DECISION MODELING USING EXCEL

UPDATED FOR MICROSOFT OFFICE XP AUGUST 2003

Michael R. Middleton

This page is intentionally mostly blank.

Copyright 2003 by Michael R. Middleton

Detailed Contents

Chapter 1 Introduction to Decision Modeling ............................................................... 9 1.1 Models to Aid Decision Making .............................................................................. 9 Components of a Decision Model ............................................................................ 10 1.2 Basic What-If Model.............................................................................................. 12 Influence Diagram Representation ........................................................................... 12 Decision Tree Representation .................................................................................. 13 Consequence Table Representation.......................................................................... 14 Chapter 2 Sensitivity Analysis Using SensIt ................................................................ 15 2.1 How to Install SensIt .............................................................................................. 15 2.2 How to Uninstall or Delete SensIt.......................................................................... 15 2.3 SensIt Overview ..................................................................................................... 16 2.4 Example Problem ................................................................................................... 16 2.5 Plot ......................................................................................................................... 17 Plot Input Variable ................................................................................................... 17 Plot Output Variable................................................................................................. 17 Plot Input Values...................................................................................................... 17 2.6 Spider ..................................................................................................................... 19 Spider Input Variables.............................................................................................. 19 Spider Output Variable............................................................................................. 19 Spider Input Changes ............................................................................................... 19 2.7 Tornado .................................................................................................................. 22 Tornado Input Variables........................................................................................... 22 Tornado Output Variable ......................................................................................... 23 Tornado Input Values............................................................................................... 23 2.8 Tornado Tips .......................................................................................................... 25 2.9 Eagle Airlines Problem .......................................................................................... 26 2.10 Seven-Variable Non-SensIt Spider Chart............................................................. 27

Detailed Contents

Chapter 3 Monte Carlo Simulation Using RiskSim .................................................... 29 3.1 Introduction ............................................................................................................ 29 3.2 Using RiskSim Functions....................................................................................... 31 3.3 Using RiskSim Functions....................................................................................... 31 3.4 Excel Error Message .............................................................................................. 32 3.5 Monte Carlo Simulation ......................................................................................... 32 3.6 Random Number Seed ........................................................................................... 33 3.7 One-Output Example.............................................................................................. 34 3.8 RiskSim Output for One-Output Example ............................................................. 35 3.9 Random Number Generator Functions................................................................... 38 RandBinomial .......................................................................................................... 38 RandCumulative....................................................................................................... 38 RandDiscrete ............................................................................................................ 40 RandExponential ...................................................................................................... 42 RandInteger .............................................................................................................. 43 RandNormal ............................................................................................................. 44 RandPoisson............................................................................................................. 45 RandTriangular ........................................................................................................ 45 RandUniform............................................................................................................ 47 3.10 RiskSim Technical Details ................................................................................... 48 3.11 Modeling Uncertain Relationships ....................................................................... 49 Base Model, Four Inputs .......................................................................................... 49 Three Inputs ............................................................................................................. 50 Two Inputs ............................................................................................................... 51 Four Inputs with Three Uncertainties....................................................................... 52 Intermediate Details ................................................................................................. 54 Chapter 4 Decision Trees Using TreePlan ................................................................... 57 4.1 TreePlan Installation .............................................................................................. 57 Occasional Use......................................................................................................... 57 Selective Use............................................................................................................ 57 Steady Use................................................................................................................ 58 4.2 Building a Decision Tree in TreePlan .................................................................... 58 4.3 Anatomy of a TreePlan Decision Tree ................................................................... 60 4.4 Step-by-Step TreePlan Tutorial.............................................................................. 62 DriveTek Problem .................................................................................................... 62 Nodes and Branches ................................................................................................. 63 Terminal Values ....................................................................................................... 64 Building the Tree Diagram....................................................................................... 65 Interpreting the Results ............................................................................................ 73 Formatting the Tree Diagram ................................................................................... 74 Displaying Model Inputs.......................................................................................... 76 Printing the Tree Diagram........................................................................................ 78 Alternative Model .................................................................................................... 79

Detailed Contents

4.5 Decision Tree Solution........................................................................................... 79 Strategy .................................................................................................................... 79 Payoff Distribution................................................................................................... 80 DriveTek Strategies.................................................................................................. 80 Strategy Choice ........................................................................................................ 84 Certainty Equivalent................................................................................................. 85 Rollback Method...................................................................................................... 87 Optimal Strategy ...................................................................................................... 88 4.6 Newox Decision Tree Problem .............................................................................. 90 4.7 Brandon Decision Tree Problem ............................................................................ 91 Decision Tree Strategies........................................................................................... 91 Chapter 5 Multiattribute Utility ................................................................................... 99 5.1 Applications of Multi-Attribute Utility .................................................................. 99 5.2 MultiAttribute Utility Swing Weights.................................................................. 100 Attribute Scores...................................................................................................... 101 Swing Weights ....................................................................................................... 102 Overall Scores ........................................................................................................ 103 5.3 Sensitivity Analysis Methods............................................................................... 104 Dominance ............................................................................................................. 105 Monetary Equivalents Assessment......................................................................... 105 Additive Utility Function ....................................................................................... 106 Weight Ratio Assessment....................................................................................... 107 Weight Ratio Sensitivity Analysis ......................................................................... 109 Swing Weight Assessment ..................................................................................... 110 Swing Weight Sensitivity Analysis ........................................................................ 112 Direct Weight Assessment and Sensitivity Analysis.............................................. 115 Summary ................................................................................................................ 117 Sensitivity Analysis Examples References ............................................................ 117 Screenshots from Excel to Word............................................................................ 118 Chapter 6 Product Mix Optimization ........................................................................ 119 6.1 Linear Programming Concepts............................................................................. 119 Formulation ............................................................................................................ 119 Graphical Solution.................................................................................................. 119 Sensitivity Analysis................................................................................................ 119 6.2 Basic Product Mix Problem ................................................................................. 121 6.3 Outdoors Problem ................................................................................................ 124 Spreadsheet Model ................................................................................................. 126 Solver Reports........................................................................................................ 127 Chapter 7 Uncertain Quantities.................................................................................. 129 7.1 Discrete Uncertain Quantities .............................................................................. 129 7.2 Continuous Uncertain Quantities ......................................................................... 129

Detailed Contents

Case A: Uniform Density ....................................................................................... 129 Case B: Ramp Density ........................................................................................... 132 Case C: Triangular Density .................................................................................... 134 Chapter 8 Simulation Without Add-Ins..................................................................... 137 8.1 Simulation Using Excel Functions ....................................................................... 137 Chapter 9 Multiperiod What-If Modeling ................................................................. 139 9.1 Apartment Building Purchase Problem ................................................................ 139 Apartment Building Analysis Notes....................................................................... 142 9.2 Product Launch Financial Model ......................................................................... 143 9.3 Machine Simulation Model .................................................................................. 147 Chapter 10 Modeling Waiting Lines .......................................................................... 155 10.1 Queue Simulation............................................................................................... 155 Chapter 11 Introduction to Data Analysis ................................................................. 161 11.1 Levels of Measurement ...................................................................................... 161 Categorical Measure............................................................................................... 161 Numerical Measure ................................................................................................ 161 11.2 Describing Categorical Data .............................................................................. 162 11.3 Describing Numerical Data ................................................................................ 162 Frequency Distribution and Histogram .................................................................. 162 Numerical Summary Measures .............................................................................. 162 Distribution Shapes ................................................................................................ 163 Chapter 12 Regression Models for Cross-Sectional Data......................................... 165 12.1 Cross-Sectional Regression Checklist................................................................ 165 Plot Y versus each X .............................................................................................. 165 Examine the correlation matrix .............................................................................. 165 Calculate the regression model with diagnostics.................................................... 165 Use the model......................................................................................................... 166 Chapter 13 Time Series Data and Forecasts.............................................................. 167 13.1 Time Series Patterns........................................................................................... 167 Chapter 14 Regression Models for Time Series Data ............................................... 171 14.1 Time Series Regression Checklist...................................................................... 171 Plot Y versus time .................................................................................................. 171 Plot Y versus each X .............................................................................................. 171 Examine the correlation matrix .............................................................................. 171 Calculate the regression model with diagnostics.................................................... 172 Use the model......................................................................................................... 172

Detailed Contents

14.2 Autocorrelation of Residuals.............................................................................. 173 Chapter 15 Sensitivity Analysis for Decision Trees................................................... 175 15.1 One-Variable Sensitivity Analysis ..................................................................... 175 15.2 Two-Variable Sensitivity Analysis..................................................................... 177 Setup for Data Table .............................................................................................. 178 Obtaining Results Using Data Table Command..................................................... 178 Embellishments ...................................................................................................... 179 15.3 Multiple-Outcome Sensitivity Analysis ............................................................. 180 15.4 Robin Pinelli's Sensitivity analysis..................................................................... 182 Chapter 16 Value of Information in Decision Trees ................................................. 185 16.1 Value of Information.......................................................................................... 185 16.2 Expected Value of Perfect Information.............................................................. 185 16.3 DriveTek Post-Contract-Award Problem ........................................................... 188 16.4 Sensitivity Analysis vs EVPI ............................................................................. 192 Chapter 17 Value of Imperfect Information.............................................................. 193 17.1 Technometrics Problem...................................................................................... 193 Prior Problem ......................................................................................................... 193 Imperfect Information ............................................................................................ 194 Revision of Probability........................................................................................... 196 Chapter 18 Modeling Attitude Toward Risk ............................................................. 197 18.1 Risk Utility Function.......................................................................................... 197 18.2 Exponential Risk Utility..................................................................................... 200 18.3 Approximate Risk Tolerance.............................................................................. 203 18.4 Exact Risk Tolerance Using Excel..................................................................... 203 18.5 Exact Risk Tolerance Using RiskTol.xla ........................................................... 207 18.6 Exponential Utility and TreePlan ....................................................................... 208 18.7 Exponential Utility and RiskSim........................................................................ 208 18.8 Risk Sensitivity for Machine Problem ............................................................... 210 18.9 Risk Utility Summary......................................................................................... 211 Concepts................................................................................................................. 211 Fundamental Property of Utility Function ............................................................. 212 Using a Utility Function To Find the CE of a Lottery............................................ 212 Exponential Utility Function .................................................................................. 212 TreePlan's Simple Form of Exponential Utility ..................................................... 212 Approximate Assessment of RiskTolerance .......................................................... 212 Exact Assessment of RiskTolerance ...................................................................... 213 Using Exponential Utility for TreePlan Rollback Values ...................................... 213 Using Exponential Utility for a Payoff Distribution .............................................. 214

Detailed Contents

Chapter 19 Modeling Marketing Decisions ............................................................... 215 19.1 Allocating Advertising Expenditures ................................................................. 215 Chapter 20 Nonlinear Product Mix Optimization .................................................... 221 20.1 Diminishing Profit Margin ................................................................................. 221 Chapter 21 Integer-Valued Optimization Models..................................................... 223 21.1 Transportation Problem...................................................................................... 223 21.2 Modified Transportation Problem ...................................................................... 224 21.3 Scheduling Problem ........................................................................................... 226 Chapter 22 Optimization Models for Finance Decisions .......................................... 229 22.1 Working Capital Management Problem............................................................. 229 22.2 Work Cap Alternate Formulations ..................................................................... 231 22.3 Stock Portfolio Problem ..................................................................................... 233 22.4 MoneyCo Problem ............................................................................................. 235

Introduction to Decision Modeling

1.1 MODELS TO AID DECISION MAKING

Decision: irrevocable allocation of resources Model: abstract representation of reality What makes decision difficult? Complexity many factors to consider; relationships among factors Uncertainty Conflicting Objectives How does modeling help? Complexity Model; consider each factor separately; consider relationships explicitly; avoid being overwhelmed Uncertainty Conflicting Objectives Sensitivity Analysis and Probability consider each objective; consider tradeoffs explicitly Goals of modeling: recommended solution, insight, clarity of action

10

Chapter 1 Introduction to Decision Modeling

Figure 1.1 Overall Model-Building Flowchart

Real World Abstraction Model

Math Model Difficult Problem Operations on Model

Implementation Model Results

Components of a Decision Model

Controllable input variables "What you can do," decision variables, alternatives Uncontrollable input variables "What you know and don't know," uncertainties, constraints Relationships how inputs are related to output, usually with intermediate variables, structure Intermediate variables useful for linking inputs to output Output variable "What you want," performance measure, overall satisfaction

1.1 Models to Aid Decision Making

11

Influence chart Rectangle for controllable inputs Rounded rectangle or oval for other variables Figure 1.2 Generic Influence Chart

Performance Measure (Output)

...

Intermediate Variables

...

Controllable Factor (Input)

...

Uncontrollable Factor (Input)

...

12

Chapter 1 Introduction to Decision Modeling

1.2 BASIC WHAT-IF MODEL

Influence Diagram Representation
Figure 1.3 Typical Influence Diagram

Net Cash Flow

Output

Total Costs Intermediate Variables Sales Revenue Total Variable Cost

Unit Price

Units Sold

Unit Variable Cost

Fixed Costs

Inputs

Figure 1.4 Typical Spreadsheet Model

1.2 Basic What-If Model

13

Figure 1.5 Formulas for Typical Spreadsheet Model

Figure 1.6 Defined Names for Typical Spreadsheet Model

Decision Tree Representation

Figure 1.7 Decision Fan and Event Fan

. . . Decision with many possible alternatives

=
Decision Fan

. . . Event with many possible outcomes

=
Event Fan

14

Chapter 1 Introduction to Decision Modeling

Figure 1.8 Conceptual Decision Tree

$ Unit Price Fixed Costs Units Sold Unit Variable Cost

Net Cash Flow

Consequence Table Representation

Figure 1.9 Professor's Summer Decision Conflicting Objectives Alternatives Develop Software Teach MBAs Vacation Cash Flow $2700 $4300 $0 Hassle-Free Yes No Yes Happy Deans Maybe Yes No Professional Fame Maybe No No

Sensitivity Analysis Using SensIt

SensIt is a sensitivity analysis add-in for Microsoft Excel 97 (and later versions of Excel) for Windows and Macintosh. It was written by Mike Middleton of the University of San Francisco and Jim Smith of Duke University.

2.1 HOW TO INSTALL SENSIT

Here are three ways to install SensIt: (1) Start Excel, and use Excels File | Open command to open the SensIt.xla file from floppy or hard drive. (2) Copy the SensIt.xla file to the Excel | Library subdirectory of your hard drive. Start Excel, and use Excels Tools | Add-Ins command to load and unload SensIt as needed. (3) Copy the SensIt.xla file to the Excel | Startup subdirectory of your hard drive, in which case the file will be opened every time you start Excel. All of SensIts functionality, including its built-in help, is a part of the SensIt.xla file. There is no separate setup file or help file.

2.2 HOW TO UNINSTALL OR DELETE SENSIT

(A) First, use your file manager to locate SensIt.xla, and delete the file from your hard drive. (B1) If SensIt is listed under Excel's add-in manager and the box is checked, when you start Excel you'll see "Cannot find ..." Click OK. Choose Tools | Add-Ins, uncheck the box for SensIt; you'll see "Cannot find ... Delete from list?" Click Yes.

(B2) If SensIt is listed under Excel's add-in manager and the box is not checked, start Excel and choose Tools | Add-Ins. Check the box for SensIt; you'll see "Cannot find ... Delete from list?" Click Yes.

16

Chapter 2 Sensitivity Analysis Using SensIt

2.3 SENSIT OVERVIEW

To run SensIt, start Excel and open the SensIt.xla file. Alternatively, install SensIt using one of the methods described above. SensIt adds a Sensitivity Analysis command to the Tools menu. The Sensitivity Analysis command has four subcommands: Plot, Spider, Tornado, and Help. Before using the SensIt options, you must have a spreadsheet model with one or more inputs and an output. All three SensIt options make it easy for you to see how sensitive the output is to changes in the inputs. Use SensIts Plot option to see how your models output depends on changes in a single input variable. Use SensIts Spider option to see how your models output depends on the same percentage changes for each of the models input variables. Use SensIts Tornado option to see how your models output depends on ranges you specify for each of the models input variables.

2.4 EXAMPLE PROBLEM

Figure 2.1 Model Display
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 A B Spreadsheet Model For Eagle Airlines Input Variables Charter Price/Hour Ticket Price/Hour Hours Flown Capacity of Scheduled Flights Proportion of Chartered Flights Operating Cost/Hour Insurance Intermediate Calculations Total Revenue Total Cost Performance Measure Annual Profit Input Cells $325 $100 800 50% 0.5 $245 $20,000 C

$230,000 $216,000

$14,000

Adapted from Bob Clemen's textbook, Making Hard Decisions, 2nd ed., Duxbury (1996).

2.5 Plot

17

Figure 2.2 Model Formulas

A 11 12 13 14 15 16 17 18 Intermediate Calculations Total Revenue Total Cost Performance Measure Annual Profit B

=(B8*B6*B4)+((1-B8)*B6*B5*B7*5) =(B6*B9)+B10

=B13-B14

2.5 PLOT
Use SensIts Plot option to see how your models output depends on changes in a single input variable.

Plot Input Variable

Plot Input Variables Cells: Option: In the Label edit box, type a cell reference, or point to the cell containing a text label and click. Required: In the Cell edit box, type a cell reference, or point to the cell containing a numeric value thats an input to your model.

Plot Output Variable

Plot Output Variables Cells: Option: In the Label edit box, type a cell reference, or point to the cell containing a text label and click. Required: In the Cell edit box, type a cell reference, or point to the cell containing a formula thats the output of your model.

Plot Input Values

Plot Input Values: Type numbers in the Start, Step, and Stop edit boxes to specify values to be used in the input variables cell. Cell references are not allowed. Send Output To: Select the destination for the output table and chart. If you send output to This Worksheet, enter a Cell reference for the top left corner of the output. Output options are not available on the Macintosh; output is always sent to a new worksheet. Click OK: SensIt Plot uses the Start, Step, and Stop values to prepare a table of values. Each value is copied to the input variable cell, the worksheet is recalculated, and the value of the output variable cell is copied to the table. (You could do this manually using the Edit | Fill | Series and Data | Table commands.) SensIt Plot uses the input and output values to prepare an XY (Scatter) chart; optionally, the text in the label cells you

18

Chapter 2 Sensitivity Analysis Using SensIt

identified are used as the charts axis labels. (You could do this manually using the ChartWizard.) Figure 2.3 SensIt Plot Dialog Box

Figure 2.4 SensIt Plot Numerical and Chart Output

Sensitivity Analysis ... Input Output Capacity of Scheduled Flights Annual Profit 40% -$6,000 42% -$2,000 44% $2,000 46% $6,000 48% $10,000 50% $14,000 52% $18,000 54% $22,000 56% $26,000 58% $30,000 60% $34,000

SensIt - Sensitivity Analysis - Plot

$30,000 Annual Profit $20,000 $10,000 $0 -$10,000 40%

45%

50%

55%

60%

Capacity of Scheduled Flights

2.6 Spider

19

2.6 SPIDER
Use SensIts Spider option to see how your models output depends on the same percentage changes for each of the models input variables. Before using Spider, arrange your model input cells in adjacent cells in a single column, arrange corresponding labels in adjacent cells in a single column, and be sure your models input cells contain base case values. For example, if your model has five inputs, the names of the five inputs could be text in A1:A5. The input cells of your model could be numbers in B1:B5; when you change a number in one of these cells, the output of your model changes; enter base case values in the input cells B1:B5 before using Spider.

Spider Input Variables

Spider Input Variables Ranges: Labels edit box: Type a range reference, or point to the range (click and drag) containing text labels. Cells edit box: Type a range reference, or point to the range containing numeric values that are inputs to your model. Each range must be adjacent cells in a single column.

Spider Output Variable

Spider Output Variables Cells: Label edit box: Type a cell reference, or point to the cell containing a text label and click. Cell edit box: Type a cell reference, or point to the cell containing a formula thats the output of your model.

Spider Input Changes

Spider Input Changes (%): Type numbers in the Start (%), Step (%), and Stop (%) edit boxes to define the percents that will be multiplied times the current value in each input variables cell. Cell references are not allowed. Send Output To: Select the destination for the output table and chart. If you send output to This Worksheet, enter a Cell reference for the top left corner of the output. Output options are not available on the Macintosh; output is always sent to a new worksheet. Click OK: SensIt Spider uses the Start (%), Step (%), and Stop (%) values and the original (base case) numeric value in each input variable cell to prepare a table of percentage change input values. For each input variable, all other input values are set at their base case values, each percentage change input value is copied to the input variable cell, the worksheet is recalculated, and the value of the output variable cell is copied to the table. The output variable values are also expressed as percentage change of the base case output value. SensIt Spider prepares two XY (Scatter) charts; the horizontal axis is percentage change of input variables; the vertical axis is model output value on one chart

20

Chapter 2 Sensitivity Analysis Using SensIt

and percentage change of model output value on the other; the input variables labels are used for chart legends. Figure 2.5 SensIt Spider Dialog Box

2.6 Spider

21

Figure 2.6 SensIt Spider Numerical Output

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 A Input Variables Values Charter Price/Hour Ticket Price/Hour Hours Flown Capacity of Scheduled Flights Proportion of Chartered Flights Operating Cost/Hour Insurance B 80% $260 $80 640 40% 0.4 $196 $16,000 C 90% $293 $90 720 45% 0.45 $221 $18,000 D 100% $325 $100 800 50% 0.5 $245 $20,000 E 110% $358 $110 880 55% 0.55 $270 $22,000 F 120% $390 $120 960 60% 0.6 $294 $24,000

Output Variable Values (Annual Profit) 80% Charter Price/Hour -$12,000 Ticket Price/Hour -$6,000 Hours Flown $7,200 Scheduled Capacity -$6,000 Chartered Proportion $8,000 Operating Cost/Hour $53,200 Insurance $18,000

90% $1,000 $4,000 $10,600 $4,000 $11,000 $33,600 $16,000

100% $14,000 $14,000 $14,000 $14,000 $14,000 $14,000 $14,000

110% $27,000 $24,000 $17,400 $24,000 $17,000 -$5,600 $12,000

120% $40,000 $34,000 $20,800 $34,000 $20,000 -$25,200 $10,000

Figure 2.7 SensIt Spider Chart Output

Sensit - Sensitivity Analysis - Spider

Annual Profit Value $60,000 $50,000 $40,000 $30,000 $20,000 $10,000 $0 -$10,000 -$20,000 76808488929610101011111212 -$30,000 % % % % % % 0 4 8 2 6 0 4 %% % % % % % % Change in Input Value

Charter Price/Hour Ticket Price/Hour Hours Flown Capacity of Scheduled Flights Proportion of Chartered Flights Operating Cost/Hour

22

Chapter 2 Sensitivity Analysis Using SensIt

Figure 2.8 SensIt Spider Chart Output After Formatting

Sensit - Sensitivity Analysis - Spider $60,000 $50,000 $40,000 Annual Profit Value $30,000 $20,000 $10,000 $0 75% -$10,000 -$20,000 -$30,000 % Change in Input Value Charter Price/Hour Ticket Price/Hour Hours Flow n Scheduled Capacity Chartered Proportion Operating Cost/Hour 80% 85% 90% 95% 100% 105% 110% 115% 120% 125% Insurance

2.7 TORNADO
Use SensIts Tornado option to see how your models output depends on ranges you specify for each of the models input variables. Before using Tornado, arrange your model input cells in adjacent cells in a single column, arrange corresponding labels in adjacent cells in a single column, and arrange Low, Base, and High input values for each input variable in three separate columns. Alternatively, the three columns containing input values can be worst case, likely case, and best case. For example, if your model has five inputs, the names of the five inputs could be text in A1:A5. The input cells of your model could be numbers in B1:B5; when you change a number in one of these cells, the output of your model changes. The Low input values could be numbers in D1:D5, chosen as the minimum possible value you think each input variable could be. The Base input values could be numbers in E1:E5, chosen as the most likely value for each input; you might also have these same numbers in B1:B5 as current inputs to your model. The High input values could be numbers in F1:F5, chosen as the maximum possible value you think each input variable could be.

Tornado Input Variables

Tornado Input Variables Ranges: Labels edit box: Type a range reference, or point to the range (click and drag) containing text labels. Cells edit box: Type a range reference, or

2.7 Tornado

23

point to the range containing numeric values that are inputs to your model. Each range must be adjacent cells in a single column.

Tornado Output Variable

Tornado Output Variables Cells: Label edit box: Type a cell reference, or point to the cell containing a text label and click. Cell edit box: Type a cell reference, or point to the cell containing a formula thats the output of your model.

Tornado Input Values

Tornado Input Values Ranges: In the Low, Base, and High edit boxes, type a range reference, or point to the range (click and drag) containing numeric values for each of your models inputs. Send Output To: Select the destination for the output table and chart. If you send output to This Worksheet, enter a Cell reference for the top left corner of the output. Output options are not available on the Macintosh; output is always sent to a new worksheet. Click OK: For each input variable, SensIt Tornado sets all other input values at their Base case values, copies the Low input value to the input variable cell, recalculates the worksheet, and copies the value of the output variable cell to the table; the steps are repeated using each High input value. For each input variable, SensIt Tornado computes the range of the output variable values, sorts the table from largest range down to smallest range, and prepares a bar chart. Figure 2.9 Example with Lower and Upper Bounds
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 A B Spreadsheet Model For Eagle Airlines Input Variables Charter Price/Hour Ticket Price/Hour Hours Flown Capacity of Scheduled Flights Proportion of Chartered Flights Operating Cost/Hour Insurance Intermediate Calculations Total Revenue Total Cost Performance Measure Annual Profit Input Cells $325 $100 800 50% 0.5 $245 $20,000 C D E F

Lower Bound Base Value Upper Bound $300 $325 $350 $95 $100 $108 500 800 1000 40% 50% 60% 0.45 0.5 0.7 $230 $245 $260 $18,000 $20,000 $25,000

$230,000 $216,000

$14,000

Adapted from Bob Clemen's textbook, Making Hard Decisions, 2nd ed., Duxbury (1996).

24

Chapter 2 Sensitivity Analysis Using SensIt

Figure 2.10 SensIt Tornado Dialog Box

Figure 2.11 SensIt Tornado Numerical and Chart Output

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 A Tornado Analysis ... B C D E F G H I J K

Input Variable Capacity of Scheduled Flights Operating Cost/Hour Hours Flown Charter Price/Hour Proportion of Chartered Flights Ticket Price/Hour Insurance

Input Values Low Base High 40% 50% 60% $230 $245 $260 500 800 1000 $300 $325 $350 0.45 0.5 0.7 $95 $100 $108 $18,000 $20,000 $25,000

Output Values (Annual Profit) Low Base High -$6,000 $14,000 $34,000 $26,000 $14,000 $2,000 $1,250 $14,000 $22,500 $4,000 $14,000 $24,000 $11,000 $14,000 $26,000 $9,000 $14,000 $22,000 $16,000 $14,000 $9,000

Percent Swing Variance $40,000 46.1% $24,000 16.6% $21,250 13.0% $20,000 11.5% $15,000 6.5% $13,000 4.9% $7,000 1.4%

SensIt - Sensitivity Analysis - Tornado

Capacity of Scheduled Flights Operating Cost/Hour Hours Flown Charter Price/Hour Proportion of Chartered Flights Ticket Price/Hour Insurance -$15,000 -$5,000 40% $260 500 $300 0.45 $95 $25,000 $5,000 $18,000 $15,000 Annual Profit $25,000 $35,000 $108 1000 $350 0.7 $230 60%

2.8 Tornado Tips

25

2.8 TORNADO TIPS

When defining the high and low cases for each variable, it is important to be consistent so that the "high" cases are all equally high and the "low" cases are equally low. For example, you might take all of the base case values to be estimates of the mean of the input variable, take low cases to be values such there is a 1-in-10 chance of the variable being below this amount, and take the high cases to be values such that there is a 1-in-10 chance of the variable being above this amount. Alternatively, you may specify low and high values that are the absolute lowest and highest possible values. When you click OK, SensIt sets all of the input variables to their base-case values and records the output value. Then SensIt goes through each of the input variables one at a time, plugs the low-case value into the input cell, and records the value in the output cell. It then repeats the process for the high case. For each substitution, all input values are kept at their base-case values except for the single input value that is setn at it low or high value. SensIt then produces a spreadsheet that lists the numerical results as shown in columns F, G, and H above. In the worksheet, the variables are sorted by their "swing" -- the absolute value of the difference between the output values in the low and high cases. "Swing" serves as a rough measure of the impact of each input variable. The rows of numerical output are sorted from highest swing at the top down to lowest swing at the bottom. Then SensIt creates a bar chart of the sorted data. "Percent variance" is a standardized measure of impact: it squares each swing, sums them up to get a "Total Variance", and reports the percentage of the "total variance" attributed to each input variable. In general, you should focus your modeling efforts on those variables with the greatest impact on the value measure. If your model has input variables that are discrete or categorical, you should create multiple tornado charts using different base case values of that input variable. For example, if your model has an input variable "Government Regulation" that has possible values 0 (zero) or 1, the low and high values will be 0 and 1, but you should run one tornado chart with base case = 0 and another tornado chart with base case = 1.

26

Chapter 2 Sensitivity Analysis Using SensIt

2.9 EAGLE AIRLINES PROBLEM

Figure 2.12 Eagle Model Display
A B C D E F 1 Spreadsheet Model For Eagle Airlines 2 3 Variable Input Cells Lower Bound Base Value Upper Bound 4 Hours Flown 800 500 800 1000 5 Charter Price/Hour $325 $300 $325 $350 6 Ticket Price/Hour $100 $95 $100 $108 7 Capacity of Scheduled Flights 50% 40% 50% 60% 8 Proportion Of Chartered Flights 0.5 0.45 0.5 0.7 9 Operating Cost/Hour $245 $230 $245 $260 10 Insurance $20,000 $18,000 $20,000 $25,000 11 Proportion Financed 0.4 0.3 0.4 0.5 12 Interest Rate 11.5% 10.5% 11.5% 13.0% 13 Purchase Price $87,500 $85,000 $87,500 $90,000 14 15 $230,000 Total Revenue 16 $220,025 Total Cost 17 18 $9,975 Annual Profit 19 20 Adapted from Bob Clemen's textbook, Making Hard Decisions

Figure 2.13 Eagle Model Formulas

A 14 15 16 17 18 19 Total Revenue Total Cost Annual Profit B C D E F =(B8*B4*B5)+((1-B8)*B4*B6*B7*5) =(B4*B9)+B10+(B13*B11*B12) =B15-B16

2.10 Seven-Variable Non-SensIt Spider Chart

27

Figure 2.14 Worst Case and Best Case Inputs Determined by Solver
Variable Hours Flown Charter Price/Hour Ticket Price/Hour Capacity of Scheduled Flights Proportion Of Chartered Flights Operating Cost/Hour Insurance Proportion Financed Interest Rate Purchase Price Total Revenue Total Cost Annual Profit Worst Case 1000 $300 $95 40% 0.45 $260 $25,000 0.5 13.0% $90,000 $239,500 $290,850 -$51,350 Base Case 800 $325 $100 50% 0.5 $245 $20,000 0.4 11.5% $87,500 $230,000 $220,025 $9,975 Best Case 1000 $350 $108 60% 0.7 $230 $18,000 0.3 10.5% $85,000 $342,200 $250,678 $91,523

2.10 SEVEN-VARIABLE NON-SENSIT SPIDER CHART

Figure 2.15 Extremes Associated With Tornado-Like Inputs
Spider Chart for Eagle Airlines $35,000

$30,000

$25,000

$20,000 Output Value: Annual Profit Capacity of Scheduled Flights Operating Cost/Hour Hours Flown Charter Price/Hour Proportion Of Chartered Flights Ticket Price/Hour Insurance

$15,000

$10,000

$5,000

$0

-$5,000

-$10,000

-$15,000 40%

60%

80%

100%

120%

140%

160%

Input Value, as Percent of Base Case

28

Chapter 2 Sensitivity Analysis Using SensIt

This page is intentionally mostly blank.

Monte Carlo Simulation Using RiskSim

3.1 INTRODUCTION

Figure 3.1 Conceptual Simulation as a Sample of Tree Endpoints

$ Unit Price Fixed Costs 3 values Units Sold ~400 values Unit Variable Cost ~500 values

Net Cash Flow

~600,000 values

30

Chapter 3 Monte Carlo Simulation Using RiskSim

Figure 3.2 Probability Distributions for Sampling Tree Endpoints

$ Unit Price Fixed Costs Units Sold Unit Variable Cost

Net Cash Flow

Discrete

Normal

Uniform

Figure 3.3 Conceptual Simulation as Influence Chart with Repeated What-Ifs

Net Cash Flow

Unit Price

Fixed Costs

Units Sold

Unit Variable Cost

$29

Constant

Discrete

Normal

Uniform

3.2 Using RiskSim Functions

31

3.2 USING RISKSIM FUNCTIONS

RiskSim is a Monte Carlo Simulation add-in for Microsoft Excel 97 (and later versions of Excel) for Windows and Macintosh. RiskSim provides random number generator functions as inputs for your model, automates Monte Carlo simulation, and creates charts. Your spreadsheet model may include various uncontrollable uncertainties as input assumptions (e.g., demand for a new product, uncertain variable cost of production, competitor reaction), and you can use simulation to determine the uncertainty associated with the model's output (e.g., annual profit). RiskSim automates the simulation by trying hundreds of what-ifs consistent with your assessment of the uncertainties. To use RiskSim, you (1) (2) (3) (4) (5) (6) create a spreadsheet model optionally use SensIt to identify critical inputs enter one of RiskSim's nine random number generator functions in each input cell of your model choose Tools | Risk Simulation from Excel's menu specify the model output cell and the number of what-if trials interpret RiskSim's histogram and cumulative distribution charts.

RiskSim facilitates Monte Carlo simulation by providing: Nine random number generator functions Ability to set the seed for random number generation Automatic repeated sampling for simulation Frequency distribution of simulation results Histogram and cumulative distribution charts

3.3 USING RISKSIM FUNCTIONS

RiskSim adds nine random number generator functions to Excel. You can use these functions as inputs to your model by typing in a worksheet cell or by using the Function Wizard. From the Insert menu choose Function, or click the Function Wizard button. RiskSim's functions are listed in a User Defined category. The nine functions are: RANDBINOMIAL(trials,probability_s) RANDCUMULATIVE(value_cumulative_table) RANDDISCRETE(value_discrete_table)

32

Chapter 3 Monte Carlo Simulation Using RiskSim

RANDEXPONENTIAL(lambda) RANDINTEGER(bottom,top) RANDNORMAL(mean,standard_dev) RANDPOISSON(mean) RANDTRIANGULAR(minimum,most_likely,maximum) RANDUNIFORM(minimum,maximum) RiskSim's RAND... functions include extensive error checking of arguments. After verifying that the functions are working properly, you may want to substitute RiskSim's FAST... functions which have minimal error checking and therefore run faster. From the Edit menu choose Replace; in the Replace dialog box, type =RAND in the "Find What" edit box, type =FAST in the "Replace with" edit box, and click the Replace All button.

3.4 EXCEL ERROR MESSAGE

When you insert a RiskSim random number generator function in a worksheet cell, the function is linked to RiskSim.xla. When you save the workbook, Excel saves the complete path to the function in RiskSim.xla. When you open the workbook, Excel looks for RiskSim.xla using the saved path. If Excel cannot find RiskSim.xla at the saved path location (e.g., if you deleted RiskSim.xla or if you opened the workbook on another computer where RiskSim.xla isn't located at the same path), Excel displays a dialog box: "This document contains links. Re-establish links?" Click No. The workbook will be opened, but any cell containing a reference to a RiskSim function will display the #REF!, #NAME?, or other error code. To fix the links, be sure that RiskSim.xla is open (e.g., File | Open | RiskSim.xla), choose Edit | Links | Change Source, and locate the RiskSim.xla file that is open.

