Re: [Gretl-devel] coint2 and warnings
On Sat, 12 Jan 2013, Andreas Noack Jensen wrote:
When I extract the p-values from coint2 with restricted exogenous
variables
I receive a warning and a column of nans
? coint2 4 endo; d_IAU; IAU --crt --silent
? eval $pvalue
0,021269 nan
0,88179 nan
0,68159 nan
Warning: generated non-finite values
Is this intended? Can I suppress the warning?
The documentation, and probably the output, could do with some
improvement here.
Here's the current situation. The $pvalue accessor is set up
to provide an n x 2 matrix, where n is the number of
potentially cointegrated series. The first column contains
p-values for the trace test and the second contains p-values
for the lambda-max test. In the case of "partial systems",
where the analysis is conditional on one or more I(1)
variables, we can use the method of Harbo, Johansen, Nielsen
and Rahbek to get asymptotic p-values for the trace test but
so far as I'm aware there isn't a standard method for
obtaining p-values for the lambda-max test in this case. In
the printed output we show lambda-max p-values computed
"ignoring exogenous variables" but in the $pvalue matrix we
fill the second column with NaNs.
I can see two possible alternatives to current practice. We
could include in the $pvalue matrix the ("wrong") lambda-max
p-values that we print; or we could drop the second column for
the partial system case. Either way, of course, we should
clearly document what we're doing.
Allin Cottrell