Abstract
Based on data from the European banking stress tests of 2014, 2016 and the transparency exercise of 2018 we construct networks of European banks and demonstrate that the latent geometry of these financial networks can be wellrepresented by geometry of negative curvature, i.e., by hyperbolic geometry. Using two different hyperbolic embedding methods, hydra+ and Mercator, this allows us to connect the network structure to the popularityvssimilarity model of Papdopoulos et al., which is based on the Poincaré disc model of hyperbolic geometry. We show that the latent dimensions of ‘popularity’ and ‘similarity’ in this model are strongly associated to systemic importance and to geographic subdivisions of the banking system, independent of the embedding method that is used. In a longitudinal analysis over the time span from 2014 to 2018 we find that the systemic importance of individual banks has remained rather stable, while the peripheral community structure exhibits more (but still moderate) variability. Based on our analysis we argue that embeddings into hyperbolic geometry can be used to monitor structural change in financial networks and are able to distinguish between changes in systemic relevance and other (peripheral) structural changes.
Introduction
Network models based on hyperbolic geometry have been successful in explaining the structural features of informational^{1}, social^{2} and biological networks^{3}. Such models provide a mathematical framework to resolve the conflicting paradigms of preferential attachment (attraction to popular nodes) and community effects (attraction to similar nodes) in networks^{4,5,6}.
Just as the geometric structure of a social network determines the diffusion of news, rumors or infective diseases between individuals^{7}, the geometric structure of a financial network influences the diffusion of financial stress between financial institutions, such as banks^{8,9,10,11}. Indeed, the lack of understanding for risks originating from the systemic interaction of financial institutions has been identified as a major contributing factor to the global financial crisis of 2008^{12}. While many recent studies have analysed the mechanisms of financial contagion in theoretical or simulationbased settings, less attention has been payed to the structural characteristics and the geometric representation of real financial networks. Although evidence of hyperbolic structure has been uncovered for international trade networks^{13}, no such analysis has been carried out for networks of financial institutions. Identifying a suitable geometric representation for such networks can help to monitor and quantify structural change, and—in the case of hyperbolic geometry—even distinguish structural change in terms of systemic importance from changes in the network’s peripheral structure. Moreover, a geometric representation can form the basis of analytic models of contagion processes and their optimal control in future research.
Here, we consider financial networks inferred from bank balance sheet data, as collected and made available by the European Banking Authority (EBA) within the European banking stress test and transparency exercises of 2014, 2016 and 2018^{14,15}. We show that these networks can be embedded into lowdimensional hyperbolic space with considerably smaller distortion than into Euclidean space of the same dimension, suggesting that the paradigm of latent hyperbolic geometry also applies to financial networks. In addition, we demonstrate that the hyperbolic geometric representation compares favorably to a degreecorrected stochastic block model^{16}—a popular nongeometric network model. Furthermore—following Papadopoulos et al.^{4}—we decompose the embedded hyperbolic coordinates into the dimensions of popularity and similarity and demonstrate that these dimensions align with systemic importance and membership in regional banking clusters respectively. Finally, the longitudinal structure of the data allows us to track changes in these dimensions over time, i.e., to track the stability of systemic importance and of the peripheral community structure over time.
Results
Inference of financial networks
Contagion in financial networks is a complex process, which can take place through several parallel (and potentially interacting) mechanisms and channels^{17}. These mechanisms include direct banktobank liabilities^{18}, bank runs^{19}, and marketmediated contagion through asset sales^{17,20,21,22} (‘firesale contagion’); see also (French et al.^{12}, p. 21ff). Here, we focus on the channel of firesale contagion, which has been singled out—both in simulation^{21} and in empirical studies^{20}—as a key mechanism of financial contagion. Moreover, the propensity of firesale contagion can be quantified from available balance sheet data, using liquidityweighted portfolio overlap (LWPO)^{22,23} as an indicator (see “Methods” for details).
Our inference of financial networks follows a twostage mechanism: First, we construct a weighted bipartite network in which banks \(B = (b_1, \ldots , b_n)\) are linked to a common pool of assets \(A = (a_1, \ldots , a_m)\), which consist of sovereign bonds classified by issuing country and by different levels of maturity. In the second step we perform a onemode projection of this network on the node set B, using the LWPO of two banks \(b_i, b_j \in B\) to determine the weight \(w_{ij}\) of the link between the corresponding nodes. For any of the years \(y \in \{2014, 2016, 2018\}\), the result is an undirected, weighted network \(N_y\) of banks, in which two banks are connected if and only if they hold common assets. The link weight \(w_{ij}\), normalized to [0, 1], represents the susceptibility of two banks \(b_i, b_j\) to financial contagion, quantified by their LWPO.
