Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Management
Forecasting
Chapter 4 - Part 2
4-1
Forecasting a Trend
Trend is increasing or decreasing pattern.
First, plot data to verify trend.
If trend exists, then moving averages and
exponential smoothing will always lag.
4-2
Plot Data
20
Actual
16
12
8
4
1
Period
4-3
MA
MA
Erro
r
10.67
13.00
15.67
4.33
6.00
?
Trend Graph
20
Actual
16
MA Forecast
12
8
4
1
Period
4-5
MA
10.67
13.00
15.67
MA
Erro
r
4.33
6.00
?
ES
11
11
12
13.5
16.25
ES
Error
3.0
5.5
?
Trend Graph
20
ES Forecast
Actual
16
MA Forecast
12
8
4
1
Period
4-7
Forecasting a Trend
Moving Averages and (simple) Exponential
Smoothing are always poor.
For a linear trend can use:
4-8
Y = a + bX
4-9
Actual
observation
Time (x)
4-10
Least Squares
Values of Dependent Variable (Y)
Actual
observation
Deviation
Deviation
Deviation
Deviation
Deviation
Deviation
Deviation
Y a bx
Time (x)
4-11
Point on
regression
line
Least Squares
Least squares line minimizes sum of squared
deviations.
This reduces large errors.
Similar to MSE.
4-12
i a bx i
Y
n
x i y i nx y
b in
x i nx
i
Y-Intercept:
a y bx
4-14
Sales
(y)
x2
1
8
8
4
11
22
39
9
13
60
16
15
25
19
95
y=13.2 xy=224 x2=55
224 5 3 13.2
b
2.6
2
55 5 3
4-15
xy
66
15
a 2.6 5.4
5
5
MA
10.67
13.00
15.67
MA
Err.
ES
ES
Err. TP
TP
Err.
11
11
4.33 12 3.0 15.8 -0.8
6.00 13.5 5.5 18.4 0.6
16.25
21.0
Small errors!
Trend Graph
TP Forecast
20
ES Forecast
Actual
16
MA Forecast
12
8
4
1
Period
4-17
4-18
Determine relationship.
Plot data.
Calculate parameters.
Forecast.
Monitor.
4-19
Linear Regression
Linear relationship between dependent & explanatory
variables.
Yi = a + b X i
Dependent variable (sales).
Independent variable
(number of ads).
4-20
Least Squares
Values of Dependent Variable (Y)
Actual
observation
Deviation
Deviation
Deviation
Deviation
Deviation
Deviation
Deviation
Y a bx
Values of Independent Variable (x)
4-21
Point on
regression
line
i a bx i
Y
n
x i y i nx y
b i n
x i nx
Slope:
a y bx
Y-Intercept:
4-22
Interpretation of Coefficients
Slope (b):
Y changes by b units for each 1 unit increase in X.
If b = +2, then sales (Y) is forecast to increase by 2 for
each 1 unit increase in advertising (X).
Y-intercept (a):
Average value of Y when X = 0.
If a = 4, then average sales (Y) is expected to be 4
when advertising (X) is 0.
4-23
Least Squares
Plot data to verify linearity!
Sales
4-25
Sales
What is sales
forecast for
small number
of ads?
Y a bx
Number of TV ads per month
4-26
Sales
Forecast is for
negative sales!
Y a bx
Number of TV ads per month
4-27
Correlation
Answers: How strong is the linear relationship
between the variables?
Coefficient of correlation - r
Coefficient of determination - r2
n x i yi x i yi
n
x
x
n
y
i
i
i
i
i
i
i
i
n
4-29
Coefficient of Correlation
r = +1
r = -1
Y
X
r = .89
r=0
Y
X
4-30
4-31
Error
Time
Time
4-32
1
1
2
2
4
0.6
1.3
2.0
2.7
3.4
4-33
Forecast (.9)
1.00
1.00
1.00
1.90
1.99
Yi
Fcast
Error
Error2
|Error|
1
2
3
4
5
Total
1
1
2
2
4
0.6
1.3
2.0
2.7
3.4
0.4
-0.3
0.0
-0.7
0.6
0.0
0.16
0.09
0.00
0.49
0.36
1.10
0.4
0.3
0.0
0.7
0.6
2.0
Yi
1
2
3
4
5
Total
1
1
2
2
4
Fcast
1.00
1.00
1.00
1.90
1.99
Error
0.0
0.0
1.0
0.1
2.01
0.3
Error2
0.00
0.00
1.00
0.01
4.04
5.05
|Error|
0.0
0.0
1.0
0.1
2.01
3.11
Which is Better???
