Professional Documents
Culture Documents
Week 5 - Forecasting Techniques
Week 5 - Forecasting Techniques
A forecast is simply a prediction of what will happen in the future. Managers must
learn to accept the fact that, regardless of the technique used, they will not be able to
develop perfect forecasts. The purpose of forecasting is only to help managers
develop appropriate strategies for the future.
Quantitative forecasting methods include time series methods and causal methods. A
time series method is appropriate when the historical data are restricted to past values
of the variable being forecast. The three time series methods discussed here are
smoothing (moving averages, weighted moving averages, and exponential
smoothing), trend projection and trend projection adjusted for seasonal influence.
Smoothing methods are appropriate for a stable time series; that is, one that exhibits
no significant trend, cyclical or seasonal effects. The moving averages approach
consists of computing an average of past values and then using this average as the
forecast for the next period. The weighted moving averages method allows for the
possibility of unequal weights for the data; thus, the moving averages method is a
special case of the weighted moving averages method in which all the weights are
equal. Exponential smoothing also is a special case of the weighted moving averages
method involving only on e parameter; the weight for the most recent observation.
When a time series consists of random fluctuations around a long-term trend line, a
linear equation may be used to estimate the trend. When seasonal effects are present,
seasonal indexes can be computed and used to deseasonalize the data and to develop
forecasts. When both seasonal and long-term trend effects are present, a trend line is
fitted to the deseasonalized data; the seasonal indexes are then used to adjust the trend
projection.
1|P age
A. Components of a Time Series
The usual assumption is that four separate components – trend, cyclical, seasonal and
irregular – combine to provide specific values for the time series.
• The trend component accounts for the gradual shifting of the time series over a
long period of time.
• Any regular pattern of sequences of values above and below the trend line is
attributable to the cyclical component of the series.
B. Smoothing Methods
•Moving averages
•Centered moving averages
•Weighted moving averages
•Exponential smoothing
2|P age
1. Moving Average Method
The moving average method consists of computing an average of the most recent n
data values for the series and using this average for forecasting the value of the time
series for the next period.
In the weighted moving average method for computing the average of the most recent
n periods, the more recent observations are typically given more weight than older
observations. For convenience, the weights usually sum to 1.
4. Exponential Smoothing
Using exponential smoothing, the forecast for the next period is equal to the forecast
for the current period plus a proportion (a) of the forecast error in the current period.
3|P age
When deciding on the better forecasting method, we need to consider their accuracy.
These are measured by:
The average of the squared forecast errors for the historical data is calculated. The
forecasting method or parameter(s) which minimize this mean squared error is then
selected.
The mean of the absolute values of all forecast errors is calculated, and the forecasting
method or parameter(s) which minimize this measure is selected. The mean absolute
deviation measure is less sensitive to individual large forecast errors than the mean
squared error measure.
Illustrated Problem 1
Corporate Triple A Bond interest rates for 12 consecutive months are 9.5, 9.3, 9.4, 9.6,
9.8, 9.7, 9.8, 10.5, 9.9, 9.7, 9.6 and 9.6.
a) Develop three- and four-month moving averages for this time series.
Which moving average provides the better forecasts? Explain.
4|P age
Solution
a.
Time- 3-Month Moving 4-Month Moving
Month Series Average Forecast (Error)2 Average (Error)2
Value Forecast
1 9.5
2 9.3
3 9.4
4 9.6 9.40 0.04
5 9.8 9.43 0.14 9.45 0.12
6 9.7 9.60 0.01 9.53 0.03
7 9.8 9.70 0.01 9.63 0.03
8 10.5 9.77 0.53 9.73 0.59
9 9.9 10.00 0.01 9.95 0.00
10 9.7 10.07 0.14 9.98 0.08
11 9.6 10.03 0.18 9.97 0.14
12 9.6 9.73 0.02 9.92 0.10
1.08 1.09
Illustrated Problem 2
For the Hawkins Company, the monthly percentages of all shipments that were
received on time over the past 12 months are 80, 82, 84, 83, 83, 84, 85, 84, 82, 83, 84
and 83.
5|P age
Solution
a.
3-Month α =2
Month Yt Moving (Error)2 Forecast (Error)2
Averages
Forecast
1 80
2 82 80.00 4.00
3 84 80.40 12.96
4 83 82.00 1.00 81.12 3.53
5 83 83.00 0.00 81.50 2.25
6 84 83.33 0.45 81.80 4.84
7 85 83.33 2.79 82.24 7.62
8 84 84.00 0.00 82.79 1.46
9 82 84.33 5.43 83.03 1.06
10 83 83.67 0.45 82.83 0.03
11 84 83.00 1.00 82.86 1.30
12 83 83.00 0.00 83.09 0.01
11.12 39.06
6|P age
Activity 3.1 (Answer at end of chapter)
Moving averages often are used to identify movements in stock prices. Daily closing
prices (in dollars per share) for SanDisk for August 16, 2002, through September 3,
2002, follow:
a) Use a five-day moving average to smooth the time series. Forecast the
closing price for September 4, 2002.
b) Use a four-day weighted moving average to smooth the time series. Use a
weight of 0.4 for the most recent period, 0.3 for the next period back, 0.2
for the third period back, and 0.1 for the fourth period back. Forecast the
closing price for September 4, 2002.
