Forecasting of Maize Yield in Bihar through Auto Regressive Integrated Moving Average (ARIMA) Models

The experiment was conducted during March, 2023 at Dr Rajendra Prasad Central Agricultural University, Samastipur, Bihar, India based on secondary maize yield data from 1990 to 2021, sourced from the Department of Economics and Statistics and India Agristat databases to investigate the forecasting of yield of maize. Various ARIMA models were developed based on Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) plots at different lags. Data from 1990 to 2019 were used for model development, while data for 2020 and 2021 were reserved for validation. Among several ARIMA models tested-namely ARIMA (0,0,1), ARIMA (1,0,0), ARIMA (1,0,1), ARIMA (0,1,1), ARIMA (1,1,1), ARIMA (0,1,2), and ARIMA (2,0,1)-the ARIMA (2,0,0) model provided the best fit for forecasting maize yield in Bihar. The significance of model parameters was assessed, and diagnostic checks, including tests for model adequacy, invertibility, stationarity, and forecast accuracy (MAPE, MAE, RMSE, % forecast error, and BIC), were conducted using t-tests and chi-square tests. The forecasted maize yields for the years 2021, 2022, 2023, and 2024 were 357 kg ha^-1, 358 kg ha^-1, 355 kg ha^-1, and 358 kg ha^-1, respectively. The forecast errors for 2021 and 2022 were 31.7% and 32.3%, respectively. These results suggested that the ARIMA(2,0,0) model was a reliable tool for short-term yield forecasting of maize in Bihar. The final ARIMA model used for forecasting was represented by the following equation: Z_t–Z_t-1=22.810+0.29 (z_t-1-z_t-2)+0.20 (z_t-2-z_t-3)+a_t.