Pith. sign in

REVIEW 3 major objections 5 minor 40 references

A few low-risk firms can, if they fail together, destroy more than 90% of an economy's output—up to 257 times the damage of their individual failures.

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 · deepseek-v4-flash

2026-08-01 10:52 UTC pith:6D3ASPAZ

load-bearing objection A genuinely new idea—simultaneous-failure amplification in firm-level supply chains—wrapped around a headline number (257x) that lives or dies by one unvalidated replaceability term in the ESRI model; worth refereeing, but the abstract overstates it. the 3 major comments →

arxiv 2607.20068 v1 pith:6D3ASPAZ submitted 2026-07-22 econ.GN q-fin.EC

Catastrophic disruption cascades driven by the nonlinearity of systemic risk

Jan Fialkowski , Shlomo Havlin , Stefan Thurner This is my paper
classification econ.GN q-fin.EC
keywords systemic risksupply chain networkscascading failuresnonlinear amplificationfirm-level economic datasupplier substitutabilityeconomic systemic risk index (ESRI)Ecuador
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 tries to establish that supply chains hide nonlinear risks: pairs or triples of firms that look harmless on their own can, if they fail at the same time, trigger cascades that destroy far more output than the sum of their individual effects. Using the complete firm-level supply network of Ecuador, the authors report joint failures that amplify damage by factors up to 257, while only about 0.14% of firm pairs show more than fourfold amplification. They trace the effect to a single mechanism—the breakdown of suppliers' replaceability—and show that rare high-risk combinations can be found with a sampling algorithm. If true, this means that monitoring only individual firms can miss catastrophic scenarios.

Core claim

The central claim is that additive intuition fails for systemic risk in production networks: the economic systemic risk of simultaneously failing firms can vastly exceed the sum of their individual systemic risks. Quantitatively, among roughly two million sampled pairs in Ecuador's national supply network, the authors identify 28 pairs, 10 triples, and one quadruple whose joint failure wipes out more than 90% of total network output, even though the median sum of their individual risks is only 0.094. The strongest case is a pair with summed individual risk 0.0036 whose joint failure reaches an ESRI (economic systemic risk index) of 0.94—an amplification factor of 257. They argue that the mec

What carries the argument

The load-bearing object is the supplier-replaceability factor σ_i(t), which the paper calls the effective market share of supplier i at time t: the ratio of that supplier's sales to the sales of all active suppliers in its 4-digit industry. At t=0 it is simply i's share of its industry; during a cascade it grows as rival suppliers in the same industry lose production. This time dependence introduces nonlinearity into the shock-propagation equation: a shock that hits several suppliers in one industry at once can push the effective market share of the remaining supplier toward 1, making it effectively irreplaceable and triggering a much larger downstream cascade than any single failure would.

Load-bearing premise

The 257-fold amplification comes from the model's assumption that replacement of a failed supplier depends only on current market shares within the same four-digit industry, leaving out price bidding, inventories, imports, and fast rewiring; if real substitution is stronger, the amplification shrinks.

What would settle it

Re-run the cascading simulation on the same Ecuador network but replace the effective-market-share formula with one that lets downstream firms immediately switch to any supplier in the same two-digit sector (or to imports); if no pair then exceeds about fourfold amplification, the 257-fold result is an artifact of the formula's industry granularity.

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

If this is right

  • If the paper is right, risk monitoring that scores firms only in isolation will systematically miss the most dangerous failure scenarios; regulators need to screen combinations of firms.
  • The amplification is a tail-risk phenomenon: it barely moves the average pairwise risk (about +0.71%) but creates rare events that can destroy most of the network's output, so aggregated statistics will not warn about it.
  • Networks with a systemic risk plateau (a small set of firms whose individual failure is already extremely damaging) are the breeding ground for strong amplification; networks without such plateaus (crustacean, banana) show almost none.
  • The identified amplification pairs are not transient: three high-amplification pairs found in the 2015 network also appear in the 2010 network, suggesting some dangerous combinations persist over years.
  • The random-chemistry style extraction procedure used here can be reused to locate hidden high-risk sets in other networks where exhaustive search is infeasible.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If real economies substitute more flexibly than the model assumes—through inventories, rapid rewiring, price-based allocation, or imports—the quantitative 257x figure is likely an upper bound; the paper itself notes these omissions, so the practical magnitude may be smaller, though the qualitative breakdown-of-compensation mechanism could persist.
  • A testable extension is to apply the same pair/triple extraction to another country with firm-level VAT data to see whether the heavy-tailed amplification distribution (exponent ≈1.8) and the dependence on a systemic-risk plateau are general features of production networks rather than an Ecuador-specific artifact.
  • The structural modes the paper identifies (same-industry pair, combined market-share pair, cross-industry cascade intersection) could be converted into early-warning motifs and monitored in real time with invoice data, before a disaster strikes.
  • For policy, the existence of rare high-amplification sets suggests that targeted interventions—such as backup capacity for specific firms or pre-negotiated substitution contracts in identified industries—may be far more cost-effective than blanket resilience measures.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. This paper studies whether the simultaneous failure of small sets of firms can produce systemic losses larger than the sum of the losses from the same firms failing individually. Using Ecuadorian VAT data, the previously published ESRI model, and a random-chemistry extraction procedure, the authors define an amplification factor alpha = ESRI(joint)/sum(individual ESRI) and report that rare pairs, triples, and quadruples can have alpha up to 257, that 0.14% of randomly sampled pairs have alpha > 4, and that the survival function of alpha follows a power law with exponent 1.79. The extraction method is validated exhaustively on the softdrink subnetwork, and a second country-year (2010) check is provided. The authors identify three structural motifs associated with strong amplification and argue that the core mechanism is a breakdown of supplier substitutability.

