## 1. Residential Real Estate and Financial and Macroeconomic Stability

## Source details

**Canonical URL:** [1. Residential Real Estate and Financial and Macroeconomic Stability](https://www.imf.org/-/media/files/publications/selected-issues-papers/2023/english/sipea2023060.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/selected-issues-papers/2023/english/sipea2023060.pdf.md)
- [Structured JSON version](/-/media/files/publications/selected-issues-papers/2023/english/sipea2023060.pdf.json)

---

### Introduction and purpose
- Purpose: Advance discussion of borrower-based macroprudential policy in Germany by explaining how borrower-based measures could strengthen financial stability, macroeconomic stability, and consumer protection; address concerns; offer approaches to initial calibrations; and hint at likely effects based on other countries’ experiences.
- Method: Microsimulation model to quantify effects of activating borrower-based measures under a housing market stress scenario starting in 2022Q2.
- Key simulation finding: Activating LTV and DSTI limits imposed in 2015 would have prevented losses in CET1 bank capital under the stress scenario of almost 6 percent, or €26 billion.
- Comparative metric: The €26 billion amount is similar to the €23 billion that was locked in through the introduction of CCyB and SSRB capital buffers in 2022.
- Caveat: Estimates are subject to significant model and scenario uncertainty.

### Germany’s current toolkit and identified gap
- Existing measures: risk weights, countercyclical capital buffer (CCyB), sectoral systemic risk buffer (SSRB), capital buffers for SIFIs, restrictions on amount of a bank’s new lending that can exceed LTV thresholds, and minimum amortization rates.
- Missing element: income-related borrower-based measures (e.g., limits on debt service-to-income (DSTI), debt-to-income (DTI), or loan-to-income (LTI) ratios).
- Legislative status: Legislation drafted; to be presented to parliament when the political environment is opportune.
- Recommendations: Germany’s Financial Stability Committee, IMF staff, and the ESRB have recommended adding income-related measures.

### Rationale for borrower-based instruments
- Definition and metrics:
  - DSTI: ratio of (typically monthly) interest and principal repayments on the new mortgage and all other debts of the borrower relative to her income.
  - DTI: value of the mortgage loan and outstanding balances of all other debts of the borrower relative to her income.
  - LTI: value of only the mortgage loan relative to the income of the borrower.
- Complementarity with lender-based tools:
  1. Target risky lending practices in ways that capital measures cannot.
  2. Bank capital might not be fully usable in a crisis due to overlapping requirements including the leverage ratio and MREL.
  3. Higher bank capital requirements tend to have a limited effect on credit growth when accompanied by increasing asset prices.
  4. Excessive borrower leverage can fuel unsustainable construction activity, real estate prices, and/or consumption during a boom that reverses during a downturn.
  5. Consumers need protection from overindebtedness even if banks could withstand associated credit losses.
- Differences vs LTV caps:
  - DSTI/DTI/LTI caps bind more tightly when house prices grow faster than incomes.
  - DSTI/DTI/LTI caps reduce likelihood borrowers are forced into default; LTV caps increase recoverability via property sale.
  - Can prevent leakages that bypass LTV caps (e.g., unsecured credit to cover down-payments).

### Simulation design (housing market “crash” counterfactual)
- Objective: Quantify how much less banks’ capital would decline under a one-year housing market stress scenario starting in 2022Q2 if LTV and DSTI limits had been imposed in 2015.
- Microsimulation setup:
  - Simulation starts in 1980; randomly draw 1,000 newly issued loans each quarter between 1982Q2 and 2022Q2 → 161,000 loans.
  - Loan characteristics drawn with replacement from HFCS; median loan maturity 10 years; maturity range 2–40 years; about 14 percent of loans are variable-rate.
  - LTVs and DSTIs drawn from Europace and Interhyp distributions; LTVs adjusted down by 7.5 ppts to convert mortgage lending values to market values.
  - Every loan fully amortizing with monthly instalments; loans issued at national-aggregate interest rates and collateralized by national house prices.
  - Mortgage credit growth is exogenous in model; no behavioral feedback on aggregate mortgage volumes.
- Counterfactual calibration (post-2015Q2, phased over 2 years):
  - LTVs between 90 and 100 percent reduce to 10 percent of new lending over 2 years.
  - LTVs above 100 percent phase out over 2 years.
  - New lending at DSTIs above 40 percent drops from 10 percent to 5 percent over 2 years and then stays there.
- Comparison assumption: no new lending (existing loans run off), so no difference in credit growth between scenarios.

### Stress and reference scenarios (2022Q3–2023Q2, exact shocks)
- Stress scenario shocks relative to reference:
  - Nominal house prices fall 40 percent.
  - Unemployment increases 2.2 percentage points (three-standard deviation increase).
  - Mortgage rates increase 4.3 percentage points.
  - Nominal disposable income reduced by 4.3 percent.
  - Liquid assets fall 5 percent.
  - Risk-free rates increase 2 percentage points at all maturities.
- Selected scenario table entries (preserved exactly):
  - Reference scenario:
    - 2022Q2: unemployment 5.10, house prices 1.00, risk free rate 0.00, mortgage rate 0.00, disposable income 0.00, liquid assets 1.00
    - 2023Q2: unemployment 5.10, house prices 1.03, risk free rate 0.10, mortgage rate 0.00, disposable income 4.41, liquid assets 1.00
  - Stress scenario:
    - 2022Q2: unemployment 5.10, house prices 1.00, risk free rate 0.00, mortgage rate 0.00, disposable income 0.00, liquid assets 1.00
    - 2022Q3: unemployment 5.59, house prices 0.88, risk free rate 0.50, mortgage rate 1.08, disposable income -1.09, liquid assets 0.99
    - 2022Q4: unemployment 6.12, house prices 0.77, risk free rate 1.00, mortgage rate 2.15, disposable income -2.19, liquid assets 0.98
    - 2023Q1: unemployment 6.70, house prices 0.68, risk free rate 1.50, mortgage rate 3.23, disposable income -3.28, liquid assets 0.96
    - 2023Q2: unemployment 7.34, house prices 0.60, risk free rate 2.00, mortgage rate 4.30, disposable income -4.38, liquid assets 0.95

