标题： 重新思考金融传染 显示英文标题

标题： Rethinking Financial Contagion

摘要： 如何以及在多大程度上，相互关联的金融体系会内生地放大外部冲击？ 本文试图调和2008年危机后出现的一些关于金融网络中传染性质和相关性的不同观点。 我们开发了一个涵盖多个网络传染模型的共同框架，并表明无论冲击分布和网络拓扑结构如何，模型之间的总体系统性损失水平都存在精确的排序关系。 我们认为，传染的程度关键取决于每个模型假设市场参与者可以获得的信息量。 在没有任何关于网络结构和外部资产价值的不确定性的情况下，著名的Eisenberg和Noe（2001）模型适用，该模型产生最低程度的传染。 这是由于损失守恒的性质：传染后的总体损失等于最初受到冲击的机构所遭受的损失。 这一特性意味着许多传染分析通过构造排除了任何损失放大，实际上将相互关联的系统视为一个整体实体，其中损失只是相互分摊的。 在更高的不确定性水平下，如债务排名模型所捕捉到的那样，损失变得非守恒，并在网络中累积。 这具有重要的政策含义：在困难时期通过减少不确定性水平（例如，获取网络的具体数据），政策制定者可以朝着更保守的情景移动。 实证上，我们在2006-2016年间欧洲最大银行的样本中比较了不同模型之间传染的幅度。 特别是，我们将传染效应作为冲击大小和被冲击的外部资产类型函数进行分析。 显示英文摘要

摘要： How, and to what extent, does an interconnected financial system endogenously amplify external shocks? This paper attempts to reconcile some apparently different views emerged after the 2008 crisis regarding the nature and the relevance of contagion in financial networks. We develop a common framework encompassing several network contagion models and show that, regardless of the shock distribution and the network topology, precise ordering relationships on the level of aggregate systemic losses hold among models. We argue that the extent of contagion crucially depends on the amount of information that each model assumes to be available to market players. Under no uncertainty about the network structure and values of external assets, the well-known Eisenberg and Noe (2001) model applies, which delivers the lowest level of contagion. This is due to a property of loss conservation: aggregate losses after contagion are equal to the losses incurred by those institutions initially hit by a shock. This property implies that many contagion analyses rule out by construction any loss amplification, treating de facto an interconnected system as a single aggregate entity, where losses are simply mutualised. Under higher levels of uncertainty, as captured for instance by the DebtRank model, losses become non-conservative and get compounded through the network. This has important policy implications: by reducing the levels of uncertainty in times of distress (e.g. by obtaining specific data on the network) policymakers would be able to move towards more conservative scenarios. Empirically, we compare the magnitude of contagion across models on a sample of the largest European banks during the years 2006-2016. In particular, we analyse contagion effects as a function of the size of the shock and the type of external assets shocked.