Significance. If the quantitative conclusions are robust, this is a significant result: it would imply that production networks contain rare 'hidden' joint vulnerabilities that are invisible to single-firm systemic-risk rankings, with direct implications for disaster risk, targeted shocks, and supply-chain monitoring. The paper has clear strengths: the code is released, the softdrink subnetwork provides an exhaustive validation of the extraction heuristic, the 2010 network provides a temporal check, and the discussion openly lists many omitted mechanisms. The central concern is that the headline amplification values are outputs of the ESRI model and, as I detail below, the entire superlinear effect is generated by a single time-dependent term in the replaceability formula, Eq. (2)/(9), which is itself an unvalidated behavioral assumption. The paper's internal logic is careful, but its central empirical claim needs to be re-framed or buttressed with robustness analysis.

major comments (3)
  1. [§4.2, Eq. (2)/(9)] The superlinear amplification that is the paper's central result is produced entirely by the time dependence of the effective market share sigma_i(t). For fixed sigma_i, the iteration in Eq. (3)/(10) is subadditive: the essential-input max term, the additive non-essential term, and the retained initial shock psi_j together satisfy F(x+y, psi1+psi2) <= F(x, psi1)+F(y, psi2) by induction, and the upstream equation is likewise subadditive; hence ESRI(psi_i+psi_j) <= ESRI(psi_i)+ESRI(psi_j). Therefore every reported alpha > 4 outcome, including alpha = 257, requires the denominator of Eq. (9) to shrink as other firms in the same 4-digit ISIC class fail. The Discussion (Sec. 3) concedes that price adjustment, inventories, rewiring, and imports are omitted and that 'such model adjustments would likely change the quantitative results,' but no robustness test of this specific functional form is
  2. [§2.2, Fig. 4] The paper describes the 'breakdown of substitutability' as the origin of the nonlinear amplification and presents three structural motifs. But this mechanism is not discovered independently; it is effectively hard-wired into Eq. (9), where a firm's replaceability decreases mechanically as the production of other firms in its industry falls. In particular, Mode C (different-industry pairs, shared downstream industry) is a direct restatement of the time-dependent denominator in Eq. (9). To support the stronger claim that this mechanism is the origin of large amplification in real supply chains, the authors should compare Eq. (9) with at least one alternative model of substitution — for instance, a fixed-replacement-rate model, a dynamic rewiring model, or a model with inventories — or validate the market-share mechanism on a few documented supply-chain shocks. Without this, Section 2.2 sho
  3. [§2.1 and Discussion, final paragraph] The counts of 28 pairs, 10 triples, and 1 quadruple with ESRI > 0.9 are obtained from a stochastic extraction procedure whose coverage is 96.8% for pairs but 'drops drastically' for triples and larger sets, as the Discussion acknowledges. The Results section presents these counts without this caveat, so a reader may read them as an enumeration of all such sets. The manuscript should state clearly in Section 2.1 that these are lower bounds under the sampling scheme, and should give the coverage probability for triples/quads in the main text (or omit the counts).
minor comments (5)
  1. [Fig. 3 and §2.1] The power-law fit with exponent 1.79 should report the fitted x_min, the confidence interval, and a goodness-of-fit comparison (e.g., lognormal or exponential alternatives). Since the tail above alpha > 4 contains only 2,836 pairs from a 2-million-pair sample, the exponent as stated is statistically under-specified.
  2. [§1] The text says the full network contains roughly 10^5 firms, 'corresponding to about 10^10 possible firm pairs.' With 86,385 firms the number of pairs is about 3.7x10^9, not 10^10. Please correct.
  3. [§4.3] The extraction procedure is described verbally as 'systematically removed until a minimal subset remains,' but the precise greedy-removal order and stopping rule are not stated. The softdrink validation alleviates concern, but a pseudocode or a precise step list would improve reproducibility.
  4. [SI B.4] The 0.71% average-increase calculation in Eq. (15) relies on the assumption that unobserved pairs behave like the random sample. The assumption is stated in the SI, but it should also be flagged in the main-text sentence citing this number.
  5. [Throughout] Typos: 'sofdrink' in Section 1; 'Not the log-scale' in SI Fig. 8 caption; 'we only examine consider pairs' in SI B.4. Also, the y-axis label of Fig. 3 uses 'Number of pairs with amplification > alpha' while the text says 'amplification factor'; please harmonize.

