SAS Exponential Smoothing

SAS Exponential Smoothing:
Data Configuration – Simple & Double Exponential Smoothing.

SAS Exponential Smoothing: Data Configuration – Seasonal/Tripple Exponential Smoothing

Select the data set, the Dependent Variable, the Time ID (which should be unique), and the season length (how many seasons, in our case it is quarters, so 4):

SAS does not provide the configuration of Alpha in PROC ESM, which is the main procedure for “modeling and forecasting” task, i.e. exponential smoothing (and ARIMA). It does enable it though in PROC FORECAST (for time decomposition), and it names alpha as “Weight” and it has a configuration for Exponential Smoothing (METHOD=EXPO) but it is not in the default GUI wizard for forecasting.

From the Model Tab > Choose Forecasting model Type: Exponential Smoothing

Then choose the exact forecasting model:
For SES: Choose Simple Exponential Smoothing
For State-Space: choose Linear.
For Double: Choose Double/Brown.
for seasonal: Choose Seasonal Multiplicative or additive.
For Trippple (Level-Trend-Seasonal): Choose Winter’s Multiplicative or additive.

Last: Choose the confidence level, the periods to forecsat and the periods to leave out (out-sample):

Hit on Run (the walking person’s logo):

Results:
SES: Simple Exponential Smoothing: SES is unable to forecast properly after some time:

Seasonal Smoothing (no Trend):

We can see that the forecasting (after the dashed line) isn’t taking in consideration the escalating trend.

Now we apply the Winter Additive Exponential Smoothing:

Leave out: 3. Forecast 12 ahead

Winters Exponential Smoothing

Winters Exponential Smoothing

Hold Back (a.k.a. Leave out, Out-Sample): 4. Forecast 12 ahead

 

Winters Exponential Smoothing

Winters Exponential Smoothing

Leave a Reply

Your email address will not be published. Required fields are marked *