Status: Draft.

⚠️ Self note: my README calls this "Hull Tactical / Hedge Fund Forecasting", but the data (code / sub_code / sub_category fields, multiple horizons) looks like the Hedge Fund Time Series Forecasting competition. I need to confirm which one it is before linking it.

Objective

Forecast a financial target per entity ID across multiple time horizons.

Approach

A LightGBM pipeline with modular feature processors, so I can add or remove feature groups without rewriting the model code.

Next steps

1. Focus on horizon 1 only before trying to cover every horizon.
2. Add lag features.
3. Test whether the category fields (code, sub_code, sub_category) work better as targets than as inputs.