Network features
The inferred networks are very dense (densities: \(\rho _{2014} = 0.86\), \(\rho _{2016} = 0.96\), \(\rho _{2018} = 1.00\)), i.e., almost all pairs of banks hold some common assets. However, the distribution of weights is highly skewed (see Fig. 1A), with most of the connections exhibiting very small weights. In other words, the networks are dominated by a ‘sparse backbone’ of a few strong connections, which represent the dominant channels of potential contagion of financial distress. The same skew is present in the distribution of node strengths (see Fig. 1B) with a few strong nodes dominating over a majority of weaker nodes in all years.
To extract more salient connectivity information from these highly connected networks, we also consider the ‘\(p\%\)Backbone’ of each network, formed by the upper \(p\%\)quantile of highestweighted edges (We have also considered disparity filtering^{24} as an alternative method for backbone extraction; see ‘Comparison of methods and robustness checks’ below.). Figure 2 shows the degree distribution and the local clustering coefficient (in dependence on node degree) for the 10% and the 25%Backbone. While there is no evidence of a scalefree degree distribution, the clustering coefficient displays an interesting pattern: It is highest for mediumdegree nodes and then decreases with increasing node degree. This indicates that highdegree nodes (i.e. highly connected banks) typically act as hubs between lowerdegree nodes (i.e. ‘normal’ banks) without a direct link.
Latent network geometry
Network representation methods
Our next objective was to uncover the latent geometric network structure and to evaluate the suitability of a hyperbolic network model. (See “Methods” for background on hyperbolic geometry.) To this end, we applied four different network representation methods (one method embedding into twodim. Euclidean space \({\mathbb{E}}_2\), two methods embedding into twodim. hyperbolic space \({\mathbb{H}}_2\), and one nongeometric method) to the financial networks \(N_{2014}, N_{2016}\) and \(N_{2018}\) and their \(p\%\)backbones for \(p = 10, 25, 50\). The first two methods, multidimensional scaling^{25} (MDS) and hydra+^{26,27}, calculate stressminimizing embeddings of the weighted network distances into Euclidean and hyperbolic geometry, respectively. The third method, Mercator^{28}, is a connectivitybased method (i.e. ignoring network weights) and uses a mix of machine learning and maximum likelihood estimation to infer latent coordinates in a popularityvssimilaritymodel of hyperbolic geometry; see GarcíaPérez et al.^{28} for details. As nongeometric representation method, we used a degreecorrected stochastic block model (dSBM)^{16}, as implemented in the Rpackage randnet^{29}, which aims to represent network structure by inferring communities and their connection probabilities; see Karrer and Newman^{16} for details. Since Mercator and dSBM are connectivitybased, they can only be meaningfully applied to the network backbones. MDS and hydra+, on the other hand, can also be applied to the full weighted networks and are directly comparable, since they minimize exactly the same objective function, but only differ in their target geometries. Figure 3 shows a comparison of the embedding quality of the different methods. For the full networks, we use stress (i.e. the root mean square error between network distances and embedded distances) as an evaluation metric, while for the backbones we use the AUPR (area under the PrecisionRecallcurve) from a network reconstruction task (see “Methods” for details).
Comparison of methods and robustness checks
Our comparison shows that Mercator, based on hyperbolic geometry, outperforms all other methods in terms of network reconstruction performance, consistently over all three years of observation and independent of the threshold used for the extraction of the network backbone. The second hyperbolic embedding method, hydra+, yields results that are better or at least comparable to dSBM, while the Euclidean embedding method MDS performs worst. Also note that hydra+ and MDS are the only methods which can be directly applied to the full weighted network, in which case hydra+ clearly outperforms MDS in terms of embedding error.
To check the robustness of our results with respect to the method of backbone extraction, we have repeated the same analysis with backbones determined by disparity filtering^{24}. The results are reported in supplementary Figure S1. While reconstruction quality deteriorates for all methods on the disparity filtered backbones, the advantage of the hyperbolic methods over the nongeometric and Euclidean methods becomes even more pronounced. Overall, we conclude that the latent geometry of the observed financial networks is—at least in low dimension—much better represented by negatively curved (hyperbolic) rather than flat (Euclidean) geometry. Moreover, the hyperbolic representations are superior even to the (nongeometric) degreecorrected stochastic block model in terms of network reconstruction quality.
