SAS 9.4 PROC ESM issues Error: unable to forecast a variable

[AlexRubin]

proc esm

data = wrk.FCST_15_16_NON_X

out = wrk.f_raw_fcst_NON_X

lead = 53 ;

id week interval = week ACCUMULATE=total ;

by target;

forecast writer_count imputed_mg / model = winters ;

run;

ERROR: Unable to forecast the variable writer_count. Forecast values set to missing.

NOTE: The above message was for the following BY group:

target=NON_X

[dw_sas]

Hi Alex,

Winters method requires not only the actual (historical) values to be positive, but also the one-step-ahead predicted values.  A plot of the series shows a sharp downward trend toward zero in the historical data, where the Winters model likely returned a non-positive one-step-ahead predicted value.  Because zero or negative values cause the Winters algorithm to fail, PROC ESM issued the error:

ERROR: Unable to forecast the variable writer_count. Forecast values set to missing.

The other seasonal models supported by PROC ESM do not require the actual values and one-step-ahead predicted values to be positive, so you might want to replace the MODEL=WINTERS option with either MODEL=ADDWINTERS or MODEL=SEASONAL for your data.

I hope this helps!

DW