3.5 MONTE CARLO SIMULATION

After specifying random number generator functions as inputs to your model, from the Tools choose Risk Simulation | One Output.

3.6 Random Number Seed

33

Figure 3.4 RiskSim Dialog Box

Optionally, select the "Output Label Cell" edit box, and point or type a reference to a cell containing the name of the model output (for example, a cell whose contents is the text label "Net Profit"). Select the "Output Formula Cell" edit box, and point to a single cell on your worksheet or type a cell reference. The output cell of your model must contain a formula that depends, usually indirectly, on the model inputs determined by the random number generator functions. Select the "Random Number Seed" edit box, and type a number between zero and one. (If you want to change the seed without performing a simulation, enter zero in the "Number of iterations" edit box.) Select the "Number Of Trials" edit box, and type an integer value (for example, 100 or 500). This value, sometimes called the sample size or number of iterations, specifies the number of times the worksheet will be recalculated to determine output values of your model.

3.6 RANDOM NUMBER SEED

The "Random Number Seed" edit box on the RiskSim dialog box allows you to set the seed for RiskSim's random number generator functions. These functions depend on RiskSim's own uniform random number function that is completely independent of Excel's built-in RAND(). Random numbers generated by the computer are actually pseudo-random. The numbers appear to be random, and they pass various statistical tests for randomness. But they are

34

Chapter 3 Monte Carlo Simulation Using RiskSim

actually calculated by an algorithm where each random number depends on the previous random number. Such an algorithm generates a repeatable sequence. The seed specifies where the algorithm starts in the sequence. A Monte Carlo simulation model usually has uncontrollable inputs (uncertain quantities using random number generator functions), controllable inputs (decision variables that have fixed values for a particular set of simulation iterations), and an output variable (a performance measure or operating characteristic of the system). For example, a simple queuing system model may have an uncertain arrival pattern, a controllable number of servers, and total cost (waiting time plus server cost) as output. To evaluate a different number of servers, you would specify the same seed before generating the uncertain arrivals. Then the variation in total cost should depend on the different number of servers, not on the particular sequence of random numbers that generates the arrivals.

3.7 ONE-OUTPUT EXAMPLE

In this example the decision maker has described his subjective uncertainty using normal, triangular, and discrete probability distributions. Figure 3.5 One-Output Example Model Display
1 2 3 4 5 6 7 8 9 A B Software Decision Analysis Unit Price Units Sold Unit Variable Cost Fixed Costs Net Cash Flow $29 739 $8.05 $12,000 $3,485 C D E F G H

Price is controllable and constant. Normal Mean = 700, StDev = 100 Triangular Min = $6, Mode = $8, Max = $11 Discrete Value Probability $10,000 0.25 $12,000 0.50 $15,000 0.25

Figure 3.6 One-Output Example Model Formulas

1 2 3 4 5 6 7 8 A Software Decision Analysis Unit Price Units Sold Unit Variable Cost Fixed Costs Net Cash Flow B

$29 =INT(RANDNORMAL(700,100)) =RANDTRIANGULAR(6,8,11) =RANDDISCRETE(E7:F9) =B4*(B3-B5)-B6

3.8 RiskSim Output for One-Output Example

35

Figure 3.7 RiskSim Dialog Box for One-Output Example

3.8 RISKSIM OUTPUT FOR ONE-OUTPUT EXAMPLE

When you click the Simulate button, RiskSim creates a new worksheet in your Excel workbook named "RiskSim Summary 1." A summary of your inputs and the output is shown in cells L1:R9 with the accompanying histogram and cumulative distribution charts.

36

Chapter 3 Monte Carlo Simulation Using RiskSim

Figure 3.8 RiskSim Summary Output for One-Output Example

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 L M N RiskSim - One Output - Summary Date (current date) Time (current time) Workbook risksamp.xls Worksheet Simulation Output Cell $B$8 Output Label Net Cash Flow Seed 0.5 Trials 500 O P Q Mean St. Dev. Mean St. Error Minimum First Quartile Median Third Quartile Maximum Skewness R $2,454 $2,794 $125 -$5,455 $536 $2,458 $4,416 $10,236 0.0028

RiskSim Histogram, (current date), (current time) 140 120 100 Frequency 80 60 40 20 0 -$6,000 -$2,000 $2,000 $6,000 $10,000 Net Cash Flow, Upper Limit of Interval

RiskSim Cumulative Chart, (current date), (current time) 1.0 0.9 Cumulative Probability 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 -$6,000 -$4,000 -$2,000 $0 $2,000 $4,000 $6,000 $8,000 $10,000 $12,000

Net Cash Flow

3.8 RiskSim Output for One-Output Example

37

The histogram is based on the frequency distribution in columns I:J. The cumulative distribution is based on the sorted output values in column C and the cumulative probabilities in column D. Figure 3.9 RiskSim Numerical Output for One-Output Example
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 B Trial Net Cash Flow 1 $1,653 2 $2,804 3 $2,280 4 $761 5 -$1,817 6 -$692 7 $623 8 $5,575 9 $1,389 10 $445 11 $2,573 12 $5,055 13 $1,430 14 $4,529 15 $701 16 -$903 17 $3,900 18 $7,282 19 $9,901 20 $285 21 $3,833 22 $4,369 23 $1,991 24 -$11 25 $1,100 26 -$1,100 27 -$5,455 C D Sorted Cumulative -$5,455 0.0010 -$4,267 0.0030 -$4,185 0.0050 -$3,898 0.0070 -$3,675 0.0090 -$3,582 0.0110 -$3,569 0.0130 -$3,562 0.0150 -$3,547 0.0170 -$3,275 0.0190 -$3,207 0.0210 -$3,137 0.0230 -$3,135 0.0250 -$3,063 0.0270 -$3,036 0.0290 -$3,008 0.0310 -$2,968 0.0330 -$2,950 0.0350 -$2,774 0.0370 -$2,649 0.0390 -$2,485 0.0410 -$2,370 0.0430 -$2,319 0.0450 -$2,219 0.0470 -$2,195 0.0490 -$1,986 0.0510 -$1,969 0.0530 E F Percent 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% G Percentile -$5,455 -$1,996 -$1,132 -$637 $77 $536 $923 $1,331 $1,823 $2,063 $2,458 $2,756 $3,138 $3,644 $4,104 $4,416 $4,867 $5,412 $5,897 $7,109 $10,236 H I J Upper Limit Frequency -$6,000 0 -$4,000 3 -$2,000 22 $0 72 $2,000 122 $4,000 128 $6,000 105 $8,000 38 $10,000 9 $12,000 1 0

The cumulative probabilities start at 1/(2*N), where N is the number of trials, and increase by 1/N. The rationale is that the lowest ranked output value of the sampled values is an estimate of the population's values in the range from 0 to 1/N, and the lowest ranked value is associated with the median of that range. Column B contains the original sampled output values. Columns F:G show percentiles based on Excel's PERCENTILE worksheet function. Refer to Excel's online help for the interpolation method used by the PERCENTILE function. The summary measures in columns Q:R are also based on Excel worksheet functions: AVERAGE, STDEV, QUARTILE, and SKEW.

38

Chapter 3 Monte Carlo Simulation Using RiskSim

3.9 RANDOM NUMBER GENERATOR FUNCTIONS

RandBinomial
Returns a random value from a binomial distribution. The binomial distribution can model a process with a fixed number of trials where the outcome of each trial is a success or failure, the trials are independent, and the probability of success is constant. RANDBINOMIAL counts the total number of successes for the specified number of trials. If n is the number of trials, the possible values for RANDBINOMIAL are the nonnegative integers 0,1,...,n. RANDBINOMIAL Syntax: RANDBINOMIAL(trials,probability_s) Trials (often denoted n) is the number of independent trials. Probability_s (often denoted p) is the probability of success on each trial. RANDBINOMIAL Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if an argument is text and the name is undefined. Returns #NUM! if trials is non-integer or less than one, or probability_s is less than zero or more than one. Returns #VALUE! if an argument is a defined name of a cell and the cell is blank or contains text. RANDBINOMIAL Example A salesperson makes ten unsolicited calls per day, where the probability of making a sale on each call is 30 percent. The uncertain total number of sales in one day is =RANDBINOMIAL(10,0.3) RANDBINOMIAL Related Function FASTBINOMIAL: Same as RANDBINOMIAL without any error checking of the arguments. CRITBINOM(trials,probability_s,RAND()): Excel's inverse of the cumulative binomial, or CRITBINOM(trials,probability_s,RANDUNIFORM(0,1)) to use the RiskSim Seed feature.

RandCumulative
Returns a random value from a piecewise-linear cumulative distribution. This function can model a continuous-valued uncertain quantity, X, by specifying points on its cumulative distribution. Each point is specified by a possible value, x, and a

3.9 Random Number Generator Functions

39

corresponding left-tail cumulative probability, P(X<=x). Random values are based on linear interpolation between the specified points. RANDCUMULATIVE Syntax: RANDCUMULATIVE(value_cumulative_table) Value_cumulative_table must be a reference, or the defined name of a reference, for a two-column range, with values in the left column and corresponding cumulative probabilities in the right column. RANDCUMULATIVE Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if the argument is text and the name is undefined. Returns #NUM! if the first (top) cumulative probability is not zero, if the last (bottom) cumulative probability is not one, or if the values or cumulative probabilities are not in ascending order. Returns #REF! if the number of columns in the table reference is not two. Returns #VALUE! if the argument is not a reference, if the argument is a defined name but not for a reference, or if any cell of the table contains text or is blank. RANDCUMULATIVE Example A corporate planner thinks that minimum possible market demand is 1000 units, median is 5000, and maximum possible is 9000. Also, there is a ten percent chance that demand will be less than 4000 and a ten percent chance it will exceed 7000. The values, x, and cumulative probabilities, P(X<=x), are entered into spreadsheet cells A1:B5. Figure 3.10 RandDiscrete Example Spreadsheet Data

The function is entered into another cell: =RANDCUMULATIVE(A1:B5) RANDCUMULATIVE Related Function FASTCUMULATIVE: Same as RANDCUMULATIVE without any error checking of the arguments.

40

Chapter 3 Monte Carlo Simulation Using RiskSim

Figure 3.11 RandCumulative Example Probability Density Function

0.0005

0.0004 Probability Density, f(x)

0.0003

0.0002

0.0001

0 0 2000 4000 6000 8000 10000 Market Demand, x, in units

Figure 3.12 RandCumulative Example Cumulative Probability Function

1

Cumulative Probability, P(X<=x

0.8

0.6

0.4

0.2

0 0 2000 4000 6000 8000 10000 Market Demand, x, in units

RandDiscrete
Returns a random value from a discrete probability distribution. This function can model a discrete-valued uncertain quantity, X, by specifying its probability mass function. The function is specified by each possible discrete value, x, and its corresponding probability, P(X=x). RANDDISCRETE Syntax: RANDDISCRETE(value_discrete_table)

3.9 Random Number Generator Functions

41

Value_discrete_table must be a reference, or the defined name of a reference, for a twocolumn range, with values in the left column and corresponding probability mass in the right column. RANDDISCRETE Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if the argument is text and the name is undefined. Returns #NUM! if a probability is negative or if the probabilities do not sum to one. Returns #REF! if the number of columns in the table reference is not two. Returns #VALUE! if the argument is not a reference, if the argument is a defined name but not for a reference, or if any cell of the table contains text or is blank. RANDDISCRETE Example A corporate planner thinks that uncertain market demand, X, can be approximated by three possible values and their associated probabilities: P(X=3000) = 0.3, P(X=4000) = 0.6, and P(X=5000) = 0.1. The values and probabilities are entered into spreadsheet cells A1:B3. Figure 3.13 RandDiscrete Example Spreadsheet Data

The function is entered into another cell: =RANDDISCRETE(A1:B3) RANDDISCRETE Related Function FASTDISCRETE: Same as RANDDISCRETE without any error checking of the arguments. RandDiscrete Example Probability Mass Function

42

Chapter 3 Monte Carlo Simulation Using RiskSim

Figure 3.14 RandDiscrete Example Probability Mass Function

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1000 2000 3000 4000 5000 6000 7000

Ma rke t D e ma nd, x, in units

Figure 3.15 RandDiscrete Example Cumulative Probability Function

1

Cumulative Probability, P(X<=x

0.8

0.6

0.4

0.2

0 0 1000 2000 3000 4000 5000 6000 7000 Market Demand, x, in units

RandExponential
Returns a random value from an exponential distribution. This function can model the uncertain time interval between successive arrivals at a queuing system or the uncertain time required to serve a customer. RANDEXPONENTIAL Syntax: RANDEXPONENTIAL(lambda) Lambda is the mean number of occurrences per unit of time.

3.9 Random Number Generator Functions

43

RANDEXPONENTIAL Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if the argument is text and the name is undefined. Returns #NUM! if lambda is negative or zero. Returns #VALUE! if the argument is a defined name of a cell and the cell is blank or contains text. RANDEXPONENTIAL Examples Cars arrive at a toll plaza with a mean rate of 3 cars per minute. The uncertain time between successive arrivals, measured in minutes, is =RANDEXPONENTIAL(3). The average value returned by repeated recalculation of RANDEXPONENTIAL(3) is 0.333. A bank teller requires an average of two minutes to serve a customer. The uncertain customer service time, measured in minutes, is =RANDEXPONENTIAL(0.5). The average value returned by repeated recalculation of RANDEXPONENTIAL(0.5) is 2. RANDEXPONENTIAL Related Functions FASTEXPONENTIAL: Same as RANDEXPONENTIAL without any error checking of the arguments. LN(RAND())/lambda: Excel's inverse of the exponential, or LN(RANDUNIFORM(0,1))/lambda to use the RiskSim Seed feature. RANDPOISSON: Counts number of occurrences for a Poisson process.

RandInteger
Returns a uniformly distributed random integer between two integers you specify. RANDINTEGER Syntax: RANDINTEGER(bottom,top) Bottom is the smallest integer RANDINTEGER will return. Top is the largest integer RANDINTEGER will return. RANDINTEGER Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if an argument is text and the name is undefined. Returns #NUM! if top is less than or equal to bottom. Returns #VALUE! if bottom or top is not an integer or if an argument is a defined name of a cell and the cell is blank or contains text.

44

Chapter 3 Monte Carlo Simulation Using RiskSim

RANDINTEGER Example The number of orders a particular customer will place next year is between 7 and 11, with no number more likely than the others. The uncertain number of orders is =RANDINTEGER(7,11). RANDINTEGER Related Function FASTINTEGER: Same as RANDINTEGER without any error checking of the arguments. RANDBETWEEN(bottom,top): Excels function for uniformly distributed integers, without RiskSims capability of setting the seed.

RandNormal
Returns a random value from a normal distribution. This function can model a variety of phenomena where the values follow the familiar bell-shaped curve, and it has wide application in statistical quality control and statistical sampling. RANDNORMAL Syntax: RANDNORMAL(mean,standard_dev) Mean is the arithmetic mean of the normal distribution. Standard_dev is the standard deviation of the normal distribution. RANDNORMAL Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if an argument is text and the name is undefined. Returns #NUM! if standard_dev is negative. Returns #VALUE! if an argument is a defined name of a cell and the cell is blank or contains text. RANDNORMAL Example The total market for a product is approximately normally distributed with mean 60,000 units and standard deviation 5,000 units. The uncertain total market is =RANDNORMAL(60000,5000). RANDNORMAL Related Function FASTNORMAL: Same as RANDNORMAL without any error checking of the arguments. NORMINV(RAND(),mean,standard_dev): Excel's inverse of the normal, or NORMINV(RANDUNIFORM(0,1),mean,standard_dev) to use the RiskSim Seed feature.

3.9 Random Number Generator Functions

45

RandPoisson
Returns a random value from a Poisson distribution. This function can model the uncertain number of occurrences during a specified time interval, for example, the number of arrivals at a service facility during an hour. The possible values of RANDPOISSON are the non-negative integers, 0, 1, 2, ... . RANDPOISSON Syntax: RANDPOISSON(mean) Mean is the mean number of occurrences per unit of time. RANDPOISSON Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if the argument is text and the name is undefined. Returns #NUM! if mean is negative or zero. Returns #VALUE! if mean is a defined name of a cell and the cell is blank or contains text. RANDPOISSON Examples Cars arrive at a toll plaza with a mean rate of 3 cars per minute. The uncertain number of arrivals in a minute is =RANDPOISSON(3). The average value returned by repeated recalculation of RANDPOISSON(3) is 3. A bank teller requires an average of two minutes to serve a customer. The uncertain number of customers served in a minute is =RANDPOISSON(0.5). The average value returned by repeated recalculation of RANDPOISSON(0.5) is 0.5. RANDPOISSON Related Functions FASTPOISSON: Same as RANDPOISSON without any error checking of the arguments. RANDEXPONENTIAL: Describes time between occurrences for a Poisson process.

RandTriangular
Returns a random value from a triangular probability density function. This function can model an uncertain quantity where the most likely value (mode) has the largest probability of occurrence, the minimum and maximum possible values have essentially zero probability of occurrence, and the probability density function is linear between the minimum and the mode and between the mode and the maximum. This function can also model a ramp density function where the minimum equals the mode or the mode equals the maximum. RANDTRIANGULAR Syntax: RANDTRIANGULAR(minimum,most_likely,maximum)

46

Chapter 3 Monte Carlo Simulation Using RiskSim

Minimum is the smallest value RANDTRIANGULAR will return. Most_likely is the most likely value RANDTRIANGULAR will return. Maximum is the largest value RANDTRIANGULAR will return. RANDTRIANGULAR Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if an argument is text and the name is undefined. Returns #NUM! if minimum is greater than or equal to maximum, if most_likely is less than minimum, or if most_likely is greater than maximum. Returns #VALUE! if an argument is a defined name of a cell and the cell is blank or contains text. RANDTRIANGULAR Example The minimum time required to complete a particular task that is part of a large project is 4 hours, the most likely time required is 6 hours, and the maximum time required is 10 hours. The function returning the uncertain time required for the task is entered into a cell: =RANDTRIANGULAR(4,6,10). RANDTRIANGULAR Related Function FASTTRIANGULAR: Same as RANDTRIANGULAR without any error checking of arguments. Figure 3.16 RandTriangular Example Probability Density Function
0.6 0.5 Probability Density, f(x) 0.4 0.3 0.2 0.1 0 0 2 4 6 8 10 Task Time, x, in hours

3.9 Random Number Generator Functions

47

Figure 3.17 RandTriangular Example Cumulative Probability Function

1

Cumulative Probability, P(X<=x

0.8

0.6

0.4

0.2

0 0 2 4 6 8 10 Task Time, x, in hours

RandUniform
Returns a uniformly distributed random value between two values you specify. As a special case, RANDUNIFORM(0,1) is the same as Excel's built-in RAND() function. RANDUNIFORM Syntax: RANDUNIFORM(minimum,maximum) Minimum is the smallest value RANDUNIFORM will return. Maximum is the largest value RANDUNIFORM will return. RANDUNIFORM Remarks Returns #N/A if there are too few or too many arguments. Returns #NAME! if an argument is text and the name is undefined. Returns #NUM! if minimum is greater than or equal to maximum. Returns #VALUE! if an argument is a defined name of a cell and the cell is blank or contains text. RANDUNIFORM Example A corporate planner thinks that the company's product will garner between 10% and 15% of the total market, with all possible percentages equally likely in the specified range. The uncertain market proportion is =RANDUNIFORM(0.10,0.15). RANDUNIFORM Related Function

48

Chapter 3 Monte Carlo Simulation Using RiskSim

FASTUNIFORM: Same as RANDUNIFORM without any error checking of the arguments.

3.10 RISKSIM TECHNICAL DETAILS

RiskSim's random number generator functions are based on a (0,1) uniformly distributed random number function called RandSeed which is not directly accessible by the user. Internally, decimal values for RandSeed are calculated by dividing a uniformly distributed random integer by 2147483647, which is RandSeed's period. Random integers are generated using the well-documented Park-Miller algorithm, where each random integer depends on the previous random integer. When RiskSim starts, the initial integer seed depends on the system clock. Unlike Excel's RAND() function, you can use RiskSim at any time to specify an integer seed, which is used as the previous random integer for the sequence of random numbers generated by the RiskSim functions. In the Risk Simulation dialog box, the "Random number seed" edit box changes the seed only for the RiskSim functions; it does not have any effect on Excel's built-in RAND() function. Each of RiskSim's random number generator functions use RandSeed as a building block. RANDBINOMIAL(trials,probability_s) uses RandSeed as the cumulative probability in Excel's built-in CRITBINOM function. RANDCUMULATIVE(value_cumulative_table) uses the value of RandSeed, R, searches to find the adjacent cumulative probabilities that bracket R, and interpolates on the linear segment of the cumulative distribution to find the corresponding value. RANDDISCRETE(value_discrete_table) compares RandSeed with summed probabilities of the input table until the sum exceeds the RandSeed value, and then returns the previous value from the input table. RANDEXPONENTIAL(lambda) uses the value of RandSeed, R, as follows. If the exponential density function is f(t) = lambda*EXP(-lambda*t), the cumulative is P(T<=t) = 1 - EXP(-lambda*t). Associating R with P(T<=t), the inverse cumulative is t = -LN(1R)/lambda. Since R and 1-R are both uniformly distributed between 0 and 1, RiskSim uses -LN(R)/lambda for the returned value. RANDINTEGER(bottom,top) returns bottom + INT(RandSeed*(top-bottom+1)). RANDNORMAL(mean,standard_dev) uses two RandSeed values in the welldocumented Box-Muller method.7

3.11 Modeling Uncertain Relationships

49

RANDPOISSON(mean) compares RandSeed with cumulative probabilities of Excel's built-in POISSON function until the probability exceeds the RandSeed value, and then returns the previous value. RANDTRIANGULAR(minimum,most_likely,maximum) uses RandSeed once. The triangular density function has two linear segments, so the cumulative distribution has two quadratic segments. The returned value is determined by interpolation on the appropriate quadratic segment. RANDUNIFORM(minimum,maximum) returns minimum + RandSeed*(maximumminimum). RANDUNIFORM(0,1) is equivalent to Excel's built-in RAND() function. RiskSim includes a FAST... version of each of the nine functions, e.g., FASTBINOMIAL, FASTCUMULATIVE, etc. The FAST... functions are identical to the RAND... functions except there is no error checking of arguments.

3.11 MODELING UNCERTAIN RELATIONSHIPS

Base Model, Four Inputs
Price is fixed. The three uncontrollable inputs are independent. Figure 3.18 Four Inputs Influence Chart
Net Cash Flow

Price

Fixed Costs

Units Sold

Unit Variable Cost

50

Chapter 3 Monte Carlo Simulation Using RiskSim

Figure 3.19 Four Inputs Display

1 2 3 4 5 6 7 8 A B Controllable Input Price $29 Uncontrollable Inputs Fixed Costs $12,000 Units Sold 700 Unit Variable Cost $8 Output Variable Net Cash Flow $2,700

Figure 3.20 Four Inputs Formulas

1 2 3 4 5 6 7 8 A Controllable Input Price Uncontrollable Inputs Fixed Costs Units Sold Unit Variable Cost Output Variable Net Cash Flow B 29 12000 700 8 =(B2-B6)*B5-B4

Three Inputs
Price is variable. Units sold depends on price. The two cost inputs are independent. Figure 3.21 Three Inputs Influence Chart
Net Cash Flow

Units Sold

Price

Fixed Costs

Unit Variable Cost

3.11 Modeling Uncertain Relationships

51

Figure 3.22 Three Inputs Display

1 2 3 4 5 6 7 8 9 A B Controllable Input Price $29 Uncontrollable Inputs Fixed Costs $12,000 Unit Variable Cost $8 Intermediate Variable Units Sold 700 Output Variable Net Cash Flow $2,700 C D E Price Units Sold $29 700 $39 550 $49 400 $59 250 Slope Intercept -15 1135

Figure 3.23 Three Inputs Formulas

1 2 3 4 5 6 7 8 9 A Controllable Input Price Uncontrollable Inputs Fixed Costs Unit Variable Cost Intermediate Variable Units Sold Output Variable Net Cash Flow B 29 12000 8 =E8+E7*B2 =(B2-B5)*B7-B4 C D Price 29 39 49 59 E Units Sold 700 550 400 250

Slope =SLOPE(E2:E5,D2:D5) Intercept =INTERCEPT(E2:E5,D2:D5)

Two Inputs
Price is variable. Units sold depends on price. Unit variable cost depends on fixed costs. Figure 3.24 Two Inputs Influence Chart
Net Cash Flow

Units Sold

Unit Variable Cost

Price

Fixed Costs

52

Chapter 3 Monte Carlo Simulation Using RiskSim

Figure 3.25 Two Inputs Display

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 A B Controllable Input Price $29 Uncontrollable Inputs Fixed Costs $12,000 Intermediate Variable Unit Variable Cost $8.00 Units Sold 700 Output Variable Net Cash Flow $2,700 C D Price $29 $39 $49 $59 Slope Intercept E Units Sold 700 550 400 250 -15 1135

Fixed Costs Unit Variable Cost $10,000 $11 $12,000 $8 $15,000 $6 a b c 0.000000166667 -0.005166666667 46

Figure 3.26 Two Inputs Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 A Controllable Input Price Uncontrollable Inputs Fixed Costs Intermediate Variable Unit Variable Cost Units Sold Output Variable Net Cash Flow B 29 12000 =E16*B4^2+E17*B4+E18 =E8+E7*B2 =(B2-B6)*B7-B4 Fixed Costs 10000 12000 15000 a b c Unit Variable Cost 11 8 6 =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2})) =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2})) =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2})) C Price 29 39 49 59 Slope Intercept D Units Sold 700 550 400 250 =SLOPE(E2:E5,D2:D5) =INTERCEPT(E2:E5,D2:D5) E

Four Inputs with Three Uncertainties

Price is variable. Units sold depends on price. Unit variable cost depends on fixed costs. Fixed costs, units sold, and unit variable cost are uncertain.

3.11 Modeling Uncertain Relationships

53

Figure 3.27 Three Uncertainties Influence Chart

Net Cash Flow

Units Sold

Unit Variable Cost

Units Sold Median

Unit Variable Cost Median

Price

Units Sold Uncertainty

Unit Variable Cost Uncertainty

Fixed Costs

Figure 3.28 Three Uncertainties Display

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 A Controllable Input Price Uncontrollable Inputs Fixed Costs Units Sold Uncertainty Unit Variable Cost Uncertainty Intermediate Variable Units Sold Median Units Sold Unit Variable Cost Median Unit Variable Cost Output Variable Net Cash Flow B $29 $12,000 10 $0.10 700 710 $8.00 $8.10 $2,839 C D Price $29 $39 $49 $59 Slope Intercept E Units Sold 700 550 400 250 -15 1135

Fixed Costs Unit Variable Cost $10,000 $11 $12,000 $8 $15,000 $6 a b c 0.000000166667 -0.005166666667 46

54

Chapter 3 Monte Carlo Simulation Using RiskSim

Figure 3.29 Three Uncertainties Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 A Controllable Input Price Uncontrollable Inputs Fixed Costs Units Sold Uncertainty Unit Variable Cost Uncertainty Intermediate Variable Units Sold Median Units Sold Unit Variable Cost Median Unit Variable Cost Output Variable Net Cash Flow B 29 12000 10 0.1 =E8+E7*B2 =B8+B5 =E16*B4^2+E17*B4+E18 =B10+B6 =(B2-B11)*B9-B4 C Price 29 39 49 59 Slope Intercept D Units Sold 700 550 400 250 =SLOPE(E2:E5,D2:D5) =INTERCEPT(E2:E5,D2:D5) E

Fixed Costs 10000 12000 15000 a b c

Unit Variable Cost 11 8 6 =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2})) =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2})) =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2}))

Intermediate Details
Price is variable. Units sold depends on price. Unit variable cost depends on fixed costs. Fixed costs, units sold, and unit variable cost are uncertain. Include revenue, total variable cost, and total costs as intermediate variables. Figure 3.30 Intermediate Details Influence Chart
Net Cash Flow

Revenue

Total Costs

Total Variable Cost

Units Sold

Unit Variable Cost

Units Sold Median

Unit Variable Cost Median

Price

Units Sold Uncertainty

Unit Variable Cost Uncertainty

Fixed Costs

3.11 Modeling Uncertain Relationships

55

Figure 3.31 Intermediate Details Display

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 A Controllable Input Price Uncontrollable Inputs Fixed Costs Units Sold Uncertainty Unit Variable Cost Uncertainty Intermediate Variable Units Sold Median Units Sold Revenue Unit Variable Cost Median Unit Variable Cost Total Variable Cost Total Costs Output Variable Net Cash Flow B $29 $12,000 10 $0.10 700 710 $20,590 $8.00 $8.10 $5,751 $17,751 $2,839 C D Price $29 $39 $49 $59 Slope Intercept E Units Sold 700 550 400 250 -15 1135

Fixed Costs Unit Variable Cost $10,000 $11 $12,000 $8 $15,000 $6 a b c 0.000000166667 -0.005166666667 46

Figure 3.32 Intermediate Details Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 A Controllable Input Price Uncontrollable Inputs Fixed Costs Units Sold Uncertainty Unit Variable Cost Uncertainty Intermediate Variable Units Sold Median Units Sold Revenue Unit Variable Cost Median Unit Variable Cost Total Variable Cost Total Costs Output Variable Net Cash Flow B 29 12000 10 0.1 =E8+E7*B2 =B8+B5 =B9*B2 =E16*B4^2+E17*B4+E18 =B11+B6 =B12*B9 =B4+B13 =B10-B14 C Price 29 39 49 59 Slope Intercept D Units Sold 700 550 400 250 =SLOPE(E2:E5,D2:D5) =INTERCEPT(E2:E5,D2:D5) E

Fixed Costs 10000 12000 15000 a b c

Unit Variable Cost 11 8 6 =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2})) =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2})) =TRANSPOSE(LINEST(E12:E14,D12:D14^{1,2}))

56

Chapter 3 Monte Carlo Simulation Using RiskSim

This page is intentionally mostly blank.

Decision Trees Using TreePlan

TreePlan is a decision tree add-in for Microsoft Excel 97 (and later versions of Excel) for Windows and Macintosh. It was developed by Professor Michael R. Middleton at the University of San Francisco and modified for use at Fuqua (Duke) by Professor James E. Smith.

4.1 TREEPLAN INSTALLATION

All of TreePlans functionality is in a single file, TreePlan.xla. Depending on your preference, there are three ways to install TreePlan. (These instructions also apply to the other Decision ToolPak add-ins: SensIt.xla and RiskSim.xla.)

Occasional Use
If you plan to use TreePlan on an irregular basis, simply use Excels File | Open command to load TreePlan.xla each time you want to use it. You may keep the TreePlan.xla file on a floppy disk, your computers hard drive, or a network server.

Selective Use
You can use Excels Add-In Manager to install TreePlan. First, copy TreePlan.xla to a location on your computers hard drive. Second, if you save TreePlan.xla in the Excel or Office Library subdirectory, go to the third step. Otherwise, run Excel, choose Tools | Add-Ins; in the Add-Ins dialog box, click the Browse button, use the Browse dialog box to specify the location of TreePlan.xla, and click OK. Third, in the Add-Ins dialog box, note that TreePlan is now listed with a check mark, indicating that its menu command will appear in Excel, and click OK. If you plan to not use TreePlan and you want to free up main memory, uncheck the box for TreePlan in the Add-In Manager. When you do want to use TreePlan, choose Tools | Add-Ins and check TreePlans box.

58

Chapter 4 Decision Trees Using TreePlan

To remove TreePlan from the Add-In Manager, use Windows Explorer or another file manager to delete TreePlan.xla from the Library subdirectory or from the location you specified when you used the Add-In Managers Browse command. The next time you start Excel and choose Tools | Add-Ins, a dialog box will state Cannot find add-in treeplan.xla. Delete from list? Click Yes.

Steady Use
If you want TreePlans options immediately available each time you run Excel, use Windows Explorer or another file manager to save TreePlan.xla in the Excel XLStart directory. Alternatively, in Excel you can use Tools | Options | General to specify an alternate startup file location and use a file manager to save TreePlan.xla there. When you start Excel, it tries to open all files in the XLStart directory and in the alternate startup file location. For additional information visit TreePlan FAQ at www.treeplan.com. After opening TreePlan.xla in Excel, the command "Decision Tree" appears at the bottom of the Tools menu (or, if you have a customized main menu, at the bottom of the sixth main menu item).

4.2 BUILDING A DECISION TREE IN TREEPLAN

You can start TreePlan either by choosing Tools | Decision Tree from the menu bar or by pressing Ctrl+t (hold down the Ctrl key and press t). If the worksheet doesn't have a decision tree, TreePlan prompts you with a dialog box with three options; choose New Tree to begin a new tree. TreePlan draws a default initial decision tree with its upper left corner at the selected cell. For example, the figure below shows the initial tree when $B$2 is selected. (Note that TreePlan writes over existing values in the spreadsheet: begin your tree to the right of the area where your data is stored, and do not subsequently add or delete rows or columns in the tree-diagram area.) In Excel 5 and 95 a terminal node is represented by a triangle instead of a vertical bar. Figure 4.1 TreePlan Initial Default Decision Tree
A 1 2 3 4 5 6 7 8 9 10 11 B C D E F G H I

Decision 1 0 0 1 0 Decision 2 0 0 0 0

4.2 Building a Decision Tree in TreePlan

59

Build up a tree by adding or modifying branches or nodes in the default tree. To change the branch labels or probabilities, click on the cell containing the label or probability and type the new label or probability. To modify the structure of the tree (e.g., add or delete branches or nodes in the tree), select the node or the cell containing the node in the tree to modify, and choose Tools | Decision Tree or press Ctrl+t. TreePlan will then present a dialog box showing the available commands. For example, to add an event node to the top branch of the tree shown above, select the square cell (cell G4) next to the vertical line at the end of a terminal branch and press Ctrl+t.. TreePlan then presents this dialog box. Figure 4.2 TreePlan Terminal Dialog Box

To add an event node to the branch, we change the selected terminal node to an event node by selecting Change to event node in the dialog box, selecting the number of branches (here two), and pressing OK. TreePlan then redraws the tree with a chance node in place of the terminal node. Figure 4.3
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 B C D E F G H I 0.5 Event 3 0 Decision 1 0 0 0 0.5 Event 4 0 1 0 Decision 2 0 0 0 0 0 0 J K L M

The dialog boxes presented by TreePlan vary depending on what you have selected when you choose Tools | Decision Tree or press Ctrl+t. The dialog box shown below is presented when you press Ctrl+t with an event node selected; a similar dialog box is

60

Chapter 4 Decision Trees Using TreePlan

presented when you select a decision node. If you want to add a branch to the selected node, choose Add branch and press OK. If you want to insert a decision or event node before the selected node, choose Insert decision or Insert event and press OK. To get a description of the available commands, click on the Help button. Figure 4.4

The Copy subtree command is particularly useful when building large trees. If two or more parts of the tree are similar, you can copy and paste "subtrees" rather than building up each part separately. To copy a subtree, select the node at the root of the subtree and choose Copy subtree. This tells TreePlan to copy the selected node and everything to the right of it in the tree. To paste this subtree, select a terminal node and choose Paste subtree. TreePlan then duplicates the specified subtree at the selected terminal node. Since TreePlan decision trees are built directly in Excel, you can use Excel's commands to format your tree. For example, you can use bold or italic fonts for branch labels: select the cells you want to format and change them using Excel's formatting commands. To help you, TreePlan provides a Select dialog box that appears when you choose Tools Decision Tree or press Ctrl+t without a node selected. You can also bring up this dialog box by pressing the Select button on the Node dialog box. From here, you can select all items of a particular type in the tree. For example, if you choose Probabilities and press OK, TreePlan selects all cells containing probabilities in the tree. You can then format all of the probabilities simultaneously using Excel's formatting commands. (Because of limitations in Excel, the Select dialog box will not be available when working with very large trees.)