Circularity Check

0 steps flagged

No significant circularity: the 257x amplification is a model output, not a fitted prediction; the mechanism statement is in-model but not presented as an independent empirical discovery.

full rationale

The paper's central quantitative claim—that simultaneous failure of small firm sets can yield ESRI amplifications up to 257—is computed by applying the previously published ESRI model [28] to the Ecuadorian supply chain network. No parameter is fitted to the amplification factors or to the identified firm sets; alpha is a function of the model equations, the network data, and the exogenous shock vector. The identification of amplifying pairs is also validated against exhaustive enumeration in the softdrink subnetwork (SI A.3), which is genuine out-of-sample support for the search algorithm. The only step that could look circular is the statement that the 'origin of these amplifications is a breakdown of the substitutability of defaulted suppliers,' since Eq. 2 (Eq. 9 in SI) defines sigma_i(t) as the replaceability factor and is the sole source of superlinearity. However, the paper does not claim to have independently discovered that mechanism; it is a stated model assumption from the prior ESRI framework. The Discussion explicitly acknowledges that changes to this assumption 'would likely change the quantitative results,' confirming that the finding is model-dependent rather than constructed to match the output. Self-citations to [28] and [29] are load-bearing in the sense that the model and data come from the same group, but they are prior published work with stated assumptions that do not include the present alpha results, so they constitute legitimate inputs rather than circular reasoning. The limitation that the quantitative headline rests on the unvalidated replaceability specification is a correctness risk, not an internal circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 7 axioms · 0 invented entities

No new physical or economic entities are introduced; 'plateau-firm' is a descriptive label for firms with large individual ESRI. The central mechanism is the market-share-based replaceability term sigma_i(t), inherited from the ESRI model, and the headline amplification is a computed model output rather than a fitted constant.

free parameters (5)
  • theta1 (success threshold) = 3
    Hand-chosen threshold in Eq. 4 for classifying a 500- or 100-firm set as exhibiting nonlinear amplification; sensitivity tested in SI A.2 but affects which pairs are found.
  • theta2 (removal threshold) = 0.9
    Hand-chosen threshold in Eq. 5 deciding when removing a firm from a successful large set is considered significant; sensitivity tested in SI A.2.
  • initial sampled set size N = 500 and 100
    Sampling design parameter for random chemistry; larger N is more efficient at finding destructive sets but can obscure low-ESRI pairs, as noted in Materials and Methods.
  • convergence tolerance epsilon = 0.01
    Stopping criterion for iterating the propagation equations (Eqs. 3, 10-11); affects final ESRI values at the margin.
  • number of sampled sets = 100,000 per size
    Computational budget that determines pair coverage (about 96.8% for pairs) and therefore which amplifying sets are identified.
axioms (7)
  • domain assumption The ESRI model [28] is an adequate representation of supply-chain shock propagation.
    Invoked throughout Materials and Methods 4.2; the central claim is a model output rather than a direct measurement of observed cascades.
  • domain assumption 4-digit ISIC industry codes are a valid proxy for what a firm produces.
    Used in Eq. 2/9 and Eqs. 6-8 to define industries and supplier replaceability; product-level data are not available.
  • domain assumption The expert survey from [16] on essential and non-essential inputs transfers to Ecuadorian firms.
    Used to construct Lambda matrices; no Ecuador-specific validation is provided.
  • domain assumption Supplier replaceability is captured by the market-share formula in Eq. 2.
    This mechanism is the direct origin of the reported amplification; if real substitution involves price bidding, inventories, or rapid rewiring, the amplification could be much smaller.
  • standard math The iterative propagation equations (Eqs. 3/10-11) converge to a well-defined ESRI for every shock vector.
    The paper stops at epsilon=0.01 but does not prove convergence or uniqueness of the fixed point, though monotonicity of the dynamics makes failure unlikely.
  • domain assumption The VAT network is static and complete for 2015 after excluding 'personas naturales'.
    Yearly aggregation omits inventories, production times, and network rewiring; the Discussion acknowledges this and interprets results as an upper bound.
  • domain assumption The 2,000,000 randomly sampled pairs are representative of unobserved pairs in the average-risk calculation.
    Used in SI B.4 to compute the 0.71% average increase; this is an explicit assumption and is not guaranteed for rare tail pairs.

pith-pipeline@v1.3.0-alltime-deepseek · 22760 in / 13528 out tokens · 124070 ms · 2026-08-01T10:52:19.110357+00:00 · methodology

0 comments
read the original abstract

Whether the COVID-19 pandemic or the Iran war, recent events have highlighted the systemic fragility of supply chains. Due to highly specific and mutual buyer-supplier dependencies, even the failure of a single firm can cause system-wide economic disruptions in the form of cascading failures up and down the supply chain network. Only recently has it become possible to quantify the systemic impact of the failure of individual firms on the total supply chain. Here, we demonstrate that the systemic risk contributions of combinations of firm failures can be drastically larger than the sum of the damage caused by the firms individually. Using a unique data set that allows us to reconstruct the national supply chain network of Ecuador at the firm-level, we find that combined failures can produce systemic risk amplifications of up to a factor of 257. However, only a tiny fraction of 0.14\% of pairs exhibit a more than 4-fold amplification of systemic risk. We develop a simple method to identify firm combinations that lead to large systemic risk amplifications. The origin of these amplifications is a breakdown of the substitutability of defaulted suppliers. We discuss the implications of the existence of rare but strong systemic risk amplification for situations that simultaneously affect multiple firms, such as natural disasters and wars.

