MGS 3100 Business Analysis Chapter 1

Final Exam Review Introduction to Modeling

Julie Liggett De Jong Julie Liggett De Jong

mgs3100@mindspring.com mgs3100@mindspring.com
www.mindspring.com/~mgs3100 www.mindspring.com/~mgs3100

THE MODELING PROCESS Take a “Black Box” View of a Model

Applied to the first two stages of decision making

Analysis
Model Results
Interpretation

Decisions Performance
Abstraction

Symbolic (Controllable) Measure(s)

World Managerial
Model
Real Judgment Parameters Consequence
World
A

(Uncontrollable) Variables

Management Intuition
Decisions
Situation
Exogenous variables Endogenous variables

1
Take a “Black Box” View of a Model
TYPES OF MODELS
Physical Model Analog Model Symbolic Model

Tangibility Tangible Intangible Intangible

Comprehension Easy Harder Hardest

Duplicate & Share Difficult Easier Easiest

Pie Price
Unit Cost, Filling Modify & Difficult Easier Easiest
Unit Cost, Dough
Unit Pie Processing Cost
Model Profit Manipulate

Fixed Cost p of Use

Scope Lowest Wider Widest

Examples Model Airplane Road Map Simulation Model

Model House Speedometer Algebraic Model
Model City Pie Chart Spreadsheet
Exogenous variables Endogenous variables Model

Modeling
Techniques Modeling
Techniques Deductive
Modeling
Deterministic Probabilistic
Models Models

Time Series
Profit Model Decision Analysis Simulation
Forecasting

Breakeven Naive / Exp Smoothing Alternatives,States,Payoffs Random numbers

Pricing for Max Profit Regression (trend) Decision Criteria Distributions
Crossover / Indifference Classical Decomposition Decision Trees Discrete Variables
point (trend + seasonality) Bayes Theorem Continuous Variables

2
Deductive Focuses on
Modeling variables

Top down/ Assumes

data poor mathematical
relationships
& parameter
values

Focuses on Depends on
variables modeler’s
knowledge &
Assumes judgments of
mathematical mathematical
relationships relationships
& parameter & data values
values

3
Inferential Inferential
Modeling Modeling

Bottom up /
data rich

Focuses on Focuses on
variables in variables in
existing data existing data
collections
ll ti collections
ll ti

Analyzes data to Analyzes data to

determine determine
relationships & to relationships & to
estimate estimate
parameter values parameter values

4
 Model building is an iterative
Places premium
process
on accurate,
readily available DEDUCTIVE MODELING

d t & jjudgments
data d t Decision
D i i Modeling
M d li
(‘What If?’ Projections, Decision
Analysis, Decision Trees)
Data Poor
D i i Modeling
Decision M d li
(‘What If?’ Projections, Profit
Model, Optimization)

about future
applicability of Low Certainty
Model Building
High Certainty
Process
data PROBABILISTIC
MODELS
DETERMINISTIC
MODELS

Data Analysis Data Analysis

(Forecasting, Simulation, (Data Warehouses,
Parameter Evaluation
Statistical Analysis, Data Rich
Parameter Estimation)

INFERENTIAL MODELING

Chapter 2 Basic Profit Model

Spreadsheet
Modeling •Selling Price (SP) •Cost
•Quantity (Q) ƒ Overhead Cost
ƒ Sales Volume ƒ Sunk Cost
ƒ Production ƒ Fixed Cost (FC)
Volume ƒ Variable Cost (VC)
ƒ Total Cost (TC)

•Demand (D) •Contribution margin

•Revenue •Breakeven point
•Profit •Crossover point.

5
Basic Profit Model Breakeven Point
Profit = Revenue – Total Cost Set Profit = 0 and solve for Q, finds
Profit = SP*Q – (VC*Q + FC) breakeven quantity:
q y
P fit = SP*Q – VC*Q – FC
Profit 0 = CM*QBE – FC
Profit = (SP – VC)*Q - FC FC = CM*QBE
Contribution Margin (CM) = SP-VC, so FC / CM = QBE
Profit=CM*Q – FC, and Q=(FC+Profit)/CM
If a profit model has fixed costs of $150K, a If a profit model has fixed costs of
sales price of $400, and variable costs of $150,000, variable costs of $250, and a
$250, at what quantity will profit be $300k? sales price of $400, what is its breakeven
quantity?

Breakeven Point Crossover Point

Set Profit = 0 and solve for Q, finds
Cross-over Point a/k/a Indifference Point
breakeven quantity:
q y
ƒProfitA=CMA*QA – FCA SetProfitA=ProfitB
0 = CM*QBE – FC
ƒProfitB=CMB*QB - FCB Solve for Q
FC = CM*QBE
FC / CM = QBE
Total Cost of Process A = Total Cost of
Process B
ƒCMA*QA – FCA = CMB*QB – FCB
If a profit model has a breakeven point of
ƒCMA*QA – CMB*QB = FCA – FCB
400 units, a sales price of $300 and a
ƒQAtoB = (FCA – FCB) / (CMA - CMB)
variable cost of $75, what are its fixed
costs?