4.3 ANATOMY OF A TREEPLAN DECISION TREE

An example of a TreePlan decision tree is shown below. In the example, a firm must decide (1) whether to prepare a proposal for a possible contract and (2) which method to use to satisfy the contract. The tree consists of decision nodes, event nodes and terminal nodes connected by branches. Each branch is surrounded by cells containing formulas,

4.3 Anatomy of a TreePlan Decision Tree

61

cell references, or labels pertaining to that branch. You may edit the labels, probabilities, and partial cash flows associated with each branch. The partial cash flows are the amount the firm "gets paid" to go down that branch. Here, the firm pays $50,000 if it decides to prepare the proposal, receives $250,000 up front if awarded the contract, spends $50,000 to try the electronic method, and spends $120,000 on the mechanical method if the electronic method fails. Figure 4.5
DECISION NODES: TreePlan formula for which alternative is optimal. PROBABILITIES: Enter numbers or formulas in these cells. TERMINAL VALUES: TreePlan formula for sum of partial cash flows along path.

Use mechanical method $80,000 PARTIAL CASH FLOWS: Enter numbers or formulas in these cells. 0.5 Awarded contract 2 $250,000 $90,000 Try electronic method Prepare proposal -$50,000 $20,000 -$50,000 ROLLBACK EVs: TreePlan formula for expected value at this point in the tree. $90,000 -$120,000 $80,000 0.5 Electronic success $150,000 $0 0.5 Electronic failure $30,000 -$120,000 BRANCH LABELS: Type text in these cells. 0.5 Not awarded contract $0 -$50,000 TERMINAL NODES Don't prepare proposal $0 $0 $0 $30,000 $150,000

1 $20,000

EVENT NODES -$50,000

The trees are "solved" using formulas embedded in the spreadsheet. The terminal values sum all the partial cash flows along the path leading to that terminal node. The tree is then "rolled back" by computing expected values at event nodes and by maximizing at decision nodes; the rollback EVs appear next to each node and show the expected value at that point in the tree. The numbers in the decision nodes indicate which alternative is optimal for that decision. In the example, the "1" in the first decision node indicates that it is optimal to prepare the proposal, and the "2" in the second decision node indicates the firm should try the electronic method because that alternative leads to a higher expected value, $90,000, than the mechanical method, $80,000. TreePlan has a few options that control the way calculations are done in the tree. To select these options, press the Options button in any of TreePlan's dialog boxes. The first choice is whether to Use Expected Values or Use Exponential Utility Function for computing certainty equivalents. The default is to rollback the tree using expected values. If you choose to use exponential utilities, TreePlan will compute utilities of endpoint cash flows at the terminal nodes and compute expected utilities instead of expected values at event nodes. Expected utilities are calculated in the cell below the certainty equivalents. You may also choose to Maximize (profits) or Minimize (costs) at decision nodes; the default is to maximize profits. If you choose to minimize costs instead, the cash flows are

62

Chapter 4 Decision Trees Using TreePlan

interpreted as costs, and decisions are made by choosing the minimum expected value or certainty equivalent rather than the maximum. See the Help file for details on these options.

4.4 STEP-BY-STEP TREEPLAN TUTORIAL

A decision tree can be used as a model for a sequential decision problems under uncertainty. A decision tree describes graphically the decisions to be made, the events that may occur, and the outcomes associated with combinations of decisions and events. Probabilities are assigned to the events, and values are determined for each outcome. A major goal of the analysis is to determine the best decisions. Decision tree models include such concepts as nodes, branches, terminal values, strategy, payoff distribution, certainty equivalent, and the rollback method. The following problem illustrates the basic concepts.

DriveTek Problem
DriveTek Research Institute discovers that a computer company wants a new tape drive for a proposed new computer system. Since the computer company does not have research people available to develop the new drive, it will subcontract the development to an independent research firm. The computer company has offered a fee of $250,000 for the best proposal for developing the new tape drive. The contract will go to the firm with the best technical plan and the highest reputation for technical competence. DriveTek Research Institute wants to enter the competition. Management estimates a cost of $50,000 to prepare a proposal with a fifty-fifty chance of winning the contract. However, DriveTek's engineers are not sure about how they will develop the tape drive if they are awarded the contract. Three alternative approaches can be tried. The first approach is a mechanical method with a cost of $120,000, and the engineers are certain they can develop a successful model with this approach. A second approach involves electronic components. The engineers estimate that the electronic approach will cost only $50,000 to develop a model of the tape drive, but with only a 50 percent chance of satisfactory results. A third approach uses magnetic components; this costs $80,000, with a 70 percent chance of success. DriveTek Research can work on only one approach at a time and has time to try only two approaches. If it tries either the magnetic or electronic method and the attempt fails, the second choice must be the mechanical method to guarantee a successful model. The management of DriveTek Research needs help in incorporating this information into a decision to proceed or not.

4.4 Step-by-Step TreePlan Tutorial

63

[Source: The tape drive example is adapted from Spurr and Bonini, Statistical Analysis for Business Decisions, Irwin.]

Nodes and Branches

Decision trees have three kinds of nodes and two kinds of branches. A decision node is a point where a choice must be made; it is shown as a square. The branches extending from a decision node are decision branches, each branch representing one of the possible alternatives or courses of action available at that point. The set of alternatives must be mutually exclusive (if one is chosen, the others cannot be chosen) and collectively exhaustive (all possible alternatives must be included in the set). There are two major decisions in the DriveTek problem. First, the company must decide whether or not to prepare a proposal. Second, if it prepares a proposal and is awarded the contract, it must decide which of the three approaches to try to satisfy the contract. An event node is a point where uncertainty is resolved (a point where the decision maker learns about the occurrence of an event). An event node, sometimes called a "chance node," is shown as a circle. The event set consists of the event branches extending from an event node, each branch representing one of the possible events that may occur at that point. The set of events must be mutually exclusive (if one occurs, the others cannot occur) and collectively exhaustive (all possible events must be included in the set). Each event is assigned a subjective probability; the sum of probabilities for the events in a set must equal one. The three sources of uncertainty in the DriveTek problem are: whether it is awarded the contract or not, whether the electronic approach succeeds or fails, and whether the magnetic approach succeeds or fails. In general, decision nodes and branches represent the controllable factors in a decision problem; event nodes and branches represent uncontrollable factors. Decision nodes and event nodes are arranged in order of subjective chronology. For example, the position of an event node corresponds to the time when the decision maker learns the outcome of the event (not necessarily when the event occurs). The third kind of node is a terminal node, representing the final result of a combination of decisions and events. Terminal nodes are the endpoints of a decision tree, shown as the end of a branch on hand-drawn diagrams and as a triangle on computer-generated diagrams. The following table shows the three kinds of nodes and two kinds of branches used to represent a decision tree.

64

Chapter 4 Decision Trees Using TreePlan

Figure 4.6 Nodes and Symbols Type of Node Written Symbol Decision square Event circle Terminal endpoint

Computer Symbol square circle triangle or bar

Node Successor decision branches event branches terminal value

Terminal Values
Each terminal node has an associated terminal value, sometimes called a payoff value, outcome value, or endpoint value. Each terminal value measures the result of a scenario: the sequence of decisions and events on a unique path leading from the initial decision node to a specific terminal node. To determine the terminal value, one approach assigns a cash flow value to each decision branch and event branch and then sum the cash flow values on the branches leading to a terminal node to determine the terminal value. In the DriveTek problem, there are distinct cash flows associated with many of the decision and event branches. Some problems require a more elaborate value model to determine the terminal values. The following diagram shows the arrangement of branch names, probabilities, and cash flow values on an unsolved tree.

4.4 Step-by-Step TreePlan Tutorial

65

Figure 4.7
Use mechanical method -$120,000 0.5 Electronic success 0.5 Awarded contract $250,000 Try electronic method -$50,000 $0 0.5 Electronic failure -$120,000 0.7 Magnetic success Prepare proposal Try magnetic method -$50,000 -$80,000 0.3 Magnetic failure -$120,000 0.5 Not awarded contract $0 $0

Don't prepare proposal $0

To build the decision tree, you use TreePlans dialog boxes to develop the structure. You enter a branch name, branch cash flow, and branch probability (for an event) in the cells above and below the left side of each branch. As you build the tree diagram, TreePlan enters formulas in other cells.

Building the Tree Diagram

1. 2. Start with a new worksheet. (If no workbook is open, choose File | New. If a workbook is open, choose Insert | Worksheet.) Select cell A1. From the Tools menu, choose Decision Tree. In the TreePlan New dialog box, click the New Tree button. A decision node with two branches appears.

66

Chapter 4 Decision Trees Using TreePlan

Figure 4.8

Figure 4.9
A 1 2 3 4 5 6 7 8 9 B C D Decision 1 0 0 1 0 Decision 2 0 0 0 0 E F G

3.

Do not type the quotation marks in the following instructions. Select cell D2, and enter Prepare proposal. Select cell D4, and enter 50000. Select cell D7, and enter Don't prepare proposal.

Figure 4.10
A 1 2 3 4 5 6 7 8 9 B C D E F G

Prepare proposal -50000 -50000 2 0 Don't prepare proposal 0 0 0 -50000

4.

Select cell F3. From the Tools menu, choose Decision Tree. In the TreePlan Terminal dialog box, select Change To Event Node, select Two Branches, and click OK. The tree is redrawn.

4.4 Step-by-Step TreePlan Tutorial

67

Figure 4.11

Figure 4.12
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 B C D E F G H 0.5 Event 3 -50000 Prepare proposal -50000 -50000 0 0.5 Event 4 -50000 2 0 Don't prepare proposal 0 0 0 0 -50000 -50000 I J K

5.

Select cell H2, and enter Awarded contract. Select cell H4, and enter 250000. Select cell H7, and enter Not awarded contract.

68

Chapter 4 Decision Trees Using TreePlan

Figure 4.13
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 B C D E F G I 0.5 Awarded contract 250000 200000 H J K

200000 Prepare proposal -50000 75000

0.5 Not awarded contract -50000 0 -50000

1 75000 Don't prepare proposal

0 0 0

6.

Select cell J3. From the Tools menu, choose Decision Tree. In the TreePlan Terminal dialog box, select Change To Decision Node, select Three Branches, and click OK. The tree is redrawn.

Figure 4.14
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 B C D E F G H I J K L Decision 5 200000 0 0.5 Awarded contract 1 250000 200000 0 200000 200000 M N O

Decision 6 200000

Prepare proposal -50000 75000 0.5 Not awarded contract 1 75000 0 -50000

Decision 7 200000 0 200000

-50000

Don't prepare proposal 0 0 0

7.

Select cell L2, and enter Use mechanical method. Select cell L4, and enter 120000. Select cell L7, and enter Try electronic method. Select cell L9, and

4.4 Step-by-Step TreePlan Tutorial

69

enter 50000. Select cell L12, and enter Try magnetic method. Select cell L14, and enter 80000. Figure 4.15
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 B C D E F G H I J K L M N O

Use mechanical method 80000 -120000 0.5 Awarded contract 2 250000 150000 -50000 150000 80000

Try electronic method 150000

Prepare proposal -50000 50000 0.5 Not awarded contract 1 50000 0 -50000

Try magnetic method 120000 -80000 120000

-50000

Don't prepare proposal 0 0 0

8.

Select cell N8. From the Tools menu, choose Decision Tree. In the TreePlan Terminal dialog box, select Change To Event Node, select Two Branches, and click OK. The tree is redrawn.

70

Chapter 4 Decision Trees Using TreePlan

Figure 4.16
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 B C D E F G H I J K L M N O P Q R S Use mechanical method 80000 -120000 80000 0.5 Event 8 0.5 Awarded contract 2 250000 150000 -50000 150000 0.5 Event 9 150000 0 Prepare proposal -50000 50000 Try magnetic method 120000 -80000 0.5 Not awarded contract -50000 0 -50000 120000 150000 150000 Try electronic method 0 150000

1 50000

Don't prepare proposal 0 0 0

9.

Select cell P7, and enter Electronic success. Select cell P12, and enter Electronic failure. Select cell P14, and enter 120000.

4.4 Step-by-Step TreePlan Tutorial

71

Figure 4.17
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 B C D E F G H I J K L M N O P Q R S Use mechanical method 80000 -120000 80000 0.5 Electronic success 0.5 Awarded contract 3 250000 120000 -50000 90000 0.5 Electronic failure 30000 -120000 Prepare proposal -50000 35000 Try magnetic method 120000 -80000 0.5 Not awarded contract -50000 0 -50000 120000 30000 150000 Try electronic method 0 150000

1 35000

Don't prepare proposal 0 0 0

10.

Select cell N18. From the Tools menu, choose Decision Tree. In the TreePlan Terminal dialog box, select Change To Event Node, select Two Branches, and click OK. The tree is redrawn.

72

Chapter 4 Decision Trees Using TreePlan

Figure 4.18
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 B C D E F G H I J K L M N O P Q R S Use mechanical method 80000 -120000 80000 0.5 Electronic success 150000 0.5 Awarded contract 3 250000 120000 Try electronic method -50000 90000 0 150000

0.5 Electronic failure 30000 -120000 0.5 Event 10 30000

Prepare proposal Try magnetic method -50000 35000 -80000 120000 0.5 Event 11 0 120000

120000

120000 0 1 35000 0.5 Not awarded contract -50000 0 -50000 120000

Don't prepare proposal 0 0 0

11.

Select cell P16, and enter .7. Select cell P17, and enter Magnetic success. Select cell P21, and enter .3. Select cell P22, and enter Magnetic failure. Select cell P24, and enter 120000.

4.4 Step-by-Step TreePlan Tutorial

73

Figure 4.19
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 B C D E F G H I J K L M N O P Q R S Use mechanical method 80000 -120000 80000 0.5 Electronic success 150000 0.5 Awarded contract 2 250000 90000 Try electronic method -50000 90000 0 150000

0.5 Electronic failure 30000 -120000 30000

0.7 Magnetic success Prepare proposal Try magnetic method -50000 20000 -80000 84000 0.3 Magnetic failure 0 -120000 1 20000 0.5 Not awarded contract -50000 0 -50000 0 0 120000 120000

Don't prepare proposal 0 0 0

12.

Double-click the sheet tab (or right-click the sheet tab and choose Rename from the shortcut menu), and enter Original. Save the workbook.

Interpreting the Results

The $30,000 terminal value on the far right of the diagram in cell S13 is associated with the following scenario: Figure 4.20 Branch Type Decision Event Decision Event

Branch Name Prepare proposal Awarded contract Try electronic method Electronic failure (Use mechanical method) Terminal value

Cash Flow $50,000 $250,000 $50,000 $120,000 $30,000

TreePlan put the formula =SUM(P14,L11,H12,D20) into cell S13 for determining the terminal value.

74

Chapter 4 Decision Trees Using TreePlan

Other formulas, called rollback formulas, are in cells below and to the left of each node. These formulas are used to determine the optimal choice at each decision node. In cell B26, a formula displays 1, indicating that the first branch is the optimal choice. Thus, the initial choice is to prepare the proposal. In cell J11, a formula displays 2, indicating that the second branch (numbered 1, 2, and 3, from top to bottom) is the optimal choice. If awarded the contract, DriveTek should try the electronic method. A subsequent chapter provides more details about interpretation.

Formatting the Tree Diagram

The following steps show how to use TreePlan and Excel features to format the tree diagram. You may choose to use other formats for your own tree diagrams. 13. From the Edit menu, choose Move or Copy Sheet (or right-click the sheet tab and choose Move Or Copy from the shortcut menu). In the lower left corner of the Move Or Copy dialog box, check the Create A Copy box, and click OK.

14.On sheet Original (2), select cell H9. From the Tools menu, choose Decision Tree. In the TreePlan Select dialog box, verify that the option button for Cells with Probabilities is selected, and click OK. With all probability cells selected, click the Align Left button. Figure 4.21

15.

Select cell H12. From the Tools menu, choose Decision Tree. In the TreePlan Select dialog box, verify that the option button for Cells with Partial Cash Flows is selected, and click OK. With all partial cash flow cells selected, click the Align Left button. With those cells still selected, choose Format | Cells. In the Format Cells dialog box, click the Number tab. In the Category list box, choose

4.4 Step-by-Step TreePlan Tutorial

75

Currency; type 0 (zero) for Decimal Places; select $ in the Symbol list box; select -$1,234 for Negative Numbers. Click OK. Figure 4.22

16.

Select cell I12. From the Tools menu, choose Decision Tree. In the TreePlan Select dialog box, verify that the option button for Cells with Rollback EVs/CEs is selected, and click OK. With all rollback cells selected, choose Format | Cells. Repeat the Currency formatting of step 16 above. Select cell S3. From the Tools menu, choose Decision Tree. In the TreePlan Select dialog box, verify that the option button for Cells with Terminal Values is selected, and click OK. With all terminal value cells selected, choose Format | Cells. Repeat the Currency formatting of step 16 above.

17.

76

Chapter 4 Decision Trees Using TreePlan

Figure 4.23
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 B C D E F G H I J K L M N O P Q R S Use mechanical method $80,000 -$120,000 $80,000 0.5 Electronic success $150,000 0.5 Awarded contract 2 $250,000 $90,000 Try electronic method -$50,000 $90,000 $0 $150,000

0.5 Electronic failure $30,000 -$120,000 $30,000

0.7 Magnetic success Prepare proposal Try magnetic method -$50,000 $20,000 -$80,000 $84,000 0.3 Magnetic failure $0 -$120,000 1 $20,000 0.5 Not awarded contract -$50,000 $0 -$50,000 $0 $0 $120,000 $120,000

Don't prepare proposal $0 $0 $0

18.

Double-click the Original (2) sheet tab (or right-click the sheet tab and choose Rename from the shortcut menu), and enter Formatted. Save the workbook.

Displaying Model Inputs

When you build a decision tree model, you may want to discuss the model and its assumptions with co-workers or a client. For such communication it may be preferable to hide the results of formulas that show rollback values and decision node choices. The following steps show how to display only the model inputs. 19. From the Edit menu, choose Move or Copy Sheet (or right-click the sheet tab and choose Move Or Copy from the shortcut menu). In the lower left corner of the Move Or Copy dialog box, check the Create A Copy box, and click OK. On sheet Formatted (2), select cell B1. From the Tools menu, choose Decision Tree. In the TreePlan Select dialog box, verify that the option button for Columns with Nodes is selected, and click OK. With all node columns selected, choose Format | Cells | Number. In the Category list box, select Custom. Select the entry in the Type edit box, and type ;;; (three semicolons). Click OK.

20.

4.4 Step-by-Step TreePlan Tutorial

77

Figure 4.24

Explanation: A custom number format has four sections of format codes. The sections are separated by semicolons, and they define the formats for positive numbers, negative numbers, zero values, and text, in that order. When you specify three semicolons without format codes, Excel does not display positive numbers, negative numbers, zero values, or text. The formula remains in the cell, but its result is not displayed. Later, if you want to display the result, you can change the format without having to enter the formula again. Editing an existing format does not delete it. All formats are saved with the workbook unless you explicitly delete a format. 21. Select cell A27. From the Tools menu, choose Decision Tree. In the TreePlan Select dialog box, verify that the option button for Cells with Rollback EVs/CEs is selected, and click OK. With all rollback values selected, choose Format | Cells | Number. In the Category list box, select Custom. Scroll to the bottom of the Type list box, and select the three-semicolon entry. Click OK. Double-click the Formatted (2) sheet tab (or right-click the sheet tab and choose Rename from the shortcut menu), and enter Model Inputs. Save the workbook.

22.

78

Chapter 4 Decision Trees Using TreePlan

Printing the Tree Diagram

23. 24. In the Name Box list box, select TreeDiagram (or select cells A1:S34). To print the tree diagram from Excel, with the tree diagram range selected choose File | Print Area | Set Print Area. Choose File | Page Setup. In the Page Setup dialog box, click the Page tab; for Orientation click the option button for Landscape, and for Scaling click the option button for Fit To 1 Page Wide By 1 Page Tall. Click the Header/Footer tab; in the Header list box select None, and in the Footer list box select None (or select other appropriate headers and footers). Click the Sheet tab; clear the check box for Gridlines, and clear the check box for Row And Column Headings. Click OK. Choose File | Print and click OK. To print the tree diagram from Word, clear the check boxes for Gridlines and for Row And Column Headings on Excels Page Setup dialog box Sheet tab. Select the tree diagram range. Hold down the Shift key and from the Edit menu choose Copy Picture. In the Copy Picture dialog box, click the option button As Shown When Printed, and click OK. In Word select the location where you want to paste the tree diagram and choose Edit | Paste.

25.

Figure 4.25
Use mechanical method $80,000 -$120,000 0.5 Electronic success $150,000 0.5 Awarded contract $250,000 Try electronic method -$50,000 $0 0.5 Electronic failure $30,000 -$120,000 0.7 Magnetic success Prepare proposal Try magnetic method -$50,000 -$80,000 0.3 Magnetic failure $0 -$120,000 0.5 Not awarded contract -$50,000 $0 $0 $120,000

Don't prepare proposal $0 $0

4.5 Decision Tree Solution

79

Alternative Model
If you want to emphasize that the time constraint forces DriveTek to use the mechanical approach if they try either of the uncertain approaches and experience a failure, you can change the terminal nodes in cells R13 and R23 to decision nodes, each with a single branch. Figure 4.26
Use mechanical method $80,000 -$120,000 0.5 Electronic success $150,000 0.5 Awarded contract $250,000 Try electronic method -$50,000 $0 0.5 Electronic failure $0 0.7 Magnetic success Prepare proposal Try magnetic method -$50,000 -$80,000 0.3 Magnetic failure $0 0.5 Not awarded contract -$50,000 $0 Use mechanical method $0 -$120,000 $0 $120,000

Use mechanical method $30,000 -$120,000

Don't prepare proposal $0 $0

4.5 DECISION TREE SOLUTION

Strategy
A strategy specifies an initial choice and any subsequent choices to be made by the decision maker. The subsequent choices usually depend upon events. The specification of a strategy must be comprehensive; if the decision maker gives the strategy to a colleague, the colleague must know exactly which choice to make at each decision node. Most decision problems have many possible strategies, and a goal of the analysis is to determine the optimal strategy, taking into account the decision maker's risk attitude. There are four strategies in the DriveTek problem. One of the strategies is: Prepare the proposal; if not awarded the contract, stop; if awarded the contract, try the magnetic method; if the magnetic method is successful, stop; if the magnetic method fails, use the mechanical method. The four strategies will be discussed in detail below.

80

Chapter 4 Decision Trees Using TreePlan

Payoff Distribution
Each strategy has an associated payoff distribution, sometimes called a risk profile. The payoff distribution of a particular strategy is a probability distribution showing the probability of obtaining each terminal value associated with a particular strategy. In decision tree models, the payoff distribution can be shown as a list of possible payoff values, x, and the discrete probability of obtaining each value, P(X=x), where X represents the uncertain terminal value associated with a strategy. Since a strategy specifies a choice at each decision node, the uncertainty about terminal values depends only on the occurrence of events. The probability of obtaining a specific terminal value equals the product of the probabilities on the event branches on the path leading to the terminal node.

DriveTek Strategies
In this section each strategy of the DriveTek problem is described by a shorthand statement and a more detailed statement. The possible branches following a specific strategy are shown in decision tree form, and the payoff distribution is shown in a table with an explanation of the probability calculations.

Strategy 1 (Mechanical): Prepare; if awarded, use mechanical. Details: Prepare the proposal; if not awarded the contract, stop (payoff = -$50,000); if awarded the contract, use the mechanical method (payoff = $80,000).

4.5 Decision Tree Solution

81

Figure 4.27
Use mechanical method $80,000

0.5 Electronic success $150,000 0.5 Awarded contract Try electronic method 0.5 Electronic failure $30,000

0.7 Magnetic success Prepare proposal Try magnetic method 0.3 Magnetic failure $0 $120,000

0.5 Not awarded contract -$50,000

Don't prepare proposal $0

Figure 4.28 Value, x $80,000 -$50,000 Probability P(X=x) 0.50 0.50 1.00

Strategy 2 (Electronic): Prepare; if awarded, try electronic. Details: Prepare the proposal; if not awarded the contract, stop (payoff = -$50,000); if awarded the contract, try the electronic method; if the electronic method is successful, stop (payoff = $150,000); if the electronic method fails, use the mechanical method (payoff = $30,000).

82

Chapter 4 Decision Trees Using TreePlan

Figure 4.29
Use mechanical method $80,000

0.5 Electronic success $150,000 0.5 Awarded contract Try electronic method 0.5 Electronic failure $30,000

0.7 Magnetic success Prepare proposal Try magnetic method 0.3 Magnetic failure $0 $120,000

0.5 Not awarded contract -$50,000

Don't prepare proposal $0

Figure 4.30 Value, x $150,000 $30,000 -$50,000 Probability P(X=x) 0.25 0.25 0.50 1.00

= 0.5 * 0.5 = 0.5 * 0.5

Strategy 3 (Magnetic): Prepare; if awarded, try magnetic. Details: Prepare the proposal; if not awarded the contract, stop (payoff = -$50,000); if awarded the contract, try the magnetic method; if the magnetic method is successful, stop (payoff = $120,000); if the magnetic method fails, use the mechanical method (payoff = $0).

4.5 Decision Tree Solution

83

Figure 4.31
Use mechanical method $80,000

0.5 Electronic success $150,000 0.5 Awarded contract Try electronic method 0.5 Electronic failure $30,000

0.7 Magnetic success Prepare proposal Try magnetic method 0.3 Magnetic failure $0 $120,000

0.5 Not awarded contract -$50,000

Don't prepare proposal $0

Figure 4.32 Value, x $120,000 $0 -$50,000 Probability P(X=x) 0.35 0.15 0.50 1.00

= 0.5 * 0.7 = 0.5 * 0.3

Strategy 4 (Don't): Don't. Details: Don't prepare the proposal (payoff = $0).

84

Chapter 4 Decision Trees Using TreePlan

Figure 4.33
Use mechanical method $80,000

0.5 Electronic success $150,000 0.5 Awarded contract Try electronic method 0.5 Electronic failure $30,000

0.7 Magnetic success Prepare proposal Try magnetic method 0.3 Magnetic failure $0 $120,000

0.5 Not awarded contract -$50,000

Don't prepare proposal $0

Figure 4.34 Value, x $0 Probability P(X=x) 1.00 1.00

Strategy Choice
Since each strategy can be characterized completely by its payoff distribution, selecting the best strategy becomes a problem of choosing the best payoff distribution. One approach is to make a choice by direct comparison of the payoff distributions.

4.5 Decision Tree Solution

85

Figure 4.35 Strategy 1 (Mechanical) Probability Value, x P(X=x) $80,000 0.50 -$50,000 0.50 1.00

Strategy 2 (Electronic) Probability Value, x P(X=x) $150,000 0.25 $30,000 0.25 -$50,000 0.50 1.00

Strategy 3 (Magnetic) Probability Value, x P(X=x) $120,000 0.35 $0 0.15 -$50,000 0.50 1.00

Strategy 4 (Don't) Probability Value, x P(X=x) $0 1.00 1.00

Another approach for making choices involves certainty equivalents.

Certainty Equivalent
A certainty equivalent is a certain payoff value which is equivalent, for the decision maker, to a particular payoff distribution. If the decision maker can determine his or her certainty equivalent for the payoff distribution of each strategy, then the optimal strategy is the one with the highest certainty equivalent. The certainty equivalent is the minimum selling price for a payoff distribution; it depends on the decision maker's personal attitude toward risk. A decision maker may be risk preferring, risk neutral, or risk avoiding. If the terminal values are not regarded as extreme (relative to the decision maker's total assets), if the decision maker will encounter other decision problems with similar payoffs, and if the decision maker has the attitude that he or she will "win some and lose some," then the decision maker's attitude toward risk may be described as risk neutral. If the decision maker is risk neutral, the expected value is the appropriate certainty equivalent for choosing among the strategies. Thus, for a risk neutral decision maker, the optimal strategy is the one with the highest expected value. The expected value of a payoff distribution is calculated by multiplying each terminal value by its probability and summing the products. The expected value calculations for each of the four strategies of the DriveTek problem are shown below.

86

Chapter 4 Decision Trees Using TreePlan

Figure 4.36 Strategy 1 (Mechanical) Probability Value, x P(X=x) $80,000 0.50 -$50,000 0.50

x * P(X=x) $40,000 -$25,000 $15,000

Strategy 2 (Electronic) Probability Value, x P(X=x) $150,000 0.25 $30,000 0.25 -$50,000 0.50

x * P(X=x) $37,500 7,500 -$25,000 $20,000

Strategy 3 (Magnetic) Probability Value, x P(X=x) $120,000 0.35 $0 0.15 -$50,000 0.50

x * P(X=x) $42,000 $0 -$25,000 $17,000

Strategy 4 (Don't) Value, x $0 Probability P(X=x) 1.00 x * P(X=x) $0 $0

The four strategies of the DriveTek problem have expected values of $15,000, $20,000, $17,000, and $0. Strategy 2 (Electronic) is the optimal strategy with expected value $20,000. A risk neutral decision maker's choice is based on the expected value. However, note that if strategy 2 (Electronic) is chosen, the decision maker does not receive $20,000. The actual payoff will be $150,000, $30,000, or -$50,000, with probabilities shown in the payoff distribution.

4.5 Decision Tree Solution

87

Rollback Method
If we have a method for determining certainty equivalents (expected values for a risk neutral decision maker), we don't need to examine every possible strategy explicitly. Instead, the method known as rollback determines the single best strategy. The rollback algorithm, sometimes called backward induction or "average out and fold back," starts at the terminal nodes of the tree and works backward to the initial decision node, determining the certainty equivalent rollback values for each node. Rollback values are determined as follows: At a terminal node, the rollback value equals the terminal value. At an event node, the rollback value for a risk neutral decision maker is determined using expected value; the branch probability is multiplied times the successor rollback value, and the products are summed. At a decision node, the rollback value is set equal to the highest rollback value on the immediate successor nodes.

In TreePlan tree diagrams the rollback values are located to the left and below each decision, event, and terminal node. Terminal values and rollback values for the DriveTek problem are shown below. Figure 4.37
Use mechanical method $80,000 $80,000 0.5 Electronic success $150,000 0.5 Awarded contract $90,000 Try electronic method $90,000 $150,000 0.5 Electronic failure $30,000 $30,000 0.7 Magnetic success Prepare proposal Try magnetic method $20,000 $84,000 0.3 Magnetic failure $0 $0 0.5 Not awarded contract -$50,000 -$50,000 $120,000 $120,000

$20,000

Don't prepare proposal $0 $0

88

Chapter 4 Decision Trees Using TreePlan

Optimal Strategy
After the rollback method has determined certainty equivalents for each node, the optimal strategy can be identified by working forward through the tree. At the initial decision node, the $20,000 rollback value equals the rollback value of the "Prepare proposal" branch, indicating the alternative that should be chosen. DriveTek will either be awarded the contract or not; there is a subsequent decision only if DriveTek obtains the contract. (In a more complicated decision tree, the optimal strategy must include decision choices for all decision nodes that might be encountered.) At the decision node following "Awarded contract," the $90,000 rollback value equals the rollback value of the "Try electronic method" branch, indicating the alternative that should be chosen. Subsequently, if the electronic method fails, DriveTek must use the mechanical method to satisfy the contract. Cell B26 has the formula =IF(A27=E20,1,IF(A27=E34,2)) which displays 1, indicating that the first branch is the optimal choice. Thus, the initial choice is to prepare the proposal. Cell J11 has the formula =IF(I12=M4,1,IF(I12=M11,2,IF(I12=M21,3))) which displays 2, indicating that the second branch (numbered 1, 2, and 3, from top to bottom) is the optimal choice. If awarded the contract, DriveTek should try the electronic method. The pairs of rollback values at the relevant decision nodes ($20,000 and $90,000) and the preferred decision branches are shown below in bold.

4.5 Decision Tree Solution

89

Figure 4.38
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 B C D E F G H I J K L M N O P Q R S Use mechanical method $80,000 $80,000 0.5 Electronic success $150,000 0.5 Awarded contract 2 $90,000 Try electronic method $90,000 $150,000 0.5 Electronic failure $30,000 $30,000 0.7 Magnetic success Prepare proposal Try magnetic method $20,000 $84,000 0.3 Magnetic failure $0 $0 1 $20,000 0.5 Not awarded contract -$50,000 -$50,000 $120,000 $120,000

Don't prepare proposal $0 $0

Taking into account event branches with subsequent terminal nodes, all branches and terminal values associated with the optimal risk neutral strategy are shown below.

90

Chapter 4 Decision Trees Using TreePlan

Figure 4.39
Use mechanical method $80,000

0.5 Electronic success $150,000 0.5 Awarded contract Try electronic method 0.5 Electronic failure $30,000

0.7 Magnetic success Prepare proposal Try magnetic method 0.3 Magnetic failure $0 $120,000

0.5 Not awarded contract -$50,000

Don't prepare proposal $0

The rollback method has identified strategy 2 (Electronic) as optimal. The rollback value on the initial branch of the optimal strategy is $20,000, which must be the same as the expected value for the payoff distribution of strategy 2. Some of the intermediate calculations for the rollback method differ from the calculations for the payoff distributions, but both approaches identify the same optimal strategy with the same initial expected value. For decision trees with a large number of strategies, the rollback method is more efficient.

4.6 NEWOX DECISION TREE PROBLEM

The Newox Company is considering whether or not to drill for natural gas on its own land. If they drill, their initial expenditure will be $40,000 for drilling costs. If they strike gas, they must spend an additional $30,000 to cap the well and provide the necessary hardware and control equipment. (This $30,000 cost is not a decision; it is associated with the event "strike gas.") If they decide to drill but no gas is found, there are no other subsequent alternatives, so their outcome value is $-40,000. If they drill and find gas, there are two alternatives. Newox could sell to West Gas, which has made a standing offer of $200,000 to purchase all rights to the gas well's production

4.7 Brandon Decision Tree Problem

91

(assuming that Newox has actually found gas). Alternatively, if gas is found, Newox can decide to keep the well instead of selling to West Gas; in this case Newox manages the gas production and takes its chances by selling the gas on the open market. At the current price of natural gas, if gas is found it would have a value of $150,000 on the open market. However, there is a possibility that the price of gas will rise to double its current value, in which case a successful well will be worth $300,000. The company's engineers feel that the chance of finding gas is 30 percent; their staff economist thinks there is a 60 percent chance that the price of gas will double.

