Stock Return Direction Forecasting Based on Good and Bad Information Spillover Networks

This paper innovatively proposes stock return direction forecasting method by using good and bad information spillover networks. First of all, we construct static and dynamic multilayer networks based on CHOLO log return decomposition and Elastic-Net-VAR models, and analyze differences and similarities between bad and good information spillover networks. It is found that the bad and good information spillover networks exhibit various topology properties and evolution behaviors. Specifically, the direction and magnitude of spillover vary greatly in bad and good information layers, and the spillover effects is stronger on the bad information layer. Moreover, the topological features of bad and good information layers change over time. Besides, a novel forecasting model (RF) based on good and bad information spillover networks is proposed for return direction forecasting of SZ50 index, and compare the RF method with other models, empirical results show that our proposed RF prediction model outperform other models for stock return direction forecasting, and forecasting power of RF model is not only statistically significant but also economically valuable. Finally, a range of methods are applied to enhance the interpretability of RF model, and we found that the core network characteristics (e.g., ACS_Good and ACS_Bad) that influence stock direction forecasting