## 1. Loan Data

## Source details

**Canonical URL:** [1. Loan Data](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019182-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019182-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019182-print-pdf.pdf.json)

---

### Introduction and motivation
- High household credit growth and leverage pose risks to financial stability (Reinhart and Rogoff (2009), Schularick and Taylor (2012), Jordà et al. (2016)).
- Downturns with financial stress produce longer, slower recoveries (Claessens et al. (2009), Abdul et al. (2011)).
- Borrower-based macroprudential tools (LTV, DSTI) have proven effective in containing excessive credit growth and bolstering borrower resilience (Claessens et al. (2013), Kuttner and Shim (2016), Akinci and Olmstead-Rumsey (2015), Alam et al. (2019), IMF (2014)).
- Objective: use Romanian loan-level data to calibrate a DSTI limit that increases borrower resilience without unduly restricting access to credit.

### Data sources, sample construction, and DSTI calculation
- Main data source: Central Credit Register covering all consumer or mortgage loans above RON 20,000 on balance sheet of banks as of June 2016; loans observed again in June 2017 for 1-year ahead performance status.
- Supplementary data for loans with original principal < RON 20,000 from the Credit Bureau; registers merged by borrower to obtain total indebtedness.
- Interest rates for debt-service computation taken from the Monetary Balance Sheet (average interest rate reported by each bank by loan class, maturity, currency).
- Monthly loan installment computed under a constant annuity assumption using residual maturity and interest rate.
- Monthly income data from the Ministry of Public Finance (latest available wages reported for the fiscal year).
- DSTI definition and formula:
  - DSTI = [ r (1 − (1 + r)^(−n) ) S ] / V
    - n = remaining loan term (months)
    - r = current monthly interest rate
    - S = remaining principal
    - V = monthly income
  - For borrower with k loans: DSTI = (Σ_{i=1}^k [ r_i S_i (1 − (1 + r_i)^(−n_i) ) ]) * 1 / V
- Winsorization and sample filters:
  - DSTI values above 300 percent and below 5 percent are winsorized at those values.
  - Borrowers with a salary below the minimum wage are eliminated from the sample.
- Initial selection as of June 2016:
  - 356,321 mortgage loans
  - 2,012,294 unsecured consumer loans
- Final merged sample (after exclusions and consolidation):
  - 200,598 borrowers with a mortgage loan
    - Of these, 59,482 (29.6 percent) also have at least one consumer loan
  - 467,969 borrowers who only have unsecured consumer loans
    - Of the 467,969: 411,000 (88 percent) have only one loan; 51,400 (11 percent) have two loans; 5,600 (1 percent) have 3 or more consumer loans
- Coverage after exclusions (relative to initial sample):
  - Mortgages: 56 percent coverage
  - Consumer loans: 23 percent coverage

### Borrower and loan descriptive statistics and distributions
- Currency and denomination:
  - Mortgage loans roughly equally split between domestic currency and foreign currency (FX).
  - FX loans almost entirely in euros; about 3000 CHF-denominated loans remain.
  - CHF loans constitute 1.5 percent of loans (2,956 loans).
- Income stratification:
  - Borrowers roughly equally split into low-income (below economy-wide average), average-earners (between average and up to twice average), and high-earners (above twice average).
  - Approximately 60 percent of loans issued under the Prima Casa program.
- Origination vintages:
  - 85 percent of loans in sample originated after 2010.
  - Loans originated before 2009 show significantly higher unconditional probability of default.
- Age distribution:
  - Almost 70 percent of mortgage borrowers are below 40 years of age.
  - More than 90 percent are below 50 years of age.
  - Older borrowers show higher average probability of default.
- DSTI distribution and income differences (mortgages):
  - Two thirds of mortgage loans have DSTI below 40 percent.
  - Additional 10 percent have DSTI between 40 percent and 50 percent.
  - Average DSTI:
    - Low-income borrowers: 48 percent
    - High-income borrowers: 19 percent
- Consumer loan characteristics:
  - Median consumer loan: 21,000 lei (median mortgage: 158,000 lei).
  - Residual maturity: median 3 years for consumer loans (median 24 years for mortgages).
  - Consumer loans maximum maturity at origination: five years; loans issued in 2015 and 2016 represent almost 80 percent of consumer loan observations.
  - Consumer loan borrowers skewed to low-income: 56 percent below average income.
  - DSTI asymmetry by income less pronounced for consumer loans: 23 percent (low-income) vs. 19 percent (high-income).
- Default rates in the sample:
  - Consumer loans: 1.5 percent probability of default
  - Mortgages: 0.1 percent probability of default

### Econometric approach
- Estimation objective: conditional relationship between DSTI level (June 2016) and one-year ahead probability of default (June 2017), controlling for borrower and loan characteristics.
- Functional form:
  - Logit model: y_i = α + Σ_{j=1}^5 β_j d_{ij} DSTI_i + Σ_k δ_k X_{ik} + ε_i
    - y_i = logit(PD_i) = log( PD_i / (1 − PD_i) )
    - d_{ij} = dummy indicating DSTI range j
    - X_{ik} = other loan/borrower characteristics
  - Piece-wise linear specification with 5 DSTI ranges; linear specification also tested as robustness.
- Regression conducted at borrower level; mortgage subsample is one-to-one mapping between borrowers and loans (borrowers with >1 mortgage excluded); consumer subsample treats borrowers with multiple consumer loans as in default if any loan becomes non-performing.

