EQC7006 Time Series Analysis
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
EQC7006 Time Series Analysis
Question 1 (5 marks)
a) –
Year |
Yt |
3 MA |
3 WMA |
1 |
42 |
|
|
2 |
69 |
70.33 |
70.00 |
3 |
100 |
94.67 |
96.00 |
4 |
115 |
115.67 |
115.50 |
5 |
132 |
129.33 |
130.00 |
6 |
141 |
142.33 |
142.00 |
7 |
154 |
155.33 |
155.00 |
8 |
171 |
168.33 |
169.00 |
9 |
180 |
185.00 |
183.75 |
10 |
204 |
204.00 |
204.00 |
11 |
228 |
226.33 |
226.75 |
12 |
247 |
|
|
|
|
|
|
|
|
(2 marks) |
(2 marks) |
b) Additive: = −
Multiplicative:
(0.5 mark / 0 mark)
(0.5 mark / 0 mark)
(4 marks)
(1 mark)
Question 2 (10 marks)
a) Each of values gives estimate of average monthly temperature for different months (January – December)
(1 mark)
b) Coefficient of T in Model 2 is not significant (p-value > 0.05). Model 1 has smaller AIC and SIC.
Thus, Model 1 should be used.
(2 marks)
c) 0 : 1 = 2 = ⋯ = 11 = 12
1 : 0 is not true (1 / 1)
(37. 1230−15.4378)/11
= = 21.0702 (1 / 1)
Reject the null hypothesis. (0.5 / 0.5)
There is seasonal variation. (1 / 1)
(4 marks)
d) MAPE = 2.1616. The average absolute errors is 2.1616% of the observed values. (1)
ACF1 = 0.4512. Correlation between two successive errors is 0.4512, considerably is not strong. (1)
Theil’s U = 1.6157. Since the value is greater than 1, naïve method produces better forecasts (1)
(3 marks)
Question 3 (7.5 marks)
a) Both ADF tests have p-value of the t-stat smaller than 0.05. ADF test results suggest that the level of the series is stationary. (0.75)
Both KPSS tests have LM-stat smaller than critical value at 5% significance value. KPSS
test results suggest that the level of the series is stationary. (0.75)
Overall decision; differencing is not needed. (0.5)
(2 marks)
b) Since the level is stationary, identify possible models by inspecting correlogram of level .
Possible models:
1) ARMA(0,6)(2,0)12
ACF: Spikes at lags 1 to 6, cut off to zero.
Exponential decay at lags 12, 24, 36.
PACF: Assuming true PAC exponential decay at lags 1, 2,3,… .
Spikes at lags 12, 24, and cut off to zero at lags 36, 48,…
(2.5 marks)
2) ARMA(6,0)(2,0)12
ACF: Assuming true AC exponential decay at lags 1,2,3,… .
Exponential decay at lags 12, 24, 36.
PACF: Spikes at lags 1 to 6, cut off to zero.
Spikes at lags 12, 24, and cut off to zero at lags 36, 48,…
(2.5 marks)
# Other appropriate models are also acceptable.
c) Since the level is stationary and ARMA models can be applied on levels, decision on inclusion of a constant term can be made based on the time plot of the series.
(0.5 mark)
Question 4 (7.5 marks)
a) Both ADF tests have p-value of the t-stat greater than 0.05. ADF test results suggest that the level of the series is non-stationary. (0.75)
Both KPSS tests have LM-stat smaller than critical value at 5% significance value. KPSS
test results suggest that the level of the series is stationary. (0.75)
Overall decision; differencing is needed. Since the series has seasonal variation, seasonal
differencing is preferable. (0.5)
(2 marks)
b) Since the level is considered as non-stationary, identify possible models by inspecting correlogram of the seasonally differenced series.
Possible models:
1) ARMA(0,5)(0,1)12
ACF: Spikes at lags 1 to 5, cut off to zero.
Spike at lags 12, and cut off to zero at lags 24, 36,… .
PACF: PAC exponential decay at lags 1, 2, 3,… .
Assuming the true PAC exponential decay at lags 12, 24, 36
(2.5 marks)
2) ARMA(2,0)(3,0)12
ACF: Exponential decay at lags 1, 2, 3,… .
Assuming true AC exponential decay at lags 12, 24, 36 …
PACF: Spikes at lags 1 and 2, cut off to zero.
Spikes at lags 12 and 36 and cut off to zero at lags 48, 60 …
(2.5 marks)
# Other appropriate models are also acceptable.
c) Since the level is non-stationary and ARMA models cannot be applied on level, decision on inclusion of a constant term cannot be made based on the time plot of the series.
(0.5 mark)
Question 5 (10 marks)
a) All inverted AR roots are inside unit circle, the estimated process is stationary (1)
c) (1 − 1 − 2 2 )(1 − 12 ) = (1 + Θ1 12) ( )1
= 1 −1 + 2 −2 + −12 − 1 −13 − 2 −14 + + Θ 1 −12 ( )2
d) January 2016:
193 = 25.4787 (1.5)
February 2016:
194 = 25.6107 (1.5)
January 2017:
25.4787 ± 1.96(√0.0663) = (24.9740, 25.9834) ( )1
25.6107 ± 1.96((√0.0754) = (25.0725, 26.1489) (1)
Question 6 (10 marks)
a) –
Statistics |
Models |
|||
ARMA(0,6)(2,0) with non-zero mean |
ARMA(0,6) with non-zero mean |
ARMA(6,0)(2,0) with non-zero mean |
ARMA(0,1)(0,1) with zero mean |
|
Q-statistic |
5.6001 |
25.737 |
8.7182 |
215.28 |
Degree of freedom |
5 |
6 |
5 |
11 |
Critical value (5% sig. level) |
11.07 |
14.07 |
11.07 |
19.68 |
Decision (Reject or do not reject the null hypothesis at 5% sig. level) |
Do not reject the null hypothesis |
Reject the null hypothesis |
Do not reject the null hypothesis |
Reject the null hypothesis |
Conclusion (The errors are white noise or not) |
Residuals are white noise |
Residuals are not white noise |
Residuals are white noise |
Residuals are not white noise |
SIC |
11.4456 |
11.5851 |
11.4661 |
12.6959 |
(0.5 mark each) (8 marks)
b) ARMA(0,6)(2,0) with non-zero mean since it is an adequate model and has the smallest value of SIC.
(1 mark)
c) (1 − Φ1 12 − Φ2 24)( − ) = (1 + 1 + 22 + 33 + 44 + 55 + 66)
(1 mark)
2022-05-19