The effects of changes in the mortgage financing code of conduct on risk profiles of newly issued mortgages : a study of the Dutch mortgage market

(1)

MSc Thesis

The Effects of Changes in the Mortgage Financing Code of Conduct on

Risk Profiles of Newly Issued Mortgages: A Study of the Dutch

Mortgage Market

Fabian K. Ruinard (10269487)

Specialization: Finance & Real Estate Finance Institution: University of Amsterdam Supervisor: Dr. M.I. Dröes

Second supervisor: Dr. F.P.W. Schilder

(2)

2

Abstract

After the financial crisis, the Netherlands adopted policy changes with regards to mortgage issuance. Rules have been applied in both 2007 and 2011 with regards to maximum loan-to-value ratios and tax deductibility of certain amortization schemes. This paper measures the effectiveness of the modified mortgage issuance policy on the risk profiles of newly issued mortgages by the means of difference-in-differences and logistic regressions. A dataset of approximately 2,700 households in the Netherlands over the period 1995-2015, which consists out of general, financial and residential information has been used. The main findings are that the changes in the mortgage issuance policy resulted in reduced loan-to-value ratios, loan payments-to-income ratios and a reduction of the amount of non-amortization mortgages. With regards to loan payments-to-income ratios only statistically significant reductions for the policy changes in 2007 have been found. Therefore, the conclusion that can be drawn from this research is that the changes in the mortgage issuance policy resulted in a lower risk profile on newly issued mortgages. Next to that, the effectiveness of the mortgage issuance policy changes increased when the policy became binding instead of being a guideline. This research also shows that lower loan-to-value ratios result in lower house prices which is an implication of the policy that should be taken into account.

Statement of Originality

This document is written by Student Fabian Ruinard who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

1. Introduction

1.1 Introduction and research question

One of the driving factors of the subprime mortgage crisis has been the relaxed underwriting criteria that mortgage issuers had in the period 2004 until 2007 (Brunnermeier, 2009). Households were able to borrow too much which resulted in high risk profiles on outstanding mortgages on banks’ balance sheets. Due to the steady increase in house prices over the period 2002-2007, most mortgagees and mortgagors perceived the residential real estate market as a safe market with little risk. However, a sharp increase in subprime mortgage lending (Demyanyk and van Hemert, 2009) in combination with an increase in the mortgage backed security (MBS)1 market made the financial institutions take on too much risk on their balance sheets.

Eventually in 2008, the optimism the financial institutions had with regards to the housing market and its stability came to an end (Lim, 2008). Due to the high amount of subprime mortgage lending, many households were forced to default on their loans after the housing prices dropped slightly. The high loan-to-values in combination with little income or collateral would quickly result in a default for underwater mortgages. The defaults triggered the house prices to drop even more which eventually resulted in a further increase in defaults. The mortgage defaults led to banks and other large financial institutions to collapse and having to be bailed out by governments in order to evade a further collapse of the financial system (Reinhart and Rogoff, 2009).

In the Netherlands, the government made a total of €20 billion available to protect financial institutions from bankruptcy (Rijksoverheid, 2016). A total of €13.75 billion has been used by the corporations Aegon (3 billion), ING (10 billion) and SNS (0.75 billion). The collapse of the financial markets and the bailouts resulted in a change in policy in the Netherlands with regards to mortgage issuance. Since 2003 a code of conduct has existed in the Netherlands which has as main objective to realize to supply norms and values for a responsible mortgage issuance policy (NVB, 2016). The code of conduct is known as the Gedragscode Hypothecaire

Financieringen (GHF). The perceived increase of risk on the mortgage market resulted in the

GHF being modified in both 2007 and 2011. These modifications concerned mainly the percentage of the income that flows to mortgage payments, rules regarding amortization of mortgages and maximum loan-to-value ratios on outstanding and newly issued mortgages (Homefinance, 2015). The GHF was not binding until 2011 and therefore the modifications in 2007 were only seen as a guideline for mortgage issuers. From 2011 and further, mortgagees were obliged to issue mortgages under the norms and values the GHF prescribes.

1_{MBSs are complex financial products that consist of bundled mortgages into one package which can then be sold as a}

type of bond where the owner would be entitled to the mortgage payments. Therefore the risk of bad loans could be shifted to the buyer of the package and mortgage issuers would have moral hazard which resulted in more and more relaxed lending standards.

(5)

5

Existing literature about the effectiveness of mortgage issuance legislation is scarce and there has been no consensus on the subject (also see Section 2). Therefore the results of this research can be of importance for governments and policy makers to see whether the policy changes in the Netherlands resulted in the desired reduction of risk on newly issued mortgages. This thesis analyzes the effects of the changes in the GHF on different variables (i.e. loan-to-value, loan payments-to-income and mortgage types) which are all perceived valid measures of the risk of an outstanding mortgage. Given the above discussion, the main research question of this thesis is:

Have changes in the Dutch code of conduct (GHF) regarding mortgage financing led to a lower risk profile on newly issued mortgages in the Netherlands?

1.2 Relevance of the research question and results

This study looks at variables such as the loan-to-value (LTV) ratio, loan payments-to-income (LPTI) ratio and mortgage types to investigate whether the risk profiles actually decreased after the policy changes. Tsatsaronis and Zhu (2004) show that LTV ratios are an important positive determinant of house prices and that a ceiling on LTVs can determine the ability of banks to lend against real estate collateral. In this thesis a regression of the natural logarithm of the market values of houses on the Original Loan-to-Value (OLTV)2_{has been done next to the main research} to show the impact of LTVs on house prices which resulted in a significantly positive effect of 2.27% for every 0.10 percentage points increase in the original loan-to-value.

Campbell and Dietrich (1983) and Archer et al. (2002) show that the LTV ratio is an important determinant of mortgage default rates and therefore is a correct variable to use as a measure for a mortgage risk profile. Due to the fact that LTVs are a positive determinant of house prices, this research shows that if the changes in the GHF have a negative effect on the LTVs, this would have a negative effect on house prices. The decrease in house prices results in a loss in financial wealth to the Dutch residents but also a deflation of a possible bubble in combination with a lower default probability. With regards to default probability the same holds for the variables LPTI and mortgage type. Higher LPTI means a larger share of income that flows to mortgage payments. Therefore a reduced LPTI ratio equals a larger buffer for the mortgagor in case of a negative income shock which reduces the risk for the mortgagor and therefore the default rate. Mortgage types without amortization, which is a variable of interest in this study, mean that at the end of the loan term the mortgagor has to pay back the full loan amount. If the mortgagor would have amortized during the loan term, possible financing problems could become apparent and resolved earlier. Therefore both the LPTI and mortgage types influence the default rates so if there can be shown that the changes in the GHF reduced the LPTI and

(6)

6

amount of non-amortization mortgages there can be concluded that the risk profiles of outstanding mortgages have reduced. This reduction of risk can result in less default rates and therefore eventually higher welfare for the Dutch economy through less incurred losses by the mortgagor (Clauretie, 1987). By taking the reasons discussed in this paragraph into account, it is clear that the effectiveness of the policy is important through either a decrease in house prices and therefore a decrease in financial wealth,but also through a decrease in default rates which should eventually result in higher welfare.