### Simulation results (cumulative over 4 quarters; losses in stress scenario in excess of reference)
- Aggregate table values (exact numbers preserved):
  - Loss (€ bn): without BBMs 51.7; with BBMs 30.4; difference -21.3
    - Standard errors: (0.33), (0.44), (0.43)
  - Loss (% CET1): without BBMs 11.9; with BBMs 7.0; difference -4.9
    - Standard errors: (0.08), (0.1), (0.1)
  - Loss (% RWA): without BBMs 1.9; with BBMs 1.1; difference -0.8
    - Standard errors: (0.01), (0.02), (0.02)
  - % increase RWA: without BBMs 3.2; with BBMs 2.2; difference -1.0
    - Standard errors: (0.03), (0.02), (0.02)
  - =(2)+(4) % change CET1: without BBMs 15.1; with BBMs 9.2; difference -5.9
    - Standard errors: (0.1), (0.12), (0.1)
- Key numeric interpretations:
  - Borrower-based measures preserve almost 6 percent of CET1 capital, or €26 billion (5.9 percent of CET1).
  - Expected credit losses are €21 billion higher, or 4.9 percent of CET1, without the LTV/DSTI limits.
  - Prevention of increasing risk-weighted assets preserves a further 1 percent of CET1 capital.
- Distributional findings by LTV bucket:
  - Most of the €21 billion in prevented losses originate among loans with LTVs between 70 and 100 percent (of market value) in 2022Q2.
  - Examples: 7 percent of losses prevented in the 80–90 percent LTV bucket; 24 percent prevented in the 90–100 percent LTV bucket.
- Uncertainty: Standard errors across ten replications are small, but model and scenario uncertainty are likely substantial.

### Interpretation, limitations, and costs
- Possible understatement of benefits:
  - PD data limitations; assuming BBMs introduced in 2015 and run for seven years before crash may understate benefits if more time allowed.
  - Simulation does not capture ex ante price-moderation effects of BBMs or reduced household leverage’s effect on consumption contraction.
- Costs not analyzed:
  - Foregone bank profits and other welfare costs not quantified; comparison of BBMs versus lender-based measures requires assessing costs alongside benefits.

### Illustrative calibration approaches
- LTV calibration by negative-equity protection:
  - 2-standard deviation 1-year price drop = 8 percent → maximum LTV = 92 percent.
  - 2-standard deviation 3-year drop = 22 percent → maximum LTV = 78 percent.
  - 5th percentile of three-year house price changes in 2022 = 14 percent → maximum LTV = 86 percent.
- DSTI calibration by protecting essential consumption:
  - Essential spending (broad): averages 54 percent of income → leaves at most 46 percent for debt-servicing and non-essential spending.
  - Essential spending (narrow): averages 42 percent → leaves at most 58 percent.
  - 2-standard deviation household income falls:
    - 1 year: 3.5 percent → DSTI ≤ 45 percent (broad) or DSTI ≤ 56 percent (narrow).
    - 3 years: 4.6 percent → DSTI ≤ 44 percent (broad) or DSTI ≤ 56 percent (narrow).
    - 10 years: 9.5 percent → DSTI ≤ 42 percent (broad) or DSTI ≤ 53 percent (narrow).

### Empirical evidence and international case studies (summary)
- Mixed evidence on effects on credit and house prices; BBMs tend to have weaker effects in AEs and stronger effects when first introduced.
- Key empirical findings:
  - Europace: almost half of new mortgages over two decades issued with LTV above 80 percent; post-2018 more loans above 100 percent.
  - Bundesbank: over past decade, >10 percent of loans issued at DSTI above 40 percent.
  - Cross-country: in 2022Q1, Germany had second-highest share of new loans with LTV above 100 percent among EU countries.
- Case study highlights:
  - Ireland (2015): CBI estimated house prices may have been 13–25 percent higher in 2019 without mortgage measures; evidence of loan bunching below thresholds.
  - Netherlands (2013 onward): average LTV on new loans fell from ~103 percent in 2011 to ~90 percent by 2018; homeownership among first-time buyers estimated to fall by 6 percent due to LTV cap.
  - New Zealand (2013): share of outstanding loans with LTV above 80 percent fell from 21 percent to 11 percent by 2016; RBNZ estimated re-introducing LTV caps would reduce house price inflation by 1–4 percentage points.
  - Austria (June 2022): LTV cap 90 percent, DSTI cap 40 percent, maximum loan term 35 years, 20 percent speed limit.
  - Denmark (2015): LTV cap 95 percent reduced share of new borrowers above 95 percent LTV.
- General conclusions from literature:
  - Activating BBMs tends to increase resilience: lower house price volatility, lower mortgage defaults, lower sensitivity of consumption to house prices.
  - Evidence on effects on aggregate output is limited; some findings suggest temporary reductions in consumer prices.

### Timing, implementation, and policy recommendations for Germany
- Timing:
  - Not ideal to activate BBMs mid-downturn; current property-market weakness suggests waiting.
  - Early activation is important because BBMs affect new lending only and take years to change outstanding loan composition.
- Recommendation:
  - Add income-related borrower-based measures to complete the macroprudential toolkit; these would complement existing lender- and borrower-based measures.
  - Activate BBMs as the economy recovers, well in advance of the next housing market downturn.
  - Consider phased approaches (e.g., start with quantitative guidance on LTV limits that are not legally binding).
- Further work:
  - More detailed impact assessments and calibration work are needed, including whether to calibrate cyclically or at a constant level throughout the cycle.

