Monetary Policy ABM Model
How do interest rate changes propagate through an economy of heterogeneous households and firms with different balance sheets, credit access, and expectations - and why does this transmission break down at the zero lower bound?
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 (Overview, Design concepts, Details + Decision-making) protocol is the standard documentation format for agent-based models, introduced by Grimm et al. (2006, 2010) and extended by Muller et al. (2013) for decision-making. The protocol is applied here to the monetary policy ABM family.
PURPOSE: The model explores how interest rate changes propagate through an economy of heterogeneous households, firms, and banks, producing distributional outcomes that representative-agent models cannot capture. The target audience is central bank researchers evaluating conventional and unconventional monetary policy transmission, distributional impact, and ZLB dynamics.
ENTITIES, STATE VARIABLES, AND SCALES: The model contains five entity types. Households ( to in research runs, 50 to 200 in toy demonstrations) carry state variables: income (continuous, positive), WEALTH (continuous, decomposed into deposits, bonds, and real assets - initialized explicitly, since both the MPC schedule and the leverage schedule are functions of the wealth percentile and neither is evaluable without it), debt (continuous, non-negative, set as with DTI declining in the wealth percentile rather than drawn independently of it), MPC (continuous, in [0.05, 0.95] on the unconstrained branch, assigned by wealth percentile), and inflation expectations (continuous, with a credibility weight ). Firms ( to in research, 10 to 50 in toy) carry: capital stock, cash holdings, outstanding loans, expected demand, and productivity. Banks ( to ) carry: deposit base, loan portfolio, bond holdings, central bank reserves, capital ratio, and lending spread. The central bank (1 agent) carries: policy rate, balance sheet (bond holdings for QE), and a communication state (forward guidance signal). The government (1 agent) carries: tax rate, spending level, and outstanding debt. The temporal scale is quarterly periods over 40 to 80 quarters.
PROCESS OVERVIEW AND SCHEDULING: Each period proceeds in fixed order: (1) central bank evaluates Taylor rule and sets policy rate, subject to ZLB, (2) central bank executes QE purchases/sales if active, (3) banks partially adjust lending rates toward the policy rate plus a margin-targeting spread, (4) households and firms form inflation expectations, (5) firms form demand expectations and make investment decisions on the real lending rate, (6) firms apply to banks for investment loans and banks screen them, (7) firms convert planned output into labor demand, the labor market matches, and the wage responds to tightness and expected inflation - so household INCOME is determined here rather than drifting exogenously, (8) households compute disposable income (wage minus debt service), (9) households allocate between consumption and saving using the MPC rule, with the constrained branch for high debt-service ratios, (10) firms set prices as a markup over unit cost and aggregate inflation is the output-weighted average of firm price changes, (11) defaults are realized (firms and households), (12) bank capital is updated and spreads adjusted, (13) asset prices adjust to the long rate determined in step 2, (14) credibility weights update and aggregate statistics are recorded. The ordering ensures that monetary policy changes in step 1 propagate through the credit channel (steps 2-6), the income channel (steps 7-8), and the demand channel (steps 9-10) within each period, with amplification and feedback in subsequent periods. Steps 7 and 10 are load-bearing and were previously missing from the runnable specification: with income evolving exogenously the income channel has no mechanism, and with inflation set by an imposed aggregate identity the Phillips relation is assumed rather than emergent.
Taylor rule with ZLB and an explicit monetary shock term. Two fixes are embedded here. First, carries a value - 3%, the neutral nominal rate implied by a 1% neutral real rate and a 2% target - so the rule is evaluable. Second, the slider enters as the shock rather than as a starting level: with a level-only reading, the rule at sees inflation at target and output at , returns exactly , and discards whatever the slider was set to, which is why all three shock presets previously set an inert parameter. The max operator imposes the zero lower bound; when the unconstrained rate is negative, conventional policy is exhausted.
