EIE3002 Tutorial 3

1. When α =1, the single exponential smoothing approaches the naïve model.

Lt =α Y t + ( 1−α ) Lt −1
F t+1 =Lt

If α =1;
Lt =1Y t + ( 1−1 ) Lt−1
F t+1 =Lt

F t+1 =Y t

2. An exponential model is developed not based on any specific theory. The parameter
is determined based on solely of the minimum error measure. The error terms tend
to be biased.

3. Given a single randomless series 2, 4, 6, 8, 10, 12, 14, 16, 18, and 20.

(a) Optimal parameter: α =1

α 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
SSE 638.74 381.95 243.59 169.82 117.40 87.31 67.28 53.39 43.40 36.00

MAPE 63.87 53.08 45.05 38.97 34.29 30.59 27.62 25.18 23.15 21.43

(b) Optimal parameters: any value of α and β .

α 0.1 0.5 1.0 0.1 0.5 1.0 0.1 0.5 1.0

β 0.1 0.1 0.1 0.5 0.5 0.5 1.0 1.0 1.0
SSE 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

MAPE 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

(c) Clearly Holt’s method is better as it allows for the trend/growth in the data.

1
4. Based on time series plot of the series, we decide to use Holt-Winters’ Multiplicative
model.

Optimal parameters: α =0.8023 , β=0.0107 and γ=1.0000 .

Forecasts for two years in advance:

Year 1957 1958

Jan 310.02 331.36
Feb 303.31 324.07
Mar 354.96 379.12
Apr 345.26 368.63
May 339.03 361.84
June 376.20 401.38
July 409.22 436.46
Aug 400.81 427.34
Sep 360.46 384.18
Oct 321.40 342.44
Nov 285.34 303.91
Dec 327.19 348.38

5. Estimated models: ^y t =0.51+2.30 t

Sample size (1990 Q1 – 2006 Q4) => T=68

Point forecast: ^y T +h=0.51+2.30 (T + h)

Example: ^y T +1=^y 69=0.51+2.30 ( 69 ) =159.21

T 68
T+h y-hat
h 1 69 159.21
2 70 161.51
3 71 163.81
4 72 166.11
5 73 168.41
6 74 170.71
7 75 173.01
8 76 175.31
9 77 177.61
10 78 179.91
11 79 182.21

2
 12 80 184.51
13 81 186.81
14 82 189.11
15 83 191.41
16 84 193.71

6. The data provided in Q1.xls file is monthly data between Jan 2001 and Dec 2012.

a) Use linear, quadratic and exponential deterministic trend regression to model

trend (use estimation set between Jan 2001 and Dec 2010).

120

100

80

60

40

20

0
0 1 0 1 0 2 0 2 0 3 0 3 0 4 0 5 0 5 0 6 0 6 0 7 0 8 0 8 09 0 9 10 1 0 1 1 1 2 1 2
an- ug- ar- ct- ay- ec- Jul- eb- ep- pr- ov- un- an- ug- ar- ct- ay- ec- Jul- eb- ep-
J A M O M D F S A N J J A M O M D F S

Linear:
Dependent Variable: Y
Method: Least Squares

Sample: 2001M01 2010M12

Included observations: 120

Variable Coefficient Std. Error t-Statistic Prob.  

C 27.71838 0.605445 45.78186 0.0000

T 0.566566 0.008685 65.23820 0.0000

R-squared 0.973023    Mean dependent var 61.99565

Adjusted R-squared 0.972794    S.D. dependent var 19.97938
S.E. of regression 3.295452    Akaike info criterion 5.239490
Sum squared resid 1281.481    Schwarz criterion 5.285948
Log likelihood -312.3694    Hannan-Quinn criter. 5.258357
F-statistic 4256.022    Durbin-Watson stat 0.172499
Prob(F-statistic) 0.000000

3
 Quadratic:
Dependent Variable: Y
Method: Least Squares

Sample: 2001M01 2010M12

Included observations: 120

Variable Coefficient Std. Error t-Statistic Prob.  

