A Comparative Statistical Approach for Forecasting Maize Yield in India Using ARIMA Models
Manish Kumar *
Department of Agricultural Statistics, Acharya Narendra Deva University of Agriculture & Technology, Ayodhya, India.
Shiv Kumar Rana
Department of Agricultural Statistics, Acharya Narendra Deva University of Agriculture & Technology, Ayodhya, India.
*Author to whom correspondence should be addressed.
Abstract
This paper investigates the scenario of maize yield in India using several conventionally developed autoregressive integrated moving average (ARIMA) models. The forecasting performances of the generated models were evaluated using akaike information criterion (AIC), root mean square error (RMSE) and mean absolute percentage error (MAPE). The best fitted model was ARIMA(2,1,0) with drift, having AIC value of 894.95, RMSE value of 148.05, and MAPE value of 7.71%. Additionally, a comparative performance assessment of the fitted models were made with the automatically generated model viz., ARIMA(1,1,2) with drift, which was obtained on using auto.arima () function in R-studio. Furthermore, the Ljung-Box test was performed for diagnostic checking of residuals of the generated models. The results of the analysis revealed that ARIMA(1,1,2) with drift model was slightly more precise as compared to ARIMA(2,1,0) with drift. The forecast values of maize yield for five consecutive years were obtained with 80% and 95% prediction intervals using ARIMA(1,1,2) with drift model. The findings of the study revealed that the trend of maize yield is significantly rising over the recent years, which is a good sign for policymakers and scientists regarding development of strategies pertaining to global food trade and nutritional security.
Keywords: ARIMA, time series, stationarity, autocorrelation, residual