No Reason to Worry About German Mortgages? An Analysis of Macroeconomic and Individual Drivers of Credit Risk

The great financial crisis has highlighted that housing markets can pose significant risks to financial stability. Housing price booms fueled by excessive mortgage credit growth and a deterioration of credit standards contributed to a buildup of systemic risks not only in the United States but also in several European countries. Subsequently, when house prices were falling and this risk materialized, large credit losses destabilized the financial system and entailed immense negative spillovers to the real economy. Although imprudent securitization and cross-border trading in such products were important channels of risk propagation in the national and the global financial systems, the national banking sectors were mostly affected by housing downturns due to the large direct exposures of banks to residential real estate (Goodhart and Hofmann 2007).

In contrast to several other countries, the housing market in Germany has not experienced a full-blown bust during the great financial crisis as residential real estate (RRE) price and credit stock growth rates remained close to zero. In 2010 residential real estate prices and mortgages volumes started to grow again and have remained on an upward trajectory since then. Even in the recent economic downturn induced by the COVID-19 pandemic, the RRE market in Germany has so far proven robust. Despite the ongoing pandemic, by mid-2020 the RRE price and credit growth dynamics did not show any signs of deceleration. Although institutional features of the German residential real estate market are generally considered to contribute to its stability, they do not make it immune to adverse macroeconomic shocks. In the past, economic stress episodes have led to a significant increase in credit risk in banks’ residential mortgage portfolios. For instance, at the beginning of the millennium the loss rates in German RRE mortgage portfolios more than tripled, as a major macroeconomic shock triggered by the burst of the dot-com bubble hit the German economy which was already weakened by structurally high unemployment rates and weak house price development.

While the data basis remains fragmented, there are now various indications that the ongoing house price inflation trend exhibits increasing upward pressure on loan-to-value (LTV) at origination as borrowers need to finance higher acquisition prices. In particular an increasing share of loans with LTVs at origination above 100% can be observed. At the same time, a loosening of lending standards, including a relaxation of LTV constraints, has been identified as an important component of earlier international housing bubbles and busts (e.g. Cerutti et al. 2017).

Yet, up to now a comprehensive analysis of credit risk drivers for RRE lending in Germany has been hindered by the lack of detailed data on LTV ratios and other lending standards. This data gap was considered so relevant that it was pointed out in a public warning issued by the European Systemic Risk Board (ESRB) to Germany as an obstacle to the identification of risks to financial stability.¹ This paper makes a first attempt at closing these analytical gaps and contributes to an improved understanding of RRE-related risks in Germany in various dimensions.

First, we estimate a regional panel VAR model in order to demonstrate the macroeconomic link between foreclosure rates, house price dynamics and the unemployment rate in Germany. The results confirm the view that the favorable macroeconomic environment is supporting the low default and foreclosure rates currently observable in Germany. At the same time, the results also suggest that the RRE credit performance is not immune to macroeconomic recessions and that a significant increase in default rates can be expected if unemployment surges and house price growth undergoes a sharp reversal.

Second, we run additional microeconomic regressions on German household data to shed light on which borrower characteristics can explain future credit performance. Our results suggest that mainly income and financial wealth metrics explain default rates in Germany. This contrasts with studies in the Anglo-American context, which often focus on the role of LTV ratios as a “strategic” default driver.

Third, we calibrate a structural model to project credit losses in the RRE mortgage portfolios for the entire German banking sector, explicitly accounting for past changes in house prices as well as historical LTV ratios at origination. Moreover, we take into account structural features of the German housing market and other important aspects of RRE lending in Germany. In the event of a surge in unemployment and a strong reversal of house price growth, our model predicts a significant increase in mortgage losses. While it is straightforward to show that high-LTV mortgages will generally make a disproportionately large contribution to losses, our results indicate that the more significant losses are only realized once lower LTV mortgages are affected as well. As such, it appears unlikely that high initial loan-to-value (ILTV) mortgages are solely responsible for causing a systemic RRE crisis.

Fourth, the obtained bank-by-bank loss projections can ultimately be used to assess system-wide first-round stress effects in the German banking sector. We thus contribute to understanding the extent to which the housing boom in Germany in recent years contributed to a buildup of systemic risks in the banking sector.

It is important to note, however, that we focus our analysis on RRE lending to private households but exclude RRE mortgages to enterprises as well as other commercial real estate (CRE) loans which are driven by different risk transmission mechanisms.