Figures

Figures reproduced from arXiv: 2607.20068 by Jan Fialkowski, Shlomo Havlin, Stefan Thurner.

Figure 1
Figure 1. Figure 1: Nonlinear amplification of systemic risk in a supply chain network. The left column illustrates a schematic example of nonlinear amplification. In panel A, a single supplier firm is shocked, marked by the red cross. As a result, this firm completely stops production and is shown in dark purple. The shock then propagates through the network and reduces production at downstream firms. Yellow indicates no pro… view at source ↗
Figure 2
Figure 2. Figure 2: Scatter plots of the ESRI for simultaneous firm failures versus the sum of individual firms’ ESRIs. Panels A and B show the ESRI of simultaneous failure, ESRI(ψi + ψj ), against the sum of the individual ESRIs, ESRI(ψi)+ESRI(ψj ), for all possible pairs in the crustacean (A) and softdrink supply chain network (B) respectively. The dashed diagonal indicates equality between the ESRI of the simultaneous fail… view at source ↗
Figure 3
Figure 3. Figure 3: Number of pairs with amplification factor larger than α. We show the number of firm pairs whose amplification factor, α, exceeds a given threshold. The blue curve represents all firm pairs in the crustacean subnetwork, where amplification factors decay rapidly and reach a maximum value of α = 15. The green curve represents all firm pairs in the softdrink network. Here, amplification factors decay more slow… view at source ↗
Figure 4
Figure 4. Figure 4: Schematic network configurations leading to strong nonlinear amplification of systemic risk. We show three representative network modes observed among firm pairs with large systemic risk. Color denotes the industry classification of a node, and its size denotes the initial market share. Red links denote buyer-supplier relationships that can completely disable the buyer if the supplier fails and cannot be r… view at source ↗
Figure 5
Figure 5. Figure 5: Sensitivity analysis of the sampling method We show scatter plots of the ESRI of the simultaneous failure of identified firm sets against the sum of the ESRIs of the constituent firms. In each plot, we vary one of the parameters of the sampling scheme used to find the sets of firms, and unless otherwise noted, we use 100,000 sets of 500 firms each, with thresholds for identification chosen at θ1 = 3 and θ2… view at source ↗
Figure 6
Figure 6. Figure 6: ESRI of simultaneous failure against the sum of individual ESRIs for the Softdrink subnetwork We show the value of the pair failure ESRI against the sum of individual ESRI values. Black dots are the results of an exhaustive search of all possible firm pairs. Blue dots represent node pairs identified with our extraction procedure. Note how, for the large ESRI pairs, our extraction method manages to identify… view at source ↗
Figure 7
Figure 7. Figure 7: Treemap of the export basket of Ecuador for the year 2015 according to the Harvard Growth Lab[37]. The distribution of exports of Ecuador shows that bananas and plantains are the second largest export with a volume of 2.94 billion dollars. The export of bananas is second only to the export of crude petroleum, with a volume of 5.75 billion dollars. The export of crustaceans and prepared or preserved fish is… view at source ↗
Figure 8
Figure 8. Figure 8: ESRI values for single firm failures for the crustacean, softdrink, and banana subnetworks as well as the full supply chain network of Ecuador The ESRI value for single firm failures, ordered by the rank of each firm, i.e., the most destructive firm is at position 1. Not the log-scale of the x-axis. The left column shows the crustacean supply chain network and the banana supply chain network, with flat pro… view at source ↗
Figure 9
Figure 9. Figure 9: Histogram of the industry distribution for the four supply chain networks. We present a histogram of the industry distribution at the section level according to the ISIC classification scheme. We present the relative number of firms in sections A to V for the full supply chain network of 2015 (black), for the softdrink supply chain network (green), the crustacean supply chain network (blue), and the banana… view at source ↗
Figure 10
Figure 10. Figure 10: Scatter plot of the ESRI for simultaneous failure of firm pairs against the sum of the individual firms’ ESRI The ESRI of simultaneous failure of pairs of firms is plotted against the sum of ESRIs for the firms in the set for the banana subnetwork. Similar to the crustacean subnetwork, no single firm has an ESRI close to 1, and most pairs are close to the identity line, where the sum of their individual E… view at source ↗
Figure 11
Figure 11. Figure 11: Scatter plots of the ESRI for simultaneous firm failures against the sum of individual firms’ ESRI The ESRI of simultaneous failure of small sets of firms is plotted against the sum of ESRIs for the firms in the sets for the full supply chain network of Ecuador in 2010. We also show the identity line as a visual aid. The colored symbols are small sets of firms identified with our extraction procedure. We … view at source ↗
Figure 12
Figure 12. Figure 12: Number of firm pairs with ESRI(ψi + ψj ) > 0.8 against the total number of pairs of firms with ESRI(ψi) < 0.1. For each supply chain network we present the total number of pairs with large systemic risk, ESRI(ψi + ψj ) > 0.8 against the total number of pairs in that network in a linear-log plot. The banana (brown) and crustacean supply chain network (blue) contain no firm pair with large systemic risk. Th… view at source ↗
Figure 13
Figure 13. Figure 13: Survival curves of the amplification factor of simultaneous node failure for pairs with ESRI > 0.01. Shown are the number of firm pairs with an amplification factor larger than a specific value, given that they reduce the network output by at least 1%. The blue line shows the results for all pairs in the crustacean subnetwork. Of these impactful node pairs, the largest amplification factor is 5.5. The gre… view at source ↗
Figure 14
Figure 14. Figure 14: Network visualizations of example pairs for the three identified modes. Panels A – C show an example pair of the first mode as well as their tier one and two customers. The additional node between the pair and the rest of the network is the "plateau-firm" that is supplied by the smaller firm. Color denotes the size of the production shock, xi , for each firm, and size is proportional to market share. Red … view at source ↗
Figure 15
Figure 15. Figure 15: Scatter plot of the DebtRank losses from simultaneous pairwise failure plotted against the sum of individual DebtRank. With a statistical reconstruction of an interbank network with 19 banks, see [38], we simulate the DebtRank algorithm for each individual bank and for all pairs of banks. In panel A we show the total equity losses in the financial system due to the simultaneous pairwise failure of two ban… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 15 canonical work pages