Bank 's lending rate under INCOMPLETE pass-through, with two independent sources of incompleteness. The first is the bank-specific adjustment speed : only a fraction of the gap between the target rate and last period's rate is closed each quarter, so the impact-period pass-through is and the loan-weighted aggregate lands in [0.4, 0.9] - exactly the range the emergence section claims and the readout reports. The second is the margin-targeting spread: adjusts toward a target margin , and because realized margins move with the rate environment, the spread partially offsets policy moves. The earlier form had a spread that did not depend on at all, so identically - perfect pass-through by construction, at every parameter setting, contradicting this file's own assumption block, its readout, and its 0.4-0.9 emergence claim. The seed made it worse by starting banks at a capital ratio equal to , so the multiplier was exactly 1 and the only spread adjustment in the model fired when the capital ratio fell below , which retained profits push away from rather than toward. Banks now start at 11% against an 8% minimum.
The single household consumption rule used everywhere in this file - the graph tab, this equation, and the step-by-step derivation all state this and nothing else. Three things were reconciled. (1) The constrained branch is real and reachable: on the debt-service ratio binds for the bottom decile at a 5% lending rate and for about 44% of households at 7%, so tightening pushes households into the corner. The corner MPC of 1 sits outside the [0.05, 0.95] schedule deliberately - the schedule prices the unconstrained branch only. (2) The real-rate term gives inflation expectations somewhere to go. Without it, is constructed, is updated, forward guidance is announced, and nothing in any decision rule reads any of it, so there is no real-rate channel and no mechanism behind the forward-guidance results. , . (3) The wealth term is now computable: comes from the bond-price block, and households carry an explicit wealth state. The distributional gradient comes from DTI heterogeneity, NOT from MPC heterogeneity: in percentage terms , in which MPC cancels exactly.
The zero lower bound as a regime switch
The ZLB is a qualitative regime change, not a parameter bound. Above the ZLB, the central bank can always offset demand shocks by cutting rates. At the ZLB, the conventional instrument is exhausted and only unconventional tools remain. ABMs handle this regime switch naturally because the operator is just a constraint on one agent's action space. DSGE models struggle because the ZLB creates a kink in the policy function that breaks perturbation-based solution methods.
Why pass-through is incomplete
In the data, pass-through is fastest and most complete at the SHORT end: short-maturity corporate and business lending rates, which reprice off policy-linked benchmarks, track the policy rate closely, while mortgage rates - especially fixed-rate mortgages priced off longer-maturity funding - move less and more slowly. The earlier statement here had that ordering inverted. The ABM produces incomplete pass-through through two channels: the bank-specific adjustment speed , which sets the impact-period response, and the margin-targeting spread, which partially offsets policy moves when banks need earnings to rebuild capital. Both are necessary. A specification in which the spread does not depend on the policy rate delivers pass-through of exactly one regardless of how the capital multiplier is parameterized.
MPC heterogeneity is the distributional engine
Fagereng, Holm, and Natvik (2021) estimate household-level MPCs from Norwegian lottery data and report a bottom-wealth-QUARTILE MPC of roughly 0.55 and a top-quartile MPC of roughly 0.05, declining smoothly in between. Those are the figures used in the sibling household-ABM reference, and they are the ones used here; the 0.05/0.85 pair previously quoted does not appear in that paper. Jappelli and Pistaferri (2014) is not a survey of the literature - it is an original study using Italian SHIW data in which households are asked directly what they would do with an unexpected windfall, and its headline finding cuts the other way: MPC heterogeneity is only WEAKLY related to cash-on-hand, which is a caution against reading the wealth-MPC gradient as tight. The ABM's MPC schedule is a stylized linear gradient in the wealth percentile, calibrated to span these estimates at maximum dispersion, not a fit to either paper. Note that the aggregate demand response to a rate change is dominated by high-LEVERAGE households rather than by high-MPC households as such: in percentage terms the MPC cancels out of and the debt-to-income ratio is what remains.
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.