The reminder of the paper is organized as follows. In the next section we present stylized facts about the German housing and mortgage markets and discuss their characteristic institutional features. We subsequently discuss the empirical literature in the field of mortgage credit risk in Section 3. Sections 4 and 5 present our analytical framework and the underlying data. We discuss our results in Section 6, while Section 7 derives related (macroprudential) policy implications. Finally, Section 8 concludes.

There is ample empirical research on credit risks in the US mortgage market driven by the experience of the subprime crisis and supported by the availability of granular mortgage data. Empirical evidence using granular mortgage data for European RRE markets is relatively less voluminous but rapidly growing. Overall, these studies provide general guidance with respect to modeling of the key mortgage credit risk components, i.e. the PD and the LGD.

Regarding the PD, studies of the US mortgage market consider two factors as the main triggers of default: negative house price shocks and income shocks. A large part of the literature for the US can accordingly be interpreted in the light of the “equity” vs “ability to pay” default hypotheses (Jackson and Kaserman, 1980). Generally speaking, a negative impact of RRE prices on the PD can be interpreted as evidence for the equity hypothesis while an unemployment shock effect lends support to the “ability to pay” hypothesis. According to the “equity” hypothesis, borrowers’ default decision will depend solely on the net housing equity value. Such strategic household defaults are modeled for instance by Cocco and Campbell (2004). In contrast to the “equity” hypothesis, the “ability to pay” hypothesis predicts that borrowers will default when they are hit by an income shock that renders their income insufficient to service their mortgage. Using Australian household-level data for the years 2006 and 2010, Read et al. (2014) provide consistent evidence and show that the probability to miss a payment is particularly high for households with relatively high debt service ratios. Fuster and Zafar (2015) show for the US that the delinquency rate drops more than half when cutting the required payment by half.

Findings of numerous studies advocate in favor of the so called “double-trigger” hypothesis. Deng et al. (2000) and Elul et al. (2010) are two studies that give empirical support to the hypothesis that simple option models in the spirit of the “equity” hypothesis are not sufficient to model default rates. The negative relationship between house price changes and PDs might stem from the fact that households with negative income shocks find it easier to avoid personal insolvency by selling their house in a boom market rather than during a bust (compare Schelkle, 2018). The “double-trigger” hypothesis also plays a role in stress-testing models: Djoudad (2012), Dey et al. (2008) and Faruqui et al. (2012) use household micro-data to assess the vulnerability of households in an adverse scenario. Although focusing on debt service variables as the main underlying factor of the propensity to default, housing wealth - or more broadly total wealth - is shown to play a significant role. Other studies use debt-to-income or debt service variables from household surveys to identify vulnerable households and combine these indicators with housing wealth indicators to approximate potential risk exposures (see e.g. Deutsche Bundesbank 2013; Ampudia et al. 2016; Albacete et al. 2014).

Our paper relates to the broader “equity” vs “ability to pay” credit risk literature and sheds further light on the role of institutional features. In general, strategic default decision are less likely to be relevant in Germany given the full recourse character of mortgages. Therefore, house price shocks are unlikely to be the single driving force behind mortgage defaults. This view is confirmed by our empirical model, which correlates mortgage defaults at the German state level not only with house price dynamics but unemployment shocks as well.

Turning to the modeling of LGDs, approaches vary depending on the degree of data granularity and specificities of the considered mortgage markets. Using macro-data to calibrate a model of mortgage losses, Hott (2015) show that loss rates in the US and Switzerland are related to house price level, the LTV ratio of mortgages, interest rates, and the unemployment rate. Various empirical studies using granular mortgage data for European RRE mortgage markets confirm these relationships (see Rodriguez and Trucharte (2007); Dietsch and Welter-Nicol (2014) and Gaffney et al. (2014) for the Spanish, French and Irish housing markets respectively). Furthermore, empirical studies modeling US loss severities indicate the predominant importance of LTV for determining LGDs (e.g. Clauretie and Herzog (1990); Lekkas et al. (1993); Calem and LaCour-Little (2004); Pennington-Cross (2010), and Qi and Yang (2009)). For European countries, similar evidence is provided by studies with different modeling approaches (e.g. Ampudia et al. 2016; Siemsen and Vilsmeier 2017; Hott 2015; Dietsch and Welter-Nicol 2014). Ampudia et al. (2016) find that the LGD at the level of the household is particularly sensitive to the value of the house. Siemsen and Vilsmeier (2017) is one of the few studies looking at the German residential real estate mortgage market. The study assess the impact of a severe decline in house prices on the solvency of German less significant institutions taking the link between LTV and LGD into account. However, lacking sufficient data for the calibration, they derive the meta-dependency between LTV and LGD based on a review of the literature. Focusing on the determinants of the recovery rates of a portfolio of 1,236 defaulted mortgages in Germany, Ingermann et al. (2016) show that collateralization of a loan is the major driver of the recovery rate for retail loans.

