REVISED

M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 52

5
C H A P T E R

Forecasting Models

TEACHING SUGGESTIONS Actual Three-Week

Teaching Suggestion 5.1: Wide Use of Forecasting. Week Bicycle Sales Moving Average
Forecasting is one of the most important tools a student can master 1 8
because every firm needs to conduct forecasts. Its useful to moti- 2 10
vate students with the idea that obscure sounding techniques such 3 9
as exponential smoothing are actually widely used in business, and 4 11 (8  10  9)/3  9
a good manager is expected to understand forecasting. Regression 5 10 (10  9  11)/3  10
6 13 (9  11  10)/3  10
is commonly accepted as a tool in economic and legal cases.
7 (11  10  13)/3  11Z\c
Teaching Suggestion 5.2: Forecasting as an Art and a Science.
Forecasting is as much an art as a science. Students should under- Alternative Example 5.2: Weighted moving average
stand that qualitative analysis (judgmental modeling) plays an im-
(weight for period n)(demand in period n))
portant role in predicting the future since not every factor can be
quantified. Sometimes the best forecast is done by seat-of-the- weights
pants methods. Bowers Bikes decides to forecast bicycle sales by weighting the
past 3 weeks as follows:
Teaching Suggestion 5.3: Use of Simple Models.
Many managers want to know what goes on behind the forecast. Weights Applied Period
They may feel uncomfortable with complex statistical models with
too many variables. They also need to feel a part of the process. 3 Last week
2 Two weeks ago
Teaching Suggestion 5.4: Management Input to the Exponential 1 Three weeks ago
Smoothing Model. 6 Sum of weights
One of the strengths of exponential smoothing is that it allows de-
cision makers to input constants that give weight to recent data. A 3-week weighted moving average appears below.
Most managers want to feel a part of the modeling process and
appreciate the opportunity to provide input. Actual
Bicycle
Teaching Suggestion 5.5: Wide Use of Adaptive Models. Week Sales Three-Week Moving Average
With todays dominant use of computers in forecasting, it is
possible for a program to constantly track the accuracy of a 1 8
2 10
models forecast. Its important to understand that a program
3 9
can automatically select the best alpha and beta weights in
4 11 [(3  9)  (2  10)  (1  8)]/6  9Z\n
exponential smoothing. Even if a firm has 10,000 products, the 5 10 [(3  11)  (2  9)  (1  10)]/6  10Z\n
constants can be selected very quickly and easily without human 6 13 [(3  10)  (2  11)  (1  9)]/6  10Z\n
intervention. 7 [(3  13)  (2  10)  (1  11)]/6  11X\c

Alternative Example 5.3: A firm uses simple exponential

ALTERNATIVE EXAMPLES smoothing with a  0.1 to forecast demand. The forecast for the
Alternative Example 5.1: week of January 1 was 500 units, whereas actual demand turned
demand in previous n periods out to be 450 units. The demand forecasted for the week of Janu-
Moving average = ary 8 is calculated as follows.
n
Bicycle sales at Bowers Bikes are shown in the middle column of the Ft1 Ft (At Ft)
following table. A 3-week moving average appears on the right. 500 0.1(450 500) 495 units

52
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 53

CHAPTER 5 FORECASTING MODELS 53

Alternative Example 5.4: Exponential smoothing is used to

forecast automobile battery sales. Two values of  are examined,
  0.8 and   0.5. To evaluate the accuracy of each smoothing
constant, we can compute the absolute deviations and MADs.
Assume that the forecast for January was 22 batteries.

