Futures Price Prediction Based On RNN-SVM Model

The futures market has always occupied an important role and dominant position in the current international financial market.With the rapid prosperity and development of domestic and overseas futures markets,people have paid great attention to the long-term fluctuations of futures prices and their trends,analysis,and market risk prediction.More and more,a reasonable analysis and prediction of stock price fluctuations in the futures market can effectively avoid market investment risks and obtain huge investment returns and returns.However,futures price fluctuations are generally difficult to predict in nature,affecting futures market stocks.There are many decisive factors for price fluctuations and trends,such as the time and trend of price changes in the futures market,the ratio of supply and demand to price changes,economic data changes,financial statements,and historical transaction data in the futures market.From the perspective of basic research in economic statistics and from a viewpoint,the commonly used methods for forecasting futures price fluctuations mainly include time series forecasting models,gray time series forecasting models,and neural network forecasting models.This paper takes the domestic futures market as the main research object and uses a deep-based machine.Learning method theory,with a Shallow structured machine learning algorithms are combined for comparison,and a model of RNN-SVM futures price fluctuation prediction is established,which provides certain data reference and reference for future investor screening.The main work of this article is as follows:Firstly,we construct a futures price time series matrix and select four futures products:Bean Main Link,Cotton Main Link,Gold Main Link,and Crude Oil Main Link.Each group has a total of 401 sample data.Therefore,after obtaining the opening price data and related variable indicators,the data is normalized and correlated,and finally the time sequence that can be used as the model input is obtained.Then four dimensionality reduction methods are introduced,namely MDS,PCA,LLE and t-SNE,there is a great correlation between the indicators selected by the Pearson correlation coefficient test,indicating that the subsequent dimensionality reduction of the data is reasonable and effective.Secondly,we establish the RNN-SVM combination model,first introduce the characteristics and algorithm process of the RNN network,and propose to use the RNN self-encoding network features to directly extract the features of the original data sequence,construct the timing difference,and MDS,PCA,LLE and t-SNE algorithms.The results arecompared and analyzed,parameter optimization and adjustment are continued as well,training is repeated until the model converges,and the last layer of sensor is converted into SVM.The extracted feature are input,and the SVM result is used as the output.We choose the mean square error and square correlation coefficient as the prediction results measuring indicators,it is found that the square correlation coefficients of the combined model for the four futures products are 0.9932,0.9923,0.9891 and 0.9765,which shows that the RNN-SVM combination model predicts the opening price of futures is reasonable.Finally,the RNN-SVM combination model is compared with the prediction results of RNN,BP,SVM and random forest.Compared with the feature extraction prediction results of RNN model and MDS,PCA,LLE and t-SNE dimensionality reduction algorithm,it is found that the combination model effect proposed best in this paper.It shows that the combined model has good predictive ability,which indirectly illustrates the effectiveness of the combined model.