AgTech: Volatility Prediction for Agricultural Commodity Exchange Trading Applied Deep Learning

The rapid advancement of computer science technology and artificial intelligence has generated heightened investor interest in quantitative trading, primarily attributable to its exceptional efficiency and consistent performance. This paper presents the development of a volatility prediction system for the agricultural commodity exchange trading domain. The system utilizes raw financial data as input and produces trading decision-support as output using a volatility prediction based on Generalized AutoRegressive Conditional Heteroskedasticity (GARCH) and Long Short-term Memory (LSTM) models. The main goal of the system is to enhance overall profitability through efficient management of trading losses. In addition, a denoising technique is utilized to reduce the influence of market noise and improve overall performance. The prototype has been trained and back-tested in the agricultural commodity market trading data from 2010-2023. The research findings suggest that the Multivariate Bidirected LSTM 2 layers model has the best accuracy of 91.38% in predicting the volatility of cotton commodities trading throughout the time frame spanning from September 22, 2023, to October 21, 2022. The GARCH model is widely utilized for volatility forecasting, but the Multivariate LSTM model has promise in offering investors a possible edge through enhanced forecasting accuracy.