As to the LGD component of our model, our approach is linked to the literature underscoring the relevance of LTV for credit risk. Our paper is the fist to present evidence on the impact of historical LTV on the expected losses in RRE mortgage portfolios of German banks.

Further, our paper relates to the stress-testing literature and proposes a risk assessment methodology. Since the global financial crisis, models of mortgage credit risk have gained increasing attention from policy makers and supervisors who rely on them in prudential stress-testing exercises. The purpose of stress tests is to assess the impact of severe macroeconomic and financial shocks on the resilience of individual financial institutions (microprudential stress tests) or entire financial systems (macroprudential stress tests). A number of national authorities have developed stress-testing tools focusing specifically on mortgage credit risk that are either run autonomously (see Albacete et al. 2014 for Austria, Faruqui et al. 2012 for Canada, to name a few) or are integrated in a broader stress-testing framework (e.g. Daniëls et al. 2017 for the Netherlands). Since 2018, the ECB has been applying a macro-micro model created for the purpose of macroprudential top-down stress testing of the banking sector (Budnik et al. 2019; EBA 2021). We contribute to the literature by developing a methodology where modeling of mortgage PDs and LGDs is accurately tailored to the specificities of the domestic housing and mortgage market.

Last but not least, our analysis is related to the growing empirical literature on borrower-based macroprudential measures. Existing papers measure the effects of lending standards on PDs and LGDs indirectly, by linking them to various metrics of household vulnerability reflecting either the liquidity position of households (e.g. Michelangeli and Pietrunti 2014; Gross and Poblaciôn 2017) or both their liquidity and solvency situation (e.g. Albacete and Lindner 2013; Ampudia et al. 2016; Barasinska et al. 2021). We contribute to this new research area by analyzing the role of lending standards in a more straightforward manner by evidencing the direct link between observed payment arrears to LTV, DTI, and DSTI metrics at the household level, and by linking LTVs to the LGDs in banks’ loan portfolios.

To our knowledge, this is the fist study to present an comprehensive assessment of mortgage credit risks in Germany from a microprudential and macroprudential perspective.

Our analytical framework is based on three building blocks. First, we estimate a regional Panel Vector Autoregressive (PVAR) model to evidence the macroeconomic link between default rates, house price dynamics and the unemployment rate in Germany. Second, based on the German Panel on Household Finances (PHF), we run microeconometric regressions to investigate the role of debt affordability and LTV ratios on future loan defaults. Third, taking into account the results from the above analyses, we calibrate a structural model to assess systemic risks in the RRE mortgage portfolios of the entire German banking sector.

In line with our analytical framework, we discuss below our results with respect to the role of macroeconomic and microeconomic default and loss rate drivers as well as the overall expected credit losses in a severely adverse scenario.

Understanding the underlying drivers of RRE mortgage risk is crucial from a policy perspective as evidenced by the Great Financial Crisis. Yet from a macroprudential point of view, it is important to assess the risks not only from a “banking” perspective but also from a “household” perspective. For the former it is important to assess the banks’ ability to absorb losses under stress which assures a continuous functioning of the financial system to supply the real economy with credit and liquidity. For the latter, the focus is on households’ ability to withstand housing market busts and related macroeconomic growth and income shocks.

Analysing both perspectives – that of households and of banks – is important as the results of our analyses indicate that the policy options suitable to strengthen the resilience under both dimensions might differ.