Absolute Absolute
Actual Forecast Deviation Forecast Deviation
Battery with with with with
Month Sales  0.8  0.8  0.5  0.5
January 20 22 2 22 2
February 21 20.40 0.6 21 0
March 15 20.880 5.88 21 6
April 14 16.176 2.176 18 4
May 13 14.435 1.435 16 3
June 16 13.287 2.713 14.5 31.5
Sum of absolute deviations: 15 16.5
MAD: 2.46 2.75

On the basis of this analysis, a smoothing constant of   0.8 is Alternative Example 5.6: The rated power capacity (in hours/
preferred to   0.5 because it has a smaller MAD. week) over the past 6 years has been:
Alternative Example 5.5: Use the sales data given below to de-
termine: (a) the least squares trend line, (b) the predicted value for Rated Capacity
Year (hrs/wk)
2000 sales.
1 115
Year Sales (Units) 2 120
3 118
1993 100 4 124
1994 110 5 123
1995 122 6 130
1996 130
1997 139 Here is an alternative way to recode years which simplifies the
1998 152
math since X  0.
1999 164

Renumbered Capacity
To minimize computations, transform the value of x (time) to sim- Year Year (x) (y) x2 xy
pler numbers. In this case, designate 1993 as year 1, 1994 as year
2, and so on. 1 2.5 115 6.25 287.5
2 1.5 120 2.25 180
3 .5 118 0.25 59
Time Sales
4 .5 124 0.25 62
Year Period (Units) x2 xy
5 1.5 123 2.25 184.5
1993 1 100 1 100 6 2.5 130 6.25 325
1994 2 110 4 220 X  0 Y  730 X2  17.5 XY  45
1995 3 122 9 366
1996 4 130 16 520 XY 45
b= = = 2.57
1997 5 139 25 695 X2 17.5
1998 6 152 36 912
1999 17 164 149 1,148 Y 730
a= = = 121.67
x  28 y  917 x 2  140 xy  3,961 n 6
y 121.67 2.57X
x 28 y 917
x= = =4 y= = = 131 Year 7 121.67 (2.57)(3.5)
n 7 n 7
xy nxy 3, 961 (7)( 4 )(131) 293 131
b= = = = 10.464
x 2 nx 2 140 (7)( 4 2 ) 28 Alternative Example 5.7: The forecast demand and actual de-
mand for 10-foot fishing boats are shown below. We compute the
a = y bx = 131 10.46( 4 ) = 89.14
tracking signal and MAD.
Therefore, the least squares trend equation is, Forecast errors 70
MAD = = = 11.7
y = a + bx = 89.14 + 10.464 x n 6
To project demand in 2000, we denote the year 2000 as x  8, RSFE 24
Tracking Signal = = = 2.1 MADs
MAD 11.7
Sales in 2000  89.14  10.464(8)  172.85
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 54

54 CHAPTER 5 FORECASTING MODELS

Table for Alternate Example 5.7

Forecast Actual Forecast Cumulative Tracking
Year Demand Demand Error RSFE Error Error MAD Signal
1 78 71 7 7 7 7 7.0 1.0
2 75 80 5 2 5 12 6.0 0.3
3 83 101 18 16 18 30 10.0 1.6
4 84 84 0 16 0 30 7.5 2.1
5 88 60 28 12 28 58 11.6 1.0
6 85 73 12 24 12 70 11.7 2.1