### Key empirical findings — mortgages
- Non-linearity in DSTI–default relationship:
  - Mortgage PD becomes sensitive to increases in DSTI at indebtedness levels about 50 percent.
  - For DSTI < 50 percent (Ranges 1 and 2), no significant relationship between DSTI increases and PD.
  - For DSTI > 50 percent, marginal effect of DSTI on PD becomes positive and significant.
- Marginal effects (preferred specification, mortgage):
  - Range =3 (50–90): point estimate ~1.558*** (percentage increase in PD for a 1 percentage point increase in DSTI).
  - Range =4 (90–120): point estimate ~1.369***.
  - Range =5 (>120): point estimate ~0.757***.
  - Example interpretation: marginal impact of DSTI in range 3 = 1.6 implies a 10-percentage point increase in DSTI increases PD by 16 percent.
- Other mortgage determinants:
  - Mortgages denominated in foreign currency are more likely to default (effect stronger for CHF).
  - Prima Casa (first-time buyer) dummy negative and significant (e.g., -1.547*** in baseline), indicating lower PD for first-time buyers.
  - Having both mortgage and consumer loan increases likelihood of default, controlling for total debt burden.
  - Longer residual maturity associated with higher PD.
  - Larger loan size associated with lower PD (negative and significant coefficient).
- Robustness:
  - Results robust to excluding loans with DSTI > 100 percent, adding LTV, adding log loan size, and replacing income dummies with log income (income becomes insignficant due to multicollinearity).

### Key empirical findings — consumer loans
- Non-linearity and thresholds:
  - Consumer loan PD elasticity to DSTI not significant below a threshold of 30 percent.
  - For DSTI > 30 percent, indebtedness has a positive and significant impact on PD.
- Marginal effects (consumer loans, as percentage change in PD for a 10-percentage point increase in DSTI):
  - 30–50 DSTI range: 8 percent increase in PD for a 10-percentage point increase in DSTI.
  - 50–70 DSTI range: 6 percent increase in PD for a 10-percentage point increase in DSTI.
  - DSTI > 70 range: 3 percent increase in PD for a 10-percentage point increase in DSTI.
- Other consumer determinants:
  - Residual maturity increases PD.
  - Borrower's age negative and significant: older borrowers less likely to default.
  - Loan size negative and significant: larger loans less likely to default.
  - Income group and log income add discriminatory power: wage below economy-wide average associated with higher PD; log income negative and significant.

### DSTI ranges and sample breakdowns
- Mortgage DSTI ranges (total consolidated mortgage sample: 200,598 borrowers):
  - Range =1 [0,30): 94,712 (47.2%)
  - Range =2 [30,50): 62,430 (31.1%)
  - Range =3 [50,90): 32,175 (16.0%)
  - Range =4 [90,120): 5,859 (2.9%)
  - Range =5 >120: 5,422 (2.7%)
- Consumer DSTI ranges (total consolidated consumer sample: 467,969 borrowers):
  - Range =1 [0,15): 190,176 (40.6%)
  - Range =2 [15,30): 126,036 (26.9%)
  - Range =3 [30,50): 107,571 (23.0%)
  - Range =4 [50,70): 26,651 (5.7%)
  - Range =5 >70: 17,535 (3.7%)

### New NBR regulation (October 2018) — main provisions and rationale
- Main provisions:
  - Ceiling for household loan indebtedness set at 40 percent of net income for leu-denominated loans and 20 percent for foreign currency loans.
  - Total indebtedness measured as ratio of monthly debt service to monthly net income (DSTI at origination).
  - For first-time buyers using property as primary residence (including Prima Casa), maximum indebtedness set at 45 percent.
  - Regulation applies to banks and NBFIs.
  - Exemption allowed: up to 15 percent of a lender's flow of new credit to households may be exempt from the DSTI limit.
  - Regulation effective date: January 1, 2019.
- Methodology and scope changes:
  - Simplified methodology: no adjustments to income for subsistence; limit applies at origination without stress adjustments.
  - Extended perimeter to non-bank financial institutions (NBFIs), important because more than 20 percent of consumer loans as of June 2018 were granted by NBFIs previously outside the perimeter.
- Calibration rationale from empirical results:
  - Mortgage PD sensitivity near DSTI ≈ 50 percent motivated setting origination DSTI ceiling at 40 percent to build buffer for shocks.
  - Tighter DSTI limit for FX loans (20 percent) and more generous cap for first-time buyers (45 percent) justified by higher PD for FX mortgages and lower PD for first-time buyers.

### Resilience to shocks — scenario analysis for a loan originated at DSTI = 40 percent (mortgage baseline)
- Baseline loan for analysis: mortgage with maturity 30 years and original interest rate 4 percent (close to median observed).
- Domestic loan scenarios (origin DSTI = 40% → target ~50%):
  - Scenario 1:
    - Interest Rate shock 2.0%
    - Income Shock 0.0%
    - FX shocks 0.0%
    - Modified DSTI 50.2%
    - Annuity factor original 0.48%
    - Annuity factor shocked 0.60%
  - Scenario 2:
    - Interest Rate shock 0.0%
    - Income Shock 20.0% (i.e., 20 percent reduction)
    - FX shocks 0.0%
    - Modified DSTI 50.0%
    - Annuity factor shocked 0.60%
  - Scenario 3:
    - Interest Rate shock 1.5%
    - Income Shock 5.0%
    - FX shocks 0.0%
    - Modified DSTI 50.1%
    - Annuity factor shocked 0.60%
  - Interpretation: a loan originated at DSTI 40 percent can absorb interest rate increases up to 200 basis points (Scenario 1) or a 20 percent reduction in income (Scenario 2) before reaching DSTI ~50 percent; combination in Scenario 3 (150 bps and 5% income drop) also reaches critical threshold.
- FX loan scenarios (origin DSTI = 20% → FX-equivalent threshold 32%):
  - Derived DSTI threshold for FX loans that equates PD to a leu mortgage at DSTI 50 percent: 32 percent.
  - Scenario 4:
    - Interest Rate shock 0.0%
    - Income Shock 0.0%
    - FX shocks 60.0%
    - Modified DSTI 32.0%
    - Annuity factor shocked 0.76%
  - Scenario 5:
    - Interest Rate shock 1.5%
    - Income Shock 5.0%
    - FX shocks 30.0%
    - Modified DSTI 32.5%
    - Annuity factor shocked 0.78%
  - Interpretation: an FX mortgage originated at DSTI 20 percent can absorb a 60 percent depreciation (Scenario 4) or a combination of 5 percent drop in income, 30 percent depreciation and 150 bps increase in interest rates (Scenario 5) before reaching DSTI ~32 percent.

