Futures Price Prediction Based On Long-short Term Memory Model

The futures of agricultural products occupy a large proportion in the futures market of our country.Corn futures,as the largest agricultural products futures varieties in the scale of trading in our country,play an important role in the futures market.It is of great significance to study the corn futures market,not only to make the policy of government macro-control and stabilize the market,but also to avoid the income risk of futures investors.At present,the research on futures market at home and abroad is mainly carried out from two aspects.On the one hand,it is based on the analysis of market fundamentals,such as the analysis of national policies,commodity supply and demand,speculative psychology and so on.Relying on the experience of the analyst,it is difficult to quantify the expression.On the other hand,it is based on the historical data of the market to construct a mathematical model to study the futures market.This kind of method is objective and persuasive,so this paper chooses this method to study the corn futures market.The price of corn futures market has the characteristics of nonlinear and high noise.The traditional statistical model is difficult to describe these properties.As a neural network with long-term memory ability,the long and short term memory model has shown excellent performance in dealing with nonlinear time series learning in recent years model.The long-short term memory is very sensitive to input.In this paper,a lot of work has been done on the input processing of the model,such as the standardization of the input data,the analysis of the abnormal values of the input data,the establishment of various types of technical indexes based on the historical data and the reduction of its dimension by the principal component analysis.Construct a set of input data sets suitable for long-and short-term memory model.In the output of the model,the aim of this paper is to forecast the settlement price of futures,but the price of settlement is fluctuating and not stable.In order to solve this problem,In this paper,the target is transformed into forecasting the fluctuation of corn futures settlement price,which can eliminate the very good volatility.After dealing with the input and target data sets,a long-term and short-term memory model of futures settlement price is built.The accuracy of forecasting settlement price on the test set is 92.06 and the numerical error is about 2/1000.