SOLUTIONS TO DISCUSSION QUESTIONS MAD is important because it can be used to help increase forecast-
AND PROBLEMS ing accuracy.
5-1. The steps that are used to develop any forecasting system 5-9. If a seasonal index equals 1, that season is just an average
are: season. If the index is less than 1, that season tends to be lower
than average. If the index is greater than 1, that season tends to be
1. Determine the use of the forecast.
higher than average.
2. Select the items or quantities that are to be forecasted.
5-10. If the smoothing constant equals 0, then
3. Determine the time horizon of the forecast.
Ft1  Ft  0(At  Ft)  Ft
4. Select the forecasting model.
This means that the forecast never changes.
5. Gather the necessary data. If the smoothing constant equals 1, then
6. Validate the forecasting model. Ft1  Ft  1(At  Ft)  At
7. Make the forecast. This means that the forecast is always equal to the actual value in
8. Implement the results. the prior period.
5-2. A time-series forecasting model uses historical data to pre- 5-11. A centered moving average (CMA) should be used if
dict future trends. trend is present in data. If an overall average is used rather than a
5-3. The only difference between causal models and time- CMA, variations due to trend will be interpreted as variations due to
series models is that causal models take into account any factors seasonal factors. Thus, the seasonal indices will not be accurate.
that may influence the quantity being forecasted. Causal models 5-12.
use historical data as well. Time-series models use only historical Actual
data. Month Shed Sales Four-Month Moving Average
5-4. Qualitative models incorporate subjective factors into the Jan. 10
forecasting model. Judgmental models are useful when subjective Feb. 12
factors are important. When quantitative data are difficult to ob- Mar. 13
tain, qualitative models are appropriate. Apr. 16
5-5. The disadvantages of the moving average forecasting May 19 (10  12  13  16)/4  51/4  12.75
June 23 (12  13  16  19)/4  60/4  15
model are that the averages always stay within past levels, and the
July 26 (13  16  19  23)/4  70/4  17.75
moving averages do not consider seasonal variations. Aug. 30 (16  19  23  26)/4  84/4  21
5-6. When the smoothing value, , is high, more weight is given Sept. 28 (19  23  26  30)/4  98/4  24.5
to recent data. When  is low, more weight is given to past data. Oct. 18 (23  26  30  28)/4  107/4  26.75
Nov. 16 (26  30  28  18)/4  102/4  25.5
5-7. The Delphi technique involves analyzing the predictions
Dec. 14 (30  28  18  16)/4  92/4  23
that a group of experts have made, then allowing the experts to re-
view the data again. This process may be repeated several times.
After the final analysis, the forecast is developed. The group of The MAD  7.78
experts may be geographically dispersed. See solution to 5-13 for calculations.
5-8. MAD is a technique for determining the accuracy of a
forecasting model by taking the average of the absolute deviations.
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 55

CHAPTER 5 FORECASTING MODELS 55

5-13.
Three- Four-
Three- Month Four- Month
Actual Month Absolute Month Absolute
Month Shed Sales Forecast Deviation Forecast Deviation
Jan. 10
Feb. 12
Mar. 13
Apr. 16 11.66 4.34
May 19 13.66 5.34 12.75 6.25
June 23 16 7 15 8
July 26 19.33 6.67 17.75 8.25
Aug. 30 22.66 7.34 21 9
Sept. 28 26.33 1.67 24.5 3.5
Oct. 18 28 10 26.75 8.75
Nov. 16 25.33 9.33 25.5 9.5
Dec. 14 20.66 56.66 23 69.25
58.35 62.25

58.35 The 3-month moving average appears to be more accurate. How-

Three-month MAD = = 6.48 ever, if weighted moving averages had been used, the results
9
might be different.
62.25
Four-month MAD = = 7.78
8

5-14.
Three-Year Weighted Three-Year Three-Year Three-Year Weighted
Year Demand Moving Averages Moving Averages Absolute Deviation Absolute Deviation
1 4
2 6
3 4 sum of the weights
a

4 5 (4  6  4)/3  423 [(2  4)  6  4]/4  412 0.34 0.55

5 10 (6  4  5)/3 5 [(2  5)  4  6]/4  50 5.55 5.55
6 8 (4  5  10)/3  613 [(2  10)  5  4]/4  714 1.67 0.75
7 7 (5  10  8)/3  723 [(2  8)  10  5]/4  734 0.67 0.75
8 9 (10  8  7)/3  813 [(2  7)  8  10]/4  80 0.67 1.55
9 12 (8  7  9)/3 8 [(2  9)  7  8]/4  814 4.55 3.75
10 14 (7  9  12)/3  913 [(2  12)  9  7]/4  10 4.67 4.55
11 15 (9  12  14)/3  1123 [(2  14)  12  9]/4  1214 3.34 2.75
Total absolute deviations: 20.36 18.5

MAD for 3-year average  2.54

MAD for weighted 3-year average  2.32
The weighted moving average appears to be slightly more accurate
in its annual forecasts.
5-15. Using Excel or QM for Windows, the trend line is
Y  2.22  1.05X
Where X  time period (1, 2, . . .) Y  demand
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 56

