STOCKHOLM UNIVERSITY

Department of Statistics

Renata Tsirpitzi
2020-03-15

Edited by Jonas Bjermo

Exercise 15 – NCT chapter 16

16.1 The data file Housing Starts shows private housing units started per thousand of
population in the United States over a period of 24 years. Use a computer to prepare a time
plot of this series and comment on the components of the series revealed by
this plot.

There is a clear cyclic component in the observed time series:

12

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
16.6 The data file Housing Starts shows private housing units started per thousand of
population in the United States over a period of 24 years. Compute a simple, centered 5-point
moving average series for the housing starts data. Draw a time plot of the smoothed series
and comment on your results.
Calculate by hand or with Excel:

Year 5ptMA

1 *
2 *
3 7,76
4 7,66
5 7,80
6 7,56
7 7,30
8 7,10
9 6,94
10 6,82
11 7,62
12 8,58
13 9,02
14 8,84
15 8,52
16 7,96
17 7,52
18 7,40
19 7,70
20 7,76 The smoothed time series still captures some of the cyclic behavior
21 7,28 present in the original data but it is not as extreme.
22 6,38
23 *
24 *
 12

10

Observed
Observerat
6
5pMA
centr 5pt MA
4

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

16.13 The data file Housing Starts shows private housing units started per thousand of
population in the United States over a period of 24 years. Using the data, employ the method
of simple exponential smoothing with smoothing constant a = 0.5 to predict housing starts in
the next 3 years.
Smoothing parameter 𝛼 = 0.5 and start value 𝑥̂1 = 𝑥1 . Calculate the smoothed values with the
formula 𝑥̂𝑡 = 𝛼𝑥𝑡 + (1 − 𝛼)𝑥̂𝑡−1 . Remember that the forecast equals the smoothed value from the
previous time point, 𝑥̃𝑡 = 𝑥̂𝑡−1 . The forecast errors are simply the difference 𝑒𝑡 = 𝑥𝑡 − 𝑥̃𝑡 .
The calculations were done using the Minitab software but it can be done using Excel.

Time 𝒙𝒕 ̂𝟏
SMOOTH, 𝒙 ̃𝒕
FITS, 𝒙 ERROR, 𝒆𝒕
1 8.5 8.50000 - -
2 6.9 7.70000 8.50000 -1.60000
3 7.1 7.40000 7.70000 -0.60000
4 7.8 7.60000 7.40000 0.40000
5 8.5 8.05000 7.60000 0.90000
6 8.0 8.02500 8.05000 -0.05000
7 7.6 7.81250 8.02500 -0.42500
8 5.9 6.85625 7.81250 -1.91250
9 6.5 6.67812 6.85625 -0.35625
10 7.5 7.08906 6.67812 0.82188
11 7.2 7.14453 7.08906 0.11094
12 7.0 7.07227 7.14453 -0.14453
13 9.9 8.48613 7.07227 2.82773
14 11.3 9.89307 8.48613 2.81387
 15 9.7 9.79653 9.89307 -0.19307
16 6.3 8.04827 9.79653 -3.49653
17 5.4 6.72413 8.04827 -2.64827
18 7.1 6.91207 6.72413 0.37587
19 9.1 8.00603 6.91207 2.18793
20 9.1 8.55302 8.00603 1.09397
21 7.8 8.17651 8.55302 -0.75302
22 5.7 6.93825 8.17651 -2.47651
23 4.7 5.81913 6.93825 -2.23825
24 4.6 5.20956 5.81913 -1.21913

Out-of-sample forecasts (future forecasts) are simply the most recent smoothed value: 𝑥̂𝑛+ℎ =
𝑥̂1 for ℎ = 1,2,3, … which is simply 𝑥̂25 = 𝑥̂26 = 𝑥̂27 = 5.20956.
Using statistical software we can get prediction intervals as well (not included in the course!).

12

10

8
Observerat
Observed
6
Smoothed
ExpUtjämnat

4 Forecasts
Prognoser

0
1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627