Re: st: rollreg - r134

[Nick Cox <njcoxstata@gmail.com>]

Kit's answer was better. If getting rid of the gaps means working with
the longest possible unbroken spell in each panel, then that is a
method easy to explain.

Nick

On Thu, May 3, 2012 at 8:27 AM, Nick Cox <njcoxstata@gmail.com> wrote:
> I won't encourage or support your doing this. It's at best a fudge. I
> think you need to write a loop from first principles.
>
> Nick
>
> On Wed, May 2, 2012 at 10:15 PM, "Fabian Schönenberger" <sch.f@gmx.ch> wrote:
>> Oh, yes, rollreg from SSC.
>>
>> OK. So I have to get rid off the gaps. I have generated a new variable:
>> by cusip: generate datecontrol=datemonthly[_n]-datemonthly[_n-1]
>>
>> If the variable is larger than 1 there is a gap in the monthly time series. Instead of dropping a cusip completely if there is datecontrol>1, I would like to keep either the observations befor or after the gap depending on the length of the time series without a datecontrol>1. The longer the better.
>>
>> How do I have to proceed or is there a more elegant way?
>
> Von: Nick Cox <njcoxstata@gmail.com>
>>
>>> -rollreg- is from SSC. Please recall that you are asked to explain
>>> where user-written programs you refer to come from.
>>>
>>> -rollreg- itself does not ever issue error 134 so far as I can see.
>>>
>>> My guess is this.
>>>
>>> guess {
>>>
>>> You are running -rollreg-, which in turn runs -tsreport-, which in
>>> turn finds gaps. -tsreport- would use  -tabdisp- for display but you
>>> hit limits on how much it can display.
>>>
>>> But -rollreg- does not allow gaps any way. So what you are finding is
>>> that -rollreg- is failing to tell you that you cannot use it because
>>> your dataset triggers a system limit first.
>>>
>>> }
>>>
>>> Nick
>>>
>>> On Wed, May 2, 2012 at 8:56 PM, "Fabian Schönenberger" <sch.f@gmx.ch>
>>> wrote:
>>> > Dear all
>>> > I have a problem running rollreg regression for my panel data. The
>>> sample is very large, about 2'400'000 observations. First, I tsset the date:
>>> >
>>> > . tsset cusip datemonthly
>>> >       panel variable:  cusip (unbalanced)
>>> >        time variable:  datemonthly, 1969m2 to 2012m2, but with gaps
>>> >                delta:  1 month
>>> >
>>> > I want to conduct rolling regressions for my panel date. I tried:
>>> >
>>> > rollreg stockpremium marketpremium, move(36) stub(capm)
>>> >
>>> > The output is:
>>> > Observations with preceding time gapstoo many values
>>> > r(134);
>>> >
>>> > I did some internet research and found out that some years ago the same
>>> problem was discussed among Statalist. However, I did not find a solution
>>> for the problem. I am not sure if the problem are too many values, or the
>>> gaps. I dropped all missing observations (monthly stock prices). However, it
>>> is possible that for some cusips the sequence of months has gaps. The
>>> reason why I want to conduct rollreg is that I am interested in how the
>>> explanatory power of marketpremium changes over time.
>>> >

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/