Representing Uncertainty by Probability and Possibility ‐ What’s the Difference? Presentation at Palisade 2011 Risk Conference Amsterdam, March 29‐30, 2011 Hans Schjær‐Jacobsen Professor, Director RD&I Copenhagen University College of Engineering Ballerup, Denmark +45 4480 5030 hsj@ihk.dk www.ihk.dk Palisade 2011 Risk Conference Copenhagen University College of Engineering 1 Agenda 1. Why do we need uncertainty management? 2. Alternative representations of uncertainty 3. Some principles of New Budgeting 4. Introducing uncertainty in the cost model 5. Numerical examples 6. Resumé and perspectives Palisade 2011 Risk Conference Copenhagen University College of Engineering 2 1. Why do we need uncertainty management? Palisade 2011 Risk Conference Copenhagen University College of Engineering 3 Cost overruns and demand shortfalls in urban rail • Average cost escalation for urban rail projects is 45% in constant prices • For 25% of urban rail projects cost escalations are at least 60% • Actual ridership is on average 51% lower than forecast • For 25% of urban rail projects actual ridership is at least 68% lower than forecast (Flyvbjerg 2007) Palisade 2011 Risk Conference Copenhagen University College of Engineering 4 2. Alternative representations of uncertainty Palisade 2011 Risk Conference Copenhagen University College of Engineering 5 Two worlds of risk and uncertainty Uncertainty Imprecision Ignorance Lack of knowledge Statistical nature Randomness Variability Palisade 2011 Risk Conference World Representation and calculation Possibility Possibility distributions [a; …; b] Interval arithmetic Global optimisation Probability Probability distributions {µ; σ} Linear approximation Monte Carlo simulation Copenhagen University College of Engineering 6 Rectangular representation [a; b] and {µ; σ} Alternative interpretations Possibility distribution [a; b] Probability distribution {µ; σ} 1 h h = 1/(b-a) μ = (a+b)/2 σ2 = (b-a)2/12 0 Palisade 2011 Risk Conference a Copenhagen University College of Engineering b 0 7 Trapezoidal representation [a; c; d; b] and {µ; σ} Alternative interpretations Possibility distribution [a; c; b] Probability distribution {µ; σ} 1 α 0 Palisade 2011 Risk Conference h h = 2/(b-a+d-c) μ = h((b3-d3)/(b-d)-(c3-a3)/(c-a))/6 σ2 = (3(r+2s+t)4+6(r2+t2)(r+2s+t)2 -(r2-t2)2)/(12(r+2s+t))2 α-cut r a s c t d Copenhagen University College of Engineering b 0 8 Triangular representation [a; c; b] and {µ; σ} Alternative interpretations Possibility distribution [a; c; b] Probability distribution {µ; σ} 1 h h = 2/(b-a) μ = (a+b+c)/3 σ2 = (a2+b2+c2-ab-ac-bc)/18 α 0 Palisade 2011 Risk Conference α-cut a c Copenhagen University College of Engineering b 0 9 Independent stochastic variables Intervals Triple estimates {μ; σ} = {μ1; σ1} # {μ2; σ2} [a; b] = [a1; b1] # [a2; b2] [a; c; b] = [a1; c1; b1] # [a2; c2; b2] μ = μ1 + μ2; a = a1 + a2; σ2 = σ12 + σ22 b = b1 + b2 μ = μ1 - μ2; a = a1 - b2; σ2 = σ12 + σ22 b = b1 - a2 μ = μ1·μ2; a = min(a1a2, a1b2, b1a2, b1b2); σ2 ≅ σ12·μ22 + σ22·μ12 b = max(a1a2, a1b2, b1a2, b1b2) μ = μ1/μ2; a = min(a1/b2, a1/a2, b1/b2, b1/a2,); σ2 ≅ σ12/μ22 + σ22·μ12/μ24, b = max(a1/b2, a1/a2, b1/b2, b1/a2), if μ2 ≠ 0 if 0 ∉ [a2; b2] Addition