{"paper_id":"2410.07222","title":"Computing Systemic Risk Measures with Graph Neural Networks","authors":["Lukas Gonon","Thilo Meyer-Brandis","Niklas Weber"],"paper_date":"2024-09-30","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2410.07222v2","page_url":"https://thequant.space/flowcharts/computing-systemic-risk-measures-with-graph-neural-networks/","scores":{"empirical_rigor":6,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes using permutation equivariant neural network architectures, specifically graph neural networks (GNNs) and extended permutation‑equivariant neural networks (X)PENNs, to approximate optimal random bailout allocations in stochastic financial networks.","quote":"We propose to use permutation equivariant architectures of neural networks like graph neural networks (GNNs) and a class that we name (extended) permutation equivariant neural networks ((X)PENNs). we characterise what we call permutation-equivariant node-labelling functions and, mo-tivated by this result, propose the neural network architecture of (X)PENNs.","summary":"This paper investigates systemic risk measures for stochastic financial networks of explicitly modelled bilateral liabilities. We extend the notion of systemic risk measures from Biagini, Fouque, Fritelli and Meyer-Brandis (2019) to graph structured data. In particular, we focus on an aggregation…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Assumption 2.10 on the aggregation function, which includes properties such as concavity and boundedness, holds for the total payment shortfall based on the Eisenberg‑Noe clearing mechanism.","type":"model","stated":"explicit","quote":"For an overview of network reconstruction techniques see for example [4]. • Under certain conditions on the aggregation function (Assumption 2.10), which in particular include aggregation functions based on the contagion model of Eisenberg and Noe, such as the total payment shortfall, we show that the systemic risk measure is well-defined. Property (i) of Assumption 2.10 is trivially fulfilled by the total payment","support":"full"},{"assumption":"Mild regularity assumptions ensure the existence of a measurable target function Hc that yields an optimal allocation for any fixed bailout capital.","type":"statistical","stated":"explicit","quote":"We further derive a reformulation of the systemic risk measure that allows us to construct an iterative algorithm for its numerical approximation. • Under mild assumptions, we show that there exists a measurable target function Hc : RN × RN×N → RN such that Hc(A, L) is an optimal allocation of the fixed bailout capital c ∈R. Due to its measura-bility, this allows for approximation via neural networks. • Furthermore,","support":"full"},{"assumption":"The Eisenberg‑Noe market‑clearing mechanism is used as the basis for the aggregation functions considered.","type":"market","stated":"explicit","quote":"2.3 Eisenberg–Noe clearing mechanism Now, we present the market clearing mechanism from the seminal work of Eisenberg and Noe [28]. Even though our systemic risk measures are defined in terms of general aggregation functions, our main focus in this paper is on aggregation functions based on the Eisenberg–Noe clearing mechanism.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Synthetic cascade and star networks","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":"Networks with N ∈ {10,20,50,100} nodes, each with two realizations (cascade and star) per ","quote":"Each respective dataset contains realisations with N ∈{10, 20, 50, 100} nodes. Specifically, for every i ∈1, . . . , N, each dataset contains two distinct realisations of the network GN: One realisation is a default cascade.","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"synthetic"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"Existence of an optimal random allocation that distributes the minimal bailout capital and secures the network is established.","kind":"theoretical","quote":"In this setting, we show the existence of an optimal random allocation that distributes the overall minimal bailout capital and secures the network. In particular, we show existence of “optimal” random bailout capital allocations and investigate their properties from a theoretical point of view.","support":"full"},{"finding":"In numerical experiments, XPENN, PENN, and GNN outperform other benchmark models, with XPENN showing a slight advantage.","kind":"performance","quote":"In the ER and CP networks, XPENN, PENN, and GNN are the best models, with XPENN maintaining a slight advantage compared to the others.","support":"full"},{"finding":"Permutation‑equivariant benchmark models (Level‑1, Linear, Default) work but are less effective than the neural network approaches.","kind":"performance","quote":"The permutation-equivariant benchmarks Level-1, Linear, and Default work, but cannot provide as effective capital allocations.","support":"full"},{"finding":"FNN, FNN(L), Constant, and Uniform models do not yield satisfactory bailout allocation results.","kind":"performance","quote":"FNN, FNN(L), Constant, and Uniform do not yield satisfactory results.","support":"full"},{"finding":"For the CPf network, FNN and FNN(L) can be useful and even outperform GNN, while PENN, XPENN, and GNN remain strong performers.","kind":"performance","quote":"In the CPf network, the results suggest again, that for a fixed interbank structure – besides PENN, XPENN, and GNN – the models FNN, FNN(L), and even Constant are useful, with FNN and FNN(L) even outperforming GNN.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":31,"citations":{"cited_by_in_corpus":1,"by_independent_authors":1,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":1,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":true,"code_url":"https://github.com/NiklasMWeber/ComputingSystemicRiskMeasures","note":null},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":2,"extracted_at":"2026-11-14T19:54:48Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}