Can we better predict financial crisis? The role of Laplacian-energy-like measure

How to accurately and effectively predict financial crisis is one of the crucial and challenging issues in the field of financial risk management. A great number of financial crisis prediction paradigms are strictly limited to macroeconomic indicators, ignoring potential network effects, and the drawbacks are actually obvious. This paper aims to propose early warning models with a novel network topological indicator, the Laplacian-energy-like measure (LEL), to deal with this issue. We construct a series of monthly stock networks of Chinese A-shares and extract the LEL. Then, with the help of the prediction methods of machine learning and ensemble learning, we establish new early warning models involving LEL. Furthermore, we formulate and implement a comprehensive validation strategy to evaluate the predictive performance of our proposed models. Compared with the traditional models with alternative indicators such as network density and investor sentiment index, our proposed models demonstrate the outstanding characteristics of higher scores on the area under the curve, F1-score, accuracy and recall. In addition, by utilizing the Shapley value approach to assess the importance of predictors across diverse early warning models, LEL consistently ranks top two among predictive factors. Finally, the robustness of our proposed models with LEL are further confirmed in the US stock market.