Logo Lanfrica

GIVEN-CHINYAMA/zambia-inflation-forecasting

Domaine:

socioeconomic

Type de record:

project
Créateur:
GIV
Hôte:
Hybrid inflation forecasting for Zambia comparing ARIMAX, XGBoost, Prophet & LSTM using World Bank + Yahoo Finance data. Includes 3-year forward forecast (2025-2027). # 📊 Zambia Inflation Forecasting ### Author: Given Chinyama | Data Scientist --- ## 🎯 Project Overview A hybrid macroeconomic forecasting framework that predicts Zambia's annual inflation rate by combining classical statistical models with modern machine learning and deep learning architectures. Rather than relying on a single model, this project benchmarks four distinct approaches — ARIMAX, XGBoost, Facebook Prophet, and LSTM — against the same test set to identify the most accurate forecasting strategy for Zambia's economic context. --- ## 📈 Final Results | Rank | Model | RMSE | MAE | MAPE | |------|-------|------|-----|------| | 🥇 1st | ARIMAX | 5.65 | 4.18 | 23.68% | | 🥈 2nd | LSTM | 6.29 | 4.54 | 25.12% | | 🥉 3rd | XGBoost | 5.72 | 4.52 | 29.10% | | 4th | Prophet | 11.50 | 8.03 | 57.35% | ### 3-Year Forward Forecast (ARIMAX) | Year | Forecast | 95% Confidence Interval | |------|----------|------------------------| | 2025 | 14.34% | 5.76% – 22.92% | | 2026 | 13.97% | 4.02% – 23.91% | | 2027 | 13.75% | 3.38% – 24.11% | --- ## 🗃️ Data Sources | Source | Data | API | |--------|------|-----| | World Bank | Zambia CPI Inflation (FP.CPI.TOTL.ZG) | `wbgapi` | | World Bank | USD/ZMW Exchange Rate (PA.NUS.FCRF) | `wbgapi` | | Yahoo Finance | Copper Spot Prices (HG=F) | `yfinance` | **Dataset:** 25 years (2000–2024) · 3 features --- ## 🔬 Methodology ### Phase 0 — Data Collection - Pulled 3 macroeconomic indicators from World Bank API and Yahoo Finance - Merged into a single aligned annual time series (2000–2024) ### Phase 1 — EDA & Preprocessing - Visualized all 3 variables over time - Correlation analysis: Copper prices show -0.66 correlation with inflation - Augmented Dickey-Fuller stationarity tests - Chronological train/test split (2000–2019 train | 2020–2024 test) ### Phase 2 — Modeling - **ARIMAX** — ARIMA with exogenous regressors; auto-tuned via AIC - **XGBoost** — Lag features + rolling windows + exogenous variables - **Prophet** — Trend + ex …