  1. [1]

    Dynamically induced cascading failures in power grids.Nature Communications, 9(1):1975, May 2018

    Benjamin Schäfer, Dirk Witthaut, Marc Timme, and Vito Latora. Dynamically induced cascading failures in power grids.Nature Communications, 9(1):1975, May 2018. ISSN 2041-1723. doi: 10.1038/s41467-018-04287-5. URL https://www.nature.com/articles/ s41467-018-04287-5

    work page doi:10.1038/s41467-018-04287-5 1975

  2. [2]

    Eugene Stanley, and Shlomo Havlin

    Limiao Zhang, Guanwen Zeng, Daqing Li, Hai-Jun Huang, H. Eugene Stanley, and Shlomo Havlin. Scale-free resilience of real traffic jams.Proceedings of the National Academy of Sciences, 116(18):8673–8678, April 2019. doi: 10.1073/pnas.1814982116. URLhttps://www.pnas.org/ doi/full/10.1073/pnas.1814982116

    doi:10.1073/pnas.1814982116 2019

  3. [3]

    Landslides, forest fires, and earthquakes: examples of self-organized critical behavior.Physica A: Statistical Mechanics and its Applications, 340(4):580–589, September 2004

    Donald L Turcotte and Bruce D Malamud. Landslides, forest fires, and earthquakes: examples of self-organized critical behavior.Physica A: Statistical Mechanics and its Applications, 340(4):580–589, September 2004. ISSN 0378-4371. doi: 10.1016/j.physa.2004.05.009. URL https://www.sciencedirect.com/science/article/pii/S0378437104005667. 32

    work page doi:10.1016/j.physa.2004.05.009 2004

  4. [4]

    Contagion Flow through Banking Networks

    Michael Boss, Martin Summer, and Stefan Thurner. Contagion Flow through Banking Networks. In Marian Bubak, Geert Dick van Albada, Peter M. A. Sloot, and Jack Dongarra, editors, Computational Science - ICCS 2004, Lecture Notes in Computer Science, pages 1070–1077, Berlin, Heidelberg, 2004. Springer. ISBN 9783540246886. doi: 10.1007/978-3-540-24688-6_138

    work page doi:10.1007/978-3-540-24688-6_138 2004

  5. [5]

    Network topology of the interbank market.Quantitative Finance, 4(6):677–684, December 2004

    Michael Boss, Helmut Elsinger, Martin Summer, and Stefan Thurner 4. Network topology of the interbank market.Quantitative Finance, 4(6):677–684, December 2004. ISSN 1469-7688. doi: 10.1080/14697680400020325. URL https://www.tandfonline.com/doi/abs/10.1080/ 14697680400020325

    work page doi:10.1080/14697680400020325 2004

  6. [6]

    Debtrank: Too central to fail? financial networks, the fed and systemic risk.Scientific Reports, 2(1):541, August 2012

    Stefano Battiston, Michelangelo Puliga, Rahul Kaushik, Paolo Tasca, and Guido Caldarelli. Debtrank: Too central to fail? financial networks, the fed and systemic risk.Scientific Reports, 2(1):541, August 2012. ISSN 2045-2322. doi: 10.1038/srep00541. URL https: //www.nature.com/articles/srep00541

    doi:10.1038/srep00541 2012

  7. [7]

