The forecasting efficiency gains obtained by building time series models in which the data are optimally aggregated have been studied from a theoretical perspective in numerous studies. However, an empirical study focused on the potential benefits of temporal disaggregation in commodity price forecasting has not been conducted. This is the case even though commodities markets are extremely important for the economic performance of the U.S. agricultural sector, where a slight difference in a prediction represents losses of million of dollars. One important commodity is cotton, which generated approximately $25.0 billion in annual revenue and was responsible for 200,000 jobs in 2008 (USDA, 2012). This study evaluates the efficiency gains in forecasting cotton cash prices using alternative ARMA models with varying levels of temporal aggregations (daily, weekly, monthly and annual). More specifically, it evaluates whether the disaggregated models can produce more accurate aggregated price predictions than the corresponding aggregated models. Likewise, this is the first study that incorporates the daily level of aggregation to evaluate the efficiency gain in forecasting.The dataset consisted of approximately 60 years of daily cotton prices (9,120 observations from 1972-2010) in which the prices were adjusted using the Consumer Price Index (CPI). The results suggested that overall, disaggregation leads to gains in efficiency; which would be consistent with the results of the theoretical studies of Tiao (1972), Koreisha and Fang (2004). Finally, the weekly model was the most efficient in forecasting the cotton prices. These results are important for cotton farmers because it could lead to better investment and hedging strategies.