📈 economics

Consensus backbones for directed economic and financial networks: sovereign risk spillovers and the emerging-market core

This paper introduces a consensus backbone method for directed economic networks to reveal a sparse, hierarchical spillover structure, demonstrating through a 36-country analysis that sovereign risk transmission is dominated by concentrated bilateral channels among emerging markets while advanced economies exhibit diffuse, symmetric spillovers.

Original authors: Jorge Antonio Chan-Lau

Published 2026-10-15

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Jorge Antonio Chan-Lau

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the vast, invisible web of the global economy, nations are constantly connected. Money flows across borders, trade links supply chains, and financial fears travel instantly from one capital to another. Economists have long tried to map these connections to understand how a shock in one place might ripple out to cause trouble elsewhere. Traditionally, they have treated these relationships like a simple web of strings, where every country is tied to every other country with a line representing the strength of their connection. While this captures the sheer volume of interaction, it often creates a tangled mess that is impossible to read. It shows that everyone is connected, but it fails to show who is pulling the strings, who is being pulled, and which specific paths a crisis is most likely to take. To make sense of this complexity, researchers need a way to strip away the noise and reveal the underlying skeleton of the system—the few critical pathways that actually carry the most significant influence.

A new study by Jorge Antonio Chan-Lau offers a fresh way to see this hidden structure. Instead of looking at the entire tangled web, the researcher developed a method to find the "consensus backbone" of the global financial system. Imagine taking a dense forest of trees and trying to find the main trails that hikers actually use. Most paths are overgrown and rarely traveled, but a few clear routes connect the major camps. This study applies that logic to the financial world, using a sophisticated mathematical approach to filter out the weak, random connections and keep only the strongest, most directional links. The method focuses on "spillovers," which are the ways that financial stress or volatility in one country spreads to another. By analyzing data from 36 different economies, the study builds a map that shows not just who is connected, but the specific direction of the flow: who is sending the shock and who is receiving it.

The researchers started with a massive amount of data regarding the risk of default for 36 countries, ranging from the United States and Germany to emerging markets like Brazil, Turkey, and South Africa. They used a standard statistical tool to measure how much the financial uncertainty in one country contributed to the uncertainty in another. This initial calculation produced a dense network with over 1,200 connections, a tangle so thick that it was difficult to see any clear pattern. To simplify this, the team applied a new filtering technique. They looked for the most dominant paths of influence, creating a series of "trees" where one country sits at the top and branches out to others. By combining several of these trees, they created a single, unified map that represents the most reliable routes of financial transmission. This process reduced the network from 1,200 connections down to just 27, a tiny fraction of the original data, yet one that preserved the essential structure of how risk moves.

The most striking discovery from this simplified map was a complete absence of the world's wealthiest nations. In the final backbone, the United States, Germany, Japan, the United Kingdom, and all other advanced economies disappeared entirely. They did not appear as the leaders sending out shocks, nor as the primary receivers. Instead, the entire active structure of the map was made up of emerging markets and commodity-linked countries. The map showed a clear hierarchy where nations like South Africa, Colombia, and Peru acted as the primary sources of financial influence, sending stress along specific channels to other countries. Meanwhile, nations like South Korea and Vietnam appeared only as the final destinations, receiving shocks but not sending them further. The advanced economies, despite their massive size and deep integration into global finance, were so evenly connected to everyone else that no single country stood out as the dominant source of a specific, directional shock. Their influence was too diffuse to show up on this particular map.

This finding challenges the common assumption that the world's largest economies are always the primary drivers of financial contagion. The study suggests that while advanced economies are deeply intertwined, their risk spreads out in all directions almost equally, like a gentle fog rather than a focused beam. In contrast, the financial risks in emerging markets travel along sharp, specific routes. A shock in one emerging market tends to hit a specific neighbor with great force before moving on, creating a clear chain of transmission. The research indicates that when we look for the specific pathways where financial stress is most likely to travel, we should be watching the emerging markets, particularly the links between Latin America and Asia, rather than the traditional hubs of the developed world.

The study also identified specific roles for the countries that did appear on the map. Some acted as "transmitters," starting the chain of events. Others served as "bridges," acting as essential middlemen that connected different parts of the network. If these bridge countries were removed from the map, the flow of information would break, and the network would fall apart. For example, several Asian nations were found to be crucial intermediaries, channeling financial stress from Latin American roots to other parts of the emerging world. This highlights a structural reality where these smaller economies play a disproportionately large role in how global risk is organized and transmitted.

