What the study found
Adding machine learning components to Vector Autoregression (VAR) models improved forecasting accuracy for several macroeconomic indicators across the eight African economies studied. The strongest gains were reported for inflation and foreign direct investment (FDI), while exchange rate results were mixed.
Why the authors say this matters
The authors conclude that hybrid VAR–machine learning models can better capture nonlinear macroeconomic relationships and may be relevant for economic planning and policy formulation in African economies. They state that these models are more effective for predicting inflation and FDI.
What the researchers tested
The researchers used annual data from 1990 to 2024 from the World Development Indicators for Nigeria, South Africa, Egypt, Angola, Morocco, Ethiopia, Tanzania, and Mozambique. They compared baseline VAR models with hybrid models that added Random Forest, Multilayer Perceptron (MLP), and XGBoost, and evaluated performance using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).
What worked and what didn't
For inflation, VAR+MLP gave the largest error reductions in Nigeria, South Africa, and Morocco, including a drop in South Africa's MAE from 1.471 to 0.888. FDI predictions improved markedly across nearly all countries, with Tanzania's MAE falling from 0.534 to 0.184 under VAR+XGBoost. For government expenditure, hybrid models outperformed VAR in Angola, South Africa, Morocco, and Cameroon, while exchange rate performance was mixed and traditional VAR did better in more stable economies such as Nigeria and Morocco.
What to keep in mind
The abstract does not describe detailed limitations beyond the mixed results for exchange rates and the country-specific differences in performance. The study is limited to the eight countries and the annual period from 1990 to 2024 described in the abstract.
Key points
- Hybrid VAR–machine learning models improved forecasting accuracy overall.
- The biggest gains were reported for inflation and foreign direct investment.
- VAR+MLP performed especially well for inflation in Nigeria, South Africa, and Morocco.
- VAR+XGBoost reduced FDI forecasting error in Tanzania from 0.534 to 0.184 MAE.
- Exchange rate forecasts showed mixed results, with traditional VAR sometimes performing better.
Disclosure
- Research title:
- Hybrid VAR models improved forecasting for several macroeconomic indicators
- Authors:
- R. N. Okafor, H. O. Obiora-Ilouno, J. O Odum
- Institutions:
- Nnamdi Azikiwe University
- Publication date:
- 2026-02-24
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.