This research looks at two points in time where changes in the GHF have been applied in the Netherlands, namely 2007 and 2011. As mentioned in Section 1.1, these points in time differ in the way that in 2007 the code was not binding and could be seen as a guideline for mortgage issuers. After 2011 the code became binding. By looking at the different effects of both time frames on the proposed variables which indicate the risk profile of a mortgage, the differences between binding and non-binding policy can be evaluated.

To answer the research question, survey panel data of approximately 2,700 households in the Netherlands over the period 1995-2015 have been used. Three difference-in-differences regressions have been done as well as a logistic regression. The method is suited due to the fact that two differences need to be distinguished. The first difference is the difference in time frame before the changes in the GHF and after. The second difference is the difference between the households who were not affected by the changes in the GHF and groups who were affected. The second difference is necessary to distinguish the regular effect of the financial crisis from the effect the GHF had on the risk profiles of the mortgages. Section 5 elaborates on the proposed methodology.

The results in this paper show that after the GHF changed in 2007, the original loan-to-value ratios (OLTV) decreased significantly by approximately 0.12 percentage points. Also the probability that a newly issued mortgage would be a non-amortization mortgage decreased by approximately 14%. When the code was modified again in 2011 and became binding, the OLTV ratios decreased by a significant 0.19 percentage points, whereas the non-amortization probability decreased by a significant 54%. The Loan Payments-to-Income ratio (LPTI) reduced by 0.10 percentage points as a result of the GHF modifications in 2007. The modifications in 2011 showed no significant reduction. There can be concluded that the changes in the GHF reduced the risk profile through the OLTV, the amount of newly issued non-amortization mortgages and the LPTI ratio. The fact that the coefficients on the 2011 changes are more negative than those for 2007 implies that the GHF becoming binding increased the effect. An additional regression in this thesis shows that the reduced loan-to-values decrease house prices as has been shown in existing literature. Therefore the policy changes do not come without a loss in housing wealth for the Dutch homeowners. However, the lower house prices can also be

(7)

7

seen as an objective to mitigate a potential bubble which would result in higher long-term welfare for the Netherlands.

The remainder of this thesis is structured in the following way: Section 2 provides information on the Dutch mortgage market and the GHF and how changes in the GHF have been applied over the years. Section 3 discusses existing literature on the subject and places this research within existing literature and explains how it contributes to the existing literature. Section 4 continues by describing the data used to test for the hypotheses and how the data has been prepared. Section 5 describes the methodology used to be able to test for the hypotheses and answer the research question. Section 6 elaborates on the results of the tests and the results are interpreted. Section 7 gives a conclusion and discussion of the thesis whereas Section 8 concludes the paper with limitations and opportunities for further research.

2. The Dutch mortgage market

2.1 Characteristics of the Dutch mortgage market

According to the CBS (2012), the financial wealth of Dutch households consisted for 60.2% out of housing in 2012. As of 2015 (CBS, 2015), 4.3 million households owned a home of which approximately 32% had a mortgage on which the loan exceeds the market value of the house. IG&H (2015) discloses in their Dutch mortgage market update that in 2015 a total of 62,000 mortgages have been sold in combination with a mortgage revenue of €253 billion. The average mortgage amount in the fourth quarter of 2015 was €252,000 which is an increase of 10% to a year earlier and approaches the level of 2008 and 2011.

The NVB (2014) has done research to the Dutch mortgage market and its characteristics. According to the NVB (2014), the Dutch mortgage market is characterized by high Loan-to-Value and Loan-to-Income ratios. For both ratios, the Netherlands ranked second after France when compared to other European countries in 2014. However, Fitch (2013) shows that the Netherlands rank among the lowest in Europe with expected default rates. The expected default rate in 2013 was approximately 3.5% in comparison to 6% in the UK, 9% in Italy and Spain, while Greece and Ireland have expected default rates of above 10%. Next to the low expected default rates, the Netherlands also have a relatively low number of foreclosures and low amount of recovery time after an unwinding process. However, the relatively low number of foreclosures can be explained by the fact that in the Netherlands the lender keeps the right to seek recourse against the mortgagor (ABN Amro, 2012). Therefore the mortgagee cannot simply hand over his house to the mortgagor in case the debt exceeds the value of the house. The high practical application of recourse in the Netherlands plays a part in the low number of foreclosures.

(8)

8

The NVB (2014) gives various risk mitigating factors for the opposing LTV and LTI ratios as compared to the expected default rates, foreclosures and recovery time from unwinding processes. The first factor is the amount of pension savings. The Netherlands has the largest amount of pension assets per capita worldwide. The pension system reduces mortgage risk due to the fact that Dutch citizens do not solely rely on the value of their house in the later stage of their life but they have a healthy pension savings account to rely on. Two other risk mitigating factors are the social security and the developed legal environment the Netherlands has. When a citizen has a certain income shock through for instance loss of job he can rely on unemployment benefits and other social security. The developed legal environment makes sure mortgage lenders have a strong position in relation to borrowers. Moody’s (2013) ranks the Netherlands the highest in Europe with regards to practical application of recourse and the legal strength for recourse as mentioned before. Another factor which reduces the risk for mortgagors on the Dutch mortgage market is the good administration of debt in the Netherlands through the credit registration office Bureau Krediet registratie. Lenders are able to see the indebtedness of mortgagors to assess their Debt-to-Income ratio (DTI). The DTI ratio gives insight in the amount of income that is indebted. Therefore banks can use the DTI ratio to have a better understanding of the riskiness the issuance of a mortgage entails for a certain mortgagor. The final important factor that reduces the mortgage risk in the Netherlands is the fact that most of the mortgages remain on the balance sheet of the lender. Therefore mortgagees cannot shift risks entirely to investors which reduces the risk of moral hazard among mortgage issuers which has been a big factor in the financial crisis of 2008 (Brunnermeier, 2009).