Latent hyperbolic coordinates
As a result of the hyperbolic embeddings we obtain for each bank node \(b_i\) latent coordinates \((r_i, \theta _i)\) in the Poincaré disc model of hyperbolic space (see “Methods”). This allows us to connect the network embedding to the popularityvssimilarity model of Papadopoulos et al.^{4} and the \({\mathbb{S}}^1\)model of Ángeles Serrano et al.^{28,30}. The hyperbolic embedding of the full banking network of 2018 produced by hydra+ is shown in Fig. 4A. The embedding of the 10%Backbone of the same network produced by Mercator is shown in Fig. 4D. Note that the hydra+embedding attempts to give a faithful representation of all distances in the weighted network, whereas Mercator only encodes connectivity information and is harder to interpret visually. This phenomenon is exacerbated by the laws of hyperbolic geometry, in which seemingly small differences in the radial coordinate can represent large differences in hyperbolic distance. With reference to Fig. 4A, the embedded network shows a clear coreperiphery structure, in line with previous studies of financial networks^{31,32} and in agreement with the pattern exhibited by the local clustering coefficient in Fig. 2.
Structural analysis
The popularityvssimilarity model of Papadopoulos et al.^{4} and the \({\mathbb{S}}^1\)model of Ángeles Serrano et al.^{28,30} used by Mercator offer a direct interpretation of the latent hyperbolic network coordinates in the Poincaré disc in terms of their popularity dimension (the radial coordinate r) and the similarity dimension (the angular coordinate \(\theta\)). In the context of financial networks, we hypothesized that the popularity dimension of a given bank aligns with its systemic importance, and that its similarity dimension is associated with subsectors of the banking system, e.g., along geographic and regional divisions. Also for the hydra+ embedding, a theoretical foundation for interpreting r as popularity dimension and \(\theta\) as similarity dimension has been given^{27}. However, due to the asymmetric distribution of banks within the Poincaré disc (Fig. 4A) for the hydra+ embedding, we calculate its geodesic polar coordinates \((r_i,\theta _i)\) with respect to the network centerofweight, rather than the center of the Poincaré disc; see “Methods” for details (The fact that both approaches—Mercator and recentered hydra+—lead to qualitatively very similar results can be seen as a validation of this methodology.). For the Mercator method we directly use the coordinates \((r_i,\theta _i)\) from the embedding of the 25%backbone and perform no additional centering.
To test the first hypothesis—the association between radial coordinate r and systemic importance—we labelled a bank as systemically important in a given year, whenever it was included in the contemporaneous list of global systemically important banks (GSIBs) as published by the Financial Stability Board^{33,34,35}; see also Table 1. Using a Wilcoxon–Mann–Whitney test, we find a significant association between radial rank and systemic importance in all years and for both methods (\(P_{2014} < .0001\), \(P_{2016} < .0001\) for both methods, \(P_{2018} = .0038\) for hydra and \(P_{2018} = 0.0001\) for Mercator). In Table 2 we report the five topranked banks (most central in terms of r) for each year.
To test the second hypothesis—the association between similarity dimension \(\theta\) and regional banking subsectors—we assigned banks to the following nine regional groups:
Spain (ES), Germany (DE), France (FR), Italy (IT), UK and Ireland (UK/IE), Nordic Region (EE/NO/SE/DK/FI/IS), Benelux Region (BE/NE/LU), Southern/Mediterranean (GR/CY/MT/PT), Central/Eastern Eur. (AT/BG/HU/LV/RO/SI).
These regions are reasonably balanced in terms of the number of banks included in the EBA panel. Using ANOVA for circular data (see “Methods”) we find a highly significant association between the angular coordinate \(\theta\) and the regional group in all three years considered (\(P < .0001\) in all years for both methods). This indicates that the peripheral community structure (away from the network core) of the EBA financial network is indeed strongly aligned with geographic and regional divisions in Europe. We have highlighted two different regional groups in Fig. 4B,C to illustrate the association between angular coordinate and regional structure.
Network structure over time
The longitudinal structure of the data set allows us to track changes in the network structure over the whole time span of observations from 2014 to 2018. Note, however, that the samples of banks included by the EBA vary substantially in size and—even when restricted to the smallest sample—are not completely overlapping; see Table 3. Nevertheless, the embedding quality of the hyperbolic methods (reported in Fig. 3) is surprisingly stable over all years. This suggests that the hyperbolic model does indeed capture intrinsic qualities of the network, rather than relying on transitory structural artefacts.
