In-Season Price Forecasting in Cotton Futures Markets Using ARIMA, Neural Network, and LSTM Machine Learning Models

Jeffrey Vitale, John Robinson

This study explores the efficacy of advanced machine learning models, including various Long Short-Term Memory (LSTM) architectures and traditional time series approaches, for forecasting cotton futures prices. This analysis is motivated by the importance of accurate price forecasting to aid U.S. cotton producers in hedging and marketing decisions, particularly in the Texas Gulf region. The models evaluated included ARIMA, basic feedforward neural networks, basic LSTM, bidirectional LSTM, stacked LSTM, CNN LSTM, and 2D convolutional LSTM models. The forecasts were generated for five-, ten-, and fifteen-day periods using historical data spanning 2009 to 2023. The results demonstrated that advanced LSTM architectures outperformed other models across all forecast horizons, particularly during periods of significant price volatility, due to their enhanced ability to capture complex temporal and spatial dependencies. The findings suggest that incorporating advanced LSTM architectures can significantly improve forecasting accuracy, providing a robust tool for producers and market analysts to better navigate price risks. Future research could explore integrating additional contextual variables to enhance model performance further.