Subtraction Multiplication Division a = a1 + a2; c = c1 + c2; b = b1 + b2 a = a1 - b2; c = c 1 - c2 ; b = b 1 - a2 a = min(a1a2, a1b2, b1a2, b1b2); c = c1c2; b = max(a1a2, a1b2, b1a2, b1b2) a = min(a1/b2, a1/a2, b1/b2, b1/a2,); c = c1/c2; b = max(a1/b2, a1/a2, b1/b2, b1/a2), if 0 ∉ [a2; b2] Table 1. Formulas for basic calculations with alternative representations of uncertain variables. Palisade 2011 Risk Conference Copenhagen University College of Engineering 10 Modelling by possibility distributions i.e. intervals, fuzzy intervals, etc. The actual economic problem is modelled by a function Y of n uncertain variables Y = Y(X1, X2,…, Xn). NB: Function can be arranged in different ways. In case of intervals Y is calculated by means of interval arithmetic (only applicable in the simple case) or global optimisation (applicable in the general case). In case of triple estimates Extreme values of Y are calculated as above. In case of fuzzy intervals As above, for all α‐cuts. Palisade 2011 Risk Conference Copenhagen University College of Engineering 11 Modelling by probability distributions The actual economic problem is modelled by a function Y of n independent uncertain variables Y = Y(X1, X2,…, Xn). Linear approximation Y is approximated by means of a Taylor series Y ≅ Y(μ1,…, μn) + ∂Y/∂X1∙(X1‐μ1) + ∂Y/∂X2∙(X2‐μ2) + … + ∂Y/∂Xn∙(Xn‐μn), where ∂Y/∂Xi is the partial derivative of Y with respect to Xi calculated at (μ1,…, μn). The expected value is given by E(Y) = μ = Y(μ1,…, μn). The variance is approximated by VAR(Y) = σ2 ≅ (∂Y/∂X1)2∙σ12 +…+ (∂Y/∂Xn)2∙σn2. Palisade 2011 Risk Conference Copenhagen University College of Engineering Monte Carlo simulation 12 Y = X(1-X), X = [0; 1] 30 25 pdf 20 Monte Carlo simulation X = RiskUniform(0; 1) Y = {0,167; 0,075} 15 10 Global optimisation X = [0; 1] Y = [0; 0,25] (Normalised as pdf) 5 0 0,00 0,05 0,10 0,15 0,20 0,25 Y = X(1-X) Palisade 2011 Risk Conference Copenhagen University College of Engineering 13 Sum of 10 identical rectangular cost elements [7; 15] 0,06 0,05 Independent variables Monte Carlo simulation N{110; 7,3} pdf 0,04 Fuzzy variables Fuzzy arithmetic [70; 150] (normalized as pdf) 0,03 0,02 0,01 0,00 60 70 80 90 100 110 120 130 140 150 160 Sum Palisade 2011 Risk Conference Copenhagen University College of Engineering 14 Sum of 10 identical trapezoidal cost elements [7; 9; 11; 15] 0,08 0,07 Independent variables Monte Carlo simulation N{106; 5,4} 0,06 pdf 0,05 Fuzzy variables Fuzzy arithmetic [70; 90; 110; 150] (normalized as pdf) 0,04 0,03 0,02 0,01 0,00 70 80 90 100 110 120 130 140 150 Sum Palisade 2011 Risk Conference Copenhagen University College of Engineering 15 Sum of 10 identical triangular cost elements [7; 10; 15] 0,08 0,07 Independent variables Monte Carlo simulation N{107; 5,2} 0,06 pdf 0,05 Fuzzy variables Fuzzy arithmetic [70; 100; 150] (normalized as pdf) 0,04 0,03 0,02 0,01 0,00 70 80 90 100 110 120 130 140 150 Sum Palisade 2011 Risk Conference Copenhagen University College of Engineering 16 Sum of 10 identical triangular cost elements [7; 10; 15] Palisade 2011 Risk Conference Copenhagen University College of Engineering 17 Sum of 10 identical triangular cost elements [7; 10; 15] Palisade 