Professional Documents
Culture Documents
M04 Heiz5577 01 Se C04
M04 Heiz5577 01 Se C04
C H A P T E R
Forecasting
DISCUSSION
QUESTIONS
1. Qualitative models
incorporate subjective factors
into the forecasting model.
Qualitative models are useful
where all previous values are
weighted with a set of
weights that decline
exponentially.
8. MAD, MSE, and MAPE
are common measures of
forecast accuracy. To find the
4 10. A time series model
predicts on the basis of the
assumption that the future is
a function of the past,
whereas an associative model
incorporates into the model
the variables of factors that
together, but move in
opposite directions.
18. Independent variable
(x) is said to explain
variations in the dependent
variable (y).
when subjective factors are 19. Nearly every industry
more accurate forecasting might influence the quantity
important. When quantitative has seasonality. The
model, forecast with each being forecast.
data are difficult to obtain, seasonality must be filtered
tool for several periods where 11. A time series is a out for good mediumrange
qualitative models may be
the demand outcome is sequence of evenly spaced data planning (of production and
appropriate.
known, and calculate MSE, points with the inventory) and performance
2. Approaches are MAPE, or MAD for each. four components of trend, evaluation.
qualitative and quantitative. The smaller error indicates seasonality, cyclical, and
Qualitative is relatively 20. There are many
the better forecast. random variation.
subjective; quantitative uses examples. Demand for raw
9. The Delphi technique 12. When the smoothing materials and component
numeric models.
involves: constant, , is large (close to parts such as steel or tires is a
3. Shortrange (under 3 1.0), more weight is given to
(a) Assembling a group function of demand for goods
months), mediumrange (3 recent data; when is low
of experts in such a such as automobiles.
months to (close to 0.0), more weight is
manner as to preclude 21. Obviously, as we go
3 years), and longrange (over given to past data.
direct communication farther into the future, it
3 years).
between identifiable 13. Seasonal patterns are of becomes more difficult to
4. The steps that should be members of the group fixed duration and repeat make forecasts, and we must
used to develop a regularly. Cycles vary in diminish our reliance on the
(b) Assembling the
forecasting length and regularity. forecasts.
responses of each
system are: Seasonal indexes allow
expert to the
(a) Determine the questions or “generic” forecasts to be ETHICAL DILEMMA
purpose and use of the problems of interest made specific to the month, This exercise, derived from
forecast week, etc., of the application. an actual situation, deals as
(c) Summarizing these
(b) Select the item or responses 14. Exponential smoothing much with ethics as with
quantities that are to be weighs all previous values forecasting. Here are a few
(d) Providing each points to consider:
forecasted with a set of weights that
expert with the summary of
(c) Determine the time decline exponentially. It can No one likes a system
all responses
4.2 (a) No, the data appear to have no consistent pattern
(see part d for graph).
Year 1 2 3 4 5 6 7 8 9 10 11 Forecas
t
Demand 7 9 5 9.0 13.0 8.0 12.0 13.0 9.0 11.0 7.0
(b) 3-year moving 7.0 7.7 9.0 10.0 11.0 11.0 11.3 11.0 9.0
(c) 3-year weighted 6.4 7.8 11.0 9.6 10.9 12.2 10.5 10.6 8.4
sense. (except for the first Trend Adjustment 374 + 368 + 381
4.1 (a) 374.33 pints
Regression is open to forecast) but is offset by one 1. Scroll through different 3
abuse as well. Models period. This is a naive values for alpha and beta. (b)
can use many years of forecast. Which smoothing constant
data yielding one 3. Generalize what happens appears to have the greater Weight
result or few years Week Pint ed
to a forecast as alpha effect on the graph?
of s Moving
yielding a increases. alpha
Use Averag
totally different As alpha increases the 2. With beta set to zero, d e
forecast. Selection of forecast is more sensitive to find the best alpha and August 360
associative variables changes in demand. observe the MAD. Now find 31
can have a major *Active Models 4.1, 4.3, and the best beta. Observe the Septe 389 381
impact on results as 4.4 appear on the CDROM mber 7 .1
well. only. Active Model 4.2 also Forecasti Error =
appears in the text. Week of Pints Forecas ng .20 Foreca38.1
Active Model t Septe
Error 410 st 368
mber .3
Exercises* August 31 360 360 0
14
0 360
=
September 7 389 360 29 5.8 365.8
110.
