It Takes Two to Tango, but More to Assess Systemic Risk: Credit Networks Through the Lens of Hypergraphs

Federico D. Forte

REVIEW 3 major objections 5 minor 13 references

Hypergraph centrality that weights each bank by its own lending share picks the institutions whose distress amplifies commercial-credit shocks most.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 08:09 UTC pith:D7UTLVK2

load-bearing objection Solid first hypergraph application to modern bank–firm credit with a usable weighted H-eigenvector; the amplification edge is real in-sample but partly circular and under-tested against size/shared-exposure nulls. the 3 major comments →

arxiv 2607.10943 v1 pith:D7UTLVK2 submitted 2026-07-12 econ.EM

It Takes Two to Tango, but More to Assess Systemic Risk: Credit Networks Through the Lens of Hypergraphs

Federico D. Forte This is my paper

classification econ.EM

keywords hypergraphssystemic riskcredit networksH-eigenvector centralityDebtRankbank-firm creditfinancial supervision

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that bank–firm credit is not just a collection of bilateral loans. When several banks finance the same firm they share a simultaneous multilateral exposure that standard graphs flatten into pairwise links. Representing those co-financing sets as hyperedges, then computing an adjusted H-eigenvector centrality that multiplies neighbors’ centrality by each bank’s own amount lent inside every hyperedge, ranks institutions whose joint distress produces the largest second-round losses. Applied to Argentine commercial-credit registry data (August 2023–December 2025), the top-five banks chosen by this measure generate roughly five extra percentage points of total credit affected relative to degree, strength or ordinary eigenvector rankings. The claim is that supervisors gain a practical, complementary lens for spotting systemically relevant lenders that traditional metrics miss.

Core claim

An H-eigenvector centrality computed on hypergraphs that have been uniformized by uplifting small hyperedges and/or projecting large ones, then re-weighted by each bank’s individual lending amount inside every hyperedge, consistently identifies the five institutions whose distress produces the greatest amplified systemic impact—measured as the share of total commercial credit that becomes affected after DebtRank-style propagation—outperforming all other centrality rankings tested in every month of the sample.

What carries the argument

Adjusted H-eigenvector centrality on uniformized hypergraphs (HIP_HE_uplift_adj / HIP_HE_upproj_adj): after hyperedges are made the same size via auxiliary-node uplift or projection, each node’s score multiplies the product of its neighbors’ centralities by the amounts that node and its neighbors actually lend inside that hyperedge.

Load-bearing premise

That turning co-financing sets into hyperedges and then forcing those hyperedges to a single common size by adding artificial nodes or by splitting large groups still keeps the economically important higher-order structure intact for both ranking and shock propagation.

What would settle it

Re-run the same DebtRank exercise on the identical Argentine data after replacing the uniformized hypergraph with the raw bipartite network (or with a simple one-mode bank–bank projection) and check whether the adjusted hypergraph ranking still produces a reliably larger amplified impact; if the ranking gap disappears or reverses, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Solid first hypergraph application to modern bank–firm credit with a usable weighted H-eigenvector; the amplification edge is real in-sample but partly circular and under-tested against size/shared-exposure nulls.

read the letter

This is the first clean hypergraph treatment of a modern bank–firm credit registry for systemic-risk ranking, plus a concrete individual-contribution weighting of H-eigenvector after Contreras-Aso uniformization. That is the real novelty. The pipeline is coherent: Argentine BCRA commercial loans (Aug 2023–Dec 2025), eleven centralities, Kendall/top-20 comparisons, monthly adapted DebtRank, and a bipartite robustness check that leaves relative rankings intact. The adjusted metrics (HIP_HE_uplift_adj / upproj_adj) win amplified impact every month, roughly 5 pp of total commercial credit above traditional degree/strength/eigenvector. Math and citations look solid; the appendices on Tudisco–Higham and uplift/projection are usable.

What it does well is make the higher-order co-financing structure explicit and give supervisors a complementary ranking that is not just “largest lenders.” The individual weight h_ie inside each hyperedge is a sensible fix for the free-rider problem of small banks riding large co-exposures.

