The M4 Competition: Results, findings, conclusion and way forward
By Spyros Makridakis, Evangelos Spiliotis, Vassilios Assimakopoulos. 2018
686 citations recorded by the source corpus.
The M4 competition is the continuation of three previous competitions started more than 45 years ago whose purpose was to learn how to improve forecasting accuracy, and how such learning can be applied to advance the theory and practice of forecasting. The purpose of M4 was to replicate the results of the previous ones and extend them into three directions: First significantly increase the number of series, second include Machine Learning (ML) forecasting methods, and third evaluate both point forecasts and prediction intervals. The five major findings of the M4 Competitions are: 1. Out Of the 17 most accurate methods, 12 were “combinations” of mostly statistical approaches. 2. The biggest surprise was a “hybrid” approach that utilized both statistical and ML features. This method’s average sMAPE was close to 10% more accurate than the combination benchmark used to compare the submitted methods. 3. The second most accurate method was a combination of seven statistical methods and one ML one, with the weights for the averaging being calculated by a ML algorithm that was trained to minimize the forecasting. 4. The two most accurate methods also achieved an amazing success in specifying the 95% prediction intervals correctly. 5. The six pure ML methods performed poorly, with none of them being more accurate than the combination benchmark and only one being more accurate than Naive2. This paper presents some initial results of M4, its major findings and a logical conclusion. Finally, it outlines what the authors consider to be the way forward for the field of forecasting.
Subfields: Management Science and Operations Research, Statistics and Probability, Evaluation & Benchmarks, Time-Series & Forecasting
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Management Science and Operations Research
- Bernie Wang
- Christoph Bergmeir
- Fotios Petropoulos
- Jan Gasthaus
- Rob J Hyndman
- Tim Januschowski
- Kasun Bandara
- Spyros Makridakis
In Statistics and Probability
- Fotios Petropoulos
- Christoph Bergmeir
- Slawek Smyl
- Spyros Makridakis
- David Salinas
- Maciej Pawlikowski
- Agata Chorowska
- Kasun Bandara
In Evaluation & Benchmarks
- Fotios Petropoulos
- Rob J Hyndman
- Christoph Bergmeir
- Slawek Smyl
- Yanfei Kang
- Kasun Bandara
- Spyros Makridakis
- Boris N. Oreshkin
In Time-Series & Forecasting
- Bernie Wang
- Jan Gasthaus
- Tim Januschowski
- Fotios Petropoulos
- Christoph Bergmeir
- David Salinas
- Valentin Flunkert
- Syama Sundar Rangapuram
Citations in the corpus
- N-BEATS: Neural basis expansion analysis for interpretable time series forecasting (2019, methodology): “Smyl [23] who had to use very different architectures hand crafted for different horizons.”
- Enhancing the Locality and Breaking the Memory Bottleneck of Transformer on Time Series Forecasting (2019, background): “f(x) = 8 >< >: A1 sin(⇡x/6) + 72 +Nx x 2 [0, 12), A2 sin(⇡x/6) + 72 +Nx x 2 [12, 24), A3 sin(⇡x/6) + 72 +Nx x 2 [24, t0), A4 sin(⇡x/12) + 72 +Nx x 2 [t0, t0 + 24), where x is an integer, A1, A2, A3 are randomly generated by uniform distribution on [0, 60], A4 = max(A1, A2) and Nx ⇠ N (0, 1).”
- Enhancing Time-Series Momentum Strategies Using Deep Neural Networks (2019, neutral): “For instance, the creation of hybrid methods that combine traditional time-series models with neural network components have been observed to outperform pure methods in either category [17] e.”
- Less Is More: Fast Multivariate Time Series Forecasting with Light Sampling-oriented MLP Structures (2022, methodology): “N-BEATS is the first pure deep-learning-based method for univariate time series forecasting that achieves SOTA in the M4 competition2 [13, 15].”
- Probabilistic Forecasting with Temporal Convolutional Neural Network (2019, methodology): “A hybrid model that combines exponential smoothing method and RNN won the M4 forecasting competition, which consists of 100,000 serieswith different seasonal patterns [21].”
- A review of artificial neural network models for ambient air pollution prediction (2019, methodology): “Finally, the application of hybrid ANN models has been highlighted in recent years (Makridakis et al., 2018).”
- Recurrent Neural Networks for Time Series Forecasting: Current Status and Future Directions (2019, background): “Compared to the previous competitions, one of the key objectives of the M4 is to increase the number of time series available for forecasting (Makridakis et al., 2018a).”
- A hybrid method of exponential smoothing and recurrent neural networks for time series forecasting (2020, positive): “Once again, the denominator of the MSIS was omitted, since the input to the NNs was already deseasonalized and normalized.”
- Weighted ensemble of statistical models (2018, methodology): “• Na¨ıve models - Na¨ıve 1, Na¨ıve 2 (Makridakis, 2018) • Exponential Smoothing (ETS) models - simple ETS [forecast::ses], ETS with automatic model choice [forecast::ets], with or without damped trend (Hyndman, Koehler, Snyder & Grose, 2002) • Theta models - Theta method (Assimakopoulos &…”
- One Fits All: Power General Time Series Analysis by Pretrained LM (2023, methodology): “6 Short-term Forecasting Setups To fully evaluate different algorithms in forecasting tasks, we also conduct short-term forecasting (with relatively short forecasting horizon) experiments on M4 Makridakis et al. (2018), contains marketing data of various frequencies.”
- Deep State Space Models for Time Series Forecasting (2018, background): “This includes monthly and quarterly time series from the tourism competition dataset [2] describing tourism demand, hourly time series from the M4 competition [20] and parts dataset [6] which contains monthly demand of spare parts at a US auto-mobile company.”