Linear exponential smoothing with PROC ESM

cw_sas — Mon, 26 Jun 2017 04:45:25 GMT

I would like to know how to interpret the parameter estimates in PROC ESM when Holt's exponential smoothing method is used.

proc esm data=t print=all ;
forecast x /model=linear;
run;

The estimate for Trend Weight is always 0.001.

i applied to various real data sets and the simulated data sets with different parameter settings.

data t;
a=0.5;
b=0.1;
do t=0 to 100;
x=a+b*t+rannor(123)*0.1;
output;
end;
run;

how can they be converted to the parameters, alpha and gamma, specified in the document?

thanks.

Re: Linear exponential smoothing with PROC ESM

cw_sas — Mon, 26 Jun 2017 13:05:13 GMT

i answer my own question.

Level weight is alpha and Trend weight is gamma.

it just happens that the all data sets i used have constant trends (T(t) = T(t-1)).

the simulation data, of course, have constant trend.

therefore, gamma=0.001 is essentially zero.