Soft spots are real but not fatal. Uniformization (modal size or p=19 covering ≥80% of volume, discard larger edges) and the free parameters (threshold, top-k=5, α) are design choices without sensitivity that isolates higher-order structure from size and shared exposure. DebtRank propagates on the same weighted incidence matrix, so the evaluation shares drivers with the proposed score; the paper itself calls the win “a priori intuitive.” Appendix D only re-runs DebtRank on the bipartite graph; it does not ablate total lending or use a size-matched null. No uncertainty bands or code/data. Those are medium concerns for a methods paper, not load-bearing contradictions.

For network systemic-risk people and higher-order finance applications this is worth reading. It deserves a serious referee, not a desk reject. I would engage, cite the application and the weighted H-eigenvector recipe, and push for public materials and a cleaner null. Bring it to reading group if the group cares about credit networks or hypergraphs; otherwise optional.

Referee Report

3 major / 5 minor

Summary. The paper applies hypergraphs to Argentine commercial bank–firm credit registry data (Aug 2023–Dec 2025) to represent multi-bank co-financing of the same firm as a single hyperedge rather than a collection of pairwise links. It compares traditional bipartite and one-mode-projection centralities (degree, strength, eigenvector) with several hypergraph measures (vectorial centrality, Tudisco–Higham family, and H-eigenvector after Contreras-Aso uniformization), and proposes an adjusted H-eigenvector that multiplies neighbors’ centralities by each bank’s own lending amount in each hyperedge (HIP_HE_uplift_adj / HIP_HE_upproj_adj). Systemic relevance is then scored by an adapted DebtRank cascade on the weighted incidence matrix: the top-five banks under each ranking are fully distressed and the amplified (indirect) share of total commercial credit affected is compared. The main claim is that the proposed individually weighted, uniformized H-eigenvector rankings identify the set of institutions with the largest amplification—on average about 5 percentage points of total credit higher than traditional metrics—in every month of the sample.

Significance. If the ranking advantage is not an artifact of size and shared-exposure factors already present in strength or projected eigenvector centrality, the paper would supply a concrete, implementable complement to existing SIFI identification tools that explicitly encodes multilateral co-financing. The empirical pipeline is coherent: multiple centrality families, Kendall and top-20 overlap comparisons, monthly cascades, and a bipartite DebtRank robustness check (Appendix D) that leaves relative rankings intact. The proposed individual-weighting formula and the adaptation of DebtRank to hyperedges are clearly stated and could be reused on other credit registries. The contribution is therefore of genuine interest to network finance and supervisory practice, provided the higher-order content of the result is isolated more carefully.

major comments (3)

§5 and Fig. 9: the central claim that HIP_HE_*_adj identify institutions with strictly greater shock-amplification capacity rests on a comparison in which both the ranking metric and the evaluation metric (DebtRank on the weighted incidence matrix H) are driven by portfolio size and shared hyperedge exposures. The paper itself notes that the result is ‘a priori intuitive’ for this reason. No ablation holds total lending fixed, reweights by relative rather than absolute contribution, or recomputes amplification under a size-matched null (e.g., configuration-model or strength-preserving rewiring of hyperedges). Without such a check, it remains open whether the ~5 pp edge isolates higher-order structure or simply re-packages BIP_strength / ONE_eigenvector information.
§4 and Appendix B: uniformization choices are load-bearing free parameters. Target order p is set either to the modal hyperedge size or to 19 (covering ≥80% of loan volume), with hyperedges larger than 19 discarded; the combinatorial uplift weight ŵ(e)=w(e)(p−k)k!/p! and the projection of large edges are taken from Contreras-Aso et al. without systematic sensitivity. Because non-uniformity is a defining feature of the data (hyperedge sizes 2–36), the claim that the transformed hypergraph preserves the economically relevant co-financing structure needs at least a leave-one-choice-out or range-of-p robustness table for the amplification ranking.
§5 / Fig. 9 / Appendix C: amplified impacts are reported as point averages with no standard errors, bootstrap intervals, or formal tests of the month-by-month dominance of HIP_HE_*_adj over the other ten metrics. Given that the paper asserts superiority ‘in every month’, even a simple paired comparison or rank-sum test across months would substantially strengthen the statistical claim.

minor comments (5)

Dates: the sample is given as August 2023–December 2025 and the version date as July 2026; a short note on data vintage / real-time availability would avoid confusion.
Table 2 and the HIP_HE_* abbreviations are dense; a one-line formula column or a short pseudocode box for the individual-weighting fixed point would help readers implement the measure.
Figure 9 and Appendix C: solid vs hatched red columns are hard to distinguish in grayscale; consider patterns or direct labels.
Notation: the incidence matrix is H in §5 but B in Appendix A; a single symbol would reduce friction.
The commercial-loan inclusion threshold (twice the microenterprise sales ceiling, inflation-adjusted) is reasonable but should be listed among free parameters and, if possible, subjected to a brief sensitivity check.