4.7 BRANDON DECISION TREE PROBLEM

Brandon Appliance Corporation, a predominant producer of microwave ovens, is considering the introduction of a new product. The new product is a microwave oven that will defrost, cook, brown, and boil food as well as sense when the food is done. Brandon must decide on a course of action for implementing this new product line. An initial decision must be made to (1) nationally distribute the product from the start, (2) conduct a marketing test first, or (3) not market the product at all. If a marketing test is conducted, Brandon will consider the result and then decide whether to abandon the product line or make it available for national distribution. The finance department has provided some cost information and probability assignments relating to this decision. The preliminary costs for research and development have already been incurred and are considered irrelevant to the marketing decision. A success nationally will increase profits by $5,000,000, and failure will reduce them by $1,000,000, while abandoning the product will not affect profits. The test market analysis will cost Brandon an additional $35,000. If a market test is not performed, the probability of success in a national campaign is 60 percent. If the market test is performed, the probability of a favorable test result is 58 percent. With favorable test results, the probability for national success is approximately 93 percent. However, if the test results are unfavorable, the national success probability is approximately 14 percent.

Decision Tree Strategies

Brandon Appliance Corporation must decide on a course of action for implementing this new microwave oven. An initial decision must be made to (1) nationally distribute the product from the start, (2) conduct a marketing test first, or (3) not market the product at all. If a marketing test is conducted, Brandon will consider the result and then decide whether to abandon the product line or make it available for national distribution. The

92

Chapter 4 Decision Trees Using TreePlan

following decision tree is based on information about cash flows and probability assignments. Figure 4.40
0.6 Success +$5,000 National 0.4 Failure -$1,000

0.93 Success +$4,965 National 0.07 Failure -$1,035

0.58 Favorable

Don't Brandon Test 0.14 Success +$4,965 National 0.86 Failure -$1,035 -$35

0.42 Unfavorable

Don't -$35

Don't $0

In a decision tree model, a strategy is a specification of an initial choice and any subsequent choices that must be made by the decision maker. How many strategies are there in the Brandon problem? Describe each strategy.

4.7 Brandon Decision Tree Problem

93

Figure 4.41 Strategy 1: National

0.6 Success +$5,000 National 0.4 Failure -$1,000

0.93 Success +$4,965 National 0.07 Failure -$1,035

0.58 Favorable

Don't Brandon Test 0.14 Success +$4,965 National 0.86 Failure -$1,035 -$35

0.42 Unfavorable

Don't -$35

Don't $0

94

Chapter 4 Decision Trees Using TreePlan

Figure 4.42 Strategy 2: Test; if Favorable, National; if Unfavorable, National

0.6 Success +$5,000 National 0.4 Failure -$1,000

0.93 Success +$4,965 National 0.07 Failure -$1,035

0.58 Favorable

Don't Brandon Test 0.14 Success +$4,965 National 0.86 Failure -$1,035 -$35

0.42 Unfavorable

Don't -$35

Don't $0

4.7 Brandon Decision Tree Problem

95

Figure 4.43 Strategy 3: Test; if Favorable, National; if Unfavorable, Don't

0.6 Success +$5,000 National 0.4 Failure -$1,000

0.93 Success +$4,965 National 0.07 Failure -$1,035

0.58 Favorable

Don't Brandon Test 0.14 Success +$4,965 National 0.86 Failure -$1,035 -$35

0.42 Unfavorable

Don't -$35

Don't $0

96

Chapter 4 Decision Trees Using TreePlan

Figure 4.44 Strategy 4: Test; if Favorable, Don't; if Unfavorable, National

0.6 Success +$5,000 National 0.4 Failure -$1,000

0.93 Success +$4,965 National 0.07 Failure -$1,035

0.58 Favorable

Don't Brandon Test 0.14 Success +$4,965 National 0.86 Failure -$1,035 -$35

0.42 Unfavorable

Don't -$35

Don't $0

4.7 Brandon Decision Tree Problem

97

Figure 4.45 Strategy 5: Test; if Favorable, Don't; if Unfavorable, Don't

0.6 Success +$5,000 National 0.4 Failure -$1,000

0.93 Success +$4,965 National 0.07 Failure -$1,035

0.58 Favorable

Don't Brandon Test 0.14 Success +$4,965 National 0.86 Failure -$1,035 -$35

0.42 Unfavorable

Don't -$35

Don't $0

98

Chapter 4 Decision Trees Using TreePlan

Figure 4.46 Strategy 6: Don't

0.6 Success +$5,000 National 0.4 Failure -$1,000

0.93 Success +$4,965 National 0.07 Failure -$1,035

0.58 Favorable

Don't Brandon Test 0.14 Success +$4,965 National 0.86 Failure -$1,035 -$35

0.42 Unfavorable

Don't -$35

Don't $0

Multiattribute Utility

5.1 APPLICATIONS OF MULTI-ATTRIBUTE UTILITY

Strategy for Dealing with Microcomputer Networking Impact on microcomputer users Productivity enhancement User satisfaction Impact on mainframe capacity Costs Upward compatibility of the network Impacts on organizational structure Risks Purchase of manufacturing machinery Price Technical features Service Choosing a manager candidate Education Management skills Technical skills Personal skills Choosing a beverage container (soft drink industry)

100

Chapter 5 Multiattribute Utility

Energy to produce Cost Environmental waste Customer service Selecting a best job Monetary compensation Geographical location Travel requirements Nature of work

5.2 MULTIATTRIBUTE UTILITY SWING WEIGHTS

Excel Workbook Clemen15.xls Conflicting Objectives: Fundamental Objectives versus Means Objectives Clemen, Making Hard Decisions, Ch. 15 Multiattribute Utility Set of Objectives should be 1) complete 2) as small as possible 3) not redundant 4) decomposable ("independent" or unrelated) Additive Utility Function Overall Score of Alternative = Sum [ Weight times Attribute Score of Alternative ] Figure 5.1 Data for Example
Attribute Life span, in years Price Color Red Portalo 12 $17,000 Red Blue Norushi 9 $10,000 Blue Yellow Standard 6 $8,000 Yellow

5.2 MultiAttribute Utility Swing Weights

101

Attribute Scores
Figure 5.2 Individual Utility for Life Span
Life Span
Scores for Life Span

Years 6 9 12

Score 0 0.5 1
Life Span Score

1.0

0.8

0.6

0.4

0.2

0.0 5 6 7 8 9 10 11 12 13

Life Span, in years

Figure 5.3 Individual Utility for Price

Price
Scores for Price

Price $17,000 $10,000 $8,000

Score 0 0.78 1
Price Score

1.0

0.8

0.6

0.4

0.2

0.0 $5,000 $10,000 Price $15,000 $20,000

102

Chapter 5 Multiattribute Utility

Figure 5.4 Individual Utility for Color

Color
Scores for Color

Color Red Blue Yellow

Score 0 0.667 1
Color Score

1.0

0.8

0.6

0.4

0.2

0.0 Red Blue Color Yellow

Swing Weights
Figure 5.5 Swing Weight Assessment Display
1 2 3 4 5 6 7 8 9 10 A Swing Weights B C D E F G

Consequence to Compare Attribute Swung from Worst to Best Life span Price Color Rank Rate Weight (Benchmark) 6 years $17,000 red 4 0 0.000 Life span 12 years $17,000 red 2 75 0.405 Price 6 years $8,000 red 1 100 0.541 Color 6 years $17,000 yellow 3 10 0.054 185

1) Hypothetical alternatives (number of attributes plus one) Benchmark alternative is worst for all attributes Each other hypothetical alternative has one attribute at best, all others at worst 2) Rank the hypothetical alternatives 3) Benchmark has rating zero, first ranked alternative has rating 100

5.2 MultiAttribute Utility Swing Weights

103

Assign level-of-satisfaction ratings to the intermediate alternatives 4) Weight equals rating divided by sum of ratings Figure 5.6 Swing Weight Assessment Formulas
1 2 3 4 5 6 7 8 9 10 A Swing Weights B C D E F G

Consequence to Compare Attribute Swung from Worst to Best Life span (Benchmark) 6 years Life span 12 years Price 6 years Color 6 years Price Color Rank Rate Weight $17,000 red 4 0 =F6/$F$10 $17,000 red 2 75 =F7/$F$10 $8,000 red 1 100 =F8/$F$10 $17,000 yellow 3 10 =F9/$F$10 =SUM(F6:F9)

Overall Scores
Figure 5.7 Swing Weight Overall Scores Display
1 2 3 4 5 6 7 8 9 10 11 12 I Overall Scores J K L M N O P Q

Attribute Life span Price Color Overall Score Best

Red Portalo Attribute Attribute Value Score 12 $17,000 Red 1.000 0.000 0.000 0.40541 Blue Norushi

Blue Norushi Attribute Attribute Value Score 9 $10,000 Blue 0.500 0.780 0.667 0.66038

Yellow Standard Attribute Attribute Value Score 6 $8,000 Yellow 0.000 1.000 1.000 0.59459

Figure 5.8 Swing Weight Overall Scores Formulas

1 2 3 4 5 6 7 8 9 10 11 12 I Overall Scores J K L M N O P Q R

Attribute Life span Price Color Overall Score Best

Attribute Value 12 $17,000 Red

Red Portalo Attribute Score 1.000 0.000 0.000

Attribute Value 9 $10,000 Blue

Blue Norushi Attribute Score 0.500 0.780 0.667

Yellow Standard Attribute Attribute Value Score 6 $8,000 Yellow 0.000 1.000 1.000 =SUMPRODUCT($G$7:$G$9,Q6:Q8)

=SUMPRODUCT($G$7:$G$9,K6:K8)

=SUMPRODUCT($G$7:$G$9,N6:N8)

=IF(K10=MAX(K10,N10,Q10),"Red Portalo",IF(N10=MAX(K10,N10,Q10),"Blue Norushi","Yellow Standard"))

104

Chapter 5 Multiattribute Utility

Figure 5.9 Sensitivity Analysis

U V W X 1 Sensitivity Analysis Data Tables 2 Life Span Rate (10 to 100) 3 4 W9 Output Formula: =J12 5 Column Input Cell: F7 6 7 Life Span Rate Best 8 9 10 Yellow Standard 10 15 Yellow Standard 11 20 Yellow Standard 12 25 Yellow Standard 13 30 Yellow Standard 14 35 Yellow Standard 15 40 Yellow Standard 16 45 Yellow Standard 17 50 Yellow Standard 18 55 Blue Norushi 19 60 Blue Norushi 20 65 Blue Norushi 21 70 Blue Norushi 22 75 Blue Norushi 23 Base Case 80 Blue Norushi 24 85 Blue Norushi 25 90 Blue Norushi 26 95 Blue Norushi 27 100 Blue Norushi 28 Y Z AA

Color Rate (0 to 75) Z9 Output Formula: =J12 Column Input Cell: F9 Color Rate 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 Best Blue Norushi Blue Norushi Blue Norushi Blue Norushi Blue Norushi Blue Norushi Blue Norushi Blue Norushi Blue Norushi Blue Norushi Yellow Standard Yellow Standard Yellow Standard Yellow Standard Yellow Standard Yellow Standard

Base Case

5.3 SENSITIVITY ANALYSIS METHODS

SENSITIVITY ANALYSIS FOR MULTI-ATTRIBUTE UTILITY USING EXCEL This paper describes several standard methods for analyzing decisions where the outcomes have multiple attributes. The example problem concerns a large company that is planning to purchase several hundred cars for use by the sales force. The company wants a car that is inexpensive, safe, and lasts a long time. Figure 1 shows data for seven cars that are being considered.

5.3 Sensitivity Analysis Methods

105

Figure 1 Attribute Data for Seven Alternatives

A 1 2 3 4 5 6 7 8 9 Attribute Cost Lifetime Safety Cost Lifetime Safety B Alta $20 10 High C Bulldog $18 10 Medium E Alternatives Cruiser Delta $16 $14 8 8 High Medium D F Egret $12 6 Medium G Fleet $10 6 Low H Garnett $15 8 Low

thousands of dollars expected years third-party rating

Other attributes might be important, e.g., comfort and prestige. The cost attribute should include operating costs, insurance, and salvage value, in addition to purchase price. It might be appropriate to combine the cost and lifetime attributes into a single attribute, e.g., cost per year. Clemen [1] suggests that a set of attributes should be complete (so that all important objectives are included), as small as possible (to facilitate analysis), not redundant (to avoid double-counting a common underlying characteristic), and decomposable (so that the decision maker can think about each attribute separately).

Dominance
An alternative can be eliminated if another alternative is better on some objectives and no worse on the others. The Garnett is more expensive than the Delta, has the same lifetime, and has a lower safety rating. So the Garnett can be eliminated from further consideration.

Monetary Equivalents Assessment

One method for comparing multi-attribute alternatives is to subjectively assign monetary values to the non-monetary attributes. For example, the decision maker may determine that each additional year of expected lifetime is worth $500, medium safety is $4,000 better than low safety, and high safety is $6,000 better than low safety. Arbitrarily using Fleet as the base case with total equivalent cost of $10,000, Figure 2 shows costs and equivalent costs, in thousands of dollars, in rows 9:11. The negative entries for Lifetime and Safety correspond to positive benefits relative to the Fleet car's base case values. Based on this method, the Egret is chosen. Sensitivity analysis, not shown here, would involve seeing how the choice depends on subjective equivalents different from the $500 per year lifetime and the $4,000 and $6,000 safety assessments. Hammond et al. [3] describe another method involving even swaps that could be used to select the best alternative.

106

Chapter 5 Multiattribute Utility

Figure 2 Monetary Equivalents for Non-Dominated Alternatives

A 1 2 3 4 5 6 7 8 9 10 11 12 13 Attribute Cost Lifetime, years Safety rating B Alta $20 10 High D E F Non-Dominated Alternatives Bulldog Cruiser Delta Egret $18 $16 $14 $12 10 8 8 6 Medium High Medium Medium Non-Dominated Alternatives Bulldog Cruiser Delta $18 $16 $14 -$2 -$1 -$1 -$4 -$6 -$4 $12 $9 $9 C G Fleet $10 6 Low

Attribute Cost Lifetime, $ Safety, $ Equiv. Cost

Alta $20 -$2 -$6 $12

Egret $12 $0 -$4 $8

Fleet $10 $0 $0 $10

Additive Utility Function

The additive multi-attribute utility function U includes individual utility functions Ui for each attribute xi, usually scaled from 0 to 1, and weights wi that reflect the decision maker's tradeoffs among the attributes. U(x1,x2,x3) = w1.U1(x1) + w2.U2(x2) + w3.U3(x3), where w1 + w2 + w3 = 1 Weights may be specified directly, as ratios, or using a swing weight procedure. Individual utility functions are assessed using the range of attribute values for the alternatives being considered. The individual utility values for Cost and Lifetime shown in Figure 3 are based on proportional scores, corresponding to linear utility functions. For example, each thousand dollar difference in cost is associated with a 0.1 difference in utility. The utility values for Safety are subjective judgments. For example, the decision maker thinks that a change in Safety from Low to Medium achieves only two-thirds of the satisfaction associated with a change from Low to High. (1)

5.3 Sensitivity Analysis Methods

107

Figure 3 Individual Utilities

A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Attribute Cost Lifetime Safety B Alta $20 10 High D E F Non-Dominated Alternatives Bulldog Cruiser Delta Egret $18 $16 $14 $12 10 8 8 6 Medium High Medium Medium C G Fleet $10 6 Low

Assess individual utility for each attribute. Cost U($20,000)=0, U($10,000)=1, linear Lifetime U(6 years)=0, U(10 years)=1, linear Safety U(Low)=0, U(Medium)=2/3, U(High)=1 Non-Dominated Alternatives Bulldog Cruiser Delta 0.200 0.400 0.600 1.000 0.500 0.500 0.667 1.000 0.667

Attribute Cost Lifetime Safety

Alta 0.000 1.000 1.000

Egret 0.800 0.000 0.667

Fleet 1.000 0.000 0.000

Compared to the assessments for individual utility, the assessments for tradeoffs are usually much more difficult to make. The following sections focus on assessments of tradeoff weights and sensitivity analysis.

Weight Ratio Assessment

One method for measuring trade-offs among the conflicting objectives is to assess weight ratios. For example, the decision maker may judge that cost is five times as important as lifetime, which may be interpreted to mean that the change in overall satisfaction corresponding to a change in cost from $20,000 to $10,000 is five times the change in overall satisfaction corresponding to a change in lifetime from 6 years to 10 years. Similarly, the decision maker may judge that a $10,000 decrease in cost is one and a half times as satisfying as a change from a low to a high safety rating. The assessments are shown in cells J4:J5 in Figure 4.

108

Chapter 5 Multiattribute Utility

Figure 4 Weight Ratio Assessment and Choice

A 1 2 3 4 5 6 7 8 9 10 11 12 13 Attribute Cost Lifetime Safety Overall Max Value Location Choice Choice B Alta 0.000 1.000 1.000 0.464 0.667 5 Egret Egret D E Non-Dominated Alternatives Bulldog Cruiser Delta 0.200 0.400 0.600 1.000 0.500 0.500 0.667 1.000 0.667 0.452 0.625 0.613 C F Egret 0.800 0.000 0.667 0.667 G Fleet 1.000 0.000 0.000 0.536 H I J Assess weight ratios. Weight Ratio Cost/Lifetime Cost/Safety Weights Cost Lifetime Safety Input 5.0 1.5

0.536 0.107 0.357

With three attributes, the two assessed weight ratios determine two equations and the requirement that the weights sum to one determines a third equation. Using algebra, a solution for the three unknown weights is shown in cells J8:J10 in Figure 5. The formula for overall utility in cell B7, with a relative reference to the attribute utilities in B3:B5 and an absolute reference to the weights in J8:J10, is copied to cells C7:G7. The MAX worksheet function determines the maximum overall utility in B7:G7, the MATCH function determines the location of that maximum in B7:G7, and the INDEX function returns the alternative name located in B2:G2. The zero argument in the MATCH function is needed to specify that an exact match is required; the zero argument in the INDEX function is used as a placeholder and could be omitted in this application without affecting the results. Cell B13 combines these functions into a single formula. Figure 5 Formulas for Weight Ratio Assessment and Choice
A 1 2 3 4 5 6 7 8 9 10 11 12 13 Attribute Cost Lifetime Safety Overall Alta 0 1 1 =SUMPRODUCT(B3:B5,$J$8:$J$10) B Non-Dominated Alternatives H I Assess weight ratios. Weight Ratio Cost/Lifetime Cost/Safety Weights Cost Lifetime Safety Input 5 1.5 J

Max Value =MAX(B7:G7) Location =MATCH(B9,B7:G7,0) Choice =INDEX(B2:G2,0,B10) Choice =INDEX(B2:G2,0,MATCH(MAX(B7:G7),B7:G7,0))

=1/(1/J4+1/J5+1) =J8/J4 =J8/J5

After deleting cells A9:B12, the single formula is in cell B9. The arrangement shown in Figure 6 is used for the remaining analyses.

5.3 Sensitivity Analysis Methods

109

Figure 6 Weight Ratio Choice for Sensitivity Analysis

A 1 2 3 4 5 6 7 8 9 Attribute Cost Lifetime Safety Overall Choice B Alta 0.000 1.000 1.000 0.464 Egret D E Non-Dominated Alternatives Bulldog Cruiser Delta 0.200 0.400 0.600 1.000 0.500 0.500 0.667 1.000 0.667 0.452 0.625 0.613 C F Egret 0.800 0.000 0.667 0.667 G Fleet 1.000 0.000 0.000 0.536

Weight Ratio Sensitivity Analysis

The decision maker specified tradeoffs using weight ratios, so it is appropriate to see whether the choice is sensitive to changes in those assessed values. To construct a twoway data table for sensitivity analysis of the weight ratios as shown in Figures 7 and 8, enter a set of values in a row, N4:R4, and another set of values in a column, M5:M13. In the top left cell of the data table, M4, enter a formula for determining the data table's output values, =B9. (To improve the appearance of the table, cell M4 is formatted with a custom three-semicolon format so that the formula result is not displayed.) Select M4:R13. Choose Data | Table. In the Data Table dialog box, specify J4 as the Row Input Cell and J5 as the Column Input Cell. Click OK. Figure 7 Coarse Two-Factor Sensitivity Analysis of Weight Ratios
1 2 3 4 5 6 7 8 9 10 11 12 13 L M N Two-Factor Sensitivity Analysis O P Q R

Cost/Safety Weight Ratio

1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00

Cost/Lifetime Weight Ratio 3.0 4.0 5.0 Cruiser Cruiser Cruiser Egret Egret Cruiser Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret

6.0 Cruiser Egret Egret Egret Egret Egret Egret Egret Egret

7.0 Cruiser Egret Egret Egret Egret Egret Egret Egret Egret

Cell P7, corresponding to the original assessments, has a border. The data table is dynamic, so the macro view may be refined near the base-case assessments by specifying different input values.

110

Chapter 5 Multiattribute Utility

Figure 8 Fine Two-Factor Sensitivity Analysis of Weight Ratios

1 2 3 4 5 6 7 8 9 10 11 12 13 L M N Two-Factor Sensitivity Analysis O P Q R

Cost/Safety Weight Ratio

1.00 1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80

Cost/Lifetime Weight Ratio 4.0 4.5 5.0 Cruiser Cruiser Cruiser Cruiser Cruiser Cruiser Cruiser Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret

5.5 Cruiser Egret Egret Egret Egret Egret Egret Egret Egret

6.0 Cruiser Egret Egret Egret Egret Egret Egret Egret Egret

Figure 8 shows that the Cost/Safety weight ratio must be less than 1.2 to affect the choice. If the decision maker regards 1.2 as "far away" from 1.5, then the Egret choice is appropriate. Otherwise, the decision maker should think more carefully about the original assessments before making a choice based on this analysis. The assessment of the Cost/Lifetime weight ratio is not as critical, because any value between 4 and 6 yields the same choice.

Swing Weight Assessment

Compared to weight ratio assessment, the swing weight method requires assessments that are similar to directly assigning an overall utility to an alternative. However, the hypothetical alternatives requiring assessment in this method are constructed so that it should be easier for the decision maker to assign overall utilities to them instead of to the actual alternatives. The swing weight method involves four steps as shown in Figure 9. 1) Develop the hypothetical alternatives. The number of hypothetical alternatives equals the number of attributes plus one. The benchmark alternative in column J is worst for all attributes. Each other hypothetical alternative, shown in columns K, L, and M, has one attribute at best and all others at worst. Rank the hypothetical alternatives, as shown in row 7. This is an intermediate step that facilitates assigning overall utilities. Assign overall utility scores reflecting overall satisfaction for the hypothetical alternatives. The benchmark worst case has score zero, and the first-ranked alternative has score 100. Then assign level-of-satisfaction scores to the intermediate alternatives, as shown in cells L9 and M9.

2) 3)

5.3 Sensitivity Analysis Methods

111

4)

Sum the scores, as shown in cell N9. In the additive utility function, the weight for each attribute equals the score divided by sum of the scores. (The algebra solution, not shown here, is based on the special zero and one individual utility values of the hypothetical alternatives.) Formulas are shown in Figure 10.

Figure 9 Hypothetical Alternatives and Weights for Swing Weight Assessment

I 1 2 3 4 5 6 7 8 9 10 11 12 13 Attribute Cost Lifetime Safety Rank Overall Score Weight J Worst $20 6 Low 4 0 0.000 K L M Hypothetical Alternatives Best Cost Best Lifetime Best Safety $10 $20 $20 6 10 6 Low Low High 1 100 0.526 3 20 0.105 2 70 0.368 Total 190 N

Decision Maker's Inputs Underlined

Figure 10 Formulas for Swing Weight Assessment

I 1 2 3 4 5 6 7 8 9 10 11 12 13 Attribute Cost Lifetime Safety Rank K L Hypothetical Alternatives Worst Best Cost Best Lifetime 20 10 20 6 6 10 Low Low Low 4 1 100 3 20 J M Best Safety 20 6 High 2 70 =M9/$N$9 Total =SUM(J9:M9) N

Overall Score 0 Weight

=J9/$N$9 =K9/$N$9 =L9/$N$9

Decision Maker's Inputs Underlined

The individual utility values are in a column, and the weights are in a row. The SUMPRODUCT function requires that the two arrays for its arguments have the same orientation, so the TRANSPOSE function converts the weights into a column format, as shown in Figure 11. The function in B7 must be array-entered; after typing the function, hold down Control and Shift while you press Enter.

112

Chapter 5 Multiattribute Utility

Figure 11 Formulas for Swing Weight Choice

A 1 2 3 4 5 6 7 8 9 Attribute Cost Lifetime Safety Overall Choice Alta 0 1 1 =SUMPRODUCT(B3:B5,TRANSPOSE($K$11:$M$11)) =INDEX(B2:G2,0,MATCH(MAX(B7:G7),B7:G7,0)) B Non-Dominated Alternatives

Figure 12 Swing Weight Choice

A 1 2 3 4 5 6 7 8 9 Attribute Cost Lifetime Safety Overall Choice B Alta 0.000 1.000 1.000 0.474 Egret D E Non-Dominated Alternatives Bulldog Cruiser Delta 0.200 0.400 0.600 1.000 0.500 0.500 0.667 1.000 0.667 0.456 0.632 0.614 C F Egret 0.800 0.000 0.667 0.667 G Fleet 1.000 0.000 0.000 0.526

Swing Weight Sensitivity Analysis

The decision maker specified tradeoffs using overall scores for the hypothetical alternatives, so it is appropriate to see whether the choice is sensitive to changes in those assessed values. Figure 13 shows the sensitivity for the Best-Lifetime score that was specified as 20 relative to the worst-case benchmark and the highest-ranked Best-Cost hypothetical alternative. The Best-Lifetime alternative is still ranked 3 as long as its score is between 0 and 70. To improve the appearance of the sensitivity analysis tables in Figure 13, the output formula cells, R13 and T13, have a three-semicolon custom format.

5.3 Sensitivity Analysis Methods

113

Figure 13 Sensitivity Analysis of Swing Weight Best-Lifetime Score

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 P Q R Single-Factor Sensitivity Analysis S T U

Best Lifetime Overall Score Base case Score is 20 Rank 3 as long as Score is between 0 and 70 Output Formula in cell R13: =B9 Data Table Column Input Cell: M9 Detail Best Lifetime Overall Score 30 31 32 33 34 35

Best Lifetime Overall Score 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70

Choice Egret Egret Egret Egret Egret Egret Egret Cruiser Cruiser Cruiser Cruiser Cruiser Cruiser Cruiser Cruiser

Choice Egret Egret Egret Egret Cruiser Cruiser

Base Case

The results in the left table Figure 13, cells Q13:R28, indicate that the Best-Lifetime score must be greater than 30 to affect the choice. A refined data table in cells T13:U19 shows that the score must be greater than 33 before the choice changes from Egret to Cruiser. If the decision maker regards 33 as "far away" from 20, then the Egret choice is appropriate. Figure 14 shows a similar sensitivity analysis for the Best-Safety score. The assessed score of 70 must be greater than 89 to affect the choice.

114

Chapter 5 Multiattribute Utility

Figure 14 Sensitivity Analysis of Swing Weight Best-Safety Score

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 W X Y Single-Factor Sensitivity Analysis Z AA AB

Best Safety Overall Score Base case Score is 70 Rank 2 as long as Score is between 20 and 100 Output Formula in cell Y13 and cell AB13: =B9 Data Table Column Input Cell: N9 Detail Best Safety Overall Score 85 86 87 88 89 90

Best Safety Overall Score 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Choice Fleet Fleet Fleet Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Cruiser Cruiser Cruiser

Choice Egret Egret Egret Egret Egret Cruiser

Base Case

To construct a two-way data table for sensitivity analysis of the swing weight assessments as shown in Figure 15, enter a set of values in a row, R4:V4, and another set of values in a column, Q5:Q13. In the top left cell of the data table, Q4, enter a formula for determining the data table's output values, =B9. (To improve the appearance of the table, cell Q4 is formatted with a custom three-semicolon format so that the formula result is not displayed.) Select Q4:V13. Choose Data | Table. In the Data Table dialog box, specify L9 as the Row Input Cell and M9 as the Column Input Cell. Click OK.

5.3 Sensitivity Analysis Methods

115

Figure 15 Sensitivity Analysis of Both Swing Weight Scores

1 2 3 4 5 6 7 8 9 10 11 12 13 P Q R Two-Way Sensitivity Analysis S T U V

Best Safety Overall Score

50 55 60 65 70 75 80 85 90

Best Lifetime Overall Score 10 15 20 Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Egret Cruiser

25 Egret Egret Egret Egret Egret Egret Egret Cruiser Cruiser

30 Egret Egret Egret Egret Egret Cruiser Cruiser Cruiser Cruiser

The table shows that the choice changes from Egret to Cruiser if the combination of assessments is changed from 20 & 70 to 30 & 75. This table could be refined to examine the exact threshold values.

Direct Weight Assessment and Sensitivity Analysis

In some situations the decision maker may be able to assign tradeoff weights directly. Figure 16 shows results using the formulas shown in Figure 17. Figure 16 Direct Weight Assessment
A 1 2 3 4 5 6 7 8 9 Attribute Cost Lifetime Safety Overall Choice B Alta 0.000 1.000 1.000 0.500 Egret D E Non-Dominated Alternatives Bulldog Cruiser Delta 0.200 0.400 0.600 1.000 0.500 0.500 0.667 1.000 0.667 0.467 0.650 0.617 C F Egret 0.800 0.000 0.667 0.667 G Fleet 1.000 0.000 0.000 0.500 H I Weights Cost Lifetime Safety J 0.500 0.100 0.400

The formula in cell B9 includes an IF function to verify that each weight is between 0 and 1, inclusive, and that the sum of the weights equals one. If not, the formula returns empty text. This formula must be array-entered; after typing the function, hold down Control and Shift while you press Enter.

116

Chapter 5 Multiattribute Utility

Figure 17 Formulas for Direct Weight Assessment

A 1 2 3 4 5 6 7 8 9 Non-Dominated Alternatives Attribute Alta Cost 0 Lifetime 1 Safety 1 Overall Choice =SUMPRODUCT(B3:B5,$J$2:$J$4) =IF(AND(SUM(J2:J4)<=1,J2:J4>=0),INDEX(B2:G2,0,MATCH(MAX(B7:G7),B7:G7,0)),"") B H I J Weights Cost 0.5 Lifetime 0.1 Safety =1-J3-J2

Figure 18 shows a two-way table for sensitivity analysis of the weights. Cell R5 corresponds to the approximate base case assessments in the weight ratio and swing weight methods. Figure 18 Sensitivity Analysis of Direct Weight Assessment
L M N O P Q R 1 Two-Factor Sensitivity Analysis 2 3 Cost Weight 4 0.1 0.2 0.3 0.4 0.5 5 Lifetime 0.1 Alta Cruiser Cruiser Cruiser Egret 6 Weight 0.2 Alta Alta Cruiser Cruiser Cruiser 7 0.3 Alta Alta Alta Cruiser Delta 8 0.4 Alta Alta Alta Bulldog Bulldog 9 0.5 Alta Alta Alta Bulldog Bulldog 10 0.6 Alta Alta Bulldog Bulldog 11 0.7 Alta Bulldog Bulldog 12 0.8 Alta Bulldog 13 0.9 Bulldog S T U V

0.6 Egret Egret Fleet Fleet Fleet Fleet Fleet

0.7 Fleet Fleet

0.8 Fleet

0.9

Figure 19 is a more detailed view. The choice formula in cell B9 is modified by placing the INDEX function inside the LEFT function so that only the first letter of the alternative's name is returned.

5.3 Sensitivity Analysis Methods

117

Figure 19 Detailed Sensitivity Analysis of Direct Weight Assessment

L M N O P Q R S T U V W X Y Z AA AB AC AD AE AF AG AH 1 Two-Factor Sensitivity Analysis 2 3 Cost Weight 4 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 5 Lifetime 0.00 A C C C C C C C C C E E E E E E F F F F F 6 Weight 0.05 A A C C C C C C C C E E E E E F F F F F 7 0.10 A A A C C C C C C C E E E E F F F F F 8 0.15 A A A A C C C C C C E E E E F F F F 9 0.20 A A A A A A C C C C C E E F F F F 10 0.25 A A A A A A A C C C D D F F F F 11 0.30 A A A A A A A A C D D D F F F 12 0.35 A A A A A A A A A D D D F F 13 0.40 A A A A A A A A B B B D F 14 0.45 A A A A A A A B B B B B 15 0.50 A A A A A A A B B B B 16 0.55 A A A A A A B B B B 17 0.60 A A A A A A B B B 18 0.65 A A A A A B B B 19 0.70 A A A A B B B 20 0.75 A A A A B B 21 0.80 A A A B B 22 0.85 A A B B 23 0.90 A A B 24 0.95 A B 25 1.00 A

The results in Figure 19 show that all alternatives in this data set are candidates depending on the tradeoffs specified by the decision maker. In general, moving left to right, if more weight is given to cost, a less expensive alternative is chosen.

Summary
This paper considered three methods for assessing tradeoffs in the additive utility function. For each method sensitivity analysis is useful for gaining insight into which tradeoff assumptions are critical. Kirkwood [2] includes Excel VBA methods for sensitivity analysis of individual utility functions in addition to weights.

Sensitivity Analysis Examples References

[1] [2] [3] Clemen, R.T. Making Hard Decisions: An Introduction to Decision Analysis, 2nd Edition. Duxbury Press, 1996. Kirkwood, C.W. Strategic Decision Making: Multiobjective Decision Analysis with Spreadsheets. Duxbury Press, 1997. Hammond, J.S., Keeney, R.L., and Raiffa, H. Smart Choices: A Practical Guide to Making Better Decisions. Harvard Business School Press, 1999.

118

Chapter 5 Multiattribute Utility

Screenshots from Excel to Word

To copy Excel displays for the figures in this paper, choose File | Page Setup | Sheet | Gridlines and File | Page Setup | Sheet | Row And Column Headings. Select the cell range, hold down the Shift key, and in Excel's main menu choose Edit | Copy Picture | As Shown When Printed. In Word, position the pointer in an empty paragraph and choose Edit | Paste.

Product Mix Optimization

6.1 LINEAR PROGRAMMING CONCEPTS

Formulation
Decision variables (Excel Solver Changing Cells) Objective function (Target Cell) Constraints and right-hand-side values (Constraints) Non-negativity constraints (Constraints)

Graphical Solution
Constraints Feasible region Corner points (extreme points) Objective function value at each corner point Total enumeration vs. simplex algorithm (search) Optimal solution

Sensitivity Analysis
Post-optimality analysis and interpretation of computer print-outs Shadow price (a marginal value) (Excel Solver Sensitivity Report, Constraints section, Shadow Price) The shadow price for a particular constraint is the amount of change in the value of the objective function corresponding to a unit change in the right-hand-side value of the constraint.

