Abstract
Many countries introduced borrower-based instruments to constrain credit to households exceeding a limit on their loan-to-value ratio, their (mortgage) debt-to-income ratio or their debt service-to-income ratio. We evaluate how well borrower-based instruments can target households that would become vulnerable after a shock. We apply the signals approach to derive “optimal” limits that minimize classification errors (either granting credit to financially vulnerable households or constraining credit to households that are not vulnerable). To illustrate, we simulate an adverse scenario using household-level data from Luxembourg. We find that combining several ratios could better target households that would become vulnerable after a shock.
This is a preview of subscription content, log in via an institution to check access.
Source: Own calculations based on waves 1, 2 and 3 of the LU-HFCS; data are multiply imputed and weighted; only households with recent HMR mortgages. Panel (a) Cumulative distribution functions are calculated across all 5 implicates each year. Vertical lines indicate lowest and highest limits envisaged by the law. We omit the upper tail of the cumulative distribution functions. Panel (b) Shares among households with recent HMR mortgages calculated across all 5 implicates in 2018
Source: Own calculations based on waves 2 and 3 of the LU-HFCS; data are multiply imputed and weighted; only households with recent HMR mortgages. Panel (a) Cumulative distribution functions are calculated across all 5 implicates for each year. Vertical lines indicate lowest and highest limits envisaged by the law. We omit the upper tail of the cumulative distribution functions. Panel (b) Shares among households with recent HMR mortgages calculated across all 5 implicates in 2018
Source: Own calculations based on waves 2 and 3 of the LU-HFCS; data are multiply imputed and weighted; only households with recent HMR mortgages. Panel (a) Cumulative distribution functions are calculated across all 5 implicates for each year. Vertical lines indicate lowest and highest limits envisaged by the law. We omit the upper tail of the cumulative distribution functions. Panel (b) Shares among households with recent HMR mortgages calculated across all 5 implicates in 2018
Source: Own calculations based on waves 1, 2 and 3 of the LU-HFCS; data are multiply imputed and weighted; only households with recent HMR mortgages. Panel (a) Cumulative distribution functions are calculated across all 5 implicates for each year. Vertical lines indicate lowest and highest limits envisaged by the law. Panel (b) Shares among households with recent HMR mortgages calculated across all 5 implicates in 2018
Source: Own calculations based on the 3rd wave of the LU-HFCS; data are multiply imputed and weighted; the statistics are calculated by pooling all implicates. Note: Peaks in average LGD difference between compliant and affected households (dashed black line) indicate best limits
Source: Own calculations based on the 3rd wave of the LU-HFCS; data are multiply imputed and weighted; classification errors are calculated by pooling all implicates
Similar content being viewed by others
Notes
See Sangaré (2019) for a study based on a DSGE modelling approach applied to Luxembourg.
Other studies reached similar conclusion using different methodologies, data and approaches. Grodecka’s (2020) theoretical exercise provides the arguments underlying our results. In a DSGE framework calibrated for Sweden, Chen et al. (2016) find that a mix of macroprudential measures is needed to deliver the maximum welfare benefit.
In the category of loans for principal residences granted to borrowers who are not first-time buyers, lending institutions may grant individual loans with an LTV ratio of up to 100%, provided that the aggregate amount of the loans benefiting from this derogation (90% < LTV ratio ≤ 100%) represents no more than 15% of the annual aggregate amount of this category of loans granted by this institution.
To facilitate reading, “debt burden ratios” in the text also include the limit on mortgage maturity.
Evidence in Ferreira (2018) suggests that such a feedback loop may have been active in Luxembourg.
The initial amount of the most recent HMR mortgage is reported directly in the survey. If the household has more than one mortgage, we adjust older mortgages using a linear approximation based on the difference between the initial value of the older loan and its outstanding amount at the survey date.
For each debt burden indicator, we run a quantile regression (results not reported) to confirm that age and wealth are not statistically significant when accounting for other household characteristics (listed in Table 12).
Mortgage rates reported by banks in Table 03.02 on www.bcl.lu declined from 2.15% in 2014 to 1.85% in 2018.
Some less regular mortgage maturities are also visible, reflecting some irregular values as reported by survey participants.
Fixed-rate mortgages usually involve higher interest rates.
Leika and Marchettini (2017) call this measure “probability of incurring distress”.