We proceed to analyze the temporal changes in the latent radial coordinate r and angular coordinate \(\theta\), corresponding to changes in systemic importance and community structure. Note that the small sample of banks included in the 2016 stress test restricts the number of banks that are included in this longitudinal analysis, cf. Table 3. The scatter plots in Fig. 5 and the corresponding Pearson’s correlations of .678 (hydra+, \(P < .0001\)) and .892 (Mercator, \(P < .0001\)) between 2014 and 2016, and .569 (hydra+, \(P = .0001\)) and .502 (Mercator, \(P =.0015\)) between 2016 and 2018 show a significant positive association between hyperbolic centrality in successive snap shots of the financial networks.
In panel Ia of Fig. 5, Nordea bank can be identified as a clear outlier, moving from a very central position in 2014 to a peripheral position in 2016. Interestingly, Nordea was one of just two banks (together with Royal Bank of Scotland) which were removed from the list of GSIBs in the subsequent update in 2018 due to decreasing systemic importance^{35}. In the Mercator embedding (panel IIa) Nordea bank does not appear as an outlier, which is likely due to the fact that some structural information is lost when the full network is reduced to its backbone.
For the angular coordinate, we account for the circular nature of the variable and compute the circular correlation^{36} of the angular coordinates between successive years. Only moderate associations between successive years can be observed at absolute circular correlation values of 0.211 (hydra+, \(P = .1877\)) and 0.01 (Mercator, \(P = .9442\)) between 2014 and 2016 and 0.225 (hydra+, \(P = 0.1385\)) and 0.383 (Mercator, \(P = .0196\)) between 2016 and 2018.
Discussion
Based on data from the EBA stress tests of 2014, 2016 and the transparency exercise of 2018, we have presented strong evidence that the latent geometry of financial networks can be wellrepresented by geometry of negative curvature, i.e., by hyperbolic geometry. Calculating embeddings into the Poicaré disc model of hyperbolic geometry has allowed us to visualize this geometric structure and to connect it to the popularityvssimilarity model of Papdopoulos et al.^{4} and the \({\mathbb{S}}^1\)model of Ángeles Serrano et al.^{28,30}. We find that the radial coordinate (‘popularity’) is strongly associated with systemic importance (as assessed by the Financial Stability Board) and the angular coordinate (‘similarity’) with geographic and regional subdivisions. A longitudinal analysis shows that—in the observation period from 2014 to 2018—systemic importance of banks within the European banking network has stayed rather stable and has been predominated by only gradual changes. The peripheral community structure has been more variable, but has remained strongly determined by geographical divisions in all years considered.
From a broader perspective, our results indicate that hyperbolic network representations could be important tools for regulators to monitor structural change in financial networks, as they are able to distinguish changes in the systemic importance (popularity) of financial institutions from ‘peripheral changes’ (similarity) which are less relevant from a regulator’s perspective. Furthermore, our research provides an empirical basis for using hyperbolic geometry as a model space for the modelling of contagion processes and their optimal control in financial (or other) networks. Instead of modelling such processes by simulation on individual networks, a geometric model space provides the opportunity of analytic models that provide deeper insights beyond a specific case.
Methods
Data preparation and inference of financial networks
The financial networks were extracted from three different publicly available data sets stemming from the stress tests (in 2014 and 2016) and the EUwide transparency exercise (in 2018) of the European Banking Authority (EBA)^{14,15}. The data sets contain detailed balance sheet information from all European banks (EU incl. UK + Norway) included in the stress test/transparency exercise of the EBA in the respective year. From these data sets we extracted the portfolio values of all sovereign bonds held by the banks, split by issuing country (38 countries) and three levels of maturity (short: 0M3M, medium: 3M2Y, long: 2Y10Y+), resulting in \(m = 38 \times 3 = 114\) different asset classes.