2.2 Non-competitive mortgage issuance environment

Next to the previously discussed risk reducing characteristics the Dutch mortgage market has, the market is also characterized by a non-competitive environment with regards to mortgage issuers (ACM, 2013). De Haan and Sterken (2011) conducted an analysis of the interest rate setting behavior of the four bank with the largest share in the Dutch mortgage market over the sample period 1997 – 2003. Their research shows no evidence which should indicate non-competitive price setting behavior. Mortgage issuers seem to be non-competitive where they are adjusting their interest rates to their funding costs. However in more recent figures, there seems to be a non-competitive environment on the mortgage market in the Netherlands. In 2013, the Authority Consumer and Market (ACM, 2013) has performed a study in order to analyze the competitiveness of the Dutch mortgage market to see whether the market can be considered oligopolistic. A reason to conduct the study was the fact that mortgage rates in the Netherlands were approximately 100 basis points higher than in surrounding countries. The ACM concluded that one of the main reasons behind the high margins are capacity restrictions banks face after

(9)

9

the financial crisis. To improve their capital ratios, banks have been lowering their loan amounts while they were increasing the interest rates. The reason why the banks were able to do that is due to the limited competition they face in the market. According to the research done by ACM (2013), the HHI index of the Dutch mortgage market was measured at 1,916 points while the C4 ratio, a second measurement for market concentration, was approximately 79.3% in 2013. ABN Amro, ING, Rabobank and SNS, the four banks with the largest share in the mortgage market, had a total market share of approximately 80% in 2012 which facilitated the charging of high interest rates.

Dijkstra et al. (2014) investigated the reasons behind the high mortgage rates in the Netherlands as well with a focus on the time period after the spring of 2009. The authors chose the spring of 2009 because at that moment the Dutch mortgage rates clearly became higher than the rates in the rest of Europe. Through a literature study Dijkstra et al. (2014) concluded that the high mortgage rates in the Netherlands have been the result of several factors which complemented each other. The high level of concentration in combination with barriers to entry the Dutch mortgage market resulted in the current players being able to earn high margins as discussed in the first part of this section. Also funding capacity restrictions such as the Basel III rules with regards to loan-to-deposit ratios made banks reluctant to finance new mortgages, especially mortgages with high loan-to-value ratios. The funding capacity restrictions created an incentive to therefore issue mortgages with higher rates. However, Dijkstra et al. (2014) conclude that the most important determinant of the high interest rates have been price leadership bans given by the European Commission in 2008. The price leadership bans banned the four largest mortgage issuing banks from offering lower rates than the three competitors with the lowest interest rates. After it became clear the price leadership bans resulted in the high mortgage pattern in the Dutch market, the bans were dismissed in 2013.

More recent figures supplied by IG&H (2015) show that the concentration on the Dutch mortgage market has reduced since 2013. IG&H measures the concentration by looking at the market share of the three largest banks (ABN Amro, ING and Rabobank) in comparison to the market share of insurers and pension funds. The market share in the mortgage market of the three largest banks has reduced steadily from approximately 63% in January 2013 to 54% in January 2014 and eventually to 48.5% in the fourth quarter of 2015. The insurers and pension funds saw their market shares increase from 24% in 2013 to 36.3% in 2015. IG&H (2014) concludes that the reduction of the market concentration has been caused by an increased number of suppliers on the market. They expect the increase in competition to have a negative effect on the mortgage rates in the future.

(10)

10

For this study the oligopolistic characteristics of the Dutch mortgage market in combination with the high interest rates should be taken into account when verifying the external validity of the results.

2.3 Gedragscode Hypothecaire Financieringen

With regards to norms and values on the issuance of mortgages in the Netherlands, the Dutch Association of Banks (NVB) and the Union of Insurers have imposed a code of conduct, the ‘’Gedragscode Hypothecaire Financieringen’’ (GHF), that should limit the risks and costs to society of mortgage financing (Rijksoverheid, 2011). The code has existed since January 1st₂₀₀₃ (NVB, 2002). The GHF is meant for every mortgage issuance on a non-commercial residential unit in the Netherlands as long as the mortgage issuing bank has signed the code of conduct. The main objective of this thesis is to test whether the changes in the GHF in 2007 and 2011 decreased the risk profile on newly issued mortgages.

2.3.1 GHF in 2007

The most important objective of the change in the GHF in 2007 has been to reduce the possibility of too high indebtedness of mortgagors (HomeFinance, 2015). The government noticed in the years preceding 2007 that mortgage issuing banks were issuing mortgages with too high loan-to-value ratios in comparison to the income the mortgagor earned, which could be labeled subprime (Rijksoverheid, 2008). The GHF subsequently made a change regarding the percentage of the income that goes to mortgage payments. Starting from 2007, mortgage issuers have to use the same percentages for every mortgage applicant. This means that households with higher incomes can have more of the income flowing to mortgage payments. Another change in the code has been the so called ‘’toetsrente’’ which is an interest rate percentage used to estimate for mortgages with a fixed interest term of less than ten years whether the mortgage is of a justified risk profile (AFM, 2016). The current ‘’toetsrente’’ interest rate is 5%. When the actual mortgage interest rate exceeds the ‘’toetsrente’’, then there has to be worked with the actual interest rate. The toetsrente has been established to prevent mortgagors to get in financial troubles when the interest rates increase after the fixed interest term.

Important to note is that the code of conduct was not binding in 2007. A mortgage issuer could deviate from the code if the issuer had for instance positive expectations about the income growth of the applicant in the future. The issuer was obliged, if there would occur a deviation, to give notice to the mortgagor and often make the mortgagor sign a statement in which the mortgagor admitted to have been given notice of the possible risks and the risk profile of the mortgage.

(11)

11 2.3.2 GHF in 2011

In 2011 the GHF became binding. This meant that mortgage issuers could not deviate from the code under any circumstances. The changes in 2011 were caused by the fact that the Netherlands was ranked number one in the world with mortgages issued with loan-to-values of more than 100% on a percentual base (NVB, 2012). Many of these mortgages also were non-amortization mortgages which made the mortgage market in the Netherlands very risky. The riskiness of the market made the government put additional pressure on the banks to tighten the GHF even more. After minister de Jager of Finances threatened the banks to make a new law that would put a maximum on loan-to-values the banks modified the code and made it binding in August 2011 (HomeFinance, 2015).