56 CHAPTER 5 FORECASTING MODELS

5-16. Using the forecasts in the previous problems we obtain the 5-17.   0.3. New forecast for year 2 is last periods forecast 
absolute deviations given in the table below. (last periods actual demand  last periods forecast):
new forecast for year 2 5,000 (0.3)(4,000 5,000)
3-Yr MA 3-Yr Wt. MA Trend line
Year Demand |deviation| |deviation| |deviation| 5,000 (0.3)( 1,000)
11 14 0.73 5,000 300
12 16 1.67 4,700
13 14 1.38 The calculations are:
14 15 0.33 0.50 1.44
15 10 5.00 5.00 2.51
Year Demand New Forecast
16 18 1.67 0.75 0.55
17 17 0.67 0.75 2.60 2 6,000 4,700  5,000  (0.3)(4,000  5,000)
18 19 0.67 1.00 1.65 3 4,000 5,090  4,700  (0.3)(6,000  4,700)
19 12 4.00 3.75 0.29 4 5,000 4,763  5,090  (0.3)(4,000  5,090)
10 14 4.67 4.00 1.24 5 10,000 4,834  4,763  (0.3)(5,000  4,763)
11 15 3.33 2.75 1.18 6 8,000 6,384  4,834  (0.3)(10,000  4,834)
Total absolute 7 7,000 6,869  6,384  (0.3)(8,000  6,384)
deviations 20.33 18.50 15.24 8 9,000 6,908  6,869  (0.3)(7,000  6,869)
9 12,000 7,536  6,908  (0.3)(9,000  6,908)
MAD (3-year moving average)  2.54 10 14,000 8,875  7,536  (0.3)(12,000  7,536)
MAD (3-year weighted moving average)  2.31 11 15,000 10,412  8,875  (0.3)(14,000  8,875)
MAD (trend line)  1.39
The trend line is best because the MAD is lowest. The mean absolute deviation (MAD) can be used to determine
which forecasting method is more accurate.

Weighted
Moving Absolute Absolute
Year Demand Average Deviation Exp. Sm. Deviation
1 4,000 5,000 1,000
2 6,000 4,700 1,300
3 4,000 5,090 1,090
4 5,000 4,500 500 4,763 237
5 10,000 5,000 5,000 4,834 5,166
6 8,000 7,250 750 6,384 1,616
7 7,000 7,750 750 6,869 131
8 9,000 8,000 1,000 6,908 2,092
9 12,000 8,250 3,750 7,536 4,464
10 14,000 10,000 4,000 8,875 5,125
11 15,000 12,250 12,750 10,412 14,588
Total: 18,500 26,808
Mean: 2,312.5 2,437

Thus, the 3-year weighted moving average model appears to be

more accurate.

Year 1 2 3 4 5 6
5-18.
Forecast 410.0 422.0 443.9 466.1 495.2 521.8

5-19.

Year Sales Forecast Using  0.6 Forecast Using  0.9

1 450
2 495 410  (0.6) (450  410)  434 410  (0.9)(450  410)  446
3 518 434  (0.6) (495  434)  470.6 446  (0.9)(495  446)  490.1
4 563 470.6  (0.6)(518  470.6)  499.0 490.1  (0.9)(518  490.1)  515.21
5 584 499  (0.6) (563  499)  537.4 515.21  (0.9)(563  515.21)  558.2
6 ? 537.4  (0.6)(584  537)  565.6 558.221  (0.9)(584  558.2)  581.4
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 57

CHAPTER 5 FORECASTING MODELS 57

5-20.

Actual  0.3 Absolute  0.6 Absolute  0.9 Absolute

Year Sales Forecast Deviation Forecast Deviation Forecast Deviation
1 450 410.0 40.0 410.0 40.0 410.0 40.0
2 495 422.0 73.0 434.0 61.0 446.0 49.0
3 518 443.9 74.1 470.6 47.4 490.1 27.9
4 563 466.1 96.9 499.0 64.0 515.2 47.8
5 584 495.2 88.8 537.4 46.6 558.2 25.8
6 ? 521.8 565.8 581.4
Total absolute deviation 372.8 259.0 190.5

MAD0.3 372.8/5 74.56

MAD0.6 259/5 51.8
MAD0.9 190.5/5 38.1
Because it has the lowest MAD, the smoothing constant   0.9
gives the most accurate forecast.
5-21.

