Research On LSTM Hybrid Model Based On Wavelet Transform And Stock Price Forecasting

The trend of the stock price index is not only related to the interests of individual investors and listed companies,but also closely related to the regulation of national economic policies.Accurately predicting the stock price index has become an increasingly important issue in investment decision-making.Establishing traditional time series models.To a certain extent,the prediction results of these models can roughly reflect the general trend,but the accuracy of the prediction results needs to be improved.Therefore,in recent years,many experts and scholars have explored the mixed model to predict the stock price index and achieved Good results.Our country's stock market is a policy market.It is not only limited by the rise and fall,but also affected by relevant national economic policies.It is difficult to predict.Many models applicable to foreign markets may not be able to accurately predict the Chinese market.At the same time,the high stock price Uncertainty of volatility,complex data components and a series of noise interference have brought difficulties to accurately predicting stock price indices.For the problem of complex data components,wavelet transform,ARIMA model and nonlinear data prediction models(including LSTM,SVR)mixed model solution.The original data is decomposed by wavelet transform using the db3 wavelet function as the wavelet transform basis function,and then reconstructed by the two interpolation method,and a set of low-frequency trend series that eliminates the influence of short-term fluctuations and retains the long-term trend is obtained,and three sets of stock price indices The mean line and the fluctuation of the mean line show short-period non-linear characteristics of high-frequency fluctuation sequences,which give play to the characteristics of the wavelet transform "mathematical microscope";establish an ARIMA model to predict low-frequency trend sequences,and establish LSTM and SVR models to predict highfrequency fluctuation sequences;and finally Choose the model with the best prediction effect for each frequency to construct a hybrid prediction model.In the empirical part,the Shanghai Stock Exchange Index,Shenzhen Stock Exchange Index,and Mingchen Health Closing Price are used as training sets to decompose the lowfrequency sequence and the high-frequency sequence.Establish the ARIMA model to predict the low-frequency sequence;establish the LSTM and SVR models to predict the highfrequency sequence;finally,compare the results of the hybrid model M-ARIMA-LSTMSVR with other models.The M-ARIMA-LSTM-SVR model is stable.