Master Thesis
MSc Business Economics: Finance & Real Estate Finance
Exploring the Link Between Pensions and Housing
Modeling Young Households’ Tenure Choice
Name: Bram Broekmeulen
Student-number: 6182054
Thesis Supervisor: Prof. Dr. M.K. Francke Date August 15th, 2015
Statement of Originality
This document is written by Student Bram Broekmeulen 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.
Abstract
In this thesis, the policy implications of a more flexible pension system, in which young households would be allowed to utilize their pension assets for the purchase of a home, are empirically investigated using 21 waves of Dutch Household Surveys spanning over two decades. This thesis builds upon the empirical literature of households’ tenure choice behavior, but is unique in its approach of modeling the effect of income on households’ homeownership propensities, as it focuses on the income component that determines the maximum mortgage one might obtain. The outcomes of the conducted survival analysis show that, controlling for other factors, a percentage-point increase in the amount households may maximally spend on housing services leads to a 73% increase in the hazard of the household realizing homeownership for the first time. Using the model’s estimation results for policy simulations further show that for the average household in the sample, the expected time until first-time homeownership decreases by 9,7% to 29,5%, depending on the time-period during which young households may utilize their pension savings to pursue homeownership.
Keywords Tenure choice, homeownership, wealth-accumulation, survival analysis, policy
Table of Contents
1 Introduction 1
2 The Dutch Housing Market 4
2.1 Demand-side Drivers of Mortgage Debt 4
2.2 Amplifying Effect of Household Debt 5
2.3 Policy Changes 6
2.4 Dutch Asset Base 8
3 Literature Review of Tenure Choice Studies 9
3.1 Consequences of Homeownership 9
3.2 Homeownership as an Investment 9
3.3 Determinants of Homeownership 10
3.4 Position of this Thesis & Hypotheses 13
4 Methodology 15 4.1 Survival Analysis 15 4.2 Model Choice 16 4.3 Variables of Interest 17 4.4 Control Variables 18 4.5 Model Specification 19
4.6 Limitations of the Model 20
5 Data and Descriptive Statistics 21
5.1 Data Sources 21
5.2 Dataset Construction 22
5.3 Descriptive Statistics 24
6 Results 28
6.1 Model Outcomes 28
6.2 Outcomes Proportional Hazards Assumption Test 32 6.3 Time to Homeownership and Policy Implications 33
7 Robustness Checks 35
8 Conclusion 38
9 Reference List 40
List of Figures and Tables
List of Figures
Figure 1: Year-on-year House Price Appreciation in the Netherlands 5
Figure 2: Underwater Mortgages per Age Cohort 6
Figure 3: International Comparison of Pension Bases 8
List of Tables
Table I: Overview of the Reviewed Literature’s Main Components 14
Table II: Survival Analysis’ Main Components 16
Table III: Overview of the Used Data Sources 21
Table IV: Covariates’ Employed Specifications 24
Table V: Covariates’ Expected Coefficients 24
Table VI: Condensed Life Table 25
Table VII: Descriptive Statistics of the Studied Sample 26 Table VIII: Pension Contributions as a Fraction of Income 27
Table IX: Results Survival Analysis 29
Table X: Test of the Proportional Hazards Assumption 32
Table XI: 30-year Annualized Increases in the Amount Spendable on Housing 33
Table XII: Results Policy Simulations 34
Table XIII: Outcomes Robustness Checks 36
Table XIV: Full Life Table 43
1
Introduction
The Dutch housing market is marked by a peculiar structure. In line with other Western European countries, its owner-occupier sector comprises the largest portion of its housing stock. Its rental sector on the other hand is quite distinct from its fellow EU-members however, as this segment of the Dutch housing market is dominated by social housing corporations, making for the largest social housing sector in the world (Pittini & Laino, 2011).1
The composition of its housing stock, dominated by owner-occupancy, is the result of a fiscal climate in which homeownership was promoted. Fueled by this fiscal climate, the Netherlands has become the most mortgage-indebted country in the world (Van Leeuwen & Bokeloh, 2012).2 This indebtedness had an amplifying effect during the build-up and
aftermath of the last financial crisis, and exposed the vulnerability of the Dutch economy (De Nederlandsche Bank, 2015; IMF, 2012).
