Reconstructing and Stress Testing Credit Networks

Ramadiah, Amanah; Caccioli, Fabio; Fricke, Daniel

doi:10.2139/ssrn.3084543

Cited by 10 publications

(17 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Generally speaking, what the authors observe is that the best performance depends on the specific indicator and the level of aggregation. However, apart from the trivial result that the MaxEnt and the Minimum-Density methods achieve, respectively, the highest sensitivity and the highest specificity, the authors find that the considered variants of ERG models (i.e., the methods inspired 63 by [179] and [104]) "consistently perform best" [178] and "are able to reconstruct adjacency matrices and weighted networks relatively well, and they are capable to preserve the statistical properties of the actual network at all (data) aggregation levels" [178]. Interestingly, the performance of reconstruction methods in reproducing dynamical (financial) indicators is also tested.…”

Section: Comparing Different Reconstruction Algorithms On Real-world mentioning

confidence: 91%

“…As the authors notice, even null models preserving degrees fail to accurately reproduce the actual level of systemic risk (defined as the probability of default of a bank [178]). However, the model inspired by [179] (followed by the model inspired by [104] and MaxEnt) "has the closest behavior to the actual network overall, while Minimum-Density shows an inconsistent performance across different aggregation levels" [178]. Generally speaking the fitness-induced ERG model performs well in replicating the binary topology because it provides a realistic estimate of the degrees in the network.…”

Section: Comparing Different Reconstruction Algorithms On Real-world mentioning

confidence: 99%

See 1 more Smart Citation

Reconstruction methods for networks: The case of economic and financial systems

et al. 2018

View full text Add to dashboard Cite

The study of social, economic and biological systems is often (when not always) limited by the partial information about the structure of the underlying networks. An example of paramount importance is provided by financial systems: information on the interconnections between financial institutions is privacy-protected, dramatically reducing the possibility of correctly estimating crucial systemic properties such as the resilience to the propagation of shocks. The need to compensate for the scarcity of data, while optimally employing the available information, has led to the birth of a research field known as network reconstruction. Since the latter has benefited from the contribution of researchers working in disciplines as different as mathematics, physics and economics, the results achieved so far are still scattered across heterogeneous publications. Most importantly, a systematic comparison of the network reconstruction methods proposed up to now is currently missing. This review aims at providing a unifying framework to present all these studies, mainly focusing on their application to economic and financial networks. 7The problem of missing information. After an initial activity aimed at determining the structure of real-world networks by measuring standard topological quantities, a more theoretical activity was started, aiming at both defining new quantities and devising proper models to explain observations [45,44,52,53,54]. Given the complexity that can arise even from a simple mathematical model based upon graphs, researchers have recently focused on the development of a topological theory: loosely speaking, topological quantities are employed to define statistical models, rather than reproduced from microscopic dynamical rules [55,56,57,58,59].Unfortunately, when moving to the validation of such models a common problem arises: very often, the data available on the real network are either incomplete or imprecise (or both). This problem is particularly evident in the case of economic and financial networks: in this case, data collection suffers from the problem of partial accounting and the presence of disclosure requirements. In order to illustrate the importance of such an issue, let us think of a bipartite, financial network whose node sets represent investors and the investments they do. Although the knowledge of the whole network structure could help regulators to take immediate countermeasures to stop the propagation of financial distress, this information is seldom available (the knowledge of the whole network of investments would pose immense problems of privacy), thus hindering the possibility of providing a realistic estimate of the extent of the contagion. As confirmed by the analysis of the various papers reported in this review, the incompleteness of network instances seems to be unavoidable [60,61]: since addressing the problem of estimating the resilience of financial networks cannot be addressed without knowing the structural details of national and cross-countries interbank networks, inf...

show abstract

Section: Comparing Different Reconstruction Algorithms On Real-world mentioning

confidence: 91%

Section: Comparing Different Reconstruction Algorithms On Real-world mentioning

confidence: 99%

Reconstruction methods for networks: The case of economic and financial systems

et al. 2018

View full text Add to dashboard Cite

show abstract

“…2 A complementary method is proposed by Anand et al (2015). Here the authors reconstruct a comparison of network reconstructions techniques has been carried out also for bipartite networks (Ramadiah et al, 2017).…”

