M5: Why it matters for time-series forecasting
The M5 Forecasting Competition is one of the most-used modern benchmarks for practical demand forecasting. It uses real Walmart retail sales data with rich hierarchy (item, department, store, state) and calendar/event effects.
Why M5 is a strong standard
- Real-world complexity: noisy retail demand, intermittent series, promotions, and seasonality.
- Hierarchical structure: forecasts must make sense across aggregation levels, not just one series.
- Scale: thousands of related series test both model quality and operational reliability.
- Community adoption: many papers and libraries report M5-style performance, so comparisons are meaningful.
What ForecastingAPI.com does with M5
- Lets you run a quick M5 demo mode from the landing page.
- Samples a configurable number of M5 series via
n_series. - Runs rolling backtests and reports SMAPE so you can compare models on historical windows.
- Then generates forward forecasts for each series.
When to use M5 demo vs your own payload
- M5 demo: great for sanity checks, regressions, and model-comparison smoke tests.
- Your payload: best for business-specific signal, seasonality, and decision support.
Note: M5 is a benchmark, not your business. Treat it as a quality baseline, then validate on your own data.