Acemoglu–Ozdaglar–Tahbaz-Salehi Systemic Network Phase Transition Model

Acemoglu–Ozdaglar–Tahbaz-Salehi systemic network phase transition model is a network-economics framework in business and economics theory that establishes how the relationship between financial network interconnectedness and systemic stability undergoes a sharp phase transition depending on the magnitude and number of negative balance-sheet shocks hitting the financial system. Formulated by economists Daron Acemoglu, Asuman Ozdaglar, and Alireza Tahbaz-Salehi in their February 2015 American Economic Review article "Systemic Risk and Stability in Financial Networks", the model unifies two previously conflicting paradigms in financial economics: the view of the Allen–Gale financial contagion model (2000) that complete interbank diversification maximizes financial stability, and the view raised after the 2008 global financial crisis that dense financial interconnections amplify systemic collapse. Embedding the Eisenberg–Noe systemic clearing model with costly premature asset liquidation, Acemoglu, Ozdaglar, and Tahbaz-Salehi prove that as long as negative shocks remain below an explicit analytical threshold φ*, a complete diversified network is the least fragile financial architecture; once negative shocks exceed φ*, however, the financial system crosses a phase transition where complete networks become the most fragile architecture, whereas weakly connected or partitioned networks act as financial firewalls that contain default cascades.

Following the 2007–2008 subprime mortgage crisis and the collapse of Lehman Brothers, central bankers and economic theorists confronted an apparent theoretical paradox regarding interbank linkages:

1. The Diversification-as-Insurance View (Allen and Gale, 2000; Freixas, Parigi, and Rochet, 2000): In the classic Allen–Gale financial contagion model and the Freixas–Parigi–Rochet systemic risk and interbank gridlock model, denser interbank linkages dilute an idiosyncratic default loss across all (n − 1) counterparties so each creditor bank absorbs a small loss within its liquid buffer without defaulting.
2. The Interconnectedness-as-Contagion View (Haldane, 2009; Blume et al., 2011): Epidemiological and network scientists argued that dense financial interconnections behave like high-contact biological networks, creating "too-interconnected-to-fail" institutions and propagating tail shocks across the entire financial system.

Acemoglu, Ozdaglar, and Tahbaz-Salehi (2015) demonstrated that both views are special regimes of a single unified phase diagram governed by the excess liquidity supply of unshocked banks relative to the aggregate liquidity deficit of shocked banks.

Timeline, Balance Sheets, and Senior Non-Financial Claims

Consider an economy with n financial institutions (banks) indexed by i ∈ {1, 2, ..., n} operating over three dates (t = 0, 1, 2):

• Date t = 0 (Borrowing and Project Investment): Each bank i borrows working capital from other banks via bilateral interbank debt contracts y_ij ≥ 0 (the nominal face value bank i promises to pay bank j at t = 1). In a regular network (such as a symmetric ring or a symmetric complete network), every bank holds identical total interbank liabilities ∑_j≠i y_ij = y and identical total interbank assets ∑_j≠i y_ji = y, so net interbank positions are zero in the absence of default.
• Asset Structure at Date t = 1: At t = 1, each bank i holds three asset sources and two liability tiers:
  1. Liquid Cash Holding (c > 0): Available in cash at t = 1.
  2. Short-Term Project Return (z_i ≥ 0): Subject to exogenous idiosyncratic negative shocks at t = 1. Unshocked banks realize the normal return z_i = a > 0, whereas a shocked bank realizes a lower return z_i = a − ε (where ε > 0 is the shock size).
  3. Long-Term Illiquid Project (R > 0): Yields full return R at t = 2 if held to maturity, but yields a discounted salvage recovery ζR (with ζ ∈ [0, 1)) if liquidated prematurely at t = 1 to pay creditors.
• Liability Seniority Structure: Each bank owes an exogenous senior obligation v > 0 at t = 1 (representing retail deposits covered by the Diamond–Dybvig model, tax obligations, or short-term commercial paper) that is legally senior to junior interbank liabilities y_ij.

