preliminary ADF, PP, KPPS testing for time-series analysis
Posted: Tue Oct 02, 2018 8:37 pm
Hello,
I am running preliminary ADF, PP and KPPS testing for a set of variables (see attached).
Most variables are ratios, the rest I have logged.
I have the following questions
1) Variable NE had negative values (NE means the net exports of a country). Because it is not a ratio, I am logging the data sets. The lowest value was -77.78. Doing some research, I found that you can add 100+ to the lowest value to make all values positive and then Log. Is this the correct method?
2) Looking at the ADF, PP and KPSS testing, what can you conclude about the testing? (The results for ADF and PP are the P-values with 1% 5% and 10%).
My conclusion is the following:
ADF: There is no unit root for the 1st difference with only an intercept for all variables, with most at 1% significant.
PP: There is no unit root for the 1st difference with only an intercept for all variables, with most at 1% significant.
KPSS: There is no unit root for the 1st difference with only an intercept for all variables, with most at 10% significant.
3) would the next step to test the variables within the equations (not shown here) under co integration analysis through johansen co integration testing? From here, I am thinking granger causality testing, then VAR lag selection, then granger causality/block exgoneity wald test, then SVAR, then impulse response functions, then possibly VEC, then variance decomposition
I am running preliminary ADF, PP and KPPS testing for a set of variables (see attached).
Most variables are ratios, the rest I have logged.
I have the following questions
1) Variable NE had negative values (NE means the net exports of a country). Because it is not a ratio, I am logging the data sets. The lowest value was -77.78. Doing some research, I found that you can add 100+ to the lowest value to make all values positive and then Log. Is this the correct method?
2) Looking at the ADF, PP and KPSS testing, what can you conclude about the testing? (The results for ADF and PP are the P-values with 1% 5% and 10%).
My conclusion is the following:
ADF: There is no unit root for the 1st difference with only an intercept for all variables, with most at 1% significant.
PP: There is no unit root for the 1st difference with only an intercept for all variables, with most at 1% significant.
KPSS: There is no unit root for the 1st difference with only an intercept for all variables, with most at 10% significant.
3) would the next step to test the variables within the equations (not shown here) under co integration analysis through johansen co integration testing? From here, I am thinking granger causality testing, then VAR lag selection, then granger causality/block exgoneity wald test, then SVAR, then impulse response functions, then possibly VEC, then variance decomposition