Comparing the Forecasting Power of Multivariate Var and Univariate Arima Models: A Case of Groundnut Prices in Bikaner District of Rajasthan Kwasi Bannor Richard, Ph.D Scholar, Sharma Amita, Assistant Professor Institute of Agribusiness Management, SK Rajasthan Agricultural University, Bikaner, India Online published on 10 March, 2015. Abstract This study explored modelling and forecasting of wholesale groundnut monthly prices in Bikaner district of Rajasthan using the Autoregressive Integrated Moving Average (ARIMA) model and Vector Autoregressive model (VAR). Augmented Dickey fuller test, DF-GLS test and Philips Perron test were used to test for unit root in price series. Autocorrelation (ACF) and Partial autocorrelation (PACF) functions were estimated to assist in deciding the appropriate orders of the Autoregressive model of order p (AR) and Moving Average model of order q (MA). The BIC test was conducted because we were considering several ARIMA models and the model (0, 1, 0) which had the lowest BIC value of 11.612 with R square figure of 84.7% and the mean absolute percentage error of 7.602% was chosen as the best fit model. Vector Autoregressive model used in Model factored in Jaipur, Jodhpur monthly groundnut market prices and cumulative rainfall and rain days per month over the period of 2005–2014. VAR model showed high forecasting error with MAPE of 0.769% compared to ARIMA model of 2.745%. Indiacting, VAR model has high forecasting accuracy compared to ARIMA. The Lagrange Multiplier test of VAR showed that there is no serial correlation in the model whereas Jacque-Bera test also indicated, the residuals of the VAR model is normally distributed indicating a good fit model. Top Keywords ARIMA model, Vector Autoregressive model, Autocorrelation function, Forecasting accuracy, Bikaner, Groundnut, Mean Absolute Percentage Error. Top |