Machine Learning and Macroeconomic Indicators for Predicting Consumer Goods Stock Prices in Nigeria

Nigeria’s fast-moving consumer goods (FMCG) sector is Africa’s most dynamic market; however, stock prices remain highly sensitive to inflation, exchange rate volatility, and oil shocks. Despite global advances in machine learning, sector-specific forecasting models for Nigerian equities are scarce. This study addresses this gap by developing hybrid models to forecast BUA Foods PLC stock prices and by evaluating the influence of macroeconomic predictors. A hybrid framework employing Mixed Data Sampling (MIDAS) integrated daily stock and Brent crude data with monthly macroeconomic indicators (USD/NGN rate, inflation, and MPR) from 2022 to 2025. Features included technical indicators (RSI and MACD) and lagged variables. We compared three models-univariate ARIMA (baseline), ARIMA–SVR, and ARIMA–LSTM-using walk-forward validation based on Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). SHAP values were used to provide model interpretability. The ARIMA–SVR hybrid proved superior, achieving an MAE of ₦1.28 and an RMSE of ₦1.66-an improvement of 82.3% and 77.2%, respectively, over the ARIMA baseline (MAE ₦7.24, RMSE ₦7.29). While the ARIMA–LSTM hybrid also outperformed the baseline (MAE ₦1.71), it lagged behind the SVR approach. SHAP analysis identified the USD/NGN exchange rate and oil prices as the most dominant predictors. Hybrid models, particularly ARIMA–SVR, significantly enhance forecasting accuracy in Nigerian consumer goods stocks by effectively capturing nonlinear macroeconomic dependencies. These findings demonstrate the value of integrating traditional time-series methods with kernel-based machine learning for volatile emerging markets, offering investors actionable insights into currency and commodity risks.