    Doyne Farmer, and John Geanakoplos

    Stefan Thurner, J. Doyne Farmer, and John Geanakoplos. Leverage causes fat tails and clustered volatility.Quantitative Finance, 12(5):695–707, 2012. doi: 10.1080/14697688.2012.674301. URL https://doi.org/10.1080/14697688.2012.674301

    arXiv 2012

  8. [8]

    Contagion in financial networks.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, August 2010

    Prasanna Gai and Sujit Kapadia. Contagion in financial networks.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, August 2010. doi: 10.1098/rspa. 2009.0410. URLhttps://royalsocietypublishing.org/doi/10.1098/rspa.2009.0410

    arXiv 2010

  9. [9]

    Debtrank: A microscopic foundation for shock propagation.PLOS ONE, 10(6):e0130406, June 2015

    Marco Bardoscia, Stefano Battiston, Fabio Caccioli, and Guido Caldarelli. Debtrank: A microscopic foundation for shock propagation.PLOS ONE, 10(6):e0130406, June 2015. ISSN 1932-6203. doi: 10.1371/journal.pone.0130406. URL https://journals.plos.org/plosone/ article?id=10.1371/journal.pone.0130406

    work page doi:10.1371/journal.pone.0130406 2015

  10. [10]

    Network valuation in financial systems.Mathematical Finance, 30(4):1181–1204, 2020

    Paolo Barucca, Marco Bardoscia, Fabio Caccioli, Marco D’Errico, Gabriele Visentin, Guido Caldarelli, and Stefano Battiston. Network valuation in financial systems.Mathematical Finance, 30(4):1181–1204, 2020. ISSN 1467-9965. doi: 10.1111/mafi.12272. URL https: //onlinelibrary.wiley.com/doi/abs/10.1111/mafi.12272

    work page doi:10.1111/mafi.12272 2020

  11. [11]

    The multi-layer network nature of systemic risk and its implications for the costs of financial crises.Journal of Financial Stability, 20:70–81, October 2015

    Sebastian Poledna, José Luis Molina-Borboa, Serafín Martínez-Jaramillo, Marco van der Leij, and Stefan Thurner. The multi-layer network nature of systemic risk and its implications for the costs of financial crises.Journal of Financial Stability, 20:70–81, October 2015. ISSN 1572-3089. doi: 10.1016/j.jfs.2015.08.001. URL https://www.sciencedirect.com/scie...

    work page doi:10.1016/j.jfs.2015.08.001 2015

  12. [13]

    Systemic risk-efficient asset allocations: Minimization of systemic risk as a network optimization problem.Journal of Financial Stability, 52:100809, February 2021

    Anton Pichler, Sebastian Poledna, and Stefan Thurner. Systemic risk-efficient asset allocations: Minimization of systemic risk as a network optimization problem.Journal of Financial Stability, 52:100809, February 2021. ISSN 1572-3089. doi: 10.1016/j.jfs.2020.100809. URL https://www.sciencedirect.com/science/article/pii/S1572308920301121

    arXiv 2021

  13. [14]

    Fire Sales, Indirect Contagion and Systemic Stress Testing

    Rama Cont and Eric Schaanning. Fire Sales, Indirect Contagion and Systemic Stress Testing. Technical Report 2955646, Norges Bank working Paper, Rochester, NY, March 2017. URL https://papers.ssrn.com/abstract=2955646. 33

    2017

  14. [15]

    DebtRank-transparency: Controlling systemic risk in financial networks.Scientific Reports, 3(1):1888, May 2013

    Stefan Thurner and Sebastian Poledna. DebtRank-transparency: Controlling systemic risk in financial networks.Scientific Reports, 3(1):1888, May 2013. ISSN 2045-2322. doi: 10.1038/ srep01888. URLhttps://www.nature.com/articles/srep01888

    2013

  15. [16]

    Maria del Rio-Chanona, François Lafond, and J

    Anton Pichler, Marco Pangallo, R. Maria del Rio-Chanona, François Lafond, and J. Doyne Farmer. In and out of Lockdown: Propagation of Supply and Demand Shocks in a Dynamic Input-Output Model. Technical Report 3788494, Rochester, NY, February 2021. URLhttps: //papers.ssrn.com/abstract=3788494

    2021

  16. [17]

    What is the minimal systemic risk in financial exposure networks?Journal of Economic Dynamics and Control, 116:103900, July

    Christian Diem, Anton Pichler, and Stefan Thurner. What is the minimal systemic risk in financial exposure networks?Journal of Economic Dynamics and Control, 116:103900, July

  17. [18]

    Pathways towards instability in financial networks.Nature Communications, 8(1):14416, February 2017

    Marco Bardoscia, Stefano Battiston, Fabio Caccioli, and Guido Caldarelli. Pathways towards instability in financial networks.Nature Communications, 8(1):14416, February 2017. ISSN 2041-

    2017

  18. [19]

