Non-stationary temporal-spatio correlation analysis of information-driven complex financial dynamics

Abstract

The explosion of information production provides a new perspective on investigating complex dynamic systems such as the financial markets. With large-scale historical data, a textual-based sentiment index is introduced to study the correlations between the external information and the price return in the stock market. In particular, a novel approach taking into account the non-stationary effect of the sentiment is proposed to compute the sentiment-return correlation function, and it reveals a non-zero correlation between the past sentiment and the future motion of the price return. Such a computation is then extended to a cross-correlation form which describes the correlations between different sentiment indexes and price returns. A stratified structure of the cross-correlation functions is observed. With the random matrix theory, the features of the stratified structure are quantitatively analyzed. Finally, an investment strategy is constructed based on the temporal correlation.