ARIMA Vs VARMA - Modelling and Forecasting of India’s Cereal Production

Abstract

In agriculture, data on various parameters such as area, production and yield are collected over time. These data collected over time are modelled using various time-series modelling techniques. In this paper, an attempt is made to model time-series data of two important food commodities viz. Rice and Wheat using Autoregressive Integrated Moving Average (ARIMA) model and its multivariate variant Vector Autoregressive Integrated Moving Average (VARMA) model. The VARMA models are advantageous over the ARIMA models since two or more series can be modelled simultaneously besides capturing the relations between different series. The performance of ARIMA and VARMA models are compared using the measures of accuracy. Time-series data on production of rice and wheat for the period 1965-2017 is utilized for modelling and forecasting using ARIMA and VARMA statistical time-series modelling techniques. It was observed that the multivariate VARMA modelling technique is not an alternative to the univariate ARIMA modelling technique in terms of efficiency since the production of these two commodities are independent of each other. Finally, forecasting of rice and wheat production for the year 2020 was carried out and is found out to be 114 million tonnes of rice and 106 million tonnes of wheat. An increase of 4.5 % in rice production and 8.8 % in wheat production over the current production values are forecasted for the year 2020. Forecasting for future years is essential as this would help the planners in planning for eventualities arising due to vagaries of monsoon such as floods or droughts.