Stock Market Forecast Research Based On Artificial Neural Network

The Stock market is an extremely complex non-linear dynamic system, and risk comes with income from investments. Therefore, the research of inherent laws and the forecast of stock market have an extremely important application value and theory significance. The investors have been exploring its inherent laws, seeking for the effective analysis methods and the tools to make a profit. But stock market prediction is a hard work, the stock price system is complicated in internal structure, polytropy in external factors, which make the traditional forecast methods and tools can't meet its need. In contrast, neural network has the capability to approximate any nonlinear system and the speciality of self- adapting and self- learning, can learn economical knowledge from its historic data, and is suited to solve problems in stock market prediction.This paper analysis the main problems in the process of stock price prediction, compares various stock forecasting methods, and discuss the feasibility of forecasting stock price trend by using BP neural network. BP neural network finds out the inherent law of stock market through learning historical data and store them in its weights to forecast the trend in the future.According to the principle of stock price prediction based on BP network, this paper establishes stock price prediction model with three-layer feedforward neural network, discusses the problem of network structure, number of hidden nodes, and choice and preprocessing of the network's input data, improves the generalization ability of neural network by using network optimization and regularization methods. The simulation experiment establishes prediction model using MATLAB software, forecasts the trend of shanghai stock index price in short-term, and contrasts the result of different methods of single step, multi-step and rolling forecast.