19th November 2026
Key takeaways
- Institutions matter more than headline ratios. Whether mortgages are fixed or floating, whether lenders have full recourse, whether home equity can be withdrawn, and how property is taxed and planned determine how interest rates transmit and house price shocks spread. "One size does not fit all", so risk indicators such as price-to-income, the credit-gap or credit growth often mislead.
- A debt-service limit set at the current interest rate is a double-edged sword. Even in fixed-rate markets such as France, where borrowers are protected from interest rate risk, it relaxes new credit when rates fall and fuels booms. A loan-to-income limit tightens automatically as prices outrun incomes.
- The short-run costs of borrower limits are real, but the medium-run story may reverse them. Studies that stop at the first-round effect on first-time buyers miss the general equilibrium effects. Also worries about adverse distributional effects are better addressed by tax, land and planning reform, rather than weaker macroprudential rules.
- The standard evidence and models are weaker than they look. Panel studies probably understate how well the tools work, and mainstream DSGE models are poorly suited to the question. I favour country-specific latent-variable models to assess variations in credit standards, and developing agent-based simulation methods.
Summary
Especially since the Global Financial Crisis, housing-related macroprudential tools have been much more widely applied. They consist of borrower-based instruments such as loan-to-value (LTV) caps, loan-to-income (LTI) and debt-service-to-income (DSTI) limits, and associated speed-limits, and lender-based capital measures. Speed limits on the share of high-LTI or high-LTV lending help address the objection that borrower limits interfere with private contracting. Galán (2020) finds that capital measures may take up to two years to show benefits in an upswing, whereas loosening them in a downturn helps immediately. Institutions such as the ESRB, the BIS, the IMF and central banks have enhanced the toolkit and surveillance of potential risks to financial stability.
The financial accelerator is institution-specific
Housing booms and busts are a recurrent source of financial crises: Schularick and Taylor (2012) show that credit growth is a powerful predictor, and the "double trigger" literature, surveyed in Aron and Muellbauer (2016), points to negative equity and cash-flow stress as the key causes of arrears and foreclosures. In a classic credit cycle, excessive weakening of credit standards fuels a credit and real estate boom. With asset prices and debt to income at high levels, affordability pressures and the arrival of negative shocks trigger a downturn. Falling house prices and payment defaults hit bank balance sheets and the economy through a "financial accelerator". However, institutional differences amplify or dampen shocks: these include the tenure structure, the share of fixed-rate lending, amortisation rules, recourse versus collective insurance, home equity withdrawal, funding models, property and transaction taxes, planning rules and how quickly residential construction responds to prices.
A large social rented sector or a large private rental sector weakens the link between house prices and consumption. Lengthening loan terms or switching to interest-only loans in response to tighter rules "effectively loosens lending standards", because it lowers the monthly payment. In France most loans carry a collective insurance scheme among lenders, so negative equity matters little as long as payments continue. That makes onerous debt service, not collateral shortfalls, the main route to instability there. In the US both matter. Moreover, fixed and floating rates shift risk differently. With floating rates, the risk lands on households, but central banks may be able to ease the pressure quickly. With fixed rates the financial system carries more of the risk, and policy rate cuts don’t help households that cannot refinance. This explains why the US crisis in 2009-12 was worse than the UK's. The inflation and interest rate shocks of 2022–24 reinforced the importance of this distinction.
Land is highly relevant for financial stability. In recent years around 70% of the average UK home’s value has been in land, the highest share in the G7. Land values are far more volatile than building costs, so the UK’s mix of regressive property taxes, tight planning, weak infrastructure and a shrinking social stock builds fragility into the system. Most macroprudential writing treats housing supply and tax policy as background. They are risk factors in their own right, especially when the policy focus is not merely to ensure the survival of the big banks but also to protect vulnerable households.
A broader case for intervention
The usual economic justification for macroprudential policy is the presence of externalities. For Korinek and Simsek (2016) there is a demand externality when interest rates are stuck at the lower bound. Their model is too narrow however: it omits real estate and the function of debt is only to smooth consumption. Non-linear dynamics and multiple equilibria visible in the 2008 crisis support a more general case. And there are two further grounds: distorted incentives in finance, and the protection of poorly informed consumers. On the last, many people buy near the top of asset price cycles and sell near the bottom, so lowering house price volatility improves the information they decide on and benefits the less sophisticated most.
