CHAPTER 4 F O R E C A S T I N G 27

C H A P T E R

Forecasting

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
DISCUSSION
QUESTIONS
1. Qualitative   models
incorporate subjective factors
into   the   forecasting   model.
Qualitative models are useful
when   subjective   factors   are
important. When quantitative
data   are   difficult   to   obtain,
qualitative   models   may   be
appropriate.
2. Approaches   are
qualitative   and   quantitative.
Qualitative   is   relatively
subjective;   quantitative   uses
numeric models.
3. Short­range   (under   3
months),   medium­range   (3
months   to  
3 years), and long­range (over
3 years).
4. The steps that should be
used   to   develop   a
forecasting
system are:
(a) Determine   the
purpose   and   use   of   the
forecast
(b) Select   the   item   or
quantities   that   are   to   be
forecasted
(c) Determine   the   time
horizon of the forecast
(d) Select   the   type   of
forecasting model to be used
(e) Gather the necessary
data
(f)  Validate  
forecasting model
(g) Make the forecast
(h) Implement  
evaluate the results
5. Any   three   of:   sales
planning,   production
planning and budgeting, cash
budgeting,   analyzing  various
operating plans.
6. There   is   no   mechanism
for   growth   in   these   models;
they   are   built   exclusively
from   historical   demand
values.   Such   methods   will
always lag trends.
7. Exponential smoothing is
a   weighted   moving   average
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 assumption that the future is
more   accurate   forecasting
model,   forecast   with   each
tool for several periods where
the   demand   outcome   is
known,   and   calculate   MSE,
MAPE,   or   MAD   for   each.
The   smaller   error   indicates
the better forecast.
9. The  Delphi   technique
involves:
(a) Assembling   a   group
of   experts   in   such   a
manner as to preclude
direct communication
between   identifiable
members of the group
(b) Assembling   the
responses   of   each
expert   to   the
questions   or
problems of interest
(c) Summarizing   these
responses
(d) Providing   each
expert   with   the   summary   of
all responses
(e) Asking   each   expert
to study the summary
of   the   responses
and respond again to
the   questions   or
the problems of interest.
(f)  Repeating   steps   (b)
through   (e)   several
times as necessary to
and obtain   convergence
in   responses.   If
convergence   has   not
been obtained by the
end   of   the   fourth
cycle,   the   responses
at   that   time   should
probably be accepted
and   the   process
terminated—little
additional
convergence is likely
if   the   process   is
continued.
10. A   time   series   model
predicts   on   the   basis   of   the
Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
a   function   of   the   past,
whereas an associative model
incorporates   into   the   model
the   variables   of   factors   that
might   influence   the   quantity
being forecast.
11. A   time   series   is   a
sequence of evenly spaced data
points   with   the
four   components   of   trend,
seasonality,   cyclical,   and
random variation.
12. When   the   smoothing
constant, , is large (close to
1.0), more weight is given to
recent   data;   when    is   low
(close to 0.0), more weight is
given to past data.
13. Seasonal patterns are of
fixed   duration   and   repeat
regularly.   Cycles   vary   in
length   and   regularity.
Seasonal   indices   allow  
“generic”   forecasts   to   be
made   specific   to   the   month,
week, etc., of the application.
14. Exponential   smoothing
weighs   all   previous   values
with   a   set   of   weights   that
decline   exponentially.   It   can
place   a   full   weight   on   the
most   recent   period   (with   an
alpha of 1.0). This, in effect,
is the naïve approach, which
places all its emphasis on last
period’s actual demand.
15. Adaptive   forecasting
refers   to   computer
monitoring   of   tracking
signals and self­adjustment if
a   signal   passes   its   present
limit.
16. Tracking   signals  alert
the user of a forecasting tool
to   periods   in   which   the
forecast   was   in   significant
error.
17. The   correlation
coefficient   measures   the
degree   to   which   the
independent   and   dependent
variables   move   together.   A
negative   value   would   mean
that as X increases, Y tends
to   fall.   The   variables   move
together,   but   move   in
opposite directions.
18. Independent   variable
(x)   is   said   to   explain
variations   in   the  dependent
variable (y).
19. Nearly   every   industry
has   seasonality.   The
seasonality   must   be   filtered
out   for   good   medium­range
planning   (of   production   and
inventory)   and   performance
evaluation.
20. There   are   many
examples.   Demand   for   raw
materials   and   component
parts such as steel or tires is a
function of demand for goods
such as automobiles.
21. Obviously,   as   we   go
farther   into   the   future,   it
becomes   more   difficult   to
make forecasts, and we must
diminish  our  reliance  on  the
forecasts.
ETHICAL DILEMMA
This   exercise,   derived   from
an   actual   situation,   deals   as
much   with   ethics   as   with
forecasting.   Here   are   a   few
points to consider:
 No one likes a system
they don’t understand,
and   most  
college   presidents
would   feel
uncomfortable   with
this one. It does offer
the   advantage   of
depoliticizing   the
funds   al­
location if used wisely
and   fairly.   But   to   do
so   means   all   parties
must have input to the
process   (such   as
smoothing   constants)
and all data need to be
open to everyone.
 The   smoothing
constants   could   be
selected by an agreed­
upon criteria  (such as
lowest MAD) or could
be   based   on   input
from   experts   on   the
board   as   well   as   the
college.
 Abuse of the system
is   tied   to   assigning
28
 CHAPTER 4 F O R E C A S T I N G 29

alphas   based   on   what
results   they   yield,
rather   than   what
alphas make the most
sense.
2. What   happens   to   the
graph   when   alpha   equals
one?
The forecast follows the
same pattern as the demand
4. At what level of alpha is
the   mean   absolute   deviation
(MAD) minimized?
alpha = .16
No,   they   are  not  the
same   values.   For   example,
an intercept of 57.81 with a
slope of 9.44 yields a MAD
of 7.17.

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
 Regression  is   open  to (except   for   the   first ACTIVE MODEL 4.3:  END-OF-CHAPTER
abuse as well. Models forecast) but is offset by one Exponential Smoothing 
can use many years of period.   This   is   a   naïve with
PROBLEMS
data   yielding   one forecast. Trend Adjustment 374 + 368 + 381
4.1   (a)    374.33 pints
result   or  few  years 3. Generalize what happens 1. Scroll   through   different 3
yielding   a   to   a   forecast   as   alpha values   for   alpha   and   beta. (b)
totally   different increases. Which  smoothing   constant
forecast.   Selection   of Weight
As   alpha   increases   the appears   to   have   the   greater Week Pint ed
associative   variables forecast is more sensitive to effect on the graph? of s Moving
can   have   a   major changes in demand. alpha Use Averag
impact   on   results   as
*Active Models 4.1, 4.2, 4.3, d e
well.
and   4.4   appear   on   our   Web August 360Error
Forecasti
site, Week of Pints Forecast 31ng  .20 Foreca
Active Model www.pearsonhighered.com/h Septe
Error 389 st 381
Exercises* eizer. August 31 360 360 mber
0 7 0 360 .1
September 7 389 360 29 5.8 = 
365.8
ACTIVE MODEL 4.1:  September 14 410 365.8 44.2 8.84 38.1
374.64
Moving Averages September 21 381 374.64 Septe
6.36 4101.272 375.91368
mber 2  .3
1. What   does   the   graph 14 =
September 28 368 375.912 –7.912 –1.5824 374.32
look like when n = 1?
96 110.
The   forecast   graph October 5 374 374.3296–.3296 –.06592 374.264
mirrors the data graph but Septe 381 36 374
one period later. mber  .6
2. With   beta   set   to   zero, 21 =
2. What   happens   to   the find   the   best   alpha   and 224.
graph   as   the   number   of observe the MAD. Now find 4
periods   in   the   moving the   best   beta.   Observe   the Septe 368
average increases? MAD. Does the addition of a mber
The   forecast   graph trend improve the forecast? 28
becomes   shorter   and alpha   =   .11,   MAD   = Octobe 374
smoother. r5
2.59; beta above .6 changes
Forec
3. What   value   for  n the   MAD   (by   a   little)   to
ast
minimizes the MAD for this 2.54.
372.9
data?
ACTIVE MODEL 4.4:  (c)
n = 1 (a naïve forecast)
Trend Projections The forecast is 374.26.
ACTIVE MODEL 4.2:  1. What is the annual trend
Exponential Smoothing in the data?
1. What   happens   to   the 10.54
graph   when   alpha   equals 2. Use the scrollbars for the
zero? slope   and   intercept   to
The graph is a straight determine   the   values   that
line.   The   forecast   is   the minimize   the   MAD.   Are
same in each period. these   the   same   values   that
regression yields?

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
30 CHAPTER 4 F O R E C A S T I N G

(d) The   three­year (c) The   banking 4.5 (c) Weighted   2   year (b) [i] Naïve

moving   average industry has a great M.A. with .6 weight for most
appears   to   give deal   of   seasonality recent year. The   coming   January   =
better in   its   processing
Year Mileage Forecast December = 23
results. requirements
1 3,000 [ii] 3­month
3,700 + 3,800
4.5   (a)    2
=3,750 miles 4,000 moving   (20 + 21
2 3 3,400 3,600 + 23)/3 = 21.33
(b) 4 3,800 3,640 [iii] 6­month
5 3,700 3,640 weighted       [(0.1  
17) + (.1  18)
Forecast   for   year   6   is   3,740     +
miles.