    Kleinnijenhuis, Thom Wetzer, and J

    Garbrand Wiersema, Alissa M. Kleinnijenhuis, Thom Wetzer, and J. Doyne Farmer. Scenario- free analysis of financial stability with interacting contagion channels.Journal of Banking & Finance, 146:106684, January 2023. ISSN 0378-4266. doi: 10.1016/j.jbankfin.2022.106684. URL https://www.sciencedirect.com/science/article/pii/S0378426622002643

    arXiv 2023

  19. [20]

    The physics of financial networks.Nature Reviews Physics, 3(7):490–507, July 2021

    Marco Bardoscia, Paolo Barucca, Stefano Battiston, Fabio Caccioli, Giulio Cimini, Diego Garlaschelli, Fabio Saracco, Tiziano Squartini, and Guido Caldarelli. The physics of financial networks.Nature Reviews Physics, 3(7):490–507, July 2021. ISSN 2522-5820. doi: 10.1038/ s42254-021-00322-5. URLhttps://www.nature.com/articles/s42254-021-00322-5

    2021

  20. [21]

    A Complex Systems Perspective on Macroprudential Regulation

    Stefan Thurner. A Complex Systems Perspective on Macroprudential Regulation. In Alissa M. Kleinnijenhuis, J. Doyne Farmer, Thom Wetzer, and Til Schuermann, ed- itors,Handbook of Financial Stress Testing, pages 593–634. Cambridge University Press, Cambridge, 2022. ISBN 9781108830737. doi: 10.1017/9781108903011.034. URL https://www.cambridge.org/core/books/...

    work page doi:10.1017/9781108903011.034 2022

  21. [22]

    Input Specificity and the Propagation of Idiosyncratic Shocks in Production Networks.The Quarterly Journal of Economics, 131(3):1543–1592, August 2016

    Jean-Noël Barrot and Julien Sauvagnat. Input Specificity and the Propagation of Idiosyncratic Shocks in Production Networks.The Quarterly Journal of Economics, 131(3):1543–1592, August 2016. ISSN 0033-5533. doi: 10.1093/qje/qjw018. URLhttps://doi.org/10.1093/ qje/qjw018

    doi:10.1093/qje/qjw018 2016

  22. [23]

    Estimating the loss of economic predictability from aggregating firm-level production networks.PNAS Nexus, 3(3):pgae064, March 2024

    Christian Diem, András Borsos, Tobias Reisch, János Kertész, and Stefan Thurner. Estimating the loss of economic predictability from aggregating firm-level production networks.PNAS Nexus, 3(3):pgae064, March 2024. ISSN 2752-6542. doi: 10.1093/pnasnexus/pgae064. URL https://doi.org/10.1093/pnasnexus/pgae064

    work page doi:10.1093/pnasnexus/pgae064 2024

  23. [24]

    An Adaptive Regional Input-Output Model and its Application to the Assessment of the Economic Cost of Katrina.Risk Analysis, 28(3):779–799, 2008

    Stéphane Hallegatte. An Adaptive Regional Input-Output Model and its Application to the Assessment of the Economic Cost of Katrina.Risk Analysis, 28(3):779–799, 2008. ISSN 1539-6924. doi: 10.1111/j.1539-6924.2008.01046.x. URLhttps://onlinelibrary.wiley.com/ doi/abs/10.1111/j.1539-6924.2008.01046.x. 34

    arXiv 2008

  24. [25]

    Firm-level propagation of shocks through supply-chain networks.Nature Sustainability, 2(9):841–847, September 2019

    Hiroyasu Inoue and Yasuyuki Todo. Firm-level propagation of shocks through supply-chain networks.Nature Sustainability, 2(9):841–847, September 2019. ISSN 2398-9629. doi: 10.1038/ s41893-019-0351-x. URLhttps://www.nature.com/articles/s41893-019-0351-x

    2019

  25. [26]

    Supply Chain Disruptions: Evidence from the Great East Japan Earthquake.The Quarterly Journal of Economics, 136(2):1255–1321, May 2021

    Vasco M Carvalho, Makoto Nirei, Yukiko U Saito, and Alireza Tahbaz-Salehi. Supply Chain Disruptions: Evidence from the Great East Japan Earthquake.The Quarterly Journal of Economics, 136(2):1255–1321, May 2021. ISSN 0033-5533. doi: 10.1093/qje/qjaa044. URL https://doi.org/10.1093/qje/qjaa044

    doi:10.1093/qje/qjaa044 2021

  26. [27]

    Firm-level production networks: what do we (really) know.INET Oxford Working Paper, 2023, 2023

    Andrea Bacilieri, András Borsos, Pablo Astudillo-Estevez, and François La- fond. Firm-level production networks: what do we (really) know.INET Oxford Working Paper, 2023, 2023. URL https://www.researchgate.net/ profile/Andrea-Bacilieri/publication/370838043_Firm-level_production_ networks_what_do_we_really_know/links/64652f319533894cac772ccc/ Firm-level-p...

    arXiv 2023

  27. [28]

    Quantifying firm-level economic systemic risk from nation-wide supply networks.Scientific Reports, 12(1): 1–13, 5 2022