Modified Eisenberg–Noe Clearing with Costly Liquidation

Let x_ij ∈ [0, y_ij] denote the actual interbank payment made by bank i to bank j at t = 1, and let x_i = ∑_j≠i x_ij be bank i's total interbank payout. Because premature liquidation of the long-term project destroys value (ζ < 1), a bank liquidates its long-term asset up to maximum salvage capacity ζR if and only if its liquid cash c + z_i plus incoming interbank receipts ∑_k≠i x_ki fall short of its total obligations v + y. If even full liquidation (ζR) cannot cover the senior claim v + y, bank i defaults on its junior interbank debt and pays its interbank creditors pro rata (x_ij = x_i(y_ij/y)) according to the piecewise-linear clearing equation:

where [u]⁺ ≡ max{u, 0}. Social welfare W equals the aggregate economic output minus the deadweight liquidation losses (1 − ζ)l_i incurred across all banks i that are forced to liquidate long-term projects prematurely, where social distress scales directly with the total number of defaulting banks.

Canonical Network Topologies (G_ring, G_complete, and G_λ)

To analyze how network geometry governs contagion, Acemoglu, Ozdaglar, and Tahbaz-Salehi parameterize the class of symmetric regular networks through convex combinations of two polar extremes:

• The Ring Network (G_ring): Bank i owes its entire interbank liability y to a single predecessor (y_i,i−1 = y, and y_ij = 0 otherwise), forming a single directed cycle 1 → n → (n − 1) → ··· → 1.
• The Complete Network (G_complete): Bank i splits its interbank liability y equally across all (n − 1) other banks (y_ij = y/(n − 1) for all j ≠ i).
• Convex-Combination Networks (G_λ): For weight λ ∈ [0, 1], each bank owes (1 − λ)y to its immediate ring neighbor and distributes the remaining fraction λy symmetrically across all (n − 1) other banks, so λ = 0 recovers the ring network and λ = 1 recovers the complete network.

Define the individual full-liquidation buffer of an unshocked bank as s ≡ c + a + ζR − v > 0 (the maximum loss an unshocked bank can absorb before defaulting on its interbank debt). When one bank experiences a negative shock of size ε > s (large enough to force the shocked bank itself into default), the net liquidity surplus of the (n − 1) unshocked banks relative to the shocked bank's deficit is controlled by the structural phase-transition threshold:

More generally, when k banks simultaneously suffer negative shocks of size ε, the phase-transition threshold is ε*(k) = (n / k)s.

Why Weakly Connected Networks Outperform Complete Networks in the Large-Shock Regime

The most striking result of the model is what happens in the large-shock regime (ε > φ* = n · s). When ε > n · s, the aggregate net liquidity of the entire n-bank financial system (n · s − ε < 0) is negative. In the complete network (G_complete), because all banks share the loss symmetrically across every counterparty without friction, the negative aggregate surplus pulls every single bank (n out of n) into simultaneous default.

By contrast, in a ring network (G_ring) or a weakly connected block/cluster network, default contagion stops after a smaller number of defaults m < n whenever senior non-financial claims v are sufficiently large. The mechanism is loss absorption by senior non-financial creditors: when Bank 1 suffers a large-scale shock ε > n · s, Bank 1's payout to its ring neighbor cannot fall below zero (x₁ = 0) because of limited liability. The remaining excess portion of the shock (ε − s − y) is borne by Bank 1's external senior creditors (who suffer a haircut on v), rather than being transmitted into the interbank network! Each subsequent bank that defaults along the ring similarly absorbs up to s + v of the shock before its interbank payment hits the zero lower bound x_i = 0, truncating the cascade and leaving the remaining (n − m) banks solvent.