4.3 Year 1 2 3 4 5 6 7 8 9 10 11 Forecast

Demand 7 9.0 5.0 9.0 13.0  8.0 12.0 13.0  9.0 11.0  7.0
Naïve 7.0 9.0 5.0  9.0 13.0  8.0 12.0 13.0  9.0 11.0 7.0
Exp. Smoothing 6 6.4 7.4 6.5  7.5  9.7  9.0 10.2 11.3 10.4 10.6 9.2

Two-Year � 420 �
Year Mileage Moving Average MAD  140 �
� 3 � �
  (0.1  20)
1 3,000 + (0.2 
4.5 (d)
2 4,000 20)
3 3,400 3,500 Forecast
  
4 3,800 3,700 Year Mileage Forecast Error
 +
5 3,700 3,600 1 3,000 3,000
(0.2
2 4,000 3,000 1,000
3 3,400 3,500

4 3,800 3,450 21)
300 5 3,700 3,625 +
MAD   100. Total (0.3
3

23)]/
The forecast is 3,663 miles.
1.0
4.6
= 20.6
Y Sales X Period
[iv] Exponential
January 20 smoothing   with
February 21
alpha = 0.3
March 15
April 14 FOct  18 + 0.3(20 - 18)  18.6
May 13 FNov  18.6 + 0.3(20 - 18.6)  19.0
Naïve  tracks   the   ups June 16
July 17 FDec  19.02 + 0.3(21 - 19.02)  1
and downs best but lags
August 18 FJan  19.6 + 0.3(23 - 19.6)  20.6
the data by one period.
September 20
Exponential   smoothing October 20 10 [v] Trend 
is   probably   better November 21 11 � x  78, x  6.5,   � y = 218, y  18.
because it smoothes the December 23 12
data and does not have � xy - nx y
Sum    218 78 b
as much variation. � x 2 - nx 2
Average   18.2
TEACHING   NOTE: 1474 - (12)(6.5)(18.2) 54.4
b   0.38
Notice   how   well 650 - 12(6.5)2 143
(a)
exponential   smoothing a  y - bx
forecasts the naïve. a  18.2 - 0.38(6.5)  15.73
4.4    (a)  FJuly  FJune + 0.2(Forecasting error) Forecast   =   15.73 
                       42 + 0.2(40 �42)  41.6 + .38(1
3)   =
        (b)  FAugust  FJuly + 0.2(Forecasting error) 20.67,
                          41.6 + 0.2(45 - 41.6)  42.3 where
next
January

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 31

4.9 (d) Table for Problem 4.9(d):
 = .1  = .3  = .5
Table for Problem 4.9 (a, b, c)
Month Price per Forecast |Error| Forecast |Error| Forecast |Error|
Chip Forecast |Error|
January $1.80 $1.80 $.00
Two-Month$1.80 $.00
Three- $1.80 $.00
Two-Month Three-
February 1.67 1.80 .13  1.80 .13
Month  1.80 .13 Month
March 1.70 Price per
1.79 .09Moving 1.76 Moving
.06  1.74 Moving.04 Moving
April Month 1.85 Chip
1.78 Average
.07  1.74 Average
.11  1.72 Average
.13 Average
May January
1.90 $1.80
1.79 .11  1.77 .13  1.78 .12
June February
1.87  1.67
1.80 .07  1.81 .06  1.84 .03
July March1.80  1.70
1.80 .00 1.735  1.83 .03  1.86 .035.06
August April 1.83  1.85
1.80 .03 1.685  1.82 1.723
.01  1.83 .165.00 .127
May  1.90 1.775 1.740 .125 .160
Septembe 1.70 1.81 .11  1.82 .12  1.83 .13
June  1.87 1.875 1.817 .005 .053
r
July  1.80 1.885 1.873 .085 .073
October 1.65 1.80 .15  1.79 .14  1.76 .11
August  1.83 1.835 1.857 .005 .027
November 1.70 1.78 .08  1.75 .05  1.71 .01
September  1.70 1.815 1.833 .115 .133
December 1.75
October 1.77
 1.65 .02 1.765  1.73 .02
1.777  1.70 .115.05 .127
Totals
November  1.70 $.8 1.675 $.8
1.727 $.81
.025 .027
December  1.75 6 1.675 6
1.683 .075 .067
MAD (total/12) $.072 $.072
Totals . $.067 .793
5

 = .5 is preferable, using MAD, to  = .1 or  = .3. One could
4.10 Year 1 2 3 4 5 6 7 8 9 10 11 Forecast
Demand 4 6 4 5.0 10.0 8.0 7.0 9.0 12.0 14.0 15.0
(a) 3-year moving 4.7  5.0 6.3 7.7 8.3  8.0  9.3 11.7 13.7
(b) 3-year 4.5  5.0 7.3 7.8 8.0  8.3 10.0 12.3 14.0
weighted
forecast becomes 56.3, or (96 + 88 + 90)
56 patients. 4.8    (a)     91.3
is   the 3 results   appear   in
13th the figure below.
month.
  
(c) Only trend provides
an   equation
that   can
extend
beyond   one
month
4.7 Present = Period (week)
6.
��1 � �1 � �1 � �1 � �
a) So:  F7  ��3 �A6 + � 4 �A5 + �4 �A4 + �6 �A3 � 1.0
�� � � � � � � � �
�1 � �1 � �1 � �1 �
 � �(52) + � �(63) + � �(48) + � �(70) = 56.76 patients,
3
� � 4
� � 4
� � �6 � or 57 patients
where                                                                         
(c) MAD   (two­month
1 1 1 1 moving   average)   =   .750/10
1.0 = � weights   ,  ,  , 
3 4 4 6 MAD = 13.5/5 = 2.7
= .075
b) If the weights are 20, 15, (d) MSE   =   66.75/5   = MAD   (three­month
15,   and 10,  there  will be 13.35 moving average) = .793/9 = .
no change in the forecast (e) MAPE   =   14.94%/5 088
because   these   are   the = 2.99% Therefore,   the   two­month
same  relative  weights   as 4.9 (a, b) The computations moving   average   seems   to
in   part   (a),   i.e.,   20/60, for both the two­ have performed better.
15/60, 15/60, and 10/60. and   three­month
c) If the weights are 0.4, 0.3, averages   appear
0.2,   and   0.1,   then   the in   the   table;   the

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
32 CHAPTER 4 F O R E C A S T I N G

             (b) |Error| = |Actual – Forecast|
4.11  (a) Year 1 2 3 4 5 6 7 8 9 10 11 Forecast
Year 1 2 3 4 5 6 7 8 9 10 11 MAD
Demand 4 6.0 4.0 5.0 10.0 8.0 7.0 9.0 12.0 14.0 15.0
Exp. smoothing
Exp. Smoothing 5 4.7 5.1 14.8 1.3  4.8 1.1
6.4 0.2 6.9 5.2 6.9 1.6  7.5 0.1  8.9 2.1 10.4 4.5 11.85.1 4.6 2.4
These calculations were completed in Excel. Calculations are slightly different in Excel OM and POM for Windows due to
rounding differences.

4.12  2 50
3 52
Actual 4 56 (45 + 50 + 52)/3 = 4
t Day Demand 5 58 (50 + 52 + 56)/3 =
1 Monday 88 52.7
6 ? (52 + 56 + 58)/3 =
2 Tuesday 72 55.3
3 Wednesday 68
 = 12.3
4 Thursday 48 MAD = 6.2
5 Friday (c) Trend projection:
Ft = Ft–1 + (At–1 – Ft–1)
Let    =   .25.   Let   Monday Year Demand Trend Projection
forecast demand = 88 1 45 42.6 + 3.2  1 = 45.8
(c) The   forecasts   are
F2 = 88 + .25(88 – 88) = 88 2 50 42.6 + 3.2  2 = 49.0
about the same. 3 52 42.6 + 3.2  3 = 52.2
+ 0 = 88
4 56 42.6 + 3.2  4 = 55.4
F3 = 88 + .25(72 – 88) = 88
5 58 42.6 + 3.2  5 = 58.6
– 4 = 84 6 ? 42.6 + 3.2  6 = 61.8
F4 = 84 + .25(68 – 84) = 84
 = 3.2
– 4 = 80
MAD = 0.64
F5 = 80 + .25(48 – 80) = 80 Y  a + bX
– 8 = 72
�XY �nXY
4.13 (a) Exponential b
�X 2 �nX 2
smoothing,  = 0.6:
a  Y �bX
Exponential
Year Demand Smoothing X Y XY X2
1 45 41
2 50 1
41.0 + 0.6(45–41) = 45
43.4 45 1
3 52 2
43.4 + 0.6(50–43.4) 5047.4
= 100 4
4 56 3
47.4 + 0.6(52–47.4) 5250.2
= 156 9
5 58 4
50.2 + 0.6(56–50.2) 5653.7
= 224 16
6 ? 5
53.7 + 0.6(58–53.7) 5856.3
= 290 25