2011 Risk Conference Copenhagen University College of Engineering 18 Sum of 10 identical triangular cost elements [7; 10; 15] Palisade 2011 Risk Conference Copenhagen University College of Engineering 19 Sum of 10 identical triangular cost elements [7; 10; 15] Palisade 2011 Risk Conference Copenhagen University College of Engineering 20 Sum of 10 identical triangular cost elements [7; 10; 15] Palisade 2011 Risk Conference Copenhagen University College of Engineering 21 Y = ∑Xi , Xi = [8; 10; 16], i = 1, …,10 0,08 0,07 Monte Carlo simulation X = RiskTriangular(8; 10; 16) (Uncorrelated variables) Y = {113,3; 5,35} 0,06 pdf 0,05 Fuzzy arithmetic X = [8; 10; 16] Y = [80; 100; 160] (Normalised as pdf) 0,04 Monte Carlo simulation X = RiskTriangular(8; 10; 16) (100% correlated variables) Y = {113,3; 17,00} 0,03 0,02 0,01 0,00 80 90 100 110 120 130 140 150 160 Y = sum of Xi, i = 1,…,10 Palisade 2011 Risk Conference Copenhagen University College of Engineering 22 n Yn = ∑ Xi, Xi = [90/n;100/n;140/n] cdf i =1 n=1 n = 15 n=5 Palisade 2011 Risk Conference n = 10 Copenhagen University College of Engineering 23 Some observations of probability vs. possibility • With numerically identical input variables probability results in less numerical output uncertainty than does possibility • Uniform probability representation is different from interval possibility representation • Probability uncertainty decreases with increasing analytical complexity whereas possibility uncertainty is independent • Possibility uncertainty corresponds to fully correlated input probability variables • Monte Carlo simulation does not generally produce possibility results Palisade 2011 Risk Conference Copenhagen University College of Engineering 24 3. Some principles of New Budgeting Palisade 2011 Risk Conference Copenhagen University College of Engineering 25 A Danish governmental initiative • ”Best realistic budget based on available knowledge” • Budget control is done by standardised budgets and logging of follow‐up results • Risk and uncertainty management is conducted during entire project • Estimates of unit prices, quantities and particular risks • Experience based supplementary budget of one third of 50% of rough budget is allocated • Likelihood of event multiplied by impact is not accepted • Acceptable to incur additional cost to reduce risk and uncertainty Palisade 2011 Risk Conference Copenhagen University College of Engineering 26 The Anchor Budget • Project with a number of activities A • Each activity: unit price p and quantity q • Total cost C of activities at time t = 0 • Subsequently, additional activities and costs may be introduced Palisade 2011 Risk Conference Copenhagen University College of Engineering 27 The Event Impact Matrix • Risk events E are identified at any time t = τ • Additional activities may be initiated • Impacts of Risk Events on all p and q are estimated • We keep track of accumulated cost impacts for all individual risk events • Impacts from interacting (co‐acting) Risk Events are pooled Palisade 2011 Risk Conference Copenhagen University College of Engineering 28 The Risk Budget • All identified Risk Events are assumed to occur • Resulting p, q