We measure liquid assets as the sum of bank deposits (mainly sight and saving accounts), stocks (publicly traded stocks, mutual funds, managed accounts, hedge funds), bonds, and potentially less liquid assets (including private businesses other than self-employment and other assets).
Basic living costs are estimated using household specific amounts spent on utilities (e.g. electricity, water, gas, telephone…) and on food consumed at home, as well as 50% of the amounts spent on food outside the home.
Unemployment benefits in Luxembourg are 80% of the last gross wage or 85% if the unemployed person receives child benefits for his/her dependents. Unemployment benefits cannot exceed 2.5 times the minimum wage during the first 6 months and 2.0 times the minimum wage in the following 6 months.
For a detailed description of the REVIS with respect to eligibility conditions and the amount of the benefit please consult: https://guichet.public.lu/en/citoyens/sante-social/action-sociale/aide-financiere/revenu-inclusion-sociale-revis.html#bloub-11
This approach has also been used to evaluate indicators of economic recessions and expansions (Berge and Jordà 2011), to evaluate the performance of investment strategies (Jordà and Taylor 2011), to evaluate indicators of real credit contractions and expansions (Jordà, 2012), and to evaluate early-warning systems for bank distress (Betz et al. 2014), for banking crises (Drehmann and Juselius 2014), or for financial crises (Candelon et al. 2012; Detken et al. 2014).
The Receiver Operating Characteristic (ROC) curve plots the True Positive Rate (TPR) against type I error for all candidate limits in our grid search. See Hanley and McNeil (1982) for a definition.
See Hsieh and Turnbull (1996) for further definitions.
This limit will only be “optimal” in the sense of minimizing classification errors. In our application, this will not necessarily correspond to minimizing the policymaker’s loss function, which would require specifying the links between classification errors, systemic risk and social welfare.
Alternatively, Buckmann et al. (2023) identify “optimal” limits by targeting specific thresholds for type I errors (or “hit rates”). However, this alternative approach to selecting the optimal point in the ROC curve appears less suitable for examining survey data without information on actual credit defaults, where hit rates would be difficult to interpret. Buckmann et al. (2023) observe banks that failed in the aftermath of the 2007–2008 global financial crisis directly in their data. Therefore, with such a condition variable, it makes sense to target for specific values of type I errors to identify “optimal” limits. For borrower-based measures, it would be similar if loan defaults were observed and the condition variable took the value one for households in arrears and zero otherwise. However, in our application, household financial vulnerability is not directly observed but estimated using a measure of the probability of default. The advantage of using the probability of default is that we can simulate adverse scenarios. This helps to find optimal limits under adverse conditions if crises episodes are not observed.
Mortgage maturity is not really a debt burden ratio (Sect. 2.2), so we only analyse it in combination with other ratios.
Given our nonparametric approach, the limited number of observations and their heterogeneous nature produces ROC curves that are kinked rather than smooth parabolas (see Fig. 7). As a result, outlying points on the ROC curve may be optimal for several values of the policy parameter, corresponding to different slopes of the tradeoff between Type I and Type II errors.
Only final household weights were considered to ensure their statistical representativeness in the population.
We adapted Fawcett’s (2006) proposal to focus on the costs of classification errors instead of plotting benefits of true positives against cost of false negatives.
Table 5 provides definitions of the classification errors and the loss function.
This includes all individuals between 16 and 64 years of age who reported their primary labour status as 1—Doing regular work for pay/self-employed/working in family business; or 2—On sick/maternity/other leave (except holidays), planning to return to work; or 3—Unemployed.
According to official figures (Statec Table B3019), the seasonally adjusted unemployment rate averaged 5.4% between April and November 2018 (when fieldwork took place for HFCS wave 3). This falls within our estimated confidence interval (3.8% to 5.6%).