120

Chapter 6 Product Mix Optimization

Range on a right-hand-side (RHS) value (Excel Solver Sensitivity Report, Constraints section, Allowable Increase/Decrease) Range over which the shadow price applies. The optimal values of the decision variables would change depending on the exact RHS value, but the current mix of decision variables remains optimal over the specified range of RHS values. Range on an objective function coefficient (Excel Solver Sensitivity Report, Changing Cells section, Allowable Increase/Decrease) Range over which an objective function coefficient could change with the current optimal solution remaining optimal (same mix and values of decision variables). The value of the objective function would change depending on the exact value of the objective function coefficient. Simplex algorithm terminology Slack, surplus, and artificial variables Basic variables (variables "in the solution," typically with non-zero values) Non-basic variables (value equal to zero) Complementary slackness

6.2 Basic Product Mix Problem

121

6.2 BASIC PRODUCT MIX PROBLEM

Figure 6.1 Display
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 A B C D E F G Small Example 1: Product mix problem Your company manufactures TVs and stereos, using a common parts inventory of power supplies, speaker cones, etc. Parts are in limited supply and you must determine the most profitable mix of products to build. TV set 250 1 1 2 1 2 Profit $75 $18,750 $23,750 Stereo 100 1 0 2 1 1 $50 $5,000 RHS Used Available 350 450 250 250 700 800 350 450 600 600

Number to Build-> Part Name Chassis Picture Tube Speaker Cone Power Supply Electronics Per Unit By Product Total

Slack 100 0 100 100 0

Figure 6.2 Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 A B C D E Small Example 1: Product mix problem Your company manufactures TVs and stereos, using a common parts inventory of power supplies, speaker cones, etc. Parts are in limited supply and you must determine the most profitable mix of products to build. F G

Number to Build-> Part Name Chassis Picture Tube Speaker Cone Power Supply Electronics

TV set Stereo RHS 250 100 Used Available Slack 1 1 =SUMPRODUCT($C$7:$D$7,C8:D8) 450 =F8-E8 1 0 =SUMPRODUCT($C$7:$D$7,C9:D9) 250 =F9-E9 2 2 =SUMPRODUCT($C$7:$D$7,C10:D10) 800 =F10-E10 1 1 =SUMPRODUCT($C$7:$D$7,C11:D11) 450 =F11-E11 2 1 =SUMPRODUCT($C$7:$D$7,C12:D12) 600 =F12-E12 Profit Per Unit $75 $50 By Product =C14*C7 =D14*D7 Total =SUMPRODUCT(C7:D7,C14:D14)

122

Chapter 6 Product Mix Optimization

Figure 6.3 Graphical Solution

600 Five Constraints 500 Chassis & Power Supply 400 Number of TVs Speaker Cone 300 Picture Tube 200

100

Feasible Region Electronics

0 0 100 200 300 400 500 600 700 Number of Stereos

Figure 6.4 Solver Solution

6.2 Basic Product Mix Problem

123

Figure 6.5 Solver Answer Report

Target Cell (Max) Cell Name $C$16 Total Profit Original Value $23,750 Final Value $25,000

Adjustable Cells Cell Name Original Value $C$7 Number to Build-> TV set 250 $D$7 Number to Build-> Stereo 100

Final Value 200 200

Constraints Cell Name $E$8 Chassis Used $E$9 Picture Tube Used $E$10 Speaker Cone Used $E$11 Power Supply Used $E$12 Electronics Used

Cell Value 400 200 800 400 600

Formula $E$8<=$F$8 $E$9<=$F$9 $E$10<=$F$10 $E$11<=$F$11 $E$12<=$F$12

Status Slack Not Binding 50 Not Binding 50 Binding 0 Not Binding 50 Binding 0

Figure 6.6 Solver Sensitivity Report

Adjustable Cells Final Reduced Objective Allowable Allowable Cell Name Value Cost Coefficient Increase Decrease $C$7 Number to Build-> TV set 200 $0.00 $75.00 $25.00 $25.00 $D$7 Number to Build-> Stereo 200 $0.00 $50.00 $25.00 $12.50 Constraints Cell $E$8 $E$9 $E$10 $E$11 $E$12 Name Chassis Used Picture Tube Used Speaker Cone Used Power Supply Used Electronics Used Final Shadow Constraint Allowable Allowable Value Price R.H. Side Increase Decrease 400 $0.00 450 1E+30 50 200 $0.00 250 1E+30 50 800 $12.50 800 100 100 400 $0.00 450 1E+30 50 600 $25.00 600 50 200

124

Chapter 6 Product Mix Optimization

Figure 6.7 Solver Limits Report

Target Cell Name $C$16 Total Profit Value $25,000

Adjustable Cell Name $C$7 Number to Build-> TV set $D$7 Number to Build-> Stereo

Value 200 200

Lower Limit 0 0

Target Result $10,000 $15,000

Upper Limit 200 200

Target Result $25,000 $25,000

6.3 OUTDOORS PROBLEM

Outdoors, Inc., has lawn furniture as one of its product lines. They currently have three items in that line: a lawn chair, a standard bench, and a table. These products are produced in a two-step manufacturing process involving the tube bending department and the welding department. The hours required by each item in each department is as follows: Product Bench Table 1.7 1.2 0.0 2.3

Department Bending Welding

Chair 1.2 0.8

Present Capacity 1,000 hours 1,200 hours

The profit contribution that Outdoors receives from manufacture and sale of one unit of each product is $3 for a chair, $3 for a bench, and $5 for a table. The company is trying to plan its production mix for the current selling season. They feel that they can sell any number they produce, but unfortunately production is further limited by available material because of a prolonged strike. The company currently has on hand 2,000 pounds of tubing. The three products require the following amounts of this tubing: 2 pounds per chair, 3 pounds per bench, and 4.5 pounds per table. In order to determine the optimal product mix, the production manager has formulated the linear programming problem as shown below.

6.3 Outdoors Problem

125

Contribution Constraint Bending Welding Tubing A.

Chair $3

Product Bench Table $3 $5 Relation <= <= <= Limit 1,000 1,200 2,000

1.2 0.8 2.0

1.7 0.0 3.0

1.2 2.3 4.5

The inventory manager suggests that the company produce 200 units of each product. Is the plan to produce 200 units of each product a feasible plan, i.e., does it satisfy all contraints? If not, which constraints are not satisfied? If the company produces 200 chairs, 200 benches, and 200 tables, how much tubing, if any, will be left over?

B.

Each of the following questions refer to the solution of the original linear programming problem. C. A local manufacturing firm has excess capacity in its welding department and has offered to sell 100 hours of welding time to Outdoors for $3 per hour. This arrangement would cost $300 and would increase welding capacity from 1,200 hours to 1,300 hours. Should Outdoors purchase the additional welding capacity? Why or why not? The marketing manager thinks that the original estimate of $3 profit contribution per chair should be changed to $2.50 per chair. Should the production manager solve the linear programming problem again using the $2.50 value, or should Outdoors go ahead with the plan to produce 700 chairs, zero benches, and 133 tables? Why or why not? A local metal products distributor has offered to sell Outdoors some additional metal tubing for 60 cents per pound. Should Outdoors buy additional tubing at this price? If so, how much would their contribution increase if they bought 500 pounds and used it in an optimal fashion? The R&D department has been redesigning the bench to make it more profitable. The new design will require 1.1 hours of tube bending time, 2 hours of welding time, and 2.0 pounds of metal tubing. If they can sell one unit of this bench with a unit contribution of $3, what effect will it have on overall contribution? Marketing has suggested a new patio awning that would require 1.8 hours of tube bending time, 0.5 hours of welding time, and 1.3 pounds of metal tubing.

D.

E.

F.

G.

126

Chapter 6 Product Mix Optimization

What contribution must this new product have to make it attractive to produce this season? H. Outdoors, Inc., has a chance to sell some of its capacity in tube bending at a price of $1.50 per hour. If it sells 200 hours at that price, how will this affect contribution? If Outdoors, Inc., feels that it must produce benches to round out its production line, what effect will production of benches have on overall contribution?

I.

Adapted from Vatter et al., Quantitative Methods in Management, Irwin, 1978.

Spreadsheet Model
Figure 6.8 Model
A B 1 Outdoors, Inc. 2 Number to Build-> 3 4 Resource Tube Bending Welding 5 Tubing 6 Per Unit 7 Profits By Product 8 Total 9 C Chair 100 1.2 0.8 2 $3 $300 $1,100 D Bench 100 1.7 0 3 $3 $300 E Table 100 1.2 2.3 4.5 $5 $500 F G H

Used 410 310 950

Available 1000 1200 2000

Slack 590 890 1050

Figure 6.9 Formulas

A B 1 Outdoors, Inc. 2 Number to Build-> 3 Tube Bending 4 Resource Welding 5 Tubing 6 Per Unit 7 Profits By Product 8 Total 9 C Chair 100 1.2 0.8 2 3 =C7*C3 =SUMPRODUCT(C3:E3,C7:E7) D Bench 100 1.7 0 3 3 =D7*D3 E F G H Table 100 Used Available Slack 1.2 =SUMPRODUCT(C$3:E$3,C4:E4) 1000 =G4-F4 2.3 =SUMPRODUCT(C$3:E$3,C5:E5) 1200 =G5-F5 4.5 =SUMPRODUCT(C$3:E$3,C6:E6) 2000 =G6-F6 5 =E7*E3

6.3 Outdoors Problem

127

Figure 6.10 Solution

A B C 1 Outdoors, Inc. Chair 2 Number to Build-> 700 3 1.2 4 Resource Tube Bending Welding 0.8 5 Tubing 2.0 6 Per Unit $3 7 Profits By Product $2,100.00 8 Total $2,766.67 9 D Bench 0 1.7 0.0 3.0 $3 $0.00 E Table 133.33 1.2 2.3 4.5 $5 $666.67 F G H

Used 1000 866.67 2000

Available 1000 1200 2000

Slack 0 333.33 0

Solver Reports
Figure 6.11 Answer Report
Target Cell (Max) Cell Name $C$9 Total Chair Original Value Final Value $1,100 $2,767

Adjustable Cells Cell Name Original Value Final Value $C$3 Number to Build-> Chair 100 700 $D$3 Number to Build-> Bench 100 0 $E$3 Number to Build-> Table 100 133.33

Constraints Cell Name $F$4 Tube Bending Used $F$5 Welding Used $F$6 Tubing Used

Cell Value Formula Status 1000 $F$4<=$G$4 Binding 866.67 $F$5<=$G$5 Not Binding 2000 $F$6<=$G$6 Binding

Slack 0 333.33 0

128

Chapter 6 Product Mix Optimization

Figure 6.12 Sensitivity Report

Adjustable Cells Final Reduced Objective Allowable Cell Name Value Cost Coefficient Increase $C$3 Number to Build-> Chair 700 $0.00 $3.00 $2.00 $D$3 Number to Build-> Bench 0 -$1.383 $3.00 $1.383 $E$3 Number to Build-> Table 133 $0.00 $5.00 $1.75 Constraints Cell Name $F$4 Tube Bending Used $F$5 Welding Used $F$6 Tubing Used Final Shadow Constraint Allowable Value Price R.H. Side Increase 1000 $1.167 1000 200 866.67 $0.00 1200 1E+30 2000 $0.80 2000 555.56 Allowable Decrease 466.67 333.33 333.33 Allowable Decrease $0.778 1E+30 $2.00

Uncertain Quantities

7.1 DISCRETE UNCERTAIN QUANTITIES

Discrete UQ: a few, distinct values Assign probability mass to each value (probability mass function). Contrast discrete UQs with continuous UQs. Continuous UQs have an infinite number of values or so many distinct values that it is difficult to assign probability to each value. Instead, for a continuous UQ we assign probability only to ranges of values.

7.2 CONTINUOUS UNCERTAIN QUANTITIES

Probability Density Functions and Cumulative Probability for Continuous Uncertain Quantities The total area under a probability density function equals one. A portion of the area under a density function is a probability. The height of a density function is not a probability. The simplest probability density function is the uniform density function.

Case A: Uniform Density

The number of units of a new product that will be sold is an uncertain quantity. What is the minimum quantity? What is the maximum quantity? 1000 units 5000 units

Are any values in the range between 1000 and 5000 more likely than others? No Represent the uncertainty using a uniform density function.

130

Chapter 7 Uncertain Quantities

Technical point: For a continuous UQ, P(X=x) = 0. For a continuous UQ, probability is non-zero only for a range of values. For convenience in computation and assessment, we may use a continuous UQ to approximate a discrete UQ, and vice versa. In Figure 1, the range of values is 5000 1000 = 4000, which is the width of the total area under the uniform (rectangular) density function. The area of a rectangle is Width * Height = Area, and the area under the uniform density function in Figure 1 must equal 1. So, Height = Area / Base. Here the Base is 5000 1000 = 4000 units. Therefore, Height = 1/4000 = 0.00025. Figure 7.1 Uniform Density Function

Uniform Density Function

Probability Density, f(x)

0.00025

0 0 1000 2000 3000 Unit Sales, x 4000 5000 6000

7.2 Continuous Uncertain Quantities

131

Figure 7.2 Figure 2

Cumulative Probability for Uniform Density

Cumulative Probability, P(X<=x)

1.00 0.75 0.50 0.25 0.00 0 1000 2000 3000 Unit Sales, x 4000 5000 6000

Both probability mass functions (for discrete UQs) and probability density functions (for continuous UQs) have corresponding cumulative probability functions. It is important to understand the relationship between a density function and its cumulative probability function. Cumulative probability can be expressed in four ways: P(X<=x) P(X<x) P(X>=x) P(X>x) probability that UQ X is less than or equal to x probability that UQ X is strictly less than x probability that UQ X is greater than or equal to x probability that UQ X is strictly greater than x inclusive left -tail exclusive left -tail inclusive right -tail exclusive right -tail

For continuous UQs the cumulative probability is the same for inclusive and exclusive. P(X<=x) is the most common type.

132

Chapter 7 Uncertain Quantities

Figure 2 is the cumulative probability function corresponding to the uniform density function shown in Figure 1. What is the probability that sales will be between 3,500 and 4,000 units? P(3500<=X<=4000) = 0.125 P(3500<=X<=4000) = P(X<=4000) P(X<=3500) = 0.750 0.625 = 0.125 Mathematical observation: The uniform density function is a constant; the corresponding cumulative function (the integral of the constant function) is linear.

Case B: Ramp Density

The number of units of a new product that will be sold is an uncertain quantity. What is the minimum quantity? What is the maximum quantity? 1000 units 5000 units

Are any values in the range between 1000 and 5000 more likely than others? Yes, values close to 5000 are much more likely than values close to 1000. Represent the uncertainty using a ramp density function. The area of a triangle is Base * Height / 2, and the area under the ramp density function in Figure 3 must equal 1. So, Height = 2 / Base. Here, the Base is 5000 1000 = 4000 units. Therefore, Height = 2 / 4000 = 0.0005.

7.2 Continuous Uncertain Quantities

133

Figure 7.3 Figure 3

Ramp Density Function

Probability Density, f(x)

0.0005

0 0 1000 2000 3000 Unit Sales, x 4000 5000 6000

Figure 7.4 Figure 4

Cumulative Probability for Ramp Density

Cumulative Probability, P(X<=x)

1.00 0.75 0.50 0.25 0.00 0 1000 2000 3000 Unit Sales, x 4000 5000 6000

134

Chapter 7 Uncertain Quantities

An important observation is that flatter portions of a cumulative probability function correspond to ranges with low probability. Steeper portions of a cumulative probability function correspond to ranges with high probability. What is the probability that sales will be between 3,500 and 4,000 units? P(3500<=X<=4000) = 0.171875 P(3500<=X<=4000) = P(X<=4000) P(X<=3500) = 0.562500 0.390625 = 0.171875 The ramp density may not be appropriate for describing uncertainty in many situations, but it is an important building block for the extremely useful triangular density function. Mathematical observation: The ramp density function is linear; the corresponding cumulative function (the integral of the linear function) is quadratic.

Case C: Triangular Density

The number of units of a new product that will be sold is an uncertain quantity. What is the minimum quantity? What is the maximum quantity? 1000 units 5000 units

Are any values in the range between 1000 and 5000 more likely than others? Yes, values close to 4000 are more likely. Represent the uncertainty using a triangular density function. The area of a triangle is Base * Height / 2, and the area under the triangular density function in Figure 5 must equal 1. So, Height = 2 / Base. Here, the Base is 5000 1000 = 4000 units. Thus, Height = 2 / 4000 = 0.0005.

7.2 Continuous Uncertain Quantities

135

Figure 7.5 Figure 5

Triangular Density Function

Probability Density, f(x)

0.0005

0 0 1000 2000 3000 Unit Sales, x 4000 5000 6000

Figure 7.6 Figure 6

Cumulative Probability for Triangular Density

Cumulative Probability, P(X<=x)

1.00 0.75 0.50 0.25 0.00 0 1000 2000 3000 Unit Sales, x 4000 5000 6000

136

Chapter 7 Uncertain Quantities

Again, an important observation is that flatter portions of a cumulative probability function correspond to ranges with low probability (the range close to 1000 and the range close to 5000 in Figure 6). Steeper portions of a cumulative probability function correspond to ranges with high probability (the range close to 4000). What is the probability that sales will be between 3,500 and 4,000 units? P(3500<=X<=4000) = 0.229167 P(3500<=X<=4000) = P(X<=4000) P(X<=3500) = 0.750000 0.520833 = 0.229167 The triangular density function is extremely useful for describing uncertainty in many situations. It requires only three inputs: minimum, mode (most likely value), and maximum. Mathematical observation: The triangular density function has two linear segments, i.e., piecewise linear; the corresponding cumulative function (the integral of each linear function) is two quadratic segments, i.e., piecewise quadratic.

Simulation Without Add-Ins

8
D E F G

8.1 SIMULATION USING EXCEL FUNCTIONS

Figure 8.1 Display
1 2 3 4 5 6 7 8 9 A B Software Decision Analysis Unit Price Units Sold Unit Variable Cost Fixed Costs Net Cash Flow $29 661 $10.92 $12,000 -$47 C RAND() 0.3502 Normal Mean = 700, StDev = 100 0.9832 Uniform Min = $6, Max = $11 0.7364 Discrete Value Probability Cumulative $10,000 0.25 0.25 $12,000 0.50 0.75 $15,000 0.25 1.00

Figure 8.2 Formulas

1 2 3 4 5 6 7 8 9 A Software Decision Analysis Unit Price Units Sold Unit Variable Cost Fixed Costs Net Cash Flow B C RAND() 29 =INT(NORMINV(C4,700,100)) =6+5*C5 =IF(C6<0.25,10000,IF(C6<0.75,12000,15000)) =B4*(B3-B5)-B6 =RAND() Normal Mean = 700, StDev = 100 =RAND() Uniform Min = $6, Max = $11 =RAND() Discrete Value 10000 12000 15000 D E F G

Probability 0.25 0.5 0.25

Cumulative 0.25 0.75 1

138

Chapter 8 Simulation Without Add-Ins

This page is intentionally mostly blank.

Multiperiod What-If Modeling

9.1 APARTMENT BUILDING PURCHASE PROBLEM

You are considering the purchase of an apartment building in northern California. The building contains 25 units and is listed for $2,000,000. You plan to keep the building for three years and then sell it. You know that the annual taxes on the property are currently $20,000 and will increase to $25,000 after closing. You estimate that these taxes will grow at a rate of 2 percent per year. You estimate that it will cost about $1,000 per unit per year to maintain the apartments, and these maintenance costs are expected to grow at a 15 percent per year rate. You have not decided on the rent to charge. Currently, the rent is $875 per unit per month, but there is substantial turnover, and the occupancy is only 75 percent. That is, on average, 75 percent of the units are rented at any time. You estimate that if you lowered the rent to $675 per unit per month, you would have 100 percent occupancy. You think that intermediate rental charges would produce intermediate occupancy percentages; for example, a $775 rental charge would have 87.5 percent occupancy. You will decide on the monthly rental charge for the first year, and you think the rental market is such that you will be able to increase it 7 percent per year for the second and third years. Furthermore, whatever occupancy percentage occurs in the first year will hold for the second and third years. For example, if you decide on the $675 monthly rental charge for the first year, the occupancy will be 100 percent all three years. At the end of three years, you will sell the apartment building. The realtors in your area usually estimate the selling price of a rental property as a multiple of its annual rental income (before expenses). You estimate that this multiple will be 9. That is, if the rental income in the third year is $200,000, then the sale price will be $1,800,000. Your objective is to achieve the highest total accumulated cash at the end of the three year period. If rental income exceeds expenses in the first or second years, you will invest the excess in one-year certificates of deposits (CDs) yielding 5 percent. Thus, total

140

Chapter 9 Multiperiod What-If Modeling

accumulated cash will include net cash flow (income minus expense) in each of the three years, interest from CDs received at the end of the second and third years, and cash from the sale of the property at the end of the third year. In your initial analysis you have decided to ignore depreciation and other issues related to income taxes. Instead of purchasing the apartment building, you could invest the entire $2,000,000 in certificates of deposits yielding 5 percent per year.

9.1 Apartment Building Purchase Problem

141

Figure 9.1 Base Case Model Display

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 A B Apartment Building Purchase Controllable Factors Unit monthly rent Uncertain Factors Annual unit maintenance Annual maint. increase Annual tax increase Gross rent multiplier Other Assumptions First year property taxes Annual rent increase CD annual yield Intermediate variable Occupancy percentage Performance measure Final cash value C D E Monthly Rent $675 $775 $875 slope intercept F Occupancy 100 87.5 75 -0.125 184.375

$775 $1,000 15% 2.0% 9.00 $25,000 7% 5% 87.50% $2,610,848 One $775 $203,438 $25,000 $25,000 $50,000 $153,438 Two $829 $217,678 $28,750 $25,500 $54,250 $163,428 $153,438 $7,672 Three $887 $232,916 $33,063 $26,010 $59,073 $173,843 $324,538 $16,227 $2,096,240 $2,610,848 $2,000,000 $2,100,000 $2,205,000 $100,000 $105,000 $110,250 $2,315,250

Unit monthly rent Annual rental income Annual maintenance cost Annual property tax Total annual expenses Operating cash flow CD investment Year-end CD interest Sale receipt Final Cash Value CD investment Year-end CD interest Final Cash Value

142

Chapter 9 Multiperiod What-If Modeling

Figure 9.2 Base Case Model Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 A Apartment Building Purchase Controllable Factors Unit monthly rent Uncertain Factors Annual unit maintenance Annual maint. increase Annual tax increase Gross rent multiplier Other Assumptions First year property taxes Annual rent increase CD annual yield Intermediate variable Occupancy percentage Performance measure Final cash value B C D E Monthly Rent 675 775 875 slope intercept F Occupancy 100 87.5 75 =SLOPE(F2:F4,E2:E4) =INTERCEPT(F2:F4,E2:E4)

775 1000 0.15 0.02 9 25000 0.07 0.05 =(F7+F6*B4)/100 =D34 One Two Three =B4 =B20*(1+$B$12) =C20*(1+$B$12) =B20*25*$B$15*12 =C20*25*$B$15*12 =D20*25*$B$15*12 =B6*25 =B11 =SUM(B23:B24) =B21-B25 =(1+$B$7)*B23 =(1+$B$8)*B24 =SUM(C23:C24) =C21-C25 =B27 =B13*C29 =(1+$B$7)*C23 =(1+$B$8)*C24 =SUM(D23:D24) =D21-D25 =C27+C29+C30 =B13*D29 =D21*B9 =SUM(D27:D32) 2000000 =B13*B36 =B36+B37 =B13*C36 =C36+C37 =B13*D36 =D36+D37

Unit monthly rent Annual rental income Annual maintenance cost Annual property tax Total annual expenses Operating cash flow CD investment Year-end CD interest Sale receipt Final Cash Value CD investment Year-end CD interest Final Cash Value

Figure 9.3 Ranges based on decision makers or experts judgment

Uncertain Factors Annual unit maintenance Annual maint. increase Annual tax increase Gross rent multiplier Low $700 10% 2.0% 7.00 Base $1,000 15% 2.0% 9.00 High $2,000 30% 3.0% 10.00

Apartment Building Analysis Notes

Influence Diagram (for single period) Modeling effect of rent on occupancy rate Linear fit: algebra (slope and intercept) XY Scatter chart; Insert Trendline Quadratic fit: if $775 yields 82.5% occupancy instead of 87.5%

9.2 Product Launch Financial Model

143

Base Case model Use Solver to find optimum rent to maximize final cash value Use Sensit.xla Plot of final cash value depending on rent; relatively insensitive Use Sensit.xla Spider Sensitivity Cases Ranges based on decision makers or experts judgment Sensit.xla Tornado chart: identify critical variables Monte Carlo simulation RiskSim.xla Triangular distributions for critical variables What is probability that final cash will be less than $2,315,250?

9.2 PRODUCT LAUNCH FINANCIAL MODEL

Figure 9.4 Original Model Display
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 B C D E F G H I J K L FINANCE The @RISK Demonstration Model : Product Launch Risk Analysis 2001-2010 2001 ======== Price No Entry Price With Entry Volume No Entry Volume With Entry Competitor Entry: Design Costs Capital Investment Operating Expense Factor Sales Price Sales Volume Sales Revenue Unit Production Cost Overhead Cost of Goods Sold Gross Margin Operating Expense Net Before Tax Depreciation Tax Taxes Owed Net After Tax Net Cash Flow NPV 10% $50,000.00 $100,000.00 0.15 $53.00 3300 $174,900 $23.33 $3,300 $80,289 $94,611 $12,043 $82,568 $20,000 $28,781 $0 $82,568 ======== $82,568 0.15 $67.31 4158 $279,875 $24.26 $6,944 $107,830 $172,045 $16,175 $155,870 $20,000 $62,500 $59,081 $96,789 ======== $96,789 0.15 $79.50 3564 $283,338 $25.23 $10,528 $100,461 $182,877 $15,069 $167,808 $20,000 $67,992 $67,992 $99,816 ======== $99,816 0.15 $63.60 3399 $216,176 $26.24 $8,904 $98,104 $118,072 $14,716 $103,357 $20,000 $38,344 $38,344 $65,013 ======== $65,013 0.15 $60.95 3300 $201,135 $27.29 $8,288 $98,354 $102,781 $14,753 $88,028 $40,493 $40,493 $47,535 ======== $47,535 0.15 $55.65 3300 $183,645 $28.38 $8,288 $101,957 $81,688 $15,294 $66,395 $30,542 $30,542 $35,853 ======== $35,853 0.15 $54.59 3432 $187,353 $29.52 $7,952 $109,264 $78,089 $16,390 $61,699 $28,382 $28,382 $33,317 ======== $33,317 0.15 $51.94 3696 $191,970 $30.70 $7,896 $121,366 $70,604 $18,205 $52,400 $24,104 $24,104 $28,296 ======== $28,296 2002 ======== 2003 ======== $70.00 $53.00 3500 3300 2004 ======== $88.20 $67.31 4340 4158 2005 ======== $119.00 $79.50 6580 3564 2006 ======== $112.70 $63.60 5565 3399 2007 ======== $99.40 $60.95 5180 3300 2008 ======== $94.50 $55.65 5180 3300 2009 ======== $91.70 $54.59 4970 3432 2010 ======== $90.30 $51.94 4935 3696

($50,000) ($23,000) $0 ($50,000) ======== ($50,000) $164,877

$0 $20,000 ($9,200) $0 $0 ======== ($100,000)

144

Chapter 9 Multiperiod What-If Modeling

Figure 9.5 Input Assumptions

A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 B C D E F G H I J K L FINANCE The @RISK Demonstration M odel : Product Launch Risk Analysis 2001-2010 2001 ======== Price No Entry Price With Entry Volume No Entry Volume With Entry Competitor Entry: Design Costs Capital Investment Operating Expense Factor Sales Price Sales Volume Sales Revenue Unit Production Cost Overhead Cost of Goods Sold Gross Margin Operating Expense Net Before Tax Depreciation Tax Taxes Owed Net After Tax Net Cash Flow NPV 10% 2002 ======== 2003 ======== $70.00 $53.00 3500 3300 2004 ======== $88.20 $67.31 4340 4158 2005 ======== $119.00 $79.50 6580 3564 2006 ======== $112.70 $63.60 5565 3399 2007 ======== $99.40 $60.95 5180 3300 2008 ======== $94.50 $55.65 5180 3300 2009 ======== $91.70 $54.59 4970 3432 2010 ======== $90.30 $51.94 4935 3696

1 $50,000.00 $100,000.00 0.15 $53.00 3300 $174,900 $23.33 $3,300 $80,289 $94,611 $12,043 $82,568 $20,000 $28,781 $0 $82,568 ======== $82,568 0.15 $67.31 4158 $279,875 $24.26 $6,944 $107,830 $172,045 $16,175 $155,870 $20,000 $62,500 $59,081 $96,789 ======== $96,789 0.15 $79.50 3564 $283,338 $25.23 $10,528 $100,461 $182,877 $15,069 $167,808 $20,000 $67,992 $67,992 $99,816 ======== $99,816 0.15 $63.60 3399 $216,176 $26.24 $8,904 $98,104 $118,072 $14,716 $103,357 $20,000 $38,344 $38,344 $65,013 ======== $65,013 0.15 $60.95 3300 $201,135 $27.29 $8,288 $98,354 $102,781 $14,753 $88,028 $40,493 $40,493 $47,535 ======== $47,535 0.15 $55.65 3300 $183,645 $28.38 $8,288 $101,957 $81,688 $15,294 $66,395 $30,542 $30,542 $35,853 ======== $35,853 0.15 $54.59 3432 $187,353 $29.52 $7,952 $109,264 $78,089 $16,390 $61,699 $28,382 $28,382 $33,317 ======== $33,317 0.15 $51.94 3696 $191,970 $30.70 $7,896 $121,366 $70,604 $18,205 $52,400 $24,104 $24,104 $28,296 ======== $28,296

($50,000) ($23,000) $0 ($50,000) ======== ($50,000) $164,877

$0 $20,000 ($9,200) $0 $0 ======== ($100,000)

Figure 9.6 Modifications for SensIt Display

A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 B Price w/ o Entry Price w/ Entry Volum e No Entry Volum e w/ Entry Com petitor Entry Design Costs Capital Investm ent Operating Expense Factor Unit Production Costs Overhead C Inputs $70.00 $53.00 3,500 3,300 1 $50,000 $100,000 15.0% 23.33 $3,300 D E F G H I J K L

FINANCE The @RISK Demonstration M odel : Product Launch Risk Analysis 2001-2010 2001 ======== Price No Entry Price With Entry Volume No Entry Volume With Entry Competitor Entry: Design Costs Capital Investment Operating Expense Factor Sales Price Sales Volume Sales Revenue Unit Production Cost Overhead Cost of Goods Sold Gross Margin Operating Expense Net Before Tax Depreciation Tax Taxes Owed Net After Tax Net Cash Flow NPV 10% 2002 ======== 2003 ======== $70.00 $53.00 3500 3300 2004 ======== $88.20 $67.31 4340 4158 2005 ======== $119.00 $79.50 6580 3564 2006 ======== $112.70 $63.60 5565 3399 2007 ======== $99.40 $60.95 5180 3300 2008 ======== $94.50 $55.65 5180 3300 2009 ======== $91.70 $54.59 4970 3432 2010 ======== $90.30 $51.94 4935 3696

1 $50,000.00 $100,000.00 0.15 $53.00 3300 $174,900 $23.33 $3,300 $80,289 $94,611 $12,043 $82,568 $20,000 $28,781 $0 $82,568 ======== $82,568 0.15 $67.31 4158 $279,875 $24.26 $6,944 $107,830 $172,045 $16,175 $155,870 $20,000 $62,500 $59,081 $96,789 ======== $96,789 0.15 $79.50 3564 $283,338 $25.23 $10,528 $100,461 $182,877 $15,069 $167,808 $20,000 $67,992 $67,992 $99,816 ======== $99,816 0.15 $63.60 3399 $216,176 $26.24 $8,904 $98,104 $118,072 $14,716 $103,357 $20,000 $38,344 $38,344 $65,013 ======== $65,013 0.15 $60.95 3300 $201,135 $27.29 $8,288 $98,354 $102,781 $14,753 $88,028 $40,493 $40,493 $47,535 ======== $47,535 0.15 $55.65 3300 $183,645 $28.38 $8,288 $101,957 $81,688 $15,294 $66,395 $30,542 $30,542 $35,853 ======== $35,853 0.15 $54.59 3432 $187,353 $29.52 $7,952 $109,264 $78,089 $16,390 $61,699 $28,382 $28,382 $33,317 ======== $33,317 0.15 $51.94 3696 $191,970 $30.70 $7,896 $121,366 $70,604 $18,205 $52,400 $24,104 $24,104 $28,296 ======== $28,296

($50,000) ($23,000) $0 ($50,000) ======== ($50,000) $164,877

$0 $20,000 ($9,200) $0 $0 ======== ($100,000)

9.2 Product Launch Financial Model

145

Figure 9.7 Modifications for SensIt Formulas

A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 B Price w/ o Entry Price w/ Entry Volum e No Entry Volum e w/ Entry Com petitor Entry Design Costs Capital Investm ent Operating Expense Factor Unit Production Costs Overhead C Inputs 70 53 3500 3300 1 50000 100000 0.15 23.33 3300 D E F

FINANCE The @RISK Demonstratio Product Launch Risk Analysis 2001-20 2001 ======== Price No Entry Price With Entry Volume No Entry Volume With Entry Competitor Entry: Design Costs Capital Investment Operating Expense Factor Sales Price Sales Volume Sales Revenue Unit Production Cost Overhead Cost of Goods Sold 2002 ======== =C2 =C3 =C4 =C5 =C6 =C7 =C8 =C9 =IF($C$24=0,E20,E21) =IF($C$24=0,E22,E23) =(E30*E31) =C10 =C11 =(E31*E33)+E34 =$E$28 =IF($C$24=0,F20,F21) =IF($C$24=0,F22,F23) =(F30*F31) =1.04*E33 6944 =(F31*F33)+F34 2003 ======== =1.26*E20 =1.27*E21 =1.24*E22 =1.26*E23 2004 ========

Figure 9.8 Data for Competitor Entry as Base Case

A 1 2 3 4 5 6 7 8 9 10 11 B Price w/ o Entry Price w/ Entry Volum e No Entry Volum e w/ Entry Com petitor Entry Design Costs Capital Investm ent Operating Expense Factor Unit Production Costs Overhead C Inputs $70.00 $53.00 3,500 3,300 0 $50,000 $100,000 15.0% 23.33 $3,300 D E Low $50.00 $40.00 3,100 2,800 0 $37,000 $60,000 6.5% 15.50 $2,800 F Base $70.00 $53.00 3,500 3,300 1 $50,000 $100,000 15.0% 23.33 $3,300 G High $90.00 $68.00 3,900 3,800 1 $63,000 $140,000 23.0% 32.00 $4,000

146

Chapter 9 Multiperiod What-If Modeling

Figure 9.9 Tornado Chart for Competitor Entry as Base Case

Sensit - Sensitivity Analysis - Tornado

Competitor Entry

Price w/ Entry$40.00

$68.00

Unit Production Costs

32.00

15.50

Volume w/ Entry

2,800

3,800

Capital Investment

$140,000

$60,000

Operating Expense Factor

23.0%

6.5%

Design Costs

$63,000

$37,000

Overhead

$4,000 $2,800

Price w/o Entry

$50.00 $90.00

Volume No Entry $0 $100,000

3,100 3,900 $200,000 $300,000 $400,000 $500,000 $600,000 $700,000

NPV 10%

Figure 9.10 Data for No Competitor Entry as Base Case

A 1 2 3 4 5 6 7 8 9 10 11 B Price w/ o Entry Price w/ Entry Volum e No Entry Volum e w/ Entry Com petitor Entry Design Costs Capital Investm ent Operating Expense Factor Unit Production Costs Overhead C Inputs $70.00 $53.00 3,500 3,300 0 $50,000 $100,000 15.0% 23.33 $3,300 D E Low $50.00 $40.00 3,100 2,800 0 $37,000 $60,000 6.5% 15.50 $2,800 F Base $70.00 $53.00 3,500 3,300 0 $50,000 $100,000 15.0% 23.33 $3,300 G High $90.00 $68.00 3,900 3,800 1 $63,000 $140,000 23.0% 32.00 $4,000