### Annex I — technical model highlights (parameters and estimation)
- Default logic: dual-trigger — borrower defaults if both an income-based distress trigger and a negative-equity trigger occur.
- Expected loss = PD × LGD; PD and LGD extend Valderrama (2022) and related work.
- Key parameter estimates (Germany; exact values preserved):
  - D = 0.168; Standard error = 1.312; t-statistic = 0.13
  - β1 = 1.017; Standard error = n.a.; t-statistic = n.a.
  - β2 = 0.014; Standard error = 1.325; t-statistic = 0.01
  - β3 = 0.075; Standard error = n.a.; t-statistic = n.a.
  - α = 2.080; Standard error = n.a.; t-statistic = n.a.
  - c = 0.045; Standard error = 1.003; t-statistic = 0.04
  - δ0 = −3.894; Standard error = 4.236; t-statistic = −0.92
  - δ1 = 0.106; Standard error = 0.119; t-statistic = 0.89
  - σ = 32,597; Standard error = 458,942; t-statistic = 0.07
- Estimation method: nonlinear least squares matching aggregate COREP PDs and LGDs; more weight on PDs; constrained and unconstrained parameter sets noted.
- Limitations: wide standard errors for several parameters; PDs in COREP are long-run averages (insensitive to short-term conditions), which may understate losses in stress scenarios and the benefits of BBMs.

*Source: IMF selected issues paper chapter "Residential Real Estate and Financial and Macroeconomic Stability" (sipea2023060).*

### 1. Residential Real Estate and Financial and Macroeconomic Stability __________________ 7

### 1. Residential Real Estate and Financial and Macroeconomic Stability

### Introduction
- Purpose: Advance discussion of borrower-based macroprudential policy in Germany by explaining how borrower-based measures could strengthen financial stability, macroeconomic stability, and consumer protection; addressing concerns; offering approaches to initial calibrations; and hinting at likely effects based on other countries’ experiences.
- Methods: Uses a microsimulation model to quantify effects of activating borrower-based measures under a housing market stress scenario starting in 2022Q2.
- Key simulation finding: Activating LTV and DSTI limits imposed in 2015 would have prevented losses in CET1 bank capital under the stress scenario of almost 6 percent, or €26 billion.
- Comparative metric: The €26 billion amount is similar to the €23 billion that was locked in through the introduction of CCyB and SSRB capital buffers in 2022.
- Caveat: Estimates are subject to significant model and scenario uncertainty.

### Germany’s Current Toolkit and the Missing Piece
- Existing measures: risk weights, countercyclical capital buffer (CCyB), sectoral systemic risk buffer (SSRB), capital buffers for SIFIs, restrictions on amount of a bank’s new lending that can exceed LTV thresholds, and minimum amortization rates (see Box 1).
- Missing element: income-related borrower-based measures (e.g., limits on debt service-to-income (DSTI), debt-to-income (DTI), or loan-to-income (LTI) ratios).
- Legislative status: Legislation has been drafted and will be presented to parliament when the political environment is opportune.
- Recommendation provenance: Germany’s Financial Stability Committee, IMF staff, and the ESRB have recommended adding income-related measures.

### What Income-Related Instruments Are and How They Work
- Definition: Instruments that restrict the ability of banks to lend large sums to homebuyers relative to their incomes by limiting the share of new lending that can exceed thresholds linked to borrower income over a specific period (e.g., three months).
- Typical metrics:
  - DSTI: ratio of (typically monthly) interest and principal repayments on the new mortgage and all other debts of the borrower relative to her income.
  - DTI: value of the mortgage loan and outstanding balances of all other debts of the borrower relative to her income.
  - LTI: value of only the mortgage loan relative to the income of the borrower.
- Example from another country: In Austria, only 10 percent of a bank’s newly issued loans over a six-month period are allowed to exceed a 40 percent DSTI ratio.

### Rationale for Adding Income-Related Instruments
- Complementarity: Borrower-based tools complement lender-based tools for at least five reasons:
  1. They can target risky lending practices in ways that capital measures cannot, reducing spillovers to other lending and potentially achieving resilience with lower capital overall.
  2. Bank capital might not be fully usable in a crisis due to overlapping requirements including the leverage ratio and MREL (IMF, 2022b).
  3. Higher bank capital requirements tend to have a limited effect on credit growth when accompanied by increasing asset prices (IMF, 2013).
  4. Excessive borrower leverage can fuel unsustainable construction activity, real estate prices, and/or consumption during a boom that reverses during a downturn, amplifying economic volatility even if borrowers do not default.
  5. Consumers need protection from overindebtedness, even if banks could withstand associated credit losses (consumer protection motivation).
- Differences vs LTV caps:
  - DSTI/DTI/LTI caps do not relax as house prices rise; they bind more tightly when house prices grow faster than incomes, guarding against price-credit feedback loops.
  - DSTI/DTI/LTI caps reduce the likelihood borrowers are forced into default, whereas LTV caps increase the likelihood banks can recover loans via property sale.
  - DSTI/DTI/LTI caps can prevent leakages when LTV caps or amortization requirements are activated (e.g., preventing provision of additional unsecured credit to bypass LTV caps or replacing high-LTV loans with high-DSTI/DTI/LTI loans).
- Macro area relevance: Completing the toolkit is important because Germany is part of a currency union; national macroprudential tools compensate for centralized euro-area monetary policy and help avoid regulatory arbitrage across borders.
- Reciprocity: If another EU country imposes income-related borrower-based measures, Germany may need similar tools to comply with ESRB reciprocity requests concerning cross-border lending.