Now, in an effort to improve the Dutch economy’s financial stability, policymakers are targeting the provisions that once induced households to take on excessive amounts of debt. One of these provisions relates to the high loan-to-values (LTVs) on mortgages provided to (aspiring) homeowners, and is in process of being lowered to a maximum of 100% by 2018. Some experts claim that this gradual reduction will not be sufficient for the purpose of limiting banks’ risk profile however, and therefore have proposed lowering the maximum LTV to 80% beyond 2018 (Commissie Structuur Nederlandse Banken, 2013). This proposal is expected to have far-reaching social consequences and therefore has been scrutinized, as it will capital-constrain aspiring homeowners and thereby prohibit them to realizing their housing preferences. Due to the structure of the Dutch housing, in which few alternatives to ownership exist (CPB, 2015), this proposal is therefore expected to have negative welfare effects (Schilder, Conijn, & Rouwendal, 2015).
In light of this measure’s negative implications, a debate has emerged in which a link between the Dutch pension base, the largest in the world in terms of Gross Domestic Product (GDP), and the Dutch housing market is proposed. Heading this debate, Bovenberg and Kortleve (2012) advocate for households having more flexibility in building up and drawing down pension assets. Recognizing the need for lower levels of mortgage debt on the one hand
1As a percentage of total housing stock 2 In terms of GDP
and the negative implications of capital-constrained households on the other, the authors propose the following three links between the Dutch pension base and housing market:
1. Allowing young workers to use part of their pension assets to fund a home purchase 2. Allowing homeowners to reduce their mortgage indebtedness by (temporarily)
lowering their pension contributions
3. Allowing homeowners to draw down home equity to pay out additional pension benefits
In the first proposal, the authors plead for young households having more autonomy with regard to their saving behavior, effectively letting them decide if they want to use their pension or their home as a vehicle for wealth accumulation. In this thesis, the effects of this proposal on the housing market are investigated by use of a tenure choice study. In this tenure choice study, the mechanisms underlying young households’ tenure choice behavior are examined by studying a sample of young renting households using the DNB Household Surveys’ 1994 to 2014 waves. To facilitate policy simulations, it is assumed that these households will employ their pension contributions to finance the purchase of a home. As pension contributions can be expressed as a percentage of income, modeling households’ transition into ownership using income as an explanatory variable might provide insights with regard to the proposal’s implications for the housing market.
In contrast to previous tenure choice studies, income is not included directly or independently, but jointly with financing costs, following the borrowing capacity formula used by Francke, van de Minne and Verbruggen (2014). Borrowing capacity serves as a proxy for the maximum mortgage one might obtain and thereby proxies a household’s access to the owner-occupier market. Given the characteristics of the Dutch housing market, this approach of modeling the impact of income on a household’s (first-time) homeownership propensities is expected to improve on the existing literature. By ways of a survival analysis, the following research question is addressed:
To what extent might allowing potential first-time buyers to utilize their pension contributions to fund a home purchase affect their homeownership propensities?
By addressing this research question, new insights on the policy implications of a more flexible pension system are derived. Moreover, the mechanisms underlying young households’ tenure choice behavior are reviewed in a dynamic setup in a country with a remarkable housing market: the Netherlands.
The remainder of this thesis is structured as follows. Section 2 gives a short description of the Dutch housing market. Section 3 provides a literature review on tenure choice studies. Section 4 describes the used methodology. Section 5 discusses the used data and provides some descriptive statistics. Section 6 provides the model’s estimation results and describes the outcomes of various policy simulations. In section 7, the results of several robustness checks are provided and finally, section 8 concludes.
2
The Dutch Housing Market
The Dutch housing market has been formed by a tradition of government interventions in both the owner-occupier- and rental sector (De Groot, Manting, & Mulder, 2013). In line with other European countries, homeownership has increased ever since the Second World War (Feijten & Mulder, 2002) and currently represents 60% of its housing stock, in line with countries such as Sweden, France and Denmark (Pittini & Laino, 2011). The amount of debt underlying this owner-occupier market is not in line with its fellow EU-members however, as the level of mortgage debt in the Netherlands is the highest in the world, equaling 108% of its GDP in 2012 (Van Leeuwen & Bokeloh, 2012).3
In this section, this level of mortgage debt is elaborated upon. First, mortgage debt’s main demand-side drivers are discussed. Second, its amplifying potential is discussed. Third, policymakers’ efforts to reduce mortgage indebtedness and the (potential) implications of such efforts are discussed. Finally, the Dutch pension base, and its potential link to the housing market, is briefly discussed.
