Bucharest
2015
Contents
Pros and cons for the estimation of an aggregated production function.....................................3
Testing the stationarity of the variables concerned................................................................3
Regression using differences of logarithms........................................................................17
We see now if the production function has constant increasing or decreasing returns to scale......17
t-Statistic
Prob.*
0.599777
-4.148465
-3.500495
-3.179617
0.9993
Coefficien
t
Std. Error
t-Statistic
Prob.
RGNP(-1)
C
@TREND(1)
0.018035
37.19860
2.293369
0.030069
50.21181
6.102186
0.599777
0.740834
0.375827
0.5515
0.4624
0.7087
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.355224
0.328358
121.1752
704804.8
-315.4791
1.544039
212.1864
147.8580
12.48937
12.60301
13.22221
0.000027
T-Statistic Dickey-Fuller value 0.59 is higher than the 1% level therefore this series is not stationary.
1st difference is stationary t Stat is indeed lower than 1% level therefore RGNP is stationary.
Null Hypothesis: D(RGNP) has a unit root
t-Statistic
Prob.*
-5.406271
-4.152511
-3.502373
-3.180699
0.0003
Coefficien
t
Std. Error
t-Statistic
Prob.
D(RGNP(-1))
C
@TREND(1)
-0.761732
36.86864
4.778447
0.140898
35.87833
1.412884
-5.406271
1.027602
3.382053
0.0000
0.3094
0.0015
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.384464
0.358271
118.3429
658237.4
-308.0794
1.874641
5.106063
147.7293
12.44317
12.55790
14.67812
0.000011
As for capital upon performing the 1st difference unit root test t-Stat is still higher than 1% is not
stationary.
Null Hypothesis: D(CAPITAL) has a unit root
Exogenous: Constant, Linear Trend
t-Statistic
Prob.*
-4.068445
-4.156734
-3.504330
-3.181826
0.0126
Coefficien
t
Std. Error
t-Statistic
Prob.
D(CAPITAL(-1))
D(CAPITAL(-1),2)
C
@TREND(1)
-0.352474
0.535092
93.02340
4.470783
0.086636
0.127157
28.96660
1.225176
-4.068445
4.208137
3.211403
3.649095
0.0002
0.0001
0.0024
0.0007
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.363456
0.321019
61.90488
172449.7
-269.5960
1.760140
13.80499
75.12702
11.16718
11.32162
8.564734
0.000131
Upon performing the 2nd difference we can see that the series is stationary. T-Stat is lower than 1%.
t-Statistic
Prob.*
-4.068445
-4.156734
-3.504330
-3.181826
0.0126
Coefficien
t
Std. Error
t-Statistic
Prob.
D(CAPITAL(-1))
D(CAPITAL(-1),2)
C
@TREND(1)
-0.352474
0.535092
93.02340
4.470783
0.086636
0.127157
28.96660
1.225176
-4.068445
4.208137
3.211403
3.649095
0.0002
0.0001
0.0024
0.0007
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.363456
0.321019
61.90488
172449.7
-269.5960
1.760140
13.80499
75.12702
11.16718
11.32162
8.564734
0.000131
Labour After computing the first unit root test we can see that the series is not stationary as t-Stat is
higher than 1% level.
Null Hypothesis: LABOR has a unit root
t-Statistic
Prob.*
-3.817639
-4.152511
-3.502373
-3.180699
0.0237
Coefficien
t
Std. Error
t-Statistic
Prob.
LABOR(-1)
D(LABOR(-1))
C
@TREND(1)
-0.249191
0.413125
14.95700
0.424917
0.065274
0.122500
3.736474
0.109178
-3.817639
3.372445
4.002973
3.891978
0.0004
0.0015
0.0002
0.0003
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.344121
0.301346
1.064630
52.13805
-71.99373
1.948348
1.537123
1.273701
3.039749
3.192711
8.044958
0.000205
Upon performing the second unit root test, we can see that t-Stat is lower than 1% therefore is stationary.
t-Statistic
Prob.*
-5.082033
-4.156734
-3.504330
-3.181826
0.0007
Coefficien
t
Std. Error
t-Statistic
Prob.
D(LABOUR(-1))
D(LABOUR(-1),2)
C
@TREND(1)
-0.839604
0.269304
0.963392
0.012310
0.165210
0.143826
0.416652
0.012175
-5.082033
1.872423
2.312225
1.011117
0.0000
0.0677
0.0254
0.3174
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.379038
0.337641
1.189596
63.68126
-75.94870
2.028437
0.019436
1.461682
3.263212
3.417647
9.156077
0.000076
t-Statistic
Prob.*
-2.327151
-4.152511
-3.502373
-3.180699
0.4121
Coefficien
t
Std. Error
t-Statistic
Prob.
LOG_RGNP(-1)
D(LOG_RGNP(-1))
C
@TREND(1)
-0.187035
0.200301
1.485080
0.006115
0.080371
0.139797
0.624874
0.002667
-2.327151
1.432802
2.376608
2.292939
0.0244
0.1587
0.0217
0.0265
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.122164
0.064913
0.019802
0.018037
127.2363
1.873090
0.032969
0.020478
-4.929452
-4.776490
2.133853
0.108850
t-Statistic
Prob.*
-6.253466
-4.152511
-3.502373
-3.180699
0.0000
Coefficien
t
Std. Error
t-Statistic
Prob.