The most important change of 2011 has been the maximum level of LTV that was allowed to be issued. The maximum would become 104% of the market value of the property excluding the transfer tax. Another imposed modification of the code was with regards to the maximum amount of a loan that could be without amortization. From 2011 and beyond mortgage issuers were only allowed to issue to a maximum of 50% LTV non-amortization. Any loan percentage over the 50% of the value of the house must have an amortization scheme. An amortization scheme reduces the risk for the mortgagor in a way that the residual of the loan reduces during the term and the mortgagor is able to identify payment delinquencies of a mortgagee at an earlier stage than with an interest-only loan on which is not being amortized. The third change has been that in only very limited cases the mortgage issuer could deviate from the code of conduct with regards to norms on the income. Banks are not allowed anymore to make their own decisions on solvability of a household by, for example, taking future income growth into account.

2.3.3 GHF in 2013

Starting from 2013, additional rules have been imposed for the issuance of mortgages (NVB, 2013) which made the code less important due to the fact that the government made a law regarding mortgage financing. From January 1st_{2013 mortgagors will only be able to deduct} their mortgage interest payments from their taxable income if they are amortizing their mortgage by an annuity payment scheme. Another change in legislation from 2013 has been the maximum loan-to-value that is allowed to be lend to a mortgagor. From 2012 the maximum LTV equals 106% (including 2% transfer tax) and it will be reduced by 1% each year until it equals 100% in 2018. The new legislation imposed in 2013 only holds for mortgages issued after January 1st_{2013 or for mortgages issued prior which have been raised or refinanced after} January 1st_2013.

(12)

12

3. Literature review

This section discusses the relevant literature available with regards to mortgage issuance legislation. The section starts with a broader picture discussing research about why mortgage legislation is important and should be imposed. Then there will be continued with existing mortgage modification policies and the effectiveness of the policies. The section concludes with research on the topic of mortgage issuance legislation and the results of these policies after which hypotheses for this thesis are extracted and proposed.

3.1 The financial crisis and its causes

Demyanyk and Van Hemert (2009) show that decreased lending standards reduced the mortgage documentation, increased the debt-to-income ratios and played a big part in the housing boom preceding the subprime mortgage crisis. Therefore, the decreased lending standards can be seen as a cause of the eventual collapse of the market. The authors also show that these problems in the mortgage market were already apparent in 2005. Brunnermeier (2009) found that one of the major reasons behind the slackened underwriting standards has been the securitization of the mortgages into mortgage backed securities which made the value of the combined mortgages very complicated. Mayer et al. (2009) confirm the issue of too relaxed mortgage lending requirements and show that the amount of subprime mortgages consisting of too low documentation standards and too high LTVs and LTIs have increased significantly between 2003 and 2005. These mortgages with slackened underwriting standards were the most prominent factor in the rise in mortgage defaults after 2007 resulting in the collapse of the housing market. Crotty (2009) investigated the causes of the financial crisis of 2008 by means of a literature study and finds among other results the same conclusions as other academics on the subject. Crotty (2009) found that the large increase in household debt in the years preceding the crisis has been a major factor which resulted in a bubble on the housing market which eventually burst.

3.2 Subprime mortgages and predatory lending

From the preceding literature it is clear that one of the causes of the housing bubble was the subprime and predatory mortgage market. Engel and McCoy (2002) define predatory lending as issuing a loan designed to earn supernormal profits or a loan involving fraud or misleading practices. Many of these predatory loans involve high interest rates, high loan-to-values and features such as negative amortization. The definition of predatory lending is in accordance with the definition of subprime mortgage lending as imposed by Geltner et al. (2014) and Gramlich (2007) who both define a subprime mortgage as a mortgage loan in which the mortgagor had little documentation and equity in combination with a high loan-to-value. According to Bernanke

(13)

13

(2007) the subprime mortgage market flourished in the years preceding the crisis through mortgage backed securities which resulted in no risk for the mortgage issuers due to them being able to sell the mortgages immediately. Avery et al. (2007) confirm the flourishing of the subprime mortgage market by showing a fast increase in subprime mortgage lending and the increase in loans on non-owner-occupied properties in the years preceding the crisis. Next to that, Avery et al. (2007) show that the amount ‘’piggyback’’ mortgages has increased, which are mortgages where a second mortgage has been taken out at the same moment as the issuance of the first mortgage.

3.3 Actions from the governments/central banks

The preceding section shows that the amount of subprime and predatory lending increased in the years preceding the crisis. This section investigates literature with regards to whether governmental institutions tried to prevent the housing bubble to exist by either monetary policy or policy to reduce the amount of subprime mortgage lending. Taylor (2009) looked at the role of the government and interventions in the financial crisis that started in August 2007. Taylor (2009) provides evidence that the governments caused and worsened the financial crisis. The governments caused it by keeping interest rates too low and worsened it by supporting financial institutions and their creditors. He suggests to use a set of principles to follow in order to reduce the chance of a similar bubble in the future. Cobham (2012) concludes that also the central banks did not do enough to prevent the housing market bubble and therefore the financial crisis. The Federal Reserve, Bank of England and the European Central Bank all were too attached to their orthodox view that monetary policy should not be used to determine asset prices. Also the central banks said that lowering the interest rates would have maybe deflated the bubble a little bit but it would have had devastating effects on the other parts of the economy. Nier (2009) drew the same conclusion as Cobham (2012) that the central banks and supervising institutions did not put in enough effort to address and tackle the increasing exposure to risks in the housing market. Next to that, Nier (2009) also investigated the tools the central banks used in order to actually reduce the risk through reducing the amount of subprime mortgage lending and the increasing volume of credit derivatives. Nier (2009) concludes that with regards to credit derivatives such as mortgage-backed securities, the central banks failed to generate a framework for these securities in order to reduce the overall systemic risk. By increasing the role of financial regulation, central banks should be more incentivized to reduce systemic risk and the frequency financial crises occur. However, as Cobham (2012) already described, Nier (2009) is aware as well of the fact that the use of policies by central banks or supervising institutions can come at a cost through possible spillover effects on other asset markets.

(14)

14 3.4 Effects of legislation on house prices

Legislation or supervision can be a tool to be able to control the housing market to restrain it from forming a bubble again. Andrews (2010) has investigated the effect of more banking supervision on house prices and volatility and found significant negative coefficients. The increase of more banking supervision has led to stricter lending requirements which decreased the demand for housing and therefore the prices and volatility. Chervachidze (2011) looked at the determinants of capitalization rates to get a better understanding of the housing market and how the bubble of 2007 could inflate. Chervachidze (2011) found debt availability to be an important determinant of cap rates and therefore of house prices. A higher availability of debt, which can be seen as lower lending standards, inflates therefore house prices and can result in a bubble.

