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?
Agent-based models · Model guide
How do interest rate changes propagate through an economy of heterogeneous households and firms with different balance sheets, credit a...
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?
Standard New Keynesian models compress the entire household sector into a representative agent with a single Euler equation linking the policy rate to consumption. That compression erases the distributional mechanics of monetary transmission: rate cuts benefit leveraged borrowers through lower debt-service costs but punish savers through lower interest income, and the net stimulus depends on the marginal propensities to consume across the income and wealth distribution, not on the average MPC. Agent-based models of monetary policy exist precisely because this heterogeneity is load-bearing for the transmission mechanism. Dosi et al. (2015) built one of the first comprehensive macro ABMs with an explicit monetary policy channel, embedding a Taylor-rule central bank inside a Keynesian economy with heterogeneous firms, banks, and workers. Poledna et al. (2023) extended the approach with a calibrated multi-sector ABM for the Austrian economy that tracks monetary transmission through firm investment, household consumption, and bank lending simultaneously. Haldane and Turrell (2018) made the case at the Bank of England that ABMs are a necessary complement to DSGE for central bank policy analysis, specifically because they handle nonlinearities and distributional effects that perturbation methods miss.
The core mechanism is a chain of heterogeneous responses to a rate change. When the central bank cuts the policy rate, the immediate effect is a reduction in bank lending rates and bond yields. But the downstream effects fan out unevenly. Firms with high leverage and tight cash flow constraints see the largest investment response because lower rates relax their borrowing constraint - firms with ample cash are unconstrained and barely respond. Households with adjustable-rate mortgages get immediate payment relief; those with fixed-rate mortgages or no mortgage at all see no direct effect on disposable income. Savers earning interest income lose purchasing power. Asset holders gain from valuation effects as discount rates fall. The aggregate stimulus is the sum of these heterogeneous responses, weighted by each agent's spending propensity. When the rate hits the zero lower bound, the conventional channel shuts off and unconventional tools (QE, forward guidance) operate through different - and less well-understood - distributional pathways.
Central banks are the primary institutional audience. The Bank of England's research directorate has published ABM work on monetary transmission (Haldane-Turrell 2018, Turrell et al. 2019) and uses ABM-derived insights to supplement DSGE-based policy analysis. The ECB's research division has explored ABM-based monetary policy evaluation through the Eurace model (Dawid et al. 2018), which includes a full monetary sector with interbank markets, a central bank with multiple instruments, and heterogeneous banks with balance sheet constraints. The IMF's research group has published work on ABMs for macroeconomic policy (Bookstaber et al. 2018, Fagiolo-Roventini 2017) that explicitly covers monetary transmission. Academic use spans from pure theory (Assenza et al. 2015 on heterogeneous expectations and monetary policy) to applied calibration (Poledna et al. 2023 with Austrian input-output tables).
The model family has evolved along several dimensions. Early versions used a single policy instrument (the short-term rate) and a compact banking sector. Current versions include multiple instruments (rate, QE, forward guidance, macroprudential overlays), heterogeneous banks with interbank lending, multi-sector production with input-output linkages, and labor markets with search frictions. The active frontier includes digital currency (CBDC transmission), climate-adjusted monetary policy (green QE), and the interaction between monetary and fiscal policy when both operate near their effective bounds. Seppecher, Salle, and Lang (2019) contributed a particularly clean specification that isolates the monetary transmission channel in an otherwise minimal ABM, making the distributional effects of rate changes analytically transparent.
The economy is assembled from five agent populations. Households are the demand-side agents: each carries an income stream, a wealth position (split between deposits, bonds, and real assets), a debt position (mortgage and/or consumer credit), a consumption rule, and a set of expectations about inflation and income growth. The household population is typically 5,000 to 100,000 in research implementations and 50 to 200 in browser-scale toy models. Firms form the second population: each holds a production technology, a capital stock, a cash position, outstanding bank loans, and an investment rule that depends on expected demand and the cost of borrowing. Banks are the third population: each maintains a balance sheet (deposits, loans, bonds, central bank reserves, equity), sets lending rates as a spread over the policy rate, and manages a capital constraint. The central bank is a single agent with a Taylor-rule interest rate, a balance sheet (for QE operations), and a forward-guidance communication channel. A compact government sector collects taxes, issues bonds, and runs a fiscal rule.
Interaction flows through four market channels. The credit market connects firms and households to banks: loan applications are evaluated against the borrower's balance sheet and the bank's capital position, with the lending rate set as a markup over the policy rate plus a risk premium. The goods market connects households to firms: consumption demand depends on disposable income (after debt service), wealth effects, and expectations. The labor market connects firms to households: firms hire based on expected output, and wages adjust to labor market tightness. The bond market connects the central bank to banks and households: the central bank buys and sells government bonds to implement QE, which affects long-term rates and bank reserve positions. Asset prices (bonds, equities, housing) emerge from the interaction of demand, supply, and the discount rate channel.
State variables update each period in a fixed sequence: (1) the central bank sets the policy rate and announces forward guidance, (2) banks adjust lending rates and credit standards, (3) firms make investment and hiring decisions, (4) the labor market clears, (5) households receive income and make consumption and saving decisions, (6) goods and credit markets clear, (7) defaults are realized and losses allocated, (8) the central bank updates its balance sheet (QE purchases/sales), (9) asset prices adjust, (10) aggregate statistics are recorded. This sequential structure makes the transmission chain from a rate change to its real-economy effect traceable step by step. The propagation lag - typically 4 to 8 quarters for peak GDP effect in calibrated models - emerges from the interaction of adjustment speeds across the five agent populations, not from a calibrated impulse-response function imposed on the model.