For each year, this data was stored as the weighted adjacency matrix P (‘portfolio matrix’) of a bipartite network. The n rows of P correspond to the banks in the sample, the m columns to the different asset classes, and the element \(P_{ik}\) to the portfolio value (in EUR) of asset k in the balance sheet of bank i. To perform a onemode projection of this bipartite network, we followed Cont and Wagalath^{23,37} as well as Cont and Schaanning^{22}: We computed the liquidityweighted portfolio overlap (LWPO) of bank i and bank j as
where \(d_k\) is the market depth for asset k^{22}. The LWPO measures the impact of a sudden liquidation of the portfolio of bank i on the portfolio value of bank j and vice versa. Hence, it quantifies the risk of firesale contagion between the banks in a financial stress scenario. The market depth of asset k was estimated from P as its total volume held by all banks in the sample, i.e., as \(d_k = \sum _{i=1}^{n} P_{ik}\). Writing D for the diagonal matrix of market depths, (1) can be succinctly written as matrix product \(L = P D^{1} P^\top\). Finally, we set the link weight \(w_{ij}\) between bank \(b_i\) and \(b_j\) in the onemode projection N of the banking network equal to the normalized LWPO between banks \(b_i\) and \(b_j\), i.e., \(w_{ij} := L_{ij} / \max _{i,j}L_{ij}\)
Background on hyperbolic geometry
The hyperboloid model
Hyperbolic geometry can be characterized as the geometry of a space of constant negative curvature, while the more familiar Euclidean geometry is the geometry of a flat space, i.e., a space of zero curvature. In the hyperboloid model of hyperbolic geometry^{38,39}, ddimensional hyperbolic space \({\mathbb{H}}_d\) is defined as the hyperboloid
In fact, \({\mathbb{H}}_d\) endowed with the Riemannian metric tensor \(ds^2 = dx_0^2  dx_1^2  \cdots  dx_d^2\) is a Riemannian manifold and \({\mathrm {d}}_H(x,y)\) is the corresponding Riemannian distance^{38,39}. The sectional curvature of this manifold is constant and equal to \(1\). Thus, \({\mathbb{H}}_d\) is indeed a model of geometry of constant negative curvature.
The Poincaré disc model
While the hyperboloid model is convenient for computations, a more preferable (and popular) model for visualizations in dimension \(d=2\) is the Poincaré disc model^{38}, which also forms the basis of the popularityvssimilarity model of Papadopoulos et al.^{4}. To obtain the Poincaré disc model, the hyperboloid \({\mathbb{H}}_2\) is mapped to the open unit disc (‘Poincaré disc’) \({\mathbb{D}} = \left\{ z \in {\mathbb{R}}^{2}:{z_{1}}^{2} + {z_{2}}^{2} < 1\right\}\), parameterized by hyperbolic polar coordinates as \(z_1 = \tanh (r/2) \cos \theta\), \(z_2 = \tanh (r/2) \sin \theta\), using the stereographic projection^{38}
where \({{\,\mathrm{atan'}\,}}\) is the quadrantpreserving arctangent (The quadrantpreserving arctangent \({{\,\mathrm{atan'}\,}}(x_2,x_1)\), welldefined unless \(x_1 = x_2 = 0\), returns the unique angle \(\theta \in [0,2\pi )\) which solves \(\tan \theta = x_2/x_1\) and points to the same quadrant as \((x_1,x_2)\). It is commonly implemented in scientific computing environments (e.g. in MATLAB or R) as atan2.). In the Poincaré disc model, the hyperbolic distance becomes
and geodesic lines are represented by arcs of (Euclidean) circles intersected with \({\mathbb{D}}\).
Stressminimizing embeddings and hyperbolic centering
Stressminimizing embeddings
Stressminimizing embedding methods aim to find—for each network node \(b_i\)—latent coordinates \(x^{i}\) in a geometric model space G, such that the geodesic distance between \(x^{i}\) and \(x^{j}\) in G matches—as closely as possible—a given dissimilarity measure \({\mathrm {d}}_{ij}\) (such as the weighted network distance) between nodes \(b_i\) and \(b_j\). This is achieved by minimizing the stress functional
which measures the root mean square error between given network distances and the corresponding distances in the model space. For Euclidean geometry, this method is wellknown as multidimensional scaling^{25,40}, or—using a weighted stress functional—as Sammon mapping^{41}. For hyperbolic space, i.e., when \({\mathrm {d}}_G^\text {geom} = {\mathrm {d}}_H\), several optimization methods for (3) have been proposed^{26,27,42}. We use the hydra+ method implemented in the package hydra for the statistical computing environment R^{43}.
Hyperbolic centering
For a point cloud \(x^1, \dots , x^n\) in \({\mathbb{H}}_d\) and nonnegative weights \(w_1, \ldots , w_n\) summing to one, the hyperbolic mean^{36} or hyperbolic center of weight^{44} can be determined as follows: Calculate the weighted Euclidean mean \({\bar{x}} = \sum w_i x^i\), and its ‘resultant length’ \(\rho = \sqrt{(\bar{x}_0)^2  ({\bar{x}}_1)^2  \cdots  ({\bar{x}}_d)^2}\), which is a measure of dispersion for the point cloud. The hyperbolic center c is then determined as \(c = {\bar{x}} / \rho\) and is again an element of \({\mathbb{H}}_d\). The point cloud can be centered at c by transforming each point as \({\tilde{x}}^i = T_{c} x^i\), where \(T_{c}\) is the hyperbolic translation matrix (‘Lorentz boost’)
In dimension \(d = 2\), the stereographic projection (2) can then be applied to convert the centered coordinates \({\tilde{x}}^i\) to centered polar coordinates \((r_i, \theta _i)\) in the Poincaré disc.