6
Class Exercise
Influence Diagram
Analyze the following plans
Plan A Plan B Plan C Profit

FC 150 000
150,000 450 000
450,000 2 850 000
2,850,000 Revenue Total Cost

VC 250 150 100 Processing Ingredient

Cost Cost
SP 400 400 400
Required
Ingredient
• How many units must be sold to breakeven Quantities
Pies Demanded
under each?
• Find the Indifference Points between A & B;
B & C; A & C Unit Pie Unit Cost Unit Cost
Pie Price Processing Cost Filling Dough Fixed Cost

Exercise: Capstone Computer

Capstone Computers assembles and sells

personal computers. Sales volume depends on
the sales price. At $1,000/each, 5,000
computers will be sold; every increase (or
decrease) of $100, sales will decrease (or
increase, respectively) by 1,000 units.
Rules of Good Model Design
• Compute the y-intercept.
y intercept and Layout
• Compute the slope.
• What is the linear equation that describes the
relationship between demand and price?

7
Present clearly labeled input Clearly label model results.
variables together.

Include units of measure where Store input variables in separate

appropriate. cells & refer to addresses in
formulas.

8
Format the spreadsheet to simplify Separate physical results from
interpretation. financial or economic results.

CHAPTER 9: Simulation
Introduction

Simulation allows you to quickly and inexpensively

acquire knowledge concerning a problem that is usually
gained through experience (which is often costly and
time consuming).
An experimental device (simulator) that “acts like”
(simulates) the system of interest.

Goal: To create an environment where

information about alternative actions can
be obtained in a quick, cost-effective
manner through experimentation.

9
When should simulation be used? Terminology to Know

Simulation is frequently used because:

• Simulation Models
1. Analytical models can be difficult or impossible to • Probability Distributions
obtain due to complicating factors.
• Cumulative Probabilities
2. Analytical models typically predict only average or
“steady-state” (long-run) behavior.
• Random Variables/Numbers (RN)
− discrete vs. continuous
3. Simulation models can be developed on a PC or
• Expected Value
workstation, with a minimum of computing and
mathematical skill.

Calculating Expected Values Calculating Expected Values

To calculate expected profit (or mean profit), What is the expected profit for the
multiply profits by respective probabilities for
each item and then sum to get the expected
g table?
following
profit.

10
 Calculating Expected Values Simulation
Question 4 of Sample Exam 1
1. Construct a Cumulative Probability
Distribution
2. Generate Random Numbers
3. Project a Random Observation of
the Variable

1. Construct a Cumulative Probability 2. Generate Random Numbers

Distribution

Week Random
Cases Prob. Cumulative
Number
3 .156 .156
1 .587
4 .287 .443
2 .266
5 .362 .805
3 .702
702
6 .195 1.000
4 .307

11
 Simulation & Random Numbers
3. Project Random Observation of
Variable Question 5 of Sample Exam1
Cases Prob. Cum Range

3 .156 .156 .000-.155

.000 .155
4 .287 .443 .156-.442
5 .362 .805 .443-.804
6 .195 1.000 .805-1.000

Week Random Cases

Number
1 .587 5
2 .266 4
3 .702 5
4 .307 4

Simulation & Random Numbers

Question 6 of Sample Exam1

STOP
HERE

12
 Quantitative Forecasting Models
Chapter 13 1.Causal
1. Causal Forecasting Models
Forecasting 2.Time
2. Time--Series Forecasting Models:
a) Curve Fitting

b) Moving Averages (Naive)

i. Simple n-Period Moving Average
ii. Weighted n-Period Moving Average

c)) Exponential
p Smoothingg
i. Basic model
ii. Holt’s Model (exponential smoothing
with trend)

d) Seasonality

Causal Forecasting Models Requirements

13
IIndependent
d d t andd dependent
d d t We mustt know
W k th
the values
l off the
th
variables must share a relationship independent variables when we
make the forecast

Important Variables: Time--Series Forecasting Models

Time

1. Curve Fitting
X : Independent variable
2. Moving Averages
a Naive
a. Nai e
Y : Value of actual dependent variable b. Simple n-Period Moving Average
c. Weighted n-Period Moving Average
Y : Average of dependent variable values (Y bar).
3. Exponential Smoothing
a. Basic model
b. Holt’s Model (exponential smoothing with
Yˆ : Forecast of dependent variable (Y hat). trend)

4. Seasonality

14
 ACTUAL THREE-MONTH FOUR-MONTH SIMPLE ACTUAL THREE-MONTH FOUR-MONTH SIMPLE
SALES SIMPLE MOVING MOVING AVERAGE SALES SIMPLE MOVING MOVING AVERAGE
MONTH ($000s) AVERAGE FORECAST FORECAST MONTH ($000s) AVERAGE FORECAST FORECAST
Jan. 20 Jan. 20
Feb. 24 Feb. 24
Mar. 27 Mar. 27
Apr.
p 31 ((20 + 24 + 27)/3
) = 23.67 Apr.
p 31 ((20 + 24 + 27)/3
) = 23.67
May 37 (24 + 27 + 31)/3 = 27.33 (20 + 24 + 27 + 31)/4 = 25.50 May 37 (24 + 27 + 31)/3 = 27.33 (20 + 24 + 27 + 31)/4 = 25.50
June 47 (27 + 31 + 37)/3 = 31.67 (24 + 27 + 31 + 37)/4 = 29.75 June 47 (27 + 31 + 37)/3 = 31.67 (24 + 27 + 31 + 37)/4 = 29.75
July 53 (31 + 37 + 47)/3 = 38.33 (27 + 31 + 37 + 47)/4 = 35.50 July 53 (31 + 37 + 47)/3 = 38.33 (27 + 31 + 37 + 47)/4 = 35.50
Aug. 62 (37 + 47 + 53)/3 = 45.67 (31 + 37 + 47 + 53)/4 = 42.00 Aug. 62 (37 + 47 + 53)/3 = 45.67 (31 + 37 + 47 + 53)/4 = 42.00
Sep. 54 (47 + 53 + 62)/3 = 54.00 (37 + 47 + 53 + 62)/4 = 49.75 Sep. 54 (47 + 53 + 62)/3 = 54.00 (37 + 47 + 53 + 62)/4 = 49.75
Oct. 36 (53 + 62 + 54)/3 = 56.33 (47 + 53 + 62 + 54)/4 = 54.00 Oct. 36 (53 + 62 + 54)/3 = 56.33 (47 + 53 + 62 + 54)/4 = 54.00
Nov. 32 (62 + 54 + 36)/3 = 50.67 (53 + 62 + 54 + 36)/4 = 51.25 Nov. 32 (62 + 54 + 36)/3 = 50.67 (53 + 62 + 54 + 36)/4 = 51.25
Dec. 29 (54 + 36 + 32)/3 = 40.67 (62 + 54 + 36 + 32)/4 = 46.00 Dec. 29 (54 + 36 + 32)/3 = 40.67 (62 + 54 + 36 + 32)/4 = 46.00

