Forecasting with ets results

问题

I load a dataframe (named stock) with this data:

day               value
2000-12-01 00:00:00 11.809242 
2000-12-01 06:00:00 10.919792 
2000-12-01 12:00:00 13.265208 
2000-12-01 18:00:00 13.005139 
2000-12-02 00:00:00 10.592222  
2000-12-02 06:00:00 8.873160 
2000-12-02 12:00:00 12.292847 
2000-12-02 18:00:00 12.609722 
2000-12-03 00:00:00 11.378299 
2000-12-03 06:00:00 10.510972  
2000-12-03 12:00:00 8.297222  
2000-12-03 18:00:00 8.110486  
2000-12-04 00:00:00 8.066154

I try to implement forecasting using ets() model with this:

library(forecast)
fs <- forecast(stock$value,h=8,model="AAN") 
fs

The output of fs is:

Point Forecast    Lo 80     Hi 80    Lo 95     Hi 95
14       8.778035 6.967009 10.589061 6.008310 11.547761
15       8.536608 6.725582 10.347635 5.766883 11.306334
16       8.295182 6.484155 10.106208 5.525456 11.064907
17       8.053755 6.242728  9.864781 5.284028 10.823481
18       7.812328 6.001301  9.623355 5.042601 10.582054
19       7.570901 5.759873  9.381928 4.801173 10.340628
20       7.329474 5.518446  9.140502 4.559746 10.099202
21       7.088047 5.277018  8.899076 4.318318  9.857776

What I observe in Forecast column is that the forecasting value goes down. Why is this happening? Does it need to set different parameters in model?

回答1:

Each of the last five observations has been less than the one before it. So you have a downward trend at the end of the historical data. The forecasting model has continued this. You chose an ETS(A,A,N) model which models local trends, and that is exactly what it is doing.

来源：https://stackoverflow.com/questions/16885828/forecasting-with-ets-results

标签

time-series

forecasting