Banking System ABM Model
How do interbank lending networks, heterogeneous bank balance sheets, and capital adequacy constraints generate systemic risk, contagion cascades, and too-big-to-fail dynamics that aggregate risk metrics miss entirely?
Understand the mathematics
Follow the same equations and assumptions one section at a time or in a continuous reading.
Hover or focus a symbol defined in this model to find its exact matches. Select it to keep the highlight; Escape clears the selection.
Odd+D Protocol: Model Specification
The ODD+D protocol (Grimm et al. 2006, 2010; Muller et al. 2013) standardizes documentation for agent-based models. The protocol is applied here to the banking system ABM family, synthesizing the common structure across Iori et al. (2006), Georg (2013), Klimek et al. (2015), and Poledna et al. (2015).
PURPOSE: The model explores how the network structure of interbank lending, combined with heterogeneous bank balance sheets and regulatory capital constraints, generates systemic risk that is invisible to single-bank stress tests. The target audience is macroprudential regulators designing capital surcharges, liquidity requirements, and resolution regimes for systemically important banks.
ENTITIES, STATE VARIABLES, AND SCALES: The model contains five entity types. Banks ( to in research runs, 10 to 30 in toy demonstrations) carry state variables: total assets (continuous, positive, Pareto-distributed across banks), equity (continuous, can go negative triggering failure), loan portfolio (vector of exposures to firms), interbank claims and obligations (bilateral exposure matrix), securities holdings (continuous, marked to market), and reserve balances (continuous, non-negative). Firms ( to in research, 50 to 200 in toy) carry: revenue, debt to each bank, default probability. Households (, typically aggregated) carry: deposit balances at each bank. The central bank carries: policy rate, standing facility rate, and emergency lending capacity. The regulator carries: minimum capital ratio, systemic risk buffer schedule, and resolution authority. The temporal scale is quarterly over 20 to 60 quarters. The spatial scale is the network topology, which is endogenous.
PROCESS OVERVIEW AND SCHEDULING: Each period proceeds in seven phases: (1) macro shock arrives and updates firm revenues and default probabilities, (2) firm defaults realize and credit losses hit bank equity, (3) the interbank market opens - banks with surplus reserves lend to banks with deficits at an endogenous rate, (4) banks below capital targets rebalance portfolios by selling securities, (5) fire-sale price impacts are computed from aggregate selling volume, (6) banks below are resolved - assets liquidated, interbank claims written down, contagion losses propagate, (7) the regulator updates systemic risk scores and the central bank adjusts the policy rate. This ordering ensures that banks attempt market-based funding before being forced into fire sales, matching the observed sequence during actual banking crises.
Bank 's equity evolution. Net interest income (NII) adds to equity. Three loss channels subtract: credit losses from firm defaults, mark-to-market losses from fire-sale price drops on securities, and contagion losses from interbank counterparty defaults. The relative magnitude of each channel determines whether the crisis is credit-driven, liquidity-driven, or contagion-driven.
Capital adequacy ratio with the risk weights used throughout this file. Loans carry full weight, interbank claims 20%, securities 50%, so at the seeded 70/20/10 mix . The 20% interbank weight is the Basel I (1988) OECD-club rule, not Basel II: Basel II's standardised approach keys bank exposures to external ratings and its IRB approach to modelled PD/LGD. The flat OECD-club weight was a deliberate regulatory choice that underpriced systemic risk - one of the structural failures exposed in 2008. Banks manage to , which is the threshold the capital-triggered fire sale keys off.
Securities price update under fire-sale pressure, indexed by SELLING ROUND rather than by period . Within a period the price impact is linear in that round's volume, but capital ratios are re-marked after every round, so the round-one discount pushes further banks below and round two sells more. It is that re-evaluation - not the price equation itself - that makes the within-period impact superlinear in the number of sellers; a single one-shot application of the formula would be strictly linear and the explanation would be false. The market depth is 1.2B, equal to the system securities float, so reads as the fraction of the float sold. The positivity constraint is binding on the model's domain: a round that sold more than of the float would return a zero or negative securities price, so at the default no single round may sell more than ten floats (never binding) but at it caps a round at two floats.
The 20% interbank risk weight problem
The flat 20% risk weight on claims against banks incorporated in OECD countries is a BASEL I rule - the 1988 Accord's 'OECD club' treatment - not a Basel II rule. Basel II replaced the club distinction with ratings-based weights in the standardised approach and modelled PD/LGD in the IRB approaches; the point is that the 20% figure and the club logic belong to 1988. Under either regime the effect was the same: banks were encouraged to increase interbank lending and hold little capital against those exposures. When Lehman failed, the low weight meant that banks had much less capital backing their interbank claims than the actual loss severity warranted. Basel III tightened bank-exposure treatment and introduced the leverage ratio as a non-risk-based backstop precisely because risk weights had underpriced systemic exposures.
Core-periphery network structure
Empirical interbank networks consistently exhibit core-periphery structure: a dense core of large, highly connected banks surrounded by a sparse periphery of smaller banks that connect mainly to core banks. Craig and von Peter (2014) documented this for the German banking system. In ABMs, this structure emerges endogenously from preferential attachment and size heterogeneity. The implication for contagion is that a core bank failure cascades rapidly through the core, while a periphery bank failure is typically absorbed by the core bank it connects to.
Robust yet fragile
Gai and Kapadia (2010) showed that highly connected financial networks have a dual character. High connectivity means liquidity is shared efficiently and small shocks are absorbed by many counterparties. But the same connectivity means that a sufficiently large shock can reach every bank. The cascade probability as a function of connectivity is non-monotonic: it first decreases (better shock absorption) then increases (wider propagation). This robust-yet-fragile property is one of the most important results in financial network theory.
Transition checks
Agent update
The aggregate is computed after individual states are updated. If the rule is nonlinear or stochastic, the aggregate can change abruptly even when the plotted line connects adjacent periods.
Each point is an aggregate after one model update. The connecting line helps the eye, but the law is discrete.
Do not read the line as a continuous differential equation unless one is stated.
Cutoff and contagion
The indicator changes value discontinuously at . Aggregates inherit that discontinuity when enough agents sit near the cutoff.
A cutoff creates a kink or jump because the update rule changes at the threshold.
Do not force a smooth explanation onto a chart section driven by threshold crossing.
Variable timing
Agent state
Agent states update at dates. Abrupt aggregate movement can come from many small state changes applied at once.
Aggregate readout
The aggregate can jump when the distribution changes shape or many agents cross a boundary.
Threshold indicator
The indicator turns on at the cutoff. That creates a discontinuity in the micro rule and possibly the aggregate.