It is also important to distinguish lender resilience from borrower resilience. A "classic credit cycle" with defaults, repossessions and job losses can occur without a banking crisis, especially where banks are profitable and well capitalised. A bank-focused framework could therefore declare success while households suffer. Tough micro-prudential supervision of lenders lets macroprudential policy put more weight on protecting households.
Debt-service limits: the French lesson
The CGFS study (2023) argues that income-based limits make better automatic stabilisers than LTV limits, because incomes are far less cyclical than house prices. But a DSTI limit is a good stabiliser only if computed at a fixed interest rate. At the current mortgage rate, it can work against stability: when rates fall, people can borrow more, which pushes up house prices and debt.
France is an example. Mortgages there are mostly long-term fixed rate, so DSTI limits based on current rates are common. For many years, lenders applied a conventional ceiling of 33% on DSTIs. My earlier work with Valerie Chauvin (2018) finds very large responses of housing debt and house prices to falling nominal rates. Falling nominal rates combined with credit liberalisation such as longer loan terms explain much of the rise in real French house prices. In floating-rate countries like Sweden and the UK, regulators use a fixed precautionary stress rate, which neither loosens when rates fall nor tightens when they rise.
The link between high DSTI at origination and default needs qualifying. It is strong in floating or short-fix markets, where rate rises hit existing borrowers. Drehmann and Juselius (2013) find that the aggregate debt service ratio forecasts crises, but do not separate fixed from floating regimes.
Better diagnostics: latent variables and extrapolative expectations
I am sceptical of the standard risk dashboards, such as the European Systemic Risk Board’s. Credit growth can reflect sustainable expansion, and a high price-to-income ratio can reflect lower interest rates or lasting supply constraints. Without a way to separate benign from non-benign causes, the signals mislead.
A structural alternative is needed. Most house price models omit long lags, expected income growth and changing loan standards. The standard VAR uses the same short lag length for all variables, so a four-year memory of past price rises, as US evidence suggests, would be totally missed. The answer is to estimate a house price equation and a mortgage debt equation jointly with a latent index of mortgage credit conditions, identified using bank lending surveys, loan duration and LTV/LTI data, and changes in policy settings. This approach forecast French non-performing loan ratios one to three years ahead. It also has the potential to test what macroprudential measures actually do to lending standards and builds in extrapolative expectations, which many equilibrium models assume away.
Why the empirical evidence probably understates effectiveness
Reviews such as Biljanovska et al. (2023) conclude that borrower-based measures curb credit and house price growth, with mild short-term growth costs and stronger resilience over the medium term. Araujo et al. (2020) find that tighter LTV limits curb leverage and credit growth, especially where house prices are volatile, with more variable effects on prices and some leakage to non-banks. Combinations of tools work best, to limit lenders offseting LTV caps by relaxing DSTI or LTI limits. Even this probably understates the case, for several reasons:
- Policies are enacted when credit and prices are already booming, and optimism about income drives both. Omitting income expectations biases effectiveness downward.
- Researchers rarely know what voluntary standards banks followed before a limit arrived. A new cap close to existing practice looks ineffective for good reason.
- Many panels use 0/1 policy dummies, which hide intensity.
- Panel studies use country fixed effects as the only heterogeneity, assuming the same slopes everywhere. This is an implausible assumption, since the effect of an LTV or DSTI limit depends on the rate regime, recourse rules and the kind of shock.
- Tightening may work better than loosening, because after a downturn banks’ own caution overrides looser rules.
I am therefore sympathetic to the CGFS (2023) qualitative study of 14 countries, which draws on practitioners’ experience. Its lessons include early action, since inaction bias is a perennial problem, guardrails that need no frequent adjustment, a clear legal mandate and operational independence for the authority, and candid public communication about costs.
Distributional effects and the case for general equilibrium
Micro studies in Ireland, Norway, Portugal, South Korea, Israel and the UK generally find that borrower-based limits hurt lower-income households and first-time buyers in the short run. In Korea, tighter LTV caps coincided with a steep rise in prices, so the poorest households lost the chance to build wealth. A cross-country study (Frost and van Stralen 2018) associates LTV and DTI limits with higher income inequality. A 2014 UK limit on high-LTI lending reduced credit to low-income borrowers and slowed house price growth, but defaults were lower when prices later corrected (Peydró et al. 2024), and limits can dampen inequality in downturns (Georgescu and Vila Martín 2021). Authorities also soften the effects with flexibility margins on first-time buyers but tighter limits on buy-to-let (CGFS 2023). The short-run costs are not in dispute, but these studies capture only first-round effects, and the inequality finding is could reflect reverse causation: countries with credit booms use the tools, and booms and the recessions that follow widen inequality.