Three- and Four- Month Simple Moving Averages y + y + y + y

yˆ = 15 14 13 12
16 4

Question 6 of Sample Exam2:

Using a Naïve forecast, what is the forecast for Q1, 2000?

A B C D E F
Enroll-
1 Year Quarter ment Forecast Error Abs Error
2 1997 1 313 313
3
4
5
2
3
4
285
312
339
313
292
307
?
20
32
20
32
Recent data is more important than
6
7
1998 1
2
359
320
331
?
28

28

old data
8 3 356 328 28 28
9 4 385 349 36 ?

yˆ = α 0 y6 + α1 y5 + α 2 y4
10 1999 1 396 376 20 20
11 2 367 ? 
12 3 397 373 24 24
13 4 423 391 32 32
14 2000 1 415
15 Bias = ?
16 Alpha = 0.75 MAD = 

15
 alpha2 = 0.167 Month Actual Sales (000) 3month WMA Fcst Absolute Error
alpha1 = 0.333 January 20
alpha0 = 0.500 February 24
SUM OF WTS= 1.00 March 27
April 31 24.83 6.17
May 37 28.50 8.50
June 47 33.33 13.67
July 53 41.00 12.00
August 62 48.33 13.67
September 54 56.50 2.50
October 36 56.50 20.50

Constraints: November
December
32
29
46.34
37.01
14.34
8.01

Sum = 99.35
MAD = 11.04

Weights are positive numbers

Assign smaller weights to older data Use Solver to find the optimal
Weights sum to 1 weights

Questions 3-5 of Sample Exam2:

Forecast for 1998 Q2 (cell D7)?
Error for 1997 Q2 (cell E3)?
Absolute error for 1998 Q4 (cell F9)?

A B C D E F
Enroll-
1 Year Quarter ment Forecast Error Abs Error
2 1997 1 313 313

Exponential Smoothing 3
4
5
2
3
4
285
312
339
313
292
307
?
20
32
20
32
6 1998 1 359 331 28 28
Forecast for t + 1 Observed in t Forecast for t 7 2 320 ?  
8 3 356 328 28 28
9 4 385 349 36 ?

ŷ t +1 = αy t + (1 − α )ŷ t 10
11
1999 1
2
396
367
376
?
20

20

12 3 397 373 24 24

Where α is a user-specified constant: 0 ≤ α ≤ 1

13
14 2000
4
1
423 391
415
32 32

15 Bias = ?
16 Alpha = 0.75 MAD = 

16
 Measures of Comparison Errors, Absolute Errors, & Errors Squared

∑ actual sales − forecast sales 1

2
A
Alpha
B
0.9
C D E F G

MAD =
all forecasts Exponential
Smoothing
number of forecasts 3 Time Demand Forecast Error Error2 Abs Error MAPE
4 1 10 8.00
5 2 14 9.80 4.20 17.64 4.20 30.00
6 3 19 13.58 5.42 29.38 5.42 28.53
actual sales − forecast sales 7 4 26 18.46 7.54 56.88 7.54 29.01

∑ actual sales
∗ 100 % 8
9
5
6
31
35
25.25
30.42
5.75
4.58
33.11
20.93
5.75
4.58
18.56
13.07
MAPE =
all forecasts 10 7 39 34.54 4.46 19.87 4.46 11.43
11 8 44 38.55 5.45 29.66 5.45 12.38
number of forecasts 12 9 51 43.46 7.54 56.92 7.54 14.79
13 10 55 50.25 4.75 22.60 4.75 8.64
14 11 61 54.52 6.48 41.93 6.48 10.62
n 15 12 54 60.35 -6.35 40.35 6.35 11.76

∑ ( actual sales − forecast sales ) 2

16
17
18
13 54.64

SUM = 49.82 369.28 62.52 188.79

MSE = t =1 19
20
MAD
MAPE
=
=
5.68
17.16
number of forecasts 21 MSE = 33.57

1. Look at original data to see seasonal 2. Deseasonalize the Data

pattern. Examine the data &
hypothesize an m-period seasonal
pattern.

