*
* Example 9.2.1 from page 324
*
open data deaths.tsm
calendar(m) 1973
data(format=free,org=columns) 1973:1 1978:12 deaths
*
* The Holt-Winters (non-seasonal) model is done with the ESMOOTH instruction with
* the option TREND=LINEAR. (TREND=NONE will do a simple "local level" exponential
* smoothing). The other options here request
*   (a) estimation with the parameters restricted to the range [0,1]
*      (the default will allow them to stray outside of that, as the recursions
*       are stable for, for instance, alpha values up to 2.0).
*   (b) initialization of the recursions based upon the just the values
*       at the start of the data set (the default uses the full data set).
*
* The FORECASTS and STEPS options put 24 steps of out-of-sample forecasts into
* the series FORECAST.
*
esmooth(trend=linear,constrain,init=start,forecasts=forecast,steps=24) deaths
*
graph(footer="Figure 9-2 Deaths data with predictions of non-seasonal HW") 2
# deaths
# forecast
*
* Bring in the extra data to compare with the forecasts
*
data(unit=input) 1979:1 1979:6 adeaths
 7798 7406 8363 8460 9217 9316
*
* Compute statistics on forecast errors and display the forecasts
* and actual data.
*
@uforeerrors(title="Holt-Winters Forecasts") adeaths forecast 1979:1 1979:6
print(picture="*.") 1979:1 1979:6 adeaths forecast