 = 25.3 Then:  X   =  15,  Y   =  261,

2
MAD = 5.06 XY = 815, X  = 55,  X = 3,
Exponential   smoothing,    = Y = 52.2 Therefore:
0.9: 815 �5 �3 �52.2
b  3.2
55 �5 �3 �3
Exponential
Year Demand Smoothing a  52.20 �3.20 �3  42.6
1 45 41 Y6  42.6 + 3.2 �6  61.8
2 50 41.0 + 0.9(45–41) =(d)
44.6
    Comparing   the
3 52 44.6 + 0.9(50–44.6 ) = 49.5
results   of   the   forecasting
4 56 49.5 + 0.9(52–49.5) = 51.8
5 58
methodologies   for   parts   (a),
51.8 + 0.9(56–51.8) = 55.6
6 ? (b), and (c).
55.6 + 0.9(58–55.6) = 57.8
Forecast Methodology MAD
 = 18.5
MAD = 3.7 Exponential smoothing,  = 0.6 5.06
Exponential smoothing,  = 0.9 3.7
(b) 3­year   moving
3-year moving average 6.2
average: Trend projection 0.64
Three-Year
Based   on   a   mean   absolute
Year Demand Moving Average
deviation  criterion,   the  trend
1 45

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 33

projection   is   to   be   preferred 4.15

over   the   exponential
Forecast Three-Year Forecast Expon
smoothing   with    =   0.6,
Year Sales Moving Average
exponential   smoothing   with Year Sales Smoothing  =
 = 0.9, or the 3­year moving
average   forecast 2005 450
2006 495 2005 450 410.0
methodologies. 2006 495 410 + 0.9(450 – 410)
2007 518
4.14 2008 563 (450 + 495 + 518)/3 = 2007 518 446 + 0.9(495 – 446)
Method 1: MAD:   (0.20   + 487.7 2008 563 490.1 + 0.9(518 –
2009 584 (495 + 518 + 563)/3 = 515.2
0.05 + 0.05 + 0.20)/4 = .125 
525.3 2009 584 515.2 + 0.9(563 –
better 558.2
2010 (518 + 563 + 584)/3 =
       MSE : (0.04 + 555.0 2010 558.2 + 0.9(584 –
0.0025 + 0.0025 + 0.04)/4 = . 581.4
021
                                    
Method 2: MAD:   (0.1   +
0.20   +   0.10  +   0.11)  /   4  =   .
4.16 (Refer   to   Solved   Problem
1275
4.1)
       MSE : (0.01 + 0.04 Year Time Period X Sales
For    =   0.3,   absolute
+ 0.01 + 0.0121) / 4 = .018   2005 1 deviations for 2005–2009 are
better 2006 2 40.0,   73.0,   74.1,   96.9,   88.8,
2007 3 respectively. So the MAD =
2008 4 372.8/5 = 74.6.
2009 5

MAD 0.3  74.6

2610
MAD 0.6  51.8
X  3,  Y  522
MAD 0.9  38.1
Y  a + bX
Because   it   gives   the   lowest
�XY - nXY 8166 - (5)(3)(522) 336 MAD,   the   smoothing
b    33.6
�X 2 - nX 2 55 - (5)(9) 10 constant   of
a  Y - bX  522 - (33.6)(3)  421.2   =   0.9   gives   the   most
accurate forecast.
y  421.2 + 33.6 x
4.18 We   need   to   find   the
y  421.2 + 33.6 �6  622.8
smoothing   constant  .   We
Year Sales Forecast Trend know   in  general   that  Ft  =
Ft–1 + (At–1 – Ft–1); t = 2,
3, 4. Choose either
2005 450 454.8
t = 3 or t = 4 (t = 2 won’t let
2006 495 488.4
2007 518 522.0 us find  because F2 = 50 =
2008 563 555.6 50 + (50 – 50) holds for any
2009 584 589.2 ). Let’s pick t = 3. Then F3
2010 622.8 = 48 = 50 + (42 – 50)
or                    48 = 50 + 42 
– 50

4.17 or                    –2 = –8
So,                 .25 = 
Forecast Exponential
Year Sales Smoothing Now we can find F5 : F5 = 
50 + (46 – 50)
2005 450 410.0
2006 495 410 + 0.6(450 – 410) = 434.0       F5 = 50 + 46 – 50
2007 518 434 + 0.6(495 – 434) = 470.6 = 50 – 4
2008 563 470.6 + 0.6(518 – 470.6) = For           = .25, F5 = 50 – 
499.0 4(.25) = 49
2009 584 499 + 0.6(563 – 499) = 537.4
2010 537.4 + 0.6(584 – 537.4) = The forecast for time period 5
565.6 = 49 units.

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
34 CHAPTER 4 F O R E C A S T I N G

4.19 Trend adjusted  4.20 Trend   adjusted

exponential smoothing:  = 
0.1,  = 0.2 Unadjusted Adjusted
Month Demand (y) Forecast Trend Forecast Error |Error| Error2
Unadjust
February 70.0 65.0 0 65.0 5.00 5.0 25.00
ed
March 68.5 65.5 0.4 65.9 2.60 2.6 6.76
Month Income Forecast Trend
April 64.8 66.16 0.61 66.77 –1.97 1.97 3.87
May 71.7 66.57 0.45 67.02 4.68 4.68 21.89
Februar 70.0 65.0 0.0 June 71.3 67.49 0.82 68.31 2.99 2.99 8.91
y July 72.8 68.61 1.06 69.68 3.12 3.12 9.76
March 68.5 65.5 0.1 Totals 419.1 20.3 76.19
April 64.8 65.9 0.16 16.42
6
May 71.7 65.92 0.13 Average   69.85 2.74 3.39 12.70
June 71.3 66.62 0.25 August Forecast 71.30 (Bias) (MAD) (MSE)
July 72.8 67.31 0.33
August 68.16 Based   upon   the   MSE   criterion,   the   exponential   smoothing   with    =   0.1,    =   0.8   is   to   be   preferred
over the exponential smoothing with    = 0.1,    = 0.2.  Its MSE of 12.70 is lower.  Its MAD of 3.39 is
also lower than that in Problem 4.19.
MAD  = 24.3/6 = 4.05, MSE FIT8  F8 + T8  27.14 + 2.45  29.59
= 113.2/6 = 18.87. Note that exponential   smoothing:    =
all numbers are rounded. 0.1,  = 0.8 F  A + ( 1 � ) ( F + T )  ( 0.2 ) ( 28)
9 8 8 8
Note:   To   use   POM   for
4.21  F5   A4 + ( 1 � ) ( F4 + T4 )  ( 0.2 ) ( 19) + ( 0.8) ( 20.14 )               + ( 0.8) ( 29
Windows   to   solve   this
problem,   a   period   0,   which            3.8 + 16.11  19.91
contains   the   initial   forecast T9   ( F9 �F8 ) + ( 1 �) T8  ( 0.4 ) ( 29.28
and   initial   trend,   must   be T5   ( F5 �F4 ) + ( 1 � ) T4  ( 0.4 ) ( 19.91 �17.82 )
              + ( 0.6 ) ( 2.4
added.            + ( 0.6 ) ( 2.32 )  0.4 ( 2.09 )
FIT9  F9 + T9  29.28 + 2.32  31.60
           + 1.39  0.84 + 1.39  2.23
FIT5  F5 + T5  19.91 + 2.23  22.14

F6  A5 + ( 1 � ) ( F5 + T5 )  ( 0.2) ( 24) + ( 0.8) ( 22.14 )

          4.8 + 17.71  22.51

T6   ( F6 �F5 ) + ( 1 � ) T5  0.4 ( 22.51 �19.91) + 0.6 ( 2.23)

         0.4 ( 2.6 ) + 1.34
          1.04 + 1.34  2.38
FIT6  F6 + T6  22.51 + 2.38  24.89

4.22   F7  A6 + (1 �)( F6 + T6 )  (0.2)(21) + (0.8)(24.89)

 4.2 + 19.91  24.11
          T7  ( F7 �F6 ) + (1 �)T6  (0.4)(24.11 �22.51)
+ (0.6)(2.38)  2.07
         FIT7  F7 + T7  24.11 + 2.07  26.18
F8  A7 + (1 � )( F7 + T7 )  (0.2)(31)
+ (0.8)(26.18)  27.14

T8   ( F8 �F7 ) + ( 1 � ) T7  0.4 ( 27.14 �24.11)