D(LOG_RGNP(-1))
C
@TREND(1)
-0.883502
0.031018
-7.45E-05
0.141282
0.008188
0.000205
-6.253466
3.788298
-0.364325
0.0000
0.0004
0.7172
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.455256
0.432075
0.020711
0.020161
124.4538
1.884766
-0.000728
0.027483
-4.858151
-4.743429
19.63950
0.000001
Log_labours first unit root test shows that the series is not stationary as t-Stat is higher than 1% level as
shown in the table below
t-Statistic
Prob.*
-2.351205
-4.152511
-3.502373
-3.180699
0.3997
Coefficien
t
Std. Error
t-Statistic
Prob.
LOG_LABOUR(-1)
D(LOG_LABOUR(-1))
C
@TREND(1)
-0.175616
0.403214
0.737934
0.002923
0.074692
0.142172
0.308771
0.001267
-2.351205
2.836099
2.389910
2.307635
0.0231
0.0068
0.0210
0.0256
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.185441
0.132317
0.012075
0.006707
151.9695
1.894808
0.015320
0.012963
-5.918781
-5.765819
3.490748
0.022964
t-Statistic
Prob.*
-5.338968
-4.156734
-3.504330
-3.181826
0.0003
Coefficien
t
D(LOG_LABOUR(-1)) -0.912550
D(LOG_LABOUR(1),2)
0.271893
C
0.015468
@TREND(1)
-5.68E-05
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.408022
0.368557
0.012424
0.006946
147.5781
1.998336
Std. Error
t-Statistic
Prob.
0.170922
-5.338968
0.0000
0.142706
0.004900
0.000126
1.905264
3.156787
-0.449136
0.0631
0.0028
0.6555
-0.000103
0.015634
-5.860332
-5.705897
10.33878
0.000027
Log_capital unit root test level shows that the series is not stationary.
Null Hypothesis: LOG_CAPITAL has a unit root
Exogenous: Constant, Linear Trend
Lag Length: 2 (Automatic based on SIC, MAXLAG=10)
t-Statistic
Prob.*
-1.393600
-4.156734
-3.504330
-3.181826
0.8506
Coefficien
t
Std. Error
t-Statistic
Prob.
-0.017250
0.012378
-1.393600
0.1704
0.983780
0.135100
7.281849
0.0000
-0.395188
0.171097
0.000408
0.135135
0.111289
0.000384
-2.924396
1.537414
1.061537
0.0054
0.1314
0.2942
0.766794
0.745594
0.003402
0.000509
211.6012
1.874476
0.031118
0.006744
-8.432702
-8.239659
36.16863
0.000000
The 1st difference shown in the table below shows that the series is not stationary.
Null Hypothesis: D(LOG_CAPITAL) has a unit root
Exogenous: Constant, Linear Trend
Lag Length: 1 (Automatic based on SIC, MAXLAG=10)
t-Statistic
Prob.*
-3.963298
-4.156734
-3.504330
-3.181826
0.0165
Coefficien
t
D(LOG_CAPITAL(-1)) -0.411039
D(LOG_CAPITAL(1),2)
0.416242
C
0.016116
@TREND(1)
-0.000124
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.291777
0.244563
0.003437
0.000532
210.5430
1.863267
Std. Error
t-Statistic
Prob.
0.103711
-3.963298
0.0003
0.135686
0.004264
4.72E-05
3.067675
3.779808
-2.619319
0.0036
0.0005
0.0120
-0.000231
0.003954
-8.430325
-8.275891
6.179783
0.001312
Upon computing the 2nd difference we can see that the series is stationary.
Null Hypothesis: D(LOG_CAPITAL,2) has a unit root
Exogenous: Constant, Linear Trend
Lag Length: 1 (Automatic based on SIC, MAXLAG=10)
t-Statistic
Prob.*
-6.268911
-4.161144
-3.506374
-3.183002
0.0000
Coefficien
t
Std. Error
t-Statistic
Prob.
-1.097258
0.175032
-6.268911
0.0000
0.382424
-0.000183
-2.06E-06
0.139129
0.001195
3.88E-05
2.748694
-0.153546
-0.053089
0.0086
0.8787
0.9579
0.486542
0.451534
0.003720
0.000609
202.4977
2.104704
7.64E-05
0.005023
-8.270737
-8.114804
13.89784
0.000002
We have generated a regression with the logarithms and we can notice r-squared having almost a perfect
value (0.99 almost 1) and durbin Watson has a low value resulting in a spurious regression.
C(1)
C(2)
C(3)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Coefficien
t
Std. Error
t-Statistic
Prob.
-1.506665
0.867848
0.363643
0.140143
0.090569
0.168440
-10.75088
9.582127
2.158891
0.0000
0.0000
0.0358
0.996586
0.996447
0.029967
0.044003
110.1587
8.664846
0.502731
-4.121487
-4.008915
0.285118
The spurious regression resulted because of the fact that log_rgnp and log_labour variables are stationary
in 1st difference, but log_capital is not stationary in 1st difference.
C(1)
C(2)
C(3)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Coefficien
t
Std. Error
t-Statistic
Prob.
0.003213
1.144980
0.402547
0.009205
0.166767
0.318464
0.349031
6.865747
1.264027
0.7286
0.0000
0.2123
0.611766
0.595590
0.013287
0.008474
149.5484
0.033678
0.020894
-5.746996
-5.633360
1.856282
The regression is not spurious because R-squared is not perfect anymore and moreover Durbin Watson
has a decent value.
Value
3.897829
3.897829
df
Probability
(1, 48)
1
0.0541
0.0483
Value
Std. Err.
0.547526
0.277328
>1 increasing
<1 decreasing
=1 constant returns to scale
Now we check if there is a co-integration relationship between rgnp, labour and capital. We cant check
given the fact that the variables are not stationary in the same order, we cannot do the error correction
model.
There is no long term relationship between the variables.