rah-salah/ethiopia-fi-forecast

# Ethiopia Financial Inclusion Forecasting Forecasting system for Ethiopia's digital financial transformation, built for **Selam Analytics**. Predicts Access (Account Ownership) and Usage (Digital Payment Adoption) trends for 2025-2027, using a curated, source-documented dataset of survey observations, market events, and estimated event impacts. ## Setup ```bash pip install -r requirements.txt ``` Then explore the data via the notebook: ```bash jupyter notebook notebooks/task1_2_exploration_eda.ipynb ``` ## Data Sources - **`data/raw/ethiopia_fi_unified_data.csv`** - the core dataset: survey observations (e.g. Global Findex account ownership), market events (e.g. Telebirr launch), estimated event impacts, and policy targets. Originally converted from `ethiopia_fi_unified_data.xlsx` (see `src/convert_raw_data.py`), then extended with 5 additional sourced records (see `data_enrichment_log.md`). - **`data/raw/reference_codes.csv`** - lookup table defining the controlled vocabulary used throughout the unified dataset (valid `record_type`, `pillar`, `indicator_direction`, `value_type` values, etc.). - Primary external sources include the World Bank Global Findex Database (2014/2017/2021/2024 editions), the National Bank of Ethiopia, individual mobile money operators (Ethio Telecom/Telebirr, Safaricom/M-Pesa), and peer-reviewed research citing Findex microdata. Every record traces back to a `source_name`/`source_url` pair. ## Unified Schema All data lives in a single flat table (`ethiopia_fi_unified_data.csv`), with each row's meaning determined by its `record_type`: | record_type | Meaning | `pillar` set? | |---|---|---| | `observation` | An actual measured value from a source (e.g. "49% account ownership, Nov 2024, Global Findex") | Yes | | `event` | A policy launch, market event, or milestone (e.g. "Telebirr Launch, May 2021") | No - an event itself isn't tied to one pillar | | `impact_link` | An analyst-estimated relationship connecting an `event` to t …