Application to financial networks
For the hydra+ embedding, the described methods were applied to the financial networks inferred from the EBA data as follows: We converted the similarity weights \(w_{ij}\) (normalized LWPO) to dissimilarities \({\mathrm {d}}_{ij} = 1  w_{ij}\). We embedded these dissimilarities by minimizing the stress functional (3), using the Rpackage hydra. For the resulting network embeddings, we calculated the capitalweighted network center c as the weighted hyperbolic mean with weights \(w_i\) proportional to the total capital \(\sum _{k=1}^m P_{ik}\) of bank i invested in all assets \(a_1, \dots a_m\). After centering at the hyperbolic center c, we calculated the coordinates \((r_i, \theta _i)\) by the stereographic projection (2).
For the Mercator embedding and the dSBM, we first extracted network backbones, both by simple thresholding and by disparity filtering^{24}. The resulting backbones were used as input for the methods provided at https://github.com/networkgeometry/mercator and the implementation of dSBM in the Rpackage randnet. Mercator outputs latent coordinates \((r_i, \theta _i)\) in the Poincaré disc, and the output of the dSBM method is a matrix of connection probabilities \({\hat{p}}_{ij}\) for each node pair.
For multidimensional scaling (MDS) the same methodology as for hydra+ was used, except that Euclidean distance (instead of hyperbolic distance) was used as \({\mathrm {d}}^\text {geom}_G\) in the objective function (3).
Analysis of embedding results
AUPR and network reconstruction
To evaluate the embedding results of the network backbones, we used the following network reconstruction task: To each pair of nodes \((b_i, b_j)\), assign the score \(d^\text {geom}_G(x_i,x_j)^{1}\), where \(d^\text {geom}_G\) is the geodesic distance in the geometric model space G, or—in case of the degreecorrected stochastic block model—the score \({\hat{p}}_{ij}\), where \({\hat{p}}_{ij}\) is the estimated connection probability between nodes i and j. Based on these scores we predict whether an edge is present between nodes \((b_i, b_j)\) or not, and construct the PrecisionRecall(PR)curve^{45} of this classifier. The area under the PRcurve (AUPR) measures the quality of this predictor, with an AUPR of 1.0 representing perfect prediction.
Wilcoxon–Mann–Whitney test
The Wilcoxon–Mann–Whitney^{46} test is a nonparametric test to decide whether the distributions of two populations are identical without assuming them to follow the normal distribution. Let X be a sample of size m from the first population and Y be a sample of size n from the second population. Consider the combined sample of size \(m+n\) ordered from least to greatest and denote the ranks of \(Y_{i}\) in this joint ordering by \(S_{i}\), \(i = 1, \ldots , n\). Then the test statistic \(W = \sum _{i=1}^{n} S_{i}\) is the sum of the ranks assigned to the values of Y.
ANOVA for circular data
With the Analysis of Variance for circular data^{36}, we test for the equality of p mean directions from independent circular (i.e. taking values on the unit circle) populations with vonMises (M) distribution and the same (unknown) concentration parameter \(\kappa\). We test the null hypothesis \(H_{0} :\mu _{1} = \ldots = \mu _{p}\), where \(\mu _{i}\) are the mean directions for the p populations following a \(\text {M}(\mu _{i}, \kappa )\) distribution. For any circular observation \(\theta\), denote \(s = \sin (\theta )\), \(c = \cos (\theta )\) and let \({\bar{s}}_i, {\bar{c}}_i\) be the averages within the ith population. Let \(n_{i}\) be the sample size, \(R_{i} = \sqrt{\bar{s}_i^2 + {\bar{c}}_i^2}\) the mean resultant length of the ith population, and let \(n = \sum _{i=1}^{p} n_{i}\) be the size of the combined sample and R the overall mean resultant length based on all n observations. The identity
has the approximate \(\chi ^{2}\) decomposition \(\chi ^{2}_{n1} = \chi ^{2}_{np} + \chi ^{2}_{p1}\)^{36} and therefore, the test statistic \(F = \frac{(\sum _{i=1}^{p} R_{i}  R) / (p1)}{\sum _{i=1}^{p} (n_{i}  R_{i}) / (np)}\) can be derived. The null hypothesis is rejected for a given confidence level \(\alpha\), when \(F > F_{p1, np; \alpha }\), where \(F_{p1, np; \alpha }\) is the \(\alpha\)quantile of the Fdistribution with \(p1\) and \(np\) degrees of freedom^{36}.
