Date:         Thu, 9 Oct 2008 13:38:17 -0400
Reply-To:     Sigurd Hermansen <HERMANS1@WESTAT.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Sigurd Hermansen <HERMANS1@WESTAT.COM>
Subject:      Re: Seasonal Adjustment of Weekly Time Series
Comments: To: "brunson@TANGLEDBANK.COM" <brunson@TANGLEDBANK.COM>
In-Reply-To:  <20081009171440.9B5E2200CC@ws6-7.us4.outblaze.com>
Content-Type: text/plain; charset="us-ascii"

Kevin:
Why not define a "fiscal year" from the first Saturday of a year through the Friday following the 52nd Saturday? SAS date functions will support blocks of weeks.
S

-----Original Message-----
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of brunson@TANGLEDBANK.COM
Sent: Thursday, October 09, 2008 1:15 PM
To: SAS-L@LISTSERV.UGA.EDU
Subject: Seasonal Adjustment of Weekly Time Series


This is a brief description of the problem, but I hope not too brief and sufficient for anyone who has encountered the problem.

The problem: panel data measured at weekly intervals periodically has extra days at year end which drift into the following year.  We have dozens of explanatory models with hundreds of variables modeling the business process. Saturday is defined as the beginning of a week; thus, eventually the entire last week of a year drifts into the following year.

There are various simple suggestions on the table: the first Saturday of a year is always week 1 and the leftover days from the previous year are added to week 52, average the the last X days for week 52, ignore the leftover days.

I know this isn't a unique problem but I haven't found a sound solution for my setting.  I found papers describing trigonometric transformations (e.g., Cleveland and Scott), but if I understand the process it would have to applied to each of the variables and that's a major undertaking.

Any help is appreciated, even pointing me in a direction to research.

Kevin