               + 0.6 ( 2.07)  2.45

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 35

Assume  Averages y = 55 �xy - n x y 650 �4(2.5)(55) 650

b  
4.23 Students   must 2
�x �nx 2
30 �4(2.5) 2 30
Appearances Demand Y
determine   the   naïve   forecast X 100
for   the   four   months.   The   20
naïve   forecast   for   March   is 3 3 5
4 6 a  y �bx
the February actual of 83, etc.
7 7
 55 �(20)(2.5)
(a) Actual Forecas |Error| 6 5
8 10 5
t
5 7
March 101 120 9 ?
The regression line is y = 5 +
20x. The forecast for May (x
April  96 114 X  =   33,  Y  =   38,  XY  = = 5) is y = 5 + 20(5) = 105.
227, X2 = 199,  X = 5.5,  Y = 4.26
May  89 110 6.33. Therefore:
A
June 108 108 227 �6 �5.5 �6.333
b  1.0286
199 �6  �5.5 �5.5 S
a  6.333 �1.0286  �5.5  .6762 Year1 Year2 Average
58 Y  .676 + 1.03 x  (rounded) Seaso Deman DemanYear1-Year2 D
MAD (for management)   14.5 n d d Demand
4 The   following   figure   shows
61.16% Fall 200 250 225.0
MAPE (for management)  both the data and the resulting
 15.29% Winter 350 300 325.0
4
equation: Spring 150 165 157.5
(b) Actual Naïve |Error| Summe 300 285 292.5
r
March 101  83 18
Average Yr1 to Yr2 � Yr1  Demand + Yr2  Demand
�
April  96 101   � �
May  89  96   Demand for season �
� 2
Sum of Ave Yr1 to Yr2  Demand
June 108  89 19 Average seasonal demand 
4
Average Yr1 to Yr2  Demand
Seasonal index = 
Average Seasonal Demand
49 New Annual Demand
ve) 
MAD (for na�  12.25 Yr3  �Seasonal index
4 4
48.49% 1200
MAPE (for na� ve)   12.12%.  �Seasonal index
4 4
 Naïve outperforms  (c) If there are nine
management. performances   by   Stone 20,000
Temple  Pilots,   the  estimated Average over all seasons:   1, 250
(c) MAD   for   the   manager’s 16
sales are:
technique   is   14.5,   while 6,000
Y9  .676 + 1.03 �9  .676 + 9.27  9.93 drums Average over spring:   1,500
MAD   for   the   naïve 4
forecast   is   only   12.25. �10 drums 1,500
Spring index:   1.2
MAPEs   are   15.29%   and (d)  R   =   .82   is   the 1, 250
12.12%,   respectively.   So correlation   coefficient,   and �5,600 �
the naïve method is better. R2  =   .68  means  68% of  the Answer:  � (1.2)  1,680 sailboats
� 4 � �
4.24 (a) Graph of demand variation   in   sales   can   be
The   observations   obviously explained   by   TV
4.27
do not form a straight line but appearances.
do   tend   to   cluster   about   a 4.25  Winter Spring Sum
straight   line   over   the   range Number of 2006 1,400 1,500 1,
shown. Accidents 2007 1,200 1,400 2,
(b) Least­squares Month (y) 2008 1,000 1,600 2,
regression: 2009 900 1,500 1,
January 30 4,500 6,000 7,
Y  a + bX February 40
�XY - nXY March 60 4.28
b April 90
�X 2 - nX 2  Totals 220
a  Y - bX Avera

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
36 CHAPTER 4 F O R E C A S T I N G

Quarte 2007 2008 2009 4.29 2011   is   25   years Then  y  =  a  +  bx, where  y  =
r beyond  1986.   Therefore,   the number sold, x = price, and
2011 quarter numbers are 101
Winter 73 65 89
through 104.
Spring 104 82 146
Summer 168 124 205 (2) (3)
Fall 74 52 98 (1) Quarter Forecast
Quarter Number (77 + .
43Q
Winter 101 120.43
Spring 102 120.86
Summer 103 121.29
Fall 104 121.72
4.30 Given Y = 36 + 4.3X
(a)            Y  =   36   +
4.3(70) = 337
(b)           Y  =   36   +
4.3(80) = 380
(c)            Y  =   36   +
4.3(90) = 423
4.31 Let x1,  x2 , K ,  x6  be the prices and y1,  y2 , K , y6
be the number sold.
4.33 (a) See the table below.
6
�xi
X  i 12,880  Average price = 3.2583
- 5(3)(180) 2,880 - 2,700                (1)
b6 
6 55 - 5(3)2 55 - 45
yi
�180
Y  i1  Average number sold = 550
18 (2)
610
6 a  180 - 3(18)  180 - 54  126
�xi yi  9,783                 (3)
i 1 y  126 + 18 x
6
2
�xi  = 67.1925                 (4)
i1

6
�xi yi - nx y
i 1 (9, 783) - 6(3.25833)(550)
b 6

2
�x i - nx
2 67.1925 - 6 (3.25833)2
i 1
-969.489
  -277.6
3.49222
a  y - bx  550 - [(-277.6)(3.25)]  1, 454.6

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 4.32 (a) x y xy x2
16 330 5,280 256
12 270 3,240 144
18 380 6,840 324
14
CHAPTER 4 F300
O R E C A S4,200
TING 196 37
60 1,280 19,560 920

So at x = 2.80, y = 1,454.6 – the   additional Therefore,   the d t

277.6($2.80) =  677.32. Now value. expected   cost   of 1 20 20 0.00
round  to  the  nearest  integer: (c) Some   other the trip is $452.50. 2 21 20 1.00
Answer: 677 lattes. quantitative (b) The 3 28 20.5 7.50
variables   would reimbursement 4 37 24.25 12.75
60 be   age   of   the request   is   much 5 25 30.63 –5.63
x  15
4 house,   number   of higher   than 6 29 27.81 1.19
1,280 bedrooms,   size   of 7 36 28.41 7.59
y  320 predicted   by   the
8 22 32.20 –10.20
4 the lot, and size of model.   This 9 25 27.11 –2.10
�xy - nx y 19,560 - 4(15)(320) the garage, etc.
360 request   should 10 28 26.05   1.95
b    18
�x 2 - nx 2 920 - 4(15)2 (d) Coefficient
20   of probably   be
determination   = questioned   by   the
a  y - bx  320 - 18(15)  50
(0.63)2  =   0.397. accountant. Cumulative error = 14.05;
Y  50 + 18 x MAD = 5 Tracking =
This   means   that (c) A number of other
only   about   39.7% 14.05/5  2.82
(b) If   the   forecast   is variables   should   be
of   the   variability included, such as: 4.38 (a) least   squares
for   20   guests,   the
in   the   sales   price 1. the   type   of equation:  Y  =   –0.158   +
bar   sales   forecast
of   a   house   is travel (air or car) 0.1308X
is   50   +   18(20)   =
explained   by   this (b) Y  =  – 0.158   +
$410.   Each   guest 2. conference fees,
regression   model 0.1308(22)   =   2.719
accounts   for   an if any
that   only   includes million
additional   $18   in
square   footage   as 3. costs   of (c) coefficient   of
bar sales.
the   explanatory entertaining customers correlation = r = 0.966
variable. 4. other coefficient   of
(b) MSE = 160/5 = 32
(c) MAPE = 13.23%/5 4.36 (a)   Given:  Y  =   90   + transportation determination =  r2
= 2.65% 48.5X1 + 0.4X2 where:  costs—cab, = 0.934
limousine, 4.39
4.34 Y  =   7.5   +   3.5X1  + Y  expected travel cost
special   tolls,   or
4.5X2 + 2.5X3 X1  number of days on the road Ye Pat X 2 Y2 X
parking ar ien Y
(a) 28 X2  distance traveled, in miles In   addition,   the   correlation X ts
(b) 43 r  0.68 (coefficient of correlation)coefficient   of   0.68   is   not Y
(c) 58
If: exceptionally   high.   It  1  36   1     
4.35 (a)  Y�  =   13,473   + indicates   that   the   model 1,29 36
Number of days on the
37.65(1860) = 83,502 explains   approximately   46% 6
road    X1  =   5   and
(b) The   predicted distance   of the overall variation in trip  2  33   4     
selling   price   is cost.   This   correlation 1,08 66
traveled  X2 = 300 9
$83,502,   but   this coefficient   would   suggest
then: that   the   model   is   not   a  3  40   9    
is   the   average
1,60 12
price for a Table for Problem 4.33 0 0
house   of  4  41  16    
this   size. Year Transistors
1,68 16
(x) (y) xy x2 126 + 18x Error Error 2 |% Error|
There   are 1 4
other   1 140  140  1 144 –4  16 100 (4/140) 
 5  40  25     =
2.86%
1,60 20
factors
 2 160  320  4 162 –2   4 100 (2/160) 
0 0 =
besides
 6  55  36 1.25%
   
square  3 190  570  9 180 10 100 1003,02
(10/190)
33 =
footage 5.26%
5 0
that   will  4 200  800 16 198  2   4 100 (2/200)
 7  60  49     =
impact 1.00%
3,60 42
the  5 210 1,050 25 216 –6  36 100 (6/210) 
0 0 =
selling  8  54  64 2.86%
   
price of a Totals 1 90 2,80 5 16 2,91 43 13.23
house.   If 5 0 0 5 0 6 2 %
x = 3 y = 180  9  58  81    
such   a
 Y = 90 + 48.5  5 + 3,36 52
house   sold   for
0.4  300 = 90 + 242.5 + 120 = particularly good one. 4 2
$95,000,   then 10  61 100    
these other factors 452.5 4.37  (a, b) 3,72 61
could   be 1 0
Perio Demand Forecas
contributing   to