17
3. Forecast using deseasonalized data 4. Seasonalize the forecast to account
for the seasonal pattern

Coal Receipts Over a Nine-Year Period

3,000

2,500

2,000

0 Tons)
Coal (000
1,500

1,000

500

1-1

1-3

2-1

2-3

3-1

3-3

4-1

4-3

5-1

5-3

6-1

6-3

7-1

7-3

8-1

8-3

9-1

9-3
Tim e (Year and Quarter)

1. Look at original data to see seasonal

pattern. Examine the data &
hypothesize an m-period seasonal
Gillett Coal Mine pattern.

18
 Deseasonalized Data 2.Deseasonalize the Data
3,000.0

2,500.0 a) Calculate a series of m-period moving

000 Tons)

2,000.0
averages, where m is the length of the
1,500.0
seasonal pattern.
pattern
Coal (0

1,000.0

500.0 b) Center the moving average in the middle of

-
1- 1- 1- 1- 2- 2- 2- 2- 3- 3- 3- 3- 4- 4- 4- 4- 5- 5- 5- 5- 6- 6- 6- 6- 7- 7- 7- 7- 8- 8- 8- 8- 9- 9- 9- 9-
the data from which it was calculated.
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
Tim e (Year & Qtr)
c) Divide the actual data at a given point in the
series by the centered moving average
2. Deseasonalized the Data corresponding to the same point.
d) Develop seasonal index
e) Divide actual data by the seasonal index

Time Coal 4 Period Time Coal 4 Period Centered

Year-Qtr Receipts Moving Average Year-Qtr Receipts Moving Average Moving Average
1-1 2,159 ----- 1-1 2,159
1-2 1,203 ----- 1-2 1,203
1-3 1,094 1,613 1,603
1-3 1,094 1,613 1-4 1,996 1,594 1,610
1-4 1,996 1,594 2-1 2,081 1,626 1,674
2-1 2,081 1,626 2-2 1,332 1,721 1,788 (1613 + 1594)/2 =
2-2 1,
,332 1,
,721 2-3 1,476 1,856 1,877

2-3 1,476 1,856

2-4 2,533 1,898 1,923 1603
2-4
(2,159+1,203+1,094+1,996)/4
2,533 1,898
= 1,613 3-1
3-2
2,249
1,533
1,948
2,063
2,005
2,061
3-1 2,249 1,948 3-3 1,935 2,060 2,055
3-4 2,523 2,050 2,058
3-2 1,533 2,063
4-1 2,208 2,066 2,064
3-3 1,935 2,060 4-2 1,597 2,061 2,087
3-4 2,523 2,050 4-3 1,917 2,112 2,163
4-1 2,208 2,066 4-4 2,726 2,213 2,255

a) Calculate a series of m-period moving b) Center the moving average in the

averages, where m is the length of the middle of the data from which it was
seasonal pattern. calculated.

19
 Time Coal 4 Period Centered Ratio of Coal Receipts to Seasonal
Year-Qtr Receipts Moving Average Moving Average Centered Moving Average Indices
Time Coal 4 Period Centered Ratio of Coal Receipts to
1-1 2,159 1.112
Year-Qtr Receipts Moving Average Moving Average Centered Moving Average 1-2 1,203 0.786
1-1 2,159 1-3 1,094 1,613 1,603 0.682 0.863
1-2 1,203 1-4 1,996 1,594 1,610 1.240 1.238
1-3 1,094 1,613 1,603 0.682 2-1 2,081 1,626 1,674 1.244 1.112
1-4 1,996 1,594 1,610 1.240 2-2 1,332 1,721 1,788 0.745 0.786
2-1 2,081 1,626 1,674 1.244 2-3 1,476 1,856 1,877 0.787 0.863
2-2 1,332 1,721 1,788 0.745 2-4 2,533 1,898 1,923 1.317 1.238
2-3 1,476 1,856 1,877 1,094 / 1,603 = 0.682
0.787 3-1 2,249 1,948 2,005 1.122 1.112
3-2 1,533 2,063 2,061 0.744 0.786
24
2- 2 533
2, 1 898
1, 1 923
1, 1 317
1.
3-3 1,935 2,060 2,055 0.942 0.863
3-1 2,249 1,948 2,005 1.122 3-4 2,523 2,050 2,058 1.226 1.238
3-2 1,533 2,063 2,061 0.744
3-3 1,935 2,060 2,055 0.942
3-4 2,523 2,050 2,058 1.226
d) Develop seasonal index for each
quarter
• Group p ratios by
yqquarter
c) Divide the actual data at a given point • Average all of the ratios to moving
in the series by the centered moving averages quarter by quarter
average corresponding to the same • Add Seasonal Indices data to table
point. • Normalize the seasonal index

