Forecasting Indian Poultry Production: An ARIMA Modeling Approach

Abstract

Reliable forecasts of agricultural production support food security planning and policy making. This study demonstrates the application of autoregressive integrated moving average (ARIMA) models for forecasting monthly poultry production in India. Monthly national production data from to 1990-2020 were modeled following standard time series procedures. The non-stationary data exhibited an upward trend, so log-transformation and differencing induced stationarity for fitting an ARIMA model. Based on the inspection of the autocorrelation and partial autocorrelation plots, an ARIMA(1,1,1) model was selected and fitted. Diagnostic testing indicated model adequacy with normally distributed residuals displaying no significant autocorrelations. The in-sample fitted values closely matched the original data, suggesting good model fit. However, long-term forecasts resulted in implausibly high estimates after reversing the data transformations. This highlighted issues with model extrapolation, likely due to the inherent nonstationary upward trend that violates the ARIMA assumptions. While the model showed merit for short-term forecasts, alternative methods such as Holt-Winters exponential smoothing may improve long-term predictions by intrinsically accounting for trends. Overall, the analysis demonstrated key steps in ARIMA forecasting, including stationarity assessment, model identification, and diagnostics, and also illustrated the importance of considering data properties and forecast horizon when selecting suitable models. The results provide an applied example of agricultural time-series modeling using ARIMA, while noting limitations and recommending future evaluations using methods tailored for trended data. This can support enhanced predictability in planning India’s poultry production. The learning of rigorous model selection and critical evaluation also helps to advance best practices for time-series analysis of agricultural data.