### Simulation and Policy Implications
- Simulation setup: Microsimulation to quantify how much less banks’ capital would decline under a one-year housing market stress scenario starting in 2022Q2 if LTV and DSTI limits had been imposed in 2015.
- Stress scenario logic: Benefits of borrower-based measures are most evident in scenarios with significant house price declines because falling house prices (i) increase probability of borrower default via negative equity and (ii) reduce bank recovery values on foreclosure.
- Quantitative result: Prevented CET1 capital loss of almost 6 percent, or €26 billion.
- Policy interpretation: Borrower-based measures can bolster financial stability and play a helpful role alongside capital-based measures in the macroprudential mix.
- Implementation note: Section F offers simple approaches to calibrating borrower-based measures in Germany; these are first steps requiring further analysis and refinement.
- Empirical context: Section G summarizes evidence from advanced economies and case studies of borrower-based measures’ effects.

### Implementation Considerations and Next Steps
- Addressing concerns: Sections C and D revisit arguments for activation and critically review potential concerns associated with activation, explaining how some concerns can be addressed or where further work is needed.
- Calibration: Section F presents simple and intuitive approaches to calibrating borrower-based measures in Germany; further work is necessary.
- Evidence base: Section G provides summary evidence and case studies from advanced economies to inform policy choices.

*Source: IMF selected issues paper chapter "Residential Real Estate and Financial and Macroeconomic Stability" (June 28, 2023).*

### 12.      There are financial stability, macroeconomic stability, and consumer protection

### 12.      There are financial stability, macroeconomic stability, and consumer protection arguments for activating borrower-based measures in Germany as the economy recovers

### Rationale for borrower-based measures
- Financial stability argument:
  - Mortgages make up 70 percent of household liabilities and 50 percent of domestic bank loans.
  - Household debt is 100 percent of income.
  - The Basel credit gap stood at 5 percent of GDP in 2022Q2.
  - Credit losses on residential real estate loans in a downside macroeconomic scenario are driven by high-LTV mortgages (Barasinska et al., 2019).
- Macroeconomic stability argument:
  - Borrower-based measures could limit amplification of housing market booms and busts and associated cyclicality in consumption and construction, even if defaults remain low.
  - The correlation between the growth rates of house prices and consumption is 27 percent.
  - Since Germany has full-recourse mortgage lending, consumption may be extra sensitive to house prices in a downturn.
- Consumer protection argument:
  - Loans with both high LTV and high DSTI pose hardship risk to borrowers if they cannot continue servicing mortgages, even where banks can foreclose.

### Current valuation and market vulnerabilities
- House price overvaluation and recent movements:
  - The Bundesbank estimates that house prices in cities were overvalued by 25–40 percent in 2022.
  - House prices are some 20 and 40 percent above historical average price-to-income and price-to-rent ratios, respectively.
  - Tightening monetary policy in 2022 and 2023 helped reduce some overvaluation, with house prices falling 9 percent between 2022Q2 and 2023Q1, but most overvaluation remains.
- Mortgage insurance and coverage:
  - Residual debt insurance (Restschuldversicherungen) covered €12.7 billion as at end-2021.
  - BaFin (2017) estimated the total sum insured to be €13.3 billion in 2015, which amounts to 1.1 percent of mortgage loans or 0.4 percent of GDP.
  - BCBS (2013) estimated mortgage insurance coverage at 10–25 percent of loans.
- Structural importance:
  - Mortgage credit growth accelerated between the global financial crisis and 2021Q3; credit growth decelerated but remained positive in 2022.
  - Large segments of new mortgage lending with loose lending standards have persisted since at least 2018.

### Evidence on lending standards and cross-country comparisons
- High-LTV and high-DSTI prevalence:
  - Europace data show that over the past two decades, almost half of new mortgages have been issued with an LTV ratio above 80 percent.
  - After 2018, more loans began to be issued at LTVs above 100 percent.
  - Bundesbank (2022) data indicate that over the past decade, more than 10 percent of loans have been issued at DSTI ratios above 40 percent.
  - A BaFin survey in summer 2022 found that 15 percent of new borrowers spend more than 50 percent of their net income on loan instalments.
- Cross-country position:
  - In 2022Q1, Germany had the second-highest share of new loans with LTV ratios above 100 percent among EU countries.
  - In 2018, Germany had the sixth-highest share of new loans with LTV ratios above 80 percent and the fourth-highest share of new loans with LTI ratios above 5.
- Data and measurement caveats:
  - Germany typically uses the conservative “mortgage lending value” concept (Beleihungswert) to calculate LTVs; this tends to produce higher LTV ratios than market-value-based calculations.
  - New WIFSta data will provide LTV ratios on a market value basis from 2023Q1 onward.
  - DSTI and LTI ratios are high in Germany as well and are not affected by the mortgage lending value.

### Concerns about activating borrower-based measures and counterarguments
- Concern #1: Existing capital buffers (CCyB and SSRB) suffice.
  - Counterpoints:
    - Germany’s bank capitalization is at the 38th percentile across G7 and advanced European countries.
    - Actual bank capital levels in Germany are lower than in countries with high shares of high-LTV loans or with LTV caps.
    - The policy question is about the optimal macroprudential mix; capital measures may have unintended consequences (e.g., increased corporate loan rates, diversion to NBFIs).
    - Borrower-based measures provide macroeconomic stability and consumer protection benefits not delivered by CCyB and SSRB.
- Concern #2: Conservative mortgage lending value inflates LTV comparisons.
  - Counterpoints:
    - Other countries also use conservative valuation standards (e.g., Austria, Czechia, Poland, Slovenia).
    - Several of those countries have nevertheless activated LTV caps.
    - DSTI and LTI ratios remain high in Germany regardless of valuation standard.
- Concern #3: Borrowers are high-net worth and thus low risk.
  - Evidence:
    - Households with an outstanding mortgage report median net worth €218,400; all German households median net worth €70,800.
    - In the euro area, medians are €156,600 (mortgage-holders) and €99,500 (all households).
    - High net worth does not imply low risk if leverage (DSTI, LTI) is high.
    - Policy design can accommodate exceptions (e.g., speed limits, pledging additional collateral).
- Concern #4: Full recourse lending limits default risk.
  - Counterpoints:
    - Full recourse reduces but does not eliminate defaults; income or liquidity shocks drive many defaults.
    - Countries with full recourse have experienced real estate-fueled financial crises (e.g., Nordic crisis, Spain).
    - Housing crashes can materially increase banks’ provisions and risk-weighted assets even with full recourse.
    - Consumer protection rationale is strongest under full recourse.
- Concern #5: Measures reduce affordability or disproportionately affect younger/lower-income borrowers.
  - Evidence and mitigation:
    - LTV limits in the Netherlands reduced homeownership among first-time buyers by 6 percent (Biesenbeek et al., 2022).
    - Impacts can be temporary as buyers accumulate larger down-payments.
    - Borrower-based measures can reduce house prices and thus improve affordability.
    - Design options to protect vulnerable groups include looser restrictions for first-time buyers, speed limits, de minimis exemptions, and tighter restrictions on speculative activity.