C 20.61469 0.292936 70.37261 0.0000

T 0.915928 0.011176 81.95333 0.0000
T^2 -0.002887 8.95E-05 -32.26767 0.0000

R-squared 0.997275    Mean dependent var 61.99565

Adjusted R-squared 0.997228    S.D. dependent var 19.97938
S.E. of regression 1.051874    Akaike info criterion 2.963706
Sum squared resid 129.4534    Schwarz criterion 3.033393
Log likelihood -174.8224    Hannan-Quinn criter. 2.992006
F-statistic 21407.62    Durbin-Watson stat 1.671133
Prob(F-statistic) 0.000000

Exponential:
Dependent Variable: Y
Method: Least Squares
Date: 10/18/12 Time: 13:54
Sample: 2001M01 2010M12
Included observations: 120
Convergence achieved after 11 iterations
Y=C(1)*EXP(C(2)*T)

Coefficient Std. Error t-Statistic Prob.  

C(1) 35.43738 0.793946 44.63449 0.0000

C(2) 0.008578 0.000261 32.81527 0.0000

R-squared 0.915795    Mean dependent var 61.99565

Adjusted R-squared 0.915081    S.D. dependent var 19.97938
S.E. of regression 5.822150    Akaike info criterion 6.377743
Sum squared resid 3999.897    Schwarz criterion 6.424201
Log likelihood -380.6646    Hannan-Quinn criter. 6.396610
Durbin-Watson stat 0.057854

b) Based on AIC and SIC, compare the in-sample estimation results in (a).

Quadratic trend model produces the smallest AIC and SIC for in sample
estimate.

4
 c) Generate point forecasts as well as 95% forecast intervals for 2011 and 2012
using all models specified in (a).

Linear:
120
Forecast: YF
115 Actual: Y
Forecast sample: 2011M01 2012M12
110 Included observations: 24
Root Mean Squared Error 9.574793
105 Mean Absolute Error 9.246030
Mean Abs. Percent Error 9.863647
100 Theil Inequality Coefficient 0.048744
Bias Proportion 0.932506
95 Variance Proportion 0.034245
Covariance Proportion 0.033248
90

85
I II III IV I II III IV
2011 2012

YF ± 2 S.E.

MSE = (9.574793)2 = 91.68.

MAPE = 9.8637.

Manual calculation for point forecasts: (do via similar way for other models).

Yt t y-hat et et^2 APE

Jan-11 92.13 121.00 96.27 -4.14 17.18 4.50
Feb-11 91.85 122.00 96.84 -4.99 24.89 5.43
Mar-11 88.96 123.00 97.41 -8.45 71.41 9.50
Apr-11 90.79 124.00 97.97 -7.18 51.58 7.91
May-11 93.60 125.00 98.54 -4.94 24.36 5.27
Jun-11 91.66 126.00 99.11 -7.44 55.40 8.12
Jul-11 92.61 127.00 99.67 -7.06 49.86 7.62
Aug-11 92.15 128.00 100.24 -8.09 65.39 8.78
Sep-11 92.20 129.00 100.81 -8.61 74.07 9.33
Oct-11 90.50 130.00 101.37 -10.87 118.19 12.01
Nov-11 91.00 131.00 101.94 -10.94 119.71 12.02
Dec-11 93.82 132.00 102.51 -8.69 75.50 9.26
Jan-12 94.46 133.00 103.07 -8.61 74.15 9.12
Feb-12 96.43 134.00 103.64 -7.21 52.02 7.48
Mar-12 94.84 135.00 104.20 -9.36 87.70 9.87
Apr-12 94.09 136.00 104.77 -10.68 114.00 11.35
May-12 93.69 137.00 105.34 -11.65 135.76 12.44
Jun-12 95.11 138.00 105.90 -10.80 116.61 11.35
Jul-12 94.65 139.00 106.47 -11.82 139.74 12.49
Aug-12 95.02 140.00 107.04 -12.02 144.38 12.65
Sep-12 96.57 141.00 107.60 -11.04 121.79 11.43
Oct-12 94.57 142.00 108.17 -13.60 185.00 14.38
Nov-12 97.32 143.00 108.74 -11.42 130.44 11.74

5
 Dec-12 97.01 144.00 109.30 -12.29 151.07 12.67

MSE MAPE
91.6757 9.8636

Manual calculation for 95% forecast intervals*: (do via similar way for other models).

