Bagging and time series

In this section, we will only illustrate the bagging technique for the ETS model. The main purpose of bagging is to stabilize the predictions or forecasts. Here, we will base the bagging on the Box-Cox and Loess-based decomposition. Using 500 such bootstrap samples, the bagging model for ETS will be obtained:

>uspop_bagg_ets <- baggedETS(uspop_sub,bootstrapped_series = 
+                               bld.mbb.bootstrap(uspop_sub, 500))
>forecast(uspop_bagg_ets,h=4);subset(uspop,start=16,end=19)
     Point Forecast Lo 100 Hi 100
1940            141    136    145
1950            158    150    165
1960            175    164    184
1970            193    178    204
Time Series:
Start = 1940 
End = 1970 
Frequency = 0.1 
[1] 132 151 179 203
>plot(forecast(uspop_bagg_ets,h=4))

Is there an advantage to using the bagging method? We can quickly check this using the confidence intervals:

>forecast(uspop_bagg_ets,h=4)
     Point Forecast Lo 100 Hi 100
1940      ...