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
38 CHAPTER 4 F O R E C A S T I N G

55 478 385 Year Patients Trend

X Y Forecast  8  94.8  54  8,987.0

Given: Y = a + bX where:  9 103.3  58 10,670.9

 1 36 29.8 + 3.28
 XY - nXY 33.1 10 116.2  61 13,502.4
b 2 2  2 33 29.8 + 3.28
 X - nX 36.3
Column Totals 854.0 478 76,129.
a  Y - bX  3 40 29.8 + 3.28
9
39.6
and X = 55, Y = 478, XY  4 41 29.8 + 3.28 Given: Y = a + bX where
=   2900,  X 2   =   385,  Y 2   = 42.9
 XY - nXY
23892,  5 40 29.8 + 3.28 b
X  5.5, Y  47.8,  Then: 46.2  X 2 - nX 2
 6 55 29.8 + 3.28 a  Y - bX
2900 - 10  5.5  47.8 2900 - 2629 271 49.4
b    3.28  7 60 29.8 + 3.28 and  X  =   854,  Y  =   478,
385 - 10  5.52 385 - 302.5 82.5
52.7 XY  =   42558.6,  X2  =
a  47.8 - 3.28  5.5  29.76  8 54 29.8 + 3.28 76129.9,
and Y = 29.76 + 3.28X. For: 56.1 Y 2   = 23892,   X   = 85.4,   Y
 9 58 29.8 + 3.28 = 47.8. Then:
X  11: Y  29.76 + 3.28 � 11  65.8 59.3
10 61 29.8 + 3.28 42558.6 - 10  85.4  47.8 42558.6
X  12: Y  29.76 + 3.28 � 12  69.1 b 
62.6 2 76129.
76129.9 - 10  85.4
Therefore:
1737.4
Year 11  65.8 patients   0.543
3197.3
Year 12  69.1 patients The  MAD  is   3.26—this   is a  47.8 - 0.543  85.4  1.43
The model “seems” to fit the approximately   7%   of   the
and      Y = 1.43 + 0.543X
data pretty well. One should, average   number   of   patients
and   10%   of   the   minimum For:
however, be more precise in
number   of   patients.   We   also X  131.2 :  Y  1.43 + 0.543(131.2)  72
judging   the   adequacy   of   the
model. see   absolute   deviations,   for X  90.6 :  Y  1.43 + 0.543(90.6)  50.6
Two possible approaches are years 5, 6, and 7 in the range  Therefore:
computation   of   (a)   the 5.6–7.3.   The   comparison   of
the  MAD  with   the   average Crime rate = 131.2
correlation coefficient, or (b)
and   minimum   number   of  72.7 patients
the mean absolute deviation.
patients   and   the Crime   rate   =   90.6
The correlation coefficient:
comparatively   large  50.6 patients
n  XY -  X  Y deviations   during  the  middle Note   that   rounding
r differences   occur   when
Also note that a crime rate of 131.2 is outside the range of the
years indicate that the forecast
n  X 2 - (  X ) 2  n  Y 2 - (  Y ) 2 
solving with Excel.
data set used to determine the regression equations, so caution is
model   is   not   exceptionally
   
advised.
accurate. It is more useful for 4.41 (a) It appears from the
10  2900 - 55  478 following   graph
 predicting general trends than
10  385 - 552  10  23892 - 4782  the actual number of patients that   the   points   do
   to be seen in a specific year. scatter   around   a
29000 - 26290 straight line.
 4.40
3850 - 3025  238920 - 228484
Crime Patient
2710 2710 s
   0.924 Year Rate X Y
825  10436 2934.3
2  1  58.3  36
r  0.853
The   coefficient   of  2  61.1  33
determination   of   0.853   is
quite   respectable—indicating  3  73.4  40
our   original   judgment   of   a
“good” fit was appropriate.  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

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 39

r the   data.   Moving   averages

 Y 2 - a  Y - b  XY smooth   out   seasonality.
months) (Millions) (1,000,000s Syx 
) n-2 Exponential   smoothing   can
Year (X) (Y) 71.59 - 0.511  27.1 - 0.159  352.9 forecast   January   next   year,  

 1  7 1.5 12 - 2 but   not   farther.   Because
 2  2 1.0 71.59 - 13.85 - 56.11 seasonality is strong, a naïve
  .163
 3  6 1.3 10 model that students create on
 4  4 1.5  .404 (rounded to .407 in POM for Windows software) their own might be best.
 5 14 2.5 (b) One model might be:
 6 15 2.7 Ft+1 = At–11
 7 16 2.4 (f)  The   correlation That   is   forecastnext
 8 12 2.0 coefficient and the
 9 14 2.7 period = actualone year earlier
coefficient   of to account for 
10 20 4.4 determination   are
11 15 3.4 seasonality.   But   this   ignores
given by:
12  7 1.7 the trend.
n  XY -  X One very good approach
Y
Given: Y = a + bX where: r
 n  X 2 - (  X ) 2  would   be   to 2  calculate   the
n Y 2 - ( Y ) 
 XY - nXY   increase from each month last

b
 X 2 - nX 2 year to each month this year,
12  352.9 - 132  27.1
 sum   all   12   increases,   and
a  Y - bX 12  1796 - 1322  12  71.59 - 27.12 
  divide by 12. The forecast for
 
and  X  =   132,  Y  =   27.1, next   year   would   equal   the
4234.8 - 3577.2
XY  =   352.9,  X2  =   1796,  value for the same month this
Y 2   =  71.59,   X   =  11,   Y = 21552 - 17424  859.08 - 734.41
year plus the average increase
2.26.  657.6 over   the   12   months   of   last
657.6
  year.  0.917
Then: 4128  124.67 64.25  11.166
(c) Using this model, the
352.9 - 12  11  2.26 352.9 - 298.3 54.6
b    0.159 and r 2  0.840 January forecast for next year
1796 - 12  112 1796 - 1452 344 becomes:
a  2.26 - 0.159  11  0.511 4.42 (a) This   problem
gives   students   a 148
and Y = 0.511 + 0.159X F25  17 +  17 + 12  29
chance to tackle a 12
(c) Given   a   tourist realistic   problem where   148   =   total   monthly
population   of in   business,   i.e., increases   from   last   year   to
10,000,000,   the not enough data to this year.
model   predicts   a make   a   good The forecasts for each of
ridership of: forecast. As can be the months of next year
Y = 0.511 + 0.159  seen   in   the then become:
10   =   2.101,   or   2,101,000 accompanying
figure,   the   data Jan. 29 July.
persons. Feb. 26 Aug.
contains   both
(d) If   there   are   no Mar. 32 Sep.
seasonal and trend
tourists   at   all,   the Apr. 35 Oct.
factors. May. 42 Nov.
model   predicts   a
ridership of 0.511, Jun. 50 Dec.
or   511,000 Both history and forecast for
persons.   One the next year are shown in the
would   not   place accompanying figure:
much   confidence
in   this   forecast,
however,   because
the   number   of
tourists   (zero)   is
outside   the   range
of   data   used   to
develop   the
model.
(e) The standard error
of   the   estimate   is Averaging   methods   are
given by: not   appropriate   with   trend,
seasonal, or other patterns in

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
40 CHAPTER 4 F O R E C A S T I N G

standard   error
of the estimate and
the  MAD,   the   0.2
constant   is
better.   However,
other   smoothing
constants   need   to
be examined.

4.43 (a)   and   (b)   See   the

following table:

Actual Smoothed

Week Value Value Forecast

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

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 41

4.44  XY - nXY
b
Week Actual Value Smoothed 87Trend Estimate Forecast Forecas  X 2 - nX 2
So: MAD: 
Value
 10.875 t
8 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
20 60.000 43.332  1.252 44.584  15.416
21 75.000 49.209  2.177 51.386  23.614
22 50.000 58.470  3.594 62.064 –12.064
23 40.000 58.445  2.870 61.315 –21.315
24 65.000 54.920  1.591 56.511   8.489
25 59.058  2.100 61.158

4.46 (a) X Y

To   evaluate   the   trend  421  2.90

adjusted   exponential
smoothing   model,   actual  377  2.93
week   25   calls   are   compared
 585  3.00
to   the   forecasted   value.   The
model   appears   to   be  690  3.45
producing   a   forecast
approximately   mid­range  608  3.66
between that given by simple
exponential   smoothing   using  390  2.88
  = 0.2 and    = 0.6. Trend
adjustment does not appear to  415  2.15
give  
 481  2.53
any significant improvement.
 729  3.22
n
�( At - Ft )  501  1.99
4.45 t 1
Tracking signal 
MAD
 613  2.75
Month At Ft
 709  3.90
May 100 100
June 80 104
 366  1.60
July 110 99
August 115 101
6885 36.96
Septemb 105 104 Column totals
er
October 110 104
Novembe 125 105 Given: Y = a + bX where:
r
Decembe 120 109
r

