Optimizing multiple regression model for rice production forecasting in Nepal

Abstract

This research, testing the possibility of use of probable predictors, has optimized multiple regression model to be used for rice production forecasting in Nepal. Fifty years (1961-2010) time series data were divided into training sample (a sample which is used to build the model) (n=35), and test sample size of 15 through which the built model was cross validated for its reliability in forecasting. This research has explored and used all the underlying principles of linear regression model building and its application in forecasting the production, mainly crop production such as rice. The model sustained with the three principle predictors: harvested area, rural population and price at harvest whereas these variables could explain 93% variation in production; the forecast variable. The model as such was parsimonious and as well the good fit with minimal (5%) mean absolute percentage error in its forecast. It therefore, for this fit, was concluded that multiple regression model could be scientifically used in forecasting, and the concerned stakeholders could thus be benefited from the this model especially for the enhanced ease, and efficiency for rice production forecasting to be used for planning purpose at national level. Future work might consider to increase the precision of the model in any aspects like making it more parsimonious and reliable than which have been purposed in this study.