To forecast volatility in global food commodity prices, in this paper a number of alternative competing models are employed, thin tailed normal distribution, and fat-tailed Student t-distribution GARCH models, beside a simple approach of forecasting volatility based on standard deviations over the previous months as a forecast of future volatility. Our results indicate the t-distribution model outperforms the other two approaches, whereas the simple standard deviation approach outperforms the normal distribution model, suggesting that the normality assumption of residuals which often taken for granted for its simplicity may lead to unreliable results of conditional volatility estimates. The paper also shows that some of the food commodity prices included in the study, such as wheat, rice, and beef exhibit long memory behavior, implying persistence of the effect of a shock for longer periods compared to other commodities in the group. The evidence of long memory process supports the view that structural changes in demand and supply side factors are more effective than short-term speculative factors.

(As provided by authors).