References
Akinci O, Olmstead-Rumsey J (2018) How effective are macroprudential policies? An empirical investigation. J Financ Intermed 33:33–57
Albacete N, Fessler P (2010) Stress testing Austrian Households. Österreichische Nationalbank Financial Stability Report 19, June
Albacete N, Lindner P (2017) Simulating the impact of borrower-based macroprudential policies on mortgages and the real estate sector in Austria—evidence from the Household Finance and Consumption Survey 2014. Financial Stability Report 33, Österreichische Nationalbank, June
Albacete N, Fessler P, Lindner P (2018) One policy to rule them all? On the effectiveness of LTV, DTI and DSTI ratio limits as macroprudential policy tools. Financial Stability Report 35, Österreichische Nationalbank, June
Ampudia M, van Vlokhoven H, Zochowski D (2016) Financial fragility of euro area households. J Financ Stab 27:250–262
Araujo J, Patnam DM, Popescu A, Valencia F, Yao W (2020) Effects of macroprudential policy: evidence from over 6,000 estimates. IMF Working Paper WP/20/67
Bańbuła P, Kotuła A, Przeworska J, Strzelecki P (2016) Which households are really financially distressed: How micro data could inform the macroprudential policy. In: IFC Bulletins chapters, Bank for International Settlements (eds), Combining micro and macro statistical data for financial stability analysis, vol 41. Basle
Berge TJ, Jordà Ò (2011) Evaluating the classification of economic activity into recessions and expansions. Am Econ J Macroecon 3(2):246–277
Betz F, Opricǎ S, Peltonen TA, Sarlin P (2014) Predicting distress in European banks. J Bank Finance 45:225–241
Buckmann M, Gallego Marquez P, Gimpelewicz M, Kapadia S, Rismanchi K (2023) The more the merrier? Evidence on the value of multiple requirements in bank regulation. J Bank Finance 149:106753
Candelon B, Dumitrescu E, Hurlin C (2012) How to evaluate an early-warning system: toward a unified statistical framework for assessing financial crises forecasting methods. IMF Econ Rev 60(1):75–113
Cantú C, Gambacorta L, Shim I (2020) How effective are macroprudential policies in Asia-Pacific? Evidence from a meta-analysis. In: Measuring the effectiveness of macroprudential policies using supervisory bank-level data, BIS Papers No 110, February
Cerutti E, Claessens S, Laeven L (2017) The use and effectiveness of macroprudential policies: new evidence. J Financ Stab 28:203–224
CGFS (2016) Experiences with the ex-ante appraisal of macroprudential instruments. CGFS Paper No. 56
Chen J, Columba F, Beaumont C (2016) Macroprudential and monetary policy interactions in a DSGE model for Sweden, IMF Working Paper WP/16/74
Chen Y, Mathä T, Pulina G, Schuster B, Ziegelmeyer M (2020) The Luxembourg Household Finance and Consumption Survey: results from the third wave. Banque centrale du Luxembourg Working Paper 142
Detken C, Weeken O, Alessi L, Bonfim D, Boucinha M, Castro C, Frontczak S, Giordana G, Giese J, Jahn N, Kakes J, Klaus B, Lang J, Puzanova N, Welz P (2014) Operationalising the countercyclical capital buffer: indicator selection, threshold identification and calibration options. ESRB Occas Pap Ser 5:1–95
Drehmann M, Juselius M (2014) Evaluating early warning indicators of banking crises: satisfying policy requirements. Int J Forecast 22(3):493–518
Epure M, Mihai I, Minoiu C, Peydró J-L (2018) Household credit, global financial cycle, and macroprudential policies: credit register evidence from an emerging country. IMF Working Paper WP/18/13
ESRB (2017) The ESRB handbook on operationalising macroprudential policy in the banking sector. Eur Syst Risk Board. https://doi.org/10.2849/04791
Fawcett T (2006) ROC graphs with instance-varying costs. Pattern Recogn Lett 27(8):882–891
Ferreira Filipe S (2018) Housing prices and mortgage credit in Luxembourg. BCL Working Paper 117. February 2018
Gambacorta L, Murcia A (2020) The impact of macroprudential policies in Latin America: an empirical analysis using credit registry data. J Financ Intermed 42:100828
Garrido F, Verbeke W, Bravo C (2018) A Robust profit measure for binary classification model evaluation. Expert Syst Appl 92:154–160
Giordana G (2016) Welfare and stochastic dominance for the measurement of banks’ domestic systemic importance: analytical framework and application. Int J Financ Econ 21:192–208
Giordana G, Ziegelmeyer M (2020) Stress testing household balance sheets in Luxembourg. Q Rev Finance Econ 76:115–138
Girshina A, Mathä TY, Ziegelmeyer M (2017) The Luxembourg Household Finance and Consumption Survey: results from the 2nd wave, BCL Working Paper 106
Godoy de Araujo DK, Barroso JBRB, Gonzalez RB (2020) Loan-to-value policy and housing finance: effects on constrained borrowers. J Financ Intermed 42:100830