2.1
Demand-side Drivers of Mortgage Debt
In comparison to other OECD countries, the rental sector in the Netherlands is large and of good quality (Feijten & Mulder, 2002). The composition of its housing stock is unique, as social housing corporations are responsible for more than three quarters of the rental sector’s dwellings. As a whole, the rental sector is heavily regulated, as only 7,4% of all rental dwellings are free from rent controls (Conijn J. , 2011)
In the social rental segment, housing corporations provide housing to low-income households based on a ‘choice-based letting system’. In this system, households have to meet strict selection criteria related to income levels and rank on a waiting list.4 Given its rank on
the waiting list, a household can enter the social rental sector if its income at the time of entering the dwelling is below an annually revised threshold.5 Since eligibility is determined
based solely on income at the time of entry and rents in the social housing sector tend to be well below market levels (Conijn & Schilder, 2011), households having obtained a home in the social segment are inclined to remain in this segment of the rental sector, rather than move to either the owner-occupancy or private rental segment when increases in income would have
3 This measure of (mortgage) indebtedness does not take savings into account.
4 The queue for a dwelling ranges from four years in the more rural areas to ten years in the big cities (Schilder,
2012).
allowed them to do so (Schilder, 2012). The combination of these selection criteria and size of the social housing segment make for a rigid housing market, in which households enjoying an income above the social housing’s income-threshold have little alternatives to owner-occupancy (De Groot, Manting, & Mulder, 2013).
The Dutch owner-occupier market on the other hand is marked by the large amount of mortgage debt underlying it. This mortgage debt contributed to the Dutch housing market experiencing extended periods of house price appreciation (displayed in Figure 1), which subsequently induced more borrowing and reinforced housing prices’ upward trend, confirming the two-way causality between credit and housing prices (Goodhart & Hofman, 2007; Oikarinen, 2009).
Figure 1
Year-on-year House Price Appreciation in the Netherlands
Figure 1 displays the year-on-year house price appreciation in the Netherlands. Spanning nearly three decades,
this graph shows that owner-occupied housing yielded positive returns for most of this period (Source: CBS).
Moreover, the tax-deductibility of mortgage interest payments, an instrument introduced in the 1990s to promote homeownership further incentivized aspiring homeowners to take out mortgages with high LTVs, as this instrument effectively lowered the cost of debt. This incentive motivated borrowers to refrain from amortizing on their mortgage, so that the benefits of tax-deductibility were maximized over time (Van Leeuwen & Bokeloh, 2012).6
2.2
Amplifying Effect of Household Debt
The discussed demand-side drivers of mortgage debt have contributed to the Netherlands becoming highly (mortgage) indebted. In the empirical literature on the relationship between
6 In addition to these demand-side factors, a supply-side factor impacting the amount of mortgage debt
outstanding – the National Mortgage Guarantee (NHG) – is important as well. However, this thesis focuses solely on the demand-side drivers of mortgage debt.
-10% -5% 0% 5% 10% 15% 20% 1 9 8 6 1 9 8 7 1 9 8 8 1 9 8 9 1 9 9 0 1 9 9 1 1 9 9 2 1 9 9 3 1 9 9 4 1 9 9 5 1 9 9 6 1 9 9 7 1 9 9 8 1 9 9 9 2 0 0 0 2 0 0 1 2 0 0 2 2 0 0 3 2 0 0 4 2 0 0 5 2 0 0 6 2 0 0 7 2 0 0 8 2 0 0 9 2 0 1 0 2 0 1 1 2 0 1 2 2 0 1 3 2 0 1 4
household indebtedness and financial stability, it has been found debt has the potential to augment macroeconomic shocks (Ogawa & Wan, 2007). This potential materialized during the last financial crisis, with nations experiencing larger run-ups in household debt pre-crisis seeing higher appreciation rates of housing prices at first, but larger drops in housing prices and subsequently, consumption levels, after the bubble burst (De Nederlandsche Bank, 2015; IMF, 2012). Due to large drops in housing prices, a lot of Dutch households, especially the (younger) ones that purchased their home after 2004, became under water (Francke, Van de Minne, & Verbruggen, 2014). This increase in underwater mortgages can be seen in Figure
2.7
Figure 2
Underwater Mortgages per Age Cohort
Figure 2 shows the portion of homeowners with underwater mortgages relative to the total number of
homeowners per age cohort. Both cohorts have become more (mortgage) indebted since the financial crisis, with young (old) homeowners seeing a 47% (11%) increase in the number of underwater mortgages over the 2008-2013 period (Source: CBS, own calculations).