Francke et al. (2014) have derived an index which represents the credit conditions on the Dutch mortgage market. The credit conditions index which shows how changes in mortgage credit supply occur, apart from changes in income and interest rates. Their main findings show that during the period of 1995-2009 the supply of mortgage credit in the Netherlands increased steadily with an exception of 2007 immediately after the financial crisis started. The decreased underwriting standards increased house prices by approximately 32% during this period. After 2009, the supply of credit decreased which resulted in a decrease in house prices of 12% on average after 2009. The results of the paper by Francke et al. (2014) show the importance of credit availability on house prices in the Netherlands and show that the conclusion of Chervachidze (2011) also holds for the Netherlands.

Dell’Ariccia et al. (2008) look at reduced lending standards and therefore a higher availability of debt as a predictor of the number of loan applications. The result is a significant positive relation which also shows the importance of lending standards for housing demand. The risk of a bubble to burst is amplified by the research of Case et al. (1995). They show that periods with high default rates, which can be triggered by a bubble that has been created due to relaxed mortgage lending standards, decreases house price indices even more. So a bubble that bursts, bursts even harder through the defaults on mortgages. By taking the aforementioned literature into account one can see how important it is too be able to measure the effectiveness of policy on more restrictive mortgage standards. The next paragraph will serve to give a better understanding of loan-to-value ratios and why they are important for this study and what determines them.

3.5 Loan-to-value’s, mortgage default determinants and loan-to-value determinants The main objective of this thesis is to examine whether the policy changes as imposed by the GHF decreased the risk profiles of newly issued mortgages. An important measure of the risk

(15)

15

profile of a mortgage is the Loan-to-Value (LTV) as higher LTVs lead to higher default rates (Qi et al., 2009). Qi et al. (2009) also find that the most important variables which determine the loss severity on a loan default are the current LTV, the LTV at origination, the loan size, the age of the loan and the state foreclosure laws. Wong et al. (2004) investigated the Hong Kong residential market in order to determine the most important factors for default rates there. They looked at the impact of negative equity on the probability of default rates and found that default rates in Hong Kong residential are significantly positively correlated with the current LTV ratio, the level of interest rates and the unemployment rates while the default rate is negatively correlated with market sentiment. Wong et al. (2004) therefore drew the same conclusion as Qi et al. (2009) that the current LTV is an important factor in the mortgage market as a determinant of default rates. Wong et al. (2004) even suggest an maximum LTV of 70% on origination in residential mortgage lending given the importance of the current LTV for the probability of mortgage defaults. Campbell and Dietrich (1983) had empirical evidence through a multinomial logit model that default rates are significantly related to both the current as the original LTV ratios.

Francke and Schilder (2014) examined the Dutch housing market and the key drivers of the probability of loss on Dutch mortgages. The authors use a discrete time hazard model with calender time- and duration-varying covariates to analyze the relationship between certain variables and events to identify the key drivers of the probability of loss. The main findings of Francke and Schilder (2014) are that the probability of loss varies with duration of the mortgage with a peak at a duration of approximately four years. Their findings with regards to the key drivers are that home equity, unemployment and divorce have the most impact on the probability of loss whereas affordability measures are of lesser importance. The conclusion in existing literature about determinants of default probabilities and severities is therefore in a consensus that LTV ratios are important. Therefore in this study the LTV will be investigated in detail which will be elaborated on in Sections 4 and 5.

Sandor et al. (1975) created a model to retrieve the determinants of the risk premium for mortgages. Important determinants are borrower-, property- and neighborhood characteristics which indicates that mortgage issuers are very precise on the information they use when issuing mortgages. Cunha et al. (2013) investigate the determinants of LTV ratios in the Netherlands. They perform a panel regression of Dutch households in the period 1992-2005 to determine which factors influence the outstanding LTVs significantly. Their findings are that LTV declines with time elapsed since mortgage issuance date. Next to that both income, net worth and marginal tax rate influence LTV positively. They also conclude that the current LTV ratios of non-amortization mortgages are on average 10% higher than on mortgages on which there is amortization. The results of the study by Cunha et al. (2013) are used in this thesis to determine control variables for the regressions which is elaborated on in Section 4.

(16)

16

Research by Epley et al. (1996) finds evidence that certain underwriting guidelines are positively correlated to default risk. Some of these guidelines are mortgage payment-to-income ratio, maturity, debt and high-risk employment. The mortgagor signals his risk profile through the amount of down payment capital he will provide while the mortgagee examines the aforementioned guidelines and decides whether he issues the mortgage. The main conclusions of the article are that the borrower has some signaling power through the LTV ratio and that the LTV, or the equity proportion of the housing capital, is of high importance to the risk level of a mortgage.

3.6 Mortgage modification policy

After the importance of mortgage lending standards became clear, several types of legislation with regard to mortgages were implemented. Mayer et al. (2011) investigated policy changes regarding mortgages. The modification they investigated mainly consisted of benefits for mortgagors that defaulted on their loans. The authors used a difference-in-differences framework to estimate the increase in defaults between households in the period after the change in the policy in comparison to households not affected by the new policy. Mayer et al. (2011) found empirical evidence of an increase in strategic defaults after modifying mortgage policy. Their conclusion for policy makers is therefore that when constructing new mortgage policy, the policy makers should keep strategic behavior by mortgagors into consideration. Haughwout et al. (2009) have investigated the likelihood of a modified mortgage in the United States to re-default the next year. They find that subprime mortgages with a modification in the form of lower monthly payments, achieved through lower interest rates or principal forgiveness, actually result in a significantly lower probability of default the next year. Haughwout et al. (2009) note that the modification in the form of lower payments creates a lock-in spillover effect on the household which results lock-in a lower mobility. They also mention a limitation of their research that it only uses data on subprime mortgages and therefore does not measure the effect of mortgage modifications on non-subprime mortgages.

Levitin (2009) tested for the United States whether the assumption that mortgage modification, which protects lenders from default losses, incentivizes the lenders to lend more at lower rates. The conclusion of the research is that the assumption does not hold and that such modifications would have little or no impact on the credit costs of mortgages and the availability. Therefore the modifications in the United States had no undesired spill-over effects in the form of more subprime mortgage lending as a result of the policy.