Górnicka L, Valderrama L (2020) Stress testing and calibration of macroprudential policy tools. IMF Working Paper WP/20/165
Grodecka A (2020) On the effectiveness of loan-to-value regulation in a multiconstraint framework. J Money Credit Bank 52(5):1231–1270
Gross M, Población García FJ (2017) Assessing the efficacy of borrower-based macroprudential policy using an integrated micro-macro model for European households. Econ Model 61:510–528
Hallissey N, Kelly R, O'Malley T (2014) Macro-prudential tools and credit risk of property lending at Irish banks. No. 10/EL/14. Central Bank of Ireland
Hanley JA, McNeil B (1982) The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143(1):29–36
Hodula M, Melecký M, Pfeifer L, Szabo M (2023) Cooling the mortgage loan market: the effect of borrower-based limits on new mortgage lending. J Int Money Financ 132:102808
Hsieh F, Turnbull BW (1996) Nonparametric and semiparametric estimation of the receiver operating characteristic curve. Ann Stat 24(1):25–40
Igan D, Kang H (2011) Do loan-to-value and debt-to-income limits work? Evidence from Korea. IMF Working Paper WP/11/297
Jordà Ò (2012) Discussion of ’Anchoring countercyclical capital buffers: the role of credit aggregates’. Int J Cent Bank 7(4):241–259
Jordà Ò, Taylor A (2011) Performance evaluation of zero net-investment strategies. NBER Working Paper, No 17150
Jurča P, Klacso J, Tereanu E, Forletta M, Gross M (2020) The effectiveness of borrower-based macroprudential measures: a quantitative analysis for Slovakia. IMF Working Paper WP/20/134
Kaminsky G, Lizondo S, Reinhart C (1998) Leading indicators of currency crises. IMF Staff Pap 45(1):1–48
Kim J, Choi K, Kim G, Suh Y (2012) Classification cost: an empirical comparison among traditional classifier, cost-sensitive classifier, and MetaCost. Expert Syst Appl 39(4):4013–4019
Kuttner KN, Shim I (2016) Can non-interest rate policies stabilize housing markets? Evidence from a panel of 57 economies. J Financ Stab 26:31–44
Leika M, Marchettini D (2017) A generalized framework for the assessment of household financial vulnerability. IMF Working Paper WP/17/228
Lim CH, Costa A, Columba F, Kongsamut P, Otani A, Saiyid M, Wezel T, Wu X (2011) Macroprudential policy: What instruments and how to use them? Lessons from country experiences. IMF Working Paper WP/11/238
Makdissi P, Wodon Q (2002) Consumption dominance curves: testing for the impact of indirect tax reforms on poverty. Econ Lett 75(2):227–235
Malovaná S, Hodula M, Gric Z, Bajzik J (2022) Borrower-based macroprudential measures and credit growth: How biased is the existing literature? Preprint version available at Research Square. https://doi.org/10.21203/rs.3.rs-1811177/v1
Meriküll J, Rõõm T (2020) Stress tests of the household sector using microdata from survey and administrative sources. Int J Cent Bank 16(2):203–248
Nier E, Popa R, Shamloo M, Voinea L (2019) Debt service and default: calibrating macroprudential policy using micro data. IMF Working Paper WP/19/182
Richter B, Schularick M, Shim I (2019) The costs of macroprudential policy. J Int Econ 118:263–282
Sangaré I (2019) Housing sector and optimal macro-prudential policy in an estimated DSGE model for Luxembourg. BCL Working Paper 129
Acknowledgements
This paper should not be reported as representing the views of the BCL or the Eurosystem. The views expressed are those of the authors and may not be shared by other research staff or policymakers in the BCL, the Eurosystem or the Eurosystem Household Finance and Consumption Network. We give special thanks to Paolo Guarda. In addition, we would like to thank Giuseppe Pulina, Emmanuel Thibault, Martial Dupaigne, our colleagues from the Financial Stability Department and internal seminar participants for useful comments.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Tables and figures for debt burden ratios
Appendix B: Signals approach—robustness analysis
This appendix provides a robustness analysis of the signals approach exercise, using an alternative criterion to identify financially vulnerable households. The benchmark exercise in the main text focused on the impact of household financial vulnerability on banks, identifying vulnerable households as those with a loss given default (LGD) greater than zero. Instead, the robustness exercise focuses on those households that could potentially face problems servicing their debt, identifying vulnerable households as those with a probability of default (PD) greater than zero. PD > 0 is a less restrictive criterion, since only some households with PD > 0 have LGD > 0, while all households with LGD > 0 have PD > 0 (see Sect. 3).
