Financial Contagion — How Bank Failures Cascade

1. The Interbank Network — Anatomy of Interconnection

Modern banks are not isolated firms. They are deeply embedded in a web of bilateral obligations: loans, deposits, repurchase agreements (repos), derivative contracts, payment systems, and shared exposures to common assets. This web constitutes the interbank network, and it is the channel through which financial distress travels.

In graph-theoretic terms, the interbank network is a directed weighted graph where:

• Nodes represent financial institutions (banks, broker-dealers, insurance companies, money market funds)
• Edges represent bilateral exposures: an edge from A to B means A has a claim on B (B owes A money)
• Edge weight is the notional value of the exposure

Real interbank networks are sparse (most pairs of banks do not have direct bilateral exposures) but concentrated — a small number of large banks act as hubs, connected to many counterparties. This core-periphery structure is empirically robust across countries and time periods. The European Central Bank, the Federal Reserve, and the Bank of England all map and monitor their domestic interbank networks as part of financial stability analysis.

2. The Eisenberg-Noe Clearing Model

The foundational mathematical framework for financial contagion is the Eisenberg-Noe model (2001). Consider a financial system with n banks. Each bank i has:

• e_i: external assets (cash, loans to the real economy, securities)
• p̄_i: total nominal liabilities (the total amount bank i owes)
• π_ij: the fraction of bank i's total liabilities owed to bank j (so Σⱼ π_ij ≤ 1, with the rest owed to external creditors)

Define the relative liability matrix L where L_ij = π_ij · p̄_i is the nominal amount bank i owes bank j. The clearing payment vector p* = (p₁*, ..., pₙ*) is the vector of actual payments each bank makes, subject to:

Eisenberg and Noe proved that a unique largest fixed point p* exists (under mild conditions), and it can be computed by iterated application of Φ starting from p = p̄ (everyone pays in full). This fixed point represents the equilibrium clearing of the financial system: who pays what to whom after accounting for all cascading defaults.

Default Condition

Bank i defaults (p_i* < p̄_i) if and only if its total assets — external assets plus incoming payments — fall short of its total obligations:

3. Cascade Mechanism — How Failures Propagate

Financial contagion spreads through two distinct channels, both captured partly by the Eisenberg-Noe framework:

Direct Contagion (Counterparty Credit Risk)

When bank A defaults, it fails to repay its loan to bank B. If this shortfall is large enough relative to B's capital buffer, B also becomes insolvent — even if B's own external assets are healthy. B's default then propagates to banks C, D, etc. This is the direct, balance-sheet channel of contagion.

Indirect Contagion (Fire Sales and Asset Price Spirals)

A more insidious channel operates through asset prices. When distressed banks must sell assets quickly (fire sales) to raise liquidity, the increased supply depresses asset prices. Other banks holding the same assets see their balance sheets deteriorate, potentially triggering further sales — a price-mediated spiral. This channel operates even without any direct bilateral exposure between banks.

The Brunnermeier-Pedersen model (2009) formalizes this as a feedback between funding liquidity (ability to borrow) and market liquidity (ability to sell assets without moving prices). When both deteriorate simultaneously — a "liquidity spiral" — the system can collapse far faster than direct contagion alone would suggest.

4. Network Topology and Systemic Fragility

Not all network structures are equally fragile. Network science provides precise predictions about which topologies amplify or dampen financial contagion.

Complete versus Ring Networks

In a complete network (every bank lends to every other bank), each bilateral exposure is small (total lending divided among many counterparties). A single bank's default causes small losses to many banks — losses that may be individually below each bank's capital buffer, so no cascade occurs. Complete networks are robust to idiosyncratic shocks but fragile to large systemic shocks.

In a ring network (each bank lends to only one other), all lending is concentrated in one bilateral link. A single default causes a large loss to one counterparty, which is more likely to trigger a cascade. Ring networks are fragile to idiosyncratic shocks.

Core-Periphery and Scale-Free Networks

The real interbank network has a core-periphery structure: a small dense core of highly interconnected banks surrounded by a periphery of less connected institutions. This structure is robust to peripheral failures (periphery banks have few connections) but extremely fragile to core failures — a property shared by scale-free networks and described by Albert-Barabási as "robust yet fragile."

5. Too-Big-to-Fail and Systemically Important Banks

The too-big-to-fail (TBTF) problem arises when a financial institution is so large, interconnected, or operationally critical that its failure would cause unacceptable damage to the broader economy. Governments face a perverse incentive: they must rescue such institutions even though rescues create moral hazard (if you know you'll be rescued, you take more risk).

