How do heterogeneous household consumption rules, saving buffers, and wealth positions shape the aggregate demand response to fiscal, monetary, and distributional shocks -- and why does the shape of the income/wealth distribution matter more than the mean?
Agent-based models · Model guide
How do heterogeneous household consumption rules, saving buffers, and wealth positions shape the aggregate demand response to fiscal, m...
How do heterogeneous household consumption rules, saving buffers, and wealth positions shape the aggregate demand response to fiscal, monetary, and distributional shocks - and why does the shape of the income/wealth distribution matter more than the mean?
Representative-agent macro models assume a single household whose marginal propensity to consume (MPC) out of a transitory income shock is pinned by the Euler equation - typically around 0.02 to 0.05 per quarter. Empirical estimates from tax rebates, lottery wins, and stimulus checks consistently report MPCs between 0.20 and 0.60, with the highest values concentrated among low-wealth, low-income households. The gap between theory and data is not a calibration failure; it is a structural consequence of aggregation. When you compress the wealth distribution into a single agent, you erase the mechanism that generates the high-MPC tail: hand-to-mouth households with near-zero liquid wealth who spend additional income immediately because they are constrained, not because they are impatient. Heterogeneous household ABMs exist to study the macro consequences of this distributional fact. Caiani et al. (2016) built a stock-flow consistent ABM where household consumption, saving, and portfolio choice depend on individual balance sheets. Assenza et al. (2015) showed that aggregate fluctuations in a macro ABM are driven by the cross-sectional dispersion of household expectations, not by a representative agent's forecast error. Lengnick (2013) provided the simplest baseline: a closed-economy ABM with heterogeneous households, firms, and a banking sector where business cycles emerge from the interaction of decentralized transactions without any aggregate shock.
The core mechanism is the MPC heterogeneity channel. A fiscal transfer of 500,000 in liquid wealth barely changes spending - the transfer is a rounding error on the buffer stock. A household with $200 in checking and no savings spends most of it within the quarter because the marginal utility of near-term consumption dominates the return to saving when the buffer is near zero. The aggregate multiplier depends on the mass of households in the high-MPC region of the wealth distribution, which is itself endogenous to the history of income shocks, wage dynamics, and prior fiscal policy. This is the distributional amplification mechanism that HANK (Heterogeneous Agent New Keynesian) models capture through Bewley-style optimization on a discretized state space. The ABM version replaces the optimization with behavioral rules - adaptive consumption functions, buffer-stock saving heuristics, and portfolio rules calibrated to household survey data - and lets the distribution evolve from decentralized interactions rather than from the fixed point of a Bellman equation.
Central banks use heterogeneous household models operationally. The ECB's Household Finance and Consumption Survey (HFCS) feeds directly into distributional analysis of monetary policy transmission. The Bank of England's COMPASS framework incorporates heterogeneous MPC estimates when evaluating fiscal-monetary interactions. The Federal Reserve's FEDS Notes series has published distributional multiplier estimates from HANK-style models that inform staff forecasts. On the ABM side, the Eurace@Unibi model (Dawid et al. 2018) includes a detailed household module with heterogeneous consumption, saving, and labor supply decisions, and has been used for EU fiscal policy evaluation. The K+S model (Dosi et al. 2013, 2015, 2017) uses heterogeneous households to study how income distribution and fiscal policy interact to shape business cycle dynamics. The Bank of Italy and the Bank of Canada have both published ABM-based analyses of household sector resilience.
The model family has evolved along several axes since the mid-2010s. Early ABMs used fixed consumption rules (spend a constant fraction of income); newer variants use adaptive rules where the consumption share depends on the household's position in the wealth distribution, recent income history, and sentiment. Portfolio choice has moved from a binary cash-or-bonds decision to multi-asset allocation with housing, equities, and deposits. Labor market modules have expanded from random matching to directed search with skill heterogeneity and on-the-job learning. The active frontier includes digital currency transmission (heterogeneous adoption of CBDC), climate transition costs distributed across the income scale, and pandemic-shock propagation through the household sector.
The economy is assembled from three agent populations plus a policy block. Households are the primary agents: each carries a state vector of income (wage plus transfer income), liquid wealth (deposits), illiquid wealth (bonds or equity, depending on the variant), a consumption rule, a saving target, and a labor status (employed/unemployed). The household population is typically 1,000 to 50,000 in research implementations and 50 to 200 in browser-scale toy models. Firms form the second population: each holds a production capacity, a wage offer, a price, a profit margin, and a hiring/firing rule. Firms absorb household consumption demand and pay wages that feed back into household income. The third population is a compact banking sector: one or more banks that take deposits from households, extend credit to firms, and set deposit and lending rates. The policy block consists of a government that collects taxes, pays transfers, and runs a fiscal rule, plus a central bank that sets the policy rate.
Interaction happens through four channels. The goods market matches households to firms: each period, households allocate their consumption budget across firms based on price and availability, firms fill orders from inventory, and unmet demand is lost (no backorders in the baseline). The labor market matches workers to firms: firms post vacancies when demand exceeds capacity, unemployed households apply, and wages are set by a bilateral rule (posted wage with noise, or Nash-style bargaining in richer variants). The deposit market connects households to banks: households deposit unspent income, banks pay a deposit rate that depends on the policy rate and the bank's funding needs. The credit market connects firms to banks: firms borrow to finance investment and working capital, banks screen based on leverage and profitability.
