Abstract
Phase space reconstruction of financial time series has become an effective approach for financial system analysis. Since the phase space obtained by the traditional method had the large embedding dimension and was susceptible to noise, this paper proposed a novel phase space reconstruction approach based on manifold learning, called manifold learning-based phase space reconstruction (MLPSR), which was applied in financial time series. In our approach, the traditional correlation dimension estimation was employed to reconstruct a rough phase space of financial time series; then manifold learning method was adopted to discover the embedding natural structure of the rough phase space, and thus the intrinsic embedding dimension was obtained. In the empirical study, our method was compared with the the traditional correlation dimension method, and showed the better perormance. Our MLPSR would reduce the phase space reconstruction error of financial time series, and provide more accurate data support for the subsequent studies about financial system.
Recommended Citation
Huang, Yan and Kou, Gang, "MANIFOLD LEARNING-BASED PHASE SPACE RECONSTRUCTION FOR FINANCIAL TIME SERIES" (2014). PACIS 2014 Proceedings. 364.
https://aisel.aisnet.org/pacis2014/364