Date:         Mon, 1 Dec 1997 15:22:48 +0100
Reply-To:     "Vandebroek, Martina" <Martina.Vandebroek@ECON.KULEUVEN.AC.BE>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         "Vandebroek, Martina" <Martina.Vandebroek@ECON.KULEUVEN.AC.BE>
Subject:      a question about the sas-procedure forecast

I try to use "exponential smoothing" to generate forecasts but I get
strange answers from SAS:
something seems to be wrong with the first smoothed value ...
 
e.g. I have the following timeseries:
Y1 = 103.5
Y2 =   94.7
Y3 = 118.6
Y4 = 109.3
Y5 = 126.1
....
 
the exponentially smoothed series with weight 0.7 should be
E1 = Y1 = 103.5
E2 = 0.7*Y1 + 0.3*E1 = 103.5
E3 = 0.7*Y2 + 0.3*E2 =   97.34
...
 
but with the statements:
        proc forecast data = timeser out = timeout2 outactual out1step
        method = expo trend = 1 weight = 0.7 ;
I get the following results
            1    JAN90    ACTUAL         0      103.500   (=Y1)
            2    JAN90    FORECAST       0      117.625   (=E1 = ?????)
            3    APR90    ACTUAL         0       94.700   (=Y2)
            4    APR90    FORECAST       0      107.738   (=E2 =!=
0.7*Y1+0.3*E1)
            5    JUL90    ACTUAL         0      118.600
            6    JUL90    FORECAST       0       98.611
 
I can't imagine where the value for E1 comes from.
 
Help will be greatly appreciated !
 
Martina Vandebroek
Department of Applied Economics
Naamsestraat 69
3000 Leuven
Belgium
e-mail: martina.vandebroek@econ.kuleuven.ac.be