Modeling and forecasting wholesale gasoline prices in Tanzania using ARIMA and neural network autoregressive models

Abstract

Fluctuations in wholesale gasoline prices significantly impact the economy of import-dependent countries like Tanzania, making accurate forecasting essential for policymakers and market participants. This study forecasts monthly wholesale gasoline prices in Dar es Salaam by comparing the Autoregressive Integrated Moving Average (ARIMA) model with a Neural Network Autoregressive (NNAR) model. Using data from January 2015 to December 2024 from the Energy and Water Utilities Regulatory Authority (EWURA), the models were evaluated. The ARIMA(1,1,1) model was identified as the best-fitting linear model, but the NNAR(11,7) model demonstrated superior accuracy. On the holdout test set, the NNAR model achieved a lower Mean Absolute Percentage Error (MAPE) of 2.3447% and a Root Mean Square Error (RMSE) of 76.6797, compared to the ARIMA model's MAPE of 2.049% and RMSE of 95.8016. The findings indicate that the NNAR model is more effective at capturing the complex and non-linear patterns in gasoline prices.