Empirical Research On The Stock Price Prediction Based On Neural Network Trained By GA-LMBP Algorithm

With the continuous development of the market economy system,the role of the stock market on the state,company and individual has become increasingly prominent.The stock market reduces transaction costs and provides liquidity.What's more,it can fix the pricing of securities through the supply and demand relationship between buyers and sellers.The price of securities affects the allocation of social resources,reflects the company's operating conditions,and guides investors to choose the trading behavior that is most suitable for them.Therefore,predicting the price of stocks will help market participants to make better decisions and help investors achieve asset appreciation.The Chinese stock market has not yet reached full efficiency;the stock price does not include all relevant information on the market,so the stock price is still predictable.In order to predict changes in stock prices accurately,models that can characterize the nonlinear characteristics of stock prices need to be used.The neural network model is inspired by the cognitive processes of the human brain,imitating the human nervous system,and has the ability to learn nonlinear mappings between input and output.The traditional BP neural network uses the steepest descent method to train the network.The training process may stop at the local optimum and the convergence speed and stability of the model are difficult to satisfy at the same time.The LMBP algorithm combines the characteristics of the Newton method and steepest descent method,but the solution of the model may still be trapped in the local optimum.Genetic algorithms allow individuals with higher levels of fitness to have greater opportunities to retain information to the next generation through selection,crossover,and mutation operations.In each iteration,the genetic algorithm introduces randomness,allowing the model to have global search capabilities.Even if the initial value falls near the local optimum,the model can let the algorithm continue its search at other positions through genetic operations with randomness.In order to improve the effect of neural network model to predict stock price and enhance the global search ability of LMBP algorithm,this paper adds the genetic algorithm to the neural network training process and composes the GA-LMBP algorithm.The algorithm firstly uses the global search ability of genetic algorithm to train the neural network to obtain the optimal individual.Then the initial weight of the neural network is set as the optimal individual obtained from the genetic algorithm,and the neural network is trained using the LMBP algorithm.In order to test the effect of the model on predicting the stock price,this paper uses the data from 2010 to 2017 of 5 stocks from different industries in the SSE 50 constituent stocks to conduct an empirical test.The input variables for the sample set are 13 technical indicators,and the output variable is the stock closing price for the following day.The results show that the GA-LMBP algorithm has the most lowest prediction error for the Shandong Gold Mining Co.,Ltd,with a MAPE of 1.22%.Comparing the results of GA-LMBP and LMBP for all five stocks,it can be found that the MAPE obtained by GA-LMBP algorithm is the lowest,indicating that the addition of genetic algorithm effectively reduces the prediction error of the model.At the same time,when using the LMBP in price prediction,the phenomenon that the predicted value and the actual value deviate greatly within a short time is also observed.However,these deviations have alleviated in the predictions of the GA-LMBP algorithm,indicating that the stability of GA-LMBP prediction is higher.Therefore,according to the experimental results,due to the addition of the global search ability of the genetic algorithm,the neural network model trained using GA-LMBP can learn the nonlinear mapping relationship between input and output better,and have lower prediction error and higher prediction stability.