[Gretl-devel] Re: kpss observations/issues
On Thu, 30 Sep 2021, Sven Schreiber wrote:
sorry for reporting a bunch of quirks so close to the release,
it's not
intentional.
Anyway, in actual use of the 'kpss' command I'm stumbling over the
following things:
- (extremely minor) The doc could mention that the usual $test and (if
applicable) $pvalue accessors can be used after running the test command.
- The $test accessor contains NA when running in a panel context.
(Example: 'kpss 3 invest' on the grunfeld dataset, and then look at
$test.) OK, the printout reveals three different test statistics, so
perhaps it would have to become a matrix instead of a scalar.
Perhaps. We can talk about that.
- Using automatic bandwidth choice in a panel yields glitch in the
output: For example, doing 'kpss -1 invest' runs fine, and according to
the doc it triggers automatic lag order (window size) choice. However,
in the printout the "-1" is repeated verbatim, whereas in a plain
time-series context the actually chosen value is displayed.
Printing -1 is not good, certainly. It's potentially complicated in
the panel case since it's quite possible that different units end up
with different lag orders due to different T_i, when the automatic
bandwidth option is chosen. But the output is now improved in git.
Allin