Study · 01 — Lab
Demand forecasting and inventory policy
Three years of data across twenty fast-moving SKUs with seasonality, B2B buying patterns and inflation adjustment.
The problem
How much will sell and when to restock, without deciding by intuition.
What I built
SARIMAX with Fourier terms, Prophet and a global LightGBM with recursive forecasting, compared under time-series validation.
What changed
The forecasts translate into a reorder point and safety stock per product on a newsvendor basis.
Tools
- Python
- LightGBM
- Prophet
- SARIMAX