Circularity Check

1 steps flagged

Adjusted H-eigenvector multiplies own lending amounts with neighbor centralities; DebtRank impact uses the same weighted exposures, so superior amplification is partly by construction (paper calls it a priori intuitive).

specific steps

self definitional [§4 (proposed weighted H-eigenvector) + §5 (DebtRank impact + “a priori intuitive” paragraph) + Fig. 9]
"H_eigenvector centrality weighted by individual contribution to each hyperedge: 𝜆𝑐𝑖^{𝑚−1}=∑ ℎ𝑖𝑒 ∏ ℎ𝑗𝑒 𝑐𝑗 ... This result was, a priori, intuitive, since the metric proposed in this paper captures the two most relevant dimensions ... how much each institution lends in the network (as a proxy for size), and, simultaneously, the centrality of its neighbors in a setting of multilateral relationships with nonlinear, multiplicative interactions. ... the largest amplified impact is generated ... by ... HIP_HE_uplift_adj and HIP_HE_upproj_adj."

The centrality score is defined directly from the same bank-hyperedge lending amounts ℎ_𝑖𝑒 (and the hyperedge structure) that enter the DebtRank recursion via the weighted incidence matrix H, L_i = ∑ H_ie and Δg_e. Ranking institutions by a nonlinear function of precisely those quantities and then evaluating the amplified impact of the resulting top-5 therefore selects high-impact sets partly by construction of the ranking metric itself; the paper explicitly notes the outcome is a priori intuitive for this reason. The reduction is not total (higher-order product still differs from pure strength), but the central claim of superior shock-amplification capacity is not independent of the metric’s built-in drivers.

full rationale

The paper is an empirical comparison of centrality rankings on Argentine credit-registry hypergraphs, not a first-principles derivation of a closed-form prediction. No parameter is fitted to a subset and then re-predicted; no uniqueness theorem or load-bearing self-citation forces the result; methods for uniformization and base H-eigenvector are taken from independent sources (Contreras-Aso et al., Benson, Tudisco-Higham). The sole circularity is design-level: the proposed HIP_HE_*_adj centrality is explicitly constructed to embed the two drivers (own lending volume + nonlinear multilateral neighbor centrality) that also govern the adapted DebtRank amplification computed from the identical weighted incidence matrix H. Selecting top-5 by that construction and then measuring their amplified impact therefore yields a higher score partly by design rather than as an independent discovery. The authors themselves label the outcome “a priori intuitive” for exactly this reason. Traditional strength already captures size, yet the multiplicative hypergraph version still outperforms; the circularity is therefore partial (score 5), not total equivalence. No ablation that freezes total lending or uses a size-matched null is reported, reinforcing that the evaluation shares the metric’s inputs. External validity and uniformization distortion remain open but are correctness, not circularity, issues.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 1 invented entities

The central claim rests on modeling co-financing as hyperedges, on uniformization choices that make H-eigenvectors well-defined, on an ad-hoc individual-weighting formula, and on a DebtRank-style cascade as the ground truth for ‘systemic impact.’ Several numeric cutoffs (loan threshold, p=19, top-5, 80% volume rule) are free design parameters. No new physical entities are postulated; the invented object is the adjusted centrality score itself.

free parameters (5)

Commercial-loan inclusion threshold = ~ARS 2.74bn (Dec 2025); real-constant over sample
Firms kept only if total credit exceeds the BCRA commercial reference (twice microenterprise sales max), inflation-adjusted; defines the network and captures ~80.3% of firm credit (§2).
Uniformization target order p = modal size ~5.3; or p=19
Hyperedges forced to modal size (uplift+project) or uplifted to 19 with larger edges discarded because they cover ≥80% of lending (§4).
Top-k shocked institutions = k=5
Systemic impact always computed for k=5 top-ranked banks per metric (§5); choice is not derived.
Tudisco–Higham max-model exponent α = α=10
α=10 used to approximate max aggregation (Appendix A3); standard in that paper but free here.
DebtRank convergence tolerance / iteration stop = near-zero Δ distress; ~48 iters
Iteration until new aggregate distress near zero; average 48 iterations/month (§5).