First, our results show that the direct credit losses in the RRE mortgage portfolios following a general market downturn are sizable but still appear manageable for the banking sector. However, our results on expected losses have to be seen in the context of a broader macroeconomic recession or a financial crisis that goes beyond the housing sector. This is particularly relevant as an isolated residential real estate shock without broader macroeconomic implications is an unlikely scenario. The results from the macroeconomic PVAR analysis suggest that significant spillover effects exist from the housing market on macroeconomic variables such as, for instance, the unemployment rate. Rising unemployment in turn might induce higher loss rates not only for RRE mortgage portfolios but also for other loans to households. Moreover, a housing market downturn is likely to occur in parallel with a significant decline in corporate credit portfolios, in particular for RRE and CRE loans to enterprises and loans to the construction sector in general. This broader macroeconomic view is supported by various empirical studies for the United States⁴⁷ and other OECD countries.⁴⁸

Hence, the overall effects are likely to go beyond the RRE mortgage portfolios and the banking system will face significantly increasing capital needs not only from increasing credit losses but also due to rising risk weights. Therefore, the required level of the capital buffers needed to withstand an adverse RRE scenario should be assessed taking a broader view. Yet from a pure “banking” perspective, broadly based capital-oriented policy measures, like the countercyclical capital buffer, are likely to be the first – because least intrusive – option to strengthen the resilience of the banking system to adverse RRE market shocks.

In addition, again from a “banking” perspective, our additional analyses in Section 6.4 question whether losses during a severe housing crisis are overwhelmingly driven by high-ILTV mortgages, as long as high-LTV loans are not financing particularly vulnerable households.

Hence, our results call for an extension of the existing macroprudential toolkit in Germany - currently consisting only of potential LTV caps and an amortization requirement.

Finally, our analysis highlights the importance of a good reference dataset with respect to lending standards for accurate assessment of vulnerabilities of individual banks or certain banking segments to RRE shocks. Data shortcomings may ultimately hinder the assessment of capital needs required to strengthen the resilience of the banking sector. Similarly, further in-depth analyses of driving forces of credit defaults at household level require significant improvements to the reference dataset. Results of empirical studies from other countries can only give indicative guidance as the characteristic institutional features of national mortgage markets are likely to induce asymmetric effects. Seeing the importance of an appropriate reference dataset for the monitoring of risks to financial stability, the German Financial Stability Committee already recommended in 2015 that the Federal Government creates a legal basis for the regular collection of data on housing loans.⁴⁹ Following this recommendation, the Federal Ministry of Finance finally created in February 2021 a legal basis for a regular collection of bank-level data on lending standards.⁵⁰ The data collection is scheduled to start in 2023.

The German housing market has shown a strong performance over the past decade, including a remarkable resilience even in the light of the current COVID-19 economic recession. On the other hand, elevated property valuations, dynamic credit growth rates and uncertainty regarding the recent evolution of credit lending standards call for a rigorous analysis of the potential risks in the banks’ credit portfolios. The favorable environment for mortgage financing over the last decade may conceal the buildup of vulnerabilities to adverse developments in the housing and labor markets.

In our paper, we analyse the various risk dimensions in the RRE mortgage market. Our PVAR analyses suggest significant links between regional mortgage foreclosure rates and the macroeconomic environment with most of the medium-term variation driven by local unemployment rates. Additional microeconomic household regressions suggest that debt service and affordability metrics as well as available financial buffers can better explain observable default rates than initial or current LTVs. Taken together, these results suggests that mortgage default decisions in Germany appear to be driven mostly by income and liquidity rather than the result of strategic decisions.

Building on the above insights, we estimate expected RRE mortgage credit losses for the entire German banking sector using a structural model. While currently the favorable macroeconomic environment is supporting the low default rates and credit losses, a combined shock to domestic housing prices and labor markets could induce significant additional credit losses in the mortgage portfolios. While our results suggest that the isolated stress effects on capital ratios would be manageable for most of the banks, the cumulative losses could still contribute substantially to the erosion of the risk-bearing capacity of the German banking sector. This is of particular relevance as the historical experience has shown that real estate-related recessions are particularly costly in terms of economic activity. At the same time, our results suggest that high-LTV mortgages are not the sole contributors to systemic risk in Germany as losses are likely to go beyond that particular lending segment during a severe housing crisis. Furthermore, some banks may be more susceptible to housing market shocks than estimated if their lending practices are less conservative than suggested by the available data. The prospective data collection on housing loans in Germany foreseen for 2023 will allow an even more differentiated study of the vulnerability of the banking sector.

Finally from a policy perspective, our results indicate that a combination of capital- and borrower-based macroprudential instruments appears most suitable to tackle buildups of financial stability risks stemming from the RRE market. In particular, borrower-based measures should most likely combine LTV and income-related restrictions. In addition, for a reliable calibration of such instruments, the timely creation of a representative and granular dataset is essential.