Date:         Wed, 16 Jan 2008 23:14:29 -0500
Reply-To:     "Howard Schreier <hs AT dc-sug DOT org>" <nospam@HOWLES.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "Howard Schreier <hs AT dc-sug DOT org>" <nospam@HOWLES.COM>
Subject:      Re: SAS rolling regression

On Wed, 16 Jan 2008 20:57:55 -0600, data _null_, <datanull@GMAIL.COM> wrote:

>Yes, I see the vague reference to ID and WEEK-ID.  You are correct as
>usual.  You will have to show me how to do the self-join.

Maybe

   create view rolling10 as
    select   rolls.companyID
           , rolls.weekID as rollid
           , windows.weekID
           , windows.y
     from weeks as rolls join weeks as windows
      on rolls.companyID = windows.companyID
     where rollid LE windows.weekID LE rollid + 9
     group by rolls.companyID, rollid
     having count(*) = 10
     order by rolls.companyID, rollid, weekID
   ;

Then the BY statement for PROC REG is

   by companyID rollid;

>
>On Jan 16, 2008 7:01 PM, Howard Schreier <hs AT dc-sug DOT org>
><nospam@howles.com> wrote:
>>
>> On Wed, 16 Jan 2008 18:24:18 -0600, data _null_, <datanull@GMAIL.COM> wrote:
>>
>> >This may be close to what you want.
>> >
>> >proc plan seed=2021472219;
>> >   factors
>> >      weekID = 105 ordered
>> >      y = 1 of 50 random / noprint;
>> >   output out=work.weeks;
>> >   run;
>> >data work.rolling10;
>> >   set work.weeks(rename=(weekID=rollID));
>> >   if _n_+9 lt nobs then do point = _n_ to _n_+9;
>> >      set work.weeks point=point nobs=nobs;
>> >      output;
>> >      end;
>> >   run;
>> >proc reg noprint outest=est;
>> >   by rollid;
>> >   model y = weekID;
>> >   run;
>> >proc print;
>> >   run;
>> >
>> >On Jan 16, 2008 5:23 PM, Dave Smith <daveyboysmith_123@yahoo.com> wrote:
>> >> hello people
>> >>
>> >> I am a newbie to SAS .. I tried searching the SAS archives but
couldnot find
>> >> what I am looking for,
>> >>
>> >> I need to run rolling regression in SAS, I have the following data
>> >>
>> >> id week-id price(response) x1 x2
>> >>
>> >> I need to run regression by id. week-id is an indicator of week number
from
>> >> 1 to 105 ,i need to run rolling regression using 10 week periods. This
means
>> >> for id 1 I run regression for week_id 1 to 10 regression and then
regression
>> >> on observation 2 to 11 etc. and then I do the same for id 2 etc
>> >>
>> >> I read on SAS listserv that using macro is not good as the calculations
>> >> become too many,
>> >>
>> >> Thank you for any help
>> >>
>> >> Dave Smith
>> >>
>>
>> I think the input is more like
>>
>>   proc plan seed=2021472219;
>>   factors
>>      companyID = 3 ordered
>>      weekID = 105 ordered
>>      y = 1 of 50 random / noprint;
>>   output out=work.weeks;
>>   run;
>>
>> So using the NOBS= result as a cutoff won't quite do it. I would probably
>> use a self-join in PROC SQL to generate the expanded data set or view.
>>