Time Coal 4 Period Centered Ratio of Coal Receipts to Seasonal Deseasonalized

Year-Qtr Receipts Moving Average Moving Average Centered Moving Average Indices Data Forecast
1-1 2,159 1.108 1,948.1 1,948.1
1-2 1,203 0.784 1,535.4 1,948.1
1-3 1,094 1,613 1,603 0.682 0.860 1,272.3 1,678.5
1-4 1,996 1,594 1,610 1.240 1.234 1,617.8 1,413.1
2-1 2,081 1,626 1,674 1.244 1.108 1,877.8 1,546.8
Time Coal 4 Period Centered Ratio of Coal Receipts to Seasonal Deseasonalized 2-2 1,332 1,721 1,788 0.745 0.784 1,700.0 1,763.0
Year-Qtr Receipts Moving Average Moving Average Centered Moving Average Indices Data 2-3 1,476 1,856 1,877 0.787 0.860 1,716.6 1,721.9
1-1 2,159 1.112 1,941.0 2-4 2,533 1,898 1,923 1.317 1.234 2,053.1 1,718.4
1-2 1,203 0.786 1,529.8
3-1 2,249 1,948 2,005 1.122 1.108 2,029.3 1,937.1
1-3 1,094 1,613 1,603 0.682 0.863 1,267.7
1-4 1,996 1,594 1,610 1.240 1.238 1,611.9
32
3- 1 533
1, 2 063
2, 2 061
2, 0 744
0. 0 784
0. 1 956.
1,956 5 1 997.
1,997 4
2-1 2,081 1,626 1,674 1.244 1.112 1,870.9 3-3 1,935 2,060 2,055 0.942 0.860 2,250.4 1,970.7
2-2 1,332 1,721 1,788 0.745 0.786 1,693.8 3-4 2,523 2,050 2,058 1.226 1.234 2,045.0 2,153.4
2-3 1,476 1,856 1,877 0.787 0.863 1,710.3
2-4 2,533 1,898 1,923 1.317 1.238 2,045.6
3-1 2,249 1,948 2,005 1.122 1.112 2,021.9
3-2
3-3
3-4
1,533
1,935
2,523
2,063
2,060
2,050
2,061
2,055
2,058
0.744
0.942
1.226
0.786
0.863
1.238
1,949.4
2,242.2
2,037.5
3. Forecast method in deseasonalized
terms
• Review the g graphed
p deseasonalized data to
e) Divide actual data by the seasonal reveal pattern
index • Use forecasting method that accounts for
the pattern in the deseasonalized data
• Use Excel’s Solver to minimize the MSE
See Q1 & Q2 on Sample Exam2.

20
 Time Coal 4 Period Centered Ratio of Coal Receipts to Seasonal Deseasonalized Seasonalize
Year-Qtr Receipts Moving Average Moving Average Centered Moving Average Indices Data Forecast Forecast
1-1 2,159 ----- ----- ----- 1.108 1,948.1 1,948.1 2,159.000
1-2 1,203 ----- ----- ----- 0.784 1,535.4 1,948.1 1,526.409
1-3 1,094 1,613 1,603 0.682 0.860 1,272.3 1,678.5 1,443.212
1-4 1,996 1,594 1,610 1.240 1.234 1,617.8 1,413.1 1,743.439
2-1 2,081 1,626 1,674 1.244 1.108 1,877.8 1,546.8 1,714.276
2-2 1,332 1,721 1,788 0.745 0.784 1,700.0 1,763.0 1,381.390
2-3 1,476 1,856 1,877 0.787 0.860 1,716.6 1,721.9 1,480.540
2-4 2,533 1,898 1,923 1.317 1.234 2,053.1 1,718.4 2,120.128
3-1 2,249 1,948 2,005 1.122 1.108 2,029.3 1,937.1 2,146.723
3-2 1,533 2,063 2,061 0.744 0.784 1,956.5 1,997.4 1,564.974
3-3 1,935 2,060 2,055 0.942 0.860 2,250.4 1,970.7 1,694.495
34
3- 2 523
2, 2 050
2, 2 058
2, 1 226
1. 1 234
1. 2 045
2,045.0 2 153
2,153.4 2 656
2,656.854

Qualitative
4. Reseasonalize the forecast to account for Forecasting
the seasonal pattern
Models
• Multiply the deseasonalized forecast by the
seasonal index for the appropriate period.
• Graph the actual Coal Receipts and
Seasonalized Forecast

Expert Consensus
Judgment Panel

21
 Coordinator requests forecasts

Delphi Coordinator receives

Individual forecasts
Delphi
Method Method
Coordinator determines
(a) Median response
(b) Range of middle
50% of answers

Coordinator sends to all experts Coordinator requests

(a) Median response explanations from any
(b) Range of middle 50% expert whose estimate
(c) Explanations is not in the middle
50%

Grassroots Forecasting Market Research

22
 Chapter 8 Terminology
Decision
Analysis States of nature

Payoff / payoff
table

Probability

The payoff table is a fundamental

component in decision analysis Three Classes of
models Decision Models

Decisions under:
State of Nature
Decision 1 2 … m
d1 r11 r12 … r1m
certainty
d2 r21 r22 … r2m
2
uncertainty
… … … … …
dn rn1 rn2 … rnm
risk

23
Decisions under certainty: Decisions under certainty:

If I know for sure that it will be raining when I If I know for sure that it will be raining when I
leave work this afternoon, should I take my leave work this afternoon, should I take my
umbrella
mbrella to work
ork toda
today? ? umbrella
mbrella to work
ork toda
today? ?

