基于非参数因果网络的风险溢出分析及多因子预测

Risk spillover analysis and multi factor forecasting based on nonparametric causal network

风险溢出是系统性金融风险产生和演化的核心动因,因此,探究风险的溢出路径与强度至关重要.为了弥补传统方法在维度与参数上的局限,本文将非参数条件互信息检验与因果网络结构学习算法相结合,提出了一种新的非参数高维因果网络构建方法来分析金融系统波动溢出的动态联动性和风险传递机制.此文在非线性合成数据集中验证了该方法的有效性和稳健性并根据因果拓扑关系构造最优预测子集对序列进行多因子预测.并将该模型应用于构建2013年1月至2019年12月期间全球81家能源公司日度股票收益的波动溢出网络,测量基本面与投资者两种维度的风险溢出强度动态变化并进行预测分析.此外,结合企业财务数据和宏观经济变量,考虑企业之间的业务异质性,探索风险溢出的决定因素.研究结果表明,1)能源产业链的上游以及高油价风险敞口的能源企业表现出较大的风险外溢效应和风险承受程度;2)除公司规模以外,企业资产收益率,边际收益等因素也影响溢出效应的强弱;3)能源公司的风险溢出在业务上存在很大差异,溢出驱动因素也有所不同,这对于在投资组合决策和监管政策设计等具有重要的参考价值;4)虽然因果预选信息选择策略结合非参数模型的短期预测效果要优于结合参数模型,但是随着预测步长的增加,参数模型的优势却更明显.

Risk spillover is the core motive of systemic financial risk generation and evolution.Therefore,it is crucial to investigate the path and intensity of risk spillover.To compensate for the limitations of traditional methods in terms of dimensionality and parameters,this paper combines a nonparametric conditional mutual information test with a causal network structure learning algorithm and proposes a new nonparametric high-dimensional causal network construction method to analyze the dynamic linkages and risk transmission mechanisms of volatility spillovers in the financial system.This paper validates the effectiveness and robustness of the method on a nonlinear synthetic dataset and constructs the optimal prediction subset for multi-factor forecasting of the series based on the causal topology.The model is also applied to construct a volatility spillover network of daily stock returns of 81 global energy companies over the period from January 2013 to December 2019 to measure the dynamic changes in the intensity of risk spillover in both fundamental and investor dimensions and to perform forecasting analysis.In addition,the determinants of risk premia are explored by combining corporate financial data and macroeconomic variables and considering business heterogeneity among firms.The findings show that 1)energy firms in the upstream of the energy chain and those with high oil price exposures exhibit larger risk spillover effects and risk tolerance levels.2)In addition to firm size,the return on corporate assets,marginal returns and other factors also have impacts on the spillover effect.3)The risk spillovers of energy companies differ significantly in business and the spillover drivers also differ,which is of important reference value in porfolio decision making,regulatory policy design,etc.4)Although the causal pre-selection information selection strategy combined with a non-parametric model has better short-term forecasting performance than the one combined with a parametric model,the advantage of the parametric model is more obvious as the forecasting step increases.