Who leads the M4 Competition benchmark?

TiDE currently leads M4 Competition with a score of 13.95 on smapi.

What is the state-of-the-art score on M4 Competition?

The state-of-the-art result on M4 Competition is 13.95 (smapi), achieved by TiDE as of 2024.

How many models are tracked on M4 Competition?

Codesota tracks 13 models on M4 Competition across 3 metrics.

When was the M4 Competition leaderboard last updated?

The M4 Competition leaderboard on Codesota includes results through 2024, with the earliest tracked result from 2018.

Codesota · Benchmark · M4 CompetitionHome/Leaderboards/Structured Data & Forecasting/Time-Series Forecasting/M4 Competition

Unknown

M4 Competition.

Name: M4 Competition Benchmark Results
Creator: Unknown
Published: 2018-01-01
License: https://creativecommons.org/licenses/by/4.0/

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

Paper ↗Leaderboard ↓

§ 01 · SOTA history

Year over year.

§ 02 · Leaderboard

Results by metric.

Found a wrong score or missing run?

Use row edits to send a sourced correction into moderation.

Add / edit result ↗Report issue ↗

smapi

Smapi is the reported evaluation metric for M4 Competition. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for smapiverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Fix
01	TiDE Long-term time series forecasting with MLP-Mixer encoder-decoder. Results from TimeMixer++ comparison table.	verified	13.95	2023	Source ↗	Looks wrong?
02	DLinear	verified	13.639	2022	Paper ↗Code ↗Source ↗	Looks wrong?
03	PatchTST	verified	13.152	2022	Paper ↗Code ↗Source ↗	Looks wrong?
04	Autoformer NeurIPS 2021. Auto-Correlation mechanism for time series decomposition. Results from TimeMixer++ comparison table.	verified	12.909	2021	Source ↗	Looks wrong?
05	FEDformer	verified	12.84	2022	Paper ↗Source ↗	Looks wrong?
06	iTransformer ICLR 2024. Inverted transformer with attention applied across variables. Results from TimeMixer++ comparison table.	verified	12.684	2024	Source ↗	Looks wrong?
07	N-HiTS	verified	11.927	2022	Paper ↗Code ↗Source ↗	Looks wrong?
08	LMS-AutoTSF Dec 2024. Learnable Multi-Scale Decomposition and Integrated Autocorrelation. Average across Yearly/Quarterly/Monthly/Others.	verified	11.871	2024	Source ↗	Looks wrong?
09	N-BEATS ICLR 2020. Neural basis expansion analysis. Results from TimeMixer++ comparison table (no ensemble).	verified	11.851	2020	Source ↗	Looks wrong?
10	TimesNet ICLR 2023. Transforms 1D time series into 2D space for temporal variation modeling.	verified	11.829	2023	Source ↗	Looks wrong?
11	TimeMixer	verified	11.723	2024	Paper ↗Code ↗Source ↗	Looks wrong?
12	TimeMixer++	verified	11.448	2024	Paper ↗Code ↗	Looks wrong?
13	ES-RNN	verified	11.374	2020	Paper ↗Code ↗	Looks wrong?

mase

Mase is the reported evaluation metric for M4 Competition. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for maseverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Fix
01	DLinear	verified	2.10	2022	Paper ↗Code ↗Source ↗	Looks wrong?
02	PatchTST	verified	1.95	2022	Paper ↗Code ↗Source ↗	Looks wrong?
03	TiDE Long-term time series forecasting with MLP-Mixer encoder-decoder. Results from TimeMixer++ comparison table.	verified	1.94	2023	Source ↗	Looks wrong?
04	Autoformer NeurIPS 2021. Auto-Correlation mechanism for time series decomposition. Results from TimeMixer++ comparison table.	verified	1.77	2021	Source ↗	Looks wrong?
05	iTransformer	verified	1.76	2023	Paper ↗Code ↗Source ↗	Looks wrong?
06	FEDformer	verified	1.70	2022	Paper ↗Source ↗	Looks wrong?
07	N-HiTS AAAI 2023. Neural hierarchical interpolation for time series. Results from TimeMixer++ comparison table.	verified	1.61	2022	Source ↗	Looks wrong?
08	LMS-AutoTSF Dec 2024. Learnable Multi-Scale Decomposition and Integrated Autocorrelation. Average across Yearly/Quarterly/Monthly/Others.	verified	1.59	2024	Source ↗	Looks wrong?
09	TimesNet ICLR 2023. Transforms 1D time series into 2D space for temporal variation modeling.	verified	1.58	2023	Source ↗	Looks wrong?
10	N-BEATS	verified	1.56	2019	Paper ↗Code ↗Source ↗	Looks wrong?
11	TimeMixer ICLR 2024. Decomposable multiscale mixing for time series forecasting.	verified	1.56	2024	Source ↗	Looks wrong?
12	ES-RNN	verified	1.54	2020	Paper ↗Code ↗	Looks wrong?
13	TimeMixer++	verified	1.49	2024	Paper ↗Code ↗	Looks wrong?

Owa

Owa is the reported evaluation metric for M4 Competition. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Owaverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Fix
01	DLinear	verified	1.05	2022	Paper ↗Code ↗Source ↗	Looks wrong?
02	TiDE Long-term time series forecasting with MLP-Mixer encoder-decoder. Results from TimeMixer++ comparison table.	verified	1.02	2023	Source ↗	Looks wrong?
03	PatchTST ICLR 2023. Transformer with patch tokenization. Results from TimeMixer++ comparison table.	verified	1.00	2023	Source ↗	Looks wrong?
04	Autoformer NeurIPS 2021. Auto-Correlation mechanism for time series decomposition. Results from TimeMixer++ comparison table.	verified	0.94	2021	Source ↗	Looks wrong?
05	iTransformer	verified	0.93	2023	Paper ↗Code ↗Source ↗	Looks wrong?
06	FEDformer ICML 2022. Frequency enhanced decomposed transformer. Results from TimeMixer++ comparison table.	verified	0.92	2022	Source ↗	Looks wrong?
07	N-HiTS	verified	0.86	2022	Paper ↗Code ↗Source ↗	Looks wrong?
08	N-BEATS	verified	0.85	2019	Paper ↗Code ↗Source ↗	Looks wrong?
09	LMS-AutoTSF Dec 2024. Learnable Multi-Scale Decomposition and Integrated Autocorrelation. Average across Yearly/Quarterly/Monthly/Others.	verified	0.85	2024	Source ↗	Looks wrong?
10	TimesNet	verified	0.85	2022	Paper ↗Code ↗Source ↗	Looks wrong?
11	TimeMixer	verified	0.84	2024	Paper ↗Code ↗Source ↗	Looks wrong?
12	ES-RNN M4 competition winner (2018). Hybrid exponential smoothing + RNN by Slawek Smyl (Uber). OWA=0.821.	verified	0.82	2018	Source ↗	Looks wrong?
13	TimeMixer++ ICLR 2025. Multi-resolution time imaging + decomposable multiscale mixing. Achieves OWA 0.821 matching original M4 competition winner.	verified	0.82	2024	Source ↗	Looks wrong?

§ 04 · Submit a result

Add to the leaderboard.

← Back to Time-Series Forecasting