Solutions Manual Production Operations Analysis 6th Edition Steven Nahmias
Solutions Manual Production Operations Analysis 6th Edition Steven Nahmias
Forecasting
Solutions To Problems From Chapter 2
2.1 Trend
Seasonality
Cycles
Randomness
2.2 Cycles have repeating patterns that vary in length and magnitude.
2.4 Marketing: New sales and existing sales forecasts. Causal models relating advertising
to sales
b) Short term forecasts are generally next day or month; Long term forecasts may be for
many months or years into the future.
c) Causal models are based on relationship between predictor variables and other
variables. Naive models are based on the past history of series only
2.6 The Delphi Method is a technique for achieving convergence of group opinion. The
method has several potential advantages over the Jury of Executive Opinion method
depending upon how that method is implemented. If the executives are allowed to reach a
consensus as a group, strong personalities may dominate. If the executives are
interviewed, the biases of the interviewer could affect the results.
2-1
Chapter 02 - Forecasting
2.7 Some of the issues that a graduating senior might want to consider when choosing a
college to attend include: a) how well have graduates fared on the job market, b) what are
the student attrition rates, c) what will the costs of the college education be and d) what
part-time job opportunities might be available in the region. Sources of data might be
college catalogues, surveys on salaries of graduating seniors, surveys on numbers of
graduating seniors going on to graduate or professional schools, etc.
2.8 The manager should have been prepared for the consequences of forecast error.
2.10 This type of criteria would be closest to MAPE, since the errors measured are relative not
absolute. It makes more sense to use a relative measure of error in golf. For example,
an error of 10 yards for a 200 yard shot would be fine for most golfers, but a similar
error for a 20 yard shot would not.
2.12 a) and b)
et
Forecast Period Actual
1 ei
MAPE = 100 = 25.61
n Di
2-2
Chapter 02 - Forecasting
46 12.89062 17.96875
20 15.0000 5.88253
15 14.66667 4.00000
2 21.81818 1.81818
75 15.55556 33.33333
35 4.761905 6.66667
A bias is indicated when this sum deviates too far from zero.
2-3
Chapter 02 - Forecasting
One-step-ahead Two-step-ahead
e1 e2
Month Forecast Forecast Demand
The one step ahead forecasts gave better results (and should have according to the
theory).
2.21 An MA(1) forecast means that the forecast for next period is simply the current period's
demand.
MAD = 78.0
(Much worse than MA(4))
2-4
Chapter 02 - Forecasting
FMarch = 43.47
FApr = 38.20
FMay = 29.12
2.23 Small implies little weight is given to the current forecast and virtually all weight is
given to past history of demand. This means that the forecast will be stable but not
responsive.
Large implies that a great deal of weight is applied to current observation of demand.
This means that the forecast will adjust quickly to changes in the demand pattern but will
vary considerably from period to period.
2-5
Chapter 02 - Forecasting
4 17.67
5 20.33
6 28.67
7 22.67
8 21.67
b) and c
6.547540 7.934
MAD-ES MAD-MA
Based on these results, ES(.15) had a lower MAD over the five weeks
d) It is the same as the exponential smoothing forecast made in week 6 for the demand
in week 7, which is 19.37 from part c).
2 2
2.25 a) = = .286
N1 7
2 2.05
b) N= N = 39
.05
2 2
From Appendix 2-A
2
c) =1.12
2
e
2
Hence 1.1 Solving gives = .1818
2
2.26 It is the same as the one step ahead forecast made at the end of March which is
31.64.
2-6
Chapter 02 - Forecasting
2.27 The average demand from Jan to June is 161.33. Assume this is the forecast for
July.
a) Month Forecast
3000
2000
1000
500
1 2 3 4 5 6
Jan Feb Mar Apr May Jun
Month
2-7
Chapter 02 - Forecasting
a = D b (n 1) / 2 = -.807.4
c) Regression equation
= -807.4 + (500.54)t
D t
July (t = 7) 2696
Aug (t = 8) 3197
Sept (t = 9) 3698
Oct (t = 10) 4198
Nov (t = 11) 4699
Dec (t = 12) 5199
d) One would think that peak usage would be in the summer months and the increasing
trend would not continue indefinitely.
2-8
Chapter 02 - Forecasting
Since this moving average is used for both one-step-ahead and multiple-step-ahead
forecasts, the total forecast for 1994 is (944.5)(12) = 11,334.)
c )
1200
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
The monthly average is about 1200 based on a usage graph of this shape. This graph
assumes peak usage in summer months. The yearly usage is (1200)(12) = 14,400
which is much closer to (b), since the moving average method does not project trend
indefinitely.
