Housing / Credit ABM Model

Representative-agent models compress the entire household sector into a single borrower-saver pair. That compression erases the distributional dynamics that drive housing booms and busts: some households are stretched thin on leverage while others hold excess equity, and the aggregate default rate depends on where the distribution sits relative to the collateral threshold, not on the mean leverage ratio alone. Housing-credit ABMs exist because this distributional story cannot be recovered from an aggregate equation. Geanakoplos (2010) formalized leverage cycles as an emergent phenomenon from heterogeneous beliefs and collateral constraints. Axtell et al. (2014) built one of the first large-scale computational housing models at the household level. Baptista et al. (2016) at the Bank of England constructed a housing ABM to evaluate macroprudential policy (LTV and LTI caps) in the UK market, and Ge (2017) extended the framework with endogenous bank behavior and mortgage contract heterogeneity.

The core mechanism is a feedback loop between house prices, collateral values, credit availability, and household balance sheets. When house prices rise, existing homeowners gain equity, banks face lower LTV ratios on their mortgage books, and credit standards loosen because the collateral buffer looks healthy. New buyers enter at higher leverage, pushing prices further. The cycle reverses when prices stall: marginal buyers hit negative equity first, defaults concentrate in the high-leverage tail, bank losses tighten lending standards, and the credit contraction feeds back into prices. This amplification mechanism is invisible in any model that only tracks the average household.

Central banks and financial stability authorities use housing-credit ABMs operationally. The Bank of England's Financial Policy Committee commissioned a housing ABM specifically to evaluate LTV and LTI caps before implementing them in 2014. The Dutch central bank (DNB) runs ABM stress tests on the Dutch mortgage market. The ECB's research directorate has published housing ABM work for euro-area macroprudential calibration. The IMF's ABBA framework (Bookstaber et al. 2018) includes a housing-credit channel as one of its core contagion pathways. Academic use spans from pure leverage-cycle theory (Geanakoplos, Thurner) to applied policy evaluation (Carro, Ge, Baptista).

The model family has evolved along several axes. Early versions used fixed behavioral rules for mortgage choice; newer variants let households choose between fixed-rate and adjustable-rate mortgages endogenously. Bank behavior has moved from a passive lending function to active portfolio management with capital constraints. Network structure has expanded from random matching to spatially embedded housing markets with neighborhood effects. The active frontier includes climate risk overlays (flood risk repricing collateral), pension-mortgage interactions, and rent-versus-buy decision modeling.

The economy is assembled from four agent populations. Households are the primary agents: each carries an income process, a wealth position, a mortgage contract (or rental status), a house (or not), and a behavioral rule for buying, selling, defaulting, and consuming. Income is drawn from a calibrated distribution with idiosyncratic shocks; wealth accumulates through saving and house price appreciation. The household population is typically 5,000 to 50,000 agents in research implementations and 30 to 100 in browser-scale toy models. Banks form the second population: each holds a mortgage portfolio, a capital buffer, and a lending rule that maps the applicant's LTV, LTI, and debt-service ratio into an accept/reject decision. The third population is a compact firm sector that absorbs aggregate demand and feeds back into household income. The fourth agent is a policy block: a central bank setting the policy rate and a macroprudential authority setting LTV/LTI caps.

Interaction happens through three channels. The housing market is a double-auction or posted-price mechanism where sellers list properties and buyers bid using their budget constraint (income, savings, and maximum mortgage). The credit market connects households to banks: a household applies for a mortgage, the bank evaluates the application against its lending standards and capital position, and the contract is either granted or rejected. The goods market closes the loop: household consumption depends on disposable income net of mortgage payments, and aggregate demand feeds back into firm revenue and household income in the next period.

State variables update each period in a fixed sequence: income shocks arrive, households make buy/sell/default decisions, the housing market clears (prices update), banks process mortgage applications and mark portfolios to market, defaults are realized and losses allocated to bank capital, the macroprudential authority checks its rule, and the central bank updates the policy rate. This sequential structure means the propagation path from a shock to its macro consequence is traceable step by step rather than collapsed into a simultaneous equilibrium condition.

The Bank of England's Financial Policy Committee used a housing-credit ABM (Baptista et al. 2016) to calibrate LTV and LTI caps before implementing them in June 2014. The model showed that a 15% LTV cap reduced the probability of a default cascade by roughly 40% in stress scenarios, while an LTI cap of 4.5 was more effective at containing household debt-service stress during rate-hike episodes. This was one of the first cases where an ABM directly informed a binding macroprudential rule at a major central bank.

The Dutch central bank (DNB) runs housing-credit ABM stress tests annually as part of its financial stability assessment. The Dutch housing market has high average LTV ratios (historically above 100% at origination for many cohorts) and a concentrated banking sector, making distributional dynamics particularly relevant. The ABM captures how a price decline interacts with the underwater-mortgage tail to generate losses that concentrate in specific banks rather than spreading evenly.

The model breaks down when the housing market is driven primarily by factors outside the credit channel. Pure supply-constraint stories (zoning, construction costs, land availability) are not well captured because the model's price mechanism is demand-driven through the credit channel. International capital flows into housing (e.g., Vancouver, London prime) require an open-economy extension. Markets with dominant cash buyers (no mortgage) reduce the leverage amplification that is the model's core mechanism. For these settings, a spatial equilibrium model or a supply-side structural model is more appropriate.