Forecasting production and import of maize in Nepal using the Box-Jenkins ARIMA and ARIMA-ETS models

Maize is the second most important crop in Nepal in terms of area and production. Despite the increase in production, the heavy import of maize in recent decades has become a policy concern in Nepal. The forecasting of maize production and imports was made from historical data from 1997 to 2024 by evaluating the Box-Jenkins Autoregressive Integrated Moving Average (ARIMA) model and the hybrid ARIMA-Error Trend Seasonal Exponential smoothing (ARIMA-ETS) model. The best-fit models were chosen based on the Akaike information criteria (AIC), Bayesian information criteria (BIC), and forecast accuracy criteria. The diagnostic test results were also analyzed to check the best-fit model. ARIMA (1,2,1) and ARIMA (1,2,1)-ETS models for production and ARIMA (1,1,2) and ARIMA (1,1,2)-ETS models for import were evaluated. Both out-of-sample cross validation and rolling/expanding window cross validation indicated that simple ARIMA (1,2,1) outperformed the hybrid ARIMA (1,2,1)-ETS for production. In contrast, hybrid ARIMA (1,1,2)-ETS showed slightly better accuracy than ARIMA (1,1,2) in the case of import. The forecasts for 2025–2029 indicate a consistent rise in maize production, while imports are projected to remain high but relatively stable across both simple ARIMA and hybrid models. The findings reveal Nepal’s persistent reliance on imported maize and imply the need for strategic planning to enhance maize production and reduce import dependency. The study findings are expected to be valuable to the policymakers working in public and private sectors as well as to the researchers who are interested in advanced forecasting techniques.