Network Tail Risk Estimation in the European Banking System

Torri, Gabriele; Giacometti, Rosella; Tichý, Tomáš

doi:10.2139/ssrn.3724390

Cited by 2 publications

(7 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, as shown in [40] in the context of using multivariate GARCH models to estimate the CoVaR with this definition of distress one has to numerically solve an integral which could be complex to implement well and quantile estimation to obtain the CoVaR will not be feasible [63]. For these reasons and because a large part of the recent literature keeps using the traditional definition [9,15,55,85] we continue to use the traditional definition of the CoVaR by Brunnermeier and Adrian. However, because for the CoES it is simpler to obtain estimates based on more expansive definition of a distress event and using the traditional definition there could lead to deceptive results we use the expanded definition there.…”

Section: Unconditionalmentioning

confidence: 99%

“…When it comes to empirical applications the ∆-CoVaR has been used in many empirical studies related to systemic risk. Recent examples which are far from exhaustive are [9,15,55,85] with [55,85] examining financial networks using the network ∆-CoVaR.…”

Section: Unconditionalmentioning

confidence: 99%

“…Hence, only one of the variables is taken to their quantile. The case of stimating the network-CoVaR between institutions i, j while controlling on a set of other institutions excluding institution i and j is addressed in [85]. The solution is to condition on the losses of all other banks besides i and j being equal to their median 5 loss while conditioning on the loss of i being equal to its VaR at level q.…”

Section: Multivariate Versionmentioning

confidence: 99%

“…Since in general it does not hold that ∆-CoVaR(i,j)=∆-CoVaR(j,i) with i = j(same holds for the ES based version) and the dataset contains 1523 instutions this would involve estimating n(n − 1) = 2.318.006 ∆-CoVaRs and ∆-CoESs for each confidence level. In the literature on financial networks methods like LASSO (Least Absolute Shrinkage and Selection Operator), Elastic Net and SCAD (Smoothly Clipped Absolute Deviation) are used to reduce this complexity [47,55,85]. However, our aim is to examine the properties of the network ∆-CoES/CoVaR irrespective of the method used to obtain the important links.…”

Section: Unconditional Network Model On Gsibsmentioning

confidence: 99%

“…We also argue that the increased heterogeneity of the ∆-CoES matrix serves a practical purpose. To this end the average row and column values of both matrices like in [85] are computed. As in [85] the average row value denotes the average network importance of each institution while the average column value denotes the average resiliency of each institution.…”

Section: Unconditional Network Model On Gsibsmentioning

confidence: 99%

See 4 more Smart Citations

$Δ-$CoES

Aleksy¹

2022

Preprint

View full text Add to dashboard Cite

In this paper the ∆-CoVaR method is extended in both the conditional and unconditional cases to be based on the Expected Shortfall (ES) using quantile regression with a more expansive distress definition. We find the resulting ∆-CoES measure to be complementary to the ∆-CoVaR and to be more effective than the ∆-CoVaR in measuring short-term changes in systemic risk and in identifying heterogeneity in the systemic risk contributions of financial institutions and linkages between institutions due to its lower robustness. For regulators, risk managers and market participants these properties are interesting from an economic standpoint when they require the increased sensitivity and heterogeneity of the ∆-CoES to set short-term capital requirements/risk limits, find problematic financial linkages, problematic financial institutions or have some kind of early warning system for the emergence of systemic risk. Lastly, the ∆-CoES is straightforward to estimate and would fit within recent regulatory frameworks such as the FRTB. To show the statistical advantages and properties empirically, the ∆-CoVaR and ∆-CoES methods are used on a large sample (from 31-12-1970 to 31-12-2020 1564 firms) of daily equity data from US financial institutions both in a system and network fashion. On a sample of 9 US-based GSIBS we also show the properties and the utility of the ∆-CoES when it comes to identifying problematic financial links.

show abstract