  Yt y-hat se 95% lower interval 95% upper interval

Jan-11 92.1282 96.2729 3.3506 89.7057 102.8401
Feb-11 91.8508 96.8395 3.3520 90.2696 103.4094
Mar-11 88.9555 97.4060 3.3534 90.8334 103.9787
Apr-11 90.7903 97.9726 3.3548 91.3972 104.5480
May-11 93.6036 98.5392 3.3562 91.9610 105.1174
Jun-11 91.6625 99.1057 3.3577 92.5247 105.6868
Jul-11 92.6109 99.6723 3.3592 93.0883 106.2563
Aug-11 92.1522 100.2389 3.3607 93.6520 106.8258
Sep-11 92.1988 100.8054 3.3622 94.2155 107.3954
Oct-11 90.5007 101.3720 3.3637 94.7791 107.9650
Nov-11 90.9971 101.9386 3.3653 95.3426 108.5346
Dec-11 93.8162 102.5051 3.3669 95.9060 109.1043
Jan-12 94.4604 103.0717 3.3685 96.4694 109.6740
Feb-12 96.4255 103.6383 3.3702 97.0328 110.2438
Mar-12 94.8401 104.2048 3.3718 97.5961 110.8136
Apr-12 94.0944 104.7714 3.3735 98.1594 111.3835
May-12 93.6864 105.3380 3.3752 98.7226 111.9533
Jun-12 95.1060 105.9045 3.3769 99.2858 112.5233
Jul-12 94.6501 106.4711 3.3786 99.8490 113.0933
Aug-12 95.0217 107.0377 3.3804 100.4121 113.6633
Sep-12 96.5682 107.6042 3.3822 100.9751 114.2334
Oct-12 94.5692 108.1708 3.3840 101.5382 114.8035
Nov-12 97.3161 108.7374 3.3858 102.1011 115.3736
Dec-12 97.0127 109.3039 3.3877 102.6641 115.9438

* y t +h∨T ± 1.96 σ^ h

6
 Quadratic:
96
Forecast: YF
Actual: Y
94 Forecast sample: 2011M01 2012M12
Included observations: 24
Root Mean Squared Error 2.800634
92 Mean Absolute Error 2.485640
Mean Abs. Percent Error 2.630536
Theil Inequality Coefficient 0.015162
90
Bias Proportion 0.731617
Variance Proportion 0.153960
88 Covariance Proportion 0.114423

86
I II III IV I II III IV
2011 2012

YF ± 2 S.E.

MSE = (2.800634)2 = 7.8436.

MAPE = 2.630536.

Exponential:
140
Forecast: YF
Actual: Y
130
Forecast sample: 2011M01 2012M12
Included observations: 24
120 Root Mean Squared Error 17.78781
Mean Absolute Error 17.08401
110 Mean Abs. Percent Error 18.19358
Theil Inequality Coefficient 0.087030
100
Bias Proportion 0.922433
Variance Proportion 0.061653
Covariance Proportion 0.015914
90

80
I II III IV I II III IV
2011 2012

YF ± 2 S.E.

MSE = (17.78781)2 = 316.4062.

MAPE = 18.19358.

d) Evaluate out-of-sample forecast accuracy for all models based on MSE and
MAPE.
MSE and MAPE for each model are given in (c). Quadratic trend model
produces the most accurate forecasts. Exponential trend model on the
other hand produces the worst forecasts.

7
7. (a) Y t =S t ×T t × E t .

(b) Magnitude of seasonality fluctuations increase proportionally with increases in the

level of the series.
2
(c) Use a non-linear model, example quadratic model Des Y t =β 0+ β 1 t + β 1 t + ε t ,
where t is a time dummy.
Use non-linear model because T t forms a curve (non-linear) and not a straight line
(linear).

Des Y t = ^β 0+ β^ 1 t+ β^ 1 t .
(d) Produce forecast of seasonally adjusted series, ^
2

Then produce forecast of Y t by computing ^

Des Y t × St .