Since 2011, the Financial Stability Board (FSB) annually designates Global Systemically Important Banks (G-SIBs) — currently around 30 institutions including JPMorgan Chase, Bank of America, ICBC, HSBC, and BNP Paribas. Designation is based on five criteria:

• Size: total exposure measure
• Interconnectedness: intra-financial system assets/liabilities
• Substitutability: how easily the bank's functions can be replaced
• Complexity: derivatives notional, OTC derivatives, cross-jurisdictional claims
• Cross-jurisdictional activity: global reach

G-SIBs face additional capital surcharges of 1–3.5% of risk-weighted assets above the Basel III minimum, precisely calibrated to reflect their systemic importance score.

6. The 2008 Crisis — A Network Perspective

The 2008 global financial crisis was not caused solely by the fall in US house prices. The primary transmission mechanism was the network structure of mortgage-backed securities (MBS) and collateralized debt obligations (CDOs), which distributed mortgage risk through the global financial system in opaque and concentrated ways.

The Securitization Chain

Mortgage originators sold loans to investment banks, which pooled them into MBS. MBS tranches were then pooled again into CDOs. CDO tranches were sometimes pooled into CDO-squared instruments. This chain created a network of claims where the ultimate risk-bearer was far removed from the original lender, and where the total notional value of derivative claims vastly exceeded the underlying assets:

The Lehman Moment

Lehman Brothers was not the largest firm to fail (Bear Stearns and Washington Mutual were rescued or acquired), but its failure was allowed to proceed without government support. The immediate network effects included: money market funds that held Lehman commercial paper "broke the buck" (net asset value fell below $1); repo markets froze as counterparties refused to lend against any collateral; and interbank lending rates (LIBOR-OIS spread) spiked from 10 basis points to over 350 basis points within days.

7. Stress Testing — Simulating Contagion

Following the 2008 crisis, regulators implemented systematic stress tests — simulations of how the banking system responds to hypothetical adverse scenarios. The US Dodd-Frank Act (2010) mandated annual Comprehensive Capital Analysis and Review (CCAR) tests for banks with assets over $50 billion; the EU conducts biennial stress tests through the European Banking Authority (EBA).

Scenario Design

A stress test scenario specifies:

• Macroeconomic shocks: GDP contraction, unemployment rate, house price decline
• Market shocks: equity price falls, credit spread widening, currency moves
• Operational shocks: loss of funding access, operational losses
• Contagion: counterparty defaults (the network channel)

The 2023 Federal Reserve stress test assumed a 40% decline in commercial real estate prices, a 38% decline in equity prices, and unemployment reaching 10%. Banks must demonstrate they can absorb losses without breaching minimum capital ratios.

Network Contagion in Stress Tests

Modern stress tests increasingly incorporate network contagion. The Bank of England's RAMSI (Risk Assessment Model for Systemic Institutions) simulates balance sheet dynamics across 15+ major UK banks over a 3-year horizon, including fire-sale feedbacks and funding contagion channels. The ECB's STAMP€ model extends similar analysis to 130 euro area banks.

8. Macroprudential Regulation and Capital Buffers

The policy response to systemic risk has shifted from microprudential regulation (ensuring individual bank safety) to macroprudential regulation (ensuring system-wide stability). The key tools are capital buffers that constrain the network's fragility:

Basel III Capital Framework

• Common Equity Tier 1 (CET1): minimum 4.5% of risk-weighted assets — the highest-quality capital (retained earnings, common shares)
• Capital Conservation Buffer: additional 2.5% of RWA, restricting dividends if breached
• Countercyclical Capital Buffer (CCyB): 0–2.5% of RWA, set by national regulators during credit booms to build resilience
• G-SIB surcharge: 1–3.5% based on systemic importance score
• Total Loss-Absorbing Capacity (TLAC): G-SIBs must hold at least 18% of RWA in instruments that can absorb losses in resolution

Central Clearing Counterparties

A major post-crisis reform was mandating that standardized derivatives be cleared through central counterparties (CCPs). A CCP interposes itself between buyer and seller, becoming the counterparty to both. This converts a complex bilateral network into a star network centered on the CCP, reducing bilateral exposures through multilateral netting. However, it also concentrates risk: CCPs themselves become systemically important nodes whose failure would be catastrophic.