### Counterfactual analysis — impact of applying DSTI limits to observed loans
- Assumption: all loans with DSTI above 40 percent (20 percent for FX loans) are limited to nominal loan value bringing DSTI down to 40 (20); preferred specification used to estimate counterfactual PDs; intercept calibrated to match average PD observed in June 2017.
- Impact on mortgage probability of default (select results):
  - Overall PD for all mortgages drops by 23 percent:
    - Total mean PD: Actual 0.50% → w/ limit 0.39% → -22.6%
  - Largest PD reductions for highest DSTI loans:
    - DSTI >120: mean PD falls ~56.2% (2.10% → 0.92%)
  - By currency:
    - Domestic Currency total mean PD 0.33% → 0.25% (-23.3%)
    - FX total mean PD 0.67% → 0.52% (-22.3%)
  - By income group:
    - Below average: mean PD 0.70% → 0.44% (-36.7%)
    - [average, 2*average]: mean PD 0.42% → 0.36% (-14.7%)
    - Above 2*average: mean PD 0.39% → 0.36% (-6.3%)
  - Text summary: Low-income borrowers see the largest decline in PD (~40 percent reduction; Table 14 reports -36.7%).
- Impact on mortgage credit volumes (select results):
  - Estimated total loan volumes fall by 23 percent:
    - Total 32.7 → 25.3 → -23%
  - Currency split:
    - Leu-denominated (Domestic Currency) volumes drop 10% (16.0 → 14.4).
    - FX denominated volumes drop 35% (16.7 → 10.9).
  - By DSTI range (largest reductions concentrated in high DSTI ranges):
    - Range 3 [50,90): Total -53% (5.2 → 2.5); FX -68% (2.7 → 0.9).
    - Range 4 [90,120): Total -72% (0.9 → 0.3).
    - Range 5 >120: Total -84% (0.9 → 0.1).
  - By income:
    - Below average: total volumes fall -38% (10.4 → 6.5); FX volumes fall -57% (4.9 → 2.1).
    - [average, 2*average]: total volumes fall -20% (12.6 → 10.1).
    - Above 2*average: total volumes fall -10% (9.7 → 8.7).
- Caveats that may reduce estimated volume impact:
  - Co-borrowers: many high-DSTI loans contracted with co-borrower; true DSTI may be lower than measured (dataset lacks co-borrower income).
  - Underreported income: absence of annual income statement and undeclared income may overstate DSTI.
  - Improved underwriting: more recent loans already account for debt-burden.
  - 15 percent exemption on new flows (allowed by regulation) not accounted for in the counterfactual and will soften actual impact.

### Policy implications and design guidance
- Empirical thresholds and regulatory calibration:
  - Mortgage PD threshold: 50 percent DSTI (marginal PD sensitivity emerges above this).
  - Consumer-loan PD threshold: about 30 percent DSTI.
  - NBR chosen operational limit: 40 percent DSTI for new household loans at origination (leu-denominated).
  - For FX denominated loans the operational DSTI limit is 20 percent.
  - Exception: first-time homebuyers (including Prima Casa) have a raised ceiling of 45 percent DSTI.
  - Lender exemption: up to 15 percent of a lender's flow of new credit to households may be exempted.
  - NBR regulation effective date: January 1, 2019.
- Design guidance:
  - Set regulatory DSTI limits below behavioral PD thresholds to ensure borrowers remain under critical stress points after shocks (example: 40 percent cap to provide buffer below mortgage PD sensitivity at 50 percent).
  - Differentiate treatment:
    - Secured loans (mortgages) can tolerate higher DSTI thresholds than unsecured consumer loans (mortgage threshold 50 percent vs. consumer threshold about 30 percent).
    - First-time buyers may warrant higher ceilings (NBR uses 45 percent for borrower-occupied first homes).
  - Inclusion of NBFIs in perimeter important given material share of consumer lending outside banks.

### Scope and limitations
- The paper does not conduct a full welfare analysis (impact on financial inclusion or steady-state intermediation) nor model second-round effects (macroeconomic feedbacks or loan pricing).
- Analysis focuses on borrower resilience and PD impacts based on loan-level microdata.
- Data limitation: lack of information on co-borrowers and household income pooling may overstate DSTI for some borrowers; robustness check on debtors with DSTI below 100 percent shows main findings remain valid.
- Analysis based on Point-in-Time (PiT) PDs aimed at explaining cross-sectional differences in PDs rather than their time evolution; 2016–2017 described as a period of relative tranquility with relatively low average PDs.

### Conclusion — core takeaways
- Empirical evidence from borrower-level Romanian credit register supports a DSTI cap calibrated at 40 percent for leu-denominated mortgages and 20 percent for FX mortgages to build resilience before the ~50 percent mortgage PD sensitivity threshold.
- Consumer loans exhibit PD sensitivity to DSTI from ~30 percent; regulation and inclusion of NBFIs therefore important.
- Counterfactuals suggest meaningful reductions in PDs (overall mortgage PD reduction ~23 percent) and a sizeable rebalancing of mortgage volumes toward domestic currency and lower DSTI loans, with most pronounced benefits for low-income borrowers.

*Source: wpiea2019182-print-pdf — 1. Loan Data — https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019182-print-pdf.pdf*

### 1. Loan Data ...........................................................................................................

### 1. Loan Data

### Introduction and motivation
- High household credit growth and leverage pose risks to financial stability (Reinhart and Rogoff (2009), Schularick and Taylor (2012), Jordà et al. (2016)); downturns with financial stress produce longer, slower recoveries (Claessens et al. (2009), Abdul et al. (2011)).
- Borrower-based macroprudential tools (LTV, DSTI) have proven effective in containing excessive credit growth and bolstering borrower resilience (Claessens et al. (2013), Kuttner and Shim (2016), Akinci and Olmstead-Rumsey (2015), Alam et al. (2019), IMF (2014)).
- This paper develops empirical analysis using Romanian loan-level data to help calibrate a DSTI limit that increases borrower resilience without unduly restricting access to credit.