7|P age
C. Trend Projection
If a time series exhibits a linear trend, the method of least squares may be used to
determine a trend line (projection) for future forecasts. Least squares, also used in
regression analysis, determines the unique trend line forecast which minimizes the
mean square error between the trend line forecasts and the actual observed values for
the time series. The independent variable is the time period and the dependent
variable is the actual observed value in the time series.
Using the method of least squares, the formula for the trend projection is:
Tt = b0 + b1t.
b1 = nStYt - St SYt
nSt 2 - (St )2
b0 = Y - b1 t
8|P age
Illustrated Problem 3
The enrolment data (figures in thousands) for a state college for the past six years are
shown.
Year 1 2 3 4 5 6
Enrollment 20.5 20.2 19.5 19.0 19.1 18.8
Develop the equation for the linear trend component for this time series. Comment on
what is happening to enrolment at this institution.
Solution
The following values are needed to compute the slope and intercept:
∑ t = 21 ∑t 2
= 91 ∑Y t = 117.1 ∑ tY t = 403.7
Computation of slope:
b1 =
∑ tYt − d∑ t ∑ Y i / n t
=
4 0 3 . 7 − ( 2 1 ) ( 1 1 7 .1 ) / 6
= − 0 .3 5 1 4
− d∑ ti / n
2
9 1 − ( 2 1) 2 / 6
∑ t 2
Computation of intercept:
9|P age
Illustrated Problem 4
Hudson Marine has been an authorised dealer for C&D marine radios for the past
seven years. The number of radios sold each year is shown.
Year 1 2 3 4 5 6 7
Number Sold 35 50 75 90 105 110 130
b) Develop the equation for the linear trend component for the time series.
c) Use the linear trend developed in part (b) to prepare a forecast for sales in year 8.
Solution
b. The following values are needed to compute the slope and intercept:
2
∑t = 28 ∑t = 140 ∑ Yt = 595 ∑ t Y t = 2815
Computation of slope:
b1 =
∑ tY − d
t ∑ t ∑ Y i / n = 2815 − (28)(595) / 7 = 155357
t
.
∑ ∑t − d
2
t i2
/ n
2
140 − (28) / 7
Computation of intercept:
b0 = Y - b1 t = 85 - 15.5357(4) = 22.857
10 | P a g e
Activity 2 (Answer at end of chapter)
The Garden Avenue Seven sells tapes of its musical performances. The following
data show sales for the past 18 months. The group’s manager wants an accurate
method for forecasting future sales.
a) Use exponential smoothing, with α = 0.3, 0.4 and 0.5. Which value of α
provides the best forecasts?
11 | P a g e
D. Trend and Seasonal Components
12 | P a g e
Illustrated Problem 5
Identify the monthly seasonal indexes for the following three years of expenses for a
six-unit apartment house in Changi. Use a 12-month moving average calculation.
Solution
Seasonal-Irregular
Month Component Values Seasonal Adjusted Seasonal Index
Index
1 0.72 0.70 0.71 0.707
2 0.80 0.75 0.78 0.777
3 0.83 0.82 0.83 0.827
4 0.94 0.99 0.97 0.966
5 1.01 1.02 1.02 1.016
6 1.25 1.36 1.31 1.305
7 1.49 1.51 1.50 1.494
8 1.19 1.26 1.23 1.225
9 0.98 0.97 0.98 0.976
10 0.98 1.00 0.99 0.986
11 0.93 0.94 0.94 0.936
12 0.78 0.80 0.79 0.787
12.05
Notes:
2. The adjustment is really not necessary in this problem since it implies more
13 | P a g e
accuracy than is warranted. That is, the seasonal component values and the seasonal
index were rounded to two decimal places.
Illustrated Problem 6
b) When does Castello Music experience the largest seasonal effect? Does this
result appear to be reasonable? Explain.
c) Deseasonalise the data and use the deseasonalised time series to identify the
trend.
d) Use the results of part (c) to develop a quarterly forecast for next year based on
trend.
e) Use the seasonal indexes developed in (a) to adjust the forecasts developed in
part (d) to account for seasonal effects.
14 | P a g e
Solution
a)
Seasonal-Irregular Seasonal
Component Values Index
Quarter
1
1.371, 1.270, 1.116, 1.200 1.239
2 0.681, 0.364, 0.776, 0.576 0.597
3 0.308, 0.533, 0.571, 0.528 0.485
4 1.333, 1.778, 1.641, 1.558 1.578
Total 3.899
Centered Seasonal-Irregular
t Sales Moving Average Component
1 4
2 2
3 1 3.250 0.308
4 5 3.750 1.333
5 6 4.375 1.371
6 4 5.875 0.681
7 4 7.500 0.533
8 14 7.875 1.778
9 10 7.875 1.270
10 3 8.250 0.364
11 5 8.750 0.571
12 16 9.750 1.641
13 12 10.750 1.116
14 9 11.750 0.766
15 7 13.250 0.528
16 22 14.125 1.558
17 18 15.000 1.200
18 10 17.375 0.576
19 13
20 35
Quarter Adjusted
Seasonal Index
1 1.271
2 0.613
3 0.498
4 1.619
b. The largest effect is in quarter 4; this seems reasonable since retail sales are
generally higher during October, November, and December.
