count model

[maxchen]

EViews V7.1 b2010-12-02

Code: Select all

%ex7 = "\Example Files\EV7 Manual Data\" %ch26 = "Chapter 26 - Discrete and Limited Dependent Variable Models" %wf = @evpath + %ex7 +%ch26 +"\strike.wf1" wfopen %wf c = 0 equation eq13.count(d=b,showopts) numb = c(1) +c(2)*ip +c(3)*feb 'by expression, here EViews use its own starting value equation eq33.count(d=b,v=1) numb c ip feb equation eq36.count(d=b,v=1,h) numb c ip feb 'diff from eq33

my questions are:
1 according to http://forums.eviews.com/viewtopic.php?f=9&t=3351, "All estimation using expression in EViews uses the workfile C". However, eq13 does not uses the workfile C
2 the user manual II p291 say, "For the negative binomial QMLE, EViews by default reports the robust QMLE standard errors". However, the std errors in eq33 and eq36 are different

[EViews Glenn]

1. I guess my off-the cuff forum comment about starting values wasn't correct for the count models. I had forgotten that we had added the expression form for these models recently.

2. The model you are estimating isn't the QML, it's the ML. You need to set a specific value for the QML parameter using the v parameter.

[maxchen]

there do have "v=1" in eq33 and eq36, or eq33 use other form of QML standard errors (not the Huber/White method)?

my summary: for negative binomial model
1 if v=num is given, the variance is fixed, thus not estimated by EViews. Otherwise estimated and reported as "SHAPE"
2 to QML, whether v=num is specified or not, option "h" should be specified to compute QML std errors.