Home/Browse/Time Series/Time Series Forecasting/M4 Competition

M4 Competition

Name: M4 Competition Benchmark Results
Creator: Unknown
License: https://creativecommons.org/licenses/by/4.0/

Unknown

100,000 time series from diverse domains (finance, demo, macro, micro, industry, other).

Paper Leaderboard

Benchmark Stats

Models13

Papers39

Metrics3

SOTA History

smapi

Higher is better

Rank	Model	Source	Score	Year	Paper
1	TiDE Long-term time series forecasting with MLP-Mixer encoder-decoder. Results from TimeMixer++ comparison table.	Editorial	13.95	2023	Source
2	DLinear AAAI 2023. Simple decomposition linear model. Results from TimeMixer++ comparison table.	Editorial	13.639	2023	Source
3	PatchTST ICLR 2023. Transformer with patch tokenization. Results from TimeMixer++ comparison table.	Editorial	13.152	2023	Source
4	Autoformer NeurIPS 2021. Auto-Correlation mechanism for time series decomposition. Results from TimeMixer++ comparison table.	Editorial	12.909	2021	Source
5	FEDformer ICML 2022. Frequency enhanced decomposed transformer. Results from TimeMixer++ comparison table.	Editorial	12.84	2022	Source
6	iTransformer ICLR 2024. Inverted transformer with attention applied across variables. Results from TimeMixer++ comparison table.	Editorial	12.684	2024	Source
7	N-HiTS AAAI 2023. Neural hierarchical interpolation for time series. Results from TimeMixer++ comparison table.	Editorial	11.927	2022	Source
8	LMS-AutoTSF Dec 2024. Learnable Multi-Scale Decomposition and Integrated Autocorrelation. Average across Yearly/Quarterly/Monthly/Others.	Editorial	11.871	2024	Source
9	N-BEATS ICLR 2020. Neural basis expansion analysis. Results from TimeMixer++ comparison table (no ensemble).	Editorial	11.851	2020	Source
10	TimesNet ICLR 2023. Transforms 1D time series into 2D space for temporal variation modeling.	Editorial	11.829	2023	Source
11	TimeMixer ICLR 2024. Decomposable multiscale mixing for time series forecasting.	Editorial	11.723	2024	Source
12	TimeMixer++ ICLR 2025. Multi-resolution time imaging + decomposable multiscale mixing. Achieves OWA 0.821 matching original M4 competition winner.	Editorial	11.448	2024	Source
13	ES-RNN M4 competition winner (2018). Hybrid exponential smoothing + RNN by Slawek Smyl (Uber). OWA=0.821.	Editorial	11.374	2018	Source

mase

Higher is better

Rank	Model	Source	Score	Year	Paper
1	DLinear AAAI 2023. Simple decomposition linear model. Results from TimeMixer++ comparison table.	Editorial	2.095	2023	Source
2	PatchTST ICLR 2023. Transformer with patch tokenization. Results from TimeMixer++ comparison table.	Editorial	1.945	2023	Source
3	TiDE Long-term time series forecasting with MLP-Mixer encoder-decoder. Results from TimeMixer++ comparison table.	Editorial	1.94	2023	Source
4	Autoformer NeurIPS 2021. Auto-Correlation mechanism for time series decomposition. Results from TimeMixer++ comparison table.	Editorial	1.771	2021	Source
5	iTransformer ICLR 2024. Inverted transformer with attention applied across variables. Results from TimeMixer++ comparison table.	Editorial	1.764	2024	Source
6	FEDformer ICML 2022. Frequency enhanced decomposed transformer. Results from TimeMixer++ comparison table.	Editorial	1.701	2022	Source
7	N-HiTS AAAI 2023. Neural hierarchical interpolation for time series. Results from TimeMixer++ comparison table.	Editorial	1.613	2022	Source
8	LMS-AutoTSF Dec 2024. Learnable Multi-Scale Decomposition and Integrated Autocorrelation. Average across Yearly/Quarterly/Monthly/Others.	Editorial	1.591	2024	Source
9	TimesNet ICLR 2023. Transforms 1D time series into 2D space for temporal variation modeling.	Editorial	1.585	2023	Source
10	N-BEATS ICLR 2020. Neural basis expansion analysis. Results from TimeMixer++ comparison table (no ensemble).	Editorial	1.559	2020	Source
11	TimeMixer ICLR 2024. Decomposable multiscale mixing for time series forecasting.	Editorial	1.559	2024	Source
12	ES-RNN M4 competition winner (2018). Hybrid exponential smoothing + RNN by Slawek Smyl (Uber). OWA=0.821.	Editorial	1.536	2018	Source
13	TimeMixer++ ICLR 2025. Multi-resolution time imaging + decomposable multiscale mixing. Achieves OWA 0.821 matching original M4 competition winner.	Editorial	1.487	2024	Source

owa

Higher is better

Rank	Model	Source	Score	Year	Paper
1	DLinear AAAI 2023. Simple decomposition linear model. Results from TimeMixer++ comparison table.	Editorial	1.051	2023	Source
2	TiDE Long-term time series forecasting with MLP-Mixer encoder-decoder. Results from TimeMixer++ comparison table.	Editorial	1.02	2023	Source
3	PatchTST ICLR 2023. Transformer with patch tokenization. Results from TimeMixer++ comparison table.	Editorial	0.998	2023	Source
4	Autoformer NeurIPS 2021. Auto-Correlation mechanism for time series decomposition. Results from TimeMixer++ comparison table.	Editorial	0.939	2021	Source
5	iTransformer ICLR 2024. Inverted transformer with attention applied across variables. Results from TimeMixer++ comparison table.	Editorial	0.929	2024	Source
6	FEDformer ICML 2022. Frequency enhanced decomposed transformer. Results from TimeMixer++ comparison table.	Editorial	0.918	2022	Source
7	N-HiTS AAAI 2023. Neural hierarchical interpolation for time series. Results from TimeMixer++ comparison table.	Editorial	0.861	2022	Source
8	N-BEATS ICLR 2020. Neural basis expansion analysis. Results from TimeMixer++ comparison table (no ensemble).	Editorial	0.855	2020	Source
9	LMS-AutoTSF Dec 2024. Learnable Multi-Scale Decomposition and Integrated Autocorrelation. Average across Yearly/Quarterly/Monthly/Others.	Editorial	0.854	2024	Source
10	TimesNet ICLR 2023. Transforms 1D time series into 2D space for temporal variation modeling.	Editorial	0.851	2023	Source
11	TimeMixer ICLR 2024. Decomposable multiscale mixing for time series forecasting.	Editorial	0.84	2024	Source
12	TimeMixer++ ICLR 2025. Multi-resolution time imaging + decomposable multiscale mixing. Achieves OWA 0.821 matching original M4 competition winner.	Editorial	0.821	2024	Source
13	ES-RNN M4 competition winner (2018). Hybrid exponential smoothing + RNN by Slawek Smyl (Uber). OWA=0.821.	Editorial	0.821	2018	Source

Submit a Result

Back to Time Series Forecasting