    Christian Diem, András Borsos, Tobias Reisch, János Kertész, and Stefan Thurner. Quantifying firm-level economic systemic risk from nation-wide supply networks.Scientific Reports, 12(1): 1–13, 5 2022. ISSN 2045-2322. doi: 10.1038/s41598-022-11522-z. URLhttps://doi.org/10. 1038/s41598-022-11522-z

    work page doi:10.1038/s41598-022-11522-z 2022

  28. [29]

    Systemic risk mitigation in supply chains through network rewiring.Scientific Reports, 16(1):12334, March 2026

    Giacomo Zelbi, Leonardo Niccolò Ialongo, and Stefan Thurner. Systemic risk mitigation in supply chains through network rewiring.Scientific Reports, 16(1):12334, March 2026. ISSN 2045-2322. doi: 10.1038/s41598-026-42549-1. URL https://www.nature.com/articles/ s41598-026-42549-1

    work page doi:10.1038/s41598-026-42549-1 2026

  29. [30]

    Protein mediated networks

    Stuart Kauffman. Protein mediated networks. random chemistry.Berichte der Bunsen- gesellschaft fuer physikalische Chemie, 98(9):1142–1147, 1994. doi: https://doi.org/10. 1002/bbpc.19940980915. URL https://onlinelibrary.wiley.com/doi/abs/10.1002/bbpc. 19940980915

    work page doi:10.1002/bbpc 1994

  30. [31]

    random chemistry

    Margaret J. Eppstein and Paul D. H. Hines. A “random chemistry” algorithm for identifying collections of multiple contingencies that initiate cascading failure.IEEE Transactions on Power Systems, 27(3):1698–1705, 2012. doi: 10.1109/TPWRS.2012.2183624

    arXiv 2012

  31. [32]

    Aaron Clauset, Cosma Rohilla Shalizi, and Mark E. J. Newman. Power-law distributions in empirical data.SIAM Rev., 51(4):661–703, 2009. doi: 10.1137/070710111

    doi:10.1137/070710111 2009

  32. [33]

    Supply chain network rewiring dynamics at the firm level.PNAS Nexus, 5(4):pgag091, April 2026

    Tobias Reisch, András Borsos, and Stefan Thurner. Supply chain network rewiring dynamics at the firm level.PNAS Nexus, 5(4):pgag091, April 2026. ISSN 2752-6542. doi: 10.1093/ pnasnexus/pgag091. URLhttps://doi.org/10.1093/pnasnexus/pgag091

    work page doi:10.1093/pnasnexus/pgag091 2026

  33. [34]

    In the eye of the storm: Heterogeneous effects of extreme climate events on supply chains.Journal of Environmental Management, 397:128244, January 2026

    Thiago Christiano Silva, Paulo Victor Berri Wilhelm, Solange Maria Guerra, and Benjamin Mi- randa Tabak. In the eye of the storm: Heterogeneous effects of extreme climate events on supply chains.Journal of Environmental Management, 397:128244, January 2026. ISSN 0301-4797. doi: 10.1016/j.jenvman.2025.128244. URL https://www.sciencedirect.com/science/artic...

    arXiv 2026

  34. [35]

    Eleven data-set: A labeled set of descriptions of goods captured from brazilian electronic invoices

    Vinícius Di Oliveira, Li Weigang, and Geraldo Pereira Rocha Filho. Eleven data-set: A labeled set of descriptions of goods captured from brazilian electronic invoices. InProceedings of the 18th International Conference on Web Information Systems and Technologies - WEBIST, pages 257–264. INSTICC, SciTePress, 2022. ISBN 978-989-758-613-2. doi: 10.5220/00115...

    work page doi:10.5220/0011524800003318 2022

  35. [36]

    Ai-generated production networks: Measurement and applications to global trade

    Thiemo Fetzer, Peter John Lambert, Bennet Feld, and Prashant Garg. Ai-generated production networks: Measurement and applications to global trade. CESifo Working Paper 11497, Munich,

  36. [37]

    The atlas of economic complexity.https://atlas.hks

    Growth Lab at Harvard University. The atlas of economic complexity.https://atlas.hks. harvard.edu, February 2026. URLhttps://atlas.hks.harvard.edu. Web application

    2026

  37. [38]

    URLhttp://arxiv.org/abs/2502.17044

    A data-driven econo-financial stress-testing framework to estimate the effect of supply chain networks on financial systemic risk, February 2025. URLhttp://arxiv.org/abs/2502.17044. arXiv:2502.17044 [q-fin]. 36

    Pith/arXiv arXiv 2025

  38. [1723]

    URLhttps://www.nature.com/articles/ncomms14416

    doi: 10.1038/ncomms14416. URLhttps://www.nature.com/articles/ncomms14416

    work page doi:10.1038/ncomms14416

  39. [2020]

    doi: 10.1016/j.jedc.2020.103900

    ISSN 0165-1889. doi: 10.1016/j.jedc.2020.103900. URLhttps://www.sciencedirect. com/science/article/pii/S0165188920300683

    arXiv 2020

  40. [2024]

    URLhttps://hdl.handle.net/10419/308393