9.3 Machine Simulation Model

147

Figure 9.11 Tornado Chart for No Competitor Entry as Base Case

Sensit - Sensitivity Analysis - Tornado

Price w/o Entry

$50.00

$90.00

Competitor Entry

Unit Production Costs

32.00

15.50

Volume No Entry

3,100

3,900

Operating Expense Factor

23.0%

6.5%

Capital Investment

$140,000

$60,000

Design Costs

$63,000

$37,000

Overhead

$4,000 $2,800

Price w/ Entry

$40.00 $68.00

Volume w/ Entry

2,800 3,800

$100,000 $200,000 $300,000 $400,000 $500,000 $600,000 $700,000 $800,000 $900,000 $1,000,00 $1,100,00 0 0 NPV 10%

9.3 MACHINE SIMULATION MODEL

Clemen's AJS Problem

148

Chapter 9 Multiperiod What-If Modeling

Figure 9.12 Process 1 Display and Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 A Process 1 Demand B Zero D 11,000 16,000 21,000 Mean $4.00 Mean 4 C D One D 8,000 19,000 27,000 E F Two D 4,000 21,000 37,000 G

P(D) 0.2 0.6 0.2 StDev $0.40

P(D) 0.2 0.4 0.4

P(D) 0.1 0.5 0.4

Var Cost Normal Machine Failure Poisson Equipment Unit Price Failure Cost Fixed Cost Discount Rate Year Demand Var Cost Failures Cash Flow NPV

$0 $8 $8,000 $12,000 10% Initial Zero 16,000 $4.00 4 $20,000 One 19,000 $4.00 4 $32,000 Two 21,000 $4.00 4 $40,000

Mode Mean Mean

$0 $74,681

Formula in B25: =-B15 Formula in C25: =C22*($B16-C23)-C24*$B17-$B18 Copy to D25:E25 Formula in B27: =B25+NPV(B19,C25:E25)

9.3 Machine Simulation Model

149

Figure 9.13 Process 2 Display

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 A Process 2 Demand B Zero D 14,000 19,000 24,000 Mean $3.50 Mean 3 C D One D 12,000 23,000 31,000 E F Two D 9,000 26,000 42,000 G

P(D) 0.3 0.4 0.3 StDev $1.00

P(D) 0.36 0.36 0.28

P(D) 0.4 0.1 0.5

Var Cost Normal Machine Failure Poisson Equipment Unit Price Failure Cost Fixed Cost Discount Rate Year Demand Var Cost Failures Cash Flow NPV

$60,000 $8 $6,000 $12,000 10% Initial Zero 19,000 $3.50 3 $55,500 One 23,000 $3.50 3 $73,500 Two 26,000 $3.50 3 $87,000

Mode Mean Mean

-$60,000 $116,563

Figure 9.14 RiskSim Functions for Process 1 and Process 2

A 20 21 22 23 24 25 26 27 Year Demand Var Cost Failures Cash Flow NPV Initial B C Zero =randdiscrete(B4:C6) =randnormal($B$9,$C$9) =randpoisson($B$12) =C22*($B16-C23)-C24*$B17-$B18 D One =randdiscrete(D4:E6) =randnormal($B$9,$C$9) =randpoisson($B$12) =D22*($B16-D23)-D24*$B17-$B18 E Two =randdiscrete(F4:G6) =randnormal($B$9,$C$9) =randpoisson($B$12) =E22*($B16-E23)-E24*$B17-$B18

=-B15 =B25+NPV(B19,C25:E25)

150

Chapter 9 Multiperiod What-If Modeling

Figure 9.15 RiskSim Output for Process 1

RiskSim - One Output - Summary Date 9-Apr-01 Time 7:07 PM Workbook AJS_WhatIf.xls Worksheet Process 1 Probability Output Cell $B$27 Output Label NPV Seed 0.5 Trials 1,000
RiskSim Histogram, 09-Apr-01, 07:07 PM 400 350 300 Frequency 250 200 150 100 50 0 -$100,000 $0 $100,000 NPV, Upper Limit of Interval $200,000

Mean St. Dev. Mean St. Error Minimum First Quartile Median Third Quartile Maximum Skewness

$90,526 $47,290 $1,495 -$59,664 $58,050 $91,460 $124,435 $234,703 -0.1034

RiskSim Cumulative Chart, 09-Apr-01, 07:07 PM 1.0 0.9 Cumulative Probability 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 -$100,000 -$50,000 $0 $50,000 $100,000 $150,000 $200,000 $250,000

NPV

9.3 Machine Simulation Model

151

Figure 9.16 RiskSim Output for Process 2

RiskSim - One Output - Summary Date 9-Apr-01 Time 7:08 PM Workbook AJS_WhatIf.xls Worksheet Process 2 Probability Output Cell $B$27 Output Label NPV Seed 0.5 Trials 1,000
RiskSim Histogram, 09-Apr-01, 07:08 PM 300 250 200 Frequency 150 100 50 0 -$100,000 $0 $100,000 $200,000 $300,000 NPV, Upper Limit of Interval

Mean St. Dev. Mean St. Error Minimum First Quartile Median Third Quartile Maximum Skewness

$116,159 $73,675 $2,330 -$70,685 $60,199 $114,335 $168,191 $347,514 0.1390

RiskSim Cumulative Chart, 09-Apr-01, 07:08 PM 1.0 0.9 Cumulative Probability 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 -$100,00 -$50,000 0 $0 $50,000 $100,00 $150,00 $200,00 $250,00 $300,00 $350,00 0 0 0 0 0 0 NPV

152

Chapter 9 Multiperiod What-If Modeling

Follow these instructions to show two or more risk profiles on the same chart. Use RiskSim to obtain the sorted values, cumulative probabilities, and XY charts for strategy A and strategy B. To add the data for strategy B to the existing plot for strategy A, select the sorted values and cumulative probabilities for strategy B (without including the text labels in row 1), and choose Edit | Copy. Click just inside the outer border of the strategy A chart to select it. From the main menu, choose Edit | Paste Special. In the Paste Special dialog box, select "Add cells as New series," select "Values (Y) in Columns," check the box for "Categories (X Values) in First Column," and click OK. Use the same method to add data for other strategies to the strategy A chart. To change the lines and markers of a data series, click a data point on the chart to select the data series, and choose Format | Selected Data Series | Patterns. If the X values are quite different for the various strategies, it may be necessary to adjust the minimum and maximum values on the Scale tab of the Format Axis dialog box.

9.3 Machine Simulation Model

153

Figure 9.17 Comparison of Process1 and Process 2

Process 1 Mean St. Dev. Mean St. Error Minimum First Quartile Median Third Quartile Maximum Skewness $90,526 $47,290 $1,495 -$59,664 $58,050 $91,460 $124,435 $234,703 -0.1034 Process 2 Mean St. Dev. Mean St. Error Minimum First Quartile Median Third Quartile Maximum Skewness $116,159 $73,675 $2,330 -$70,685 $60,199 $114,335 $168,191 $347,514 0.1390

RiskSim Cumulative Chart

1.0 0.9

Process 1
0.8 0.7

Process 2

Cumulative Probability

0.6 0.5 0.4 0.3 0.2 0.1 0.0 -$100,000

-$50,000

$0

$50,000

$100,000

$150,000

$200,000

$250,000

$300,000

$350,000

NPV

154

Chapter 9 Multiperiod What-If Modeling

This page is intentionally mostly blank.

Modeling Waiting Lines

10

10.1 QUEUE SIMULATION

A warehouse has one dock used to unload railroad freight cars. Incoming freight cars are delivered to the warehouse during the night. It takes exactly half a day to unload a car. If more than two cars are waiting to be unloaded on a given day, the unloading of some of the cars is postponed until the following day. The cost is $100 per day for each car delayed. Past experience has indicated that the number of cars arriving during the night have the frequencies shown in the table below. Furthermore, there is no apparent pattern, so that the number arriving on any night is independent of the number arriving on any other night. Figure 10.1 Arrival Frequency Number of cars Relative arriving frequency 0 0.23 1 0.30 2 0.30 3 0.10 4 0.05 5 0.02 6 or more 0.00 1.00 Concepts for Queuing (waiting-line) Models Arrival pattern Service time Number of servers Queue discipline

156

Chapter 10 Modeling Waiting Lines

Performance measures Equilibrium Average waiting time Average number of customers in line System utilization, rho = mean arrival rate / mean service rate Stable system: rho < 1 Figure 10.2 Influence Chart for Simulation Model
Total Cost

Cost of Delays

Number Delayed

Number Delayed

Number Delayed

Number To Unload

Actually Unloaded

Number To Unload

Actually Unloaded

Number To Unload

Actually Unloaded

Number of Arrivals Day 1

Number of Arrivals Day 2

Number of Arrivals Day N

Unloading Capacity

10.1 Queue Simulation

157

Figure 10.3 Simulation Model Spreadsheet Model Display

A B C D E 2 1 Unloading Capacity 2 Random Number of Number Actually 3 Day Number Arrivals To Unload Unloaded 4 1 0.812 3 3 2 5 2 0.524 2 3 2 6 3 0.671 2 3 2 7 4 0.250 1 2 2 8 5 0.940 3 3 2 9 6 0.771 2 3 2 10 7 0.026 0 1 1 11 8 0.178 0 0 0 12 9 0.683 2 2 2 13 10 0.727 2 2 2 14 40 0.082 0 0 0 44 41 0.425 1 1 1 45 42 0.826 3 3 2 46 43 0.855 3 4 2 47 44 0.971 3 5 2 48 45 0.429 1 4 2 49 46 0.592 2 4 2 50 47 0.085 0 2 2 51 48 0.018 0 0 0 52 49 0.678 2 2 2 53 50 0.510 2 2 2 54 55 Total 86 56 57 1.72 58 Daily Average F G H Daily Delay Cost $ Annual Delay Cost $ I 100 16,500

Number Delayed 1 1 1 0 1 1 0 0 0 0 0 0 1 2 3 2 2 0 0 0 0 33 0.66

Figure 10.4 Simulation Model Spreadsheet Model Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 A Unloading Capacity B 2 Random Number =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() =RAND() Number of Arrivals =IF(B5<0.2,0,IF(B5<0.5,1,IF(B5<0.8,2,3))) =IF(B6<0.2,0,IF(B6<0.5,1,IF(B6<0.8,2,3))) =IF(B7<0.2,0,IF(B7<0.5,1,IF(B7<0.8,2,3))) =IF(B8<0.2,0,IF(B8<0.5,1,IF(B8<0.8,2,3))) =IF(B9<0.2,0,IF(B9<0.5,1,IF(B9<0.8,2,3))) =IF(B10<0.2,0,IF(B10<0.5,1,IF(B10<0.8,2,3))) =IF(B11<0.2,0,IF(B11<0.5,1,IF(B11<0.8,2,3))) =IF(B12<0.2,0,IF(B12<0.5,1,IF(B12<0.8,2,3))) =IF(B13<0.2,0,IF(B13<0.5,1,IF(B13<0.8,2,3))) =IF(B14<0.2,0,IF(B14<0.5,1,IF(B14<0.8,2,3))) =IF(B44<0.2,0,IF(B44<0.5,1,IF(B44<0.8,2,3))) =IF(B45<0.2,0,IF(B45<0.5,1,IF(B45<0.8,2,3))) =IF(B46<0.2,0,IF(B46<0.5,1,IF(B46<0.8,2,3))) =IF(B47<0.2,0,IF(B47<0.5,1,IF(B47<0.8,2,3))) =IF(B48<0.2,0,IF(B48<0.5,1,IF(B48<0.8,2,3))) =IF(B49<0.2,0,IF(B49<0.5,1,IF(B49<0.8,2,3))) =IF(B50<0.2,0,IF(B50<0.5,1,IF(B50<0.8,2,3))) =IF(B51<0.2,0,IF(B51<0.5,1,IF(B51<0.8,2,3))) =IF(B52<0.2,0,IF(B52<0.5,1,IF(B52<0.8,2,3))) =IF(B53<0.2,0,IF(B53<0.5,1,IF(B53<0.8,2,3))) =IF(B54<0.2,0,IF(B54<0.5,1,IF(B54<0.8,2,3))) =SUM(C5:C54) =C56/50 Number To Unload =C5 =F5+C6 =F6+C7 =F7+C8 =F8+C9 =F9+C10 =F10+C11 =F11+C12 =F12+C13 =F13+C14 =F43+C44 =F44+C45 =F45+C46 =F46+C47 =F47+C48 =F48+C49 =F49+C50 =F50+C51 =F51+C52 =F52+C53 =F53+C54 Actually Unloaded =MIN(D5,$C$1) =MIN(D6,$C$1) =MIN(D7,$C$1) =MIN(D8,$C$1) =MIN(D9,$C$1) =MIN(D10,$C$1) =MIN(D11,$C$1) =MIN(D12,$C$1) =MIN(D13,$C$1) =MIN(D14,$C$1) =MIN(D44,$C$1) =MIN(D45,$C$1) =MIN(D46,$C$1) =MIN(D47,$C$1) =MIN(D48,$C$1) =MIN(D49,$C$1) =MIN(D50,$C$1) =MIN(D51,$C$1) =MIN(D52,$C$1) =MIN(D53,$C$1) =MIN(D54,$C$1) Number Delayed =D5-E5 =D6-E6 =D7-E7 =D8-E8 =D9-E9 =D10-E10 =D11-E11 =D12-E12 =D13-E13 =D14-E14 =D44-E44 =D45-E45 =D46-E46 =D47-E47 =D48-E48 =D49-E49 =D50-E50 =D51-E51 =D52-E52 =D53-E53 =D54-E54 =SUM(F5:F54) =F56/50 C D E F G H Daily Delay Cost 100 I Annual Delay Cost =250*F58*I1

Day 1 2 3 4 5 6 7 8 9 10 40 41 42 43 44 45 46 47 48 49 50 Total Daily Average

158

Chapter 10 Modeling Waiting Lines

Figure 10.5 Simulation Model Dynamic Histogram Display

K 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 L 50-Day Trial 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ M 17,000 12,500 28,000 2,500 16,000 6,000 9,500 10,500 13,500 7,000 15,500 21,500 16,000 9,000 4,500 8,500 8,500 15,000 9,000 2,000 10,500 16,500 18,500 8,500 5,500 13,500 5,000 23,500 58,000 11,500 7,000 7,000 6,000 7,500 9,500 7,500 12,500 8,500 14,000 7,000 31,000 22,000 40,000 10,500 8,500 N O Minimum Maximum Mean StDev P $ 1,500 $ 58,000 $ 12,845 $ 9,016 Q R S Interval Max Frequency 5000 11 10000 39 15000 24 20000 10 25000 8 30000 3 35000 2 40000 2 45000 0 50000 0 55000 0 60000 1 65000 0 70000 0 75000 0 80000 0 85000 0 90000 0 95000 0 100000 0 More 0 T U V

Simulation
45 40 Frequency of 100 50-Day Trials 35 30 25 20 15 10 5 0 5000 20000 35000 50000 65000 80000 95000 Annual Cost of Delays

10.1 Queue Simulation

159

Figure 10.6 Simulation Model Dynamic Histogram Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 L 50-Day Trial 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 M =I3 =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) =TABLE(,K1) N O P Minimum =MIN(M2:M101) Maximum =MAX(M2:M101) Mean StDev =AVERAGE(M2:M101) =STDEV(M2:M101) Q R Interval Max 5000 10000 15000 20000 25000 30000 35000 40000 45000 50000 55000 60000 65000 70000 75000 80000 85000 90000 95000 100000 More S Frequency =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21) =FREQUENCY(M2:M101,R2:R21)

160

Chapter 10 Modeling Waiting Lines

This page is intentionally mostly blank.

Introduction to Data Analysis

11

object of analysis: person, thing, business entity, etc. characteristic of interest: weight, hair color, diameter, sales, etc. measurement of the characteristic: pounds, blond/brunette/red/etc., inches, dollars, etc.

11.1 LEVELS OF MEASUREMENT

called measurement scales by some authors important distinctions because analysis and summary methods are very different two general levels of measurement, each with two specific levels

Categorical Measure
also called qualitative measure assign a category level to each object of analysis Nominal Measure: simple classification, "assign a name" Ordinal Measure: ranked categories, "assign an ordered classification"

Numerical Measure
also called quantitative measure assign a numerical value to each object of analysis Interval Measure:, rankings and numerical differences are meaningful Ratio Measure: natural zero and numerical ratios are meaningful

162

Chapter 11 Introduction to Data Analysis

11.2 DESCRIBING CATEGORICAL DATA

List each categorical level with frequencies (counts) or relative frequencies (percentages). Use an Excel pivot table to obtain frequencies. Use an Excel bar chart, column chart, or pie chart. To display the relationship between two categorical measures, use a two-way classification table. For nominal data, the appropriate summary measure is the mode (most frequently occurring level) For ordinal data, the appropriate summary measures are the mode and median (the middle-ranked category level with approximately 50% of the counts below and approximately 50% above). Do not assign meaningless numerical values to the categorical levels. Do not use the mean and standard deviation.

11.3 DESCRIBING NUMERICAL DATA

Frequency Distribution and Histogram
Determine the range (maximum minus minimum), generally use between 5 and 15 equally-spaced intervals, and pick "nice" numbers for the upper limit of each interval (Excel "bins"). Use Excel's Histogram analysis tool, or use Excel's FREQUENCY array-entered worksheet function with an Excel Column chart (vertical bars).

Numerical Summary Measures

Appropriate summary measures for central tendency ("What's a typical value?") include mean (average, most appropriate for mound-shaped data), median, and mode. Appropriate summary measures for dispersion ("How typical is the typical value?") include range, standard deviation (most appropriate for mound-shaped distributions), and fractiles (first quartile, or 25th percentile, is a value with approximately 25% of the values below it and approximately 75% of the values above). Appropriate summary measures for shape are Excel's SKEW worksheet function and Pearson's coefficient of skewness.

11.3 Describing Numerical Data

163

Distribution Shapes
Figure 11.1 Positively Skewed Distribution (Skewed to the Right)

Frequency

Value

In a distribution with positive skew, the mean is greater than the median. Figure 11.2 Negatively Skewed Distribution (Skewed to the Left)

Frequency

Value

In a distribution with negative skew, the mean is less than the median.

164

Chapter 11 Introduction to Data Analysis

Figure 11.3 Mound-Shaped Distribution (Symmetric)

Frequency

Value

In a symmetric distribution, the mean and median are equal. Figure 11.4 Bimodal Distribution

Frequency

Value

In a bimodal distribution, there is often a distinguishing characteristic for the two groups of data that have been combined into a single distribution.

Regression Models for Cross-Sectional Data

12

12.1 CROSS-SECTIONAL REGRESSION CHECKLIST

Plot Y versus each X
1 2 3 Verify that the relationship agrees with your prior judgment, e.g., positive vs negative relationship, linear vs nonlinear, strong vs weak Identify outliers or unusual observations and decide whether to exclude Determine whether the relationship is linear; if not, consider using a nonlinear form, e.g., quadratic (include X and X^2 in the model)

Examine the correlation matrix

4 Identify potential multicollinearity problems, i.e., high correlation between a pair of X variables; if so, consider using only one X of the pair in the model

Calculate the regression model with diagnostics

5 Verify that the sign of each regression coefficient agrees with your prior judgment, i.e., positive vs negative relationship; otherwise, consider excluding that X and rerun the regression Examine each plot of residuals vs X; if there is a non-random pattern (e.g., Ushape or upside-down-U-shape), use a nonlinear form for that X in a new model Identify key X variables by comparing standardized regression coefficients, usually computed by multiplying an X coefficient by the standard deviation of that X and dividing by the standard deviation of Y. This dimensionless standardized regression coefficient measures how much Y (in standard deviation units) is affected by a change in X (in standard deviation units).

6 7

166

Chapter 12 Regression Models for Cross-Sectional Data

If a goal is to find a model with small standard error of estimate (approx. standard deviation of residuals), use the t-stat screening method. Disregard the tstat for the intercept. If there are X variables with a t-stat between -1 and +1, remove the single X variable whose t-stat is closest to zero, and rerun the regression. Remove only one X variable at a time. Before using the final model, examine each plot of residuals vs X to verify that the random scatter is the same for all values of X. If there is more scatter for higher values of X, consider using a log transformation of X in the model (instead of using X itself). If the scatter is not uniform with respect to X, the standard error of estimate may not be a useful measure of uncertainty because it overstates the uncertainty for some values of X and understates the uncertainty for other values of X.

Use the model

10 If the purpose is to identify unusual observations, examine the residuals directly for large negative or large positive values, or examine the standardized residuals (each residual divided by the standard deviation of residuals) for values more extreme than +2 or -2 or for values more extreme than +3 or -3. If the purpose is to make predictions, use the X values for a new observation to compute a predicted Y. Use the standard error of estimate to provide an interval estimate, e.g., an approximate 95% prediction interval that ranges from two standard errors below to two standard errors above the predicted Y. Avoid extrapolation, i.e., do not make predictions using X values outside the range of the original data.

11

Time Series Data and Forecasts

13

13.1 TIME SERIES PATTERNS

Meandering time series pattern: Small changes from period to period, possible larger changes over a longer period of time Use an autoregressive model Figure 13.1 Typical Meandering Time Series Pattern

Value

Time

168

Chapter 13 Time Series Data and Forecasts

Figure 13.2 Typical Long-Term Trend Time Series Patterns

Positive Nonlinear

Positive Linear

Value

Negative Linear

No Trend

Time

Figure 13.3 Typical Quarterly Seasonal Time Series with Linear Trend

Value 1

9 10 11 12 13 14 15 16 17 18 19 20 Quarter

13.1 Time Series Patterns

169

Figure 13.4 Quarterly Seasonal Pattern with Nonlinear Trend

Value
1 2 3

4 5

6 7

9 1 1 1 1 1 1 1 1 1 1 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 0 1 2 3 4 5 6 7 8 9

Quarter

Strong seasonal pattern, no trend during first 12 quarters, positive trend during middle 12 quarters, no trend during last 12 quarters

170

Chapter 13 Time Series Data and Forecasts

This page is intentionally mostly blank.

Regression Models for Time Series Data

14

14.1 TIME SERIES REGRESSION CHECKLIST

Relevant explanatory variables (X) for time series data related to business activity (Y), e.g., sales over time, include several general types: a b c d Internal business activity, like advertising, promotion, research and development Competitor business activity, like competitor sales and competitor advertising Industry activity, like number of competitors and market size General economic activity, like personal disposable income

Plot Y versus time

1 Identify any systematic pattern to help determine an appropriate model

Plot Y versus each X

2 3 4 Verify that the relationship agrees with your prior judgment, e.g., positive vs negative relationship, linear vs nonlinear, strong vs weak Identify outliers or unusual observations and decide whether to exclude Determine whether the relationship is linear; if not, consider using a nonlinear form, e.g., quadratic (include X and X^2 in the model)

Examine the correlation matrix

5 Include a time period variable in the correlation matrix. For example, if there are n equally-spaced time periods, include a variable in your data set with values 1,2,...,n. Identify potential multicollinearity problems, i.e., high correlation between a pair of X variables; if so, consider using only one X of the pair in the model

172

Chapter 14 Regression Models for Time Series Data

Calculate the regression model with diagnostics

7 Verify that the sign of each regression coefficient agrees with your prior judgment, i.e., positive vs negative relationship; otherwise, consider excluding that X and rerun the regression Examine each plot of residuals vs X; if there is a non-random pattern (e.g., Ushape or upside-down-U-shape), use a nonlinear form for that X in a new model In addition to the residual plots generated automatically by Excel's Regression tool, prepare and examine a plot of residuals vs time. If there is a snake-like pattern of residuals, consider adding lag Y as an explanatory variable. Optionally, compute the Durbin-Watson statistic to detect autocorrelation of residuals. Identify key X variables by comparing standardized regression coefficients, usually computed by multiplying an X coefficient by the standard deviation of that X and dividing by the standard deviation of Y. This dimensionless standardized regression coefficient measures how much Y (in standard deviation units) is affected by a change in X (in standard deviation units). If a goal is to find a model with small standard error of estimate (approx. standard deviation of residuals), use the t-stat screening method. Disregard the tstat for the intercept. If there are X variables with a t-stat between -1 and +1, remove the single X variable whose t-stat is closest to zero, and rerun the regression. Remove only one X variable at a time. Before using the final model, examine each plot of residuals vs X to verify that the random scatter is the same for all values of X. If there is more scatter for higher values of X, consider using a log transformation of X in the model (instead of using X itself). If the scatter is not uniform with respect to X, the standard error of estimate may not be a useful measure of uncertainty because it overstates the uncertainty for some values of X and understates the uncertainty for other values of X.

8 9

10

11

12

Use the model

13 If the purpose is to identify unusual observations, examine the residuals directly for large negative or large positive values, or examine the standardized residuals (each residual divided by the standard deviation of residuals) for values more extreme than +2 or -2 or for values more extreme than +3 or -3. If the purpose is to make predictions, use the X values for a new observation to compute a predicted Y. Use the standard error of estimate to provide an interval estimate, e.g., an approximate 95% prediction interval that ranges from two standard errors below to two standard errors above the predicted Y. Note that a

14

14.2 Autocorrelation of Residuals

173

time series forecast usually extrapolates beyond the original range of data, so the standard error of estimate is a minimum indication of the uncertainty surrounding a forecast.

14.2 AUTOCORRELATION OF RESIDUALS

Figure 14.1 Undesirable Extreme Negative Autocorrelation

Residual 1

8 Time

10

11

12

13

14

15

174

Chapter 14 Regression Models for Time Series Data

Figure 14.2 Undesirable Extreme Positive Autocorrelation

Residual 1

8 Time

10

11

12

13

14

15

Figure 14.3 Desirable Zero Autocorrelation

Residual 1

8 Time

10

11

12

13

14

15

Sensitivity Analysis for Decision Trees

15

15.1 ONE-VARIABLE SENSITIVITY ANALYSIS

One-Variable Sensitivity Analysis using an Excel data table 1. 2. 3. Construct a decision tree model or financial planning model. Identify the model input cell (H1) and model output cell (A10). Modify the model so that probabilities will always sum to one. (That is, enter the formula =1-H1 in cell H6.)

Figure 15.1 Display for One-Variable Sensitivity Analysis

A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 B C D E F G H 0.6 High sales +$300 Introduce product -$300 +$100 +$600 0.4 Low sales +$100 -$200 +$300 =1-H1 -$200 1 +$100 Model Output Cell Don't introduce $0 $0 $0 I J K L Model Input Cell

4. 5. 6.

Enter a list of input values in a column (N3:N13). Enter a formula for determining output values at the top of an empty column on the right of the input values (=A10 in cell O2). Select the data table range (N2:O13).

176

Chapter 15 Sensitivity Analysis for Decision Trees

7.

From the Data menu choose the Table command.

Figure 15.2
M 1 2 3 4 5 6 7 8 9 10 11 12 13 14 N O +$100 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 P =A10

8.

In the Data Table dialog box, select the Column Input Cell edit box. Type the model input cell (H1), or point to the model input cell (in which case the edit box displays $H$1). Click OK.

Figure 15.3

9.

The Data Table command substitutes each input value into the model input cell, recalculates the worksheet, and displays the corresponding model output value in the table. Optional: Change the formula in cell O2 to =CHOOSE(B9,Introduce,Dont).

10.

15.2 Two-Variable Sensitivity Analysis

177

Figure 15.4
M 1 2 3 4 5 6 7 8 9 10 11 12 13 14 N O P(High Sales) Exp. Value 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 0 0 0 0 0 50 100 150 200 250 300 P

15.2 TWO-VARIABLE SENSITIVITY ANALYSIS

Two-Variable Sensitivity Analysis using an Excel data table Figure 15.5 Decision Tree for Strategy Region Table
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 B C D E F G H I J K L M N O P Q R S
Use mechanical method +$80,000 -$120,000 +$80,000 0.50 Electronic success +$150,000 0.50 Awarded contract 2 +$250,000 +$90,000 Try electronic method -$50,000 +$90,000 $0 +$150,000

0.50 Electronic failure +$30,000 -$120,000 +$30,000

0.70 Magnetic success Prepare proposal Try magnetic method -$50,000 +$20,000 -$80,000 +$84,000 0.30 Magnetic failure $0 -$120,000 1 +$20,000 0.50 Not awarded contract -$50,000 $0 -$50,000 $0 $0 +$120,000 +$120,000

Don't prepare proposal $0 $0 $0

178

Chapter 15 Sensitivity Analysis for Decision Trees

Optional: Activate the Base Case worksheet. From the Edit menu, choose Move Or Copy Sheet. In the Move Or Copy dialog box, check the box for Create A Copy, and click OK. Double-click the new worksheet tab and enter Strategy Region Table.

Setup for Data Table

Select cell P11, and enter the formula =1P6. Select cell P21, and enter the formula =1 P16. In cell U3 enter P(Elec OK). In cell V3 enter 1, and in cell V4 enter 0.9. Select cells V3:V4. In the lower right corner of cell V4, click the fill handle and drag down to cell V13. With cells V3:V13 still selected, click the Increase Decimal button once so that all values are displayed with one decimal place. Select columns V:AG. (Select column V. Click and drag the horizontal scroll bar until column AG is visible. Hold down the Shift key and click column AG.) From the Format menu choose Column | Width. In the Column Width edit box type 5 and click OK. In cell W1 enter P(Mag OK). In cell W2 enter 0 (zero), and in cell X2 enter 0.1. Select cells W2:X2. In the lower right corner of cell X2, click the fill handle and drag right to cell AG2. With cells W2: AG2 still selected, click the Increase Decimal button once so that all values are displayed with one decimal place. Select cell V2 and enter the formula =CHOOSE(J11,"Mech","Elec","Mag"). With the base case assumptions the formula shows Elec. Figure 15.6
U 1 2 3 P(Elec OK) 4 5 6 7 8 9 10 11 12 13 W X P(Mag OK) Elec 0.0 0.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 V Y 0.2 Z 0.3 AA 0.4 AB 0.5 AC 0.6 AD 0.7 AE 0.8 AF 0.9 AG 1.0

Obtaining Results Using Data Table Command

Select the entire data table, cells V2:AG13.

15.2 Two-Variable Sensitivity Analysis

179

From the Data menu, choose Table. In the Table dialog box, type P16 in the Row Input Cell edit box, type P6 in the Column Input Cell edit box, and click OK. With cells V2:AG13 still selected, click the Align Right button. Figure 15.7
U 1 2 3 P(Elec OK) 4 5 6 7 8 9 10 11 12 13 V Elec 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 W X P(Mag OK) 0.0 0.1 Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech Mech Mech Mech Mech Mech Y 0.2 Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech Z 0.3 Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech AA 0.4 Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech AB 0.5 Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech AC 0.6 Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech AD 0.7 Elec Elec Elec Elec Elec Elec Mag Mag Mag Mag Mag AE 0.8 Elec Elec Elec Elec Elec Mag Mag Mag Mag Mag Mag AF 0.9 Elec Elec Elec Elec Mag Mag Mag Mag Mag Mag Mag AG 1.0 Elec Elec Elec Mag Mag Mag Mag Mag Mag Mag Mag

Embellishments
Select cells U1:AG13, and click the Copy button. Select cell AI1, right-click, and from the shortcut menu choose Paste Special. In the Paste Special dialog box, click the Values option button, and click OK. Right-click again, choose Paste Special, click the Formats option button, and click OK. Select columns AJ:AU. Choose Format | Cells | Width, type 5, and click OK. Select cell AJ2, right-click, and from the shortcut menu choose Clear Contents. Select cells AK2:AU2, move the cursor near the border of the selection until it becomes an arrow, click and drag the selection down to cells AK14:AU14. Similarly, select cell AK1 and move its contents down to cell AP15. Also, move the contents of cell AI3 to cell AI8. Select cell AN1, and enter Strategy Region Table.

180

Chapter 15 Sensitivity Analysis for Decision Trees

Figure 15.8
AI 1 2 3 4 5 6 7 8 P(Elec OK) 9 10 11 12 13 14 15 AJ AK AL AM AN AO AP AQ Strategy Region Table Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.3 Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.4 Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech Mech Mech Mech Mech Mech 0.5 0.6 P(Mag OK) AR AS AT AU

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.0

Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.1

Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.2

Elec Elec Elec Elec Elec Elec Mag Mag Mag Mag Mag 0.7

Elec Elec Elec Elec Elec Mag Mag Mag Mag Mag Mag 0.8

Elec Elec Elec Elec Mag Mag Mag Mag Mag Mag Mag 0.9

Elec Elec Elec Mag Mag Mag Mag Mag Mag Mag Mag 1.0

Apply borders to appropriate ranges and cells to show the strategy regions. Apply shading to cell AR8 to show the base case strategy. Figure 15.9
AI 1 2 3 4 5 6 7 8 P(Elec OK) 9 10 11 12 13 14 15 AJ AK AL AM AN AO AP AQ Strategy Region Table Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.3 Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.4 Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech Mech Mech Mech Mech Mech 0.5 0.6 P(Mag OK) AR AS AT AU

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.0

Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.1

Elec Elec Elec Elec Elec Elec Mech Mech Mech Mech Mech 0.2

Elec Elec Elec Elec Elec Elec Mag Mag Mag Mag Mag 0.7

Elec Elec Elec Elec Elec Mag Mag Mag Mag Mag Mag 0.8

Elec Elec Elec Elec Mag Mag Mag Mag Mag Mag Mag 0.9

Elec Elec Elec Mag Mag Mag Mag Mag Mag Mag Mag 1.0

15.3 MULTIPLE-OUTCOME SENSITIVITY ANALYSIS

Sensitivity Analysis for Multiple-Outcome Event Probabilities Choose one of the outcome probabilities that will be explicitly changed. For example, focus on P(Low Sales).

