A hybrid information capturing methodology for price volatility and its application to financial markets

Abstract

This paper is aimed to solve uncertainty in financial data mining by using fuzzy support vector machine (FSVM), and prior wavelet denoising and subsequent error adjustment based on generalized autoregressive conditional heteroskedasticity (GARCH) model are also supplemented respectively. The hybrid information capturing methodology proposed above can thus be supposed to address complex nonlinear dynamics behind the noise of financial data. To this end, a key approach to ascertain membership values for financial sample data is introduced in order to build the FSVM in terms of statistical characteristics of the financial data from the prior wavelet denoising stage. Moreover, the GARCH model is also employed in the final step so that the test errors from the preliminary test based on FSVM are deeply analyzed to capture missed price volatility information which are often neglected by existing approaches involving the traditional SVM models. The methodology proposed is thus enabled to sufficiently tackle the styled facts, such as nonlinearity, instability, strong noise, skewed distribution, and so on, because two factors of influencing price volatility, that is, the market factor and the time series factor, are all accounted for, and its prediction outperformance appears in empirical analysis of S&P 500 index.