Research On Corn Futures Price Prediction

Agricultural products have always been one of the most important varieties in the futures market.Agricultural products were the earliest varieties that began futures trading in the world.In the early history of futures trading,agricultural products were not only the dominant varieties of futures trading,but also had large family members due to their long development time.As the largest variety traded in China,corn’s status is self-evident.For the vast number of corn growers in China,corn futures is a very important tool for them to ensure planting profits.Therefore,it is of great practical significance to study the price prediction model of corn futures.Due to the nonlinear and high volatility of corn futures prices,traditional linear models such as ARIMA and GARCH models cannot be accurately predicted.Than the linear model,nonlinear model with good prediction accuracy and variable nonlinear relationship between mining capacity,and LSTM neural network model is the most famous of these,the learning ability of mathematics in nonlinear time series is very good,so this paper choose the LSTM neural network model to forecast the corn futures prices.In this paper,we construct a LSTM neural network with a four-layer network structure and 40 hidden layer neurons.The Dropout structure is added to reduce possible overfitting problems of the model.In terms of input variables,this paper selects five factors including daily closing price of rice futures,daily spot price of corn,annual yield of corn in China,daily closing price of CBOT corn futures and daily spot price of corn starch by analyzing the fundamental factors that affect corn futures.For the evaluation of model results,in addition to the traditional root mean square error（RMSE）and mean absolute percentage error（MAPE）,the determination coefficient is also added to evaluate the fit and over-fit degree of the model.In the end,the model performed well in the test set,with RMSE=54.23,MAPE=1.86% and R₂2=0.957.Compared with SVR model and RNN model,the results also show certain advantages.