15.3 Multiple-Outcome Sensitivity Analysis

181

Keep same relative likelihood (base case) for the other probabilities. Figure 15.10
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 B C D E F G H 0.2 High Sales +$2,500 I J K L M N P(Low Sales) 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 O OptStrat Don't Don't Don't Don't Intro Intro Intro Intro Intro Intro Intro

+$1,500 +$1,500

Intro -$1,000 +$400

0.5 Medium Sales +$500 +$1,500 0.3 Low Sales +$500 Base ->

1 +$400 +$500 -$500

-$500

Don't $0 $0 $0

Figure 15.11
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 B C D E F G H =(0.2/(0.2+0.5))*(1-H11) High Sales I J K L M N P(Low Sales) 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 O OptStrat =CHOOSE(B13,"Intro","Don't")

Intro

=(0.5/(0.2+0.5))*(1-H11) Medium Sales

Base -> 0.3 Low Sales

Don't

182

Chapter 15 Sensitivity Analysis for Decision Trees

15.4 ROBIN PINELLI'S SENSITIVITY ANALYSIS

Figure 15.12 Decision Tree and Multi-Attribute Utility (Robin Pinelli)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 A B C D E F G Robin Pinelli, Clemen2, pp. 150-151 Non-TreePlan Formulas V6 =V8/V3 V7 =V8/V2 V8 =1/(1/V2+1/V3+1) O6 =$V$6*Q6+$V$7*R6+$V$8*S6 Select O6:O10; click and drag fill handle to O51. H I J K L M N O Overall Utility 0.15 Snowfall 100 cm 48.83 75 25 56 P Q R S Individual Utility Snowfall Magazine T U Weight Ratio Mag/Snow Mag/Income Weights Income Snowfall Magazine V Input 1.50 3.00

Income

0.167 0.333 0.500

0.60 Disp. Income $1500

0.70 Snowfall 200 cm 57.17 75 50 56

0.15 Snowfall 400 cm 73.83 Madison Publishing 55.08 0.15 Snowfall 100 cm 40.50 25 25 56 75 100 56

0.40 Disp. Income $1300

0.70 Snowfall 200 cm 48.83 25 50 56

0.15 Snowfall 400 cm 65.50 25 100 56

0.15 Snowfall 150 cm 29.17 100 37.5 0

MPR Manufacturing 35.96

0.70 Snowfall 230 cm 35.83 100 57.5 0

0.15 Snowfall 320 cm 43.33 100 80 0

Pandemonium Pizza 50.00 50.00 0 0 100

15.4 Robin Pinelli's Sensitivity analysis

183

Figure 15.13 Sensitivity Analysis of Weight-Ratio Input Assumptions

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 X Y Sensitivity Analysis Z AA AB AC AD AE AF AG AH Mag/Income Weight Ratio 1.0 1.5 2.0 Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison Madison

Mag/Snow Weight Ratio

1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00

2.5 Madison Madison Madison Madison Madison Madison Pizza Pizza Pizza

3.0 Madison Madison Madison Madison Madison Pizza Pizza Pizza Pizza

3.5 Madison Madison Madison Madison Pizza Pizza Pizza Pizza Pizza

4.0 Madison Madison Madison Madison Pizza Pizza Pizza Pizza Pizza

4.5 Madison Madison Madison Pizza Pizza Pizza Pizza Pizza Pizza

5.0 Madison Madison Madison Pizza Pizza Pizza Pizza Pizza Pizza

Mag/Snow Weight Ratio

0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25

Mag/Income Weight Ratio 0.25 0.50 0.75 1.00 MPR MPR MPR MPR MPR MPR MPR Madison MPR MPR MPR Madison MPR MPR MPR Madison MPR MPR MPR Madison MPR MPR MPR Madison MPR MPR MPR Madison MPR MPR MPR Madison MPR MPR MPR Madison

1.50 Madison Madison Madison Madison Madison Madison Madison Madison Madison

2.00 Madison Madison Madison Madison Madison Madison Madison Madison Madison

2.50 Madison Madison Madison Madison Madison Madison Madison Madison Madison

3.00 Madison Madison Madison Madison Madison Madison Madison Madison Pizza

3.50 Madison Madison Madison Madison Madison Madison Madison Pizza Pizza

Formulas Y3 Y17

=CHOOSE(B34,"Madison","MPR", "Pizza") =CHOOSE(B34,"Madison","MPR", "Pizza")

Data Tables Y3:AH11 and Y17:AH26 V3 Row Input Cell V2 Column Input Cell

184

Chapter 15 Sensitivity Analysis for Decision Trees

This page is intentionally mostly blank.

Value of Information in Decision Trees

16

16.1 VALUE OF INFORMATION

Useful concept for Evaluation potential information-gathering activities Comparing importance of multiple uncertainties

16.2 EXPECTED VALUE OF PERFECT INFORMATION

Several computational methods Flipping tree, moving an event set of branches, appropriate for any decision tree Payoff table, most appropriate only for single-stage tree (one set of uncertain outcomes with no subsequent decisions) Expected improvement All three methods start by determining Expected Value Under Uncertainty, EVUU, which is the expected value of the optimal strategy without any additional information.

186

Chapter 16 Value of Information in Decision Trees

Figure 16.1 Basic Probability Decision Tree

High Sales Introduce Product

Low Sales Success Prediction

Don't Introduce

High Sales Introduce Product

Low Sales Market Survey Inconclusive

Don't Introduce

High Sales Introduce Product

Low Sales Failure Prediction

Don't Introduce

High Sales Introduce Product

Low Sales Don't Survey

Don't Introduce

16.2 Expected Value of Perfect Information

187

Figure 16.2 DriveTek EVPI Magnetic Success/Failure

Use mechanical method +$80,000 +$80,000 0.5 Electronic success +$150,000 0.5 Awarded contract 2 +$90,000 Try electronic method +$90,000 +$150,000 0.5 Electronic failure +$30,000 +$30,000

0.7
Magnetic success Prepare proposal Try magnetic method +$20,000 +$84,000 +$120,000 +$120,000

0.3
Magnetic failure $0 $0

No Additional Information 1 +$20,000 +$20,000 1 0.5 Not awarded contract -$50,000 -$50,000

Don't prepare proposal $0 $0

Use mechanical method +$80,000 +$80,000 0.5 Electronic success +$150,000 0.5 Awarded contract 3 +$120,000 Try electronic method +$90,000 +$150,000 0.5 Electronic failure +$30,000 +$30,000

1.0
2 +$30,500 Prepare proposal Try magnetic method +$35,000 +$120,000 +$120,000 Magnetic success +$120,000

0.0
Magnetic failure $0

0.7
"Magnetic Success" 1 +$35,000 0.5 Not awarded contract

$0

-$50,000 -$50,000

Don't prepare proposal $0 $0

Use mechanical method +$80,000 +$80,000 0.5 Electronic success +$150,000 +$30,500 0.5 Awarded contract 2 +$90,000 Try electronic method +$90,000 +$150,000 0.5 Electronic failure +$30,000 +$30,000

Perfect Prediction

0.0
Magnetic success Prepare proposal Try magnetic method +$20,000 $0 +$120,000 +$120,000

1.0
Magnetic failure $0

0.3
"Magnetic Failure" 1 +$20,000 0.5 Not awarded contract

$0

-$50,000 -$50,000

Don't prepare proposal $0 $0

188

Chapter 16 Value of Information in Decision Trees

16.3 DRIVETEK POST-CONTRACT-AWARD PROBLEM

DriveTek decided to prepare the proposal, and it turned out that they were awarded the contract. The $50,000 cost and $250,000 up-front payment are in the past. The current decision is to determine which method to use to satisfy the contract. The following decision trees show costs as negative cash flows, so the decision criterion is to maximize expected cash flow. An alternative formulation (not shown here) would show all costs as positive values and would minimize expected cost. Figure 16.3 EVUU
Use mechanical -120000 -120000 -120000 0.5 Electronic success -50000 Try electronic 2 -110000 -50000 -110000 0 -50000

0.5 Electronic failure -170000 -120000 -170000

0.7 Magnetic success -80000 Try magnetic -80000 -116000 0 -80000

0.3 Magnetic failure -200000 -120000 -200000

16.3 DriveTek Post-Contract-Award Problem

189

Figure 16.4 EVPP Elec

Use mechanical -120000 -120000 -120000 1 Electronic success -50000 0.5 "Electronic success" 2 0 -50000 Try electronic -50000 -50000 0 -50000

0 Electronic failure -170000 -120000 -170000

0.7 Magnetic success -80000 Try magnetic -80000 -116000 0 -80000

0.3 Magnetic failure -200000 -120000 -200000

-83000

Use mechanical -120000 -120000 -120000 0 Electronic success -50000 0.5 "Electronic failure" 3 0 -116000 Try electronic -50000 -170000 0 -50000

1 Electronic failure -170000 -120000 -170000

0.7 Magnetic success -80000 Try magnetic -80000 -116000 0 -80000

0.3 Magnetic failure -200000 -120000 -200000

190

Chapter 16 Value of Information in Decision Trees

Figure 16.5 EVPP Mag

Use mechanical -120000 -120000 -120000 0.5 Electronic success -50000 0.7 "Magnetic success" 3 0 -80000 Try electronic -50000 -110000 0 -50000

0.5 Electronic failure -170000 -120000 -170000

1 Magnetic success -80000 Try magnetic -80000 -80000 0 -80000

0 Magnetic failure -200000 -120000 -200000

-89000

Use mechanical -120000 -120000 -120000 0.5 Electronic success -50000 0.3 "Magnetic failure" 2 0 -110000 Try electronic -50000 -110000 0 -50000

0.5 Electronic failure -170000 -120000 -170000

0 Magnetic success -80000 Try magnetic -80000 -200000 0 -80000

1 Magnetic failure -200000 -120000 -200000

16.3 DriveTek Post-Contract-Award Problem

191

Figure 16.6 EVPP Both

EVPP Both Use mechanical -120000 -120000 -120000 1 Electronic success -50000 0.7 "Magnetic success" 2 0 -50000 Try electronic -50000 -50000 0 -50000

0 Electronic failure -170000 -120000 -170000

1 Magnetic success -80000 Try magnetic 0.5 "Electronic success" 0 -50000 -80000 -80000 0 -80000

0 Magnetic failure -200000 -120000 -200000

Use mechanical -120000 -120000 -120000 1 Electronic success -50000 0.3 "Magnetic failure" 2 0 -50000 Try electronic -50000 -50000 0 -50000

0 Electronic failure -170000 -120000 -170000

0 Magnetic success -80000 Try magnetic -80000 -200000 0 -80000

1 Magnetic failure -200000 -120000 -200000

-71000

Use mechanical -120000 -120000 -120000 0 Electronic success -50000 0.7 "Magnetic success" 3 0 -80000 Try electronic -50000 -170000 0 -50000

1 Electronic failure -170000 -120000 -170000

1 Magnetic success -80000 Try magnetic 0.5 "Electronic failure" 0 -92000 -80000 -80000 0 -80000

0 Magnetic failure -200000 -120000 -200000

Use mechanical -120000 -120000 -120000 0 Electronic success -50000 0.3 "Magnetic failure" 1 0 -120000 Try electronic -50000 -170000 0 -50000

1 Electronic failure -170000 -120000 -170000

0 Magnetic success -80000 Try magnetic -80000 -200000 0 -80000

1 Magnetic failure -200000 -120000 -200000

192

Chapter 16 Value of Information in Decision Trees

16.4 SENSITIVITY ANALYSIS VS EVPI

Working Paper Title: Do Sensitivity Analyses Really Capture Problem Sensitivity? An Empirical Analysis Based on Information Value Authors: James C. Felli, Naval Postgraduate School and Gordon B. Hazen, Northwestern University Date: March 1998 The most common methods of sensitivity analysis (SA) in decision-analytic modeling are based either on proximity in parameter-space to decision thresholds or on the range of payoffs that accompany parameter variation. As an alternative, we propose the use of the expected value of perfect information (EVPI) as a sensitivity measure and argue from first principles that it is the proper measure of decision sensitivity. EVPI has significant advantages over conventional SA, especially in the multiparametric case, where graphical SA breaks down. In realistically sized problems, simple one- and two-way SAs may not fully capture parameter interactions, raising the disturbing possibility that many published decision analyses might be overconfident in their policy recommendations. To investigate the extent of this potential problem, we re-examined 25 decision analyses drawn from the published literature and calculated EVPI values for parameters on which sensitivity analyses had been performed, as well as the entire set of problem parameters. While we expected EVPI values to indicate greater problem sensitivity than conventional SA due to revealed parameter interaction, we in fact found the opposite: compared to EVPI, the one- and two-parameter SAs accompanying these problems dramatically overestimated problem sensitivity to input parameters. This phenomenon can be explained by invoking the flat maxima principle enunciated by von Winterfeldt and Edwards. http://www.mccombs.utexas.edu/faculty/jim.dyer/DA_WP/WP980019.pdf

Value of Imperfect Information

17

17.1 TECHNOMETRICS PROBLEM

Prior Problem
Technometrics, Inc., a large producer of electronic components, is having some problems with the manufacturing process for a particular component. Under its current production process, 25 percent of the units are defective. The profit contribution of this component is $40 per unit. Under the contract the company has with its customers, Technometrics refunds $60 for each component that the customer finds to be defective; the customers then repair the component to make it usable in their applications. Before shipping the components to customers, Technometrics could spend an additional $30 per component to rework any components thought to be defective (regardless of whether the part is really defective). The reworked components can be sold at the regular price and will definitely not be defective in the customers' applications. Unfortunately, Technometrics cannot tell ahead of time which components will fail to work in their customers' applications. The following payoff table shows Technometrics' net cash flow per component. Figure 17.1 Payoff Table

Component Condition Good Defective

What should Technometrics do?

Technometrics' Choice Ship as is Rework first +$40 +$10 -$20 +$10

How much should Technometrics be willing to pay for a test that could evaluate the condition of the component before making the decision to ship as is or rework first?

194

Chapter 17 Value of Imperfect Information

Imperfect Information
An engineer at Technometrics has developed a simple test device to evaluate the component before shipping. For each component, the test device registers positive, inconclusive, or negative. The test is not perfect, but it is consistent for a particular component; that is, the test yields the same result for a given component regardless of how many times it is tested. To calibrate the test device, it was run on a batch of known good components and on a batch of know defective components. The results in the table below, based on relative frequencies, show the probability of a test device result, conditional on the true condition of the component. Figure 17.2 Likelihoods

Test Result Positive Inconclusive Negative

Component Condition Good Defective 0.70 0.10 0.20 0.30 0.10 0.60

For example, of the known defective components tested, sixty percent had a negative test result. An analyst at Technometrics suggested using Bayesian revision of probabilities to combine the assessments about the reliability of the test device (shown above) with the original assessment of the components' condition (25 percent defectives). Technometrics uses expected monetary value for making decisions under uncertainty. What is the maximum (per component) the company should be willing to pay for using the test device?

17.1 Technometrics Problem

195

Figure 17.3 Decision Tree Model

A B C D 1 EVSI 2.25 2 $ 3 4 5 6 7 No add'l info 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 2 24 $27.25 25 26 27 28 29 30 31 32 33 34 35 36 37 Test 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 E F G H I J K L 0.7500 Good M N O P Q R S

$40.00 Ship as is $25.00 0.2500 Defective -$20.00 1 $25.00 EVUU Rework first $10.00 $10.00 0.9545 Good $40.00 Ship as is $37.27 0.5500 Positive 1 $37.27 Rework first $10.00 $10.00 0.6667 Good $40.00 Ship as is $20.00 0.2250 Inconclusive 1 $27.25 EVSP $20.00 Rework first $10.00 $10.00 0.3333 Good $40.00 Ship as is $0.00 0.2250 Negative 2 $10.00 Rework first $10.00 $10.00 0.6667 Defective -$20.00 -$20.00 $40.00 0.3333 Defective -$20.00 -$20.00 $40.00 0.0455 Defective -$20.00 -$20.00 $40.00 -$20.00 $40.00

196

Chapter 17 Value of Imperfect Information

Revision of Probability
Figure 17.4 Display
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 U Prior Likelihood Positive Inconclusive Negative Joint Positive Inconclusive Negative Posterior Positive Inconclusive Negative V 0.75 Good 0.7 0.2 0.1 Good 0.525 0.150 0.075 Good 0.9545 0.6667 0.3333 W X Y 0.25 = P(Main) Bad 0.1 = P(Info | Main) 0.3 0.6 Bad Preposterior 0.025 0.550 = P(Info) 0.075 0.225 0.150 0.225 Bad 0.0455 = P(Main | Info) 0.3333 0.6667

Figure 17.5 Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 U Prior Likelihood Positive Inconclusive Negative Joint Positive Inconclusive Negative Posterior Positive Inconclusive Negative V 0.75 Good 0.7 0.2 0.1 Good =V$1*V3 =V$1*V4 =V$1*V5 Good =V8/$X8 =V9/$X9 =V10/$X10 0.25 Bad 0.1 0.3 0.6 Bad =W$1*W3 =W$1*W4 =W$1*W5 Bad =W8/$X8 =W9/$X9 =W10/$X10 W X = P(Main) = P(Info | Main) Y

Preposterior =SUM(V8:W8) =SUM(V9:W9) =SUM(V10:W10)

= P(Info)

= P(Main | Info)

Modeling Attitude Toward Risk

18

18.1 RISK UTILITY FUNCTION

A certainty equivalent is a certain payoff value which is equivalent, for the decision maker, to a particular payoff distribution. If the decision maker can determine his or her certainty equivalent for the payoff distribution of each strategy in a decision problem, then the optimal strategy is the one with the highest certainty equivalent. The certainty equivalent, i.e., the minimum selling price for a payoff distribution, depends on the decision maker's personal attitude toward risk. A decision maker may be risk preferring, risk neutral, or risk avoiding. If the terminal values are not regarded as extreme relative to the decision maker's total assets, if the decision maker will encounter other decision problems with similar payoffs, and if the decision maker has the attitude that he or she will "win some and lose some," then the decision maker's attitude toward risk may be described as risk neutral. If the decision maker is risk neutral, the certainty equivalent of a payoff distribution is equal to its expected value. The expected value of a payoff distribution is calculated by multiplying each terminal value by its probability and summing the products. If the terminal values in a decision situation are extreme or if the situation is "one-of-akind" so that the outcome has major implications for the decision maker, an expected value analysis may not be appropriate. Such situations may require explicit consideration of risk. Unfortunately, it can be difficult to determine one's certainty equivalent for a complex payoff distribution. We can aid the decision maker by first determining his or her certainty equivalent for a simple payoff distribution and then using that information to infer the certainty equivalent for more complex payoff distributions. A utility function, U(x), can be used to represent a decision maker's attitude toward risk. The values or certainty equivalents, x, are plotted on the horizontal axis; utilities or expected utilities, u or U(x), are on the vertical axis. You can use the plot of the function

198

Chapter 18 Modeling Attitude Toward Risk

by finding a value on the horizontal axis, scanning up to the plotted curve, and looking left to the vertical axis to determine the utility. A typical risk utility function might have the general shape shown below if you draw a smooth curve approximately through the points. Figure 18.1 Typical Risk Utility Function
1.0 0.9 0.8 Utility U(x) or Expected Utility 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 -$50,000

-$25,000

$0

$25,000

$50,000

$75,000

$100,000

$125,000 $150,000

Monetary Value x or Certainty Equivalent

Since more value generally means more utility, the utility function is monotonically nondecreasing, and its inverse is well-defined. On the plot of the utility function, you locate a utility on the vertical axis, scan right to the plotted curve, and look down to read the corresponding value. The concept of a payoff distribution, risk profile, gamble, or lottery is important for discussing utility functions. A payoff distribution is a set of payoffs, e.g., x1, x2, and x3, with corresponding probabilities, P(X=x1), P(X=x2), and P(X=x3). For example, a payoff distribution may be represented in decision tree form as shown below.

18.1 Risk Utility Function

199

Figure 18.2 Figure 2 Payoff Distribution Probability Tree

P(X=x 1) x1 P(X=x 2 ) x2 P(X=x 3 ) x3

The fundamental property of a utility function is that the utility of the certainty equivalent CE of a payoff distribution is equal to the expected utility of the payoffs, i.e, U(CE) = P(X=x1)*U(x1) + P(X=x2)*U(x2) + P(X=x3)*U(x3). It follows that if you compute the expected utility (EU) of a lottery, EU = P(X=x1)*U(x1) + P(X=x2)*U(x2) + P(X=x3)*U(x3), the certainty equivalent of the payoff distribution can be determined using the inverse of the utility function. That is, you locate the expected utility on the vertical axis, scan right to the plotted curve, and look down to read the corresponding certainty equivalent. If a utility function has been determined, you can use this fundamental property to determine the certainty equivalent of any payoff distribution. Calculations for the Magnetic strategy in the DriveTek problem are shown below. First, using a plot of the utility function, locate each payoff x on the horizontal axis and determine the corresponding utility U(x) on the vertical axis. Second, compute the expected utility EU of the lottery by multiplying each utility by its probability and summing the products. Third, locate the expected utility on the vertical axis and determine the corresponding certainty equivalent CE on the horizontal axis. Figure 18.3 Calculations Using Risk Utility Function P(X=x) x U(x) P(X=x)*U(x) 0.50 -$50,000 0.00 0.0000 0.15 $0 0.45 0.0675 0.35 $120,000 0.95 0.3325 0.4000 EU -$8,000 CE

200

Chapter 18 Modeling Attitude Toward Risk

18.2 EXPONENTIAL RISK UTILITY

Instead of using a plot of a utility function, an exponential function may be used to represent risk attitude. The general form of the exponential utility function is U(x) = A B*EXP(x/RT). The risk tolerance parameter RT determines the curvature of the utility function reflecting the decision makers attitude toward risk. Subsequent sections cover three methods for determining RT. EXP is Excel's standard exponential function, i.e., EXP(z) represents the value e raised to the power of z, where e is the base of the natural logarithms. The parameters A and B determine scaling. After RT is determined, if you want to plot a utility function so that U(High) = 1.0 and U(Low) = 0.0, you can use the following formulas to determine the scaling parameters A and B. A = EXP (Low/RT) / [EXP (Low/RT) EXP (High/RT)] B = 1 / [EXP (Low/RT) EXP (High/RT)] The inverse function for finding the certainty equivalent CE corresponding to an expected utility EU is CE = RT*LN[(AEU)/B], where LN(y) represents the natural logarithm of y. After the parameters A, B, and RT have been determined, the exponential utility function and its inverse can be used to determine the certainty equivalent for any lottery. Calculations for the Magnetic strategy in the DriveTek problem are shown in Figure 4.

18.2 Exponential Risk Utility

201

Figure 18.4 Exponential Risk Utility Results

Computed values are displayed with four decimal places, but Excel's 15-digit precision is used in all calculations. For a decision maker with a risk tolerance parameter of $100,000, the payoff distribution for the Magnetic strategy has a certainty equivalent of -$7,676. That is, if the decision maker is facing the payoff distribution shown in A9:B12 in Figure 4, he or she would be willing to pay $7,676 to be relieved of the obligation. Formulas are shown in Figure 5. To construct the worksheet, enter the text in column A and the monetary values in column B. To define names, select A2:B4, and choose Insert | Name | Create. Similarly, select A6:B7, and choose Insert | Name | Create. Then enter the formulas in B6:B7. Enter formulas in C10 and D10, and copy down. Finally, enter the EU formula in D13 and the CE formula in D15. The defined names are absolute references by default.

202

Chapter 18 Modeling Attitude Toward Risk

Figure 18.5 Exponential Risk Utility Formulas

Figure 6 shows results for the same payoff distribution using a simplified form of the exponential risk utility function with A = 1 and B = 1. This function could be represented as U(x) = 1EXP(x/RT) with inverse CE = RT*LN(1EU). The utility and expected utility calculations are different, but the certainty equivalent is the same. Figure 18.6 Simplified Exponential Risk Utility Results

18.3 Approximate Risk Tolerance

203

18.3 APPROXIMATE RISK TOLERANCE

The value of the risk tolerance parameter RT is approximately equal to the maximum value of Y for which the decision maker is willing to accept a payoff distribution with equally-likely payoffs of $Y and $Y/2 instead of accepting $0 for certain. Figure 18.7 Approximate Risk Tolerance
0.5 Heads +$Y Play 0.5 Tails -$Y/2

Don't $0

For example, in a personal decision, you may be willing to play the game shown in Figure 7 with equally-likely payoffs of $100 and $50, but you might not play with payoffs of $100,000 and $50,000. As the better payoff increases from $100 to $100,000 (and the corresponding worse payoff increases from $50 to $50,000), you reach a value where you are indifferent between playing the game and receiving $0 for certain. At that point, the value of the better payoff is an approximation of RT for an exponential risk utility function describing your risk attitude. In a business decision for a small company, the company may be willing to play the game with payoffs of $200,000 and $100,000 but not with payoffs of $20,000,000 and -$10,000,000. Somewhere between a better payoff of $200,000 and $20,000,000, the company would be indifferent between playing the game and not playing, thereby determining the approximate RT for their business decision.

18.4 EXACT RISK TOLERANCE USING EXCEL

A simple payoff distribution, called a risk attitude assessment lottery, may be used to determine the decision maker's attitude toward risk. This lottery has equal probability of obtaining each of the two payoffs. It is good practice to use a better payoff at least as large as the highest payoff in the decision problem and a worse payoff as small as or smaller than the lowest payoff. In any case, the payoffs should be far enough apart that the decision maker perceives a definite difference in the two outcomes. Three values must be specified for the fifty-fifty lottery: the Better payoff, the Worse payoff, and the Certainty Equivalent, as shown in Figure 8.

204

Chapter 18 Modeling Attitude Toward Risk

Figure 18.8 Risk Attitude Assessment Lottery

0.5 Better Payoff Certainty Equivalent = 0.5 Worse Payoff

According to the fundamental property of a risk utility function, the utility of the certainty equivalent equals the expected utility of the lottery, so the three values are related as follows. U(CertEquiv) = 0.5*U(BetterPayoff) + 0.5*U(WorsePayoff) If you use the general form for an exponential utility function with parameters A, B, and RT, and if you simplify terms, it follows that RT must satisfy the following equation. Exp(CertEquiv/RT) = 0.5*Exp(BetterPayoff/RT) + 0.5*Exp(WorsePayoff/RT) Given the values for CE, Better, and Worse, you could use trial-and-error to find the value of RT that exactly satisfies the equation. In Excel you can use Goal Seek or Solver by creating a worksheet like Figure 9. Enter the text in column A. Enter the assessment lottery values in B2:B4. Enter a tentative RT value in B6. Select A2:B4, and use Insert | Name | Create; repeat for A6:B6 and A8:B9. Note that the parentheses symbol is not allowed in a defined name, so Excel changes U(CE) to U_CE and EU(Lottery) to EU_Lottery.

18.4 Exact Risk Tolerance Using Excel

205

Figure 18.9 Formulas for Risk Tolerance Search

Figure 18.10 Tentative Values for Risk Tolerance Search

Figure 10 shows tentative values for the search. From the Tools menu, choose Goal Seek. In the Goal Seek dialog box, enter B11, 0, and B6. If you point to cells, the reference appears in the edit box as an absolute reference, as shown in Figure 11. Click OK.

206

Chapter 18 Modeling Attitude Toward Risk

Figure 18.11 Goal Seek Dialog Box

The Goal Seek Status dialog box shows that a solution has been found. Click OK. The worksheet appears as shown in Figure 12. Figure 18.12 Results of Goal Seek Search

The difference between U(CE) and EU(Lottery) is not exactly zero. If you start at $250,000, the Goal Seek converges to a difference of 6.2E05 or 0.000062, which is closer to zero, resulting in a RT of $243,041. If extra precision is needed, use Solver. With Solver's default settings, the difference is 2.39E08 with RT equal to $243,261. If you change the precision from 0.000001 to 0.00000001 or an even smaller value in Solver's Options, the difference will be even closer to zero.

18.5 Exact Risk Tolerance Using RiskTol.xla

207

18.5 EXACT RISK TOLERANCE USING RISKTOL.XLA

The Goal Seek and Solver methods for determining the risk tolerance parameter RT yield static results. For a dynamic result, use the risktol.xla add-in function. A major advantage of risktol.xla is that it facilitates sensitivity analysis. Whenever an input to the function changes, the result is recalculated. The function syntax is RISKTOL(WorsePayoff,CertEquiv,BetterPayoff,BetterProb). When you open the risktol.xla file, the function is added to the Math & Trig function category list. The function returns a very precise value of the risk tolerance parameter for an exponential utility function. The result is consistent with CertEquiv as the decision makers certainty equivalent for a two-payoff assessment lottery with payoffs WorsePayoff and BetterPayoff, with probability BetterProb of obtaining BetterPayoff and probability 1 BetterProb of obtaining WorsePayoff. In case of an error, the RISKTOL function returns: #N/A if there are too few or too many arguments. The first three arguments (WorsePayoff, CertEquiv, and BetterPayoff) are required; the fourth argument (BetterProb) is optional, with default value 0.5. #VALUE! if WorsePayoff >= CertEquiv, or CertEquiv >= Better Payoff, or BetterProb (if specified) <= 0 or >= 1. #NUM! if the search procedure fails to converge. In Figure 13, the text in cells A2:A4 has been used as defined names for cells B2:B4, and the text in cell A6 is the defined name for cell B6, as shown in the name box. After opening the risktol.xla file, enter the function name and arguments, as shown in the formula bar. If one of the three inputs change, the result in cell B6 is recalculated. Figure 18.13 Exact Risk Tolerance Using RiskTol.xla

208

Chapter 18 Modeling Attitude Toward Risk

18.6 EXPONENTIAL UTILITY AND TREEPLAN

TreePlan's default is to rollback the tree using expected values. If you choose to use exponential utilities in TreePlan's Options dialog box, TreePlan will redraw the decision tree diagram with formulas for computing the utility and certainty equivalent at each node. For the Maximize option, the rollback formulas are U = AB*EXP(X/RT) and CE = -LN((A-EU)/B)*RT, where X and EU are cell references. For the Minimize option, the formulas are U = A-B*EXP(X/RT) and CE = LN((A-EU)/B)*RT. TreePlan uses the name RT to represent the risk tolerance parameter of the exponential utility function. The names A and B determine scaling. If the names A, B, and RT don't exist on the worksheet when you choose to use exponential utility, they are initially defined as A=1, B=1, and RT=999999999999. You can redefine the names using the Insert | Name | Define or Insert | Name | Create commands. To plot the utility curve, enter a list of X values in a column on the left, and enter the formula =AB*EXP(X/RT) in a column on the right, where X is a reference to the corresponding cell on the left. Select the values in both columns, and use the ChartWizard to develop an XY (Scatter) chart. If RT is specified using approximate risk tolerance values, you can perform sensitivity analysis by (1) using the defined name RT for a cell, (2) constructing a data table with a list of possible RT values and an appropriate output formula (usually a choice indicator at a decision node or a certainty equivalent), and (3) specifying the RT cell as the input cell in the Data Table dialog box.

18.7 EXPONENTIAL UTILITY AND RISKSIM

After using RiskSim to obtain model output results, select the column containing the Sorted Data, copy to the clipboard, select a new sheet, and paste. Alternatively, you can use the unsorted values, and you can also do the following calculations on the original sheet containing the model results. This example uses only ten iterations; 500 or 1,000 iterations are more appropriate. Use one of the methods described previously to specify values of RT, A, and B. Since the model output values shown in Figures 14 and 15 range from approximately $14,000 to $176,000, the utility function is defined for a range from worse payoff $0 to better payoff $200,000. RT was determined using risktol.xla with a risk-seeking certainty equivalent of $110,000. To obtain the utility of each model output value in cells A2:A11, select cell B2, and enter the formula =AB*EXP(A2/RT). Select cell B2, click the fill handle in the lower right corner of the cell and drag down to cell B11. Enter the formulas in cells A13:C13 and the labels in row 14.

18.7 Exponential Utility and RiskSim

209

Figure 18.14 Risk Utility Formulas for RiskSim

1 2 3 4 5 6 7 8 9 10 11 12 13 14 A Sorted Data 14229.56 32091.92 51091.48 66383.79 69433.32 87322.23 95920.93 135730.71 154089.36 175708.87 B Utility =A-B*EXP(-A2/RT) =A-B*EXP(-A3/RT) =A-B*EXP(-A4/RT) =A-B*EXP(-A5/RT) =A-B*EXP(-A6/RT) =A-B*EXP(-A7/RT) =A-B*EXP(-A8/RT) =A-B*EXP(-A9/RT) =A-B*EXP(-A10/RT) =A-B*EXP(-A11/RT) C

=AVERAGE(A2:A11) =AVERAGE(B2:B11) =-LN((A-B13)/B)*RT Exp. Value Exp.Util. CE

Figure 18.15 Risk Utility Results for RiskSim

1 2 3 4 5 6 7 8 9 10 11 12 13 14 A Sorted Data $ 14,230 $ 32,092 $ 51,091 $ 66,384 $ 69,433 $ 87,322 $ 95,921 $ 135,731 $ 154,089 $ 175,709 $ 88,200 Exp. Value B Utility 0.05862 0.13462 0.21851 0.28841 0.30260 0.38767 0.42966 0.63382 0.73363 0.85600 C

0.40435 $ 90,757 Exp.Util. CE

210

Chapter 18 Modeling Attitude Toward Risk

18.8 RISK SENSITIVITY FOR MACHINE PROBLEM

Figure 18.16
A 1 Process 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 B NPV $107,733 $39,389 $125,210 $66,032 $32,504 $138,132 $83,000 $48,178 $20,130 $31,445 $19,739 $4,641 $92,368 $102,585 $106,411 $110,528 $171,524 $87,698 $123,907 $69,783 $144,052 $131,461 $34,938 $75,551 $32,144 $61,719 $139,568 $89,107 $94,158 $81,459 $139,258 $58,190 -$13,104 $36,529 $91,239 $147,155 $154,168 $180,770 $112,313 C Utility 0.102133 0.038623 0.117689 0.063899 0.031982 0.129016 0.079649 0.047036 0.019928 0.030956 0.019546 0.00463 0.08823 0.097498 0.100945 0.104639 0.15762 0.083963 0.116538 0.067404 0.134157 0.123187 0.034335 0.072768 0.031633 0.059853 0.130266 0.085252 0.089861 0.07823 0.129997 0.056529 -0.01319 0.035869 0.0872 0.13684 0.142872 0.165372 0.106235 D E Process 2 F NPV $86,161 $58,417 $171,058 $263,843 $37,180 $254,027 $118,988 $133,862 $26,597 $187,063 $88,060 $114,837 $130,638 $138,882 $226,909 $156,102 $193,209 $92,004 $163,780 $22,176 $135,190 $61,013 $184,907 $70,967 -$10,251 $89,645 $119,405 $96,670 $114,124 $208,778 $24,580 $155,958 $198,519 $167,568 $36,676 $225,777 $195,738 $53,467 $213,920 G Utility 0.082554 0.056744 0.157228 0.231906 0.036498 0.224329 0.112181 0.125289 0.026247 0.170608 0.084294 0.108489 0.122465 0.12967 0.203006 0.144528 0.17569 0.087898 0.151071 0.021932 0.12645 0.059189 0.168819 0.068507 -0.010304 0.085744 0.112551 0.092144 0.107853 0.188425 0.02428 0.144405 0.180056 0.154281 0.036011 0.202104 0.177773 0.052063 0.192587 H I RT $1,000,000 J K L AJS, Clemen2 pp. 428-430

Process 1 Process 2 ExpUtility CertEquiv ExpValue 0.085527 0.107258 $89,407 $113,458 $90,526 $116,159

Goal Seek CE2 - CE1 $24,050

NPV values from RiskSim Summary Cell I2 has defined name RT Formulas C2 =1-EXP(-B2/RT) Copy down to C1001 G2 =1-EXP(-F2/RT) Copy down to G1001 J6 =AVERAGE(C2:C1001) K6 =AVERAGE(G2:G1001) J8 =-RT*LN(1-J6) K8 =-RT*LN(1-K6) J10 =AVERAGE(B2:B1001) K10 =AVERAGE(F2:F1001) J16 =K8-J8

18.9 Risk Utility Summary

211

Figure 18.17
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 M N O RiskTolerance CE Process 1 CE Process 2 $5,000 $10,000 $15,000 $20,000 $25,000 $30,000 $35,000 $40,000 $45,000 $50,000 $55,000 $60,000 $65,000 $70,000 $75,000 $80,000 $85,000 $90,000 $95,000 $100,000 Formulas N2 O2 Data Table I2 -$25,597 $3,504 $23,904 $37,468 $46,811 $53,528 $58,541 $62,404 $65,459 $67,930 $69,966 $71,672 $73,119 $74,363 $75,443 $76,389 $77,224 $77,966 $78,631 $79,229 -$37,262 -$10,097 $10,897 $26,409 $38,010 $46,998 $54,184 $60,067 $64,972 $69,122 $72,675 $75,749 $78,431 $80,791 $82,882 $84,746 $86,417 $87,924 $89,288 $90,529 P Q R S T U V W X Y Z
AJS

$100,000 Process 2 $80,000 Process 1 $60,000 Certainty Equivalent

$40,000

$20,000

$0

-$20,000

-$40,000

=J8 =K8

$0

$10,000

$20,000

$30,000

$40,000

$50,000

$60,000

$70,000

$80,000

$90,000

$100,000

Risk Tolerance Parameter for Exponential Utility

Column Input Cell

18.9 RISK UTILITY SUMMARY

Concepts
Strategy, Payoff Distribution, Certainty Equivalent Figure 18.18 Utility Function
Utility Function 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 -50000

Utility or Expected Utility, U(x)

-25000

25000

50000

75000

100000

125000

150000

Value or Certainty Equivalent, x

212

Chapter 18 Modeling Attitude Toward Risk

Fundamental Property of Utility Function

The utility of the CE of a lottery equals the expected utility of the lottery's payoffs. U(CE) = EU = p1*U(x1) + p2*U(x2) + p3*U(x3)

Using a Utility Function To Find the CE of a Lottery

1. 2. 3. U(x): Locate each payoff on the horizontal axis and determine the corresponding utility on the vertical axis. EU: Compute the expected utility of the lottery by multiplying each utility by its probability and summing the products. CE: Locate the expected utility on the vertical axis and determine the corresponding certainty equivalent on the horizontal axis.

