turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Forecasting
- /
- Winters Method -- Additive (theory and/or syntax q...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-01-2009 01:47 PM

I'm attempting to self-study time series analysis and the SAS/ETS system. Using Proc ARIMA, I think I've established some decent methods for identifying/evaluating trend, seasonality, stationarity, and potential models.

However, today I walked through the process of fitting models automatically using the Time Series Forecasting System (point & click), and a "Winters Method -- Additive" model was suggested as the best. When I select to "Show Source Statements" nothing is shown for this model, so I'm not sure how to reproduce it programmatically (except in Proc Forecast, but I don't see how to evaluate model fit with this procedure).

My question is, how might I have identified "Winters Method -- Additive" as a candidate model, and compared with other candidate models, programmatically? I realize this may be more of a theoretical question than syntax, but either form of insight would be welcome. I'd prefer to understand the process than to rely on an automatic procedure. Thanks!

****** (update) ******

I want to add, I did read that "The ARIMA model equivalency to the additive version of Winters method is the ARIMA(0,1,p+1)(0,1,0)p model." When I tried the following code,

identify var=sales(1,12);

estimate q=13;

I get the error message that there is not enough data to fit the model. Am I misinterpreting (p+1)? Thanks again!

Message was edited by: sassygrl

However, today I walked through the process of fitting models automatically using the Time Series Forecasting System (point & click), and a "Winters Method -- Additive" model was suggested as the best. When I select to "Show Source Statements" nothing is shown for this model, so I'm not sure how to reproduce it programmatically (except in Proc Forecast, but I don't see how to evaluate model fit with this procedure).

My question is, how might I have identified "Winters Method -- Additive" as a candidate model, and compared with other candidate models, programmatically? I realize this may be more of a theoretical question than syntax, but either form of insight would be welcome. I'd prefer to understand the process than to rely on an automatic procedure. Thanks!

****** (update) ******

I want to add, I did read that "The ARIMA model equivalency to the additive version of Winters method is the ARIMA(0,1,p+1)(0,1,0)p model." When I tried the following code,

identify var=sales(1,12);

estimate q=13;

I get the error message that there is not enough data to fit the model. Am I misinterpreting (p+1)? Thanks again!

Message was edited by: sassygrl

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-04-2009 06:33 AM

Hello -

Although the Time Series Forecasting System (TSFS) uses PROC ARIMA behind the scenes to generate the parameter estimates and forecasts for the ARIMA models, the TSFS uses SCL and C routines to generate the parameter estimates and forecasts for the exponential smoothing models. There is no procedure in the SAS/ETS product in SAS 9.1.3 which can be used to reproduce the forecasts for the exponential smoothing models generated by the TSFS. Although PROC FORECAST can be used to fit a class of exponential smoothing models, this procedure does not optimize the smoothing weights and it uses a different start-up algorithm than the one used in TSFS.

In SAS 9.2, we have a new procedure in the SAS/ETS product called PROC ESM which can be used to reproduce or come very close to the results generated for the exponential smoothing models fit and forecasted by the Time Series Forecasting System.

Link to the SAS 9.2 documentation on PROC ESM:

http://support.sas.com/documentation/cdl/en/etsug/60372/HTML/default/esm_toc.htm

Hope that helps,

Udo

Although the Time Series Forecasting System (TSFS) uses PROC ARIMA behind the scenes to generate the parameter estimates and forecasts for the ARIMA models, the TSFS uses SCL and C routines to generate the parameter estimates and forecasts for the exponential smoothing models. There is no procedure in the SAS/ETS product in SAS 9.1.3 which can be used to reproduce the forecasts for the exponential smoothing models generated by the TSFS. Although PROC FORECAST can be used to fit a class of exponential smoothing models, this procedure does not optimize the smoothing weights and it uses a different start-up algorithm than the one used in TSFS.

In SAS 9.2, we have a new procedure in the SAS/ETS product called PROC ESM which can be used to reproduce or come very close to the results generated for the exponential smoothing models fit and forecasted by the Time Series Forecasting System.

Link to the SAS 9.2 documentation on PROC ESM:

http://support.sas.com/documentation/cdl/en/etsug/60372/HTML/default/esm_toc.htm

Hope that helps,

Udo

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-04-2009 08:57 AM

Wonderful insight. Thank you very much!!