*
* Examples of exponential smoothing from pp 161-172
*
open data fsales.dat
calendar(q) 2001
data(format=free,org=columns) 2001:1 2006:4 fsales
*
graph(footer="Figure 4-9 Time plot of quarterly sales data",vlabel="Sales(thousand of francs")
# fsales
*
* Simple exponential smoothing
*
esmooth(estimate,initial=start,fit=sesfit) fsales
@uforeerrors fsales sesfit 10 24
*
* Holt's model
*
esmooth(trend=linear,estimate,constrain,initial=start,fit=holtfit) fsales
@uforeerrors fsales holtfit 10 24
*
graph(footer="Figure 4-10 Forecasts from SES and Holt's method",$
 key=upleft,klabels=||"Observed Data","SES forecast","Holt's forecast"||) 3
# fsales
# sesfit
# holtfit
*
* Holt's seasonal model using a linear trend. (There's also a TREND=EXPON option
* for an exponential trend).
*
esmooth(trend=linear,seasonal=multiplicative,initial=start,constrain,estimate,fit=seasfit) fsales
@uforeerrors fsales seasfit 10 24
*
* The ESMOOTH instruction includes all nine of the Pegels models. To select a
* model automatically (from the in-sample fit), use the options TREND=SELECT and
* SEASONAL=SELECT. In this case, it selects the exponential trend with
* multiplicative seasonal, which gives a slightly better fit than the linear
* trend with multiplicative seasonal.
*
esmooth(trend=select,seasonal=select,initial=start,constrain) fsales