Search

The interaction and uncertainty of stock market is exactly critical for traders and investors. Stock price prediction is a hot topic of research due to the returns and risks that coexist in financial markets. But grasping the complex fluctuations of stock price is a highly challenging task that attracts a lot of attention from researchers. To introduce a reliable forecasting model, a multiscale modelling strategy is proposed based on the machine learning method and the econometric model. On the basis of recognizing different frequency components of time series, the optimized support vector machine is used to realize the nonlinear features of stock prices. Owing to the advantages of statistical models for low frequency sub-series, it is more appropriate to capture the linear features. In this study, three stock closing price series of different industrial companies in China were used as the sample data. The multiscale strategy plays a significantly positive role in enhancing forecasting performance through horizontal comparison analysis. The statistical significance of the proposed model was examined through the Diebold Mariano test. In addition, the sensitivity of forecasting results to optimization methods was also discussed in-depth. The comparisons and discussion at multiple levels indicate the accuracy and practicality of the multiscale forecasting model. • A new model is proposed for stock price time series with multiscale modelling strategy. • The fluctuation feature in stock price data is identified and measured. • Different fitting and generalization ability of models to construct the forecast engine. • That empirical results gives the performance of the proposed model statistically verified. [ABSTRACT FROM AUTHOR]