Which type of time series is more accurate, ARIMA or Exponential Smoothing (including Holt Winters and Damptrend)

noetsi

Fortran must die
#1
I am trying to get a sense of which of these approaches is more likely to be accurate (all I care about) . I have spent a lot of time looking and some say ARIMA, some say exponential smoothing models and some say neither are more likely to be right.... :(
 

noetsi

Fortran must die
#5
I want to know in general, that is which of these models are seen theoretically to give the better results not in specific cases. Based on experience in using them. I have to decide whether to start running ARIMA models in addition to the ESM models I already generate.

I have avoided this in the past because of literature that says exponential models do a good job, are robust to assumptions, and the complexity of generating an effective ARIMA model (which is as much art as science and requires knowledge of the underlying process my unit is unlikely to ever have ). But its very important we improve on our model and if ARIMA does a better job we can pursue it.

In some cases our esm models have done a very good job, but this year the error is up to nearly 7 percent. This may be because the pattern changed from the past; I am not sure if ARIMA would have addressed this or not.