2-9
Chapter 02 - Forecasting
a) slope = 500.54
value of regression in June = -807.4 + (500.54)(6) = 2196
S0 = 2196 = .15
G0 = 500.54 = .10
2.31 This observation would lower future forecasts. Since it is probably an "outlier" (non-
representative observation) one should not include it in forecast calculations.
2.32 Both regression and Holt's method are based on the assumption of constant linear trend.
It is likely in many cases that the trend will not continue indefinitely or that the observed
trend is just part of a cycle. If that were the case, significant forecast errors could result.
2.33
Month Yr 1 Yr 2 Dem1/Mean Dem2/Mean Avg (factor)"
2-10
Chapter 02 - Forecasting
We used the Quick and Dirty Method here. The average demand for the two years
was (689 + 726)/2 = 707.5.
2.34 a)
(1) (2)
Centered MA Ratio
Quarter Demand MA Centered MA on periods (1)/(2)
1 12 42.440 0.2828
2 25 42.440 0.5891
41.25
3 76 41.750 1.8204
42.25
4 52 41.25 43.125 1.2058
44.00
5 16 42.25 43.375 0.3689
42.75
6 32 44.00 44.000 0.7272
45.25
7 71 42.75 45.000 1.5778
44.75
8 62 45.25 46.375 1.3369
48.00
9 14 44.75 49.625 0.2821
51.25
10 45 48.00 49.375 0.9114
47.50
11 84 51.25 49.500 1.6970
12 47 47.50 49.500 0.9494
The four seasonal factors are obtained by averaging the appropriate quarters (1, 5, 9 for
first; 2, 6, 10 for the second, etc.)
0.3112
0.7458
1.6984
1.1641
4
= 1.0214
3, 9163
2-11
Chapter 02 - Forecasting
b)
Deseasonalized
Quarter Demand Factor Series
1 12 0.318 37.74
2 25 0.758 32.98
3 76 1.735 43.80
4 52 1.189 43.73
5 16 0.318 50.31
6 32 0.758 42.22
7 71 1.735 40.92
8 62 1.189 52.14
9 14 0.318 44.03
10 45 0.758 59.37
11 84 1.735 48.41
12 47 1.189 39.53
c) 47.40
Dt
3. ct = -2N+1 = t 0
Vi N 1/ 2 j G0
16
c-7 = = 0.36
45.25 5/ 2 1..56
32
c-6 = = 0.71
45.25 5/ 2 2.56
2-12
Chapter 02 - Forecasting
71
c-5 = = 1.56
43.25 5/ 2 3.56
62
c-4 = = 1.35
45.25 5/ 2 4.56
14
c-3 = = 0.30
47.5 5/ 2 1.56
45
c-2 = = 0.95
47.5 5/ 2 2.56
84
c-1 = = 1.76
47.5 5/ 2 3.56
47
c0 = = 0.97
47.5 5/ 2 4.56
Sum = 3.98
2-13
Chapter 02 - Forecasting
= .3345
The sum of the factors is 4.02. Norming each of the factors by multiplying by
4/4.02 = .995 gives the final factors as:
c1 = .34
2-14
Chapter 02 - Forecasting
c2 = .84
c3 = 1.65
c4 = 1.17
The forecasts for all of 1995 made at the end of 1993 are:
2.42. ARIMA(2,1,1) means 2 autoregressive terms, one level of differencing, and 1 moving
average term. The model may be written ut a0 a1ut 1 a2ut 2 t b1 t 1
where ut Dt Dt 1 . Since ut (1 B) Dt , we have
c) Dt Dt 1 a0 a1 ( Dt 1 Dt 2 ) a2 ( Dt 2 Dt 3 ) t b1 t 1 or
Dt a0 (1 a1 ) Dt 1 a1Dt 2 a2 ( Dt 2 Dt 3 ) t b1 t 1
b) 2 Dt b0 (1 b1B b2 B 2 ) t
2.44. The ARMA(1,1) model may be written Dt a0 a1Dt 1 b1 t 1 t . If we substitute for
Dt 1 , Dt 2 ,... one can easily see this reduces to a polynomial in ( t , t 1 ,...) and if we substitute for
t , t 1 ,... we see that this reduces to a polynomial in Dt 1 , Dt 2 ,... . .
2-15
Chapter 02 - Forecasting
b) (0.2)(0.8)4 = 0.08192
200(0.08192) = 16.384 Change = -16.384 (He should decrease the forecast by
16.384)
Forecast Observed
Quarter Failures (ES(.1)) Error (et)
MAD1 = 50 (given)
The MAD obtained from direct computation is 46.6, so this method gives a pretty
good approximation after eight periods. It has the important advantage of not
requiring the user to save past error values in computing the MAD.
