The Price Forecasting Research Of Financial Products Based On Improved BAS-Elman Neural Network

Finance is the basis of the normal operation of social life and is closely related to everyone,and the transaction of various financial products is the carrier of financial operation.The prediction of its price trend plays an important role in people's investment,consumption and risk prevention as well as the formulation of monetary and fiscal policies by government sector.At present,the difficulty of using artificial neural network model to predict the price of financial products is that it is easy to fall into the local minimum,the training time is too long,the simulation accuracy is low and so on.Firstly,we compare and analyze the principles of various neural network models on the basis of studying the current financial product price forecasting methods.Considering the nonlinear dynamic characteristics of the financial product price,this paper proposes a financial product price forecasting method that combines the improved Beetle Antennae Search algorithm(BAS)and the Elman neural network with dynamic feedback characteristics.We make the search step change coefficient of BAS algorithm be iterated according to logsig function to optimize its search performance.Then write Particle Swarm Optimization algorithm(PSO),Genetic algorithm(GA),basic BAS algorithm and improved BAS algorithm programs to optimize the test function of Rosenbrock function respectively.Compare and analysis the simulation results and use the improved BAS algorithm with the best performance to optimize the initial weight threshold of Elman neural network.Building an improved BAS-Elman neural network to simulate the prediction of gold price and stock price,and compared with the prediction results of basic Elman neural network,GA-Elman neural network,Back Propagation(BP)neural network and other literature findings.Through many experiments and simulations by Matlab show that the mean square error(MSE)and relative error percentage(MAPE)mean were 0.94575 and 0.27032 respectively by the improved BAS-Elman neural network for 91 trading days.The average MSE and MAPE of the 51 stock trading days are predicted to be 3.39895 and 4.3399 respectively.Compared with other financial product price forecasting models,the method in this paper has effectively improved the accuracy of prediction,and the improved BAS-Elman neural network is effective and feasible in financial product price forecasting.