Linear Regression Model:
MSE = Error2 / n = 1.10 / 5 = 0.220
MAD = |Error| / n = 2.0 / 5 = 0.400
MAPE = [|Error|/Actual]/n = 1.2/5 = 0.24 = 24%
Exponential Smoothing Model:
MSE = Error2 / n = 5.05 / 5 = 1.01
MAD = |Error| / n = 3.11 / 5 = 0.622
MAPE = [|Error|/Actual]/n = 1.0525/5 = 0.2105 = 21%
4-36
Tracking Signal
Measures how well the forecast is predicting
actual values.
To use:
deviation (MAD).
4-37
MAD
0
-
Tracking signal
Acceptable range
Lower control limit
Time
4-38
RSFE
MAD
y i y i
i 1
MAD
4-39
(error )
MAD
100
Act Error
RSFE
90
4-40
Cum
|Error|
MAD
TS
100
Act Error
RSFE
-10
-10
90
Cum
|Error|
MAD
TS
Error
Error==Actual
Actual--Forecast
Forecast
==90
90--100
100==-10
-10
RSFE
RSFE== Errors
Errors ==-10
-10
4-41
100
Act Error
RSFE
-10
-10
90
Cum
|Error|
MAD
10
Cum
Cum|Error|
|Error|== |Errors|
|Errors|
== 10
10
4-42
TS
100
Act Error
RSFE
-10
-10
90
Cum
|Error|
10
MAD
10.0
MAD
MAD== |Errors|/n
|Errors|/n
==10/1
10/1==10
10
4-43
TS
100
Act Error
RSFE
-10
-10
90
Cum
|Error|
10
TS
TS==RSFE/MAD
RSFE/MAD
==-10/10
-10/10==-1
-1
4-44
MAD
10.0
TS
-1
Act Error
RSFE
-10
-10
100
90
99
94
4-45
Cum
|Error|
10
MAD
10.0
TS
-1
Act Error
RSFE
-10
100
90
-10
99
94
-5
Cum
|Error|
10
MAD
10.0
Error
Error==Actual
Actual--Forecast
Forecast
==94
94--99
99==-5
-5
4-46
TS
-1
Act Error
RSFE
100
90
-10
-10
99
94
-5
-15
Cum
|Error|
10
MAD
10.0
RSFE
RSFE== Errors
Errors
==(-10)
(-10)++(-5)
(-5)==-15
-15
4-47
TS
-1
Act Error
RSFE
Cum
|Error|
100
90
-10
-10
10
99
94
-5
-15
15
MAD
10.0
Cum
CumError
Error== |Errors|
|Errors|
==10
10++55==15
15
4-48
TS
-1
Act Error
RSFE
Cum
|Error|
MAD
100
90
-10
-10
10
10.0
99
94
-5
-15
15
7.5
MAD
MAD== |Errors|/n
|Errors|/n
==15/2
15/2==7.5
7.5
4-49
TS
-1
Act Error
RSFE
Cum
|Error|
MAD
TS
100
90
-10
-10
10
10.0
-1
99
94
-5
-15
15
7.5
-2
TS
TS==RSFE/MAD
RSFE/MAD
==-15/7.5
-15/7.5==-2
-2
4-50
Act Error
RSFE
Cum
|Error|
MAD
TS
100
90
-10
-10
10
10.0
-1
99
94
-5
-15
7.5
-2
98
113
15
15
3
0
4-51
1
0
Act Error
RSFE
Cum
|Error|
MAD
TS
100
90
-10
-10
10
10.0
-1
99
94
-5
-15
7.5
-2
98
113
15
105
95
-10
-10
104
119
15
110
140
30
35
15
3
0
4
0
5
5
8
5
4-52
1
0
10
-1
11
.45
14.2 2.47
Forecast
Actual demand
4
Month
4-53
Tracking Signal
Tracking Signal
3
2
1
0
-1
-2
-3
0
4
Time
4-54
1
1
2
2
4
0.6
1.3
2.0
2.7
3.4
4-56
Forecast (.9)
1.00
1.00
1.00
1.90
1.99
Yi
Fcast
Error
1
2
3
4
5
1
1
2
2
4
0.6
1.3
2.0
2.7
3.4
0.4
-0.3
0.0
-0.7
0.6
4-57
MAD
0.4
0.35
0.233
0.35
0.40
TS
1.0
0.29
0.43
-1.71
0.0
Yi
1
2
3
4
5
1
1
2
2
4
Fcast
1.00
1.00
1.00
1.90
1.99
4-58
Error
0.0
0.0
1.0
0.1
2.01
MAD
0.0
0.0
0.33
0.275
0.622
TS
0.0
0.0
3.0
4.0
5.0
Tracking Signals
Exponential Smoothing
Tracking Signal
3
2
1
0
-1
Linear Regression
-2
-3
0
4
Year
4-59
4-60
Forecasting Summary
Determine purpose of forecast first.
Plot data.
Use several appropriate methods.
Continually monitor, evaluate and adjust
methods to improve forecasts.
4-61