Year Sales Three-Year Moving Average

1 450
2 495
3 518
4 563 (450  495  518)/3  487.667
5 584 (495  518  563)/3  525.333
6 ? (518  563  584)/3  555

5-22.
Time
Period Sales
Year X Y X2 XY
1 1 450 1 450
2 2 495 4 990
3 3 518 9 1554
4 4 563 16 2252
5 5 2,584 125 2920
2,610 55 8166

b 33.6
a 421.2
Y 421.2 33.6X
Projected sales in year 6,
Y 421.2 (33.6)(6)
622.8
5-23.
Three-Year Moving Time-Series
Year Actual Sales Average Forecast Absolute Deviation Forecast Absolute Deviation
1 450 454.8 4.8
2 495 488.4 6.6
3 518 522.0 4.0
4 563 487.7 75.3 555.6 7.4
5 584 525.3 58.7 589.2 5.2
6 ? 555.0 622.8
Total absolute deviation 134.0 28.0
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 58

58 CHAPTER 5 FORECASTING MODELS

MAD0.3 74.56 (see Problem 5-20)

MADmoving average 134/2 67
MADregression 28/5 5.6
Regression (trend line) is obviously the preferred method because
of its low MAD.
5-24. To answer the discussion questions, two forecasting mod-
els are required: a three-period moving average and a three-period
weighted moving average. Once the actual forecasts have been
made, their accuracy can be compared using the mean average dif-
ferences (MAD).
a, b.

Period Month Demand Average Weighted Average

4 Apr. 10 13.67 14.5

5 May 15 13.33 12.67
6 June 17 13.67 13.5
7 July 11 14 15.17
8 Aug. 14 14.33 13.67
9 Sept. 17 14 13.50
10 Oct. 12 14 15
11 Nov. 14 14.33 14
12 Dec. 16 14.33 13.83
13 Jan. 11 14 14.67
14 Feb. 13.67 13.17

c. MAD for moving average is 2.2. MAD for weighted aver-

age is 2.72. Moving average forecast for February is 13.6667.
Weighted moving average forecast for February is 13.1667.
Because a three-period average forecasting method is used,
forecasts start for period 4. As can be seen, the MAD for the mov-
ing average is 2.2, and the MAD for the weighted moving average
is 2.7. Thus, based on this analysis, the moving average appears to
be more accurate. The forecast for February is about 14.
d. There are many other factors to consider, including sea-
sonality and any underlying causal variables such as advertis-
ing budget.
5-25. a.

Sum of
Absolute
Actual Forecast Forecast
Week Miles (Ft) Error RSFE Errors MAD Track Signal
1 17 17.00
2 21 17.00 4.00 4.00 4.00 4.00 1
3 19 17.80 1.20 5.20 5.20 2.60 2
4 23 18.04 4.96 10.16 10.16 3.39 3
5 18 19.03 1.03 9.13 11.19 2.80 3.3
6 16 18.83 2.83 6.30 14.02 2.80 2.25
7 20 18.26 1.74 8.04 15.76 2.63 3.05
8 18 18.61 0.61 7.43 16.37 2.34 3.17
9 22 18.49 3.51 10.94 19.88 2.49 4.21
10 20 19.19 0.81 11.75 20.69 2.30 5.11
11 15 19.35 4.35 7.40 25.04 2.50 2.96
12 22 18.48 3.52 10.92 28.56 2.60 4.20
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 59

CHAPTER 5 FORECASTING MODELS 59

b. The total MAD is 2.60.

c. RSFE is consistently positive. Tracking signal exceeds 5
MADs at week 10. This could indicate a problem.
5-26. a, b. See the accompanying table for a comparison of
the calculations for the exponentially smoothed forecasts using
constants of 0.1 and 0.6.
c. Students should note how stable the smoothed values for
the 0.1 smoothing constant are. When compared to actual
week 25 calls of 85, the 0.6 smoothing constant appears to do
a better job. On the basis of the forecast error, the 0.6 con-
stant is better also. However, other smoothing constants need
to be examined.

