An Ensemble De-noising Method for High Frequency Financial Data

High-frequency financial data are characterized by unbalanced, non-linear and low signal-noise ratio, which often represents a challenge on the study of financial market microstructure. There has been little research on the de-noising method for high-frequency financial data, with the wavelet analysis as the current major method. Considering that the effect of wavelet analysis is restricted by the signal-noise ratio, we introduced phase space reconstruction and independent component analysis method for analyzing high-frequency financial data. The qualitative and quantitative analyses have shown that high-frequency financial data is chaotic in the time series and suitable to use the phase space reconstruction method. Furthermore, we propose the ensemble de-noising method for the high-frequency financial data. The numerical experiments results show that the de-noising effectiveness of our proposed methods is better than that of wavelet analysis. The improvement is about 2 times and more from the view of prediction precision based on the support vector machine. Our proposed ensemble de-noising method may also become a basis for general studies of financial market microstructure.