Data availability
The data analysed during the current study are available from the website of the European Banking Authority at https://www.eba.europa.eu/riskanalysisanddata/euwidestresstesting and https://eba.europa.eu/riskanalysisanddata/euwidetransparencyexercise/2018.
References
Shavitt, Y. & Tankel, T. On the curvature of the internet and its usage for overlay construction and distance estimation. In IEEE INFOCOM 2004, vol. 1 (IEEE, 2004).
Muscoloni, A., Thomas, J. M., Ciucci, S., Bianconi, G. & Cannistraci, C. V. Machine learning meets complex networks via coalescent embedding in the hyperbolic space. Nat. Commun. 8, 1–19 (2017).
AlanisLobato, G., Mier, P. & AndradeNavarro, M. A. Manifold learning and maximum likelihood estimation for hyperbolic network embedding. Appl. Netw. Sci. 1, 1–14 (2016).
Papadopoulos, F., Kitsak, M., Serrano, M. Á., Boguná, M. & Krioukov, D. Popularity versus similarity in growing networks. Nature 489, 537 (2012).
Papadopoulos, F., Psomas, C. & Krioukov, D. Network mapping by replaying hyperbolic growth. IEEE/ACM Trans. Netw. (TON) 23, 198–211 (2015).
Barabasi, A.L. Luck or reason. Nature 486, 507–509 (2012).
Brockmann, D. & Helbing, D. The hidden geometry of complex, networkdriven contagion phenomena. Science 342, 1337–1342 (2013).
Cont, R., Moussa, A. & Santos, E. B. Network structure and systemic risk in banking systems. In Network Structure and Systemic Risk in Banking Systems (ed. JeanPierre Fouque, J. A. L.) (Cambridge University Press, Cambridge, 2010).
Battiston, S., Gatti, D. D., Gallegati, M., Greenwald, B. & Stiglitz, J. E. Liaisons dangereuses: increasing connectivity, risk sharing, and systemic risk. J. Econ. Dyn. Control 36, 1121–1141 (2012).
Roukny, T., Bersini, H., Pirotte, H., Caldarelli, G. & Battiston, S. Default cascades in complex networks: topology and systemic risk. Sci. Rep. 3, 2759 (2013).
Battiston, S., Caldarelli, G., May, R. M., Roukny, T. & Stiglitz, J. E. The price of complexity in financial networks. Proc. Natl. Acad. Sci. 113, 10031–10036 (2016).
French, K. et al. The Squam Lake report: fixing the financial system. J. Appl. Corp. Finance 22, 8–21 (2010).
GarcíaPérez, G., Boguñá, M., Allard, A. & Serrano, M. Á. The hidden hyperbolic geometry of international trade: world trade atlas 1870–2013. Sci. Rep. 6, 33441 (2016).
European Banking Authority. EUwide stress testing. https://www.eba.europa.eu/riskanalysisanddata/euwidestresstesting.
European Banking Authority. EUwide transparency exercise. https://eba.europa.eu/riskanalysisanddata/euwidetransparencyexercise/2018.
Karrer, B. & Newman, M. E. Stochastic blockmodels and community structure in networks. Phys. Rev. E 83, 016107 (2011).
Caccioli, F., Farmer, J. D., Foti, N. & Rockmore, D. Overlapping portfolios, contagion, and financial stability. J. Econ. Dyn. Control. 51, 50–63. https://doi.org/10.1016/j.jedc.2014.09.041 (2015).
Eisenberg, L. & Noe, T. H. Systemic risk in financial systems. Manag. Sci. 47, 236–249 (2001).
Brown, M., Trautmann, S. T. & Vlahu, R. Understanding bankrun contagion. Manag. Sci. 63, 2272–2282 (2017).
Shleifer, A. & Vishny, R. W. Liquidation values and debt capacity: a market equilibrium approach. J. Finance 47, 1343–1366. https://doi.org/10.1111/j.15406261.1992.tb04661.x (1992).
Glasserman, P. & Young, H. P. How likely is contagion in financial networks?. J. Bank. Finance 50, 383–399. https://doi.org/10.1016/j.jbankfin.2014.02.006 (2015).