ACTIVE MODEL 4.1: September 14 410 365.8 44.2 8.84 374.64
4
Moving Averages September 21 381 374.64 6.36
Septe 381
1.272 375.912
374
September 28 368 375.912 –7.912 –1.5824 374.329
1. What does the graph mber
6
.6
look like when n = 1? 21 =
October 5 374 374.3296–.3296 –.06592 374.263
224.
The forecast graph 6
4
mirrors the data graph but MAD. Does the addition of a Septe 368
one period later. trend improve the forecast? mber
2. What happens to the alpha = .11, MAD = 28
graph as the number of 2.59; beta above .6 changes Octobe 374
periods in the moving the MAD (by a little) to r5
average increases? 2.54. Forec
ast
The forecast graph
becomes shorter and ACTIVE MODEL 4.4: 372.9
miles.
Two-Year
Year Mileage Moving Average
� 420 �
MAD 140 �
1 3,000
� 3 � �
2 4,000
3 3,400 3,500 4.5 (d)
4 3,800 3,700 Forecast
5 3,700 3,600 Year Mileage Forecast Error
1 3,000 3,000
2 4,000 3,000 1,000
3 3,400 3,500
300 4 3,800 3,450
MAD 100.
3 5 3,700 3,625
Total
The forecast is 3,663 miles.
4.6
Y Sales X Period
January 20
February 21
March 15
April 14
Naive tracks the ups May 13
and downs best but lags June 16
the data by one period. July 17
Exponential smoothing August 18
is probably better September 20
October 20 10
because it smoothes the
November 21 11
data and does not have
December 23 12
as much variation.
Sum 218 78
TEACHING NOTE: Average 18.2
Notice how well
exponential smoothing
forecasts the naive. (a)
[v] Trend
� x 78, x 6.5, � y = 218, y 18.17
1474 - (12)(6.5)(18.2) 54.4
b 0.38
650 - 12(6.5)2 143
a 18.2 - 0.38(6.5) 15.73
29 CHAPTER 4 F O R E C A S T I N G
4.9 (d) Table for Problem 4.9(d):
(b) |Error| = |Actual – Forecast|
= .1 = .3 = .5
Year 1 2 3 4 5 6 7 8 9 10 11 MAD
Month Price
Exp.per Forecast 1 |Error|
smoothing 1.3 1.1 Forecast
0.2 |Error|
5.2 1.6 Forecast
0.1 2.1|Error|
4.5 5.1 4.6 2.4
Chip
These calculations were completed in Excel. Calculations are slightly different in Excel OM and POM for Windows due to
January $1.80 $1.80 $.00 $1.80 $.00 $1.80 $.00
rounding differences.
February 1.67 1.80 .13 1.80 .13 1.80 .13
March 1.70 1.79 .09 1.76 .06 1.74 .04
April 1.85 1.78 .07 1.74 .11 1.72 .13
May 1.90 1.79 .11 1.77 .13 1.78 .12
June 1.87 1.80 .07 1.81 .06 1.84 .03
July 1.80 1.80 .00 1.83 .03 1.86 .06
August 1.83 1.80 .03 1.82 .01 1.83 .00
Septembe 1.70 1.81 .11 1.82 .12 1.83 .13
r
October 1.65 1.80 .15 1.79 .14 1.76 .11
November 1.70 1.78 .08 1.75 .05 1.71 .01
December 1.75 1.77 .02 1.73 .02 1.70 .05
Totals $.8 $.8 $.81
6 6
MAD (total/12) $.072 $.072 $.067
5
= .5 is preferable, using MAD, to = .1 or = .3. One could
3 52 43.4 + 0.6(50–43.4) =
4.10 Year 1 2 3 4 5 6 7 8 9 10 11 Forecast
4 56 47.4 + 0.6(52–47.4) =
Demand 4 6 4 5.0 10.0 8.0 7.0 9.0 12.0 14.0 15.0 5 58 50.2 + 0.6(56–50.2) =
(a) 3-year moving 4.7 5.0 6.3 7.7 8.3 8.0 9.3 11.7 13.7
6 ? 53.7 + 0.6(58–53.7) =
(b) 3-year 4.5 5.0 7.3 7.8 8.0 8.3 10.0 12.3 14.0
weighted
4.12 = 25.3
MAD = 5.06
Actual
t Day Demand Exponential smoothing, =
0.9:
1 Monday 88
Exponential
2 Tuesday 72
Year Demand Smoothing = 0.9
3 Wednesday 68
1 45 41
4 Thursday 48 2 50 41.0 + 0.9(45–41) = 44.6
5 Friday 3 52 44.6 + 0.9(50–44.6 ) = 49
4 56 49.5 + 0.9(52–49.5) = 51
Ft = Ft–1 + (At–1 – Ft–1) 5 58 51.8 + 0.9(56–51.8) = 55
Let = .25. Let Monday 6 ? 55.6 + 0.9(58–55.6) = 57
forecast demand = 88 = 18.5
(c) The forecasts are
F2 = 88 + .25(88 – 88) = 88 MAD = 3.7
about the same.