15 | P a g e
c. Note: To simplify the calculations the seasonal indexes in problem 29 have been
round to two decimal places.
Yt
t (deseasonalized) tYt t2
1 3.15 3.15 1
2 3.28 6.56 4
3 2.00 6.00 9
4 3.09 12.36 16
5 4.72 23.60 25
6 6.56 39.36 36
7 8.00 56.00 49
8 8.64 69.12 64
9 7.87 70.83 81
10 4.92 49.20 100
11 10.00 110.00 121
12 9.88 118.56 144
13 9.45 122.85 169
14 14.75 206.50 196
15 14.00 210.00 225
16 13.58 217.28 256
17 14.17 240.89 289
18 16.39 295.02 324
19 26.00 494.00 361
20 21.60 432.00 400
210 202.05 2783.28 2870
16 | P a g e
d.
y Trend Forecast
21 20.55
22 21.55
23 22.54
24 23.54
e.
Trend Seasonal Quarterly
Year Quarter Forecast Index Forecast
6 1 20.55 1.27 26.10
2 21.55 0.61 13.15
3 22.54 0.50 11.27
4 23.54 1.62 38.13
Air pollution control specialists in Singapore monitor the amount of ozone, carbon
dioxide and nitrogen dioxide in the air on an hourly basis. The hourly time series data
exhibit seasonality, with the levels of pollutants showing similar patterns over the
hours in the day. On July 15, 16 and 17, the observed levels of nitrogen dioxide in
Orchard Road for the 12 hours from 6 a.m. to 6 p.m. were as follows:
July 15 25 28 35 50 60 60 40 35 30 25 25 20
July 16 28 30 35 48 60 65 50 40 35 25 20 20
July 17 35 42 45 70 72 75 60 45 40 25 25 25
a) Identify the hourly seasonal indexes for the 12 hourly daily readings.
b) Based on the seasonal indexes in part (a), the trend equation developed for the
deseasonalised data is Tt = 32.983 + 0.3922t. Using only the trend equation,
develop forecasts for the 12 hours for July 18.
c) Use the seasonal indexes from part (a) to adjust the trend forecasts in part (b).
17 | P a g e
Solutions to Activities
Activity 1
a.
5-Day Moving Forecast
Day Time-Series Average Forecast Error (Error)2
Value
1 14.45
2 15.75
3 16.45
4 17.40
5 17.32
6 15.96 16.27 -0.31 0.10
7 16.45 16.58 -0.13 0.02
8 15.60 16.72 -1.12 1.25
9 15.09 16.55 -1.46 2.12
10 16.42 16.08 0.34 0.11
11 16.21 15.90 0.31 0.09
12 15.22 15.95 -0.73 0.54
b. The weighted moving average forecasts for days 5-12 are 16.49, 17.01, 16.71,
16.57, 16.10, 15.60, 15.09, 16.42, 16.21 and 15.22
c. The exponential smoothing forecasts for days 2-12 are 14.45, 15.36, 16.12, 17.02,
17.23, 16.34, 16.42, 15.85, 15.32, 16.09 and 16.17
d.
Method MSE
Moving Averages 0.60
Weighted Moving Average 0.65
Exponential Smoothing 0.87
Moving Averages is the best of the three approaches because it has the smallest
MSE.
18 | P a g e
Activity 2
Note: Results were obtained using the Forecasting module of The Management
Scientist.
a.
Smoothing Constant MSE
α = 0.3 4,492.37
α = 0.4 2,964.67
α = 0.5 2,160.31
The α = 0.5 smoothing constant is better because it has the smallest MSE.
b. Tt = 244.778 + 22.088t
MSE = 357.81
c. Trend projection provides much better forecasts because it has the smallest
MSE. The reason MSE is smaller for trend projection is that sales are
increasing over time; as a result, exponential smoothing continuously
underestimates the value of sales. If you look at the forecast errors for
exponential smoothing you will see that the forecast errors are positive for
periods 2 through 18.
Activity 3
a. Use a twelve period moving averages. After centering the moving averages, you
should obtain the following seasonal indexes:
Hour Seasonal Index Hour Seasonal Index
1 0.771 7 1.207
2 0.864 8 0.994
3 0.954 9 0.850
4 1.392 10 0.647
5 1.571 11 0.579
6 1.667 12 0.504
b. The hours of July 18 are number 37 to 48 in the time series. Thus the trend
component for 7:00 a.m. on July 18 (period 37) would be
19 | P a g e
5 49.06 11 51.42
6 49.46 12 51.81
20 | P a g e