>>For a bivariate case, if the trace test rejects c=0 and does not
>>reject c=1, I report c=1. If it is the other way around, then I report
>>c=0 as the test result.

>>As you know, sometimes the results can be contradictory so that c=0
>>and c=1 are rejected (or not rejected) simultaneously.

>>My question is that would it be OK to report "inconclusive" in those
>>cases? Or am I expected to follow another further procedure?

Three things to think about (might make you recognize that your case
is more common than many realize):

1.You are searching for results using different lag numbers and the
null hypothesis probabilities are based on knowing the right lag
number beforehand. Of course no one knows the right lag number in a
real study (only known in constructed data cases), but once you
perform a search you should be willing to be skeptical of the test
statistic probabilities.

2. Not rejecting both c=0 and =1 is not an anomaly if you understand
you are only computing the odds each hypothesis is incorrect. You are
not computing the odds of 'correctness' given the results from the
other test. Failing to reject at the 5% level is just that-- failure
to say an hypothesis is blatantly wrong (+ recognizing point 1 that
the 5% number may be misleading).

3. Deciding c=1 vs c=0, i.e. testing whether two time series need to
be differenced or do not need to be differenced to create a stationary
cointegrating relationship is often not as interesting or
controversial as researchers believe it is. Assuming you are only
constructing the co-integrating vector for modeling purposes and this
is a first step, you might find similar results either way. Or once
one understands the data and its source you might conclude c=0 or c=1
is implausible. For example, one might reject c=0 and accept c=1 (or
vice-versa) when testing whether the UK and the US$ price levels are
cointegrated, when the true answer depends on how accurately the price
levels are computed. (I.e. I'd argue they must be differenced at
least once to control for measurement differences before seeing any
Johansen tests.)

(this post might be considered a test of whether an econometric
methodology list connected to GRETL would be worthwhile or fills a
need).
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users