and cost for each activity is calculated • Total cost for project is calculated • Deviations from Anchor Budget is calculated Palisade 2011 Risk Conference Copenhagen University College of Engineering 29 Risk events modify Anchor Budget For the i’th activity Ai, we get the modified estimated cost Ciτ at time τ Ciτ = = (pi + ∆pi1 + ∆pi2 + … + ∆pij + … + ∆pim) · (qi + ∆qi1 + ∆qi2 + … + ∆qij + … + ∆qim) = Ci + pi · (∆qi1 + ∆qi2 + … + ∆qij + … + ∆qim) + (∆pi1 + ∆pi2 + … + ∆pij + … + ∆pim) · qi + ∆pi1 · (∆qi1 + ∆qi2 + … + ∆qij + … + ∆qim) + ∆pi2 · (∆qi1 + ∆qi2 + … + ∆qij + … + ∆qim) +… + ∆pij · (∆qi1 + ∆qi2 + … + ∆qij + … + ∆qim) +… + ∆pim · (∆qi1 + ∆qi2 + … + ∆qij + … + ∆qim) Palisade 2011 Risk Conference Copenhagen University College of Engineering 30 Convenient set‐up for calculations Event Impact Matrix at t = τ Anchor Budget at t = 0 Activity p q Cost E1 ∆p E2 ∆q ∆Cost ∆p E3 ∆q ∆Cost ∆p12 ∆q12 ∆p22 ∆q22 ∆p32 ∆q32 ∆C12τ ∆C22τ ∆C32τ ∆p Interaction ∆Cost ∆q ∆Cost ∆p13 ∆q13 ∆c1 ∆p23 ∆q23 ∆p33 ∆q33 ∆C13τ ∆C23τ ∆C33τ Risk Budget at t = τ Sum ∆Cost p p1τ p2τ p3τ q1 ∆c3τ ∆C1τ ∆C2τ ∆C3τ Cost A1 p1 q1 C1 ∆p11 ∆q11 A2 p2 q2 C2 ∆p21 ∆q21 A3 p3 q3 C3 ∆p31 ∆q31 A4 p4 q4 C4 ∆p41 ∆q41 ∆C41τ ∆p42 ∆q42 ∆C42τ ∆p43 ∆q43 ∆C43τ ∆c4τ ∆C4τ p4τ q4τ C4τ A5 p5 q5 C5 ∆p51 ∆q51 ∆C51τ ∆p52 ∆q52 ∆C52τ ∆p53 ∆q53 ∆C53τ ∆c5τ ∆C5τ p5τ q5τ C5τ ∆C·3τ ∆cτ ∆Cτ Sum C ∆C·1τ ∆C·2τ τ q ∆C11τ ∆C21τ ∆C31τ ∆c2τ τ C1τ q2τ C2τ q3τ C3τ Cτ Table 1. Convenient set-up for calculations, n = 5, m = 3. Anchor Budget Palisade 2011 Risk Conference Event Impact Matrix Copenhagen University College of Engineering Risk Budget 31 Event Impact Matrix at t = τ Anchor Budget at t = 0 Activity p q Cost A1 100 1,000 100,000 A2 50 10,000 500,000 A3 200 500 100,000 A4 1,000 150 150,000 E1 E2 ∆p ∆q ∆Cost 20 100 32,000 ∆p E3 ∆q ∆Cost 25 2,500 30 ∆Cost ∆Cost p 500 35,000 120 1,125 135,000 –10,000 50 9,800 490,000 15,000 230 15,000 10 10,000 150 850,000 ∆q –200 –10,000 A5 Sum ∆p 47,000 300 12,500 45,000 35,000 Risk Budget at t = τ Interaction ∆Cost 500 Sum q Cost 500 115,000 10,000 1,000 160 160,000 45,000 300 45,000 95,000 150 945,000 Table 2. Numerical test example without event and impact uncertainty. A1 E1: ∆Cost = (p+∆p)·(q+∆q) - Cost = p·∆q + ∆p·q + ∆p·∆q = 100·100 + 20·1,000 + 20·100 = 32,000 A1 E2: ∆Cost = (p+∆p)·(q+∆q) - Cost = p·∆q + ∆p·q + ∆p·∆q = 100·25 + 0·1,000 + 0·25 = 2,500 A1 Interaction: ∆Cost = 20 · 25 = 500 Palisade 2011 Risk Conference Copenhagen University College of Engineering 32 4. Introducing uncertainty in the cost model Palisade 2011 Risk Conference Copenhagen University College of Engineering 33 Uncertain impacts and likelihoods • Uncertain impact of Risk Events ‐ Uncertainty for all p and q from Anchor Budget ‐ Additional activities may become necessary ‐ How to estimate and represent uncertainty? ‐ Calculate uncertain Risk Budget • Likelihood of Risk Events occurring ‐ Estimate probabilities of Risk Events occurring ‐ Construct probability distribution of total project cost Palisade 2011 Risk Conference Copenhagen University College of Engineering 34 Triangular representation [a; c; b] Uncertain impact on unit price p = 100: ∆p = [18; 20; 25] Uncertain impact on quantity q = 1000: ∆q = [95; 