Date: Wed, 16 Dec 2009 18:30:08 -0500
Reply-To: Francois van der Walt <francoisw@GJI.COM.AU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Francois van der Walt <francoisw@GJI.COM.AU>
Subject: Re: Trend matching
I have quickly read your sample data and plotted it using proc gplot. From
looking at the graphs of your sample data I can see that analyzing and
exploring for seasonality will not help you. I also do not think that trying
to combine it with similarly trended other products will give you any
If I had to answer this I would look at only the last 3 points to estimate
the trend. I clearly do not know the bigger picture like, how many regions
or what other considerations their is from the bigger project.
I am more than happy to show you the graphs and share more of my experience
working on similar data for many years.
On Wed, 16 Dec 2009 17:52:59 -0500, Sid N <nsid31@GMAIL.COM> wrote:
>Thank you for your reply. Yes, I am looking at processes that have declining
>trends. This is one of the approaches that I am taking as part of a larger
>forecasting project. As some of the regions do not have sufficient data
>points, I am not sure how useful traditional forecasting tools are in
>predicting future production.
>My approach in comparing production from different regions and then
>predicting production based on trends from matching regions also involves
>other similar characteristics that are shared by these regions. In the
>original example, based on the characteristics that these regions share, I
>could have matched trends from B with C, but A seemed a closer match based
>on the cumulative production.
>I will look at the SAS/ETS documentation to see if I could use any of the
>On Wed, 16 Dec 2009 17:30:24 -0500, Sigurd Hermansen <HERMANS1@WESTAT.COM>
>>Why are you supposing that future numbers in B will have the same trend as
>past numbers in A? It would help to know how you might use the results of
>your analysis. Are you looking at processes that deplete resources and
>thereafter have rapidly declining trends? I recall seeing Box-Jenkins
>estimates of cycles in time series and transfer functions that use cycle
>parameter estimates to predict cycles in other series. For that see
>documentation on-line for the SAS/ETS product.
>>From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Sid N
>>Sent: Wednesday, December 16, 2009 5:11 PM
>>Subject: Trend matching
>>I am looking for appropriate analysis steps or procedures that will enable
>>me to match production trends based on cumulative production. Below dataset
>>has historical production reported from multiple regions.
>>input region $ year prod;
>>A 2000 100
>>A 2001 200
>>A 2002 250
>>A 2003 275
>>A 2004 450
>>A 2005 500
>>A 2006 300
>>A 2007 150
>>A 2008 75
>>A 2009 25
>>B 2004 125
>>B 2005 200
>>B 2006 250
>>B 2007 300
>>B 2008 425
>>C 2000 150
>>C 2001 450
>>C 2002 500
>>C 2003 550
>>C 2004 900
>>C 2005 1000
>>C 2006 750
>>C 2007 600
>>D 2006 125
>>D 2007 500
>>D 2008 550
>>D 2009 600
>>If the production history of a region spans <= 5 years (as in B and D) then
>>the production trend is matched with trends from other regions which have
>>longer production history (based on cumulative production). From the above
>>dataset, B (cumulative production = 1300) is matched with A (1275 over 5
>>years) and D (1775) with C (1650 over 4 years). Assuming the same future
>>trend as A, production from B is predicted for the remaining number of
>>years. That is, production from region B is predicted for 5 more years.
>>Likewise, future production from D is predicted based on production from C.
>>In this case, production from D is predicted for 4 more years.
>>The example above is only a subset of the actual data. Therefore, trend
>>matching in my actual dataset may need to be done by comparing data from a
>>larger number of regions. I am looking for an efficient way to accomplish
>>this. Any suggestions?
>>Thank you in advance.