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
42 CHAPTER 4 F O R E C A S T I N G

and  X  = 6885,  Y  = 36.96, (b,   c) Amit   wants   to 4.48 (a) 27  7.70 5.497 2.20
XY  =   20299.5,  X 2   = forecast  by 28 10.10 6.818 3.28
Quarter Contracts Sales Y 29 15.20 8.787 6.41
3857893, exponential X
Y 2   =   110.26,   X   =   529.6, smoothing   (setting (Continued)
1   153  8
Y  = 2.843, Then: February’s   forecast
2   172 10
equal   to   January’s
20299.5 - 13  529.6  2.843 20299.5 - 19573.5 3   197 15
b  sales) with alpha  
4   178  9
3857893 - 13  529.62 3857893 - 3646190
0.1.   Barbara   wants
5   185 12
726 to   use   a   3­period
  0.0034 6   199 13
211703 moving average.
7   205 12
a  2.84 - 0.0034  529.6  1.03 Sales Amit 8
Barba   226 16
and Y = 1.03 + 0.0034X Totals 1,51 95
January 400 — 5
As   an   indication   of   the
February 380 400 Average 189.375    
usefulness   of   this 11.875
March 410 398
relationship, we can calculate
April 375 399.2 396.67
b  = (18384 – 8    189.375  
the correlation coefficient: May 405 396.8 388.33
11.875)/(290,413   –   8  
MAD
n  XY -  X  Y 189.375
r
 n  X 2 - (  X )  n  Y 2 - ((d)
2 2  189.375) = 0.1121
Y )
Note that Amit has more
a  =   11.875   –   0.1121  
   
forecast   observations,
189.375 = –9.3495
13  20299.5 - 6885  36.96 while Barbara’s moving
 average   does   not   start
Sales ( y) = –9.349 + 0.1121
13  3857893 - 68852  13  110.26 - 36.962  (Contracts)
  until month 4. Also note

that the MAD for Amit (b)
263893.5 - 254469.6
 is   an   average   of   4 r  (8 � 18384 - 1515 � 95) ((8 � 290,413 - 15152 )(8 � 1183 - 952 ))
50152609 - 47403225 1433.4 - 1366.0 
numbers,   while
9423.9 Barbara’s is only 2.  0.8963

2749384  67.0 Amit’s   MAD   for Sxy  1183 - (-9.3495 � 95) - (0.112 � 18384 / 6)  1.3408
9423.9 exponential   smoothing
  0.692 r 2  .8034
1658.13  8.21 (16.11)   is   lower   than
that   of   Barbara’s 4.49 (a)
r 2  0.479 moving average (19.17). Method  Exponential Smoothing
A   correlation   coefficient   of So his forecast seems to 0.6 =
0.692 is not particularly high. be better. Year Deposits Forecast
The   coefficient   of (Y )
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
variation.   Therefore,   while 4  0.26 0.241
the   model   does   provide   an 5  0.25 0.252
estimate   of   GPA,   there   is 6  0.30 0.251
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
(b) X  350: Y  1.03 + 0.0034 � 350  2.22 12  0.33 0.255
(c) 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 to 16  2.30 1.315
interpret significant digits. Also 17  2.80 1.906
note that X = 800 is outside the 18  2.80 2.442
range  of  the   data   set  used  to 19  2.70 2.656
determine   the   regression 20  3.90 2.682
relationship,   so   caution   is 21  4.90 3.413
advised. 22  5.30 4.305
23  6.20 4.90
4.47 (a) There   is  not  a 24  4.10 5.680
strong   linear   trend   in   sales 25  4.50 4.732
over time. 26  6.10 4.592

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 43

4.49 (a)  (Continued) appropriate
choice.   Measures
Method  Exponential Smoothing
0.6 = of   error   and
Year Deposits Forecast goodness­of­fit are
(Y ) really   irrelevant.
30  18.10 12.6350 Exponential
31  24.10 15.9140 smoothing
32  25.60 20.8256 provides a forecast
33  30.30 23.69 only   of   deposits
34  36.00 27.6561 for the next year—
35  31.10 32.6624 and  thus  does  not
36  31.70 31.72
Forecasting Summary Table
37  38.50 31.71 Exponential Linear Regression
38  47.90 35.784 Method used: Smoothing (Trend Analysis) Linear Regression
39  49.10 43.0536 Y = –18.968 + Y = –17.636 +
40  55.80 46.6814       1.638  YEAR       13.59364 
41  70.10 52.1526 GSP
42  70.90 62.9210 MAD  3.416   10.587   10.255
43  79.10 67.7084 MSE  34.39  171.817  204.919
44  94.00 74.5434 Standard error using  6.075   13.416   14.651
TOTALS 787.30  n – 2 in denominator
Correlation coefficient   0.846    0.813
AVERAGE    17.8932 31 31 24.10 13 1.20  0.50 –1
32 32 25.60 14 1.20  0.95 –1 address   the   five­
Next period forecast = 86.2173 15 1.20  1.70 –1
33 33 30.30 year   forecast
34 34 36.00 16 1.60  2.30  4 problem.   In   order
17 1.50  2.80  2
35 35 31.10 to   use   the
Method  Linear Regression 36 (Trend36
Analysis) 31.70
18 1.60  2.80  4
regression   model
Year Period (X ) Deposits 37 37 38.50 19 1.70  2.70  5 based   upon   GSP,
 1  1 0.25 38 38 47.90
20 1.90  3.90  8 one   must   first
 2  2 0.24 39 39 49.1021 1.90  4.90  8 develop   a   model
 3  3 0.24 40 40 55.8022 2.30  5.30 13 to   forecast   GSP,
 4  4 0.26 41 41 70.1023 2.50  6.20 16 and   then   use   the
 5  5 0.25 42 42 70.9024 2.80  4.10 20 forecast of GSP in
43 43 79.1025 2.90  4.50 21 the   model   to
 6  6 0.30
44 44 94.0026 3.40  6.10 28 forecast   deposits.
 7  7 0.31
27 3.80  7.70 34
 8  8 0.32 TOTALS 990.0 787.30 This   requires   the
28 4.10 10.10 38
 9  9 0.24 0
29 4.00 15.20 36
development   of
10 10 0.26 AVERAGE 22.50 17.893 30 4.00 18.10 36 two  models—one
11 11 0.25 31 3.90 24.10 35 of   which   (the
12 12 0.33 32 3.80 25.60 34 model   for   GSP)
33 3.80 30.30 34 must   be   based
13 13 0.50 Method  Least squares–Simple
34 Regression
3.70 36.00on 32 solely   on   time   as
14 14 0.95 35 4.10 31.10 38 the   independent
a b 36 4.10 31.70 38
15 15 1.70 variable   (time   is
–17.636 13.5936 37 4.00 38.50 36
16 16 2.30 the   only   other
Coefficients GSP Deposit 38 4.50 47.90 43
17 17 2.80 : s variable   we   are
39 4.60 49.10 44
18 18 2.80 Year (X) (Y ) Forecast
40 4.50 55.80 43
given).
19 19 2.70 (b) One could make a
 1 0.40  0.25 –12 41 4.60 70.10 44
20 20 3.90  2 0.40  0.24 –12 42 4.60 70.90 44 case   for   exclusion
21 21 4.90  3 0.50  0.24 –10 43 4.70 79.10 46 of   the   older   data.
22 22 5.30  4 0.70  0.26 –8 44 5.00 94.00 50 Were   we   to
23 23 6.20  5 0.90  0.25 –5 TOTALS exclude   data   from
24 24 4.10  6 1.00  0.30 –4 AVERAGE roughly the first 25
25 25 4.50  7 1.40  0.31  1
years, the forecasts
26 26 6.10  8 1.70  0.32  5
Given   that   one for   the   later   years
27 27 7.70  9 1.30  0.24  
0.036086 wishes  to  develop would   likely   be
28 28 10.10 a   five­year considerably   more
10 1.20  0.26 –1
29 29 15.20 forecast,   trend accurate.   Our
11 1.10  0.25 –2
30 30 18.10 12 0.90  0.33 –5 analysis   is   the argument would be

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
44 CHAPTER 4 F O R E C A S T I N G

that  a  change  that 4.51 0

caused an increase 0
in   the   rate   of Period Demand Exponentially Smoothed           
Forecast
growth   appears   to M
have taken place at 1  7 5 A
the   end   of   that 2  9 5 + 0.2 D
period.   Exclusion 3  5 5.4 + 0.2  
of   this   data, 4  9 6.12 + 0.2 =
5 13 5.90 + 0.2
however,   would  
6  8 6.52 + 0.2
not   change   our 7 Forecast 7.82 + 0.2
2
choice   of .
forecasting   model 4.52 2
because   we   still 0
Actual Forecast
need   to   forecast
deposits   for   a 95 100
future   five­year 108 110 (b) 3­month   weighted
period. 123 120 moving average
130 130

ADDITIONAL
HOMEWORK MAD =  10/4 = 2.5,  MSE =
38/4 = 9.5
PROBLEMS
These problems, which appear 4.53 (a) 3­month   moving
on   www.myomlab.com, average:
provide   an   additional   13
problems that you may wish to Three-Month
assign.
Month Sales Moving Average
4.50
Week 1 2 3 4 5 January
6 7 8 11 9 10 Forecast
February 14
Registration 22 21 25 27 35 March
29 33 3716 41 37
(a) Naïve 22 21 25 27 April
35 29 3310 37 (1141+ 14 37
+ 16)/3 =
(b) 2-week moving 21.5 23 26 31 32 31 35 39 39
(c) 4-week moving 23.7 May
27 29 3115 33.5(14
35+ 16 37
+ 10)/3 =
5
June 17 (16 + 10 + 15)/3 =

July 11 (10 + 15 + 17)/3 =

August 14 (15 + 17 + 11)/3 =

September 17 (17 + 11 + 14)/3 =

October 12 (11 + 14 + 17)/3 =

November 14 (14 + 17 + 12)/3 =

December 16 (17 + 12 + 14)/3 =

January 11 (12 + 14 + 16)/3 =

February (14 + 16 + 11)/3 =


 
=
 
2
2
.

