A Comparative study of ARIMA, feedforward neural network and hybrid ARIMA–FFNN models for stock price forecasting: evidence from Econet Wireless Zimbabwe (2020–2024)

Abstract

This study developed and compared the forecasting performance of the Autoregressive Integrated Moving Average (ARIMA), Feedforward Neural Network (FFNN), and Hybrid ARIMA–FFNN models using monthly closing stock prices of Econet Wireless Zimbabwe from January 2020 to December 2024. The ZiG/USD exchange rate and inflation rate were incorporated as explanatory variables in the FFNN and hybrid models. A quantitative, explanatory, and comparative research design based on the positivist paradigm was adopted. Forecast accuracy was evaluated using the Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The Hybrid ARIMA–FFNN model produced the best forecasting performance, achieving the highest explanatory power (R² = 0.6221) and the lowest forecast error (RMSE = 96.71), outperforming both the standalone ARIMA and FFNN models. The findings demonstrate that combining linear and non-linear modelling techniques improves stock price forecasting under conditions of economic instability and structural change. The study recommends the adoption of hybrid econometric and machine learning models to enhance forecasting accuracy in the Zimbabwean stock market.