100; 125] Uncertain impact on cost p∙q ∆Cost = (p+∆p)∙(q+∆q) − p∙q = p∙∆q + ∆p∙q + ∆p∙∆q ∆Cost = [9,500; 10,000; 12,500] + [18,000; 20,000; 25,000] + [1,710; 2,000; 3,125] = [29,210; 32,000; 40,625] Palisade 2011 Risk Conference Copenhagen University College of Engineering 35 Triangular representation {µ; σ} Uncertain impact on unit price p = 100: ∆p = {21.0; 1.47} Uncertain impact on quantity q = 1000: ∆q = {106.7; 6.56} Uncertain impact on cost p∙q ∆Cost = (p+∆p)∙(q+∆q) − p∙q = p∙∆q + ∆p∙q + ∆p∙∆q ∆Cost = {33,907; 1,802} (by Monte Carlo simulation) min = 29,551, max = 39,966 Palisade 2011 Risk Conference Copenhagen University College of Engineering 36 5. Numerical examples Palisade 2011 Risk Conference Copenhagen University College of Engineering 37 Event Impact Matrix at t = τ Activity A1 E1 E2 ∆p ∆q ∆Cost [18; 25] [95; 125] [29,210; 40,625] ∆p E3 ∆q ∆Cost [21; 31] [2,100; 3,100] ∆p A2 A3 ∆q ∆Cost [-210; -175] [-10,500; -8,750] [25; 45] [12,500; 22,500] A4 [7,000; 16,000] A5 ∆Cost ∆Cost [378; 775] [31,688; 44,500] [-10,500, -8,750] [7,000; 16,000] [145; 165] [41,710; 63,125] Sum [12,500; 22,500] [7; 16] Sum Interaction [9,100; 19,100] [280; 350] [40,600; 57,750] [40,600; 57,750] [30,100; 49,000] [378; 775] [81,288; 132,000] Table 3. Event Impact Matrix using interval uncertainty representation [a; b]. (Anchor Budget of Table 2). Palisade 2011 Risk Conference Copenhagen University College of Engineering 38 Risk Budget at t = τ Activity p q Cost A1 [118; 125] [1,116; 1,156] [131,688; 144,500] A2 50 [9,790; 9,825] [489,500; 491,250] A3 [225; 245] 500 [112,500; 122,500] A4 1,000 [157; 166] [157,000; 166,000] A5 [145; 165] [280; 350] [40,600; 57,750] Sum [931,288; 982,000] Table 4. Risk Budget using interval uncertainty representation [a; b]. (Anchor Budget of Table 2 and Event Impact Matrix of Table 3). Palisade 2011 Risk Conference Copenhagen University College of Engineering 39 Risk Budget at t = τ Activity p q Cost A1 [118; 120; 125] [1,116; 1,125; 1,156] [131,688; 135,000; 144,500] A2 50 [9,790; 9,800; 9,825] [489,500; 490,000; 491,250] A3 [225; 230; 245] 500 [112,500; 115,000; 122;500] A4 1,000 [157; 160; 166] [157,000; 160,000; 166,000] A5 [145; 150; 165] [280; 200; 350] [40,600; 54,000; 57,750] Sum [931,288; 945,000; 982,000] Table 5. Risk Budget using triple estimate uncertainty representation [a; c; b]. (Combination of Table 2 and 4). Palisade 2011 Risk Conference Copenhagen University College of Engineering 40 Event Impact Matrix at t = τ Activity E1 ∆p A1 ∆q E2 ∆Cost {21.0; 1.47} {106.7; 6.56} {33,907; 1,802} ∆p E3 ∆q ∆Cost {25.7; 2.05} {2,567; 205} ∆p A2 A3 ∆q {195; 7.36} {33.3; 4.25} ∆Cost {-9,750; 368} {16,667; 2,125} A4 A5 ∆Cost ∆Cost {539; 57.9} {37,012; 1,851} {-9,750; 368} {11,000; 1,871} {153.3; 4.25} {310; 14.7} {47,534; 2,619} {50,573; 2,807} Sum {16,667; 2,125} {11.0; 1.87} {11,000; 1,871} Sum Interaction {13,567; 1,884} {47,534; 2,619} {37,784; 2,643} {539; 57.9} {102,462; 4,305} Table 6. Event Impact Matrix using triangular probability input distributions {µ; σ}. (Anchor Budget of Table 2). Palisade 2011 Risk Conference Copenhagen University College of Engineering 41 Risk Budget at t = τ Activity p q Cost A1 {121.0; 