axioms (5)

domain assumption Co-financing of the same firm(s) by m banks is economically a simultaneous multilateral hyperedge, not merely a set of bilateral edges.
Core modeling premise of §§1 and 3; without it the hypergraph representation is optional re-encoding.
standard math Uplift (auxiliary node ★ with combinatorial weight correction) and clique-style projection preserve a well-defined positive H-eigenvector ranking of original banks under Perron–Frobenius-type tensor conditions.
Invoked via Contreras-Aso et al. (2024) in §4 and Appendix B; required for non-uniform hypergraphs.
domain assumption Adapted DebtRank bank–hyperedge–bank iteration with (1−h_i) caps correctly measures economically meaningful amplified systemic impact of initial full distress.
§5 evaluation criterion; adapted from Battiston et al., Silva et al., Aoyama et al.
ad hoc to paper Individual hyperedge weights h_ie should enter multiplicatively with neighbors’ centralities in the H-eigenvector fixed-point (proposed formula in §4).
Author’s refinement; not derived from a uniqueness theorem, motivated by avoiding small participants free-riding on large co-lenders.
domain assumption Binary or amount-weighted incidence of commercial loans above the threshold is a sufficient map of systemic credit interconnections (retail/consumer excluded).
§2 sample construction; commercial focus justified by concentration but is a modeling choice.

invented entities (1)

HIP_HE_uplift_adj / HIP_HE_upproj_adj (individually weighted H-eigenvector centrality) no independent evidence
purpose: Rank banks by nonlinear combination of own lending in each hyperedge and neighbors’ centralities after uniformization.
New score defined in §4; evaluated only inside this paper’s DebtRank exercise on the same Argentine hypergraphs—no external falsifiable prediction (e.g., future default events) is tested.

pith-pipeline@v1.1.0-grok45 · 23763 in / 3865 out tokens · 35827 ms · 2026-07-14T08:09:07.820955+00:00 · methodology

0 comments

read the original abstract

This paper provides the first analysis of credit relationships between financial institutions and firms through the lens of hypergraphs. Unlike traditional network approaches, which rely on pairwise connections, this framework explicitly represents the shared exposure of multiple financial institutions to the same firm as a simultaneous multilateral relationship. The approach is applied empirically to Credit Registry data from the Central Bank of Argentina, covering the period from August 2023 to December 2025 and focusing on commercial loans between banks and firms. Traditional centrality metrics are compared with hypergraph-specific measures to identify systemically relevant institutions. The paper also proposes an adjusted version of H-eigenvector centrality that nonlinearly weights both the centrality of neighboring institutions and each creditor's lending amount, in order to assess the relevance of a bank within the network. The systemic impact of shocking the top-ranked institutions according to each centrality metric is then estimated through an adaptation of the DebtRank algorithm. The results show that the proposed framework identifies institutions with greater shock-amplification capacity, providing a complementary tool for financial supervision and regulation.

discussion (0)

Reference graph

Works this paper leans on

13 extracted references · 3 canonical work pages

[1]

too-big-to-fail

Introduction Credit relationships constitute one of the fundamental pillars of financial systems. They are the core activity of banks and usually account for the largest share of their assets. At the same time, credit is one of the main drivers of firms’ investment and growth, and consequently of economic activity in general. For this reason, the relation...

2012
[2]

commercial

The commercial credit network in Argentina The main data source used in this paper is the Credit Registry of the Central Bank of Argentina (BCRA), covering the period from August 2023 to December 2025. We focus on loans classified as “commercial”, defined as loans granted to firms in connection with their productive activities. Accordingly, personal, mort...

2023
[3]

Basic topological metrics: bipartite networks, one-mode projections and hypergraphs As mentioned above, the literature has usually addressed this type of interconnection s through the concept of bipartite networks. For example, this perspective has been applied to the analysis of credit networks in Japan (Fujiwara et al., 2009), Italy (De Masi & Gallegati...

2009
[4]

Identification of systemically important financial institutions Bipartite networks One of the main contributions of network analysis and graph theory to the literature on financial systems has been the development of tools for identifying the most central institutions in a network, thereby helping to quantify and mitigate the systemic risk associated with...

