Forecasting global monthly cotton prices: the superiority of NNAR models over traditional models.

Accurate forecasting of agricultural commodity prices is essential for informed decision-making by farmers, traders, and policymakers. This study evaluates and compares the predictive performance of traditional statistical and machine learning models in forecasting global monthly cotton prices. Price volatility and nonlinear patterns in cotton markets present challenges for conventional models such as the Auto Regressive Integrated Moving Average (ARIMA), which often fail to capture complex dynamics. The novelty of this research lies in systematically comparing traditional statistical models (ARIMA, ETS, STL, TBATS, Theta), machine learning models (Neural Network Auto-Regressive [NNAR]), and hybrid approaches to determine the best forecasting tool. Performance was evaluated using Root Mean Square Error (RMSE), Mean Error (ME), Mean Absolute Error (MAE), Mean Percentage Error (MPE), and Mean Absolute Percentage Error (MAPE). Results revealed that the NNAR (26, 1, 14) [12] model outperformed all models, achieving the lowest RMSE (1.16383774), MAE (0.832275572), and MAPE (1.19%), indicating high predictive accuracy and minimal bias. The 30-month forecast for cotton prices using the NNAR model indicates fluctuations between approximately $0.66 and $0.74 per pound, following a cyclical pattern without a clear long-term trend. These findings highlight the strength of advanced machine learning techniques, particularly NNAR, in capturing complex nonlinear patterns, improving forecasting reliability, and supporting effective decision-making in volatile cotton markets. This study provides practical insights for stakeholders seeking to anticipate cotton price changes and make informed decisions in the global market.