Forecasting the Annual Rice Production in Nepal Using the Box-Jenkins ARIMA Modelling Process

Abstract

Rice is one of the main staple food crops in Nepal with the production level insufficient to meet the domestic demand of the country. The study tries to determine the best fitted ARIMA model for forecasting purpose by employing the Box Jenkins methodology, using the FAOSTAT dataset from 1961-2023. Multiple ARIMA models with the order of Autoregressive (AR) and Moving average (MA) ranging between 0 to 2 were considered. The data set of annual rice production was found stationary at the first differencing.  ARIMA (0,1,1) model was selected as the best model based on criteria such as adjusted r square, standard error of regression, Akaike Information Criteria, and Schwarz Information Criteria. Parameter estimation revealed the significance of the first lagged value of moving average at 1%, which indicates the model’s effectiveness in explaining the data for forecasting. The diagnostic checking of the ARIMA (0,1,1) model indicated that the residuals are random and the model is good fit for data. The Chow break point test indicated no structural break in the model. Hence, ARIMA (0,1,1) model can best capture the annual rice production in Nepal. Further, the model was used for in-sample and out-sample forecasting which revealed that the model had consistent and low mean absolute percentage error. The forecast was extended for the periods from 2024-2030, which predicted an average annual growth rate of 0.983%, indicating the positive increment in the annual rice production. Policy makers should leverage accurate forecast to enhance rice production and ensure food security through broader structural reforms and supportive policy interventions.