+ 0 = 88 (b) 3year moving
4.11 (a) Year 1 2 3 4 5 6 7 8 9 10 11 Forecast
Demand 4 6.0 4.0 5.0 10.0 8.0 7.0 9.0 12.0 14.0 15.0
Exp. Smoothing 5 4.7 5.1 4.8 4.8 6.4 6.9 6.9 7.5 8.9 10.4 11.8
F3 = 88 + .25(72 – 88) = 88 average:
– 4 = 84 Three-Year
F4 = 84 + .25(68 – 84) = 84 Year Demand Moving Average
– 4 = 80 1 45
F5 = 80 + .25(48 – 80) = 80 2 50
– 8 = 72 3 52
4 56 (45 + 50 + 52)/3 = 49
4.13 (a) Exponential 5 58 (50 + 52 + 56)/3 =
smoothing, = 0.6: 52.7
6
Exponential ? (52 + 56 + 58)/3 =
Year Demand Smoothing 55.3
1 45 41 = 12.3
2 50 41.0 + 0.6(45–41) = 43.4
CHAPTER 4 F O R E C A S T I N G 30
4.18 We need to find the all numbers are rounded. 0.4 ( 2.6 ) + 1.34
smoothing constant . We Note: To use POM for 1.04 + 1.34 2.38
know in general that Ft = Windows to solve this
FIT6 F6 + T6 22.51 + 2.38 24.89
Ft–1 + (At–1 – Ft–1); t = 2, problem, a period 0, which
3, 4. Choose either contains the initial forecast
and initial trend, must be 4.22 F7 A6 + (1 �)( F6 + T6 ) (0.2)(21) + (0.8)(24.89)
t = 3 or t = 4 (t = 2 won’t let
us find because F2 = 50 = added. 4.2 + 19.91 24.11
50 + (50 – 50) holds for T7 ( F7 �F6 ) + (1 �)T6 (0.4)(24.11 �22.51)
any ). Let’s pick t = 3. Then + (0.6)(2.38) 2.07
F3 = 48 = 50 + (42 – 50)
FIT7 F7 + T7 24.11 + 2.07 26.18
or 48 = 50 + 42
– 50 F8 A7 + (1 � )( F7 + T7 ) (0.2)(31)
or –2 = –8 + (0.8)(26.18) 27.14
So, .25 = T8 ( F8 �F7 ) + ( 1 � ) T7 0.4 ( 27.14 �24.11)
Now we can find F5 : F5 =
+ 0.6 ( 2.07) 2.45
50 + (46 – 50)
F5 = 50 + 46 – 50
= 50 – 4
For = .25, F5 = 50 –
4(.25) = 49
The forecast for time period 5
= 49 units.