As in the benchmark exercise (LGD > 0), using individual ratios is less effective than a classification rule combining several ratios. However, looking at the individual ratios, the LTV ratio outperforms all the other ratios in terms of AUROC (Table 12). The optimal limits for the LTV ratio are 185%, 95% and 75% depending on the loss function (\(\theta\) equal to 0.25, 0.5 and 0.75, respectively). Policies that combine ratios are less effective than in the benchmark exercise (Table 13) except for rule 1. Moreover, optimal limits are outside the legal range, in particular for the MDI ratio (Table 14).
Figure 7 depicts the Receiver Operating Characteristic (ROC) curves for each debt burden ratio in the adverse scenario. The ROC curve plots type I and type II classification errors for all the limits evaluated in the grid search. The y-axis reports the True Positive Rate (1 – type II error) and the x-axis reports the False Positive Rate (type I error). The origin represents the highest value considered for the limit (misses all vulnerable households but avoids misclassifying non-vulnerable households). Moving away from the origin along the dashed 45° line, the value of the limit declines, reducing type II errors but raising type I errors. Therefore, the line represents a poor performance (linear combinations of the maximum type II error at the origin with the maximum type I error at the top right corner).
Source: Own calculations based on the 3rd wave of the LU-HFCS; data are multiply imputed and weighted; the classification errors are calculated by pooling all the implicates
Receiver operating characteristic (ROC) curves of individual debt burden ratios in the adverse scenario.
In panel (a) of Fig. 7, the condition variable is unity for households with PD > 0. On this definition of financial vulnerability, the DI ratio (yellow), MDI ratio (green), DSI ratio (grey) and LTV ratio (red) deviate from the 45° line and therefore, are effective at identifying financially vulnerable households. For all ratio, the ROC rises below the 45° line near the origin. Thus, as the limit is lowered from very high levels, type I error increases more rapidly than type II error diminishes. This suggests that high limits on debt burden ratios are ineffective at identifying financially vulnerable households. However, the ROC soon jumps above the 45° line, suggesting that at lower levels of the limit, type II error does decline at least as rapidly as type I error increases. This suggests that, in the adverse scenario most financially vulnerable households are characterized by debt burden ratios at the centre of the evaluated range.
In panel (b) of Fig. 7, the condition variable is unity for households with LGD > 0. The LTV ratio performs slightly worse than with PD > 0 in panel (a), while the other three ratios perform better.
Appendix C: Unemployment shock
The unemployment shock is likely to focus on households or individuals with certain characteristics. To allow for the non-uniform distribution of the shock, we estimate a logit model for the probability that an individual is unemployed using survey data for all household members who were part of the active work force in the 2018 wave (Table 15).Footnote 35
The vector of explanatory variables includes the following individual and household characteristics: gender, age, country of birth, marital status, highest educational attainment, household size, homeowner/tenant status, net wealth quintile. Explanatory variables also include the following characteristics that were only available in the 2018 wave: region dummies, current labour status or (if unemployed) previous labour status, language skills and current sector of employment (based on NACE codes) or previous sector of employment (if unemployed). Standard errors are clustered at the household level.
The unemployment rate observed across all individuals in the active work force covered by the 2018 HFCS wave was 4.7%.Footnote 36 After estimating the logit model, we adjust the intercept term to simulate the average unemployment rate increasing to 12%. Following Albacete and Fessler (2010) or Meriküll and Rõõm (2020), for each individual we draw a random number from the uniform distribution over the interval (0,1) and compare it to the logit-determined probability for the given individual to determine whether he or she becomes unemployed. We then adjust household net income as described in Sect. 3 and recalculate the financial margin, probability of default and loss given default at the household level. Sections 4.2 and 5 report estimated statistics for the adverse economic scenario that are averages across 1000 Monte Carlo iterations of this process. Each iteration requires simulating the employment status of every active individual.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Giordana, G., Ziegelmeyer, M. Using household-level data to guide borrower-based macro-prudential policy. Empir Econ 66, 785–827 (2024). https://doi.org/10.1007/s00181-023-02477-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-023-02477-9