2023
[5]

The first approach proposed to quantify centrality values in non-uniform hypergraphs was the notion of vectorial centrality score. Instead of just a single centrality value assigned to each node, a centrality vector is constructed for each node 𝑖, 𝒗𝑖 = ℝ𝑚−1, where each component of the vector represents the centrality value of that node in the correspondi...

2022
[6]

log -exp

A second proposal was put forward by Tudisco & Higham (2021), who introduce a family of spectral centrality measures for non -uniform hypergraphs , which allows a simultaneous computation of the relative importance of both the nodes and hyperedges. The central idea is to extend the logic of eigenvector centrality in traditional graphs to settings in which...

2021
[7]

uniformization

Finally, the most recent proposal (and the one discussed in greater depth later in this paper ), is that of Contreras-Aso et al. (2024). They propose a “uniformization” method for making hyperedge sizes equal in hypergraphs that are originally non -uniform. Once the baseline hypergraphs have been “made” uniform, H-eigenvector centrality measures can be co...

2024
[8]

To this end, we compute the impact on the network resulting from a distress shock to the top five institutions ranked by each of the eleven metrics considered here (Table 2)

Systemic impact quantification In this section, we assess the implications of identifying relevant agents in our financial network according to each of the centrality measures described above. To this end, we compute the impact on the network resulting from a distress shock to the top five institutions ranked by each of the eleven metrics considered here ...

2012
[9]

Concluding remarks This paper proposes the use of hypergraphs to provide a more comprehensive analysis of the structure of credit relationships between financial institutions and firms. In this context, the joint exposures of multiple banks to a firm can be more realistically understood as an underlying multilateral relationship among all institutions len...

2024
[10]

maximum exposure

Appendices Appendix A. Node and edge nonlinear eigenvector centrality for hypergraphs (Tudisco & Higham, 2021) Let 𝐻 = (𝑉, 𝐸) be a hypergraph with 𝑛 nodes and 𝑚 hyperedges. The structure of the hypergraph can be represented by an incidence matrix 𝐵 ∈ ℝ𝑛×𝑚 , whose elements are defined as 𝐵𝑖,𝑒 = { 1 if 𝑖 ∈ 𝑒, 0 otherwise The framework proposed by Tudisco & ...

2021
[11]

References Accominotti, O., Lucena -Piquero, D., & Ugolini, S. (2023). Intermediaries’ substitutability and financial network resilience: A hyperstructure approach. Journal of Economic Dynamics and Control, 153, 104700. https://doi.org/10.1016/j.jedc.2023.104700 Akgüller, Ö., & Balcı, M. A. (2026). Detecting Financial Contagion Through Higher-Order Networ...

work page doi:10.1016/j.jedc.2023.104700 2023
[12]

D., & Montes -Rojas, G

https://doi.org/10.1007/s00181-011-0512-x Elosegui, P., Forte, F. D., & Montes -Rojas, G. (2022). Network structure and fragmentation of the Argentinean interbank markets. Latin American Journal of Central Banking , 3(3), 100066. https://doi.org/10.1016/j.latcb.2022.100066 Fang, L., Zhao, T., Yu, J., Guo, Q., Li, X., & Zhang, C. (2025). TD-HCN: A trend-dr...

work page doi:10.1007/s00181-011-0512-x 2022
[13]

https://doi.org/10.1140/epjds/s13688-017-0114-8 Qi, L., & Luo, Z. (2017). Tensor analysis: Spectral theory and special tensors. Society for Industrial and Applied Mathematics. Silva, T. C., Alexandre, M. da S., & Tabak, B. M. (2018). Bank lending and systemic risk: A financial- real sector network approach with feedback. Journal of Financial Stability , 3...

work page doi:10.1140/epjds/s13688-017-0114-8 2017

[1] [1]

too-big-to-fail

Introduction Credit relationships constitute one of the fundamental pillars of financial systems. They are the core activity of banks and usually account for the largest share of their assets. At the same time, credit is one of the main drivers of firms’ investment and growth, and consequently of economic activity in general. For this reason, the relation...

2012

[2] [2]

commercial

The commercial credit network in Argentina The main data source used in this paper is the Credit Registry of the Central Bank of Argentina (BCRA), covering the period from August 2023 to December 2025. We focus on loans classified as “commercial”, defined as loans granted to firms in connection with their productive activities. Accordingly, personal, mort...

