Application of time series models in forecasting maize crop production in Tanzania

This study focused on application of time series models in forecasting maize crop production in Tanzania using historical maize crop production covering the period 1961-2010. The specific objectives were to identify an appropriate time series model for forecasting maize crop production and to give 15 years ahead forecasts of maize crop production from 2011 to 2025. In the first stage of time series analysis, a plot of maize crop production was characterized with linear trend. For that case various models were applied including Linear trend, Double moving average, Quadratic trend, Brown’s linear exponential smoothing, ARIMA (0,1,1) with constant, ARIMA (1,0,1), ARIMA (2,1,0) with constant, ARIMA (3,0,0) and ARIMA (2,1,2). For the Box Jenkins ARIMA methodology, autocorrelation function (ACF) and partial autocorrelation function (PACF) were used to detect stationarity as well as identifying the number of parameters in the model. The models were estimated and followed by diagnostic checking using Box-Pierce Test. Finally all eight (8) models were evaluated for the forecast accuracy using various criteria which involved Akaike’s information criteria (AIC). The linear trend model was found to be the appropriate model. Using a liner trend model, the quantity of maize crop production was forecasted to be 3,900,010 tonnes for the year 2013 and 4,729,660 tonnes for the year 2025 with lower and upper limits of 3,672,660 and 5,786,670 respectively. Therefore applying a linear trend model in forecasting maize crop production will help in foreseeing production shortfalls and thus facilitate proper food policy decisions.