Rain

Take Umbrella 0
Do Not -7.00

Decisions under risk:

We calculate Expected Returns
Multiple states of nature
Probabilities used to capture likelihood
Historical frequencies vs subjective estimates

24
The expected return (ERi) associated The Newsvendor Model
with decision i is
A B C D E F
1 Selling Price 75
m 2 Purchase Cost 40

ERi = Σrijj=1pj = ri1p1 + ri2p2 + … + rimpm 3

4
Goodwill Cost 50

5 States of Nature
6 Decision 0 1 2 3 Expected Return
7 0 0 -50 -100 -150 -85
The decision is based on the 8
9
1
2
-40
-80
35
-5
-15
70
-65
20
-12.5
22.5
10 -120 -45 30 105 7.5
maximum expected return. In other 11
3

12 Probabilities 0.1 0.3 0.4 0.2

words, i* is the optimal decision
where:
Calculate the Expected Return
ERi* = maximum overall i of ERi

Decisions under Decisions under

uncertainty uncertainty

Multiple states of
nature

25
Decisions under Decisions under
uncertainty uncertainty

Multiple states of Laplace

nature Maximin
Maximax
Don’t know what
Minimax regret
state of nature will
occur

Maximin Maximin

extremelyy
conservative or
pessimistic
approach to
making decisions

26
Maximin Maximin

Evaluate Select decision

minimum possible yielding
return associated maximum value
max
with each
of min
minimum
decision.
returns.
returns

A B C D E F
Maximin 1
2
Selling Price
Purchase Cost
75
40
3 Goodwill Cost 50
4

Review 5
6 Decision 0
States of Nature
1 2 3 Expected Return
7 0 -50 -100 -150 -85
homework
0
8 1 -40 35 -15 -65 -12.5
9 2 -80 -5 70 20 22.5
problem 10
11
3 -120 -45 30 105 7.5

12 Probabilities 0.1 0.3 0.4 0.2

See Q3 of expected return maximum possible

sample exam3 EVPI = with perfect -
information
expected return
without sample
information

27
 DECISIONS UNDER UNCERTAINTY
A B C D E F
1 Selling Price 75
2
3
Purchase Cost
Goodwill Cost
40
50
Prior probabilities: Probabilities that are initial
4 estimates, such as P(S) and P(W).
5 States of Nature
6 Decision 0 1 2 3 Expected Return Sonorola has estimated the prior probabilities
7 0 -50 -100 -150 -85
as P(S) = 0.45 and P(W) = 0.55.
0
8 1 -40 35 -15 -65 -12.5
9 2 -80 -5 70 20 22.5
10
11
3 -120 -45 30 105 7.5
Joint probabilities
12 Probabilities 0.1 0.3 0.4 0.2
Marginal probabilities
C l l t the
Calculate th EVPI
Posterior probabilities: Conditional probabilities,
such as P(S|E).
Q6 of sample exam3 Bayes’ Theorem is used to determine the
posterior probabilities.

Calculating Posterior Probabilities:

1. Enter given Reliabilities (conditional P(E|W)

probabilities). P(D|W)
2. Calculate Joint Probabilities byy multiplying
py g
Reliabilities by Prior Probabilities.
3. Compute Marginal Probabilities by summing P(S) P(W)

the entries in each row.

4. Generate Posterior Probabilities by dividing P(E&S)

each row entry of joint probability table by its

row sum.

P(W|E)
P(W|D)

28
 Question 2 of Sample Exam3 Decision Trees

Reliabilities Graphical device for analyzing decisions under

Good Credit Risk Bad Credit Risk risk
Steady Job 0.90 0.25 P(S|B)
P(S|G)
No Steady Job 0.10 0.75
P(N|B)
P(N|G)
0.25 0.75
Useful when there is are sequences of decisions.
Prior Probabilities
P(G) P(B)
Joint & Marginal Probabilities
0.225 0.1875 0.4125
Bayes’ Theorem is used to incorporate new
Steady Job
No Steady Job 0.025 0.5625 0.5875 information (posterior probabilities) into the
process.
process
Posterior Probabilities
Steady Job 0.55 0.45
No Steady Job 0.04 0.96

CREATING A DECISION TREE To use the decision tree to find the

optimal decision, we must append:
A square node represents a point at which • Probabilities for each branch emanating from
a decision must be made. each
h circular
i l node.
d
Each line (branch) leading from the
square represents a possible decision. • Terminal values (the return associated with
each terminal position).
A circular node represents an event (a
situation when the outcome is not
certain).
Each line (branch) leading from the circle
represents a possible outcome.

29
FOLDING BACK

To solve a decision tree, one works backward

(i.e., from right to left) by folding back the tree.
Question 10 of sample exam3

1. Fold back the terminal branches by calculating

an expected value for each terminal node.

Expected terminal value = 30(0.45) + (-8)(0.55) = 9.10

Incorporating New Information

Marketing has the following “track record” in
Conditional Probability: For two events A and B, predicting the market:
the conditional probability [P(A|B)], is the
probability of event A occurs given that event B
will occur. P(E|S) = 0.6
P(D|S) = 1 - P(E|S) = 0.4
For example, P(E|S) is the conditional probability
that marketing gives an encouraging report given
that the market is in fact going to be strong.
strong
P(D|W) = 0.7
P(E|W) = 1 - P(D|W) = 0.3

30
INCORORATING POSTERIOR THE EXPECTED VALUE OF SAMPLE INFORMATION
PROBABILITIES IN THE DECISION TREE
maximum possible maximum possible
30
S EVSI = expected return - expected return
IV W
P(W|E)
-8 with sample without sample
20 i f
information
ti i f
information
ti
A S
B V P(W|E)
II 7
W
C 5
S
E
VI
W 15
Q8 & Q9 of sample exam3
I 30
S
D P(W|D)
VII W -8
A 20
S
III B VIII P(W|D)
W 7
C 5
S
IX
W
15

THE EXPECTED VALUE OF PERFECT INFORMATION

expected return maximum possible

Chapter 15
EVPI = with perfect - expected return Statistical
information without sample
i f
information
ti Process
Control
EVPI = 30(0.45) + 15(0.55) – 12.85 ≈ 8.90

Q6 of sample exam3
The EVPI is an upper bound of how much one
would be willing to pay for sample information.