### Timing and policy implementation
- Cyclical considerations:
  - From a cyclical perspective, activating borrower-based measures is not ideal mid-dowturn because the real estate market is entering a downturn and lender-based measures were recently tightened.
  - A new upward cycle in house prices and mortgage lending could begin within the next few years.
- Importance of early activation:
  - Borrower-based measures only affect new lending and thus take years to materially alter outstanding loan composition.
  - Early activation is especially important when introducing these measures for the first time because:
    - Measures may need to be phased in gradually to reduce disruptions.
    - Learning-by-doing may be needed to optimize calibration.
  - Legislation to strengthen Germany’s toolkit should be passed soon to allow time for implementing regulations, monitoring mechanisms, and consultations.

### Microsimulation evidence and methodology (housing market “crash” counterfactual)
- Objective:
  - Quantify benefits of LTV and DSTI limits by simulating a one-year housing market crash scenario starting in 2022Q2 and comparing bank capital outcomes if limits had been imposed and fully phased-in in 2015.
- Simulation setup:
  - The simulation starts in 1980.
  - Randomly draw 1,000 newly issued loans each quarter between 1982Q2 and 2022Q2.
    - This procedure results in 161,000 loans issued over 161 quarters.
  - For each new loan, draw characteristics (outstanding principal balance, maturity, fixed vs. variable rate status, liquid assets) with replacement from all three waves of the ECB’s Household Finance and Consumption Survey (HFCS).
    - Outliers removed by filtering households with the largest 1 percent of loan balances (both in euros and relative to loan maturity), and those with the largest or smallest half a percent of loan maturities, liquid assets, or financial wealth.
    - Median loan maturity is 10 years; range of maturities is 2–40 years.
    - About 14 percent of loans are variable-rate.
    - The correlation between the natural logarithm of the loan balance and the natural logarithm of liquid assets is 14 percent.
    - The correlation between loan maturity in years and the natural logarithm of liquid assets is -13 percent.
  - Each new loan gets an LTV and DSTI ratio drawn from the distributions of Europace and Interhyp data, respectively.
    - Within each bucket, assume uniform distribution (no bunching).
    - Adjust LTVs down by 7.5 ppts to convert mortgage lending values to market values (Barasinska et al., 2019 estimate an LTV of 80 percent based on mortgage lending value corresponds to 70–75 percent on market value).
  - Additional assumptions:
    - Every loan is fully amortizing, repayable in monthly instalments.
    - Loans are issued at national-aggregate interest rates and collateralized by a house whose price follows national house prices.
  - Model scope:
    - The loan book grows initially but stabilizes after about 20 years because new loans roughly offset maturing loans.
    - Mortgage credit growth is exogenous in this model and unaffected by the macroeconomic scenario; the model does not describe behavior of overall mortgage credit in Germany.
- Analysis approach:
  - Trace each loan’s characteristics (principal outstanding, interest rate, LTV, DSTI) over time as it amortizes and as house prices and interest rates evolve.
  - Apply the same housing market crash scenario to portfolios with and without borrower-based measures and compare impacts on bank capital relative to a reference business-as-usual scenario.

*Source: IMF selected issues paper section 12 (sipea2023060).*

### 28.      Each loan’s creditworthiness at each point in time is summarized by its expected loss,

### 28.      Each loan’s creditworthiness at each point in time is summarized by its expected loss,

### Modeling expected loss and simulation setup
- Expected loss for each loan = probability of default (PD) × loss given default (LGD).
- PD and LGD are modeled by extending Valderrama (2022), Górnicka and Valderrama (2020), and Harrison and Mathew (2008); specific equations for PD and LGD are given in Annex I.
- New-loan simulation (paragraphs 27–28) is repeated under a counterfactual where LTV and DSTI limits were imposed in 2015:
  - Same loans issued in outstanding balance, maturity and interest rate.
  - LTV and DSTI ratios drawn from distributions with lower LTV and DSTI ratios; within-bucket distributions assumed uniform (no bunching).
  - Calibration of counterfactual distributions:
    - After 2015Q2, LTVs between 90 and 100 percent (of the mortgage lending value) reduce to 10 percent of new lending over 2 years.
    - LTVs above 100 percent phase out over 2 years.
    - New lending at DSTIs above 40 percent drops from 10 percent to 5 percent over 2 years and then stays there.
  - Scenarios amount to imposing borrower-based measures with speed limits.
- Assumption for scenario comparison: no new lending (existing loans run off), so no difference in credit growth between scenarios.