State variables update each period in a fixed sequence: (1) the government collects taxes and pays transfers, (2) firms set prices and post vacancies, (3) the labor market clears (hiring, firing, wage payments), (4) households receive income and make consumption/saving decisions, (5) the goods market clears (purchases, inventory adjustment), (6) firms compute profits and make investment decisions, (7) banks collect deposits, process loan applications, and update balance sheets, (8) the central bank updates the policy rate, (9) aggregate statistics are recorded. This sequential structure means the propagation path from a fiscal shock through household income, consumption, firm revenue, and back into wages is traceable step by step.
The primary application is evaluating the aggregate demand impact of fiscal transfers with different distributional designs. Dosi et al. (2013, 2015) used the K+S model to show that progressive fiscal policy - higher transfers to low-income households - generates larger output multipliers than uniform transfers, precisely because it channels resources to the high-MPC tail of the wealth distribution. The multiplier difference is not small: targeted transfers to the bottom quintile produce multipliers 1.5x to 2.5x larger than uniform transfers of the same total size in calibrated simulations. This result directly informs the design of stimulus programs, where policymakers face a choice between broad-based and targeted disbursement.
The second major application is monetary policy transmission through the household distribution. A rate cut by the central bank lowers the return to saving, which the representative-agent model converts into a smooth increase in aggregate consumption via the intertemporal substitution channel. In the heterogeneous household ABM, the transmission is uneven: wealthy savers with large deposit balances lose interest income (a contractionary wealth effect), while constrained households with little savings barely notice the rate change but may benefit indirectly if firms expand hiring in response to cheaper credit. The net aggregate effect depends on the relative mass of these groups. Dawid et al. (2018) document this two-channel transmission in the Eurace model and show that the net multiplier of a rate cut can switch sign depending on the wealth distribution - a result impossible to obtain from a representative agent.
The model breaks down when household behavior is well approximated by a representative agent - that is, when the wealth distribution is tight enough that MPC heterogeneity is quantitatively irrelevant. This can happen in economies with strong social insurance (Nordic countries with high replacement rates and universal benefits), where most households have substantial buffer stocks and the high-MPC tail is thin. The model also struggles when the dominant channel is not consumption but asset prices: if the question is about equity market bubbles or housing price dynamics, the household consumption ABM does not have the asset-pricing machinery to generate realistic speculative dynamics. For those questions, a dedicated financial market ABM or housing-credit ABM is more appropriate. Finally, the model is poorly suited to long-run growth analysis because the firm sector is typically compact and innovation dynamics are stylized.
An individual household with state vector (income, liquid wealth, illiquid wealth, consumption rule parameters, labor status, MPC). Decisions: consume, save, allocate portfolio, search for job.
Household i's consumption in period t. Depends on current income, wealth buffer relative to target, and adaptive MPC. The key behavioral rule that generates aggregate demand dynamics.
Household i's liquid asset holdings. The distance between current wealth and the target buffer determines whether the household is in the high-MPC or low-MPC regime.
Household i's spending response to a marginal dollar of income. Endogenous to the wealth-to-income ratio: high when w/y is low (constrained), low when w/y is high (buffer-stock saver).
A producing firm with state vector (output capacity, inventory, price, wage offer, profit, leverage). Absorbs household consumption, pays wages, and borrows from banks.
Total goods produced and sold across all firms. Emergent from the sum of individual firm sales, which depend on the sum of individual household consumption decisions.
Fiscal transfer to households. The distributional design of transfers (uniform, means-tested, top-up) is the primary policy lever in the model.
Central bank instrument. Feeds into deposit rates and lending rates, affecting the return to saving and the cost of firm borrowing. Transmission to aggregate demand runs through the heterogeneous MPC distribution.
Households use adaptive heuristics (buffer-stock rules, rule-of-thumb spending, habit persistence) rather than solving a dynamic stochastic optimization problem.
If violated: Full optimization would push all households toward the Euler equation MPC (~0.03), eliminating the high-MPC tail that drives the distributional amplification mechanism.
Households draw initial income and wealth from a calibrated joint distribution (typically log-normal income, Pareto or double-Pareto wealth) with realistic Gini coefficients.
If violated: Identical starting positions collapse the model into a representative-agent economy after a few periods as idiosyncratic shocks are symmetric.
Households choose firms to buy from based on local information (posted prices, proximity in network, or random matching), not on a Walrasian auctioneer.
If violated: Walrasian clearing would eliminate involuntary inventory accumulation, firm-level demand uncertainty, and the possibility of rationing - all of which contribute to business cycle dynamics in the ABM.
Labor, goods, deposit, and credit markets clear in a fixed order each period, not simultaneously.
If violated: Simultaneous clearing would require a general-equilibrium fixed point, erasing the propagation sequence the model is designed to study.
Households in the baseline specification cannot borrow against future income. Wealth is bounded below at zero (or a small negative limit in extensions).
If violated: Allowing unconstrained household borrowing weakens the buffer-stock motive and reduces the mass of constrained households, compressing MPC heterogeneity.
Each household faces idiosyncratic income risk (job loss, wage changes) that is not perfectly insurable through markets or transfers.
If violated: Complete insurance eliminates the precautionary saving motive, the buffer-stock target becomes irrelevant, and the wealth distribution degenerates.
No trade, no foreign capital, no exchange rate channel. All income is generated and spent domestically.
If violated: Open-economy extensions exist (Caiani et al. 2016 multi-country) but add significant complexity without changing the core household heterogeneity mechanism.
Continue reading
Open the concept, data series, policy setting, or neighboring model that anchors this page.