Join the TS-Arena Benchmark
Test your forecasting models against the best in real-time benchmark challenges
Get Started
Ready to test your forecasting models against the best? Participate actively in our benchmark challenges! Create an account in the console and mint your own API key. Find detailed participation instructions in our Git repository.
Already have an account? Sign in to the console. If you would rather not manage an account, write to us and we will set you up.
How it Works
Create an Account and an API Key
Register in the console, confirm your email address, then mint an API key there. Naming an organization creates it with you as its admin and shows it on the leaderboard; leave it blank to take part as an individual. If your organization already exists, ask its admin to invite you instead.
Register Your Model
Use your API key to register models for challenges in the registration phase.
Submit Forecasts
Obtain time-series context via our API and submit your forecasts before registration closes. Every forecast is committed before the values it predicts exist, which is what makes leakage impossible here.
Wait for the Ground Truth
Once registration closes, no further forecasts are accepted for that round. The round then plays out on live data, and nothing can be scored until every real value it covers has arrived. For a long horizon that wait is the length of the horizon itself.
Evaluation
When the ground truth for a round is complete, your forecast is scored automatically. The check runs every 30 minutes, so results follow shortly after the last value lands rather than the moment it does. The error metrics themselves are in the code on GitHub.
Where You Appear
A scored model shows up in that challenge's leaderboard and in the aggregate views (domain, category, frequency). The overall ranking is stricter: a model enters it only once it has taken part in every challenge, and in at least half of each challenge's rounds since it joined, so nobody gains by covering only the challenges that suit them. Rankings are recomputed four times a day. You can follow your model on the leaderboards and inspect individual forecasts in the interactive plots.
What You Get
Personal API Key
For model registration and submission
Benchmark Datasets
Access to datasets and challenge specifications
Rankings & Leaderboards
Real-time performance tracking
Detailed Metrics
Comprehensive evaluation across time series
Community Support
Connect with forecasting researchers
Requirements
Help Us Validate Model Implementations
Transparency is at the heart of our live benchmarking project. We have integrated a wide range of state-of-the-art time series forecasting models to provide the community with real-time performance insights.
To ensure that every model is evaluated under optimal conditions, we invite the original authors and maintainers to review our implementations. If you are a developer of one of the featured models, we would value your feedback on:
- •Model configuration and hyperparameters
- •Data preprocessing steps
- •Implementation-specific nuances
Our goal is to represent your work as accurately as possible. Please visit our GitHub repository to review the code or open an issue for any suggested improvements.
View on GitHub