Hybrid ARIMA-machine learning models for predicting inflation and economic growth in Nigeria

Abstract

Accurate forecasting of inflation and economic growth remains a critical policy challenge for Nigeria, where volatile macroeconomic conditions, structural breaks, and persistent shocks have rendered traditional univariate methods increasingly unreliable. This paper develops and evaluates a family of hybrid ARIMA-machine learning models for predicting Nigeria's consumer price index (CPI) and gross domestic product (GDP) growth rates using monthly data from January 2010 to December 2024. The hybrid framework decomposes each series into a linear component captured by ARIMA and a nonlinear residual component captured by one of three machine learning algorithms, i.e. artificial neural network (ANN), support vector regression (SVR), and random forest (RF). Rolling-window out-of-sample experiments demonstrate that all hybrid models substantially outperform their standalone counterparts. The ARIMA-ANN hybrid achieves the lowest root mean square error (RMSE) of 0.324 for inflation forecasting, compared with 0.517 for standalone ARIMA and 0.468 for standalone ANN. For GDP growth, the ARIMA-RF hybrid achieves RMSE 0.187, a 34% improvement over ARIMA alone. Diebold-Mariano tests confirm that all forecast improvements are statistically significant at the 1% level. The results support hybrid modelling as a viable path toward more reliable macroeconomic forecasting in emerging economies characterized by structural instability and regime switching.