2023

[3] [3]

Basic topological metrics: bipartite networks, one-mode projections and hypergraphs As mentioned above, the literature has usually addressed this type of interconnection s through the concept of bipartite networks. For example, this perspective has been applied to the analysis of credit networks in Japan (Fujiwara et al., 2009), Italy (De Masi & Gallegati...

2009

[4] [4]

Identification of systemically important financial institutions Bipartite networks One of the main contributions of network analysis and graph theory to the literature on financial systems has been the development of tools for identifying the most central institutions in a network, thereby helping to quantify and mitigate the systemic risk associated with...

2023

[5] [5]

The first approach proposed to quantify centrality values in non-uniform hypergraphs was the notion of vectorial centrality score. Instead of just a single centrality value assigned to each node, a centrality vector is constructed for each node 𝑖, 𝒗𝑖 = ℝ𝑚−1, where each component of the vector represents the centrality value of that node in the correspondi...

2022

[6] [6]

log -exp

A second proposal was put forward by Tudisco & Higham (2021), who introduce a family of spectral centrality measures for non -uniform hypergraphs , which allows a simultaneous computation of the relative importance of both the nodes and hyperedges. The central idea is to extend the logic of eigenvector centrality in traditional graphs to settings in which...

2021

[7] [7]

uniformization

Finally, the most recent proposal (and the one discussed in greater depth later in this paper ), is that of Contreras-Aso et al. (2024). They propose a “uniformization” method for making hyperedge sizes equal in hypergraphs that are originally non -uniform. Once the baseline hypergraphs have been “made” uniform, H-eigenvector centrality measures can be co...

2024

[8] [8]

To this end, we compute the impact on the network resulting from a distress shock to the top five institutions ranked by each of the eleven metrics considered here (Table 2)

Systemic impact quantification In this section, we assess the implications of identifying relevant agents in our financial network according to each of the centrality measures described above. To this end, we compute the impact on the network resulting from a distress shock to the top five institutions ranked by each of the eleven metrics considered here ...

2012

[9] [9]

Concluding remarks This paper proposes the use of hypergraphs to provide a more comprehensive analysis of the structure of credit relationships between financial institutions and firms. In this context, the joint exposures of multiple banks to a firm can be more realistically understood as an underlying multilateral relationship among all institutions len...

2024

[10] [10]

maximum exposure

Appendices Appendix A. Node and edge nonlinear eigenvector centrality for hypergraphs (Tudisco & Higham, 2021) Let 𝐻 = (𝑉, 𝐸) be a hypergraph with 𝑛 nodes and 𝑚 hyperedges. The structure of the hypergraph can be represented by an incidence matrix 𝐵 ∈ ℝ𝑛×𝑚 , whose elements are defined as 𝐵𝑖,𝑒 = { 1 if 𝑖 ∈ 𝑒, 0 otherwise The framework proposed by Tudisco & ...

2021

[11] [11]

References Accominotti, O., Lucena -Piquero, D., & Ugolini, S. (2023). Intermediaries’ substitutability and financial network resilience: A hyperstructure approach. Journal of Economic Dynamics and Control, 153, 104700. https://doi.org/10.1016/j.jedc.2023.104700 Akgüller, Ö., & Balcı, M. A. (2026). Detecting Financial Contagion Through Higher-Order Networ...

work page doi:10.1016/j.jedc.2023.104700 2023

[12] [12]

D., & Montes -Rojas, G

https://doi.org/10.1007/s00181-011-0512-x Elosegui, P., Forte, F. D., & Montes -Rojas, G. (2022). Network structure and fragmentation of the Argentinean interbank markets. Latin American Journal of Central Banking , 3(3), 100066. https://doi.org/10.1016/j.latcb.2022.100066 Fang, L., Zhao, T., Yu, J., Guo, Q., Li, X., & Zhang, C. (2025). TD-HCN: A trend-dr...

work page doi:10.1007/s00181-011-0512-x 2022

[13] [13]

https://doi.org/10.1140/epjds/s13688-017-0114-8 Qi, L., & Luo, Z. (2017). Tensor analysis: Spectral theory and special tensors. Society for Industrial and Applied Mathematics. Silva, T. C., Alexandre, M. da S., & Tabak, B. M. (2018). Bank lending and systemic risk: A financial- real sector network approach with feedback. Journal of Financial Stability , 3...

work page doi:10.1140/epjds/s13688-017-0114-8 2017