CHAPTER 4 F O R E C A S T I N G 32
6
�xi yi - nx y
i 1 (9, 783) - 6(3.25833)(550)
b 6
2 2 67.1925 - 6 (3.25833)2
�x i - nx
i 1
-969.489
-277.6
3.49222
a y - bx 1, 454.6
4.32 (a) x y xy x2
16 330 5,280 256
12 270 3,240 144
18 380 6,840 324
14
CHAPTER 300
4 F O R E C A S4,200
TING 196 34
60 1,280 19,560 920
3 73.4 40
4 75.7 41
5 81.1 40
(b) Developing the
6 89.0 55 regression
relationship, we have:
7 101.1 60 (Summe Tourists Ridership
CHAPTER 4 F O R E C A S T I N G 36
standard error
of the estimate and
the MAD, the 0.2
constant is
better. However,
other smoothing
constants need to
be examined.
Actual Smoothed
t A(t) Ft ( =
0.2)
1 50 +50.0
2 35 +50.0
3 25 +47.0
4 40 +42.6
5 45 +42.1
6 35 +42.7
7 20 +41.1
8 30 +36.9
9 35 +35.5
10 20 +35.4
11 15 +32.3
12 40 +28.9
13 55 +31.1
14 35 +35.9
15 25 +36.7
16 55 +33.6
17 55 +37.8
18 40 +41.3
19 35 +41.0
20 60 +39.8
21 75 +43.9
22 50 +50.1
23 40 +50.1
24 65 +48.1
25 +51.4
MAD
(c) Students should
note how stable
the smoothed
values are for =
0.2. When
compared to actual
week 25 calls of
85, the smoothing
constant, = 0.6,
appears to do a
slightly better job.
On the basis of the
CHAPTER 4 F O R E C A S T I N G 38
4.44 XY - nXY
b
Week Actual Value Smoothed 87Trend Estimate Forecast Forecas X 2 - nX 2
So: MAD: 10.875
Value 8 t
a Y - bX
t At Ft ( = 0.3)
39 Tt ( = 0.2) FITt Error
3.586
1 50.000 50.00010.875 0.000 50.000 0.000
2 35.000 50.000 0.000 50.000 –15.000
3 25.000 45.500 –0.900 44.600 –19.600
4 40.000 38.720 –2.076 36.644 3.356
5 45.000 37.651 –1.875 35.776 9.224
6 35.000 38.543 –1.321 37.222 –2.222
7 20.000 36.555 –1.455 35.101 –15.101
8 30.000 30.571 –2.361 28.210 1.790
9 35.000 28.747 –2.253 26.494 8.506
10 20.000 29.046 –1.743 27.303 –7.303
11 15.000 25.112 –2.181 22.931 –7.931
12 40.000 20.552 –2.657 17.895 22.105
13 55.000 24.526 –1.331 23.196 31.804
14 35.000 32.737 0.578 33.315 1.685
15 25.000 33.820 0.679 34.499 –9.499
16 55.000 31.649 0.109 31.758 23.242
17 55.000 38.731 1.503 40.234 14.766
18 40.000 44.664 2.389 47.053 –7.053
19 35.000 44.937 1.966 46.903 –11.903 Note: To use POM for Windows
20 60.000 43.332 1.252 44.584 15.416 to solve this problem, a period 0,
21 75.000 49.209 2.177 51.386 23.614 which contains the initial fore
22 50.000 58.470 3.594 62.064 –12.064 cast and initial trend, must be
23 40.000 58.445 2.870 61.315 –21.315
added.
24 65.000 54.920 1.591 56.511 8.489
25 59.058 2.100 61.158
4.46 X Y
To evaluate the trend
adjusted exponential 421 2.90
smoothing model, actual
377 2.93
week 25 calls are compared
to the forecasted value. The 585 3.00
model appears to be
producing a forecast 690 3.45
approximately midrange
between that given by simple 608 3.66
exponential smoothing using
= 0.2 and = 0.6. Trend 390 2.88
adjustment does not appear to
give 415 2.15
any significant improvement.
481 2.53
n 729 3.22
�( At - Ft )
4.45 t 1
Tracking signal 501 1.99
MAD
Month At Ft 613 2.75
May 100 100
709 3.90
June 80 104
July 110 99
366 1.60
August 115 101
Septemb 105 104
6885 36.96
er Column totals
October 110 104
Novembe 125 105
r Given: Y = a + bX where:
Decembe 120 109
r
39 CHAPTER 4 F O R E C A S T I N G
and X = 6885, Y = 36.96, February’s 4.48 (a) 27 7.70 5.497 2.20
2 28 10.10 6.818 3.28
XY = 20299.5, X = forecast equal to Quarter Contracts Sales Y 29 15.20 8.787 6.41
3857893, January’s sales) X
2
Y = 110.26, X = 529.6, with alpha 0.1. (Continued)
1 153 8
Y = 2.843, Then: Barbara wants to 2 172 10
use a 3period 3 197 15
20299.5 - 13 529.6 2.843 20299.5 - 19573.5
moving
b 4 178 9
3857893 - 13 529.62 average.
