Ata method’s performance in the M4 competition

Abstract

Like the previous M competitions, M4 competition resulted in great contributions to the field of forecasting. Ata method which is a new forecasting method alternative to exponential smoothing, competed in this competition with five different models. The results obtained from these five models are discussed in detail in this paper. According to various error metrics, the models perform better than their exponential smoothing based counters. Despite their simplicity, they are ranked satisfactorily high compared to the other methods. In addition, the forecasting accuracy of simple combinations of these Ata models and ARIMA are given for the M4 competition data set. The combinations work significantly better than models that are much more complex. Therefore, besides the fact that Ata models perform well alone, Ata should be considered as a candidate for being included in combinations of forecasts.

Keywords

• Exponential smoothing
• forecasting
• M4-competition
• time series

References

1. [1] B. Cetin and I. Yavuz, Comparison of forecast accuracy of Ata and exponential smoothing, J. Appl. Stat. 48 (13-15), 2580-2590 , 2020.
2. [2] R.J. Hyndman, A brief history of forecasting competitions, Int. J. Forecast. 36 (1), 7-14, 2020.
3. [3] R.J. Hyndman and B. Billah, Unmasking the Theta method, Int. J. Forecast. 19 (2), 287-290, 2003.
4. [4] S. Makridakis, E. Spiliotis and V. Assimakopoulos, The M4 Competition: 100,000 time series and 61 forecasting methods, Int. J. Forecast. 36 (3), 54-74, 2020.
5. [5] S. Makridakis, E. Spiliotis and V. Assimakopoulos, The M4 Competition: Results, findings, conclusion and way forward, Int. J. Forecast. 34 (4), 802-808, 2018.
6. [6] S. Makridakis and M. Hibon, The M3-Competition: results, conclusions and implications, Int. J. Forecast. 16 (4), 451-476, 2000.
7. [7] S. Makridakis, Mcompetitions, Mcompetitions/M4-methods: Data, benchmarks, and methods submitted to the M4 forecasting competition, https://github.com/M4Competition/M4-methods
8. [8] J.M. Merigo and M. Jose, Measuring Errors with the OWA Operator, Lecture Notes in Business Information Processing 115, 24-33, 2012.

9. [9] G. Yapar, H. Selamlar, S. Capar and I. Yavuz, ATA method, Hacettepe J. Math. Stat. 48 (6), 1838-1844, 2019.
10. [10] G. Yapar Modified simple exponential smoothing, Hacettepe J. Math. Stat. 47 (3), 741-754, 2018.
11. [11] G. Yapar, S. Capar, H. Selamlar and I. Yavuz, Modified Holt’s linear trend method, Hacettepe J. Math. Stat. 47 (5), 1394-1403, 2018.