### Data sources and sample
- Main data source: Central Credit Register covering all consumer or mortgage loans above RON 20,000 on balance sheet of banks as of June 2016.
- Loans are observed again in June 2017 to register performance status 1- year ahead (performing vs. non-performing).
- For loans with original principal less than RON 20,000, data come from the Credit Bureau; both registers are used to obtain total indebtedness.
- Interest rates for debt-service computation are taken from the Monetary Balance Sheet (average interest rate reported by each bank by loan class, maturity, currency).
- Monthly loan installment computed under a constant annuity assumption using residual maturity and interest rate.
- Monthly income data from the Ministry of Public Finance (latest available wages reported for the fiscal year).
- Sample sizes reported in the analysis: more than 200,000 mortgages and 460,000 consumer loans.

### Key empirical findings
- Non-linearity in DSTI–default relationship:
  - For mortgage borrowers, probability of default (PD) increases with marginal increases in DSTI only from a threshold measured at 50 percent.
  - For borrowers who only carry consumer loans (unsecured), the threshold where increases in DSTI raise PD is measured at about 30 percent.
- Additional associations:
  - Loans denominated in foreign currency are associated with a higher probability of default.
  - First-time buyers are less likely to default compared to other borrowers (confirming Kelly et al. (2015)).

### Counterfactual analysis and quantitative impacts
- Using observed DSTI distribution and estimated PD–DSTI relationship, a counterfactual applying the National Bank of Romania (NBR) DSTI threshold of 40 percent to all loans in the sample indicates:
  - Probability of default would have been lower by 23 percent.
  - Largest reduction in default probabilities occurs for low-income borrowers.
- Policy-relevant thresholds from empirical results:
  - Mortgage PD threshold: 50 percent DSTI.
  - Consumer-loan PD threshold: about 30 percent DSTI.
  - NBR chosen operational limit: 40 percent DSTI for new household loans at origination.
  - For FX denominated loans the operational DSTI limit is 20 percent.
  - Exception: first-time homebuyers (including Prima Casa) have a raised ceiling of 45 percent DSTI.
  - Lender exemption: up to 15 percent of a lender's flow of new credit to households may be exempted.
  - NBR regulation effective date: January 1, 2019.

### Romanian regulatory context and prior framework
- Romania has used macroprudential caps on DSTI and LTV since 2003.
- Previous NBR Regulation No. 17/2012 required banks' internal DSTI limits after considering shocks; specified stress assumptions included:
  - FX depreciation shocks: 35.5 percent for EUR, 40.9 percent for U.S. dollars, and 52.6 percent for loans in other FX.
  - Interest rate shock: 0.6 percentage point increase.
  - Income shock: 6 percent reduction.
- Amendments implemented in October 2018 replaced the prior framework with a simple numerical limit (40 percent) and extended scope to all household loans and to non-banks; became effective January 1, 2019.

### Policy implications and design guidance
- Empirical thresholds suggest setting regulatory DSTI limits below behavioral PD thresholds to ensure borrowers remain under critical stress points after shocks.
- Differentiated treatment justified:
  - Secured loans (mortgages) can tolerate higher DSTI thresholds than unsecured consumer loans (mortgage threshold 50 percent vs. consumer threshold about 30 percent).
  - First-time buyers may warrant higher ceilings (NBR uses 45 percent for borrower-occupied first homes, including Prima Casa).
- A DSTI limit at 40 percent (with a 20 percent limit for FX loans) aims to preserve resilience while limiting adverse effects on credit access; exemptions (15 percent of new flow) provide limited flexibility.

### Scope and limitations
- The paper does not conduct a full welfare analysis (impact on financial inclusion or steady-state intermediation) nor does it model second-round effects (macroeconomic feedbacks or loan pricing).
- The analysis focuses on borrower resilience and PD impacts based on loan-level microdata.

*Source: wpiea2019182-print-pdf - 1. Loan Data — https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019182-print-pdf.pdf*

### 2016. The information from the two data sources is consolidated by borrower using the

### wpiea2019182-print-pdf - 2016. The information from the two data sources is consolidated by borrower using the

### DSTI definition and calculation
- DSTI is calculated as the monthly loan installment:
  - DSTI = [ r (1 − (1 + r)^(−n) ) S ] / V
  - where n = remaining loan term (months), r = current monthly interest rate, S = remaining principal, V = monthly income.
  - The first term on the right-hand-side is the annuity factor.
  - Concept corresponds to current DSTI (takes account of remaining principal, maturity, and current interest rates).
- For a borrower with k loans, DSTI is the sum of the loan installments divided by income:
  - DSTI = (Σ_{i=1}^k [ r_i S_i (1 − (1 + r_i)^(−n_i) ) ]) * 1 / V
- Winsorization and sample filters:
  - DSTI values above 300 percent and below 5 percent are winsorized at those values.
  - Borrowers with a salary below the minimum wage are eliminated from the sample.