The Robust-Yet-Fragile Property

Acemoglu, Ozdaglar, and Tahbaz-Salehi formalize the "robust-yet-fragile" nature of densely interconnected financial networks (earlier simulated in random graphs by Gai and Kapadia, 2010):

• Robustness to High-Probability, Small-to-Moderate Shocks: Under standard probability distributions where small and moderate shocks (ε < φ*) occur with high frequency, increasing network completeness λ from 0 to 1 monotonically lowers the probability of contagious defaults.
• Fragility to Low-Probability Tail Shocks: Conditional on a rare macroeconomic or sectoral tail shock (ε > φ* or a high number k of correlated defaults), the same complete network maximizes the number of bank failures and social deadweight liquidation losses.

Financial Network Externalities and Market Failure in Endogenous Formation

In the final section of their paper, Acemoglu, Ozdaglar, and Tahbaz-Salehi endogenize the formation of the financial network at Date t = 0 as a strategic game among borrowing and lending banks (solved via Nash equilibrium and coalition-proof stability). When Bank A lends to Bank B, the bilateral interest rate charged by Bank A prices Bank B's direct default risk (including who else Bank B lends to). However, once Bank B has already borrowed from Bank C, Bank B and Bank A do not internalize the negative network externality that their new credit link imposes on Bank C's counter-party risk exposure. Because of this financial network externality, decentralized interbank markets systematically over-connect or fail to form socially optimal ring/firewall buffers, creating a rigorous welfare justification for macroprudential capital surcharges scaled to network centrality and interbank exposure limits.

How does the Acemoglu–Ozdaglar–Tahbaz-Salehi (2015) model generalize Allen and Gale (2000)?

The Allen–Gale financial contagion model examined an arbitrarily small unanticipated liquidity shock ε > 0 and concluded that complete networks are always more stable than incomplete ring networks. Acemoglu, Ozdaglar, and Tahbaz-Salehi (2015) show that Allen and Gale's conclusion holds inside the small-shock regime (ε < φ* = n · s), but reverses above the phase-transition threshold (ε > φ*), where complete networks collapse completely while ring or partitioned networks contain the crisis.

What is the role of senior non-financial obligations (v) in halting contagion in weakly connected networks?

Because junior interbank payments x_i are bounded below by zero under limited liability (x_i ≥ 0), a bank that suffers an extremely large negative shock cannot pass a negative payment to its interbank creditors; once x_i hits zero, any residual loss is absorbed by the bank's external senior creditors (v). In a ring or weakly connected network, a few defaulting banks hit x_i = 0 and force their external senior claimants to absorb the tail of the shock, acting as a circuit breaker that shields downstream banks.

Why do decentralized banks fail to form the socially optimal financial network on their own?

When Bank i chooses its interbank lending counterparties and interest rates, it cares about the solvency of its direct debtors, but it does not internalize how its lending links alter the probability that a shock propagates through its own balance sheet to infect its creditors and creditors' creditors. This unpriced financial network externality causes laissez-faire interbank equilibria to deviate from the Pareto-efficient network architecture.

1. Acemoglu, D., Ozdaglar, A., & Tahbaz-Salehi, A. (2015). "Systemic Risk and Stability in Financial Networks." American Economic Review, 105(2), 564–608. https://doi.org/10.1257/aer.20130456
2. Allen, F., & Gale, D. (2000). "Financial Contagion." Journal of Political Economy, 108(1), 1–33. https://doi.org/10.1086/262109
3. Eisenberg, L., & Noe, T. H. (2001). "Systemic Risk in Financial Systems." Management Science, 47(2), 236–249. https://doi.org/10.1287/mnsc.47.2.236.9835
4. Gai, P., & Kapadia, S. (2010). "Contagion in Financial Networks." Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 466(2120), 2401–2423. https://doi.org/10.1098/rspa.2009.0410
5. Elliott, M., Golub, B., & Jackson, M. O. (2014). "Financial Networks and Contagion." American Economic Review, 104(10), 3115–3153. https://doi.org/10.1257/aer.104.10.3115