2.47 c1 = 0.7
c2 = 0.8
c3 = 1.0
c4 = 1.5
2-16
Chapter 02 - Forecasting
2.48
Dept yr 1 yr 2 yr 3 ratio 1 ratio 2 ratio 3 average
The multiplicative factors in the final column give the percentages above or below the
grand mean when multiplied by 100.
2.49 a) and b)
1 238
2 220
3 195
4 245 217.67 -27.33 27.33 747.11 11.16
5 345 220.00 -125.00 125.00 15625.00 36.23
6 380 261.67 -118.33 118.33 14002.78 31.14
7 270 323.33 53.33 53.33 2844.44 19.75
8 220 331.67 111.67 111.67 12469.44 50.76
9 280 290.00 10.00 10.00 100.00 3.57
10 120 256.67 136.67 136.67 18677.78 113.89
11 110 206.67 96.67 96.67 9344.44 87.88
12 85 170.00 85.00 85.00 7225.00 100.00
13 135 105.00 -30.00 30.00 900.00 22.22
14 145 110.00 -35.00 35.00 1225.00 24.14
15 185 121.67 -63.33 63.33 4011.11 34.23
16 219 155.00 -64.00 64.00 4096.00 29.22
17 240 183.00 -57.00 57.00 3249.00 23.75
18 420 214.67 -205.33 205.33 42161.78 48.89
19 520 293.00 -227.00 227.00 51529.00 43.65
20 410 393.33 -16.67 16.67 277.78 4.07
21 380 450.00 70.00 70.00 4900.00 18.42
22 320 436.67 116.67 116.67 13611.11 36.46
23 290 370.00 80.00 80.00 6400.00 27.59
24 240 330.00 90.00 90.00 8100.00 37.50
2-17
Chapter 02 - Forecasting
2.49 c)
MA(6) has about the same MAD and higher MSE and MAPE.
2.50
Month Sales ES(.1) Error Abs Err Sq Err Per Err Alpha
1 238 225 -13.00 13.00 169.00 5.46 0.1
2 220 226.30 6.30 6.30 39.69 2.86
3 195 225.67 30.67 30.67 940.65 15.73
4 245 222.60 -22.40 22.40 501.63 9.14
5 345 224.84 -120.16 120.16 14437.78 34.83
6 380 236.86 -143.14 143.14 20489.51 37.67
7 270 251.17 -18.83 18.83 354.47 6.97
8 220 253.06 33.06 33.06 1092.65 15.03
9 280 249.75 -30.25 30.25 915.07 10.80
10 120 252.77 132.77 132.77 17629.15 110.65
11 110 239.50 129.50 129.50 16769.56 117.72
12 85 226.55 141.55 141.55 20035.72 166.53
13 135 212.39 77.39 77.39 5989.65 57.33
2-18
Chapter 02 - Forecasting
The error turns out to be a declining function of for this data. Hence, = 1 gives the
lowest error.
2.51 a)
(Yi) (X i)
Sales Births
Year ($100,000) Preceding Year
1
2 6.4 2.9
3 8.3 3.4
4 8.8 3.5
5 5.1 3.1
6 9.2 3.8
7 7.3 2.8
8 12.5 4.2
Xi = 81.75, Yi = 507.48
2 2
b = SX Y = 4.158
SX X
a = y - bx = -5.8
2-19
Chapter 02 - Forecasting
c)
US Births Forecasted
(in 1,000,000) Births
(Xi)
Year Using ES(.15)
1 2.9
2 3.4
3 3.5
4 3.1
5 3.8 3.2
6 2.8 3.3
7 4.2 3.2
8 3.7 3.4
9 3.4
10 3.4
Hence, forecasted births for years 9 and 10 is 3.4 million.
Substituting gives Yt = -5.8 + (4.158)(3.4) = 8.3372 for sales in each of years 9 and
10. Hence the forecast of total aggregate sales in these years is (8.3372)(2) = 16.6744 or
$1,667,440.
Xi Yi
Month Ice Cream Sales XiYi
1 325 325
2 335 670
3 172 516
4 645 2580
5 770 3850
6 950 5700
Sxx = 105
Sxy = 14709
2-20
Chapter 02 - Forecasting
b = Sxy/Sxx = 140.1
a = Y - bX = 42.5
We would not be very confident about this answer since it assumes the trend
observed over the first six months continues into month 30 which is very unlikely.
b) Xi Yi
Park Ice Cream
attendees Sales XiYi
880 325 286000
976 335 326960
440 172 75680
1823 645 1175835
1885 770 1451450
2436 950 2314200
Sxx = 17,153,756
Sxy = 6,798,070
b = Sxy/Sxx = 0.396302
a = Y -bX = 24.6316
Yi = -24.63 + 0.4Xi
2-21
Chapter 02 - Forecasting
c)
6000
5000
4000
3000
Attendees
2000
1000
2 4 6 8 10 12 14 16 18 20
Months
Using the regression equation Yi = -24.63 + 0.4Xi derived in part (b) we obtain
the ice cream sales predictions below.