### Data sources, sample construction, and exclusions
- Initial selection as of June 2016:
  - 356,321 mortgage loans
  - 2,012,294 unsecured consumer loans
- Exclusions applied (examples):
  - borrowers with more than one mortgage (to exclude speculative house purchases)
  - loans that are restructured or refinanced
  - loans converted from FX to RON
  - loans flagged “unlikely to pay” under the EBA definition of nonperformance
  - loans with overdue payments more than 90 days
  - loans with residual maturity less than 12 months (drop before June 2017)
- Final coverage after exclusions (relative to initial sample):
  - Mortgages: 56 percent coverage
  - Consumer loans: 23 percent coverage
- Reasons for coverage reduction:
  - Mortgages: main drop due to lack of information on debtors’ income (23 percent)
  - Consumer loans: many loans settle earlier than one year so drop out when re-examined in June 2017 (43 percent)
  - Exclusion of borrowers with more than one mortgage accounts for 4 percent of total mortgages
- Merged sample (June 2016 and June 2017 observations):
  - 200,598 borrowers with a mortgage loan
    - Of these, 59,482 (29.6 percent) also have at least one consumer loan
  - 467,969 borrowers who only have unsecured consumer loans
    - Of the 467,969: 411,000 (88 percent) have only one loan; 51,400 (11 percent) have two loans; 5,600 (1 percent) have 3 or more consumer loans
- Regression analysis conducted at borrower level (each borrower appears once)
  - Mortgage subsample: one-to-one mapping between borrowers and loans (borrowers with >1 mortgage excluded)
  - Consumer subsample: borrowers with multiple consumer loans considered in default if any loan becomes non-performing
- Data limitation highlighted:
  - Lack of information on co-borrowers and household income pooling may overstate DSTI for some borrowers
  - Robustness check: regression on debtors with DSTI below 100 percent (Table 4, column 2) shows main findings remain valid

### Borrower and loan characteristics (descriptive)
- Currency and denomination:
  - Mortgage loans roughly equally split between domestic currency and foreign currency (FX)
  - FX loans almost entirely in euros; about 3000 CHF-denominated loans remain
  - CHF loans constitute 1.5 percent of loans (2,956 loans)
- Income stratification:
  - Borrowers roughly equally split into low-income (below economy-wide average), average-earners (between average and up to twice average), and high-earners (above twice average)
  - Approximately 60 percent of loans issued under the Prima Casa program
- Origination vintages:
  - 85 percent of loans in sample originated after 2010
  - Loans originated before 2009 show significantly higher probability of default (unconditional)
- Age distribution:
  - Almost 70 percent of mortgage borrowers are below 40 years of age
  - More than 90 percent are below 50 years of age
  - Older borrowers show higher average probability of default
- DSTI distribution and income differences:
  - Two thirds of mortgage loans have DSTI below 40 percent
  - Additional 10 percent have DSTI between 40 percent and 50 percent
  - Average DSTI:
    - Low-income borrowers: 48 percent
    - High-income borrowers: 19 percent
- Consumer loan characteristics:
  - Median consumer loan: 21,000 lei (median mortgage: 158,000 lei)
  - Residual maturity: median 3 years for consumer loans (median 24 years for mortgages)
  - Consumer loans have maximum maturity of five years at origination; loans issued in 2015 and 2016 represent almost 80 percent of consumer loan observations
  - Consumer loan borrowers skewed to low-income: 56 percent below average income
  - DSTI asymmetry by income less pronounced for consumer loans: 23 percent (low-income) vs. 19 percent (high-income)
- Default rates in the sample:
  - Consumer loans: 1.5 percent probability of default
  - Mortgages: 0.1 percent probability of default
- Unconditional correlations:
  - Probability of default rises rapidly for DSTI levels above 50 percent
  - Probability of default drops modestly with income level; for consumer loans it drops sharply for income levels above the 40th percentile

### Econometric model (mortgages) — specification and estimation
- Objective:
  - Establish conditional relationship between DSTI level (June 2016) and one-year ahead probability of default (as of June 2017), controlling for borrower and loan characteristics
- Functional form:
  - Logit model: y_i = α + Σ_{j=1}^5 β_j d_{ij} DSTI_i + Σ_k δ_k X_{ik} + ε_i
    - y_i = logit(PD_i) = log( PD_i / (1 − PD_i) )
    - d_{ij} = dummy indicating DSTI range j
    - X_{ik} = other loan/borrower characteristics
  - Piece-wise linear specification with 5 DSTI ranges (defined in Table (3))
  - Also test linear specification (β_j equal) as robustness

### Mortgage regression findings and robustness checks
- Non-DSTI explanatory variables (column (1) baseline):
  - Foreign currency loans more likely to default; effect stronger for CHF loans
  - Prima Casa loans have lower probability of default, ceteris paribus
  - Having both mortgage and consumer loan increases likelihood of default, controlling for total debt burden
  - Longer residual maturity associated with higher probability of default
- DSTI marginal effects:
  - For DSTI < 50 percent (ranges 1 and 2): no significant relationship between DSTI increases and probability of default
  - For DSTI > 50 percent: marginal effect of DSTI on probability of default becomes positive and significant
- Robustness specifications:
  - Column (2): eliminate loans with DSTI > 100 percent (on basis that single borrower cannot support DSTI > 100 percent) — results remain robust
  - Column (3): add LTV — coefficient on LTV positive and significant
  - Column (4): add log loan size — coefficient negative and significant (larger loans less likely to default)
  - Column (5): add dummies for high-income (≥ twice average) and low-income (< average) — income dummies not statistically significant once DSTI and loan size controlled
  - Column (6): add log income instead of income dummies — income not statistically significant (multicollinearity between income, loan size, and DSTI)
  - Preferred specification in column (7): includes loan size (not income), plus dummies for residence, lending bank, and year of origination (coefficients not reported)
- Interpretation of loan size finding:
  - Given same DSTI, larger loan size may indicate higher wage or more favorable loan terms (lower interest rate or longer original term); such borrowers are less likely to default

### Quantitative marginal impacts (preferred specification)
- Table (5) reports marginal impact of a change in DSTI on percentage change in probability of default (dPD/dDSTI divided by PD)
- Example from preferred specification (column (7)):
  - Marginal impact of DSTI in range 3 = 1.6
    - Interpretation: a 10-percentage point increase in DSTI increases probability of default by 16 percent
- Figure (4) reported results for a 10-percentage point increase in DSTI:
  - DSTI levels between 50 to 90 percent: increase in probability of default of 16 percent
  - DSTI between 90 and 120 percent: increase in probability of default of 14 percent
  - DSTI above 120 percent: increase in probability of default of 8 percent
  - Confidence intervals shown around point estimates; marginal impacts not statistically significant for DSTI < 50 percent
- Linear specification robustness (Table (6)):
  - DSTI positively and significantly correlated with higher probability of default in all tested specifications

