Which is better, exponential smoothing or ARMA models for solving short time series
It depends on what you want to do: here are some links:
https://booksc.xyz/book/9870734/18ec97
Article Joint Estimation: SPC Method for Short Run Autocorrelated Data
Best wishes, D. Booth
Dear Dr. Marwan,
I 'm interested in this brainstorming question and wish to know the answer.
Because I think the answer will be based on data quality and model identification.
Accept my best regards,
Omar A. Ali
I think ARMA. Because, in exponential smoothing the correlation will weaken with past data.
Do not forget that most of the exponential smoothing (ES) methods have underlying ARIMA model representations. For example simple ES corresponds to an ARIMA(0,1,1) model. An exception is multiplicative Winters method which can only be approximated by the additive Winters method on logs. What differs is that ES methods make use simple recurrences, at least for given smoothing constants (often roughly estimated), while ARIMA models require specification and estimation. A comparison is available in my paper with Laurence Broze.
Article Exponential smoothing: Estimation by maximum likelihood
Also note that the R forecast package combines powerful ES with ARIMA.
I think The answer is determined after knowing the nature of the model..
This link would be useful for you:
https://www.researchgate.net/post/What_is_the_difference_between_ARIMA_and_Exponential_smoothing_model_as_both_are_based_on_their_past_values_and_error_term
This may be found only by empirical comparison and depends on data features. This framework can be used to perform the empirical comparison:
Chapter Forecast Evaluation Techniques for I4.0 Systems
For more on error metrics, please let me recommend:
Chapter Forecast Error Measures: Critical Review and Practical Recommendations
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series