3857893 - 3646190 5 185 12
726 6 199 13
0.0034 Sales Amit Barba
211703 7 205 12
a 2.84 - 0.0034 529.6 1.03 January 400 —
8 226 16
Totals 1,51 95
and Y = 1.03 + 0.0034X February 380 400
March 410 398 5
As an indication of the Average 189.375
April 375 399.2 396.67
usefulness of this May 405 396.8 388.33 11.875
relationship, we can calculate MAD b = (18384 – 8 189.375
the correlation coefficient:
11.875)/(290,413 – 8
(d) Note that Amit has more 189.375
n XY - X Y
r forecast observations, 189.375) = 0.1121
n X 2 - ( X ) 2 n Y 2 - ( Y )while Barbara’s moving
2
a = 11.875 – 0.1121
average does not start 189.375 = –9.3495
13 20299.5 - 6885 36.96 until month 4. Also note Sales (y) = –9.349 + 0.1121
that the MAD for Amit
13 3857893 - 68852 13 110.26 - 36.962 (Contracts)
is an average
of 4
(b)
263893.5 - 254469.6 numbers, while
Barbara’s is only 2. r (8 � 18384 - 1515 � 95) ((8 � 290,413 - 15152 )(8 � 1183 - 952 ))
50152609 - 47403225 1433.4 - 1366.0
Amit’s MAD for 0.8963
9423.9 exponential smoothing
Sxy (1183 - (-9.3495 � 95) - (0.112 � 18384 / 6) 1.3408
2749384 67.0 (16.11) is lower than
9423.9 that of Barbara’s r 2 .8034
0.692 moving average (19.17).
1658.13 8.21 4.49 (a)
2 So his forecast seems to
r 0.479 be better. Method Exponential Smoothing
A correlation coefficient of 0.6 =
Year Deposits Forecast
0.692 is not particularly high.
(Y )
The coefficient of
2
determination, r , indicates 1 0.25 0.25
that the model explains only 2 0.24 0.25
47.9% of the overall 3 0.24 0.244
4 0.26 0.241
variation. Therefore, while
5 0.25 0.252
the model does provide an 6 0.30 0.251
estimate of GPA, there is 7 0.31 0.280
considerable variation in 8 0.32 0.298
GPA, which is as yet 9 0.24 0.311
unexplained. For 10 0.26 0.268
11 0.25 0.263
X 350: Y 1.03 + 0.0034 � 350 2.22 12 0.33 0.255
X 800: Y 1.03 + 0.0034 � 800 3.75 13 0.50 0.300
14 0.95 0.420
Note: When solving this 15 1.70 0.738
problem, care must be taken 16 2.30 1.315
to interpret significant digits. 17 2.80 1.906
4.47 (a) There is not a 18 2.80 2.442
19 2.70 2.656
strong linear trend in sales
20 3.90 2.682
over time. 21 4.90 3.413
(b, c) Amit wants to 22 5.30 4.305
forecast by 23 6.20 4.90
exponential 24 4.10 5.680
smoothing (setting 25 4.50 4.732
26 6.10 4.592
CHAPTER 4 F O R E C A S T I N G 40
1 14 21
Northern Airlines—Engine
2 17 24 Maintenance Cost:
3 25 27 Cost = 20.57 + 0.0026
4 25 32 Airframe age
5 35 29 Coefficient of
determination = 0.6124
6 35 37 Coefficient of
correlation = 0.7825
7 45 43
Southeast Airlines—
8 50 43 Airframe Maintenance Cost:
Cost = 4.60 + 0.0032
9 60 54 Airframe age
Coefficient of
10 60 66 determination = 0.3905
43 CHAPTER 4 F O R E C A S T I N G