Research On Roles Of Trading Volume And Feature Extraction In The Short-term Prediction Of Stock Market

There are still many researches on stock market prediction in recent years.And improving the prediction accuracy is the goal of scholars.The methods to improve the prediction accuracy,in recent researches,mainly focus on two aspects: the improvements of prediction models and the feature extraction of the input data.Hybrid prediction models reduce the error brought by a single forecasting model.They had made good progress,but the results of existing neural network models depend on the input data.Noise among financial datum also has significant impacts on the prediction effects.The feature extraction methods used for financial data now are signal processing or simple dimension reduction analysis.There is no definite economic significance.Therefore,how to extract the feature variables with economic significance from the perspective of finance and verifying the roles of feature extraction in the short-term prediction of stock market are still the focus.The research on the relationship between volume and price has been paid much attention by scholars.A large number of studies show that there is a positive correlation between trading volume and price changes.Many scholars also give explanations from the aspect of information and micro-structure.However,there is no consistent conclusion on whether the volume is helpful to the prediction of price and the rate of return.Therefore,further studies on the effects of trading volume on the short-term prediction of stock market are necessary.Rate of return is the target of forecasts in this paper.The paper uses BPNN as the predict model and mean square error,mean absolute percentage error as evaluations.The main work of this paper are as follows:First,we use wavelet analysis to de-noise the closing price.We also construct comparative models to test whether the wavelet de-noising is helpful to improve the accuracy of prediction.Then,a prediction model based on the back propagation neural network is constructed to check whether the trading volume has an effect on the yield forecast.Finally,the paper proposes 21 feature variables from the view of relations between volume and price.And they are used to predict the rate of return.These are used to explore the roles of feature extraction in prediction.Four prediction models are constructed,analyzed and compared in this paper.The results show that feature extraction,volume and wavelet de-noising can reduce the prediction mean squared error,making prediction to achieve better effects.The conclusions are as follows: First,the prediction results of the de-noised closing price are better than that of the original closing price.The results show that the wavelet de-noising can not only reduce the noise in the sample data,but also retain the useful signal.Second,the mean square error is reduced by 56.1% when the trading volume is added to the prediction model.And the mean absolute percentage error is also decreased.There is a correlation between trading volume and the rate of return,which is helpful to the prediction of the rate of return.Third,compared with wavelet de-noising and trading volume,feature extraction can improve the prediction accuracy more significantly.The data processing based on feature extraction provides more typical and effective information for the neural network model,which improves the prediction accuracy.