Stock Index Forecast with Back Propagation Neural Network Optimized by Genetic Algorithm
Wei Qin Shen, Mian Xing · 2009
Stock index forecast is not an easy job as it is subject to influence of various factors. Since 1980s, many researchers have used Back Propagation Neural Network BPNN to forecast stock price fluctuations. However, there are some limitations with BPNN. With slow convergent speed and low learning efficiency, BP learning algorithm is easy to get in local minimum and is far from being perfect in stock forecasting. The genetic algorithm is a sort of selfadaptive optimized search algorithm based on natural selection and natural inheritance. It can be applied in different areas of parameter space in the colony generation subrogation toward the optimal direction, which the search could easily find and couldn’t get in local minimization. In view of this, we adopt the genetic algorithm to train the BPNN to overcome the above shortcomings. By adding genetic algorithm we built an optimized stock index prediction model for Shanghai composite index. Through empirical analysis, we come to the conclusion that the above model optimized by genetic algorithm possesses better function approximating capacity, and has ideal result for the short-term stock index forecast.