### Stress scenario versus reference scenario (scenario horizon: 4 quarters, 2022Q3–2023Q2)
- Stress scenario captures a housing market crash accompanied by tighter monetary policy and rising inflation; reference scenario represents benign conditions.
- Stress scenario shocks applied (effects described relative to reference scenario):
  - Nominal house prices fall 40 percent.
  - Unemployment increases 2.2 percentage points (three-standard deviation increase).
  - Mortgage rates increase 4.3 percentage points (twice the increase observed so far in the latest tightening cycle).
  - Nominal disposable income reduced by 4.3 percent (to capture rising living costs).
  - Liquid assets fall 5 percent.
  - Risk-free rates increase 2 percentage points at all maturities.
  - Higher mortgage rates increase monthly debt servicing and DSTI on variable-rate loans.
  - Higher risk-free rates reduce household prepayment penalties in the model.
- Reference scenario cumulative movements over the horizon:
  - House prices increase cumulative 3.3 percent.
  - (Registered) unemployment stable at 5.1 percent (2022Q2 level).
  - Mortgage rates stable.
  - Nominal disposable income increases 4.5 percent.
  - Liquid assets and risk-free rates stable.

### Scenario values from Table 2 (selected quarterly entries, exact values preserved)
- Reference scenario (columns: unemployment (percent), house prices (index), risk free rate (cumulative change, ppts), mortgage rate (cumulative change, ppts), disposable income (cumulative change, percent), liquid assets (index)):
  - 2022Q2: 5.10, 1.00, 0.00, 0.00, 0.00, 1.00
  - 2023Q2: 5.10, 1.03, 0.10, 0.00, 4.41, 1.00
- Stress scenario (same ordering):
  - 2022Q2: 5.10, 1.00, 0.00, 0.00, 0.00, 1.00
  - 2022Q3: 5.59, 0.88, 0.50, 1.08, -1.09, 0.99
  - 2022Q4: 6.12, 0.77, 1.00, 2.15, -2.19, 0.98
  - 2023Q1: 6.70, 0.68, 1.50, 3.23, -3.28, 0.96
  - 2023Q2: 7.34, 0.60, 2.00, 4.30, -4.38, 0.95

### Plausibility and comparison with other exercises
- Relative responses of house prices and unemployment are within ranges of recent data and literature:
  - 2018 EU-wide banking stress test: in that scenario, house prices fall 13.6 percent more and unemployment is 0.9 percentage points higher in the adverse scenario in year one; multiplying both by three gives responses similar to the scenario here.
  - The 40 percent fall in house prices would decrease the house price index from 160 to 96, which corresponds to a 1.4 percentage point rise in the unemployment rate (alternative unemployment calculation yields 1.6 percentage points using labor force survey data).
- The scenario’s increase in unemployment is designed to capture the potential absence of discretionary Kurzarbeit expansions as in COVID-19; the unemployment response relative to house prices is weaker here than in Barasinska et al. (2019) or the 2023 EU-wide banking sector stress tests.

### Results: financial stability and consumer-protection benefits of borrower-based measures (Table 3; cumulative over 4-quarter horizon, losses in stress scenario in excess of reference scenario)
- Aggregate results (exact table values):
  - (1) loss (€ bn): without BBMs 51.7; with BBMs 30.4; difference -21.3
    - Standard errors under each estimate (in parentheses, same units): (0.33), (0.44), (0.43)
  - (2) loss (% CET1): without BBMs 11.9; with BBMs 7.0; difference -4.9
    - Standard errors: (0.08), (0.1), (0.1)
  - (3) loss (% RWA): without BBMs 1.9; with BBMs 1.1; difference -0.8
    - Standard errors: (0.01), (0.02), (0.02)
  - (4) % increase RWA: without BBMs 3.2; with BBMs 2.2; difference -1.0
    - Standard errors: (0.03), (0.02), (0.02)
  - =(2)+(4) % change CET1: without BBMs 15.1; with BBMs 9.2; difference -5.9
    - Standard errors: (0.1), (0.12), (0.1)
- Key numeric interpretations:
  - Borrower-based measures preserve almost 6 percent of CET1 capital, or €26 billion (5.9 percent of CET1).
  - Expected credit losses are €21 billion higher, or 4.9 percent of CET1, without the LTV/DSTI limits.
  - Prevention of increasing risk-weighted assets preserves a further 1 percent of CET1 capital.
- Methodological note on conversions and RWA estimate:
  - Loss rates are aggregated as loan balance-weighted averages and accumulated over four periods; to convert to euros, aggregate loss rate multiplied by stock of residential mortgage loans (€1,731bn).
  - Increase in RWAs estimated by four-quarter change in aggregate risk weight multiplied by share of residential real estate in IRB RWAs (assumed 20 percent) and share of IRB RWAs in RWAs (assumed 50 percent).
- Distributional findings (by LTV bucket):
  - Most of the €21 billion in prevented losses originate among loans with LTVs between 70 and 100 percent (of market value) in 2022Q2, despite their small share of the portfolio.
  - Loss prevention by LTV bucket examples:
    - 7 percent of losses are prevented in the 80–90 percent LTV bucket.
    - 24 percent of losses are prevented in the 90–100 percent LTV bucket.
- Uncertainty and caveats:
  - Standard errors across ten replications are small, implying negligible sampling uncertainty given the large loan sample; standard errors do not capture model or scenario uncertainty, which are likely substantial. Annex I quantifies some model uncertainty from parameter estimation.

### Interpretation, limitations, and costs
- Potential understatement of benefits:
  - Benefits could be understated due to limitations of PD data.
  - Simulation assumes borrower-based measures introduced in 2015 and run for seven years before crash; allowing more time to change mortgage-market composition would likely increase benefits (underscores importance of early activation).
  - Simulation does not capture potential of borrower-based measures to limit house price overvaluation ex ante, or to limit household leverage and thereby consumption contraction.
- Costs not analyzed in this simulation:
  - Simulation does not investigate costs; a key financial-stability cost could be foregone bank profits.
  - When comparing borrower-based (BBMs) to lender-based macroprudential measures, costs should be compared alongside benefits.