### Additional notes
- The analysis is based on Point-in-Time (PiT) PDs aimed at explaining cross-sectional differences in PDs rather than their time evolution
- 2016-2017 period described as a period of relative tranquility with relatively low average PDs
- Mortgage regressions also repeated including debtors with more than one mortgage (these account for 4 percent of the original sample) as an alternative sample (details not reported here)

*Source: wpiea2019182-print-pdf - 2016. The information from the two data sources is consolidated by borrower using the*

### 14.8 thousand loans. The results are reported in Tables (7) and (8). We exclude the variable

### wpiea2019182-print-pdf - 14.8 thousand loans. The results are reported in Tables (7) and (8). We exclude the variable

### Mortgage loans — empirical findings and thresholds
- Sample and ranges:
  - Total consolidated mortgage sample: 200,598 borrowers.
  - DSTI ranges assessed: Range =1 [0,30) 94,712 (47.2%); Range =2 [30,50) 62,430 (31.1%); Range =3 [50,90) 32,175 (16.0%); Range =4 [90,120) 5,859 (2.9%); Range =5 >120 5,422 (2.7%).
- Key empirical results:
  - Mortgage probability of default (PD) becomes sensitive to increases in DSTI at indebtedness levels about 50 percent.
  - For DSTI levels below 50 percent, there is no significant relationship between an increase in DSTI and the probability of default; for DSTI larger than 50 percent, the marginal effect of DSTI on PD becomes positive and significant.
  - Residual maturity increases PD (positive and significant coefficients).
  - Mortgages denominated in foreign currency are more likely to default (e.g., CHF denominated coefficients positive and significant in multiple specifications).
  - First-time buyer (Prima Casa) dummy is negative and significant (e.g., -1.547*** in baseline), indicating lower PD for first-time buyers.
- Marginal effects (mortgage model, preferred specification):
  - Range =3 (50–90): point estimate ~1.558*** (percentage increase in PD for a 1 percentage point increase in DSTI).
  - Range =4 (90–120): point estimate ~1.369***.
  - Range =5 (>120): point estimate ~0.757***.
  - Interpretation: point estimates indicate percentage increase in PD for a 1 percentage point increase in DSTI (Table 5 note).

### Consumer loans — empirical findings and thresholds
- Sample and ranges:
  - Total consolidated consumer sample: 467,969 borrowers.
  - DSTI ranges assessed: Range =1 [0,15) 190,176 (40.6%); Range =2 [15,30) 126,036 (26.9%); Range =3 [30,50) 107,571 (23.0%); Range =4 [50,70) 26,651 (5.7%); Range =5 >70 17,535 (3.7%).
- Key empirical results:
  - Consumer loan PD elasticity to DSTI is not significant below a threshold of 30 percent.
  - For DSTI values above 30 percent, indebtedness has a positive and significant impact on PD.
  - Residual maturity increases PD (positive and significant).
  - Borrower's age is negative and significant: older borrowers are less likely to default.
  - Loan size has a negative and significant impact on default probability.
  - Income group and log income add discriminatory power: borrowers with wage below the economy-wide average are more likely to default; log income is negative and significant.
- Marginal and graphical interpretations:
  - Table (11) and Figure (5) express marginal effects as percentage changes in PD for a 10-percentage point increase in DSTI:
    - 30–50 DSTI range: 8 percent increase in PD for a 10-percentage point increase in DSTI.
    - 50–70 DSTI range: 6 percent increase in PD for a 10-percentage point increase in DSTI.
    - DSTI above 70 percent range: 3 percent increase in PD for a 10-percentage point increase in DSTI.
- Comparative risk:
  - Consumer loans are generally riskier than mortgages: as of June 2018, the NPL ratio of consumer loans was 5.2 percent, compared to 3.0 percent for mortgages.
  - In the sample, observed default rate: consumer loans 1.5 percent versus mortgages 0.1 percent.

### New NBR regulation on DSTI limits (October 2018)
- Main provisions:
  - Ceiling for household loan indebtedness set at 40 percent of net income for leu-denominated loans and 20 percent for foreign currency loans.
  - Total indebtedness measured as ratio of monthly debt service to monthly net income (DSTI at origination).
  - For first-time buyers using the property as primary residence (including Prima Casa), maximum indebtedness set at 45 percent.
  - Regulation applies to banks and NBFIs.
  - Exemption allowed: up to 15 percent of a lender's flow of new credit to households may be exempt from the DSTI limit.
- Methodology and scope changes:
  - Simplified methodology: no adjustments to income for subsistence; limit applies at origination without stress adjustments.
  - Levels the playing field between banks and non-bank lenders (more than 20 percent of consumer loans as of June 2018 granted by NBFIs previously outside perimeter).
  - Rationale: Prima Casa borrowers have lower PD and can support higher DSTI ceteris paribus.