Predicted
Month Attendees Ice Cream Sales
12 5100 2015.37
13 5350 2115.37
14 5600 2215.37
15 5800 2295.37
16 5900 2335.37
17 5950 2355.37
18 5980 2367.37
2-22
Chapter 02 - Forecasting
2.53 The method assumes that the "best" based on a past sequence of specific demands will
be the "best" for future demands, which may not be true. Furthermore, the best value
of the smoothing constant based on a retrospective fit of the data may be either larger or
smaller than is desirable on the basis of stability and responsiveness of forecasts.
2.54
Year Demand S sub t G sub t Forecast alpha beta |error| error^2
0 8 0.2 0.2
1981 0.2 6.44 7.69 8.00 7.80 60.84
1982 4.3 12.16 7.29 14.13 9.83 96.59
1983 8.8 17.33 6.87 19.46 10.66 113.58
1984 18.6 23.08 6.64 24.19 5.59 31.30
1985 34.5 30.68 6.84 29.72 4.78 22.85
1986 68.2 43.65 8.06 37.51 30.69 941.74
1987 85.0 58.37 9.39 51.71 33.29 1108.00
1988 58.0 65.81 9.00 67.77 9.77 95.37
14.05 308.78
MAD MSE
The forecast error appears to decrease with decreasing values of and . That is, =
= 0 appears to give the lowest value of the forecast error.
2.55 a) We are given in problem 22 that the forecast for January was 25.
Hence e1 = 25-23.3 = 1.7 = E1 and M1 = |e1 | = 1.7 as well. Hence 1 = 1.
E2 = (.1)(-48.9)(.9)(1.7) = -3.36
M2 = (.1)(48.9) + (.9)(1.7) = 6.42
2 = 3.36/6.42 = .5234
2-23
Chapter 02 - Forecasting
Comparison of Methods
Obviously Trigg-Leach performed much worse for this 3-month period than did
ES(.12). (The respective MAD's are 21.7 for ES and 31.1 for Trigg-Leach.)
b) Consider only the period July to December as in problem 36. As in part (a) 7 = 1.
Use E6 = 567.1 - 480 = 87.
F7 = 480
e7 = 480 - 500 = -20
E7 = (.2)(-20) + (.8)(87) = 65.6
M7 = (.2)(20) + (.8)(87) = 73.6
7 = 65.6/73.6 = .89
8 = 37.9/149.3 = .25
9 = 21.9/171.6 = .13
2-24
Chapter 02 - Forecasting
Performance Comparison
Trigg-Leach
Month Demand Forecast |Error|
7 500 480 20
8 950 498 452
9 350 611 261
10 600 585 15
11 870 586 284
12 740 663 77
MAD = 185
c10 = 1.2
c9 = 1.1
c8 = 0.8
c7 = 0.9
b) D11 = 128
2-25
Chapter 02 - Forecasting
11
C = 4.009
t
t8
Xi Yi XiYi
2.57 a)
1 649.8 649.8
2 705.1 1410.2
3 772.0 2316.0
4 816.4 3265.6
5 892.7 4463.5
6 963.9 5783.4
7 1015.5 7108.5
8 1102.7 8821.6
9 1212.8 10915.2
10 1359.3 13593.0
11 1472.8 16200.8
nn 1
Sxy = n i Di D i
i1 2
= 1112
1174,527.6 10,963.0 = 96,245.6
2
SXX = n n 12n 1 n n 1 11 12 23 11 12 = 1210
2
2 2
2
2 2
6 4 6 4
10,963.0 66
a = Y b X 79.54 = 519.4
11 11
2-26
Chapter 02 - Forecasting
Updating Equations
Di Si GI |Error| |Error|2
Obs Yr
b)
Forecast Forecast
Year GNP % MA(6) GNP |Error| ES(.2) GNP |Error|
1964 649.8
1965 705.1 8.51%
1966 772.0 9.49%
1967 816.4 5.75%
1968 892.7 9.35%
1969 963.9 7.98%
1970 1015.5 5.35%
1971 1102.7 8.59%
1972 1212.8 9.98%
1973 1359.3 12.08%
1974 1472.8 8.35%
1975 1598.4 8.53% 8.72% 1601.3 2.9 8.54% 1598.6 0.2
1976 1782.8 11.54% 8.81% 1739.3 43.5 8.54% 1734.9 47.9
1977 1990.9 11.67% 9.84% 1958.3 32.6 9.14% 1945.7 45.2
2-27
Chapter 02 - Forecasting
c) One would expect that a causal model might be more accurate. Large-scale
econometric models for predicting GNP and other fundamental economic time series
are common.
2-28