Cont, R. & Schaanning, E. Fire sales, indirect contagion and systemic stress testing. Norges Bank Working Paper 02/2017 (2017).
Cont, R. & Wagalath, L. Fire sales forensics: measuring endogenous risk. Math. Finance 26, 835–866 (2016).
Serrano, M. Á., Boguná, M. & Vespignani, A. Extracting the multiscale backbone of complex weighted networks. Proc. Natl. Acad. Sci. 106, 6483–6488 (2009).
Kruskal, J. B. & Wish, M. Multidimensional Scaling Vol. 11 (Sage, Thousand Oaks, CA, 1978).
Chowdhary, K. & Kolda, T. G. An improved hyperbolic embedding algorithm. J. Complex Netw. 6, 321–341 (2017).
KellerRessel, M. & Nargang, S. Hydra: a method for strainminimizing hyperbolic embedding of network and distance based data. J. Complex Netw. 8, cnaa002 (2020).
GarcíaPérez, G., Allard, A., Serrano, M. Á. & Boguñá, M. Mercator: uncovering faithful hyperbolic embeddings of complex networks. New J. Phys. 21, 123033 (2019).
Li, T., Levina, E. & Zhu, J. randnet: Random Network Model Selection and Parameter Tuning. R package version 0.2 (2019).
Serrano, M. A., Krioukov, D. & Boguná, M. Selfsimilarity of complex networks and hidden metric spaces. Phys. Rev. Lett. 100, 078701 (2008).
Boss, M., Elsinger, H., Summer, M. & Thurner, S. Network topology of the interbank market. Quant. Finance 4, 677–684 (2004).
Langfield, S., Liu, Z. & Ota, T. Mapping the UK interbank system. J. Bank. Finance 45, 288–303 (2014).
Financial Stability Board. 2014 update of list of global systemically important banks (GSIBs). https://www.fsb.org/2014/11/2014updateoflistofglobalsystemicallyimportantbanks/.
Financial Stability Board. 2016 list of global systemically important banks (GSIBs). https://www.fsb.org/2016/11/2016listofglobalsystemicallyimportantbanksgsibs/.
Financial Stability Board. 2018 list of global systemically important banks (GSIBs). https://www.fsb.org/2018/11/2018listofglobalsystemicallyimportantbanksgsibs/.
Mardia, K. V. & Jupp, P. E. Directional Statistics (Wiley, Hoboken, 2009).
Cont, R. & Wagalath, L. Running for the exit: distressed selling and endogenous correlation in financial markets. Math. Finance: Int. J. Math. Stat. Financ. Econ. 23, 718–741 (2013).
Ratcliffe, J. G. Foundations of Hyperbolic Manifolds Vol. 3 (Springer, Berlin, 1994).
Cannon, W. J., Floyd, W. J., Kenyon, R. & Parry, W. R. Hyperbolic geometry. In Flavors of Geometry 31st edn (ed. Levy, S.) 59–115 (MSRI Publications, Cambridge, 1997).
Borg, I. & Groenen, P. Modern multidimensional scaling: theory and applications. J. Educ. Meas. 40, 277–280 (2003).
Sammon, J. W. A nonlinear mapping for data structure analysis. IEEE Trans. Comput. 100, 401–409 (1969).
Zhao, X., Sala, A., Zheng, H. & Zhao, B. Y. Efficient shortest paths on massive social graphs. In 7th International Conference on Collaborative Computing: Networking, Applications and Worksharing (CollaborateCom) 77–86. (IEEE, 2011).
R Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, Vienna, 2019).
Galperin, G. A concept of the mass center of a system of material points in the constant curvature spaces. Commun. Math. Phys. 154, 63–84 (1993).
Saito, T. & Rehmsmeier, M. The precisionrecall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLoS ONE 10, e0118432 (2015).
Wilcoxon, F. Individual comparisons by ranking methods. Biometr. Bull. 1, 80–83 (1945).
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Contributions
M.K.R. conceived the study, S.N. prepared the data, M.K.R. and S.N. analyzed the results. M.K.R wrote the main manuscript text and all authors reviewed the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
KellerRessel, M., Nargang, S. The hyperbolic geometry of financial networks. Sci Rep 11, 4732 (2021). https://doi.org/10.1038/s41598021833284
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598021833284
This article is cited by

Convexity of Sets and Quadratic Functions on the Hyperbolic Space
Journal of Optimization Theory and Applications (2022)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.