### Illustrative calibration approaches for borrower-based measures in Germany
- LTV calibration by protection against negative equity:
  - 2-standard deviation drop in prices of existing German houses over 1 year = 8 percent → maximum LTV to avoid negative equity = 92 percent (borrower makes down-payment at least 8 percent).
  - 2-standard deviation shock over 3 years = 22 percent → maximum LTV = 78 percent.
  - Using distributional estimate: 5th percentile of three-year house price changes in 2022 = 14 percent drop → maximum LTV = 86 percent.
- DSTI calibration by protecting essential consumption under income shocks:
  - Two definitions of essential spending:
    - Broad definition (includes food, transport and communication, medical, education and financial services expenditures, and saved income): averages 54 percent of income between 2021Q3 and 2022Q3 → leaves at most 46 percent for debt-servicing and non-essential spending.
    - Narrow definition (drops savings): averages 42 percent → leaves at most 58 percent for debt-servicing and non-essential spending.
  - 2-standard deviation falls in household income:
    - 1 year: 3.5 percent → to have at least 54 percent for essentials, DSTI ≤ 45 percent; to have at least 42 percent for essentials, DSTI ≤ 56 percent.
    - 3 years: 4.6 percent → DSTI ≤ 44 percent (broad) or ≤ 56 percent (narrow).
    - 10 years: 9.5 percent → DSTI ≤ 42 percent (broad) or ≤ 53 percent (narrow).
  - These yield illustrative DSTI limits for fixed-rate mortgages under alternative essential-spending definitions and income-shock horizons.

### International evidence (introductory note)
- Evidence from advanced economies tentatively suggests that activating borrower-based measures for the first time could reduce house prices, household credit, and inflation, with no clear effect on output.
- Recent empirical papers use methods addressing endogeneity (methods of moments, propensity score matching, narrative approaches) and sometimes control for concurrent policy changes (monetary policy, amortization requirements).
- More evidence is available on LTV caps than other instruments.

*Source: IMF staff calculations and discussion in SIPEA2023060 (Germany chapter).*

### 38.      Credit and house prices. There is mixed evidence on the effects of borrower-based measures

### 38.      Credit and house prices. There is mixed evidence on the effects of borrower-based measures

### Evidence on effects of borrower-based measures on credit and house prices
- Literature review by Gatt (2023) concludes evidence on a link between LTV caps and credit is ‘suggestive.’
- Borrower-based measures:
  - Seem to have weaker effects in advanced economies (AEs) and stronger effects when first introduced than when subsequently tightened.
- Cerutti et al. (2017):
  - Do not find evidence of an effect of LTV or DTI caps on aggregate credit growth or house price growth in AEs.
  - Find that tighter DTI caps reduce household credit growth in AEs.
- IMF (2018) finds limited evidence in Sweden and Norway of any effects on credit.
- Alam et al. (2019) find that tighter LTV caps reduce household credit in AEs when the initial LTV cap is loose (at 100 percent or higher), but not when the initial LTV cap is tight.
- Mokas and Giuliodori (2021) find a drop in real household credit, and a statistically insignificant change in real house prices, in response to LTV caps in an EU sample.
- Possible reasons for ambiguous estimated effects:
  - Complexity in design (multiple thresholds and exemptions).
  - Policies typically anticipated through consultations and phase-in periods.
  - Outcomes affected with significant and variable lags.
  - Borrower-based measures are still relatively new in many countries, limiting evidence across housing market and economic cycles.

### Output and inflation
- Richter et al. (2019):
  - Do not find evidence for an effect of LTV caps on output in AEs.
  - Find that tighter LTV caps reduce consumer prices in AEs temporarily, up to about two years, after which there is no statistically significant evidence of any effect.

### Resilience effects
- Activating borrower-based measures tends to increase resilience, observed as:
  - Lower volatility of house prices.
  - Lower rates of default on mortgages.
  - Lower sensitivity of consumption to house prices (Verbruggen et al. 2015; IMF, 2023b).
- Applicability caveat:
  - Unclear to what extent evidence on resilience applies to AEs because much evidence mixes AEs with EMs.
- Mechanisms:
  - Tighter LTV cap: borrower must take a greater equity stake, reducing likelihood of exhausting equity if house prices fall.
  - Tighter DSTI/DTI/LTI cap: borrower more likely to be able to afford debt service payments and regular consumption if income falls unexpectedly.

### Case studies: observed effects and design features
- Ireland (introduced LTV and LTI caps in 2015):
  - Up to 2006, share of loans with LTVs greater than 95 percent had grown up to 20 percent.
  - Policy has separate limits for first-time buyers, second-and-subsequent buyers, and buy-to-let investors.
  - Policy initially had a sliding scale LTV limit for first-time buyers (later abolished).
  - After introduction in 2015 and tightening in 2018, house price growth slowed.
  - CBI (2019) found house prices may have been 13-25 percent higher in 2019 without the mortgage measures.
  - New mortgages bunched below the LTV thresholds (IMF, 2022c).
  - Acharya et al. (2020) found policy reduced credit-house price spirals and reallocated credit from cities to other areas and from low-income to high-income borrowers.
- Netherlands (LTV and DSTI caps from January 2013):
  - LTV cap started at 106 percent in 2012 and gradually reduced by 1 percentage point per year to 100 percent in 2018.
  - 2013 DSTI cap around 18.5 percent of gross income for low-income households and increased up to 46.5 percent for high-income households.
  - By 2018, DSTI caps tightened to 13-36 percent across the household income distribution.
  - Average LTV on new loans fell from around 103 percent in 2011 to about 90 percent by 2018.
  - Between 2014 and 2019, share of new loans with LTV above 90 percent declined from 75 to 66 percent for first-time buyers and from 55 to 47 percent for second-and-subsequent-time buyers (DNB, 2019).
  - IMF (2023c) assessed LTV limits were successful in reducing household mortgage debt from 106 percent of GDP in 2012 to 93 percent in 2020.
  - Nevertheless, house prices increased strongly—after bottoming out in 2013, they increased 90 percent by 2022.
  - Biesenbeek et al. (2022) estimated homeownership among first-time buyers fell by 6 percent due to the LTV cap.
  - The Netherlands also limited mortgage interest deductibility, placed restrictions on non-amortizing mortgages, and introduced minimum risk weights on residential mortgages of IRB banks (12 percent for that part of each loan with an LTV of up to 55 percent and 45 percent on the remaining part of the loan).
- New Zealand (LTV caps introduced 2013):
  - At introduction, 21 percent of outstanding loans had LTVs above 80 percent.
  - Policy had a speed limit, differentiated by city, and exempted newly constructed homes.
  - By 2016, 11 percent of outstanding loans had LTVs above 80 percent.
  - After introduction in 2013 and tightening in 2016, house price growth and household credit growth both slowed.
  - Impact on house prices unclear due to policies to boost housing supply and monetary tightening from 2021.
  - New Zealand removed LTV caps during COVID-19, re-introduced at pre-pandemic settings in March 2021, and tightened further in May and November.
  - RBNZ (2021) assessed re-introducing LTV caps would strengthen financial stability and provide substantial medium-term real economic benefits via higher resilience, with small short-term economic costs by moderating house price inflation and restricting credit to households with low liquidity but high creditworthiness.
  - RBNZ (2021) judged re-introducing LTV caps would reduce house price inflation by 1-4 percentage points, citing Armstrong et al. (2019) and other papers.
- Austria (introduced measures in June 2022, to take effect in August 2022):
  - LTV cap of 90 percent introduced to preserve financial stability.
  - 20 percent speed limit.
  - DSTI cap of 40 percent and a maximum loan term of 35 years.
  - In 2021H1, half of new mortgage loans had LTVs above 90 percent, and 18 percent had DSTI above 40 percent.
- Denmark (introduced an LTV cap of 95 percent in 2015):
  - After measures, share of new borrowers with LTV above 95 percent declined markedly.
  - Share of borrowers with LTV between 80 and 90 percent declined slightly.