### Resilience to shocks — shock scenarios for a loan originated at DSTI = 40 percent (Table 12)
- Loan baseline for analysis:
  - Mortgage with maturity 30 years and original interest rate 4 percent (close to median observed).
- Domestic loan scenarios (origin DSTI = 40% → target ~50%):
  - Scenario 1: Interest Rate shock 2.0%; Income Shock 0.0%; FX shocks 0.0% → Modified DSTI 50.2%; Annuity factor original 0.48%; Annuity factor shocked 0.60%.
  - Scenario 2: Interest Rate shock 0.0%; Income Shock 20.0% (i.e., 20 percent reduction); FX shocks 0.0% → Modified DSTI 50.0%; Annuity factor shocked 0.60%.
  - Scenario 3: Interest Rate shock 1.5%; Income Shock 5.0%; FX shocks 0.0% → Modified DSTI 50.1%; Annuity factor shocked 0.60%.
  - Interpretation: loan originated at DSTI 40 percent can absorb interest rate increases up to 200 basis points (Scenario 1) or a 20 percent reduction in income (Scenario 2) before reaching DSTI ~50 percent; combination in Scenario 3 (150 bps and 5% income drop) also reaches critical threshold.
- FX loan scenarios (origin DSTI = 20% → FX-equivalent threshold 32%):
  - Derived DSTI threshold for FX loans that equates PD to a leu mortgage at DSTI 50 percent: 32 percent.
  - Scenario 4: Interest Rate shock 0.0%; Income Shock 0.0%; FX shocks 60.0% → Modified DSTI 32.0%; Annuity factor shocked 0.76%.
  - Scenario 5: Interest Rate shock 1.5%; Income Shock 5.0%; FX shocks 30.0% → Modified DSTI 32.5%; Annuity factor shocked 0.78%.
  - Interpretation: an FX mortgage originated at DSTI 20 percent can absorb a 60 percent depreciation (Scenario 4) or a combination of 5 percent drop in income, 30 percent depreciation and 150 bps increase in interest rates (Scenario 5) before reaching DSTI ~32 percent.

### Counterfactual analysis — impact of DSTI limits on PDs and credit volumes
- Method:
  - Assume all loans with DSTI above 40 percent (20 percent for FX loans) would be limited to nominal loan value bringing DSTI down to 40 (20); use preferred specifications to estimate counterfactual PDs; calibrate intercept to match average PD observed in June 2017.
- Impact on mortgage probability of default (Table 13 and Table 14):
  - Overall PD for all mortgages drops by 23 percent (Total mean PD: Actual 0.50% → w/ limit 0.39% → -22.6%).
  - Largest PD reductions for highest DSTI loans (DSTI >120): mean PD falls ~56.2% (2.10% → 0.92%).
  - Impact similar across currency: Domestic Currency total mean PD 0.33% → 0.25% (-23.3%); FX total mean PD 0.67% → 0.52% (-22.3%).
  - By income group (Table 14):
    - Below average: mean PD 0.70% → 0.44% (-36.7%).
    - [average, 2*average]: mean PD 0.42% → 0.36% (-14.7%).
    - Above 2*average: mean PD 0.39% → 0.36% (-6.3%).
  - Low-income borrowers see the largest decline in PD (~40% reduction reported in text and 36.7% in Table 14).
- Impact on mortgage credit volumes (Table 15 and Table 16):
  - Estimated total loan volumes fall by 23 percent (Total 32.7 → 25.3 → -23%).
  - Currency split:
    - Leu-denominated (Domestic Currency) volumes drop 10% (16.0 → 14.4).
    - FX denominated volumes drop 35% (16.7 → 10.9).
  - By DSTI range: largest volume reductions concentrated in high DSTI ranges:
    - Range 3 [50,90): Total -53% (5.2 → 2.5); FX -68% (2.7 → 0.9).
    - Range 4 [90,120): Total -72% (0.9 → 0.3).
    - Range 5 >120: Total -84% (0.9 → 0.1).
  - By income (Table 16):
    - Below average: total volumes fall -38% (10.4 → 6.5); FX volumes fall -57% (4.9 → 2.1).
    - [average, 2*average]: total volumes fall -20% (12.6 → 10.1).
    - Above 2*average: total volumes fall -10% (9.7 → 8.7).
- Caveats and factors that may reduce estimated volume impact:
  - Co-borrowers: many high-DSTI loans contracted with co-borrower; true DSTI may be lower than measured (dataset does not include co-borrower income).
  - Underreported income: absence of annual income statement and undeclared income may overstate DSTI.
  - Improved underwriting: more recent loans already account for debt-burden; more than 15 percent of loans in sample granted in 2009 or earlier.
  - 15 percent exemption on new flows is not accounted for in the counterfactual and will soften actual impact.

### Policy implications and regulatory design rationale
- Calibration rationale:
  - Mortgage PD sensitivity near DSTI ≈ 50% motivated setting origination DSTI ceiling at 40% to build buffer for shocks.
  - Tighter DSTI limit for FX loans (20%) and more generous cap for first-time buyers (45% for Prima Casa) are supported by empirical findings:
    - FX mortgages more likely to default; first-time buyers less likely to default.
- Consumer loans:
  - Threshold at which PD becomes sensitive to DSTI is lower for consumer loans (~30%), reflecting higher unconditional PD and less severe default consequences compared to mortgages.
  - Regulation extends to NBFIs, important because more than 20 percent of consumer loans (as of June 2018) are granted by NBFIs previously outside the perimeter.
- Resilience benefit quantified:
  - Default rates would have been lower by 23 percent had all mortgages complied with 40 percent (20 percent for FX) DSTI rule.
  - Largest reduction concentrated among low-income households (close to 40 percent reduction in default rates).
  - Regulation likely to rebalance lending towards domestic currency and lower DSTI loans.

### Conclusion — core takeaways
- Empirical evidence from borrower-level Romanian credit register supports a DSTI cap calibrated at 40 percent for leu-denominated mortgages and 20 percent for FX mortgages to build resilience before the ~50 percent mortgage PD sensitivity threshold.
- Consumer loans exhibit PD sensitivity to DSTI from ~30 percent; consumer loan regulation and inclusion of NBFIs therefore important.
- Counterfactuals suggest meaningful reductions in PDs and a sizeable rebalancing of mortgage volumes toward domestic currency and lower DSTI loans, with most pronounced benefits for low-income borrowers.