My claim is that conclusions drawn from first-round effects may be reversed in general equilibrium. Many people, including politicians, do not understand system dynamics. Tighter limits could lower house prices, debt and gearing and reduce price overshooting, which would help access to housing in the medium run. The mortgage interest tax deduction is an example: many assume it helps first-time buyers, but Sommer and Sullivan (2018) find that eliminating it lowers prices, raises homeownership and improves welfare. In the UK, owner-occupation among 30–34-year-olds fell after credit was eased from 1997 to 2007 and rose slightly after 2014, when borrower-based limits tightened.
A critique of DSGE models and the case for agent-based modelling
Dynamic stochastic general equilibrium models are the usual tool for welfare questions, but most recent versions deal inadequately with housing. They use two representative agents (saver and borrower), assume house prices clear the market instantly, include no extrapolative expectations, make households infinitely lived (ruling out inheritance and generational effects), and assume arbitrary and extreme preference differences between the two groups. Garriga and Hedlund’s (2020) heterogeneous-agent model with search frictions is a better direction.
More investment is needed into agent-based modelling (ABM), which simulates fully heterogeneous households, firms and banks from micro-data. The Polish model of Kaszowska-Mojsa et al. (2024) suggests that LTV limits adopted after 2014 widened income inequality over the medium run, and that combinations of limits can be designed to reduce this. ABMs have weaknesses, such as rules of behaviour that can seem overly simple, and simulations must be checked against macro data, but have great potential.
A formal objective function and growth-at-risk
A formalisation of the policy maker’s objective function has been suggested. I agree with Chavleishvili et al. (2021) that leaning against the cycle has net benefits even at zero risk aversion, but omitting an inflation term from the objective is a problem. An authority housed in a central bank cannot ignore monetary policy, so the objective should combine average growth, downside risk and inflation, and be tested for robustness to the weights. An important advance has been the development of the Growth-at-risk (GaR) framework which avoids arguing over what counts as a crisis. A simple target, as Aikman (2022) proposes, would limit the probability of a GDP fall of more than 5% over three years to a set maximum. A greater weight on borrower resilience would also shift attention to residential rather than commercial property, with distributional consequences.
The real conclusion: do less with macroprudential tools
My final argument will be controversial to some, though it is supported by CGFS (2023). It is that macroprudential tools are the second or third best option when markets are deeply dysfunctional, and the regulator should not carry the whole burden. Because credit restrictions can widen inequality in the short run, valid concerns about inequality are better met by tax reform. The list includes:
- Removing or limiting mortgage interest deductions, which raise leverage and are regressive.
- Proportional or progressive property taxes linked to current values.
- Split-rate taxes that tax land more heavily than buildings, which stabilise prices and address inequality.
- Planning reform and land value capture to fund infrastructure and social housing.
- Cutting transfer taxes such as stamp duty, to improve flexibility.
- Tougher conduct regulation to limit predatory lending and mis-selling, and arrears protocols that avoid rapid foreclosures, which amplify downturns (Guren and McQuade 2020).
If these were done, macroprudential authorities would need to restrict credit less.
Professor John Muellbauer is a Senior Research Fellow of Nuffield College and a Senior Fellow of the Institute for New Economic Thinking at the Oxford Martin School. He is a Fellow of the British Academy, of the Econometric Society and a CEPR Fellow. He is co-creator with Angus Deaton of the Almost Ideal Demand System (AIDS) in 1980, ranked among the top twenty papers in the first century of the American Economic Review. He is best known for his work on household economics, housing markets and on finance-real economy interactions. He has worked with many central banks and the OECD and is a frequent contributor to CEPR’s VoxEU columns. Details of his contributions can be found on his CEPR profile. He is heavily involved in policy debates. He regularly presents on housing market reform in the UK, including in the British Academy’s 2025 Keynes Lecture, and has proposed a Green Land Value Tax, a form of property taxation with global relevance.