(17)

17 3.7 Literature on mortgage issuance policy

Regarding the effects of new legislation on mortgage issuance to reduce subprime lending there has been done little research so far. An example that has been studied in existing literature is the predatory lending law which has been accepted and enacted in 1999 in North Carolina in the United States. The key provisions of the lending law were limitations on loan terms such as negative amortization, prepayment penalties, call provisions and debt-to-income ratios (Harvey and Nigro, 2004). Harvey and Nigro (2004) investigated the effect of the new predatory lending law on the probability that one would be denied a new mortgage. Next to that they looked at the change in lending to minorities and lower income households after the new legislation. Their results indicate a clear deduction in denial probability after the change in legislation. Harvey and Nigro (2004) also find a deduction in subprime originations but they attribute it to the large decline in loan applications rather than to the differences in denial probabilities. For minorities and lower income households the growth rates of applications were significantly small which Harvey and Nigro (2004) attribute to the fact that these households rely more on the subprime market. Elliehausen and Staten (2004) found that the North Carolina predatory lending law significantly reduced the number of closed-end mortgages. The number of subprime mortgages declined by a significant 14 percent accompanied by a decline in the number of mortgages to lower income borrowers of 27 percent of the quarterly county originations before the enactment of the law. The observations were only significant in North Carolina and not in surrounding states which strengthens the robustness of the results.

Ho and Cross (2006) found a different result when studying the effect of the predatory lending law of 1999 in North Carolina. They find a reduction in the rejection rate of about ten percent which has not been the intent of the law when it was enacted. Next to that, Ho and Cross (2006) conclude that strengths of different predatory lending laws depend on the coverage and the number of restrictions the law has. Laws with more extensive restrictions come with a larger decrease in lending activities while a law with broad coverage actually increases the number of applications. They conclude that the design of a lending law is very important for the effect that the law has.

A different example of lending legislation are the maximum LTV and Debt-to-Income (DTI) limits in Korea which have been examined by Igan et al. (2011). The authors investigate whether LTV and DTI maxima have an effect on household leverage, house prices and residential real estate market activity in Korea. Their findings are that the number of transactions reduced significantly while price appreciation reduced over a six-month window after the maxima were imposed. Overall Igan et al. (2011) conclude that in Korea the maxima on LTV and DTI ratios have been a successful way to slow down real estate booms and the risks that such a market boom accompanies.

(18)

18 3.8 Proposed hypotheses

Taking existing literature into account it is clear that mortgage legislation is important to be sure a housing bubble and a financial crisis as the crisis of 2008 will not occur again. The existing literature is scarce and without consensus on the effectiveness of mortgage issuing policy on risk profiles of newly issued mortgages. This thesis positions itself as an extension of current research on the topic of mortgage issuance legislation which could be used by policy makers when proposing new policy or laws on the subject. In this research therefore the following hypotheses will be tested:

Hypothesis 1: The changes in the GHF in 2007 and 2011 have led to lower Loan-To-Values on newly issued mortgages. The imposed changes in the GHF of 2007 did not include rules with regards to

loan-to-values but were implied in order to reduce the default rates of mortgagors and is therefore expected to have had a negative influence on LTVs on newly issued mortgages already. The changes in the GHF in 2011 imposed a maximum loan-to-value which has as objective to reduce the LTVs on newly issued mortgages. Therefore there is expected that the changes in the GHF in 2011 have led to a lower LTV on newly issued mortgages. The negative effect of the 2011 changes in GHF on LTV is expected to be larger than the 2007 changes due to the fact that the code became binding from 2011 and that only from 2011 there were actually rules imposed which involved the LTV ratio.

Hypothesis 2: The changes in the GHF in 2007 and 2011 have led to lower Loan Payment-to-Incomes on newly issued mortgages. The changes in the GHF in 2007 are expected to have had a

negative effect on the Loan Payments-to-Income (LPTI) ratio due to the fact that the changes in the code involved rules with regards to the maximum percentage of loan payments that is allowed to flow to mortgage payments through both the percentage of the income that can flow to mortgage payments and the toetsrente as described in Section 2.3.1. The changes in the code in 2011 are expected to have had a negative effect on the LPTI ratios through the aforementioned rules as well. The expectation is that the changes in 2011 have a larger negative effect due to the fact that the code became binding in 2011.

Hypothesis 3: The changes in the GHF in 2007 and 2011 have led to a lower probability that a newly issued mortgage is of a type without amortization during the loan term. The changes in the

GHF in 2007 are expected to have had a negative effect on the probability that a newly issued mortgage is of a non-amortization type due to the fact that the changes were implied in order to decrease the risk profiles of outstanding mortgages. The changes in the GHF in 2011 are expected to have had a negative effect on the probability that a newly issued mortgage is of a

(19)

19

type with a high risk profile as well due to fact that the changes involved restrictions on the amount of LTV that could be non-amortizable. Also restrictions with regards to tax deductibility of interest payments of non-amortization loans were imposed. For hypothesis 3 holds as well that the negative effect of the changes in 2011 on the probability is expected to be larger than the effect of the 2007 changes through the non-bindingness of the code in 2007.

The next section will elaborate more on the data that will be used to test for the proposed hypotheses.

4. Data and Descriptive Statistics

This section starts with a discussion on the data used for the research and where it has been retrieved from. The following paragraphs in this section discuss data preparation and transformation followed by descriptive statistics of the key variables.

4.1 Data source and sample selection

For this thesis a survey conducted by the University of Tilburg in the Netherlands will be used. The survey is named the DNB Household Survey and consists of annual financial and general data of approximately 2,700 Dutch households over the period 1993-2015 of which the years 1995-2015 are usable for this thesis (EUI, 2015). All people with the age of 16 years old or higher in each household have been interviewed. The recruitment for the survey has occurred on a random national sample taken from the private postal address issue file. If a household wants to exit the survey, it will be replaced by another household with the same characteristics as the household that dropped out. However, the attrition bias that occurs through households leaving the sample should be taken into account and is further discussed in Section 4.5. The survey data are collected each year through the Internetpanel of CentERdata (DNB Household Survey, 2015). The CentERdatabase is the source where the data has been retrieved from. The DNB Household Survey consists of six categories with information about the participating households and its members. The categories are labeled as follows in the database:

1. General Information on the Household 2. Household and Work

3. Accommodation and Mortgages 4. Health and Income

5. Assets and Liabilities

6. Economic and Psychological Concepts

The data give an opportunity to measure the effects of legislation on mortgage issuance. The current literature on the topic is scarce due to the fact that legislation on mortgage issuance has

(20)

20

been imposed only recently and because data on the subject of people’s personal financial documentation is limited due to confidentiality. Because of the availability of household’s borrowing and income information this research can examine the effects of the changes in the GHF on several variables indicating the risk on outstanding mortgages. The results of the thesis can be used for further policy decision making by either governments or central banks. The categories of interest in the database for this thesis are General Information on the Household,

Household and Work, Accommodation and Mortgages and Health and Income. Variables of

interest in these categories are the annual income of the household, the mortgage, the type of mortgage, the loan amount, the year the mortgage has been taken out, whether the mortgage has Nationale Hypotheek Garantie (NHG), the gross income of the household and the amount of savings. Further elaboration on the variables used and variable transformation is disclosed in Sections 4.2, 4.3 and 4.4. The survey data have been collected in separate files for each year the survey has been taken. All data has been appended per category and merged on the category which year the survey had been taken and the household number.

