david006-DS/GH-Yield

Intelligent Treasury Bill Rate Forecasting System for Ghana — ML + Quant Finance + MLOps # GH-Yield: Ghana Treasury Bill Rate Forecasting System **An end-to-end intelligent treasury management system that forecasts Ghana's 91-day T-bill rates, generates optimal investment signals, quantifies risk, and serves predictions through a production API with an interactive dashboard.** --- ## The Problem Ghana's 91-day Treasury bill rate swung from **14%** (2021) to **35.7%** (February 2023) to **11%** (December 2025). For a treasury manager at a Ghanaian bank, timing this cycle on a GHS 50 million position translates to millions in quarterly earnings either captured or lost. GH-Yield asks: **can a systematic, data-driven approach time these rate movements better than intuition?** --- ## Architecture ``` Data Sources → Feature Engineering → Model Training → Strategy Engine → API / Dashboard | | | | | Bank of Ghana 29 features 8 ML models BUY/HOLD/WAIT FastAPI + FRED API Technical + Walk-forward CV Risk metrics Streamlit Yahoo Finance Yield curve + MLflow tracking Backtesting Docker World Bank Macro + Regime Ensemble Monte Carlo ``` --- ## Key Features **Data Pipeline** - Automated extraction from Bank of Ghana PDF statistical bulletins using pdfplumber - FRED API integration for US T-bill rates and Federal Funds Rate - Yahoo Finance for USD/GHS exchange rate - World Bank API for Ghana inflation and GDP - Alignment of daily, monthly, and annual data to a common monthly frequency **Feature Engineering (29 features across 4 groups)** - **Technical (15):** Lag features, momentum, rolling statistics, z-score mean reversion signal, rate acceleration - **Yield Curve (4):** Term structure slope, curvature, short-end and long-end spreads - **Macro / Policy (7):** Real rate, policy spread, Ghana-US rate differential, FX momentum, inflation momentum - **Regime (3):** Volatility regime, rate direction regime, policy stance class …