For the ADF test, monthly data, in some cases lag choices of 12, 8, or 4 give a p-value of less than 0.10, but when inspecting the correlogram the residuals do not behave well. in this case, i went up to jointly test lags 13-16, and got p-value less than 0.10, and got well-behaved white-noise, so i used 16 lags for ADF test- was this the right thing to do?
I also encounter a similar problem sometimes for the Johansen procedure, with the varsoc, in lectures we do varsoc with max(10) lags, but when reviewing if the residuals are well-behaved-they are not well behaved, even with 9, or 10 lags as recommended by information criteria. therefore, i did varsoc again, with maximum 15 lags, and found that with 13 legs, recommended by information criteria, the residuals are well-behaved in the var estimate.- in cases like these, for ADF and johansen, is this how to approach the problem, to increase lag length until residuals are well-behaved, or should i have done something different?
I also encounter a similar problem sometimes for the Johansen procedure, with the varsoc, in lectures we do varsoc with max(10) lags, but when reviewing if the residuals are well-behaved-they are not well behaved, even with 9, or 10 lags as recommended by information criteria. therefore, i did varsoc again, with maximum 15 lags, and found that with 13 legs, recommended by information criteria, the residuals are well-behaved in the var estimate.- in cases like these, for ADF and johansen, is this how to approach the problem, to increase lag length until residuals are well-behaved, or should i have done something different?
Comment