The use of a univariate time series model to short term forecast the behaviour of beef production in Baja California, México

Abstract

The Box-Jenkins methodology was used to select an ARMA model to forecast beef production in Baja California, Mexico. The series of bovine carcasses processed monthly in the state's slaughterhouses between 2003 and 2010 was used. Because the inspection of the series graph and correlogram did not determine a stationary behavior, an augmented Dickey-Fuller test was performed and it was found that the series was stationary. As a result of identifcation procedure, an AR (1) and an ARMA (2, 1) models were selected and estimated using ordinary least squares. The estimated models were compared using the signifcance of the regression coeffcient and the Akaike information and Schwartz Bayesian criteria. A diagnostic check was done examining the goodness of ft of the models by plotting the residuals; the Q statistic was used to test for autocorrelation. Because the results were similar, a predictive effcacy evaluation of two models was carried out using a group of forecast error statistics. The result of these tests indicated that the ARMA (2,1) had a better forecasting capability, this was supported by plotting together a forecasted series with the actual series and the out-of sample prediction for January of 2011. The results support the use of ARMA models to obtain reliable short term forecasts of beef production in Baja California.