Actual Smoothed Smoothed

Week, Value, Value, Forecast Value, Forecast
t At Ft (  0.1) Error Ft (  0.6) Error
1 50 50
2 35 50 15 50 15
3 25 48 23 41 16
4 40 46 6 31 8
5 45 45 0 37 9
6 35 45 10 42 7
7 20 44 24 38 18
8 30 42 12 27 3
9 35 41 6 29 6
10 20 40 20 32 12
11 15 38 23 25 10
12 40 36 4 19 21
13 55 36 19 32 23
14 35 38 3 46 11
15 25 38 13 39 14
16 55 37 18 31 24
17 55 38 16 45 10
18 40 40 0 51 12
19 35 40 5 44 10
20 60 40 20 39 21
21 75 42 33 51 23
22 50 45 5 66 16
23 40 45 5 56 16
24 65 45 20 46 18
25 47 58
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 60

60 CHAPTER 5 FORECASTING MODELS

5-27. Using data from Problem 5-26, with   0.9 5-30. Using QM for Windows, we select Forecasting - Time
Series and multiplicative decomposition. Then specify Centered
Actual Smoothed Moving Average and we have the following results:
Value Value Forecast a. Quarter 1 index 0.8825; Quarter 2 index 0.9816;
Week At Ft Error Quarter 3 index 0.9712; Quarter 4 index 1.1569
1 50 50 b. The trendline is Y 237.7478 3.6658X
2 35 50 15
c. Quarter 1: Y 237.7478 3.6658(17) 300.0662
3 25 36 11
4 40 26 14 Quarter 2: Y 237.7478 3.6658(18) 303.7320
5 45 39 6 Quarter 3: Y 237.7478 3.6658(19) 307.3978
6 35 44 9
Quarter 4: Y 237.7478 3.6658(20) 311.0636
7 20 36 16
8 30 22 8 d. Quarter 1: 300.0662(0.8825) 264.7938
9 35 29 6 Quarter 2: 303.7320(0.9816) 298.1579
10 20 34 14 Quarter 3: 307.3978(0.9712) 298.5336
11 15 21 6
12 40 16 24 Quarter 4: 311.0636(1.1569) 359.8719
13 55 38 17 5-31. Letting
14 35 53 18 t time period (1, 2, 3, . . . , 16)
15 25 37 12
Q1 1 if quarter 1, 0 otherwise
16 55 26 29
17 55 52 3 Q2 1 if quarter 2, 0 otherwise
18 40 55 15 Q3 1 if quarter 3, 0 otherwise
19 35 41 6
Note: if Q1 Q2 Q3 0, then it is quarter 4.
20 60 36 24
21 75 58 17 Using computer software we get
22 50 73 23 Y 281.6 3.7t 75.7Q1 48.9Q2 52.1Q3
23 40 52 12 The forecasts for the next 4 quarters are:
24 65 41 24
25 62
Y 281.6 3.7(17) 75.7(1) 48.9(0) 52.1(0) 268.7
Y 281.6 3.7(18) 75.7(0) 48.9(1) 52.1(0) 299.2
MAD  14.48
Y 281.6 3.7(19) 75.7(0) 48.9(0) 52.1(1) 299.7
Note that in this problem, the initial forecast (for the first period) was Y 281.6 3.7(20) 75.7(0) 48.9(0) 52.1(0) 355.4
not used in computing the MAD. Either approach is considered valid.
5-32. For a smoothing constant of 0.2, the forecast for year 11
5-28. Exponential smoothing with   0.1 is 6.489.
Month Income Forecast Error Year Rate Forecast |Error|
Feb. 70.0 65.0 1 7.2 7.2 0
March 68.5 65.0  0.1 (70  65)  65.5 3.0 2 7 7.2 0.2
April 64.8 65.5  0.1(68.5  65.5)  65.8 1.0 3 6.2 7.16 0.96
May 71.7 65.8  0.1(64.8  65.8)  65.7 6.0 4 5.5 6.968 1.468
June 71.3 65.7  0.1(71.7  65.7)  66.3 5.0 5 5.3 6.674 1.374
July 72.8 66.3  0.1(71.3  66.3)  66.8 6.0 6 5.5 6.400 0.900
Aug. 66.8  0.1(72.8  66.8)  67.4 7 6.7 6.220 0.480
8 7.4 6.316 1.084
MAD  4.20
9 6.8 6.533 0.267
Note that in this problem, the initial forecast (for the first period) was 10 6.1 6.586 0.486
not used in computing the MAD. Either approach is considered valid. 11 6.489
5-29. Exponential smoothing with   0.3 MAD = 0.722
For a smoothing constant of 0.4, the forecast for year 11 is 6.458.
Month Income Forecast Error
Year Rate Forecast |Error|
Feb. 70.0 65.0
March 68.5 66.5 2.0 1 7.2 7.2 0
April 64.8 67.1 2.3 2 7 7.2 0.2
May 71.7 66.4 5.3 3 6.2 7.12 0.92
June 71.3 68.0 3.3 4 5.5 6.752 1.252
July 72.8 69.0 3.8 5 5.3 6.251 0.951
Aug. 70.1 6 5.5 5.871 0.371
7 6.7 5.722 0.978
MAD  3.34 8 7.4 6.113 1.287
Based on MAD,   0.3 produces a better forecast than   0.1 9 6.8 6.628 0.172
(of Problem 5-28). 10 6.1 6.697 0.597
11 6.458
Note that in this problem, the initial forecast (for the first period) was
not used in computing the MAD. Either approach is considered valid. MAD = 0.673
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 61