### Conclusions and policy recommendations (Germany context)
- Paper explains borrower-based measures could strengthen:
  - Financial stability.
  - Macroeconomic stability.
  - Consumer protection in Germany.
- Potential concerns highlighted:
  - Overall regulatory burden (when combined with existing lender-based measures).
  - Impact on housing affordability.
- Countervailing arguments and open work:
  - Housing market risk-versus-resilience balance in Germany compared to peer countries.
  - Ambiguous effects on housing affordability.
  - Need for more detailed impact assessments.
- Microsimulation analysis:
  - Shows that activating borrower-based measures (restricting lending at LTVs above 90 percent of the mortgage lending value or at DSTIs above 40 percent) could preserve as much capital in the banking system as the capital buffer requirements activated in 2022, subject to significant uncertainty.
- Recommendation:
  - Income-related borrower-based measures should be added to complete the macroprudential toolkit; these would complement existing lender- and borrower-based macroprudential measures.
- Timing and calibration:
  - Borrower-based measures should be activated as the economy recovers, well in advance of the next housing market downturn.
  - Given current weakness in property markets, now is not the ideal time to activate borrower-based measures.
  - The economy is expected to recover over the next few years and vulnerabilities will again build absent activated borrower-based measures.
  - One phased approach: start by introducing quantitative guidance on LTV limits, which are not legally binding.
  - Further work should examine merits of calibrating measures on a cyclical basis versus calibrating at levels appropriate throughout the cycle.

### Box 1 — Germany’s existing borrower-based measures (summary)
- Legal authority:
  - Section 48u of the German Banking Act (Gesetz über das Kreditwesen, KWG) gives BaFin powers to impose an LTV cap or an amortization requirement on new loans for purchase (or construction) of residential real estate.
  - LTV cap is specified relative to the market value of the property, not the mortgage lending value.
  - Powers can be used to preserve financial stability or functioning of the financial system, not for objectives like macroeconomic stability or consumer protection.
  - Amortization requirement defined as maximum period within which a certain fraction of a loan must be repaid or, for interest-only loans, a maximum loan term.
- Design features:
  - LTV limit allows exemptions and some differentiation between borrowers.
  - Measures have a speed limit (proportion of new loans not subject to measures) and a de minimis limit (exemption for small loans).
  - Law does not explicitly mention differentiated limits by borrower type, but authorities signaled differentiated application is within discretion if warranted.
- Leakages and scope:
  - Leakages to NBFIs can be avoided because similar instruments exist for insurers and investment funds in the Insurance Supervision Law (paragraph 308b) and Capital Investment Code (paragraph 5 (8a)) respectively.
  - Measures apply to lending by subsidiaries and branches of foreign banks in Germany, but not to cross-border lending by foreign banks into Germany; BaFin could request reciprocation via the ESRB if leakages material.
- Activation timing:
  - Measures can be activated pre-emptively. IMF (2022b) recommended authorities adjust legislation to allow pre-emptive activation; a joint working group concluded no such legislative obstacles.
  - Measures can also be imposed in non-legally binding form (quantitative guidance) or by instructing banks not to lend towards taxes and transaction costs in a house purchase.
- Other legal restrictions:
  - Pfandbrief Act Section 14: mortgages funded by covered bonds cannot have LTV more than 60 percent of the mortgage lending value; only 13 percent of outstanding mortgages were funded by covered bonds as of September 2022.
  - Building and Loan Associations Act Section 7: mortgages extended by building societies cannot have LTV more than 80 percent of the mortgage lending value; only 10 percent of outstanding residential real estate loans were with building and loan associations as of September 2022.
  - Note on calculations: 13 percent share is ratio of €228 bn in Pfandbriefe to €1,758 bn in domestic home loans; 10 percent is €185 bn in building loans out of €1,758 bn in domestic home loans.

*Source: Excerpt from sipea2023060 — “Credit and house prices. There is mixed evidence on the effects of borrower-based measures”*

### Annex I. Technical Details of the Microsimulation

---


_Source: https://www.imf.org/-/media/files/publications/selected-issues-papers/2023/english/sipea2023060.pdf_