers being the barbers. The queue forms as customers wait for a barber to serve them. The customers are served usually with the first-come-first-served discipline. The service mechanism involves the barbers and equipment. 17.2-2. (a) Average number of customers in the shop, including those getting their haircut: 1 =0(%) +1() +2(g) +3(4) +4(4) =2 (&) [win queue [probability | product oO 0 1 0 20 3,1 O25 4,2 0.0625 Average number of customers waiting in the shop: Ly = 0.375 (©) Expected number of customers being served: 4 + 2(% ++ @ waka} 0 minutes x W,=%= = 0.094 hours = 5.625 minutes Hence, each customer will be in the shop for half an hour on the average. This includes the time to get a haircut. The average waiting time for a customer before getting a haircut is 5.625 minutes. (©) W — W, = 0.406 hours = 24.36 minutes 17.2-3. (a) A parking lot is a queueing system for providing parking. The customers are the cars and the servers are the parking spaces. The service time is the amount of time a car stays parked in a space and the queue capacity is zero. (b)— L = 0(0.1) +1(0.2) + 2(0.4) + 3(0.3) = 1.9 cars Ly = 0 cars 0.95 hours 7, ly 0 W, = 2 =§ =Ohours (©) A car spends an average of 57 minutes in a parking space. 17-117.2-4, (a) FALSE, The queue is where customers wait before being served. (b) FALSE. Queueing models conventionally assume infinite capacity (©) TRUE. The most common is first-come-first-served. 1725, (a) A bank is a queueing system with people as the customers and tellers as the servers. (>) Wy = 1 minute Ww 142 = 3 minutes Ly = AW, = 8(2) = 0.667 customers 2 customers 17.26. The utilization factor p represents the fraction of time that the server is busy. The server is busy except when there is nobody in the system. P) is the probability of having zero customers in the system, so p = 1 — Pp. 17.2-7. De = 2A fe = yn, La = 2, > HB = PR = 1 17.2-8, (a) {14 when nobody is in the system (Lg +1 otherwise > L = Polg + (1 — Po)(Lg +1) = Ly + (1 Ph) (b) L = AW = AW, + 1/) = AW, + /e= Ly +p La + (1— Po) => p= (1- Ph) (L=Lyt+p= 17.2.9, 1 = Sub, = SinPy+ SnP, = SnP, + Sol s)Py + SosPs nt ne nt as fee wt wd ray SnPy + Ly +s Py ‘Py + Ly +8(1— 3 Pa) 17.217.3-1. Part | Customers Servers (@)_| Customers waiting for checkout | Checkers (b)_| Fires Firefighting units (©_| Cars Toll collectors (@)_| Broken bicycles Bicycle repairpersons ‘(© _| Ships to be loaded or unloaded | Longshoremen & equipment (®_| Machines needing operator Operator ()_| Materials to be handled Handling equipment (h)_| Calls for plumbers Plumbers @_| Custom orders Customized process @_| Typing requests Typists 174.1. ‘9 1/2 forn=1 A= 1/2 orn > and jy = {7 frn>2 (a) P{next arrival before 1:00} = 1 — e~¥/? = 0.393 P{next arrival between 1:00 and 2:00} = (1 — e(/?)2)(1 — eV) = 0.239 P{next arrival after 2:00} = e~(V/?}? = 0.368 (b) Probability that the next arrival will occur between 1:00 and 2:00 given no arrivals between 12:00 and 1:00 is (1 —e ¥) = 0.393 (©) P{no arrivals between 1:00 and 2:00} = 2H — 1/2 — 0.607 Lajie™ P{one arrival between 1:00 and 2:00} 7 12 — 0.303 P {two or more arrivals between 1:00 and 2:00} = 1 (a) P{none served by 2:00} 0.368 P {none served by 1:10} 0.846 P {none served by 1:01} 101/60) — 0.983 174-2. An = 2 for n > 0 => P{n arrivals in an hour} = =~ (a) P{O arrivals in an hour} = #5* = 0.135 (b) — P{2 arrivals in an hour} = #5* = 0.270 (©) P{5 or more arrivals in an hour} = 1 ~ > P{(n arrivals in an hour} i — Be? — (4/3) e~? — (2/3)e-* = 0.527 173174-3. Expected pay: 100 - P{T < 2} +80- P{T > 2} = 100 — 20- P{T > 2} PA Tag > 2} = e-#? = 0.607 P{Tepeciat > 2} 22 = 0.368 Expected increase in pay: 20[/{Tig > 2} — P{Tepcciar > 2}] = 4.78 17.4-4, Given the memoryless property, the system becomes a two-server after the first completion occurs. Let T be the amount of time after t=1 until the next service completion occurs. P{T EW) = 20 minutes, var(W) = 0.0347 1741748. (a) FALSE, £(1 = Ta and var(T) = 1/02, p.775. (b) FALSE. "The exponential distribution clearly does not provide a close approximation to the service-time distribution for this type of situation,” second paragraph, p.776. (©) FALSE. A new arrival would have an expected waiting time, before entering service of 1/np, second last paragraph, p.777. 