31
 Take periodic Variation
samples from
process

Plot sample points

on control chart
1 Common Causes
1.
Determine if 9 Variation inherent in a process
process is within 9 Eliminated through system
improvements
limits

Attribute
Variation measures

Product
characteristic
evaluated with
a discrete
choice:
2 Special Causes
2.
9 Variation due to identifiable factors
Good / bad
9 Modified through operator or
Yes / No
management action Pass / Fail

32
 Attribute Attribute
measures measures

Product Product
characteristic characteristic
evaluated with evaluated with
a discrete a discrete
choice: choice:
Good / bad
Yes / No Good / bad
Pass / Fail Yes / No
Pass / Fail

Variable measures Variable measures

Measurable product Measurable product

characteristic: characteristic:

Length, size
Length size, weight,
weight Length, size
Length size, weight,
weight
height, time, velocity height, time, velocity

33
Variable measures

Measurable product
characteristic

Length, size
Length size, weight,
weight Control Charts
height, time, velocity

Process Control Chart

Upper Out of control

control
li it
limit

Process
average

Graphs that establish process control Lower

control
limits limit

1 2 3 4 5 6 7 8 9 10
Sample number
Figure 15.1

34
 To develop Control Charts: Control Charts
Control
Measures Description
Charts
9Use in-control data p Chart Attributes Calculates percent defectives
in sample
p

9If non-random causes are present, find r Chart

(range
Variables Reflects the amount of
dispersion in a sample
them and discard data chart)
x bar Chart Variables Indicates how sample results
(mean chart) relate to the process average
9Correct control chart limits Cp Process Measures the capability
p y of a
(Process Capability process to meet design
Capability specifications
Ratio)
Cpk Process Indicates if the process mean
(Process Capability has shifted away from design
Capability target
Index)

p-Chart Example ~
p-Chart
Western Jeans Company
UCL = p + zσp 20 samples of 100 pairs of jeans
LCL = p - zσp
NUMBER OF PROPORTION
where SAMPLE DEFECTIVES DEFECTIVE

z = the number of standard deviations from the process 1 6 .06

average 2 0 .00
p = the sample proportion defective; an estimate of the 3 4 .04
process average
: : :
σp = the standard deviation of the sample proportion,
computed as: : : :
20 18 .18
p(1 - p)
σp = 200
n
Example 15.1

35
p-Chart Example ~ p-Chart Example ~
Western Jeans Company Western Jeans Company
20 samples of 100 pairs of jeans 20 samples of 100 pairs of jeans

NUMBER OF PROPORTION NUMBER OF PROPORTION

SAMPLE DEFECTIVES DEFECTIVE SAMPLE DEFECTIVES DEFECTIVE p = 0.10
1 6 .06 1 6 .06
2 0 .00 2 0 p(1.00
- p) 0.10(1 - 0.10)
total defectives UCL4= p + z = 0.10 + 3
3 4 p = .04 3 .04
n 100
total sample observations
: : : : :
UCL = 0.190 :
: : = 200: / 20(100) : : :
p(1 - p) 0.10(1 - 0.10)
20 18 = 0.10
.18 20 LCL
18= p - z .18 = 0.10 - 3
n 100
200 200
LCL = 0.010
Example 15.1 Example 15.1

p-Chart Example ~
Western Jeans Company Range ( R ) Chart
0.20

0.18 UCL = 0.190

0 16
0.16
UCL = D4R LCL = D3R
0.14
Proportion defective

0.12
∑R
0.10
p = 0.10
R= k
0.08
where:
P

0.06

0.04
R = range of each sample
0.02 LCL = 0.010
k = number of samples
2 4 6 8 10 12 14 16 18 20
Sample number

36
Factors for R-Chart: D3 & D4
SAMPLE SIZE FACTOR FOR x-CHART FACTORS FOR R-CHART
n A2 D3 D4

2 1.88 0.00 3.27

3 1.02 0.00 2.57
4 0 73
0.73 0 00
0.00 2 28
2.28
5 0.58 0.00 2.11
6 0.48 0.00 2.00
7 0.42 0.08 1.92
8 0.37 0.14 1.86
9 0.44 0.18 1.82
10 0.11 0.22 1.78
11 0.99 0.26 1.74
12 0.77 0.28 1.72
13 0.55 0.31 1.69
14 0.44 0.33 1.67
15 0.22 0.35 1.65
16 0.11 0.36 1.64
17 0.00 0.38 1.62
18 0.99 0.39 1.61
19 0.99 0.40 1.61
20 0.88 0.41 1.59 R-Chart Example ~
Table 15.1 Goliath Tool Company

R-Chart Example ~ R-Chart Example ~

Goliath Tool Company Goliath Tool Company

OBSERVATIONS (SLIP-RING DIAMETER, CM) ∑R 1.15 UCL = D4R = 2.11(0.115) = 0.243

R= = = 0.115
SAMPLE k 1 2 3 4 5 x R k 10 LCL = D3R = 0(0.115) = 0
1 5.02 5.01 4.94 4.99 4.96 4.98 0.08
2 5.01 5.03 5.07 4.95 4.96 5.00 0.12 0.28 –
3 4.99 5.00 4.93 4.92 4.99 4.97 0.08 0.24 –
4 UCL = 0.243
R = max5.03
– 4.91
min = 5.01
5.024.98 4.89
– 4.94 = 4.96
0.08 0.14
0.20 –
5 4.95 4.92 5.03 5.05 5.01 4.99 0.13
6 4.97 5.06 5.06 4.96 5.03 5.01 0.10