The researchers were careful to note that this result depends on how the data is measured. If they had simply looked at the raw size of the connections without adjusting for the different scales of the economies, the advanced nations would have appeared on the map simply because they are so large. However, by adjusting the data to treat every country's influence fairly, regardless of its size, the true structure of the risk transmission emerged. This suggests that the dominance of advanced economies in global finance is a matter of volume, while the dominance of emerging markets in risk transmission is a matter of structure. The study does not claim that advanced economies are unimportant; rather, it suggests that their importance is spread so thinly across the system that it does not form the clear, directional lines that this specific method is designed to find.

Ultimately, this work provides a new lens for watching the global economy. It moves beyond the idea that bigger countries are always the most dangerous or influential. Instead, it reveals a world where the most critical pathways of financial risk are found in the connections between developing nations. By focusing on these specific, directional links, policymakers and analysts can better understand where a crisis might start and how it is likely to travel. The study offers a clearer, more focused view of the financial world, showing that sometimes the most important stories are not told by the giants, but by the specific, concentrated paths that connect the rest of the world.

=== SUMMARY ===

Technical Summary: Consensus Backbones for Directed Economic and Financial Networks

1. Problem Statement

Economic and financial systems are characterized by dense, high-dimensional networks of interdependence (e.g., cross-border exposures, trade flows, and risk spillovers). While methods like the Minimum Spanning Tree (MST) have been widely used to extract sparse, interpretable "backbones" from these networks, they operate on undirected graphs. This approach discards the inherent asymmetry of many economic relationships, such as creditor-debtor hierarchies or directional risk spillovers.

Existing directed analogues, such as single arborescences (directed spanning trees), capture hierarchical propagation but are often overly restrictive, representing only a single path structure and ignoring redundant channels through which shocks can travel. Furthermore, standard variance decomposition networks (e.g., Diebold–Yilmaz) are fully connected and dense, making it difficult to distinguish dominant spillover paths from noise without arbitrary thresholding. The paper addresses the need for a method that can extract a sparse, directed, and interpretable structure from dense economic networks while preserving redundancy (alternative spillover routes) and directionality.

2. Methodology

The paper proposes the Consensus Backbone, a framework that aggregates multiple maximum arborescences into a unified, sparse subgraph. The methodology proceeds through the following steps:

2.1 Network Construction and Normalization

The input is a weighted directed graph $G=(V, E)$ , where edge weights $w_{ij}$ represent the influence of node $i$ on node $j$ (e.g., risk spillovers). To address scale effects where high-magnitude weights might overshadow structural importance, the adjacency matrix is normalized. The paper evaluates three modes:

Row-stochastic: Normalizes outgoing weights (emission profiles).
Column-stochastic: Normalizes incoming weights (receiver vulnerability).
Sinkhorn (Doubly Stochastic): Iteratively balances both inflows and outflows. The paper identifies this as the default for exploratory analysis as it isolates the latent spillover structure by controlling for node-specific scale effects.

2.2 Maximum Arborescence and Branching Packs

Using the Chu–Liu/Edmonds algorithm, the method identifies a maximum arborescence rooted at a specific node $r$ . This is a directed spanning tree that maximizes the sum of edge weights while ensuring exactly one path exists from the root to every other reachable node.

To capture alternative routes, the method constructs a branching pack: a collection of $k$ edge-disjoint arborescences rooted at the same node. This is achieved by iteratively solving the maximum arborescence problem, penalizing previously selected edges (via a multiplicative factor $\gamma$ ) to encourage diversity without rendering the graph infeasible.

2.3 Consensus Backbone Construction

The Consensus Backbone is defined as the union of branching packs computed across multiple root nodes.

Root Selection: Roots are chosen endogenously as the top- $k_r$ nodes by out-strength in the normalized matrix.
Edge Aggregation: The final backbone is the union of all arborescences from all roots. Edges appearing in multiple arborescences receive a higher frequency score, indicating structural robustness.
Selection of $k^*$ : The number of arborescences per root ( $k$ ) is determined by a composite diagnostic criterion balancing four metrics: weight ratio, density, reachability, and the preservation of out-strength rankings ( $R^2$ ). The optimal $k^*$ is selected at the point of diminishing returns.

3. Empirical Application

The method is applied to a 36-country sovereign risk spillover network using Credit Default Swap (CDS) data from December 2008 to May 2025.

