Hi, I am a student using Stata 13 trying to do a cointegration test on exchange rate index and CPI index.
I have been having some problems with the Johansen procedure. When I try it, using different specifications, generally the trace and maximum eigenvalue test indicate at least one cointegrating relationship. However, after imposing constraints, and doing the vec command [vec ls1 lp1 lp2, lags(3) trend(rconstant) rank(1) bconstraints(1) alpha], I have a problem, it goes up to 16,000 iterations, and says ‘convergence not achieved’, ‘beta overidentified’, and the speed of adjustment parameter has the same sign as the corresponding coefficient.
Well, I went back to do the Augmented Dickey-Fuller tests on the different series [I skipped this part, assuming all were I(1), because I just wanted to learn the varsoc, vecrank, vec commands etc.]- so, when doing the ADF test, yes some were pure random walk and therefore I(1), however the price index series were all of the form random walk + drift + deterministic trend, and as I understand differencing this alone won’t make it stationary, I also have to de-trend the first difference, so are these series still I(1)? Is this main problem I have, the main cause of the results above, that there is a trend, or could it be something else? I would be most grateful for any sort of help or advice you can give, on how to proceed, since I have asked and around and read the books but I didn’t know where else to turn too.
Thank you for reading.
I have been having some problems with the Johansen procedure. When I try it, using different specifications, generally the trace and maximum eigenvalue test indicate at least one cointegrating relationship. However, after imposing constraints, and doing the vec command [vec ls1 lp1 lp2, lags(3) trend(rconstant) rank(1) bconstraints(1) alpha], I have a problem, it goes up to 16,000 iterations, and says ‘convergence not achieved’, ‘beta overidentified’, and the speed of adjustment parameter has the same sign as the corresponding coefficient.
Well, I went back to do the Augmented Dickey-Fuller tests on the different series [I skipped this part, assuming all were I(1), because I just wanted to learn the varsoc, vecrank, vec commands etc.]- so, when doing the ADF test, yes some were pure random walk and therefore I(1), however the price index series were all of the form random walk + drift + deterministic trend, and as I understand differencing this alone won’t make it stationary, I also have to de-trend the first difference, so are these series still I(1)? Is this main problem I have, the main cause of the results above, that there is a trend, or could it be something else? I would be most grateful for any sort of help or advice you can give, on how to proceed, since I have asked and around and read the books but I didn’t know where else to turn too.
Thank you for reading.
Comment