*Source: NBR and IMF staff estimates as reported in the content unit.*

### REFERENCES

### REFERENCES

### Macroprudential policy design, evaluation, and staff guidance
- International Monetary Fund, 2014, “Staff Guidance Note on Macroprudential Policy,” IMF Policy Paper, November.
- Committee on the Global Financial System, 2016, “Experiences with the Ex-Ante Appraisal of Macroprudential Instruments,” Technical Report.
- European Systemic Risk Board, 2018, “National Measures of Macroprudential Interest in the EU/EEA,” Technical Report.
- Akinci, O. and J. Olmstead-Rumsey, 2015, How Effective are Macroprudential Policies? An Empirical Investigation, International Finance Discussion Papers 1136, Board of Governors of the Federal Reserve System (U.S.).
- Alam, Z., A. Alter, J. Eisemann, G. Gelos, H. Kang, M. Narita, E. Nier, and N. Wang, 2019, “Digging Deeper—Evidence on the Effect of Macroprudential Policies from a New Database,” IMF Working Paper 19/66.
- Claessens, S., S. R. Ghosh, and R. Mihet. 2013. “Macro-Prudential Policies to Mitigate Financial System Vulnerabilities,” Journal of International Money and Finance, Vol. 39, pp. 153–185.
- Gambacorta, L. and A. Murcia, 2017, “The Impact of Macroprudential Policies and their Interaction with Monetary Policy: An Empirical Analysis Using Credit Registry Data,” BIS Working Papers 636, Bank for International Settlements.

### Borrower-based measures, housing finance, and household indebtedness
- Igan, D. and H. Kang, 2011, “Do Loan-to-Value and Debt-to-Income Limits Work? Evidence from Korea,” IMF Working Papers, Vol. 11, March.
- Neagu, F., L. Tatarici, and I. Mihai, 2015, “Implementing Loan- to-Value and Debt Service-To-Income Measures: A Decade of Romanian Experience,” Occasional Papers 15, National Bank of Romania.
- Balás, T., A. Banai, and Z. Hosszú, 2015, “Modelling Probability of Default and Optimal PTI Level by Using a Household Survey,” Acta Oeconomica, Vol. 65, No. 2, pp. 183–209, June.
- Gross, M. and J. Poblacion, 2017, “Assessing the Efficacy of Borrower-Based Macroprudential Policy Using an Integrated Micro-Macro Model for European Households,” Economic Modelling, Vol. 61, pp. 510–528.
- Jimenez, G., S. Ongena, J.   P., and J. Saurina, 2017, “Macro-prudential Policy, Countercyclical Bank Capital Buffers, and Credit Supply: Evidence from the Spanish Dynamic Provisioning Experiments,” Journal of Political Economy, Vol. 125, No. 6. pp. 2126–2177.
- Kelly, R., T. O’Malley, and C. O’Toole, 2015, “Designing Macroprudential Policy in Mortgage Lending: Do First Time Buyers Default Less?” Research Technical Papers 02/RT/15, Central Bank of Ireland.
- D'Alessio, G. and S. Iezzi, 2013, Household Over-Indebtedness: Definition and Measurement with Italian Data, February 22, 2013, Bank of Italy Occasional Paper, No. 149.
- McCarthy, Y., 2014, “Disentangling the Mortgage Arrears Crisis: The Role of the Labour Market, Income Volatility and Housing Equity,” Research Technical Papers. Central Bank of Ireland.
- May, O. and M. Tudela, 2005, “When is Mortgage Indebtedness a Financial Burden to British Households? A Dynamic Probit Approach,” Bank of England Working Papers 277, Bank of England.

### Credit cycles, crises, and macroeconomic interactions
- Claessens, S., M. A. Kose, and M. E. Terrones, 2009, “What Happens During Recessions, Crunches and Busts?” Economic Policy, Vol. 24, No. 60, pp. 653–700.
- Schularick, M. and A. M. Taylor, 2012, “Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and Financial Crises, 1870–2008,” American Economic Review, Vol. 102, No. 2, pp. 1029–61.
- Jordà, O., M. Schularick, and A. M. Taylor, 2016, “The Great Mortgaging: Housing Finance, Crises and Business Cycles,” Economic Policy, Vol. 31, No. 85, pp. 107–152.
- Mian, A. and A. Sufi, 2009, “The Consequences of Mortgage Credit Expansion: Evidence from the U.S. Mortgage Default Crisis,” The Quarterly Journal of Economics, Vol. 124, No. 4, pp. 1449–1496.
- Reinhart, C. M. and K. S. Rogoff, 2009, “The Aftermath of Financial Crises,” American Economic Review, Vol. 99, No. 2, pp. 466–72.
- Acharya, V. V., M. P. Richardson et al, 2009, Restoring Financial Stability: How to Repair a Failed System, Vol. 542: John Wiley & Sons.

### Microdata, credit registries, empirical methods, and measurement
- Epure, M., I. Mihai, C. Minoiu, and J. Peydro. 2017, “Household Credit, Global Financial Cycle, and Macroprudential Policies: Credit Register Evidence from an Emerging Country,” Working Papers 1006, Barcelona Graduate School of Economics.
- Gambacorta, L. and A. Murcia, 2017, (see above) BIS Working Papers 636.
- Fuster, A. and B. Zafar, 2014, “The Sensitivity of Housing Demand to Financing Conditions: Evidence from a Survey,” Staff Reports 702, Federal Reserve Bank of New York.
- Kukk, M, 2016, “What are the Triggers for Arrears on Debt? Evidence from Quarterly Panel Data,” Bank of Estonia Working Papers, WP2016–9, Bank of Estonia.
- Maddaloni, A. and J. L. Peydro, 2011, “Bank Risk-Taking, Securitization, Supervision, and low Interest Rates: Evidence from the Euro-Area and the U.S. Lending Standards,” The Review of Financial Studies, Vol. 24, No. 6.
- King, G. and L. Zeng, 2001, “Logistic Regression in Rare Events Data,” Political Analysis, Vol. 9, No. 2.

### Inequality, transmission channels, and related topics
- Voinea, L., H. Lovin, and A. Cojocaru, 2018, “The Impact of Inequality on the Transmission of Monetary Policy,” Journal of International Money and Finance, Vol. 85, No. C, pp. 236–250.
- Neagu, F., L. Tatarici, and I. Mihai, 2015, (see above) Occasional Papers 15, National Bank of Romania.

*References list as provided in the source PDF.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019182-print-pdf.pdf_