Professional Documents
Culture Documents
ASSIGNMENT - 2
SUBJECT- TSA
FRANCE GDP DATA
France GDP
480,000
440,000
400,000
360,000
320,000
280,000
240,000
200,000
160,000
1975 1980 1985 1990 1995 2000 2005 2010 2015
France GDP data for the year 1975-2019 is not stationary as seen from the line graph above.
Calculating Log of gdp data, we get LGDP which is then plotted as line graph as follows:
LGDP
13.2
13.0
12.8
12.6
12.4
12.2
12.0
1975 1980 1985 1990 1995 2000 2005 2010 2015
1|Page
After this we will calculate the first difference of the data by the equation=
DLGDP= LGDP -LGDP(-1)
The graph of first difference of log of France GDP DATA is DLGDP and is as follows:
DLGDP
.05
.04
.03
.02
.01
.00
-.01
-.02
-.03
1975 1980 1985 1990 1995 2000 2005 2010 2015
2|Page
ACF is significant at 3 lag values and Pacf is significant at 2 lags.
AR(2) and Ma(3)
We will check equations at ARIMA(p,d,q) combining various lag values at which ACF and PACF are
significant.
ARIMA(0,1,1)
3|Page
ARIMA(0,1,2)
ARIMA(0,1,3)
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 13:13
Sample (adjusted): 1975Q2 2019Q4
Included observations: 179 after adjustments
Failure to improve likelihood (non-zero gradients) after 7 iterations
Coefficient covariance computed using outer product of gradients
MA Backcast: 1974Q3 1975Q1
4|Page
ARIMA(1,1,0)
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 13:15
Sample (adjusted): 1975Q3 2019Q4
Included observations: 178 after adjustments
Convergence achieved after 2 iterations
Coefficient covariance computed using outer product of gradients
ARIMA(1,1,1)
1|Page
ARIMA(2,1,0)
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 13:18
Sample (adjusted): 1975Q3 2019Q4
Included observations: 178 after adjustments
Failure to improve likelihood (non-zero gradients) after 8 iterations
Coefficient covariance computed using outer product of gradients
MA Backcast: 1975Q1 1975Q2
ARIMA(1,1,3)
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 13:20
Sample (adjusted): 1975Q3 2019Q4
Included observations: 178 after adjustments
Failure to improve likelihood (non-zero gradients) after 12 iterations
Coefficient covariance computed using outer product of gradients
MA Backcast: 1974Q4 1975Q2
2|Page
ARIMA(2,1,0)
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 13:22
Sample (adjusted): 1975Q4 2019Q4
Included observations: 177 after adjustments
Convergence achieved after 2 iterations
Coefficient covariance computed using outer product of gradients
ARIMA(2,1,1)
3|Page
ARIMA(2,1,2)
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 13:28
Sample (adjusted): 1975Q4 2019Q4
Included observations: 177 after adjustments
Failure to improve likelihood (non-zero gradients) after 19 iterations
Coefficient covariance computed using outer product of gradients
MA Backcast: 1975Q2 1975Q3
ARIMA(2,1,3)
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 13:29
Sample (adjusted): 1975Q4 2019Q4
Included observations: 177 after adjustments
Failure to improve likelihood (non-zero gradients) after 45 iterations
Coefficient covariance computed using outer product of gradients
MA Backcast: 1975Q1 1975Q3
4|Page
So, making the table for different models -
Then we will do residual diagnostics of the other model with maximum adjusted R square:
ARIMA(2,1,3) correlogram is as follows:
5|Page
Date: 12/15/20 Time: 14:09
Sample (adjusted): 1975Q4 2019Q4
Q-statistic probabilities adjusted for 5 ARMA terms
On visualizing Arima (2,1,3) residuals correlogram we saw that acf and pacf is not significant for any
lag values, so it is stationary and the best model.
So, Best model is Arima (2,1,3), Now we will use this model to forecast values of next time periods.
The best model equation can be found from the following table :
But U2008q1=0
6|Page
Dependent Variable: DLGDP
Method: ARMA Conditional Least Squares (Gauss-Newton / Marquardt
steps)
Date: 12/15/20 Time: 14:16
Sample (adjusted): 1975Q4 2019Q4
Included observations: 177 after adjustments
Failure to improve likelihood (non-zero gradients) after 45 iterations
Coefficient covariance computed using outer product of gradients
MA Backcast: 1975Q1 1975Q3
The values of Yt-1 and Yt-2 can be found from the excel file, whereas the resid folder in Eviews will
give the models Uts terms(Ut-1 and Ut-2):
7|Page
2020Q1= 476666.3394
510,000
Forecast: France
GREAT_BRITF
Actual:
Actual: GREAT_BRITAIN_GDP
France_GDP
500,000
Forecast
Forecast sample:
sample: 2018Q1
2018Q1 2019Q4
2019Q4
Included observations:
Included observations: 8
8
490,000
Root
Root Mean
Mean Squared
Squared Error
Error 6185.639
6185.639
Mean Absolute
Mean Absolute Error
Error 5472.793
5472.793
480,000 Mean
Mean Abs.
Abs. Percent
Percent Error
Error 1.165665
1.165665
Theil Inequality Coef. 0.006570
Theil Inequality Coef. 0.006570
470,000 Bias Proportion
Bias Proportion 0.782797
0.782797
Variance Proportion
Variance Proportion 0.189892
0.189892
460,000 Covariance Proportion
Covariance Proportion 0.027311
0.027311
Theil U2
Theil U2 Coefficient
Coefficient 3.762907
3.762907
450,000 Symmetric
Symmetric MAPE
MAPE 1.157100
1.157100
I II III IV I II III IV
2018 2019
France ± 2 S.E.
Table-A-These are the forecasted values for the gdp data. Table-B These are the actual values of
GDP data.
8|Page