Range
0.16 – R = 0.115
7 5.05 5.01 5.10 4.96 4.99 5.02 0.14
8 5.09 5.10 5.00 4.99 5.08 5.05 0.11 0.12 –
9 5.14 5.10 4.99 5.08 5.09 5.08 0.15
0.08 –
10 5.01 4.98 5.08 5.07 4.99 5.03 0.10
50.09 1.15 0.04 – LCL = 0
0– | | | | | | | | | |
Example 15.3 1 2 3 4 5 6 7 8 9 10
Example 15.3 Sample number

37
x-Chart Calculations x-Chart Example
= =
UCL = x + A2R LCL = x - A2R ∑x 50.09
=
OBSERVATIONS (SLIP-RING DIAMETER, CM)
x= = = 5.01 cm
SAMPLE k 1 k2 3 10 4 5 x R
1 5.02 5.01 4.94 4.99 4.96 4.98 0.08
2 5.01 = 5.03 5.07 4.95 4.96 5.00 0.12
x1 + x2 + ... xk UCL
4.99= x5.00
+ A24.93
R = 5.01
4.92 + 4.99
(0.58)(0.115) = 5.08
x= =
3 4.97 0.08
k 4 5.03 4.91 5.01 4.98 4.89 4.96 0.14
LCL = - A R = 5.01 - (0.58)(0.115) = 4.94
5 4.95= x4.92 2 5.03 5.05 5.01 4.99 0.13
6 4.97 5.06 5.06 4.96 5.03 5.01 0.10
where 7 5.05 5.01 5.10 4.96 4.99 5.02 0.14
8 5.09 5.10 5.00 4.99 5.08 5.05 0.11
x= = the average of the sample means 9 5.14 5.10 4.99 5.08 5.09 5.08 0.15
10 5.01 4.98 5.08 5.07 4.99 5.03 0.10
R bar = the average range values 50.09 1.15

Example 15.3

Question 15 from Sample Exam3:

Average Thickness: 005 inch
Average Range: 0015 inch Using x- and R-charts together
Sample Size: 3

x-Chart: 9 Each measures the process

Sample
differently
A2 D3 D4
Size
2 1.88 0 3.27
9 Both process average (x bar chart)
3 1 02
1.02 0 2 57
2.57
and variability
y ((R chart)) must be in
4 0.73 0 2.28
control
5 0.58 0 2.11
6 0.48 0 2.00

38
 Sample Size Determination Sample Size Determination

9 Attribute control charts (p chart) 9 Variable control charts (R- & x bar- charts)
• 50 to 100 parts in a sample • 2 to 10 parts in a sample

Process Capability Process Capability Measures

• Control limits (the “Voice of the Process” or

the “Voice of the Data”): based on natural Process Capability Ratio (Cp )
variations ((common causes))
ttolerance
l range
Cp = process range
• Tolerance limits (the “Voice of the Customer”):
customer requirements
upper specification limit -
lower specification limit
• Process Capability: A measure of how =
“
“capable”
bl ” th
the process iis to
t meett customer
t 6σ
requirements; compares process limits to
tolerance limits

39
Process Capability Measures Computing Cpk
Process Capability Index ( Cpk ) Munchies Snack Food Company

Net weight specification = 9.0 oz ± 0.5 oz

Process mean = 8.80 oz
= Process standard deviation = 0.12 oz
x - lower specification limit
,
3σ
Cpk = minimum =
upper specification limit - x =
x - lower specification limit
3σ ,
Cpk = minimum 3σ
=
upper specification limit - x
Design
D i 3σ
Specifications

8.80 - 8.50 9.50 - 8.80

= minimum , = 0.83
3(0.12) 3(0.12)
Example 15.7

Process

Question 14 of Sample Exam3 The Process Capability Index

Computing Cpk
Net weight specification = 5 inches. ± 0.004 inches
Process mean = 5.001 inches Cpk < 1 Not Capable
p
Process standard deviation = 0.001 inches
Cpk > 1 Capable at 3σ
Cpk > 1.33 Capable at 4σ
Cpk > 1.67 Capable at 5σ
Cpk > 2 Capable at 6σ

40
 Six Sigma Improvement Methods
DMAIC vs. DMADV

Define

Measure
Six Sigma equals 3.4 defects
per million opportunities Continuous
Analyze
Improvement Reengineering

Improve Design

Control Validate

Chapter 11
Implementation Prototype Institutionalized
Model Model

Effort: 1X Effort: 10X-100X

Institutionalized
Modeling
Modeling
Application
Application
Effort: 10X Effort: 100X – 1000X

41
 The Separation of Players Curse The Curse of Scope Creep

Narrow Wide

Modeler
Model(s) Single Multiple
Objective(s) Single Multiple
Activity Focused Diffused
Modeler,
Project Curse of Decision Players Few Many
Manager, Client Maker
Decision Player Separation
Maker,
Stakeholders Few Many
Client
Effort Low High
Cost Low High
Project Manager
Development Risk Low High
Single player Multiple players

The Curse of Scope Creep

Narrow Wide
Coordination & project Informal Formal
management
Project visibility Low High
Economies of scale:
- Information systems None Many
- Model & database maint. None Many
Model use & support Deteriorates Wide
Potential org. impact Low High

42