Networks, Shocks, and Systemic Risk

Daron Acemoğlu, Asuman Ozdaglar, Alireza Tahbaz-Salehi2015 · DOI

Abstract

This research develops a unified framework to analyze how network interactions facilitate the propagation and amplification of microeconomic shocks. By nesting various classes of network games, macroeconomic risk models, and financial interaction theories, the authors provide a comprehensive characterization of equilibrium under the assumption of small shocks. This framework clarifies the role of network structure in translating microeconomic volatility into macroeconomic outcomes, allowing for a formal ranking of different network topologies based on aggregate performance and resolving long-standing contradictions in the literature regarding systemic risk.

Introduction

In the aftermath of the 2008 financial crisis, the concept of "interconnectedness" became central to the study of systemic risk. However, the resulting academic literature was often fragmented, producing seemingly contradictory conclusions: while some models suggested that dense network linkages mitigate risk through loss-sharing, others argued that these same links act as conduits for destabilizing contagion.

Acemoglu, Ozdaglar, and Tahbaz-Salehi address this "bewildering array" of findings by proposing a general model that integrates disparate sub-fields—including production (input-output) networks, network games, and financial contagion—into a single mathematical language. Their work identifies the fundamental economic "curvatures" that determine whether a specific network structure enhances or undermines aggregate stability.

The Unified Framework: Beyond Linear Approximations

The authors develop a reduced-form model where $n$ agents possess "states" (representing metrics such as output, solvency, or asset prices) that are determined by the states of their neighbors. The model is built around two primary components: an interaction function $f$, which dictates how agents influence one another, and an aggregation function $g$, which defines the resulting macroeconomic outcome.

Historically, many network models relied on linear (first-order) approximations. The authors demonstrate that these linear models often suffer from a "certainty equivalence" property, where the expected macroeconomic outcome remains identical regardless of the network's structure. To break this equivalence and understand how different topologies respond to idiosyncratic shocks, the authors employ second-order Taylor expansions. By analyzing these second-order terms—the curvatures or non-linearities of the system—they can distinguish how structures like "Star" networks versus "Complete" networks perform from an ex ante perspective.

Key Insights: The Curvature of Risk

The Density Flip (Concave vs. Convex)

The paper’s most significant contribution is identifying the "aha moment" regarding network density. The impact of interconnectedness depends entirely on the curvature of the interaction function:

  • Concave Interactions: If the interaction function is concave, denser networks (such as the Complete network) are beneficial. They allow shocks to be dissipated across many nodes, effectively "washing out" volatility.
  • Convex Interactions: If the interaction function is convex, denser networks are detrimental. In this regime, the network structure serves to amplify shocks, accelerating systemic collapse.

Financial Phase Transitions

This distinction explains the "financial contagion" paradox. In a healthy economy, interbank linkages often operate in a concave regime, providing stability through diversification. However, as shocks increase in magnitude or asset levels dwindle, the system can hit a tipping point where the interaction becomes convex. At this juncture, the same links that previously provided safety suddenly facilitate a systemic meltdown.

Rethinking Centrality

In linear models, Bonacich centrality is typically the standard for identifying systemically important agents. The authors show that in a non-linear world, this metric is insufficient. Two agents with identical Bonacich centralities can have vastly different impacts on systemic risk depending on the local curvature of their interactions. Furthermore, the authors introduce Concentration Centrality for regular networks, which measures how "concentrated" an agent's influence is. In concave environments, agents whose influence is spread thinly across many neighbors are less systemically risky than those with concentrated influence.

Formal Characterizations

The paper provides several foundational theorems to support this framework:

  • Theorem 1: Establishes that an equilibrium always exists and is generically unique under mild regularity conditions, specifically within the context of contraction mappings.
  • Theorems 2 & 3: Provide explicit first- and second-order approximations for the macro state. These formulas illustrate how the Leontief matrix—representing the web of direct and indirect links—interacts with the curvatures of $f$ and $g$ to produce aggregate outcomes.
  • Network Ranking: The authors formally rank the "Complete" network as the optimal structure when interactions are concave and the worst-performing structure when they are convex, bridging the gap between previously conflicting studies.

Limitations and Future Directions

While the framework provides a powerful analytical tool, it relies on the assumption of small shocks to maintain the validity of Taylor expansions. This may not fully capture the dynamics of extreme "black swan" events that exist far from the equilibrium point. Additionally, the current model assumes common knowledge regarding shock realizations; the authors suggest that incorporating information asymmetries—where agents have imperfect knowledge of their neighbors' health—is a vital next step. Finally, the model treats network structures as static and exogenous. Future research into endogenous network formation, where agents choose their links in anticipation of potential shocks, would further refine our understanding of systemic resilience.

Summary updated

Abstract

This chapter develops a unified framework for the study of how network interactions can function as a mechanism for propagation and amplification of microeconomic shocks. The framework nests various classes of games over networks, models of macroeconomic risk originating from microeconomic shocks, and models of financial interactions. Under the assumption that shocks are small, we provide a fairly complete characterization of the structure of equilibrium, clarifying the role of network interactions in translating microeconomic shocks into macroeconomic outcomes. This characterization enables us to rank different networks in terms of their aggregate performance. It also sheds light on several seemingly contradictory results in the prior literature on the role of network linkages in fostering systemic risk.
BibTeX
@misc{acemoglu2015networks,
  author = {Daron Acemoğlu and Asuman Ozdaglar and Alireza Tahbaz-Salehi},
  title = {{Networks, Shocks, and Systemic Risk}},
  year = {2015},
  doi = {10.3386/w20931},
}
Figure