Temporal and Spatial Attention Network Model Based Evolution Model for Bulk Commodity Price Fluctuation Risk

Abstract

Various factors have a dynamic and nonlinear influence on the price fluctuation of bulk commodities, to illustrate, market supply and demand, raw material prices, downstream product prices, seasonal factors, international prices, and the macroeconomic environment. This paper sets up a risk evolution model based on Graph Multi-Attention Network (GMAN) to enhance the bulk commodity price volatility prediction and price fluctuation risk measurement. Node2vec, applied as graph representations learning algorithm, is used to map the relationship between geographic location, market supply and demand, macroeconomic environment, and price volatility. Furthermore, the spatio-temporal attention mechanism is exerted to analyze the evolution of price fluctuation combining spatial elements and time series. A transform attention layer is applied to simulate the direct relationship between historical and future time steps. It converts the encoded historical price volatility to generate a sequential representation of future time steps. This paper utilizes the soybean electronic trading dataset provided by the demonstration unit. The experimental results show that the model can effectively predict soybean price volatility, influencing the assessment and early warning of soybean price fluctuation risk.