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 45

(c) Based   on   a Sales
Month   Mean  9.17 Average4.56 To Absolute
Three-Month yMoving   compute (b)   Correlation
Absolute x  3.5
Moving seasonalized or adjusted sales
Deviation coefficient:
Deviation forecast, we just multiply each n  XY -  X  Y
January 11 y  5.27 + 1.11x
criterion,   14the seasonalized   index   by   the r
February n  X 2 - (  X ) 2  n  Y 2
3­month   moving Period   7   forecast   = appropriate trend forecast.
March 16   
average   10 with 13.07
April (1  11 + 2  14 + 3  16)/6 = 14.50 Y� 4.50 Y�
 Index 5  70 - 15  20
MAD  =   2.2   is15  to Period
(1  14 + 2  16  12+  forecast
3  10)/6   = = 12.67Seasonal Trend forecast
May 2.33 
June be   preferred   17
over 18.64,   but   this   is   far
(1  16 + 2  10 + 3  15)/6 = 13.50 Hence, for 3.50 5  55 - 152  5  130 - 2
July the   3­month 11 outside
(1  10 + 2  15   the
+3   range
 17)/6  of = 15.17Quarter I: Y�4.17  
I  1.25  120,000  150,000
August weighted   moving 14 valid data.
(1  15 + 2  17 + 3  11)/6 = 13.67 0.33 350 - 300
�  
Septemberaverage   17 with (1  17 + 2  11 + 3  14)/6 = 13.50 II  0.90  140,000  126,000
Quarter II: Y3.50
October MAD = 2.72. 12 (1  11 + 2  14 + 3  17)/6 = 15.00 3.00
� 275 - 225  650 - 400 
Quarter III: Y  0.75  160,000  120,000
III
November 14 (1  14 + 2  17 + 3  12)/6 = 14.00 0.00 50
(d) Other   factors 16
  that �   0.45
December (1  17 + 2  12 + 3  14)/6 = 13.83 IV  1.10  180,000  198,000
Quarter IV: Y2.17
111.80
January might be included11 (1  12 + 2  14 + 3  16)/6 = 14.67 3.67
February in   a   more (1  14 + 2  16 + 3 
4.57
11)/6 = 13.17
complex   model Mon Tue Wed Thu Fri Sat
       = 27.17
are   interest   rates Week 1 210 178 250 215 160 180
and   cycle   or MAD = 2.72
Week 2 215 180 250 213 165 185
seasonal factors. Week 3 220 176 260 220 175 190
4.54 (a) Week 4 225 178 260 225 176 190
Averages 217. 17 255 218. 16 186.3 Overall average =
Actua Cumulativ 5 8 3 9 204
l
Wee Miles Foreca Error Error (a) Seasonal indices: The   correlation   coefficient
k st indicates   that   there   is   a
1.066 (Mon) 0.873 (Tue) 
1 17 17.00 0.00 positive
1.25 (Wed) 
2 21 17.00 –4.00 correlation   between   bank
3 19 17.80 –1.20 1.07 (Thu) 0.828 (Fri) 
0.913 (Sat) deposits and consumer price
4 23 18.04 –4.96 –
indices in
5 18 19.03 +1.0 (b) To   calculate   for
3 Monday   of   Week
6 16 18.83 +2.8 5   =   201.74   +
3
0.18(25)  1.066 =
7 20 18.26 –1.74
219.9   rounded   to
8 18 18.61 +0.6
1 220
9 22 18.49 –3.51 – Forecast 220 (Mon)
10 20 19.19 –0.81 – 180 (Tue) 258 (Wed)
11 15 19.35 +4.3 221 (Thu) 171
5 (Fri) 189 (Sat)
12 22 18.48 –3.52 –
4.58 (a) 4000   +
(b) The  MAD  = 0.20(15,000) = 7,000
28.56/12 = 2.38
(b) 4000   +
(c) The  cumulative 0.20(25,000) = 9,000
error  and tracking
signals  appear  to   4.59 (a) 35   +   20(80)   +
be   consistently 50(3.0) = 1,785
negative,   and   at (b) 35   +   20(70)   +
week   10,   the 50(2.5) = 1,560
tracking   signal 4.60 Given: X = 15, Y =
exceeds 5 MADs. 2 2
20, XY = 70, X  = 55, Y
Birmingham,   Alabama—i.e.,
= 130,  X  = 3,  Y =  4
4.55 y x x2 as   one   variable   tends   to
(a)  XY - nXY increase   (or   decrease),   the
 7  1  1    b 
 9  2  4  X 2 - nX 2 other   tends   to   increase   (or
 5  3  9 a  Y - bX decrease).
11  4 16
70 - 5  3  4 70 4.60 (c) Standard
- 60 10   error   of
10  5 25 b   
the estimate: 1
13  6 36 55 - 5  32 55 - 45 10
55 21 91 a  4 - 1 3  4 - 3  1
Y  1 + 1X

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
46 CHAPTER 4 F O R E C A S T I N G

Standard   error   of   the Selection of an appropriate

 Y 2 - a  Y - b  XY

130 - 1  20 - 1  70
Syx   estimate: time   series   forecasting
n-2 3 model based upon a plot of


130 - 20 - 70

40
 13.3  3.65  Y 2 - a  Y - b  XY 94 the data.
- 1  20 - 1  70
Syx  
3 3 n-2 The 5  - 2importance   of

combining   a   qualitative
4.61 94 - 20 - 70
  1.333  1.15 model   with   a   quantitative
X Y 3 model   in   situations   where
2 4 technological   change   is
1 1 4.62 Using software, the  occurring.
4 4
5 6 regression equation is:  1. A   plot   of   the   data
3 5 Games lost = 6.41 + 0.533   indicates a linear trend (least
Column Totals 15 20 days rain. squares)   model   might   be
appropriate   for   forecasting.
Given: Y = a + bX where: Using linear trend you obtain
CASE STUDIES the following:
 XY - nXY
b
 X 2 - nX 2 1 SOUTHWESTERN x y x2 xy
a  Y - bX UNIVERSITY: B  1 480     1    48
This is the second in a series  2 436     4    87
and X = 15, Y = 20, XY = of integrated case studies that  3 482     9  1,44
2 2
70, X  = 55, Y  = 94,  X  = run throughout the text.  4 448    16  1,79
3,  5 458    25  2,29
1. One   way   to   address   the
Y  = 4. Then:  6 489    36  2,93
case   is   with   separate  7 498    49  3,48
70 - 5  4  3 70 - 60 forecasting   models   for   each
b   1.0  8 430    64  3,44
55 - 5  32 55 - 45 game.   Clearly,   the  9 444    81  3,99
a  4 - 1  3  1.0 homecoming   game   (week   2) 10 496   100  4,96
and   the   fourth   game   (craft 11 487   121  5,35
and  Y  =   1.0   +   1.0X.   The festival) are unique attendance 12 525   144  6,30
correlation coefficient: situations. 13 575   169  7,47
14 527   196  7,37
n  XY -  X  Y 15 540   225  8,10
r Game Model 16 502   256  8,03
 n  X 2 - (  X ) 2  n  Y 2 - (  Y ) 2  17 508   289  8,63
     1 y = 30,713 + 2,534x
18 573   324 10,31
 2 y = 37,640 + 2,146x
5  70 - 15  20 350 - 300  3 y = 36,940 + 1,560x 19 508   361  9,65
 
20 498   400  9,96
5  55 - 152  5  94 - 202  275 - 225 470 - 400   4 y = 22,567 + 2,143x
21 485   441 10,18
    5 y = 30,440 + 3,146x
Total 22 526   484 11,57
50 50
   0.845 23 552   529 12,69
50  70 59.16 24 587   576 14,08
(where  y  = attendance and  x 25 608   625 15,20
= time) 26 597   676 15,52
27 612   729 16,52
2. Revenue   in   2010   =
28 603   784 16,88
(239,000)   ($50/ticket)   = 29 628   841 18,21
$11,950,000 30 605   900 18,15
Revenue   in   2011   = 31 627   961 19,43
(250,530)   ($52.50/ticket) 32 578 1,024 18,49
= $13,152,825 33 585 1,089 19,30
34 581 1,156 19,75
3. In   games   2   and   5,   the 35 632 1,225 22,12
forecast   for   2011   exceeds 36 656 1,296 23,61
stadium   capacity.   With   this Totals 666 19,366 16,206 378,6
appearing to be a continuing
trend, the time has come for a Average 18.5 537.9 450.2 10,5
new or expanded stadium.