The used variables3_{have been truncated in order to diminish the influence of extreme} values on the research. Because the data consists of survey answers several answers were extremely unrealistically high values. By truncating the data these values have actually been omitted from the research to reduce the bias these values would result in. Appendix A elaborates on the boundaries and criteria used to truncate as well as on the number of observations lost by truncating.

Because there are multiple members for each household being surveyed each year the household number has been extended in the analysis with a decimal involving the number of the household member. By doing so, double values in the panel data with regards to household number and year have been made unique which makes the data suited for panel regressions. Section 5 elaborates on the methodology used in this research.

4.2 Dependent variables

To create the variables of interest several transformations have been done. The variable Original Loan-to-Value (OLTV) has been created as the original loan amount divided by the current market value of the house at the time the survey was done. The variable Current Loan-to-Value (CLTV) has been created as the original loan amount divided by the market value of the house at time t. Both variables have interesting meaning due to the fact that the OLTV is actually a representation of the LTV at which a mortgage has been issued. The CLTV is a representation of the loan value as a fraction of the current market value of the house. The CLTV gives insight in whether mortgages are under water or nearing 100% to see whether households have a decent

(21)

21

gap between their mortgage debt and house value. Next to that, existing literature as discussed in Section 3.5, showed the importance of both the CLTV and OLTV in mortgage default rates and therefore both ratios are included in the analysis. However, the CLTV is used in a regression disclosed in the appendix because it is not part of the main research which will be elaborated on in Section 5. The third dependent variable of interest is the Loan Payments-to-Income (LPTI) ratio. To create the LPTI ratio, the loan payments have been annualized for every observation. Subsequently, the annualized loan payments are divided by gross income at time t in order to get the LPTI ratio for each household member at time t. Gross income is measured as the gross salary for the household member at time t-1.

The other dependent variables of interest are dummies for each different mortgage type. The dummies are used in order to determine whether the probability that a non-amortization mortgage has been issued has decreased after the changes in the GHF. This will be explained further in Section 5. The mortgage types have been categorized into six categories of which three different levels of risk have been assigned. To determine the level of risk, interest rates charged for each mortgage are being compared among the four banks with the largest market share in the mortgage market in the Netherlands. First the mortgage types have been categorized in the following categories:

1. Annuity Mortgage 2. Linear Mortgage 3. Investment Mortgage 4. Bank Savings Mortgage 5. Interest Only

6. Life Insurance

Mortgage types noted as ‘’other’’ have been excluded from the analysis because there is no way to measure the risk profile of the mortgages of which the type is unknown.

After assigning the mortgage types to the six defined categories the interest rates for each category have been retrieved from the four banks with the largest mortgage market share in the Netherlands. These banks are ING, ABN Amro, Rabobank and SNS Bank (Banken.nl, 2015) (ACM, 2013). After retrieving the interest rates for the different mortgage types for each bank a clear division in two different risk profiles for the mortgage types became visible. All banks charge the highest rates for investment mortgages, interest only mortgage, bank savings mortgage and life insurance mortgage. A lower rate is charged for both annuity mortgages and linear mortgages. See Table 1 for an overview of the different interest rates the banks charge on the different mortgage types. The interest rates are the interest rates a mortgagor needs to pay on the outstanding loan amount.The higher risk a lender perceives he incurs on a loan, the more

(22)

22

he would like to be compensated for this risk through higher interest rates. Therefore the interest rates can be perceived as the difference in risk level for each mortgage type for the lender. The more risk the mortgagee has, the higher interest he charges on the outstanding loan amount.

Table 1 – Overview of mortgage interest rates on a 20-year loan term with an LTV of ≤

65% without NHG (April 12

th

₂₀₁₆₎

(1) (2) (3) (4) (5) (6)

Mortgage issuer Annuity Linear Investment Bank

savings Interest only insurance Life

ABN Amro 2.80% 2.80% 3.00% 3.00% 3.00% 3.00%

ING 3.20% 3.20% 3.40% 3.40% 3.40% 3.40%

Rabobank 2.70% 2.70% 3.00% 3.00% 3.00% 3.00%

SNS Bank 2.65% 2.65% 2.85% 2.85% 2.85% 2.85%

After investigating the interest rates per mortgage type, the division in risk profiles is proposed as follows in this thesis:

1. Annuity & Linear Mortgages (risk level 1)

2. Bank Savings, Life Insurance, Interest Only and Investment Mortgages (risk level 2)

The differences in risk can be explained due to the fact that the mortgages with risk level 2 all do not amortize during the loan term. The interest only mortgage is a mortgage in which the mortgagor only pays interest and pays off the loan amount at the end of the loan term (Geltner et al., 2014). The investment mortgage is a mortgage that is linked to an investment portfolio. The investment portfolio serves as the collateral and is used to pay off the loan at the end of the loan term. Because there is no amortization during the loan term and the mortgagee depends on the returns on an investment portfolio, which can consist out of stocks and bonds, the investment mortgage is perceived to have a high risk profile. Life insurance and bank savings mortgage types do not amortize during the loan term as well. For the life insurance mortgage, the mortgagor does not amortize and instead he has to get life insurance. At the end of the loan term, the mortgagor pays off the loan with the capital that has been built up through the life insurance account. For the bank savings mortgage, the mortgagor does not amortize but has to put a payment in a savings account periodically. At the end of the term the money in the savings account is used to pay off the loan amount.

For all the mortgages in risk category 2 holds that if the mortgagor defaults on the loan the full loan amount is defaulted on because there is no amortization during the loan term. When a mortgagor amortizes during the loan term, the loan amount decreases and therefore the loss will be smaller for the mortgagee. As is clear from Table 1, the four largest mortgage banks in the

(23)

23

Netherlands all include a risk premium between 0.20% and 0.30% on non-amortization mortgages on a fixed rate ≤65% LTV with a 20-year term. The higher interest rates reflect the higher perceived risk. Therefore after the analysis with respect to different risk levels on mortgages there has been made a dummy NonAmordum which equals 1 if the mortgage is of risk level 2 and therefore has no amortization during the loan term. The dummy variable

NonAmordum will be the dependent variable to test for hypothesis 3 as discussed in Section 3.8.

