I ran the code for the Hatemi J. cointegration and this is what I got;
******** Modified ADF Test ***********
t-statistic = -5.9745181
AR lag = 11.000000
first break point(ADF) = 0.43055556
second break point(ADF) = 0.52083333
******** Modified Phillips Test ********
Zt = -5.967
First breakpoint(Zt) = 0.160
Second breakpoint(Zt) = 0.264
Za = -59.611
First breakpoint(Za) = 0.153
Second breakpoint(Za) = 0.271
if model==3;
x1=ones(n,1)~dummy1~dummy2~seqa(1,1,n)~x;
elseif model==4;
x1=ones(n,1)~dummy1~dummy2~(dummy1).*x~(dummy2).*x;
elseif model==2;
x1=ones(n,1)~dummy1~dummy2~x;b,se,t 23.753 3.038 7.818 -16.648 34.761 -0.479 3.589 34.637 0.104 0.000 0.000 -7.798 0.102 0.038 2.683 -3.062 0.672 -4.561 -0.090 4.423 -0.020 -0.109 0.074 -1.469 3.795 3.371 1.126 0.097 4.428 0.022 0.329 0.134 2.459 -1.064 3.312 -0.321
Please how do I decipher which t-statistic is significant and at what level?
I am new at this. Thanks
4 Answers
0
Hello,
The t-statistic should be compared to the critical values published in the original Hatemi-J paper, "Tests for cointegration with two unknown regime shifts with an application to financial market integration". These critical values depend on the number of independent variables included in the regression.
Alternatively, if you use the Hatemi-J procedure available in the tspdlib library, the critical values for the $Z_t$, $Z_a$, and the $ADF$ statistic are returned with the coint_hatemiJ procedure.
The file coint_hatemiJ.e is included in the tspdlib library and can be used as a template for running the test with your own data.
Note that the tspdlib library is available for free and can be downloaded and installed directly from within GAUSS using the GAUSS Package Manager.
0
Thanks for your reply. However I tried running the analysis with The file coint_hatemiJ.e you included and am getting an error message like this ; datafile.src not seen or something. Is there something am not doing well?
0
I recommend working with the Hatemi-J procedure from the tspdlib library. The code has been optimized to make it more user friendly and provides the critical values that you need to interpret the results.
However, it is difficult to diagnose errors without more specific details about the error. Please email me directly and I will help you address these errors.
In your email, please attach the program file you ran when you got the error message and a screenshot of the error message.
In the meantime, you may also want to review our GAUSS Video Blog, “Running Publicly Available Code.”
0
How to interpret ? "break point = 0,256 or 0,564"
obs: 39 var: 3 and results:
******** Modified ADF Test ***********
t-statistic = -5.142
AR lag = 7.000
first break point(ADF) = 0.179
second break point(ADF) = 0.333
******** Modified Phillips Test ********
Zt = -5.784
First breakpoint(Zt) = 0.256
Second breakpoint(Zt) = 0.564
Za = -36.809
First breakpoint(Za) = 0.256
Second breakpoint(Za) = 0.564
if model==3;
x1=ones(n,1)~dummy1~dummy2~seqa(1,1,n)~x;
elseif model==4;
x1=ones(n,1)~dummy1~dummy2~(dummy1).*x~(dummy2).*x;
elseif model==2;
x1=ones(n,1)~dummy1~dummy2~x;
b,se,t
-19202.173 23659.483 -0.812
22457.477 26826.491 0.837
3348.339 19597.128 0.171
0.249 0.194 1.283
1.028 0.334 3.075
-0.093 0.218 -0.426
-0.891 0.438 -2.035
-0.218 0.180 -1.207
1.031 0.377 2.737
Your Answer
4 Answers
Hello,
The t-statistic should be compared to the critical values published in the original Hatemi-J paper, "Tests for cointegration with two unknown regime shifts with an application to financial market integration". These critical values depend on the number of independent variables included in the regression.
Alternatively, if you use the Hatemi-J procedure available in the tspdlib library, the critical values for the $Z_t$, $Z_a$, and the $ADF$ statistic are returned with the coint_hatemiJ procedure.
The file coint_hatemiJ.e is included in the tspdlib library and can be used as a template for running the test with your own data.
Note that the tspdlib library is available for free and can be downloaded and installed directly from within GAUSS using the GAUSS Package Manager.
Thanks for your reply. However I tried running the analysis with The file coint_hatemiJ.e you included and am getting an error message like this ; datafile.src not seen or something. Is there something am not doing well?
I recommend working with the Hatemi-J procedure from the tspdlib library. The code has been optimized to make it more user friendly and provides the critical values that you need to interpret the results.
However, it is difficult to diagnose errors without more specific details about the error. Please email me directly and I will help you address these errors.
In your email, please attach the program file you ran when you got the error message and a screenshot of the error message.
In the meantime, you may also want to review our GAUSS Video Blog, “Running Publicly Available Code.”
How to interpret ? "break point = 0,256 or 0,564"
obs: 39 var: 3 and results:
******** Modified ADF Test ***********
t-statistic = -5.142
AR lag = 7.000
first break point(ADF) = 0.179
second break point(ADF) = 0.333
******** Modified Phillips Test ********
Zt = -5.784
First breakpoint(Zt) = 0.256
Second breakpoint(Zt) = 0.564
Za = -36.809
First breakpoint(Za) = 0.256
Second breakpoint(Za) = 0.564
if model==3;
x1=ones(n,1)~dummy1~dummy2~seqa(1,1,n)~x;
elseif model==4;
x1=ones(n,1)~dummy1~dummy2~(dummy1).*x~(dummy2).*x;
elseif model==2;
x1=ones(n,1)~dummy1~dummy2~x;
b,se,t
-19202.173 23659.483 -0.812
22457.477 26826.491 0.837
3348.339 19597.128 0.171
0.249 0.194 1.283
1.028 0.334 3.075
-0.093 0.218 -0.426
-0.891 0.438 -2.035
-0.218 0.180 -1.207
1.031 0.377 2.737