Payment Systems ABM Model
How do banks' individual liquidity management strategies, payment queuing decisions, and bilateral netting behavior generate systemic gridlock risk, cascading settlement failures, and the need for central bank intraday liquidity provision in large-value payment systems?
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 applied to the payment system ABM family follows the standard Grimm et al. (2006, 2010) documentation format extended by Muller et al. (2013) for decision-making. The baseline specification covers Galbiati-Soramaki (2011), Arciero et al. (2009), and Beyeler et al. (2007).
PURPOSE: The model studies how individual bank liquidity management strategies interact through a shared settlement network to produce system-level outcomes - settlement efficiency, gridlock frequency, liquidity usage, and failure rates - that no single bank intends or controls. The target audience is central bank payment system operators evaluating design changes (netting algorithms, intraday credit policy, participation rules) before live deployment.
ENTITIES, STATE VARIABLES, AND SCALES: The model contains three entity types. Banks ( to in calibrated runs, 5 to 15 in toy demonstrations) carry state variables: reserve balance (continuous, non-negative if no overdraft; can go negative up to credit limit), payment queue (ordered list of obligation records), intraday credit line (continuous, determined by collateral), liquidity hoarding parameter (continuous, 0 to 1), and bilateral exposure limits. The central bank (single agent) carries: total intraday credit extended, intervention trigger thresholds, and the netting algorithm specification. The payment generator (exogenous process) produces the daily matrix of payment obligations. The temporal scale is one business day at minute-level or payment-level resolution (480 minutes or 200-2000 payment events). No multi-day dynamics in the baseline.
PROCESS OVERVIEW AND SCHEDULING: Each tick (minute or event) proceeds: (1) new payment obligations arrive according to the intraday timing distribution, (2) banks evaluate their queues and release payments where their strategy permits, (3) released payments settle immediately if the sender has sufficient balance (reserves plus credit), (4) failed submissions return to the queue, (5) at netting-interval boundaries the LSM scans for offsetting cycles, (6) the central bank checks gridlock conditions and intervenes if triggered, (7) state variables are recorded. End-of-day processing: any remaining queued payments are either force-settled using end-of-day standing facility borrowing (at penalty rate) or declared failures.
Bank 's reserve balance update. Incoming settled payments add to reserves; outgoing released payments subtract. The timing of inflows versus outflows is the core strategic variable.
Bank releases a payment to if it is FUNDABLE and either (a) the post-payment balance stays above the hoarding threshold , or (b) the deadline is imminent. Read the structure carefully, because getting it wrong empties the model's headline risk metric. The deadline clause overrides the hoarding threshold - that is behavioural, a bank choosing to spend its cushion. It does NOT override the funding gate, which is the credit limit the ODD imposes on every bank and is not a matter of choice. Written the other way, as a plain OR over all three conditions, the deadline term releases payments the sender cannot pay for; since every deadline falls inside the business day, the queue is then guaranteed to be empty at and settlement failures are identically zero - along with the failure readout, the end-of-day accounting step, and every outage scenario that is supposed to produce them. Multiplication by the funding gate, not disjunction, is what makes a variable.
Gridlock is declared when queue depth exceeds threshold AND no settlements have occurred in the past minutes. Both conditions must hold - a large queue with ongoing settlement is congestion, not gridlock.
Gridlock as coordination failure
Gridlock in payment systems is structurally identical to a coordination game. Every bank would be better off if all submitted payments early (aggregate liquidity is sufficient). But each bank individually prefers to wait for incoming payments before sending outgoing ones, because the bank that pays first bears the liquidity cost. This Nash equilibrium of mutual delay is inefficient - it is the textbook definition of a coordination failure. The netting algorithm and central bank intervention exist precisely to break this equilibrium.
RTGS versus DNS versus hybrid
Pure RTGS (real-time gross settlement) settles each payment individually and immediately but requires massive intraday liquidity. Pure DNS (deferred net settlement) batches all payments and settles only net positions at end-of-day, minimizing liquidity but creating large unsettled exposures. Modern systems are hybrids: RTGS with an embedded LSM that captures netting benefits while maintaining real-time finality. The netting interval is the dial that moves between architectures, but read its endpoints correctly, because the obvious reading is inverted at one end. Setting the interval to 1 does NOT give pure RTGS - it gives continuous NETTING, the LSM scanning for offsetting cycles every single tick, which is the opposite of gross settlement. Pure RTGS is what you get by DISABLING the LSM altogether, and that is not a value of this parameter at all; it is a separate switch. Pure DNS sits at the other end, with the interval equal to minutes, so the only settlement event is the end-of-day batch. The Graph tab's slider runs 5 to 120 minutes and therefore reaches neither limit: it spans the hybrid interior that real systems actually occupy, and both endpoints are conceptual limits to reason about rather than settings you can select.
Too connected to delay
In concentrated payment networks, the top 3-5 banks are large and highly connected. A delay by a hub bank affects its direct counterparties and their counterparties, propagating through 2-3 network hops within an hour. What Soramaki et al. (2007) actually established is the TOPOLOGY that makes such a short path possible: the daily Fedwire network has thousands of participants but a small, densely interconnected core through which most value flows, and an average shortest path of only about 2.6 hops. That paper contains no cascade simulation and no statistic about the share of 'cascade propagation distance' attributable to the top 5 banks - cascade depth is an output of simulation models such as Beyeler et al. (2007), not of the topology study. Keep the two separate: topology explains why a hub delay CAN reach everyone quickly; a simulation is needed to say how much delay it actually causes. Either way, the short core path is why central banks monitor individual bank settlement behaviour in real time.
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.