2.3
Policy Changes
In an effort to improve the financial stability of the Dutch economy, policymakers are now targeting the main demand-side drivers of its mortgage debt: the tax-deductibility of mortgage interest payments and its high LTV ratios.
As of 2013, the tax-deductibility of mortgage interest payments will only apply conditional on amortizing mortgage loans.8 Moreover, the maximum LTV ratio will be
(gradually) lowered to 100% by 2018. When it reaches 100%, households will not be able to finance the acquisition costs of a home, about five to six percent of the home’s value, by using their mortgage anymore, leading to some significant out-of-pocket expenses for aspiring
7 A mortgage is ‘under water’ when the amount of debt underlying a dwelling exceeds the dwelling’s (market)
value (LTV is greater than one).
8 This applies only to newly issued loans. Loans issued before 2013 keep their original structure.
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 2006 2007 2008 2009 2010 2011 2012 2013 2014 Homeowners Under 35 Homeowners Over 35
homeowners (Van Leeuwen & Bokeloh, 2012). As the first measure is already in effect, the focus in the remainder of this thesis is on the implications of lowering the maximum LTV ratios to below 100% beyond 2018.
By lowering the maximum LTV ratio, policymakers are effectively capital-constraining aspiring homeowners. Capital constraints have been found to significantly reduce homeownership propensities, especially for young households (Zorn, 1988; Linneman & Wachter, 1989; Haurin, Hendershott, & Wachter, 1997). Given the structure of the Dutch housing market, the impact of capital constraints on households’ homeownership propensities is expected to be more pronounced in the Netherlands than in other countries, as there exist few alternatives to owner-occupancy.
For this reason, several studies have been conducted on how lowering the maximum LTV ratios below 100% might affect the Dutch housing market (CPB, 2015; De Nederlandsche Bank, 2015; Schilder, Conijn, & Rouwendal, 2015). Of these studies, the paper by Schilder et al. (2015) is most relevant for this thesis, as the authors use micro-data to empirically test the policy implications of a maximum LTV below 100%. The authors investigate the effect of this measure on the housing market by studying the duration of the saving period households experience before being able to make the down payment on the home they intend to purchase. They find that (on average) first-time buyers will have to save four and half years before being able to make the required down payment. This finding is based on the assumption that aspiring homeowners do not consume more than the minimum living expenses and, although it has been found that households increase their saving rate when they intend to purchase a home (Haurin, Hendershott, & Wachter, 1997), Schilder et al. (2015) note that the assumption that households will not consume more than the minimum living expenses is unrealistic. Therefore, the authors also examine the duration of the saving period under the assumption that only half the amount saved under the first assumption is saved. Under this assumption, first-time buyers are found to have to save eight and a half years before being able to finance a down payment. Given the structure of the Dutch housing market, this lengthy saving period is likely to have negative welfare effects. Based on these results, the authors conclude that unless significant changes in the rental sector take place, reducing the maximum LTV below 100% is undesirable.
2.4
Dutch Asset Base
In addition to having the highest level of mortgage debt in the world, the Dutch asset base is large in international comparison as well. As is shown in Figure 3, the Netherlands have the highest level of pension savings in the world (OECD, 2014). These savings are mainly the result of the large supplementary occupational pension contributions, which are a function of (gross) income and are subtracted from one’s income before being paid to the relevant pension fund by the employer.9
Figure 3
International Comparison of Pension Bases
Figure 3 shows the world’s largest pension bases in terms of GDP. The magnitude of the Dutch pension base is
mainly driven by supplementary occupational pension contributions made by the employer out of the employee’s income (Source: OECD).
Given the problems of the Dutch housing market and the size of the Dutch pension base, several links between the two have been proposed by Bovenberg and Kortleve (2012). With these links, the authors plead for more flexibility in households’ saving behavior. This proposed flexibility works two ways, as households might use their pension assets to either finance the purchase of a home or reduce their mortgage indebtedness on the one hand, but could also draw down home equity to pay out additional pension benefits on the other.10 In
light of the negative implications of capital constraints, the structure of the Dutch housing market and the size of the Dutch pension base, the remainder of this thesis is focused on how allowing young (renting) households to use parts of their pension assets to fund the purchase of a home might affect the housing market.
