I found the only difference between ARIMA and Exponential smoothing model is the weight assignment procedure to its past lag values and error term. In that case Exponential should be considered much better that ARIMA due to its weight assigning method. But in many of literature i found ARIMA is best as compared exponential why is that?
Also both of them are time series model and suitable only in case of describing linear relationship so how it is beneficial when we want to compare them with explanatory models and non linear models specially neural network where we don't need any assumption about the data structure.
Prity -
A few thoughts:
I think simple exponential smoothing might get by with fewer assumptions, though there are more complex forms of it. You may or may not include seasonality. It may be best if you want to emphasize the most recent members of a series, and think older data are far less relevant. (I think that is what you meant by 'weights.')
ARIMA, I think, may be better if you have more auxiliary information to use.
But for any model, the best one to use is the one which best describes your data without overfitting. Because you are forecasting, unforeseen changes in relevant phenomena may be particularly unfortunate for a complex, overfit model. Still, you could go back in your series, and experiment/test to see which models would have worked best for forecasting data which you now have. That may be of some help. But remember that forecasting definitely assumes a pattern will hold.
Cheers - Jim
.
exponential smoothing has evolved in the recent years (mainly the 90's) into a very large class of models with automatic model choice and error estimates, thus becoming quite competitive with classical arima models
.
this might explain the conflicting views about the respective merits of exponential smoothing and arima you can find in the litterature
.
this said, the two classes are not equivalent : there are exponential smoothing models which are not in the arima class and vice versa
https://www.otexts.org/fpp/8/10
https://www.bauer.uh.edu/gardner/docs/pdf/Exponential-Smoothing.pdf
.
Both are useful techniques in forecasting a time series. While exponential smoothing technique depends upon the assumption of exponential decrease in weights for past data and ARIMA is employed by transforming a time series to stationary series and studying the the nature of the stationary series through ACF and PACF and then accounting auto-regressive and moving average effects in a time series, if present.
Exponential smoothing is a simple procedure to study time series (Xt) not used to analysis ... but ARIMA is agood procedure to analysis time series and used (I) by take differences of time series to become more staionary..
and this procedure used Box-Jenkenis ..
Thank you about this question
Indeed exponential smoothing is a non-parametric technique while Box-Jenkins is a parametric technique and we know that parametric techniques are usually more efficient than non-parametric techniques in prediction.
As I seen in the papers, recently some new prediction techniques such as neural networks, wavelets, decision trees and support vector machines have been used too much. Actually I think these new techniques lead to better predictions than classic techniques.
Thanks a lot for your suggestions James
Yes i completely agree that for any model, the best one model will be that one which describes available data more precisely without over-fitting.
Thank you for suggestion Fabrice Clerot . Both of link are really helpful.
ARIMA and Exponential smoothing model both are useful for forecasting time series data. The major difference is auto regressive term in ARIMA(p,d,q) is zero.
ARIMA(0,1,1) without constant = simple exponential smoothing
ARIMA(0,1,1) with constant = simple exponential smoothing with growth
- Badges
- Science topic