1.47} {1,132; 6.87} {137,012; 1,851} A2 50 {9,805; 7.36} {490,250; 368} A3 {233.3; 4.25} 500 {116,667; 2,125} A4 1,000 {161.0; 1.87} {161,000; 1,871} A5 {153.3; 4.25} {310.0; 14.7} {47,534; 2,619} Sum {952,462; 4,305} Table 7. Risk Budget using triangular probability input distributions {µ; σ}. (Anchor Budget of Table 2 and Event Impact Matrix of Table 6). Palisade 2011 Risk Conference Copenhagen University College of Engineering 42 Probabilities of combinations pr1=0.6 pr1=0.3 pr3=0.2 pdf cdf E1 E2 E3 Cost Cτ Cost Cτ [a; c; b] {µ; σ} 850,000 850,000 no no no 0.224 0.224 no yes no 0.096 0.320 [859,100; 862,500; 869,100] {863,567; 1,882} no no yes 0.056 0,376 [880,100; 885,000; 899,000] {887,783; 2,642} yes no no 0.336 0.712 [891,710; 897,000; 913,125] {900,573; 2,810} no yes yes 0.024 0.736 [889,200; 897,500; 918,100] {901,350; 3,251} yes yes no 0.144 0.880 [901,188; 910,000; 933,000] {914,679; 3,380} yes no yes 0.084 0.964 [921,810; 932,000; 962,125] {938,356; 3,853} yes yes yes 0.036 1.000 [931,288; 945,000; 982,000] {952,462; 4,305} Table 8. Distributions of Cτ for representations by triple estimates and probabilities. Palisade 2011 Risk Conference Copenhagen University College of Engineering 43 Distribution of cost Cτ (triple estimates, Table 8) 1.0 E1, E2, E3 E1, E3 E1, E2 cdf E2, E3 E1 E2 0.0 800,000 Palisade 2011 Risk Conference E3 900,000 Copenhagen University College of Engineering Cost 1,000,000 44 Triangular probability Monte Carlo simulation of Risk Budget. (Anchor Budget of Table 2, Event Impact Matrix of Table 6). Uncorrelated input parameters Uncorrelated input parameters Most poss.: 945,000 100% correlated input parameters Max: 982,000 Min: 931,288 Palisade 2011 Risk Conference Copenhagen University College of Engineering 45 6. Resumé and perspectives Palisade 2011 Risk Conference Copenhagen University College of Engineering 46 Resumé and perspectives • Anchor Budget, unit prices, quantities • Impacts of individual Risk Events • Impacts of co‐acting Risk Events • Impacts on individual Activities • Uncertainty by probabilities and possibilities • Under‐ or overestimating uncertain impacts • Research into practical applications End of presentation Palisade 2011 Risk Conference Copenhagen University College of Engineering 47 Back to Table 6 Palisade 2011 Risk Conference Copenhagen University College of Engineering 48 Back to Table 6 Palisade 2011 Risk Conference Copenhagen University College of Engineering 49 Back to Table 6 Palisade 2011 Risk Conference Copenhagen University College of Engineering 50 Back to Table 6 Palisade 2011 Risk Conference Copenhagen University College of Engineering 51 Back to Table 6 Palisade 2011 Risk Conference Copenhagen University College of Engineering 52 Back to Table 7 Palisade 2011 Risk Conference Copenhagen University College of Engineering 53 Back to Table 7 Palisade 2011 Risk Conference Copenhagen University College of Engineering 54 Back to Table 7 Palisade 2011 Risk Conference Copenhagen University College of Engineering 55 Back to Table 7 Palisade 2011 Risk Conference Copenhagen University College of Engineering 56 Thank You! Palisade 2011 Risk Conference Copenhagen University College of Engineering 57