Financial Network Contagion: How Interconnectedness Amplifies Systemic Risk

What happens when a single bank fails in a densely interconnected financial system? Does the web of relationships absorb the blow, or does it transmit the damage everywhere at once? The answer, it turns out, depends on how large the initial shock is relative to the capacity of the network to absorb it, and the relationship between shock size and system fragility is sharply nonlinear.

Acemoglu, Ozdaglar, and Tahbaz-Salehi (2015) address this question with a formal model of systemic risk in financial networks, published in the American Economic Review. Their central contribution is identifying a phase transition: the same network structure that provides resilience against small shocks becomes the mechanism of catastrophic contagion when shocks exceed a critical threshold.

The Dual Nature of Interconnectedness

Prior to this paper, two competing views dominated the debate on financial interconnectedness. One school, following Allen and Gale (2000), argued that a more complete network of interbank claims improves stability by allowing losses to be shared across many counterparties. If Bank A fails and owes money to ten other banks, each absorbs only a tenth of the loss. The other school observed that connections create channels for contagion, enabling distress to travel from a failing institution to its creditors, their creditors, and so on, in a cascade.

Acemoglu, Ozdaglar, and Tahbaz-Salehi reconcile these views by showing that both are correct, but in different regimes. The network topology does not have a fixed effect on stability. Its role reverses depending on the magnitude of the shock hitting the system.

Small Shocks: Connectivity as Insurance

In the model, financial institutions hold bilateral claims and obligations. When one bank suffers a negative shock, it may impose partial losses on its creditors. For shocks below the critical threshold, a densely connected network functions as mutual insurance. Losses distribute across many counterparties, each absorbing a small fraction. No single creditor suffers enough damage to trigger its own default, and the cascade dies out after the first round.

This is the scenario that underlies pre-crisis thinking about financial interconnectedness. Risk sharing through diversified interbank exposures was viewed as an unambiguous stabilizer.

Large Shocks: Connectivity as Contagion

The paper's most consequential finding concerns shocks above the critical threshold. When the initial loss is large enough that even a diversified share of it impairs the solvency of creditor banks, the network's connectivity transforms from a stabilizing force into an accelerant.

Each creditor bank that absorbs its fraction of the loss now faces its own solvency pressure. If that pressure is severe enough, it defaults on its own obligations, passing losses to the next layer of counterparties. In a densely connected network, this cascade has more pathways to travel and reaches more institutions at each step. The loss does not dilute as it propagates; it amplifies.

The result is a discontinuity in the relationship between shock size and total losses. Below the threshold, system-wide losses scale gradually with the initial shock. Above it, the same incremental increase in shock size triggers a disproportionate jump in aggregate losses as the cascade engulfs institutions that were not directly exposed to the original disturbance. This phase transition is the paper's signature result.

Topology Matters: Which Structures Are Most Fragile?

Not all network structures are equally vulnerable. Acemoglu, Ozdaglar, and Tahbaz-Salehi compare several canonical topologies:

The complete network offers maximum diversification for small shocks but maximum contagion for large ones. The ring network localizes cascades but provides less diversification. Core-periphery structures, which empirical research by Craig and von Peter (2014) identifies as the dominant topology in real interbank markets, inherit the worst of both: the core banks are well-diversified against modest disturbances, but a shock large enough to bring down a core bank spreads rapidly to every periphery institution connected to it.

Elliott, Golub, and Jackson (2014) extend this analysis by incorporating cross-holdings of assets and equity, showing that indirect exposures through common portfolio positions create additional contagion channels beyond direct bilateral claims.

Empirical Relevance: The 2008 Template

The model provides a precise lens for interpreting the 2008 financial crisis. Pre-crisis, the interbank network had evolved toward a concentrated core-periphery structure, with a small number of globally systemic institutions (Lehman Brothers, AIG, Bear Stearns) serving as highly connected hubs. The subprime mortgage losses that originated the crisis were, in absolute terms, modest relative to the total assets of the global banking system. But those losses were concentrated in institutions at the core of the network, and the shock exceeded the threshold at which the network's connectivity switched from stabilizing to destabilizing.

As correlation patterns broke down during the crisis, the standard portfolio diversification assumptions failed precisely because the contagion mechanism was operating through network channels that correlations do not capture. The funding liquidity spirals documented by Brunnermeier and Pedersen (2009) represent a complementary amplification mechanism: as network contagion impaired bank solvency, funding markets seized up, adding a liquidity dimension to the solvency cascade.

Regulatory Implications and Portfolio Consequences

Glasserman and Young (2016) survey the broader literature on financial network contagion and note that the phase-transition insight has directly influenced post-crisis regulation. Capital surcharges for systemically important institutions, central clearing mandates for derivatives, and network-based stress testing all reflect the recognition that interconnectedness is not a simple good or bad, but a conditional property that depends on the magnitude of shocks the system must absorb.

For portfolio construction, the practical implication is that standard risk models underestimate tail risk in the financial sector because they treat bank failures as independent events when they are, in reality, connected through the network. Concentrated exposure to financial-sector equities or credit carries a latent contagion risk that only materializes during large shocks, precisely when traditional hedges also fail.

Where the Model Falls Short

The framework assumes static network topology, but banks restructure exposures in response to emerging distress, sometimes amplifying rather than mitigating cascades. Battiston et al. (2012) show that endogenous network formation, where banks choose counterparties strategically, can produce structures even more fragile than fixed-topology models predict.

The model also abstracts from behavioral dynamics: panic, uncertainty about counterparty exposures, and strategic liquidity hoarding all played roles in 2008 that mechanical loss transmission cannot fully capture. These limitations suggest the phase-transition threshold in real networks may be lower than predicted, making the system more fragile than the formal analysis indicates.