All,
I am running a Johansen test to test for co-integration and do not find an statistically significant result in either the max eigenvalue or the trace statistic. Not wanting to let it go, both the SBIC and HQIC report minimum values at I(1). As I was building the table, I thought about having to defend this decision and came up with the following argument:
The trace statistic and max eigenvalue tests show the minimum number of equations that could exist for co-integration, which means that there are could be more, but we accept the lowest number of equations as signaled by either statistic. Not having a result here does not mean there isn't a co-integration equation, but it is common to find. ( At least the test did not report I(0), right?)
So then we can use an information criterion to identify the number of co-integration equations, which are the same in this case for both the SBIC and the HQIC. A discussion on the matter with my econometrician led to theoretical tests for proving which tests are best and being able to rationalize the choice.
I am left with reporting all and noting the information criterion as my selections for I(1).
any thoughts? Is it ok to use Information Criteria in lieu of trace statistics and eigenvalues? Should I even report them?
thanks, been thinking about this for a while,
vikingo
FYI: I'm using Stata 12 on futures and spot commodity prices
I am running a Johansen test to test for co-integration and do not find an statistically significant result in either the max eigenvalue or the trace statistic. Not wanting to let it go, both the SBIC and HQIC report minimum values at I(1). As I was building the table, I thought about having to defend this decision and came up with the following argument:
The trace statistic and max eigenvalue tests show the minimum number of equations that could exist for co-integration, which means that there are could be more, but we accept the lowest number of equations as signaled by either statistic. Not having a result here does not mean there isn't a co-integration equation, but it is common to find. ( At least the test did not report I(0), right?)
So then we can use an information criterion to identify the number of co-integration equations, which are the same in this case for both the SBIC and the HQIC. A discussion on the matter with my econometrician led to theoretical tests for proving which tests are best and being able to rationalize the choice.
I am left with reporting all and noting the information criterion as my selections for I(1).
any thoughts? Is it ok to use Information Criteria in lieu of trace statistics and eigenvalues? Should I even report them?
thanks, been thinking about this for a while,
vikingo
FYI: I'm using Stata 12 on futures and spot commodity prices