CHAPTER 5 FORECASTING MODELS 61

For a smoothing constant of 0.6, the forecast for year 11 is 6.401. 5-33. To compute a seasonalized or adjusted sales forecast, we
just multiply each seasonal index by the appropriate trend forecast.
Year Rate Forecast |Error|
Y  seasonal index  Y trend forecast
1 7.2 7.2 0
2 7 7.2 0.2 Hence for:
3 6.2 7.08 0.88 Quarter I: YI  (1.30)($100,000)  $130,000
4 5.5 6.552 1.052
5 5.3 5.921 0.621 Quarter II: Y  (0.90)($120,000)  $108,000
II

6 5.5 5.548 0.048 Quarter III: YIII  (0.70)($140,000)  $98,000

7 6.7 5.519 1.181
Quarter IV: Y  (1.10)($160,000)  $176,000
IV
8 7.4 6.228 1.172
9 6.8 6.931 0.131
5-34.
10 6.1 6.852 0.752
11 6.401 (Average demand (year 1 demand) + (year 2 demand)

MAD = 0.604 for season) 2
For a smoothing constant of 0.8, the forecast for year 11 is 6.256.

Year Rate Forecast |Error| Overall average (sum of all values)

demand =
1 7.2 7.2 0 8
2 7 7.2 0.2
3 6.2 7.04 0.84 (average for season)
Season index =
4 5.5 6.368 0.868 overall average demand
5 5.3 5.674 0.374
6 5.5 5.375 0.125 new annual demand
7 6.7 5.475 1.225 Year 3 demand =
4
8 7.4 6.455 0.945
1, 200
9 6.8 7.211 0.411 = season index
10 6.1 6.882 0.782 4
11 6.256

MAD = 0.577
The lowest MAD is 0.577 for a smoothing constant of 0.8.

Solution Table for Problem 5-34

Average
Year 1 Year 2 (Average Year 1- Season Season Year 3
Season Demand Demand Year 2 Demand) Demand Index Demand
Fall 200 250 225.0 250 0.90 270
Winter 350 300 325.0 250 1.30 390
Spring 150 165 157.5 250 0.63 189
Summer 300 285 292.5 250 1.17 351

5-35. Using Excel, the trend equation is Y 1582.61 612.37X.

