Learn how easy it is to sync an existing GitHub or Google Code repo to a SourceForge project! See Demo

Close

#12 Crash with lmtest after tsls

closed-fixed
None
7
2007-12-10
2007-11-29
Sven S.
No

The following fairly minimal script crashes gretl every time (1.7.0 on windows):

open denmark
tsls IDE 0 LRY ; 0 LRM
lmtest 4 --arch --white

-sven

Discussion

  • Logged In: YES
    user_id=1368575
    Originator: NO

    This problem is tied to the feature that enables you to specify more than one option in one go for the lmtest command.

    The crash happens because White's heteroskedasticity test is not (yet) available for models estimated via tsls.

    Of course this must be fixed, but I'm uncertain on what the proper way to do it is. If a test statistic is not available, like in this case, we have several options:

    a) just skip it as if it wasn't there
    b) flag an error and stop execution
    c) bypass the problem by disallowing multiple options to the lmtest command.

    Option (c) may seem drastic, but IMO is the most sensible, because the lmtest also produces the $test and $pvalue accessors, and it seems a bit silly to return just the last ones, as we do today.

     
  • Sven S.
    Sven S.
    2007-11-30

    Logged In: YES
    user_id=1330711
    Originator: YES

    Jack,

    (c) would be ok for me, I don't see much of a problem because of the small redundance. Besides, usually I want a different lag order for autocorrelation tests and for Arch tests.

    However, on second thought it occurs to me: (c) would break existing scripts, so there's a backward-compatibility problem. This leads me to maybe suggest (d), which also affects other cases without multiple options: flag a *warning*, not an error. I don't see why a script must be halted when something isn't implemented, gretl could just print out "sorry, not available" and proceed.

    -sven

     
  • Allin Cottrell
    Allin Cottrell
    2007-12-08

    Logged In: YES
    user_id=330339
    Originator: NO

    I tend to agree with Jack: allowing two or more tests under a single "lmtest" command
    seemed like a good idea at the time, but now I reckon it's confusing and potentially
    error-prone.

    Current CVS enforces only one test per "lmtest" and prevents the above crash.

     
  • Allin Cottrell
    Allin Cottrell
    2007-12-08

    • assigned_to: nobody --> allin
    • status: open --> open-remind
     
  • Sven S.
    Sven S.
    2007-12-09

    Logged In: YES
    user_id=1330711
    Originator: YES

    So if the crash is gone I don't understand why the bug is still open.

    Is testing needed?

    thanks,
    sven

     
  • Allin Cottrell
    Allin Cottrell
    2007-12-10

    • status: open-remind --> closed-fixed