KRASI1234/Ghana-Inflation-Forecating

# Ghana Inflation Forecasting. A Machine Learning Study > **Can machine learning outperform traditional econometrics for inflation forecasting during a crisis?** > This project answers that using 120 months of Ghanaian macroeconomic data; including the 2022 crisis where prices peaked at **54.1% YoY**. --- ## 1. Executive Summary Ghana's inflation surged from ~7% in 2019 to a peak of **54.1% in December 2022**, driven by cedi depreciation, fuel price shocks, and fiscal pressures. This project benchmarks **13 forecasting models** on monthly CPI data from January 2015 to December 2024. The best model, **Bayesian Ridge**, achieved an **R² of 0.919** and reduced forecast error by **72.9%** compared to the SARIMAX-GARCH baseline on the unseen 2023–2024 test period. | Model | RMSE | MAPE | R² | |---|---|---|---| | Bayesian Ridge | 3.04 | 9.1% | **0.919** | | Rolling XGBoost | 3.24 | **7.8%** | 0.907 | | Weighted Blend | 3.43 | 10.1% | 0.896 | | Huber Regression | 4.68 | 13.7% | 0.807 | | Elastic Net | 4.99 | 16.6% | 0.780 | | XGBoost | 5.90 | 15.3% | 0.693 | | Random Forest | 8.23 | 26.9% | 0.402 | | SARIMAX-GARCH *(baseline)* | 11.18 | 29.2% | -0.103 | | SVR | 12.41 | 22.0% | -0.358 | --- ## 2. Business Problem The **Bank of Ghana** and economic policymakers rely on inflation forecasts to set the Monetary Policy Rate, issue forward guidance to banks and households, and plan fiscal interventions before inflation spirals out of control. **The core challenge:** Traditional models (ARIMA, SARIMAX-GARCH) are calibrated on historical data and assume a stable relationship between variables. Ghana's 2022–2024 crisis represented a **2.3× regime shift**. The test period mean (31.6%) was more than double the training period mean (13.7%). Static models trained on pre-crisis data fail in this environment. **Key question answered:** Which forecasting approach (traditional or ML) best handles a sudden inflation regime shift with only 83 training observations? --- ## 3. Met …