4.3 Independent variables of interest

Because a difference-in-differences model will be used in this research, several dummy variables will have to be created in order to test for the proposed hypotheses. The independent variables of interest for this study are the dummies that equal 1 if the mortgage has been issued under new legislation and 0 otherwise. Therefore two dummies have been created: a dummy that equals 1 if the mortgage has been issued in the time frame 2007-2010 and a dummy that equals 1 if the mortgage has been issued in the time frame 2011-present. Due to the fact that the change in the GHF in 2011 occurred in August, the dummy equals 1 for mortgages issued in the time frame 2012-present due to data limitations with regards to the exact date of issuance per mortgage. For every other year in which mortgages have been issued in the sample a separate dummy has been created to control for time fixed effects in the regressions. Next to these dummies, also dummies which indicate whether a household member is part of the treatment group are included. Sections 5.1 and 5.2 elaborate on the identification strategy used for the treatment group. Eventually the variables on which the coefficients will be tested are the interactions of the treatment dummy and the dummies for the different legislation periods. Section 5 will discuss the proposed methodology and testable hypotheses.

4.4 Control variables

The control variables in the regression are determinants of LTV ratios as found by Cunha et al. (2013). Cunha et al. (2013) investigated the Dutch mortgage market to determine which variables actually determine the LTV ratio households have in the Netherlands. The research concluded that important determinants were the income, which is measured as gross income in this analysis, the age of the mortgagor and time elapsed since mortgage commencement. The gross income is measured as the gross salary the household member earned at time t-1. Next to these variables also the net worth is an important determinant of the LTV according to Cunha et al. (2013). However, due to data limitations, a proxy for net worth in the form of savings will be used in this analysis because savings should also improve the solvency of the mortgagor and therefore should affect the LTV positively. There has been decided to also use dummies to control for NHG (National Mortgage Guarantee) and whether the mortgagor is living together or

(24)

24

by him- or herself. NHG serves as a sort of insurance that limits the risk for both the mortgagor and mortgagee. If a mortgage has NHG, which can be achieved through paying a fee, the Stichting Waarborgfonds Eigen Woningen becomes liable for the debt (NHG, 2015). If the mortgagor defaults on the loan due to certain circumstances such as unemployment, divorce or disease, the NHG will help to try to figure the loan out. If the situation cannot be saved, the NHG will pay the outstanding loan debt under certain conditions. The NHG should therefore be a determinant of LTV as well due to the fact that the mortgagee has more protection in case of default. Interest rates charged on mortgages with NHG are significantly lower than on mortgages that do not have NHG (Mortgage rates, 2016). Whether the household member is living together or apart should influence the LTV as well due to the fact that often two people have more collateral and/or income than a one-person household. Next to the determined control variables, also year dummies will be included in the regressions to control for year fixed effects. Household fixed effects, which are entity fixed effects in the proposed panel regressions, are included as well. To decrease the bias in the proposed regressions regional dummies are used to control for the location effects in the data. A dummy per province and a dummy that indicates the amount of urbanization of the location have been created. For urbanization five different degrees were available in the data ranging from No urbanization to Very high degree of urbanization. For every degree a dummy has been created. At last house characteristics are expected to determine the dependent variables as well. The houses in the data sample have been categorized in five different house types for which a dummy has been created per type. The other two variables that have been used as house characteristics are a dummy that takes into account whether the house has a garden or not and the logarithm of the house size. The logarithm has been used to account for the law of diminishing returns, which means that the marginal effects of an increase in house size on the dependent variables becomes smaller the larger the house size is.

4.5 Attrition and possible informational bias

Important to note is the fact that there is possible attrition bias in the data set used in this thesis. Due to the fact that the data are survey data over a 20-year period involving a fixed set of households the probability that certain households or household members would drop from the sample is large. Over the sample period many households have entered and dropped from the sample. Every year approximately 2,700 households took part in the survey. Over the entire sample period of 1995-2015 a total of 8,063 unique households have taken part in the sample. In the first year of the survey a total of 2,773 unique households participated. So over the remaining sample period many households have left and entered the sample which could result in attrition bias (Stock and Watson, 2012).

(25)

25

Due to the data being survey data the values are also subject to informational bias. Informational bias is defined in this research as bias through possible incorrect values of observations. The questions in the DNB Household survey consist of specific questions with regards to quantitative variables such as the market value of the house, payments to the mortgages and the amount of savings. Due to the fact that there has been no control on the correctness of the survey values of these variables, there can be some bias in the data through possible incorrectness of these variable values which should be taken into account when drawing conclusions.

4.6 Descriptive statistics

After the data preparation, a descriptive statistics table of the key variables in this research has been constructed. Next to the descriptive statistics, graphs have been formed to visualize the key dependent variables over time to get a view of how the ratios changed over time. Both the descriptive statistics and the graphs are disclosed and discussed in this section. Table 2 Provides descriptive statistics on the key variables used in the regressions (1995-2015).

Table 2 - Descriptive statistics of key variables

(1) (2) (3) (4) (5) (6)

Variable mean median sd min max N

CLTV 0.48 0.44 0.27 0 1.091 13,824

OLTV 0.94 0.94 0.46 0 1.580 13,920

LPTI 0.25 0.17 0.29 0 1.589 7,855

Market value house 315,369 270,000 182,437 23,000 3,350,000 18,388

Original loan amount 142,046 120,000 106,614 0 3,400,000 13,921

Loan payments 9,284 7,445 8,261 0 227,880 13,786

Gross Income 50,745 45,000 31,597 2,500 140,000 13,433

Savings 17,185 5,060 27,577 -18 122,645 20,820

Mortgage age 9.89 8.00 7.921 0 43 14,085

The average loan-to-value measured as original loan amount over the market value of the house at time t in the sample is 0.48 with a standard deviation of 0.27. It appears to be relatively low for homeowners but it can be explained by the fact that it is constructed as original loan amount divided by the market value of the house in the year the survey has been conducted. This means that for old mortgages with an original loan-to-value of for example 0.80 in a market with rising real estate prices, the loan-to-value in later years would be significantly lower than when the mortgage was issued. When the loan-to-value is measured as the original loan amount divided by the purchase price of the house the average is 0.94 with a standard deviation of 0.46. This is substantially higher than the CLTV ratio, which can be explained through rising house prices. The average loan payments-to-income ratio is 0.25 with a standard deviation of 0.29. Approximately 25% of a household members gross income flows to mortgage payments on an

The effects of changes in the mortgage financing code of conduct on risk profiles of newly issued mortgages : a study of the Dutch mortgage market

MSc Thesis