SOLUTIONS TO INTERNET HOMEWORK PROBLEMS
5-39. With a  0.4, forecast for 2004  10,339 and MAD 
For 2008, X 19; Y 1582.61 612.37(19) 13217.6
837. With a  0.6, forecast for 2004  10,698 and MAD  612.
For 2009, X 20; Y 1582.61 612.37(20) 13830.0 5-40. Using Excel, the trend line is: GDP  6142.7
For 2010, X 21; Y 1582.61 612.37(21) 14442.4 441.4(time). For 2004 (time  12) the forecast is GDP  6142.7
The MSE from the Excel output is 1654334.7. 441.4(12)  11,439.5.
5-36. a. With a smoothing constant of 0.3, the forecast for 2008 5-41. The trend line found using Excel is: Patients  29.73
is 11211.2 with MSE 3246841. 3.28(time). Note these coefficients are rounded. For the next
b. Using QM for Windows, the best smoothing constant is 3 years (time  11, 12, and 13) the forecasts for the number of
1.0. This gives the lowest MSE of 1443842. patients are:
Patients  29.73 3.28(11)  65.8
5-37. Using Excel, the trend equation is Y 1.1940 0.0095X.
Patients  29.73 3.28(12)  69.1
For January of 2007, X 13; Y 1.1940 0.0095(13) 1.318. Patients  29.73 3.28(13)  72.4
For February of 2007, X 14; Y 1.1940 0.0095(14) 1.327. The coefficient of determination is 0.85, so the model is a fair
5-38. The forecast for January 2007 would be 1.286. model.
The MSE with the trend equation is 0.0003. The MSE with this
exponential smoothing model is 0.0010.
 REVISED
M05_REND6289_10_IM_C05.QXD 5/7/08 4:42 PM Page 62

62 CHAPTER 5 FORECASTING MODELS

5-42. The trend line found using Excel is: Crime Rate  51.98 5-46. The trend line (coefficients from Excel are rounded) for
6.09(time). Note these coefficients are rounded. For the deposits is:
next 3 years (time  11, 12, and 13) the forecasts for the crime Deposits  18.968 1.638(time)
rates are: For 2003, 2004, and 2005, time  45, 46, and 47 respectively. The
Crime Rate  51.98 6.09(11)  118.97 forecasts are:
Crime Rate  51.98 6.09(12)  125.06 Deposits  18.968 1.638(45)  54.7
Crime Rate  51.98 6.09(13)  131.15 Deposits  18.968 1.638(46)  56.4
The coefficient of determination is 0.96, so this is a very good Deposits  18.968 1.638(47)  58.0
model. The trend line (coefficients from Excel are rounded) for GSP is:
5-43. The regression equation (from Excel) is: Patients  1.23 GSP  0.090 0.112(time). The forecasts are:
0.54(crime rate). Note these coefficients are rounded. If the crime GSP  0.090 0.112(45)  5.1
rate is 131.2, the forecast number of patients is:
GSP  0.090 0.112(46)  5.2
Patients  1.23 0.54(131.2)  72.1 GSP  0.090 0.112(47)  5.4
If the crime rate is 90.6, the forecast number of patients is: 5-47. The regression equation from Excel is
Patients  1.23 0.54(90.6)  50.2 Deposits  17.64 13.59(GSP)
The coefficient of determination is 0.90, so this is a good model. In the scatterplot of this data that follows, the pattern appears to
5-44. With a  0.6, forecast for 2003  86.2 and MAD  change around 1985. There are definitely different relationships
3.42. With a  0.2, forecast for 2003  63.87 and MAD  7.23. before 1985 and after 1985, so perhaps the model should be devel-
The model with a  0.6 is better since it has a lower MAD. oped with 1985 as the first year of data.
5-45. With a  0.6, forecast for 2003  4.86 and MAD 
0.23. With a  0.2, forecast for 2003  4.52 and MAD  0.48.
The model with a  0.6 is better since it has a lower MAD.

Deposits and GSP over Time

100

80

60 DEPOSITS

40 GSP

20

0
1950 1960 1970 1980 1990 2000 2010
Time