We propose a tool to predict risks to economic growth and international business cycles spillovers: the gross domestic product (GDP)-Network conditional value at risk (CoVaR). Our methodology to assess Growth-at-Risk is composed of two building blocks. In the first step, we apply a machine learning methodology, namely the network-based NETS by Barigozzi and Brownlees, to identify significant linkages between pair of countries. In the second step, applying the CoVaR methodology by Adrian and Brunnermeier, and exploiting international statistics on trade flows and GDPs, we derive the entire distribution of Economic Growth spillover exposures at the bilateral, country and global level for different quantiles of tail events on economic growth. We find that Economic Growth Spillover probability distribution is time-varying, left-skewed and in some cases bi- or even multi-modal. Second, we prove that our two-step approach outperforms alternative one-step quantile regression models in predicting risks to economic growth. Finally, we show that Global exposure to economic growth tail events is decreasing over time.

中文翻译：

---

GDP 网络 CoVaR：评估风险增长的工具

我们提出了一种预测经济增长和国际商业周期溢出风险的工具：国内生产总值（GDP）-风险网络条件价值（CoVaR）。我们评估风险增长的方法由两个组成部分组成。在第一步中，我们应用机器学习方法，即 Barigozzi 和 Brownlees 的基于网络的 NETS，以确定两个国家之间的重要联系。在第二步中，应用 Adrian 和 Brunnermeier 的 CoVaR 方法，并利用贸易流量和 GDP 的国际统计数据，我们推导出双边、国家和全球层面经济增长溢出风险敞口的整个分布，用于经济尾部事件的不同分位数。生长。我们发现经济增长溢出概率分布是随时间变化的，左偏，在某些情况下是双模甚至多模的。其次，我们证明我们的两步法在预测经济增长风险方面优于替代的一步分位数回归模型。最后，我们表明，随着时间的推移，全球经济增长尾部事件的风险正在下降。