Study On Shanghai Composite Index Prediction Based On Improved Wavelet Neural Network

Withthe investment activitiesin the stock marketto be increasingly frequent, the stockmarket urgently needs an effective prediction method to help people increaseinvestmentincome. The stock market is a highly complex financial market, and its variationsubject to a variety of factors such as macro and micro、politicaland economic、domestic andforeign、behaviorand psychological,which makeit hard for the peopleto find the"real"variationof stock market. The traditional prediction methods used in stock index prediction are thestatistical regression method and time series prediction method. Statistical regression methodis a causal relationship between the input and output variables, in which the variablesrequired to meet certain statistical assumptions; Time series method are based on time series’inertia deduction,and it must ascertain or presumed the time series’ variation. Because theseconditions are difficult to meet in practice, these two traditional prediction models can hardlymeet the requirements ofstock market prediction. In recent years, artificialintelligenceprediction system begin to be applied to predict the stock index and have a rapidgrowth in use, in which the neural networks isthe representative. However, due to the singleintelligent prediction methodhas different kinds of flaw and shortages，the prediction effectare always subject to a certain extent. Therefore, when establish the prediction model in theapplication of intelligent technology, people start to combinevarious intelligenttechnologieswith each other thus they can complement each other in order to achieve betterprediction effect.Thesuperior performanceof wavelet neural networkmake it widely used in the economicfield, many scholars have establishthe stock prediction model based on wavelet neuralnetwork. Although it can have an improvedpredictionresults compare to general neuralnetwork, study algorithm of wavelet neural network still hasdeficiencyof localconvergence.For this reason, this study adapt bacterial foraging optimization algorithm and anadaptive inertia weight particle swarm optimization to further optimize waveletneuralnetwork.The method is to use these two optimization algorithms as learning algorithmsofwavelet neural network. Then the optimized wavelet neural networkcan get better networkweights、pancoefficientsand scaling coefficients of wavelet function. This can make the entirenetwork more reasonable、get better generalizationability、better searchingabilityand betterprediction performance.Then establish the improved wavelet neural networkprediction model based ondata samples of Shanghai Composite Index. And comparetheir predictions results with the BP neural network prediction model、 wavelet neural network prediction model and BFO-BP neural networkprediction model.Results show that the BFO wavelet neural network prediction modeland theAIW-PSOwavelet neural network prediction model have betterprediction effects than thethree existing models.