9 In this study, the focus is solely on pension contributions made out of an individual’s income.
10 The possibility of drawing down home equity to pay out additional pension benefits has been investigated by
Conijn et al. (2014). 0% 20% 40% 60% 80% 100% 120% 140% 160% 180% Finland Ireland Chile Canada United States United Kingdom Australia Switzerland Iceland Netherlands
3
Literature Review of Tenure Choice Studies
In this section, the relevant literature on homeownership is discussed. First the consequences of homeownership are discussed. Then, the potential of homeownership as a vehicle by which one might accumulate wealth is described. Subsequently, the determinants of first-time homeownership, based on a review of tenure choice studies, are discussed. Finally, the position of this thesis relative to the existing literature is described, and relevant hypotheses are developed.
3.1
Consequences of Homeownership
Homeownership is found to have consequences on both the household- and national level. In a study by Dietz and Haurin (2003), perspectives from disciplines such as economics, demography, psychology, political science and public policy studies are examined to evaluate the effectiveness of existing public policies related to tenure choice behavior. Based on an extensive review of the empirical literature related to the consequences of homeownership, the authors conclude that homeownership impacts households’ saving behavior, mobility, labor force participation, maintenance propensities, social engagement, health and child-outcomes (Dietz & Haurin, 2003). Moreover, emotional reasons for homeownership exist, as homeownership is associated with a sense of security and continuity, having full control over one’s housing situation, and status (Megbolugbe & Linneman, 1993; Helderman A. C., 2007). On a national level, homeownership has been found to impact urban form, segregation and crime levels (Dietz & Haurin, 2003). Given the range of themes impacted by homeownership, understanding the process by which households become homeowners is important. To this purpose, tenure choice studies that allow for the identification of the mechanisms underlying this process are frequently performed.
3.2
Homeownership as an Investment
For most households, the purchase of a home constitutes the single largest investment they will ever make. Moreover, it causes a non-financial cost, as the risk inherent in buying a home cannot be easily undone (Mulder & Wagner, 1998). Although first-time homeownership typically is not a conscious investment in the housing market, its characteristics mark its potential to serve as wealth-creating vehicle (Clark, Deurloo, & Dieleman, 1994).
Investing in a home provides a household with the opportunity to make a leveraged investment. Noting the amplifying effect of debt, this leverage has the potential to augment
housing wealth during periods of house price appreciation. Moreover, when a household makes mortgage payments, it is essentially saving through its own home. This way of saving is attractive in the Netherlands, as home-equity is tax-exempt, and therefore, costs of ownership are lower (Schilder, 2012).
Noting that there are few empirical studies on the impact of homeownership on household wealth accumulation, Di, Belsky and Liu (2007) find evidence that homeownership, measured by its duration, has a strong positive impact on households’ (non-housing) wealth accumulation. The authors make use of data from the Panel Study of Income Dynamics (PSID) from 1989 to 2001 and address wealth-induced endogeneity by controlling for households’ tendency to save. While appreciation rates were near their long-term average during the observation period and decreasing prices were observed as well, the number of realized losses was low, as owners adjusted their holding period when their home is was under water, in line with the empirical literature on loss aversion in the housing market (Engelhardt, 2003; Genesove & Mayer, 1997). Based on their empirical analysis, the authors conclude that the non-housing wealth of those who owned for eight years was $12,500 higher than that of those who were renters during the observation period after controlling for relevant covariates. This finding suggests that homeownership may help build up household wealth by means other than purely leveraged returns from house price appreciation, making homeownership a valid vehicle for wealth-accumulation (Di, Belsky, & Liu, 2007).
In a study focused on low-income households, similar results are found by Boehm and Schlottman (2008). In their study, the authors employ a dynamic probability model which allows for modeling the likelihoods of renters moving into and out of homeownership to predict potential housing wealth accumulation. Using data from the Panel Study of Income Dynamics (PSID) covering a nine-year observation period from 1984 to 1992, the authors model families’ housing wealth accumulation as a function of their tenure choices during the observation period, appreciation rates of the neighborhoods in which they reside(d) and level of housing expenditures. Although not a guarantee for accumulating wealth, the authors conclude by stating that homeownership serves as an important vehicle for accumulating wealth among lower-income households.