Data: Weekly changes in the logit of 5-year probability of default (PD) derived from USD-denominated CDS spreads.
Model: A Vector Autoregression (VAR) model (VAR(8)) is estimated to capture dynamic interrelationships.
Network Generation: A Generalized Forecast Error Variance Decomposition (GFEVD) is computed to generate a dense, directed adjacency matrix representing predictive variance contributions.
Specification: The primary analysis uses Sinkhorn normalization and a 52-week forecast horizon.

4. Key Results

4.1 Sparsification Efficiency

The consensus backbone dramatically reduces network complexity. In the Sinkhorn-normalized specification:

The backbone retains only 27 edges (2.1% of the original 1,260 edges).
It captures 3.8% of the total network weight.
Despite low weight coverage, it preserves a coherent, navigable spillover hierarchy, unlike simple thresholding which fragments the network.

4.2 The "Emerging-Market Core"

A striking empirical finding is the complete absence of advanced economies from the consensus backbone under Sinkhorn normalization.

Active Nodes: The entire active backbone consists of 15 emerging market and commodity-linked sovereigns (e.g., South Africa, Colombia, Peru, Turkey, Chile, Brazil, Mexico, and several Asian economies).
Isolated Nodes: All 21 advanced economies in the sample (including the US, UK, Germany, Japan, and France) appear as isolates with zero in-degree and out-degree in the backbone.
Interpretation: This suggests that advanced-economy spillovers are diffuse and nearly symmetric, failing to dominate any single directional path after scale adjustment. In contrast, emerging market spillovers are concentrated along specific bilateral channels, forming a dominant hierarchical structure.

4.3 Structural Roles

The backbone identifies specific structural roles among the active nodes:

Transmitters: Countries with high relative out-strength (e.g., South Africa, Colombia, Peru).
Receivers: Countries acting as terminal nodes (e.g., South Korea, Vietnam).
Bridges: Structural intermediaries (e.g., China, Indonesia, Malaysia, Philippines, Thailand) that connect Latin American root nodes to broader emerging market receivers.

4.4 Robustness

The findings are robust across:

VAR Lag Orders: The backbone structure remains stable for lag orders 4 through 8.
Forecast Horizons: The hierarchy is consistent across horizons of 4 to 52 weeks, with the 26-week and 52-week structures being identical.
Root Selection: The backbone structure saturates once the top two out-strength roots are included.
Normalization: The "dropout" of advanced economies is specific to Sinkhorn normalization; under raw weights, the backbone is denser and includes advanced economies, confirming that their prominence in raw networks is driven by scale rather than concentrated directional channels.

5. Comparison with Existing Methods

The paper benchmarks the consensus backbone against three alternatives:

Undirected MST: The MST forces connectivity across all 36 nodes (including advanced economies) and loses directionality. It answers a different question (bilateral closeness) rather than identifying dominant directional transmission paths.
Simple Thresholding: Selecting the top- $N$ edges by weight retains more aggregate weight (6.3% vs. 3.8%) but results in a fragmented network (23 connected components) with lower reachability. The backbone sacrifices raw weight to preserve a connected, interpretable topology.
Diebold–Yilmaz Net Rankings: While there is a moderate positive correlation ( $\rho = 0.64$ ) between backbone out-strength and standard net spillover rankings, they diverge significantly. Countries with high aggregate net spillovers (e.g., Indonesia, Philippines) may have low backbone centrality if their spillovers are diffuse, whereas countries with concentrated directional channels (e.g., Colombia, Peru) rank higher in the backbone.

6. Significance and Claims

The paper claims to be the first application of directed consensus backbones to economic and financial networks. Its significance lies in three areas:

Methodological Contribution: It bridges three research streams—undirected backbone extraction, arborescence-based network science, and variance decomposition—providing a principled method to extract sparse, directed structures that preserve redundancy (alternative paths) and directionality.
Empirical Insight: The application reveals a fundamental asymmetry in sovereign risk transmission. It challenges the view that advanced economies are the primary hubs of directional spillover, suggesting instead that their influence is diffuse and symmetric, while emerging markets form a concentrated, hierarchical transmission core.
Policy Relevance: The method offers a tool for macro-prudential surveillance that shifts focus from aggregate connectedness (which favors large, integrated economies) to the dominant directed routes of stress propagation. This allows policymakers to identify specific structural transmitters and bridges (particularly in emerging markets) that may be critical for risk containment.

The paper concludes that the consensus backbone is a robust, interpretable tool for summarizing dense dependence networks, though it notes that the results are conditional on the estimated model and normalization choice, and that the "dropout" of advanced economies reflects the specific topology of scale-invariant spillover channels rather than a lack of systemic importance.