The FORECAST Procedure

Example 16.3 Forecasting Petroleum Sales

This example uses the double exponential smoothing method to forecast the monthly U. S. sales of petroleum and related products series (PETROL) from the data set SASHELP.USECON. These data are taken from Business Statistics, published by the U.S. Bureau of Economic Analysis.

The following statements plot the PETROL series: 

title1 "Sales of Petroleum and Related Products";
proc sgplot data=sashelp.usecon;
   series x=date y=petrol / markers;
   xaxis values=('1jan80'd to '1jan92'd by year);
   yaxis values=(8000 to 20000 by 1000);
   format date year4.;
run;

The plot is shown in Output 16.3.1.

Output 16.3.1: Sales of Petroleum and Related Products

Sales of Petroleum and Related Products


The following statements produce the forecast: 

proc forecast data=sashelp.usecon interval=month
              method=expo trend=2 lead=12
              out=out outfull outest=est;
   id date;
   var petrol;
   where date >= '1jan80'd;
run;

The following statements print the OUTEST= data set: 

title2 'OUTEST= Data Set: EXPO Method';
proc print data=est;
run;

The PROC PRINT listing of the output data set is shown in Output 16.3.2.

Output 16.3.2: The OUTEST= Data Set Produced by PROC FORECAST

Sales of Petroleum and Related Products
OUTEST= Data Set: EXPO Method

Obs _TYPE_ DATE PETROL
1 N DEC91 144
2 NRESID DEC91 144
3 DF DEC91 142
4 WEIGHT DEC91 0.1055728
5 S1 DEC91 14165.259
6 S2 DEC91 13933.435
7 SIGMA DEC91 1281.0945
8 CONSTANT DEC91 14397.084
9 LINEAR DEC91 27.363164
10 SST DEC91 1.17001E9
11 SSE DEC91 233050838
12 MSE DEC91 1641203.1
13 RMSE DEC91 1281.0945
14 MAPE DEC91 6.5514467
15 MPE DEC91 -0.147168
16 MAE DEC91 891.04243
17 ME DEC91 8.2148584
18 RSQUARE DEC91 0.8008122 




            

            


            
The plot of the forecast is shown in Output 16.3.3. 
            


title1 "Sales of Petroleum and Related Products";
title2 'Plot of Forecast: EXPO Method';
proc sgplot data=out;
   series x=date y=petrol / group=_type_;
   xaxis values=('1jan89'd to '1jan93'd by qtr);
   yaxis values=(10000 to 20000 by 1000);
   refline '15dec91'd / axis=x;
run;



  

            


Output 16.3.3: Forecast of Petroleum and Related Products

               

                  
Forecast of Petroleum and Related Products