Forecasting

Forecasting techniques (pg. 436 Exhibit 11.1)
1.  Statistical (Time Series, Causal)
2.  Judgement/Qualitative (Expert opinion, Market Survey, Delphi)

Time series analysis
1.  Simple moving average
2.  Weighted moving average
3.  Exponential smoothing
4.  Regression analysis

An Example

Month

Demand

3-month Moving Average

3-month Wt. Moving Average 

(weights: 0.2, 0.3, 0.5)

Exponential Smoothing 

(alpha = 0.1)

1

650

 

 

 

2

700

 

 

0.1*650+0.9*650 = 

650

3

810

 

 

0.1*700+0.9*650 =

655

4

800

(650+700+810)/3 = 720

0.2*650+0.3*700+0.5*810 = 

745

0.1*810+0.9*655 = 

670.5

5

900

(700+810+800)/3 = 770

0.2*700+0.3*810+0.5*800 = 

783

0.1*800+0.9*670.5 = 

683.5

6

700

(810+800+900)/3 = 837

0.2*810+0.3*800+0.5*900 = 

852

0.1*900+0.9*683.5 = 705.2

7

 

(800+900+700)/3 = 800

0.2*800+0.3*900+0.5*700 =

 780

0.1*700+0.9*705.2 =

704.7

Illustration

Pg.472 Problem 2

Exercise:

pg473 Problem 9

Forecasting errors
Mean Absolute Deviation (MAD) (pg.448)
 

An Example
 

Demand

3-month Moving Average

Deviation

Absolute

Deviation

800

720

800-720 = 80

80

900

770

900-770 = 130

130

700

837

700-837 = -137

137

Sum of Absolute Deviation = 80+130+137 = 347

MAD = 347/3 = 115.7
 

Demand

3-month Wt. Moving Average

Deviation

Absolute

Deviation

800

745

800-745 = 55

55

900

783

900-783 = 117

117

700

852

700-852 = -152

152

Sum of Absolute Deviation = 55+117+152 = 324

MAD = 324/3 = 108
 

Demand

Exponential Smoothing

Deviation

Absolute

Deviation

800

670.5

800-670.5 = 129.5

129.5

900

683.5

900-683.5 = 216.5

216.5

700

705.2

700-705.2 = -5.2

5.2

 

Sum of Absolute Deviation = 129.5+216.5+5.2 = 351.2

 

MAD = 351.2/3 = 117.1

 

Hence, the 3-mth weighted moving average has the lowest MAD and is the best forecast method among the three.

 

Control limits for a range of MADs (Pg.450 Exhibit 11.11)

Number of MADs

Accuracy

+/- 1

57%

+/- 2

88.9%

+/- 3

98.3%

+/- 4

99.9%

With 57% accuracy, the forecast demand for July using 3-mth Wt. Moving Average = 780 +/- 108 (672 to 888)

With 88.9% accuracy, the forecast demand for July using 3-mth Wt. Moving Average = 780+/- 2*108 (564 to 996)

Exercise:

pg.471 Problem 3, 11


Regression Analysis

Assumptions
1.  Linear -- the past data and future projections are fall about a straight line (least squares method: minimize the sum of squared forecast error)
2.  Time is the independent variable, x

Y = a + bx

An example
 

 

Month (x)

Profit (y)

xy

x2

 

1

31

31

1

 

2

40

80

4

 

3

30

90

9

 

4

34

136

16

 

5

25

125

25

 

6

20

120

36

Total

21

180

582

91

Average

3.50

30.00

97.00

15.17

b = (582-6*3.5*30)/(91-6*3.5*3.5) = -2.7

a = 30-(-2.7)*3.5 = 39.6

Y = 39.6 - 2.7x


Coding the time variable to simplify computation (i.e., 

Case 1: odd number of time elements
 

Time

Code

January

-2

February

-1

March

0

April

1

May

2

Case 2: even number of time elements
 

Time

Code

January

-3

February

-1

March

1

April

3

If then


Example
 

 

Month

Profit (y)

Code (x)

x2

xy

 

1

31

-5

25

-155

 

2

40

-3

9

-120

 

3

30

-1

1

-30

 

4

34

1

1

34

 

5

25

3

9

75

 

6

20

5

25

100

Total

21

180

0

70

-96

Average

3.50

30.00

0.00

11.67

-16.00

b = -96/70 = -1.4

a = 30

Y = 30 - 1.4x


Forecasting error

Standard error of estimate

or

 

Month (x)

Profit (y)

xy

y2

 

1

31

31

961

 

2

40

80

1600

 

3

30

90

900

 

4

34

136

1156

 

5

25

125

625

 

6

20

120

400

Total

21

180

582

5642

= 5.25 or
 
 

= 5.25
 
 

Number of Syx

Accuracy

+/- 1

68%

+/- 2

95.5%

+/- 3

99.7%

Formula review (pg.466)

Exercise:

Pg.471 Problems 17

Using Excel
1.  Click Tools, Click Data Analysis
2.  Choose Moving Average/Exponential Smoothing/Regression
 

Method

Parameter

Excel terminology

Reminder

N-mth Moving average

Interval

Output range should be one cell lower than the input range

Exponential Smoothing

1-a

Damping factor

Output range should be at the same row as the input range

Regression

a

b

Intercept

X variable or Label

Label should be checked if you include the column heading in your input ranges