3.3
Determinants of Homeownership
As stated in section 3.1, the mechanisms underlying the process by which renters realize homeownership are frequently investigated by use of tenure choice studies. Up until the
second half of the nineties, these studies typically had a cross-sectional character and focused on life-course characteristics and contemporaneous tenure status. Clark, Deurloo and Dieleman (1994) were among the first to recognize that this approach of modeling tenure choice behavior did not accurately reflect the dynamic process underlying it. Therefore, in addition to life-course characteristics, the authors took the macro-economic context into account as well, and hypothesized that this would improve our understanding of households’ tenure choice behavior.
Using micro-level data from the Panel Study of Income Dynamics (PSID), Clark et al. (1994) model the time until a move into ownership takes place by ways of a survival analysis, which is capable of taking the time-sequence of renters moving into ownership into account. The authors do so using a Cox Proportional Hazard (Cox) model, noting the strength of the model’s flexibility. The results of the model indicate that household characteristics such as race, number of earners, number of rooms in the previous home and income important determinants of households transitioning into ownership.11 12 Regional variables, the amount
of new construction and inflation are found to stimulate couples’ homeownership propensities as well. The mortgage interest rate, hypothesized to have a negative impact on homeownership propensities since it increases borrowing costs, turns out positive and insignificant. The authors attribute this odd finding to the mortgage interest rate’s high correlation with the amount of new construction and inflation, effectively stating that this unlikely finding is related to the observation period. Similar results are found when the empirical analysis is done for families rather than couples. Based on the model’s outcomes, Clark et al. (1994) conclude that the availability of sufficient income, household stability, location and economic context affect renters’ path to homeownership.
Similar to Clark et al. (1994), Andrew, Haurin and Munasib (2006) model the event of renters becoming owner-occupiers by use of a Cox model. In their study however, the authors explicitly focus on young households becoming homeowners for the first time, as this event marks the moment as of which households will experience the consequences of homeownership as discussed in section 3.1.
Andrew et al. (2006) perform a survival analysis for both the United States (1979-2000) and Britain (1991-2002) and evaluate how the determinants of first-time homeownership behave across different markets. The hypothesized determinants of first-time
11 The race variable is defined as being ‘non-black’.
12 The number of rooms in the previous home indicates whether the previous home was ‘crowded’ and serves as
homeownership are mainly derived from the ‘user-cost of homeownership’ model, which captures the cost of ownership relative to tenancy based on variables such as income, (expected) house price appreciation and the planned length of stay (Haurin & Gill, 2002; Andrew, Hauren, & Munasib, 2006). As the authors find in their descriptive analysis that homeownership increased both in times of increasing income and in times in which income was stable but mortgage interest rates declined, they choose to include income and the mortgage interest rate simultaneously. Income is included directly, while the mortgage interest rate is interacted with the housing price index to proxy for financing costs. The results of their empirical model confirm the need for including income and financing costs simultaneously and show that policies targeted towards increasing income or easing credit market condition appear to yield the most elastic responses in terms of changes in expected homeownership (Andrew, Hauren, & Munasib, 2006). The model’s coefficients are not consistent throughout the international comparison, and the authors attribute observed disparities to differences in the housing markets and mortgage markets, noting that Britain has a relatively large social housing segment and stricter down payment requirements.
Boehm and Schlottman (2014) find similar results in their investigation of the timing and likelihood of tenure choice decisions in Germany and the United States. Unlike previous tenure choice studies, the authors examine the (overall) probability of households being homeowners during an observation period spanning 1997 – 2007. They do so by employing the same probability model used in their study on the effects of homeownership on wealth-accumulation (Boehm & Schlottmann, 2008), allowing for jointly modeling the likelihoods of renters achieving homeownership, transitioning back into tenancy and moving into owner-occupancy for a second time, making for a better approximation of actual homeownership rates (Boehm & Schlottman, 2014). For households’ initial transition into ownership, income, education level and household size all have a significant positive influence on the likelihood of renters transitioning into ownership. The mortgage interest rate enters the model indirectly through the costs of owning relative to renting, and is found to have a significantly negative impact on renters’ homeownership propensities. In line with Andrew et al. (2006), the authors find that the coefficient for having large savings in Germany is twice the size of the one found for the United States, consistent with the less strict down payment requirements in the United States.