17.4.9, Let U = min{T),..., Ty} PIU =Ty} = fj°P{T, P= 2h 2) P= (6-3P)— 2%) /4= 4P @)> R= (6-}R-3-3R) =P 4) > i= (2% iP SS MH+ Fh + 5+ P+ pha = h=P= 2 @ = sp, = 2 P= BP = 5Po Po= BAP = Ph P= BP =P Po Sage Po = 3h Pot P+ P+Ps=1>R=Fh=$,.Rh=2,R=Ph=8 nP, =0- Py +1 P+2-Pyt3- P44 P= 2 Ly =S(n-1)Py =0- Py +1- Pp $2-Py+3- Y= B a 2-Pyt3 Prt 2 Pratl Ps 17717.5-4, @ 2 2 2 a Asp’ ots 2 @ ~ oO 22 OF (b) P= BR=P Pam hinPo = 3 P= Ben = (1) "FR, ian * S() Beware Rabe, a Sah (©) The mean arrival rate to the system and the mean service rate for each server when it is busy serving customers are both 2. 175+ @) 15, pF ©. 0_e8_@ 15 is IS (b) (1) 15) = 15; (2) 15P) + 15P; (3) 10P; + 15P, (4) 5P2 = 15P ) Pot Pit Prt Py =1 © Os=h=% (2) > Pr = (2/3)Pa (3) > Ps = (2/9)Po OShM=h=HhaPB=H The same equations can be obtained as follows: PL= ie Py = Phy Py = Py age @) L=O- P41: Prt 2-Pe+3- Py A= 15-Py tO P+ 5- P= = 981 2 W = & = % = 0.106 hours 178175-6. (a) Let the state represent the number of machines that are broken down. @) 45 C) 110, @) 8 18 (b) BP P+ P+ P= wae © X= 5° Pot Wy P= gp = 0.093 L=0-R4+1-P4+2-P = = 1072 Ly =0-Pi+1-P,= 2 = 0.330 W = 4 = 4 = 11.556 hours & ~ 3.556 hours @ Pot P= B= 0.742 (O} +h 0.464 (#) The birth-and-death process is a special case of continuous time Markov chains 175-7. (a) a x a a a a. ee era ee parse FO? ai ewe (b) WPL =APy APy + (H+ OP2 = (He + AYP AP uaa + (fe nO) Pay = (+ A+ (n= 1)8)P a 17-9175-8. @ (b) Py = © . = =89n(t)"=8.1. 4, =8 = SonP. = 15.n(3) bb giyss Ig=L-(1-P)=4 Wa f= BWR Rh 175-9, (a) Let the state represent the number of documents received, but not completed, (b) P, below corresponds to the steady-state probability that documents are received but not completed. () L, =S(n-)P, = net Wakes m=4 17-10P= BP = 2%, Py = AP) =2P, = fame LP) = Py SP =e Rh =1s By =e, By = 2e7* forn > 1 17.511, (a) () 5Py + 4Py = 9P) 5Pasi + 4Pn-2 = 9Pa (o) 1717.5-12, (a) Let n be the number of customers in the system, Balance equations: Py = P, Py = Py + Py,2Py = Pi + Py, 2Py = Pr, Py = Ps (b) Let the state (s, q) be the number of customers in service and in queue respectively. Balance equations: Pp = Pip + Pao 2Pio = Poo + Pu + Pas 2P = Pio 2Py = Prat Pro 2Pay = Poo, Par = Por 175-13, (a) Let the state (my, n2) be the number of type 1 and type 2 customers in the systems. 2 10 (b) Balance equations: 12Py = 5Poo 15P = 12(Pn + Po) 22Pi» = Py + 24P 29 24Px = 10Pio Poo + Pro + Por + Paw = 1 iy = ©. P, Pro = tre Po = yy = 2 Pin = (d) Type 1 customers are blocked when the system is in state (2,0) or (0,1), so the fraction of type 1 customers who cannot enter the system is Py + Py: = 55/187. Type 2 customers are blocked when the system is in state (2, 0), (0,1) or (1,0), so the fraction of type 2 arrivals that are blocked is Pay + Pig + Pyy = 115/187. 17-1217.6-1. KeyCorp deploys queueing theory as part of its Service Excellence Management System (SEMS) to improve productivity and service in its branches. The main objective of this study is to enhance customer satisfaction by reducing wait times without increasing the staffing costs. To do this, first a system that collects data about various phases of customer transactions is developed. Then, a preliminary analysis is conducted to determine the number of tellers required for at most 10% of customers to wait more than five minutes. The underlying model is an M/M/k queue with an average service time of 246 seconds. The arrival and service rates, and y. are estimated from the data. By using steady state equations, measures such as average queue length, average waiting time, and probability of having zero customers waiting are computed. The analysis revealed that with