The Bank of England's research directorate has used ABM-derived insights to supplement DSGE-based monetary policy analysis. Haldane and Turrell (2018) argued that ABMs capture three features DSGE cannot: endogenous crises, nonlinear amplification, and distributional effects of policy. The Eurace model (Dawid et al. 2018), funded by the EU and developed at the University of Bielefeld, provides the most complete ABM implementation of monetary transmission for the euro area, with heterogeneous banks, an interbank market, a central bank with multiple instruments, and input-output linkages across sectors. Poledna et al. (2023) calibrated a monetary ABM to Austrian national accounts and input-output tables, showing that the distributional effects of a 100bp rate cut differ by a factor of 3 across income quintiles - information that aggregate DSGE impulse responses cannot provide.
The model's second major application is ZLB and unconventional policy analysis. When the policy rate hits zero, the conventional interest rate channel is severed and the central bank must operate through QE (balance sheet expansion), forward guidance (expectation management), and yield curve control (long-rate targeting). ABMs handle the ZLB naturally because it is simply a constraint on one agent's decision variable, not a singularity in a log-linearized system. Assenza et al. (2015) used an ABM with heterogeneous expectations to show that forward guidance effectiveness depends critically on how many agents anchor expectations to the central bank's communication versus extrapolating from recent experience. When the share of backward-looking agents is high, forward guidance has almost no effect - a result that matches the empirical evidence on the forward guidance puzzle better than rational expectations models.
The model breaks down in three settings. First, when monetary transmission operates primarily through the exchange rate (small open economies with flexible exchange rates and high trade openness), the domestic credit and demand channels that the model focuses on are secondary. An open-economy ABM or a Mundell-Fleming variant is more appropriate. Second, when the question requires welfare-theoretic foundations for optimal policy design, the ABM cannot provide the social welfare function needed to evaluate policy optimality - DSGE with heterogeneous agents (HANK models) is the right tool. Third, when the institutional setting involves a currency board, dollarization, or membership in a monetary union without an independent central bank, the model's central bank agent has no instruments and the entire transmission story is inapplicable.
An individual household with state vector (income, deposits, debt, asset holdings, MPC, inflation expectations). Decisions: consume, save, borrow, adjust portfolio.
A producing firm with state vector (capital, cash, outstanding loans, expected demand, productivity). Decisions: invest, hire, set prices, apply for credit.
A financial intermediary with state vector (deposits, loan portfolio, bond holdings, reserves, capital ratio). Decisions: set lending rate, approve/reject loans, manage capital buffer.
The central bank's short-term interest rate instrument, set by a Taylor-type rule responding to inflation and output gaps. The anchor for all other rates in the economy.
The stock of government bonds held on the central bank's balance sheet. QE purchases increase bank reserves and compress long-term yields; tapering reverses both effects.
Agent-specific expected inflation rate. Heterogeneous across households: some extrapolate recent inflation, others anchor to the central bank target. The distribution of expectations determines forward guidance effectiveness.
Bank k's markup over the policy rate when setting loan rates. Depends on the bank's capital ratio, perceived borrower risk, and competitive pressure. The spread is the key friction in the interest rate pass-through channel.
Household i's share of an additional dollar of disposable income that goes to consumption. Heterogeneous across the income distribution: constrained households have MPC near 1, wealthy households near 0.05. The distribution of MPCs determines the aggregate demand response to a rate change.
Households and firms form expectations using adaptive heuristics (e.g., weighted average of past inflation and the central bank target) rather than model-consistent rational expectations.
If violated: Rational expectations eliminate the distributional dynamics that are the model's reason for existing. With RE, forward guidance works perfectly, the ZLB puzzle disappears, and all agents respond identically to a rate change - exactly the compression the ABM is designed to avoid.
Households and firms draw from calibrated joint distributions of income, wealth, and debt, with persistent idiosyncratic shocks to income and productivity each period.
If violated: Homogeneous agents collapse the model into a representative-agent setup and eliminate the distributional transmission channel. The MPC distribution, which is the primary driver of heterogeneous monetary transmission, becomes a point mass.
Banks set lending rates as a spread over the policy rate. The spread adjusts sluggishly and depends on the bank's capital position, so rate changes do not pass through one-for-one or instantaneously to borrowing costs.
If violated: Perfect pass-through would make the banking sector irrelevant to transmission. The model would reduce to a direct link between the policy rate and household/firm borrowing costs, eliminating the bank capital channel.
Credit, goods, labor, and asset markets clear in a fixed order each period rather than simultaneously in general equilibrium.
If violated: Simultaneous clearing would require a general-equilibrium fixed point that erases the propagation sequence and lag structure the model is designed to study.
The policy rate cannot fall below zero (or a small negative bound). When the constraint binds, conventional monetary policy is exhausted and only unconventional instruments remain.
If violated: Without a ZLB, there is no need for unconventional policy and no regime switch in the transmission mechanism. The model becomes a smooth, symmetric story about rate changes.
No exchange rate channel, no foreign capital flows, no imported inflation pass-through. The baseline monetary transmission operates entirely through domestic credit, demand, and asset price channels.
If violated: Open-economy extensions (e.g., Dawid et al. 2018 Eurace with cross-border banking) exist but are not part of the baseline specification. For small open economies, the exchange rate channel can dominate the domestic credit channel.
Fiscal policy follows an exogenous rule (fixed deficit target or debt brake). There is no strategic interaction between the central bank and the treasury.
If violated: Fiscal-monetary interaction at the ZLB is an active research frontier. When both authorities are constrained, the transmission mechanism changes qualitatively - helicopter money, yield curve control, and MMT-adjacent policies enter the picture.
Continue reading
Open the concept, data series, policy setting, or neighboring model that anchors this page.