Professional Documents
Culture Documents
Analysis
Model Results
Interpretation
Decisions Performance
Abstraction
(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
Modeling
Techniques Modeling
Techniques Deductive
Modeling
Deterministic Probabilistic
Models Models
Time Series
Profit Model Decision Analysis Simulation
Forecasting
2
Deductive Focuses on
Modeling variables
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
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
INFERENTIAL MODELING
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?
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
7
Present clearly labeled input Clearly label model results.
variables together.
8
Format the spreadsheet to simplify Separate physical results from
interpretation. financial or economic results.
CHAPTER 9: Simulation
Introduction
9
When should simulation be used? Terminology to Know
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
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
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
c)) Exponential
p Smoothingg
i. Basic model
ii. Holt’s Model (exponential smoothing
with trend)
d) Seasonality
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
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
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
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
15 Bias = ?
16 Alpha = 0.75 MAD =
16
Measures of Comparison Errors, Absolute Errors, & Errors Squared
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
17
3. Forecast using deseasonalized data 4. Seasonalize the forecast to account
for the seasonal pattern
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)
18
Deseasonalized Data 2.Deseasonalize the Data
3,000.0
2,000.0
averages, where m is the length of the
1,500.0
seasonal pattern.
pattern
Coal (0
1,000.0
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
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
22
Chapter 8 Terminology
Decision
Analysis States of nature
Payoff / payoff
table
Probability
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
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
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
Multiple states of
nature
25
Decisions under Decisions under
uncertainty uncertainty
Maximin Maximin
extremelyy
conservative or
pessimistic
approach to
making decisions
26
Maximin Maximin
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
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.
P(W|E)
P(W|D)
28
Question 2 of Sample Exam3 Decision Trees
29
FOLDING BACK
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
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
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
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
average
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
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
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
p-Chart Example ~
Western Jeans Company Range ( R ) Chart
0.20
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
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
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
39
Process Capability Measures Computing Cpk
Process Capability Index ( Cpk ) Munchies Snack Food Company
Process
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
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
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