the current service time, the bank needed over 500 new employees. Hiring so many new tellers was too costly and physically impossible. Alternatively, the bank could achieve its goal by reducing the average service time. The investigation of the collected data helped to identify potential improvements in service. Accordingly, customer processing is reengineered, proficiency of tellers is improved and efficient schedules are obtained. Heuristic algorithms are incorporated in the model to make it more realistic. The model allowed KeyCorp to reduce the processing time by 53%. As a result of this, the customer wait time has decreased and the percentage of customers who wait more than five minutes is reduced to 4%. In addition to increased customer satisfaction, the new system resulted in the reduction of operating costs. Savings from personnel expenses is estimated to be $98 million over five years whereas the cost of the new system was only half a million dollars. The reports generated from the data are used in obtaining better schedules and identifying service components that are open to improvement. Efficient scheduling and reduced personnel released 15% of the capacity, which can now be used for more profitable investments. KeyCorp also gained more credibility by using a systematic approach in making decisions. KeyCorp management, customers, employees and shareholders all benefit from this study. 176-2. (a) M/M/1 queue with A= 2,0 = 4 p=1/2 = Py =1—p=1/2and P, = (1 p)p" = (1/2)""" 1 Proportion of the time the storage space will be adequate: }> P,, = 31/32 = 0.97 0) - Total = 0.9717.6-3. A= 30,2 = 50 > p= 0.6, Py = 17.6-4, p = 0.4 (proportion of time no one is waiting) (a) W~ Exp(— A), W () ie PIW > W} = (w= Ne (w-Afe —p ift<0 1— pel itt >0 PW, > W,} wy(W, 17.6-5. © the equalities Py = 1— 4 and W, = — Use the equalities Py = 1— 4 and Wy = zy. 17.6-6. The system without the storage restriction is an M/M/1 queue with A= 4 and x = 5, The proportion of the time that n square feet floor space is adequate for waiting jobs is SOMEP, Hence, the goal is to find n, such that [YAP > q for j= 1,2,3 and 4g; = 0.5, gp = 0.9, gs = 0.99. nytt gel nit ER rae Sea ae a-o( 4S Je aeo 2<1-g = a in(l-a) © (nj+2)Inp MEY —2, p=08 Part | g) eee — 2] Floor space required @ [05 1106 2 @&) [09 8319 9 © [99 [18.638 19 17.6-7. (a) TRUE. A customer does not wait before the service begins if and only if there is no one in the system, so the long-run probability that the customer does not wait is 1-P (b) FALSE. The expected number of customers in the system is L = p/(1— p), so it is not proportional to p. (©) FALSE. When p is increased from 0.9 to 0.99, L increases from 9 to 99. When it is increased from 0,99 to 0.999, L increases from 99 to 999. 17-14176-8. (a) FALSE, A temporary return to the state where no customers are present is possible (b) TRUE. Since \ > 1, the queue grows without bound, (©) TRUE. Since A < 2, the system can reach steady-state conditions. 17.6-9. (@) TRUE. "W ha (b) FALSE. "W, does not quite have an exponential distribution, because P{W, = 0} > 0," p.787 an exponential distribution with parameter u(1 ~ p)," p.787. (©) TRUE. "S42 represents the conditional waiting time given n customers already in the system. As discussed in Sec. 17.7, S,+1 is known to have an Erlang distribution,” p.787. 17.6-10. (a) = 2 customers, W = Wo = jaaeay = aogananj = ts hours, Ly = AW, = 20 Po=l-p pe pe There is a 29.6% chance of having more than 2 customers at the checkout stand, (6) Time in minutes: aay A Usaaaaa (sal aival Tate) 0.5 (mean service rate) "Results 1 (@eervers é 1.99389998 nt Pw?) = Oe if rssasaaas heh + F| 4] 0.202222909 1 i 3) 0148148149 Pow = BB slimeaees| wheni= 8 ibseranig SL o.osae95748 Time in hours baa 20 (ren atival at) 30 mean sevice rate) Rests 1 eervers 7 ee BW = nae a oaaaIsa ah a +} 0.222212200] 2} o. 14548149 0.06086607 Pra = NB | Sees ‘ ceseesecer| i) cvsseesen ‘when t= | 01083333 1 c.n4aea674a| 17-15