Exponential Utility Function

General form: U(x) = A B*EXP(x/RT) Parameters A and B affect scaling. Parameter RT (RiskTolerance) depends on risk attitude and affects curvature. Inverse: CE = RT*LN[(AEU)/B]

TreePlan's Simple Form of Exponential Utility

Set A and B equal to 1. U(x) = 1 EXP(x/RT) CE = RT*LN(1EU)

Approximate Assessment of RiskTolerance

Refer to the Clemen textbook, Figure 13.12, on page 478.

18.9 Risk Utility Summary

213

Figure 18.19 Assessing ApproximateRisk Tolerance

Risk tolerance parameter for an exponential utility function is approximately equal to the maximum amount Y for which the decision maker will play. 0.5 Heads +$Y Play 0.5 Tails -$Y/2
-$5 -$50 -$500 -$5,000 -$50,000 -$100,000 -$150,000 +$10 +$100 +$1,000 +$10,000 +$100,000 +$200,000 +$300,000

Don't $0
$0 $0 $0 $0 $0 $0 $0

more risk aversion

less risk aversion

Exact Assessment of RiskTolerance

The RISKTOL.XLA Excel add-in file adds the following function to the Math & Trig function category list: RISKTOL(WorsePayoff,CertEquiv,BetterPayoff,BetterProb) The first three arguments are required, and the last argument is optional with default value 0.5. WorsePayoff and BetterPayoff are payoffs of an assessment lottery, and CertEquiv is the decision maker's certainty equivalent for the lottery. RISKTOL returns #N/A if there are too few or too many arguments, #VALUE! if WorsePayoff >= CertEquiv, or CertEquiv >= Better Payoff, or BetterProb (if specified) <= 0 or >= 1, and #NUM! if the search procedure fails to converge. For example, consider a 50-50 lottery with payoffs of $100,000 and $0. A decision maker has decided that the certainty equivalent is $43,000. If you open the RISKTOL.XLA file and type =RISKTOL(0,43000,100000) in a cell, the result is 176226. Thus, the value of the RiskTolerance parameter in an exponential utility function for this decision maker should be 176226.

Using Exponential Utility for TreePlan Rollback Values

1. 2. 3. Select a cell, and enter a value for the RiskTolerance parameter. With the cell selected, choose Insert Name | Define, and enter RT. From TreePlan's Options dialog box, select Use Exponential Utility. The new decision tree diagram includes the EXP and LN functions for determining U(x) and the inverse.

214

Chapter 18 Modeling Attitude Toward Risk

Using Exponential Utility for a Payoff Distribution

Enter the exponential utility function directly, using the appropriate value for RiskTolerance. If the payoff values are equally-likely, use the AVERAGE function to determine the expected utility; otherwise, use SUMPRODUCT. Enter the inverse function directly to obtain the certainty equivalent.

Modeling Marketing Decisions

19

19.1 ALLOCATING ADVERTISING EXPENDITURES

Figure 19.1 Quick Tour
A B C D E F G H I 1 Quick Tour of Microsoft Excel Solver Q1 Q2 Q3 Q4 Total 2 Month 0.9 1.1 0.8 1.2 3 Seasonality 4 3,592 4,390 3,192 4,789 15,962 5 Units Sold $143,662 $175,587 $127,700 $191,549 $638,498 6 Sales Revenue 89,789 109,742 79,812 119,718 399,061 7 Cost of Sales 53,873 65,845 47,887 71,831 239,437 8 Gross Margin 9 8,000 8,000 9,000 9,000 34,000 10 Salesforce 10,000 10,000 10,000 10,000 40,000 11 Advertising 21,549 26,338 19,155 28,732 95,775 12 Corp Overhead 39,549 44,338 38,155 47,732 169,775 13 Total Costs 14 $14,324 $21,507 $9,732 $24,099 $69,662 15 Prod. Profit 10% 12% 8% 13% 11% 16 Profit Margin 17 $40.00 Product Price 18 $25.00 Product Cost 19 20 21 The following exam ples show you how to work with the m odel above to solve for one value or several 22 values to m axim ize or m inim ize another value, enter and change constraints, and save a problem m odel. 23

216

Chapter 19 Modeling Marketing Decisions

23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47

Row 3

Contains Fixed values

Explanation Seasonality factor: sales are higher in quarters 2 and 4, and lower in quarters 1 and 3.

=35*B3*(B11+3000)^ 0.5 Forecast for units sold each quarter: row 3 contains the seasonality factor; row 11 contains the cost of advertising.

=B5*$B$18

Sales revenue: forecast for units sold (row 5) tim es price (cell B18).

=B5*$B$19

Cost of sales: forecast for units sold (row 5) tim es product cost (cell B19).

=B6-B7

Gross m argin: sales revenues (row 6) m inus cost of sales (row 7).

10 11 12

Fixed values Fixed values =0.15*B6

Sales personnel expenses. Advertising budget (about 6.3% of sales). Corporate overhead expenses: sales revenues (row 6) tim es 15%.

19.1 Allocating Advertising Expenditures

217

A B C D E F G H I 13 =SUM(B10:B12) Total costs: sales personnel expenses (row 10) plus 48 advertising (row 11) plus overhead (row 12). 49 50 15 =B8-B13 Product profit: gross m argin (row 8) m inus total costs 51 (row 13). 52 53 16 =B15/ B6 Profit m argin: profit (row 15) divided by sales revenue 54 (row 6). 55 56 18 Fixed values Product price. 57 58 19 Fixed values Product cost. 59 60 61 This is a typical m arketing m odel that shows sales rising from a base figure (perhaps due to the sales 62 personnel) along with increases in advertising, but with dim inishing returns. For exam ple, the first 63 $5,000 of advertising in Q1 yields about 1,092 increm ental units sold, but the next $5,000 yields only 64 about 775 units m ore. 65 66 You can use Solver to find out whether the advertising budget is too low, and whether advertising 67 should be allocated differently over tim e to take advantage of the changing seasonality factor. 68 69 Solving for a Value to Maximize Another Value 70 One way you can use Solver is to determ ine the m axim um value of a cell by changing another cell. The 71 two cells m ust be related through the form ulas on the worksheet. If they are not, changing the value in 72 one cell will not change the value in the other cell. 73 74 For exam ple, in the sam ple worksheet, you want to know how m uch you need to spend on advertising 75 to generate the m axim um profit for the first quarter. You are interested in m axim izing profit by changing 76 advertising expenditures. 77 On the Tools m enu, click Solver. In the Set target cell box, type b15 or 78 select cell B15 (first-quarter profits) on the worksheet. Select the Max option. 79 In the By changing cells box, type b11 or select cell B11 (first-quarter advertising) 80 on the worksheet. Click Solve. 81 82 83 You will see m essages in the status bar as the problem is set up and Solver starts working. After a 84 m om ent, you'll see a m essage that Solver has found a solution. Solver finds that Q1 advertising of 85 $17,093 yields the m axim um profit $15,093. 86 After you exam ine the results, select Restore original values and click OK to 87 discard the results and return cell B11 to its form er value. 88 89 90 Resetting the Solver Options 91 92 If you want to return the options in the Solver Parameters dialog box to their original settings so that 93 you can start a new problem , you can click Reset All. 94

218

Chapter 19 Modeling Marketing Decisions

A B C D E F G H I 95 Solving for a Value by Changing Several Values 96 97 You can also use Solver to solve for several values at once to m axim ize or m inim ize another value. For 98 exam ple, you can solve for the advertising budget for each quarter that will result in the best profits for 99 the entire year. Because the seasonality factor in row 3 enters into the calculation of unit sales in row 5 100 as a m ultiplier, it seem s logical that you should spend m ore of your advertising budget in Q4 when the 101 sales response is highest, and less in Q3 when the sales response is lowest. Use Solver to determ ine 102 the best quarterly allocation. 103 On the Tools m enu, click Solver. In the Set target cell box, type f15 or select 104 cell F15 (total profits for the year) on the worksheet. Make sure the Max option is 105 selected. In the By changing cells box, type b11:e11 or select cells B11:E11 106 (the advertising budget for each of the four quarters) on the worksheet. Click Solve. 107 108 After you exam ine the results, click Restore original values and click OK to 109 discard the results and return all cells to their form er values. 110 111 112 You've just asked Solver to solve a m oderately com plex nonlinear optim ization problem ; that is, to find 113 values for the four unknowns in cells B11 through E11 that will m axim ize profits. (This is a nonlinear 114 problem because of the exponentiation that occurs in the form ulas in row 5). The results of this 115 unconstrained optim ization show that you can increase profits for the year to $79,706 if you spend 116 $89,706 in advertising for the full year. 117 118 However, m ost realistic m odeling problem s have lim iting factors that you will want to apply to certain 119 values. These constraints m ay be applied to the target cell, the changing cells, or any other value that 120 is related to the form ulas in these cells. 121 122 Adding a Constraint 123 124 So far, the budget recovers the advertising cost and generates additional profit, but you're reaching a 125 point of dim inishing returns. Because you can never be sure that your m odel of sales response to 126 advertising will be valid next year (especially at greatly increased spending levels), it doesn't seem 127 prudent to allow unrestricted spending on advertising. 128 129 Suppose you want to m aintain your original advertising budget of $40,000. Add the constraint to the 130 problem that lim its the sum of advertising during the four quarters to $40,000. 131 On the Tools m enu, click Solver, and then click Add. The Add Constraint 132 dialog box appears. In the Cell reference box, type f11 or select cell F11 133 (advertising total) on the worksheet. Cell F11 m ust be less than or equal to $40,000. 134 The relationship in the Constraint box is <= (less than or equal to) by default, so 135 you don't have to change it. In the box next to the relationship, type 40000. Click 136 137 OK, and then click Solve. 138 After you exam ine the results, click Restore original values and then click OK 139 to discard the results and return the cells to their form er values. 140 141

19.1 Allocating Advertising Expenditures

219

A B C D E F G H I 142 The solution found by Solver allocates am ounts ranging from $5,117 in Q3 to $15,263 in Q4. Total 143 Profit has increased from $69,662 in the original budget to $71,447, without any increase in the 144 advertising budget. 145 146 Changing a Constraint 147 148 When you use Microsoft Excel Solver, you can experim ent with slightly different param eters to decide 149 the best solution to a problem . For exam ple, you can change a constraint to see whether the results 150 are better or worse than before. In the sam ple worksheet, try changing the constraint on advertising 151 dollars to $50,000 to see what that does to total profits. 152 On the Tools m enu, click Solver. The constraint, $ F$ 11<=40000, should 153 already be selected in the Subject to the constraints box. Click Change. In 154 the Constraint box, change 40000 to 50000. Click OK, and then click Solve. 155 Click Keep solver solution and then click OK to keep the results that are 156 displayed on the worksheet. 157 158 159 Solver finds an optim al solution that yields a total profit of $74,817. That's an im provem ent of $3,370 160 over the last figure of $71,447. In m ost firm s, it's not too difficult to justify an increm ental investm ent of 161 $10,000 that yields an additional $3,370 in profit, or a 33.7% return on investm ent. This solution also 162 results in profits of $4,889 less than the unconstrained result, but you spend $39,706 less to get there. 163 164 Saving a Problem Model 165 166 When you click Save on the File m enu, the last selections you m ade in the Solver Parameters 167 dialog box are attached to the worksheet and retained when you save the workbook. However, you 168 can define m ore than one problem for a worksheet by saving them individually using Save Model in 169 the Solver Options dialog box. Each problem m odel consists of cells and constraints that you 170 entered in the Solver Parameters dialog box. 171 172 When you click Save Model, the Save Model dialog box appears with a default selection, based 173 on the active cell, as the area for saving the m odel. The suggested range includes a cell for each 174 constraint plus three additional cells. Make sure that this cell range is an em pty range on the 175 worksheet. 176 On the Tools m enu, click Solver, and then click Options. Click Save Model. 177 In the Select model area box, type h15:h18 or select cells H15:H18 on the 178 worksheet. Click OK. 179 180 181 Note You can also enter a reference to a single cell in the Select model area box. Solver will use 182 this reference as the upper-left corner of the range into which it will copy the problem specifications. 183 184 185 To load these problem specifications later, click Load Model on the Solver Options dialog box, 186 type h15:h18 in the Model area box or select cells H15:H18 on the sam ple worksheet, and then 187 click OK. Solver displays a m essage asking if you want to reset the current Solver option settings with 188 the settings for the m odel you are loading. Click OK to proceed.

220

Chapter 19 Modeling Marketing Decisions

Figure 19.2 Quick Tour Influence Chart

SolvSamp.xls Quick Tour (Row number for each variable)

Profit Margin 16

Prod. Profit 15

Gross Margin 8

Cost of Sales 7

Sales Revenue 6

Corporate Overhead 12

Total Costs 13

Units Sold 5

Product Cost 19

Seasonality 3

Advertising 11

Product Price 18

Overhead Rate (15%)

Salesforce 10

Nonlinear Product Mix Optimization

20

20.1 DIMINISHING PROFIT MARGIN

Figure 20.1 Product Mix Problem
A B C D E F G H I 1 Example 1: Product mix problem with diminishing profit margin. 2 Your com pany m anufactures TVs, stereos and speakers, using a com m on parts inventory 3 of power supplies, speaker cones, etc. Parts are in lim ited supply and you m ust determ ine 4 the m ost profitable m ix of products to build. But your profit per unit built decreases with 5 volum e because extra price incentives are needed to load the distribution channel. 6 7 TV set Stereo Speaker 8 100 100 100 Number to Build-> 9 Inventory No. Used 10 Part Name 1 1 0 450 200 11 Chassis 1 0 0 Diminishing 250 100 12 Picture Tube 2 2 1 Returns 800 500 13 Speaker Cone 1 1 0 Exponent: 450 200 14 Power Supply 2 1 1 0.9 600 400 15 Electronics Profits: 16 $4,732 $3,155 $2,208 By Product 17 Total $ 10,095 18 19 20 This m odel provides data for several products using com m on parts, each with a different profit m argin 21 per unit. Parts are lim ited, so your problem is to determ ine the num ber of each product to build from the 22 inventory on hand in order to m axim ize profits. 23

222

Chapter 20 Nonlinear Product Mix Optimization

23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

Problem Specifications Target Cell Changing cells Constraints D18 D9:F9 C11:C15<=B11:B15 Goal is to m axim ize profit. Units of each product to build. Num ber of parts used m ust be less than or equal to the num ber of parts in inventory. D9:F9>=0 Num ber to build value m ust be greater than or equal to 0. The form ulas for profit per product in cells D17:F17 include the factor ^ H15 to show that profit per unit dim inishes with volum e. H15 contains 0.9, which m akes the problem nonlinear. If you change H15 to 1.0 to indicate that profit per unit rem ains constant with volum e, and then click Solve again, the optim al solution will change. This change also m akes the problem linear.

Integer-Valued Optimization Models

21

21.1 TRANSPORTATION PROBLEM

Figure 21.1 Transportation Problem
A B C D E F G H I 1 Example 2: Transportation Problem. 2 Minim ize the costs of shipping goods from production plants to warehouses near m etropolitan dem and 3 centers, while not exceeding the supply available from each plant and m eeting the dem and from each 4 m etropolitan area. 5 Number to ship from plant x to warehouse y (at intersection): 6 Total San Fran Denver Chicago Dallas New York 7 Plants: 5 1 1 1 1 1 8 S. Carolina 5 1 1 1 1 1 9 Tennessee 5 1 1 1 1 1 10 Arizona ----------11 3 3 3 3 3 12 Totals: 13 180 80 200 160 220 Demands by Whse --> 14 Supply Shipping costs from plant x to warehouse y (at intersection): 15 Plants: 310 10 8 6 5 4 16 S. Carolina 260 6 5 4 3 6 17 Tennessee 280 3 4 5 5 9 18 Arizona 19 $19 $17 $15 $13 $19 $ 83 20 Shipping: 21 22 The problem presented in this m odel involves the shipm ent of goods from three plants to five regional 23 warehouses. Goods can be shipped from any plant to any warehouse, but it obviously costs m ore to 24 ship goods over long distances than over short distances. The problem is to determ ine the am ounts 25 to ship from each plant to each warehouse at m inim um shipping cost in order to m eet the regional 26 dem and, while not exceeding the plant supplies. 27

224

Chapter 21 Integer-Valued Optimization Models

27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46

Problem Specifications Target cell Changing cells B20 C8:G10 Goal is to m inim ize total shipping cost. Am ount to ship from each plant to each warehouse. Constraints B8:B10<=B16:B18 Total shipped m ust be less than or equal to supply at plant. C12:G12>=C14:G14 Totals shipped to warehouses m ust be greater than or equal to dem and at warehouses. C8:G10>=0 Num ber to ship m ust be greater than or equal to 0. You can solve this problem faster by selecting the Assume linear model check box in the Solver Options dialog box before clicking Solve. A problem of this type has an optim um solution at which am ounts to ship are integers, if all of the supply and dem and constraints are integers.

21.2 MODIFIED TRANSPORTATION PROBLEM

Figure 21.2 Display
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 A B C D Modified Example 2: Transportation Problem. E F G H I

Minimize the costs of shipping goods from production plants to warehouses near metropolitan demand centers, while not exceeding the supply available from each plant and meeting the demand from each metropolitan area. Number to ship from plant to warehouse Warehouse Shipped Chicago Dallas New York from plant Plant San Fran Denver S. Carolina 1 1 1 1 1 5 Tennessee 1 1 1 1 1 5 Arizona 1 1 1 1 1 5 Shipped to warehouse 3 3 3 3 3 Warehouse demand 180 80 200 160 220 Shipping cost from plant to warehouse Warehouse Chicago Dallas San Fran Denver $10 $8 $6 $6 $5 $4 $3 $4 $5 $83

Plant supply 310 260 280

Plant S. Carolina Tennessee Arizona Total shipping cost

$5 $3 $5

New York $4 $6 $9

21.2 Modified Transportation Problem

225

Figure 21.3 Formulas

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 A B C Modified Example 2: Transportation Problem. D E F G H I

Minimize the costs of shipping goods from production plants to warehouses near metropolitan demand centers, while not exceeding the supply available from each plant and meeting the demand from each metropolitan area. Number to ship from plant to warehouse Warehouse Shipped Plant San Fran Denver Chicago Dallas New York from plant S. Carolina 1 1 1 1 1 =SUM(C10:G10) Tennessee 1 1 1 1 1 =SUM(C11:G11) Arizona 1 1 1 1 1 =SUM(C12:G12) Shipped to warehouse =SUM(C10:C12) =SUM(D10:D12) =SUM(E10:E12) =SUM(F10:F12) =SUM(G10:G12) Warehouse demand 180 80 200 160 220 Shipping cost from plant to warehouse Plant S. Carolina Tennessee Arizona Total shipping cost San Fran $10 $6 $3 Denver $8 $5 $4 Warehouse Chicago $6 $4 $5 Dallas $5 $3 $5 New York $4 $6 $9

Plant supply 310 260 280

=SUMPRODUCT(C10:G12,C19:G21)

226

Chapter 21 Integer-Valued Optimization Models

21.3 SCHEDULING PROBLEM

Figure 21.4 Personnel Scheduling
A B C D E F G H I J K L M 1 Example 3: Personnel scheduling for an Amusement Park. 2 For em ployees working five consecutive days with two days off, find the schedule that m eets dem and 3 from attendance levels while m inim izing payroll costs. 4 5 Days off Employees Sun M on Tue Wed Thu Fri Sat 6 Sch. Sunday, Monday 7 A 0 0 0 1 1 1 1 1 Monday, Tuesday 8 B 8 1 0 0 1 1 1 1 Tuesday, Wed. 9 C 0 1 1 0 0 1 1 1 Wed., Thursday 10 D 10 1 1 1 0 0 1 1 Thursday, Friday 11 E 0 1 1 1 1 0 0 1 Friday, Saturday 12 F 7 1 1 1 1 1 0 1 Saturday, Sunday 13 G 0 0 1 1 1 1 1 0 14 Schedule Totals: 15 25 25 17 17 15 15 18 25 16 Total Demand: 17 22 17 13 14 15 18 24 18 19 Pay/Employee/Day: $40 $ 1,000 20 Payroll/Week: 21 22 The goal for this m odel is to schedule em ployees so that you have sufficient staff at the lowest cost. In 23 this exam ple, all em ployees are paid at the sam e rate, so by m inim izing the num ber of em ployees working 24 each day, you also m inim ize costs. Each em ployee works five consecutive days, followed by two days 25 off. 26 27 Problem Specifications 28 D20 Goal is to m inim ize payroll cost. 29 Target cell 30 D7:D13 Em ployees on each schedule. 31 Changing cells 32 D7:D13>=0 Num ber of em ployees m ust be greater than or equal 33 Constraints to 0. 34 35 D7:D13=Integer Num ber of em ployees m ust be an integer. 36 37 F15:L15>=F17:L17 Em ployees working each day m ust be greater than or 38 equal to the dem and. 39 40 Rows 7-13 1 m eans em ployee on that schedule works that day. 41 Possible schedules 42 43 In this exam ple, you use an integer constraint so that your solutions do not result in fractional num bers of 44 em ployees on each schedule. Selecting the Assume linear model check box in the Solver Options 45 dialog box before you click Solve will greatly speed up the solution process.

21.3 Scheduling Problem

227

Figure 21.5 Personnel Scheduling with Corrections

A B C D E F G H I J K L M 1 Example 3: Personnel scheduling for an Amusement Park. (with corrections) 2 For em ployees working five consecutive days with two days off, find the schedule that m eets dem and 3 from attendance levels while m inim izing payroll costs. 4 5 Days off Employees Sun M on Tue Wed Thu Fri Sat 6 Sch. Sunday, Monday 0 0 0 1 1 1 1 1 7 A Monday, Tuesday 8 1 0 0 1 1 1 1 8 B Tuesday, Wed. 0 1 1 0 0 1 1 1 9 C Wed., Thursday 10 1 1 1 0 0 1 1 10 D Thursday, Friday 0 1 1 1 1 0 0 1 11 E Friday, Saturday 7 1 1 1 1 1 0 0 12 F Saturday, Sunday 0 0 1 1 1 1 1 0 13 G 14 Schedule Totals: 25 25 17 17 15 15 18 18 15 16 Total Demand: 22 17 13 14 15 18 24 17 18 Pay/Employee/Day: $40 19 $ 5,000 Payroll/Week: 20 21 22 The goal for this m odel is to schedule em ployees so that you have sufficient staff at the lowest cost. In 23 this exam ple, all em ployees are paid at the sam e rate, so by m inim izing the num ber of em ployees working 24 each day, you also m inim ize costs. Each em ployee works five consecutive days, followed by two days 25 off. 26 27 Problem Specifications 28 D20 Goal is to m inim ize payroll cost. 29 Target cell 30 D7:D13 Em ployees on each schedule. 31 Changing cells 32 D7:D13>=0 Num ber of em ployees m ust be greater than or equal 33 Constraints to 0. 34 35 D7:D13=Integer Num ber of em ployees m ust be an integer. 36 37 F15:L15>=F17:L17 Em ployees working each day m ust be greater than or 38 equal to the dem and. 39 40 Rows 7-13 1 m eans em ployee on that schedule works that day. 41 Possible schedules 42 43 In this exam ple, you use an integer constraint so that your solutions do not result in fractional num bers of 44 em ployees on each schedule. Selecting the Assume linear model check box in the Solver Options 45 dialog box before you click Solve will greatly speed up the solution process.

228

Chapter 21 Integer-Valued Optimization Models

This page is intentionally mostly blank.

Optimization Models for Finance Decisions

22
J

22.1 WORKING CAPITAL MANAGEMENT PROBLEM

Figure 22.1 Working Capital Management
A B C D E F G H I 1 Example 4: Working Capital Management. 2 Determ ine how to invest excess cash in 1-m onth, 3-m onth and 6-m onth CDs so as to 3 m axim ize interest incom e while m eeting com pany cash requirem ents (plus safety m argin). 4 Yield Term Purchase CDs in months: 5 Interest 1.0% 1 1, 2, 3, 4, 5 and 6 6 1-mo CDs: Earned: 4.0% 3 1 and 4 7 3-mo CDs: Total 9.0% 6 1 $ 7,700 8 6-mo CDs: 9 M onth 1 M onth 2 M onth 3 M onth 4 M onth 5 M onth 6 End 10 M onth: $400,000 $205,000 $216,000 $237,000 $158,400 $109,400 $125,400 11 Init Cash: 100,000 100,000 110,000 100,000 100,000 120,000 12 M atur CDs: 1,000 1,000 1,400 1,000 1,000 2,300 13 Interest: 100,000 100,000 100,000 100,000 100,000 100,000 14 1-mo CDs: 10,000 10,000 15 3-mo CDs: 10,000 16 6-mo CDs: 75,000 (10,000) (20,000) 80,000 50,000 (15,000) 60,000 17 Cash Uses: $205,000 $216,000 $237,000 $158,400 $109,400 $125,400 $187,700 18 End Cash: 19 -290000 20 21 22 If you're a financial officer or a m anager, one of your tasks is to m anage cash and short-term investm ents in a 23 way that m axim izes interest incom e, while keeping funds available to m eet expenditures. You m ust trade off 24 the higher interest rates available from longer-term investm ents against the flexibility provided by keeping funds 25 in short-term investm ents. 26 27 This m odel calculates ending cash based on initial cash (from the previous m onth), inflows from m aturing 28 certificates of deposit (CDs), outflows for new CDs, and cash needed for com pany operations for each m onth. 29 30 You have a total of nine decisions to m ake: the am ounts to invest in one-m onth CDs in m onths 1 through 6; 31 the am ounts to invest in three-m onth CDs in m onths 1 and 4; and the am ount to invest in six-m onth CDs in 32 m onth 1. 33

230

Chapter 22 Optimization Models for Finance Decisions

A 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

Problem Specifications Target cell Changing cells H8 B14:G14 B15, E15, B16 Constraints B14:G14>=0 B15:B16>=0 E15>=0 B18:H18>=100000 Ending cash m ust be greater than or equal to $100,000. The optim al solution determ ined by Solver earns a total interest incom e of $16,531 by investing as m uch as possible in six-m onth and three-m onth CDs, and then turns to one-m onth CDs. This solution satisfies all of the constraints. Suppose, however, that you want to guarantee that you have enough cash in m onth 5 for an equipm ent paym ent. Add a constraint that the average m aturity of the investm ents held in m onth 1 should not be m ore than four m onths. The form ula in cell B20 com putes a total of the am ounts invested in m onth 1 (B14, B15, and B16), weighted by the m aturities (1, 3, and 6 m onths), and then it subtracts from this am ount the total investm ent, weighted by 4. If this quantity is zero or less, the average m aturity will not exceed four m onths. To add this constraint, restore the original values and then click Solver on the Tools m enu. Click Add. Type b20 in the Cell Reference box, type 0 in the Constraint box, and then click OK. To solve the problem , click Solve. To satisfy the four-m onth m aturity constraint, Solver shifts funds from six-m onth CDs to three-m onth CDs. The shifted funds now m ature in m onth 4 and, according to the present plan, are reinvested in new three-m onth CDs. If you need the funds, however, you can keep the cash instead of reinvesting. The $56,896 turning over in m onth 4 is m ore than sufficient for the equipm ent paym ent in m onth 5. You've traded about $460 in interest incom e to gain this flexibility. Investm ent in each type of CD m ust be greater than or equal to 0. Goal is to m axim ize interest earned. Dollars invested in each type of CD.

22.2 Work Cap Alternate Formulations

231

22.2 WORK CAP ALTERNATE FORMULATIONS

Figure 22.2 Working Capital Management Horizontal Time

232

Chapter 22 Optimization Models for Finance Decisions

Figure 22.3 Working Capital Management Vertical Time

22.3 Stock Portfolio Problem

233

22.3 STOCK PORTFOLIO PROBLEM

Figure 22.4 Efficient Stock Portfolio
A B C D E F G H I J 1 Example 5: Efficient stock portfolio. 2 Find the weightings of stocks in an efficient portfolio that m axim izes the portfolio rate of 3 return for a given level of risk. This worksheet uses the Sharpe single-index m odel; you 4 can also use the Markowitz m ethod if you have covariance term s available. 5 6.0% 3.0% Market variance 6 Risk-free rate 15.0% 100.0% Maximum weight 7 Market rate 8 Beta ResVar Weight *Beta *Var. 9 0.80 0.04 20.0% 0.160 0.002 10 Stock A 1.00 0.20 20.0% 0.200 0.008 11 Stock B 1.80 0.12 20.0% 0.360 0.005 12 Stock C 2.20 0.40 20.0% 0.440 0.016 13 Stock D 0.00 0.00 20.0% 0.000 0.000 14 T-bills 15 100.0% 1.160 0.030 16 Total Variance Return 17 7.1% Portfolio Totals: 16.4% 18 19 Minimize Risk: D21:D29 20 Maximize Return: A21:A29 0.1644 0.07077 21 5 5 22 TRUE TRUE 23 TRUE TRUE 24 TRUE TRUE 25 TRUE TRUE 26 TRUE TRUE 27 TRUE TRUE 28 TRUE TRUE 29 30 31 One of the basic principles of investm ent m anagem ent is diversification. By holding a portfolio of several 32 stocks, for exam ple, you can earn a rate of return that represents the average of the returns from the 33 individual stocks, while reducing your risk that any one stock will perform poorly. 34 35 Using this m odel, you can use Solver to find the allocation of funds to stocks that m inim izes the portfolio 36 risk for a given rate of return, or that m axim izes the rate of return for a given level of risk. 37 38 This worksheet contains figures for beta (m arket-related risk) and residual variance for four stocks. In 39 addition, your portfolio includes investm ents in Treasury bills (T-bills), assum ed to have a risk-free rate of 40 return and a variance of zero. Initially equal am ounts (20 percent of the portfolio) are invested in each 41 security. 42 43 Use Solver to try different allocations of funds to stocks and T-bills to either m axim ize the portfolio rate of 44 return for a specified level of risk or m inim ize the risk for a given rate of return. With the initial allocation 45 of 20 percent across the board, the portfolio return is 16.4 percent and the variance is 7.1 percent. 46 K

234

Chapter 22 Optimization Models for Finance Decisions

A 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76

Problem Specifications Target cell Changing cells Constraints E18 E10:E14 E10:E14>=0 E16=1 G18<=0.071 Beta for each stock Variance for each stock B10:B13 C10:C13 Goal is to m axim ize portfolio return. Weight of each stock. Weights m ust be greater than or equal to 0. Weights m ust equal 1. Variance m ust be less than or equal to 0.071.

Cells D21:D29 contain the problem specifications to m inim ize risk for a required rate of return of 16.4 percent. To load these problem specifications into Solver, click Solver on the Tools m enu, click Options, click Load Model, select cells D21:D29 on the worksheet, and then click OK until the Solver Parameters dialog box is displayed. Click Solve. As you can see, Solver finds portfolio allocations in both cases that surpass the rule of 20 percent across the board. You can earn a higher rate of return (17.1 percent) for the sam e risk, or you can reduce your risk without giving up any return. These two allocations both represent efficient portfolios. Cells A21:A29 contain the original problem m odel. To reload this problem , click Solver on the Tools m enu, click Options, click Load Model, select cells A21:A29 on the worksheet, and then click OK. Solver displays a m essage asking if you want to reset the current Solver option settings with the settings for the m odel you are loading. Click OK to proceed.

22.4 MoneyCo Problem

235

22.4 MONEYCO PROBLEM

Figure 22.5 Display
A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 CD rate = Time 1 Time 2 Time 3 Time 4 Max to invest Amount Invested B C Return on investments 0.06 A B -1.00 -1.00 1.15 1.40 $500 $100 $500 $100 $500 $100 D E F G H I J K L M

C -1.00 1.28

E -1.00

CD1 -1.00 1.06

CD2 -1.00 1.06

CD3

-1.00 1.15 $500 $100

1.32

-1.00 1.06

$500 $1,000,000 $1,000,000 $1,000,000 $100 $100 $100 $100 Cash in $1,000 $0 $0 Feasible Cash out $600 $21 $34 $493 Final balance

Time 1 Time 2 Time 3 Time 4

Cash flows from investments -$100 -$100 -$100 $0 $115 $0 $0 $0 $128 $140 $0 $0

$0 $0 -$100 $115

$0 -$100 $0 $132

-$100 $106 $0 $0

$0 -$100 $106 $0

$0 $0 -$100 $106

Legend data cells changing cells computed cells input assumptions, uncontrollable, constraints decision variables, controllable intermediate and output variables, target Defined Names Amount_Invested = $B$11:$I$11 Cash_out = $K$14:$K$17 Final_balance = $K$17 Max_to_invest = $B$9:$I$9

Figure 22.6 Formulas

A 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 CD rate = Time 1 Time 2 Time 3 Time 4 Max to invest Amount Invested B C Return on investments 0.06 A B -1.00 -1.00 1.15 1.40 $500 $100 $500 $100 $500 $100 D E F G H I J K L

C -1.00 1.28

CD1 -1.00 -1.00 =1+$B$2 1.32

CD2 -1.00 =1+$B$2

CD3

-1.00 1.15 $500 $100

-1.00 =1+$B$2

$500 $1,000,000 $1,000,000 $1,000,000 $100 $100 $100 $100 =IF(AND(Amount_Invested< Cash in Cash out $1,000 =SUM(B14:J14) $0 =SUM(B15:J15) $0 =SUM(B16:J16) =SUM(B17:J17) Final balanc

Time 1 Time 2 Time 3 Time 4

Cash flows from investments =B4*B$11 =C4*C$11 =D4*D$11 =B5*B$11 =C5*C$11 =D5*D$11 =B6*B$11 =C6*C$11 =D6*D$11 =B7*B$11 =C7*C$11 =D7*D$11

=E4*E$11 =E5*E$11 =E6*E$11 =E7*E$11

=F4*F$11 =F5*F$11 =F6*F$11 =F7*F$11

=G4*G$11 =G5*G$11 =G6*G$11 =G7*G$11

=H4*H$11 =H5*H$11 =H6*H$11 =H7*H$11

=I4*I$11 =I5*I$11 =I6*I$11 =I7*I$11

Array-entered (Control+Shift+Enter) formula in K11: =IF(AND(Amount_Invested<=Max_to_invest,Cash_out>=0),"Feasible","Not Feasible") Enter =B4*B$11 in cell B14 and copy to cells B14:I17 Enter =SUM(B14:J14) in cell K14 and copy to K14:K17

236

Chapter 22 Optimization Models for Finance Decisions

Figure 22.7 Solver Dialog Box