Section: Introductionmentioning

confidence: 99%

Assessing systemic risk due to fire sales spillover through maximum entropy network reconstruction

Gangi

Lillo

Pirino

2018

Journal of Economic Dynamics and Control

View full text Add to dashboard Cite

Monitoring and assessing systemic risk in financial markets is of great importance but it often requires data that are unavailable or available at a very low frequency. For this reason, systemic risk assessment with partial information is potentially very useful for regulators and other stakeholders. In this paper we consider systemic risk due to fire sales spillovers and portfolio rebalancing by using the risk metrics defined by Greenwood et al. (2015). By using a method based on the constrained minimization of the Cross Entropy, we show that it is possible to assess aggregated and single bank's systemicness and vulnerability, using only the information on the size of each bank and the capitalization of each investment asset. We also compare our approach with an alternative widespread application of the Maximum Entropy principle allowing to derive graph probability distributions and generating scenarios and we use it to propose a statistical test for a change in banks' vulnerability to systemic events. JEL codes: C45;C80;G01;G33.

show abstract

“…When P (A) factorizes f ij corresponds to the quantity denoted with p ij in the literature [20], that is the linkage probability the is independent from link to link. Differentiating (12) with respect to β out i and β in i , ∀i, yields the system of 2N coupled equations…”

mentioning

confidence: 99%

“…for model A we determine the vectors β out and β in by solving the system of equations (13) and for model B, we compute the quantities (18). For both models A and B, we used as prior binary distribution the linkage probabilities from the dcGM model (See (25) in the Appendix for details), since methods using such probability form have won several independent horse races [10][11][12].…”

mentioning

confidence: 99%

A faster horse on a safer trail: generalized inference for the efficient reconstruction of weighted networks

2020

View full text Add to dashboard Cite

The knowledge of the structure of a financial network gives valuable information for the estimation of systemic risk and other important properties. However, since financial data are typically subject to confidentiality, network reconstruction techniques become necessary to infer both the presence of connections (i.e. the topology) and their intensity (i.e. the link weights) from partial information. Recently, various horse races have been conducted to compare the performance of these methods. These comparisons were based on arbitrarily chosen metrics of network similarity. Here we establish a generalised likelihood approach to the rigorous definition, estimation and comparison of methods of reconstruction of weighted networks. The crucial ingredient is the possibility of separately constraining the topology and the link weights to follow certain "tight" empirical relationships that connect them to the observed marginals. We find that a statistically rigorous estimation of the probability of each link weight should incorporate the whole distribution of reconstructed network topologies, and we provide a general procedure to do this. Our results indicate that the best method is obtained by "dressing" the best-performing method available to date with a certain exponential distribution of link weights. The method is fast, unbiased and accurate and reproduces the empirical networks with highest generalised likelihood.Network reconstruction is an active field of research within the broader field of complex networks [20]. Addressing the network reconstruction problem means facing the double challenge represented by the estimation of topology and link weights. The task at hand consists in determining both binary and weighted ensemble distributions, and to understand the interplay between them. Among the methods proposed so far, some assume that the binary and weighted constraints jointly determine the final configuration in terms of both topology and weights while others attribute weights to the binary configuration using a completely separate methodology [17,21]. Amidst the former ones, a special mention is deserved by the Enhanced Configuration Model [15] . This is defined by simultaneously constraining the degrees and the strengths of nodes which jointly affect the estimation of the two sets of quantities, the linkage probabilities and the weight estimates. Since these are jointly determined on the basis of the same information (i.e. constraints), this implies the impossibility to include purely topological additional information. Examples of algorithms belonging to the second group are those iteratively adjusting the link weights (e.g. via the RAS recipe [19]) on top of some previously-determined topological structure, in such a way to satisfy the constraints concerning strengths a posteriori. This approach has encountered critiques in [16]. It is important to notice that this kind of procedure assigns weights deterministically, and therefore the likelihood of observing any real matrix is exactly zero, assuming conti...

show abstract

Reconstructing and Stress Testing Credit Networks

Cited by 10 publications

References 49 publications

Reconstruction methods for networks: The case of economic and financial systems

Reconstruction methods for networks: The case of economic and financial systems

Assessing systemic risk due to fire sales spillover through maximum entropy network reconstruction

A faster horse on a safer trail: generalized inference for the efficient reconstruction of weighted networks

Contact Info

Product

Resources

About