As important components of production raw materials and daily necessities, predicting the future long-term price trends of agricultural products is of great significance. The prices of bulk agricultural commodity usually fluctuate widely and have complex relationships with multiple factors, which makes long-term price forecasting difficult. We propose a long sequence time-series forecasting method based on the Multi-decomposition Attentional Gate Recurrent Unit (MAGRU) model, addressing two issues of existing bulk commodity price forecasting methods: weak response to price volatility and inadequate utilization of other covariates related to prices. The cotton price index is used as an example to study and analyze long sequence time-series forecasting algorithms. We select six covariates from three perspectives to improve the quality of prediction. Experimental results show that our model achieved an average decrease of 49.6%, 60.4%, 51.2%, and 36.8% in MSE compared to LSTM, TCN, GRU, and DLinear, respectively. This effectively reduces errors in predicting long-term prices of bulk commodities.