2 DIGITAL CELL 
PHONE, INC.
Objectives:

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
 CHAPTER 4 F O R E C A S T I N G 47

�xy - nx y 378,661 - 36 �18.5Analysis for Management,

�537.9 20390.0 years   ago.   This   system 2003 51.80 43.4
b    5.25
 

2
�x - nx 2
16,206 - (36 �18.52 ) ed.,   Prentice
10th  3885.0   Hall actually   protects   managers 2004 54.92 38.5
Publishing. from   large   sales   variations 2005 69.70 51.4
a  y - bx  537.9 - 5.25 �18.5  440.85 2006 68.90 58.7
Cases   =   432.28   +   5.73 outside   their   control.   One
n �xy - �x �y (time),  r2  =   .93,   MAD   = could   also   justify   a   50%– 2007 63.72 45.4
r 2008 84.73 50.2
[ n �x 2 - ( �x )2 ][ n �y 2 - ( �y)212.84
] 30%–20%   model   or   some
2009 78.74 79.6
other variation.
(36)(378,661) - c) Additive seasonal model,
(666)(19,366)
 4. Other   predictors   of   cafe
2 Cases   =   444.29   +   5.06
[(36) �(16,206) - (666) ][(36)(10,558,246) - (19,366)2 ] sales could include season of
(time),  r2  =   .86,   MAD   = Southeast Airline D
13,631,796 - 12,897,756
20.02 year   (weather);   hotel
 Airframe Cost Engine
[(583,416) - (443,556)][380,096,856) - (375,041,956)] occupancy; spring break from Year per Aircraft per Airc
d)   Additive   seasonal   model,
colleges;   beef   prices;
737,040 with
734,040  centered   moving 2003 13.29 18.8
  promotional budget; etc.
[139,860][5,054,900] averages.
706,978,314,000 2004 25.15 31.5
Cases   =   431.31   +   5.72 5. Y  a + bx 2005 32.18 40.4
734,040 (time),  r2  =   .94,   MAD   = 2006 31.78 22.1
  .873 Month Advertising Guest Count
840,820 12.28 X Y
2007 25.34 19.6
2008 32.78 32.5
r 2  .76 The two methods that use  1 14 21 2009 35.56 38.0
y  440.85 + 5.25 (time) the   average   of   all   data   have  2 17 24
   r = 0.873 indicating a 
   very   similar   results,   and   the  3 25 27 Utilizing   the   software
reasonably good fit two CMA  methods also  look  4 25 32 package   provided   with   this
quite close. As suggested with  5 35 29 text,   we   can   develop   the
The   student   should this analysis, CMA is typically following   regression
report the linear trend results, the better technique.  6 35 37
equations for the variables of
but   deflate   the   forecast interest:
 7 45 43
obtained   based   upon VIDEO CASE Northern   Airlines—
qualitative information about
industry   and   technology
STUDY  8 50 Airframe Maintenance Cost:
43
 Cost = 36.10 + 0.0026
trends. FORECASTING AT   9 60 54
Because there is limited  Airframe age
HARD ROCK CAFE  Coefficient   of
seasonality   in   the   data,   the 10 60 66
linear   trend   analysis   above There   is   a   short   video   (8 determination = 0.7695
Totals 36 37
provides a good r2 of .76. minutes)   available   from  Coefficient   of
6 6
However, a more precise Prentice   Hall   and   filmed Averag 36.6   37.6 correlation = 0.8772
forecast   can   be   developed specifically  for  this  text  that e Northern   Airlines—Engine
addressing   the   seasonality supplements this case.  Maintenance Cost:
issue,   which   is   done   below. 1. Hard   Rock   uses  Cost = 20.57 + 0.0026

Methods a and c yield  r2  of . forecasting for (1) sales (guest  Airframe age

85 and .86, respectively, and
methods b and d, which also
center   the   seasonal
adjustment, yield r2 of .93 and
.94, respectively.
2. Four   approaches   to
decomposition of The Digital
Cell  Phone  data  can  address
seasonality, as follows:
a)   Multiplicative   seasonal
model,
Cases   =   443.87   +   5.08
(time),  r2  =   .85,   MAD   =
20.89
b)   Multiplicative   Seasonal
Model,   with   centered
moving   averages  (CMA),
which   is   not   covered   in
our text but can be seen in
Render,  Stair,   and
Hanna’s  Quantitative
15,772 - 10 �36.6 �37.6
counts)   at   cafes,   (2)   retail b  0.7996 Coefficient
�.8
   of
sales,   (3)   banquet   sales,   (4) 15,910 - 10 �36.62 determination = 0.6124
concert   sales,   (5)   evaluating a  37.6 - 0.7996 �36.6  8.3363 �8.3  Coefficient   of
managers,   and   (6)   menu Y  8.3363 + 0.7996 X correlation = 0.7825
planning.   They   could   also At $65,000; y  8.3 + .8 (65) Southeast   Airlines—
employ   these   techniques   to   8.3 +  52 = 60.3, or 60,300 Airframe Maintenance Cost:
forecast:   retail   store   sales   of guests.  Cost =  4.60 + 0.0032
individual   (SKU)   product  Airframe age
For   the   instructor   who   asks
demands; sales of each entrée;  Coefficient   of
other questions than this one:
sales at each work station, etc. determination = 0.3905
         r2  0.8869
2. The POS system captures Std. error  5.062  Coefficient   of
all the basic sales data needed correlation = 0.6249
to   drive   individual   cafe’s Southeast   Airlines—Engine
scheduling/ordering. It also is ADDITIONAL CASE Maintenance Cost;
aggregated   at   corporate   HQ. STUDY*  Cost = –0.67 + 0.0041
Each   entrée   sold   is   counted  Airframe age
as one guest at a Hard Rock THE NORTH­SOUTH   Coefficient   of
Cafe. AIRLINE determination = 0.4600
3. The weighting system is  Coefficient   of
Northern Airline Data
subjective, but is reasonable. correlation = 0.6782
More weight is given to each Airframe Cost
of the past 2 years than to 3 Year per Aircraft

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.
48 CHAPTER 4 F O R E C A S T I N G

The   following   graphs for  Northern  Airlines.  Southeast   Airlines: management

portray   both   the   actual   data There   is   certainly The   relationships practices.
and   the   regression   lines   for reason   to   conclude, between mainte­nance  The   difference
airframe   and   engine however, that airframe costs and airframe age between   maintenance
maintenance   costs   for   both age   is   not   the   only for Southeast Airlines procedures of the two
airlines. important factor. are   much   less   well airlines   should   be
defined.   It   is   even investigated.
more   obvious   that  The   data   with   which
airframe age is not the she   is   currently
*This case study appears on our only   important   factor working   does   not
companion Web site, at  —perhaps   not   even provide   conclusive
www.pearsonhighered.com/heize the   most   important results.
r, and at www.myomlab.com. factor. Concluding Comment:
Overall, it would seem that: The   question   always   arises,
 Northern   Airlines   has with this case, as to whether
the   smallest   variance the data should be merged for
in   mainte­nance   costs the two  airlines,   resulting in
—indicating   that   its two   regressions   instead   of
day­to­day four. The solution provided is
management   that   of   the   consultant   who
of   maintenance   is was hired to analyze the data.
working pretty well. The   airline’s   own   internal
analysts   also   conducted
 Maintenance   costs regressions,   but  did  merge
seem   to   be   more   a the data sets. This shows how
function   of   airline statisticians can take different
than of airframe age. views of the same data.
 The   airframe   and
engine   maintenance
costs   for   Southeast
Airlines   are   not   only
lower, but more nearly
similar than  those for
Northern   Airlines.
From   the   graphs,   at
least,   they   appear   to
be rising more sharply
with age.
 From   an   overall
perspective,   it   appears
that  Southeast   Airlines
may   perform   more
efficiently   on   sporadic
or   emergency 
repairs,   and   Northern
Airlines   may   place
more   emphasis   on
Note that the two graphs preventive
have been drawn to the same maintenance.
scale   to   facilitate Ms.   Young’s   report   should
comparisons between the two conclude that:
airlines.
 There   is   evidence   to
Comparison: suggest   that
 Northern   Airlines: maintenance   costs
There   seem   to   be could be made to be a
modest   correlations function   of   airframe
between   maintenance age  by   implementing
costs and airframe age more   effective

Copyright © 2011 Pearson Education, Inc. publishing as Prentice Hall.