Focusing on the United States exclusively, Herbert and Tsen (2007) study the implications of down payment assistance on minorities’ and low-income-households’ homeownership propensities. To this purpose, the authors use a proportional hazard model in
which the effects of financial, demographic and economic determinants are included. Unlike Clark et al. (1994) and Andrew et al. (2006) however, the authors do not leave the baseline hazard unspecified, as they employ a parametric proportional hazard specification in which the baseline hazard is assumed to have a Weibull distribution. Herbert and Tsen (2007) study a four-year period between 1995 and 2000 and find that the coefficients of most independent variables are consistent with the formulated hypotheses. Financing costs are accounted for by ways of changes in the mortgage interest rate relative to the start of the observation period, and is found to significantly impact one’s homeownership propensities as it lowers the hazard ratio by 30.5% for every percentage-point increase in the mortgage interest rate relative to the start of the observation period.13 A percentage-point increase in one’s income on the other
hand, increases one’s hazard rate by 34.75%.14 In sum, these findings support the notion for
including both financial determinants in modeling renter’s transition into homeownership.
3.4
Position of this Thesis & Hypotheses
The analysis employed in this thesis aims to extend and improve on the existing literature in several ways. Both income and financing costs have been accounted for, as they have been found to significantly impact the affordability of owner-occupied housing. The way in which these two financial determinants are assumed to influence renters’ homeownership propensities is assumed to be different from the ways proposed in the existing literature on homeownership however and is derived based on calculations of the National Institute for Family Finance Information (Nibud). Given the tendency of young households in the Netherlands to take out mortgages with LTV greater than one, the maximum mortgage a household can obtain is expected to be the most important financial determinant of young households, and can be proxied by households’ borrowing capacity. Decomposing the borrowing capacity equation (done in the methodology section) allows for the identification of an income and financing cost component. The income component, defined as the maximum amount one can spend on housing services annually, is likely to be a more accurate determinant of first-time homeownership than the previously employed income specifications. Building on the existing empirical literature and the institutional background in the Netherlands, the following hypotheses are tested:
13 Significant at the 10% level 14 Significant at the 5% level
Hypothesis I (H1): An increase in one’s income, measured as the amount one is allowed to
maximally spend on housing services, significantly increases households’ homeownership propensities, controlling for other variables.
Hypothesis II (H2): Given the structure of the Dutch housing market, modeling the effects of
income and financing costs through the maximum mortgage obtainable improves on previous approaches, as it better reflects the process underlying young renters’ transition into owner-occupancy.
By focusing on young households, which typically have little savings, this thesis addresses wealth-induced endogeneity, a phenomenon related to the likely simultaneous causality between wealth and homeownership (Herbert & Tsen, 2007). In Table I an overview of this study’s main features, compared with previous tenure choice studies, is provided.
Table I
Overview of the Reviewed Literature’s Main Components
Table I provides an overview of the reviewed literature on tenure choice studies. For each study, the geographic
focus, observation period, sample of interest, data format, employed methodology, number of observations and the way by which income and financing costs are modeled is stated.
4
Methodology
In this section, the methodology by which the hypotheses are tested is described. First, the type of analysis employed is discussed. Second, the used model is described. Third, the variable of interest is discussed. Fourth, important control variables are described. Fifth, the model’s specification is stated. Finally, the model’s limitations are discussed.
4.1
Survival Analysis
In line with the reviewed literature, this thesis models households’ tenure choice behavior by ways of a survival analysis. This type of analysis is concerned with the length of time before an event takes place and typically marks the transition from one state into another. It allows for both temporal and cross-sectional variation and is capable of coping with the difficulties that arise with episodes that have not (yet) resulted in the subject experiencing the event of interest (Jenkins, 2005; Deurloo, Dieleman, & Clark, 1997).
In survival analysis, one is interested in modeling the hazard rate ℎ(𝑡). The hazard rate represents a subject’s probability of moving from one state to the next at t conditional on not having made the transition up until t. Here, t represents ‘analysis time’, the passing of time since which a subject has been at risk of experiencing the event of interest (Hosmer, Lemeshow, & May, 2008). In this thesis, analysis time is defined as the number of years since which the head of the household has turned 18, while the hazard rate represents the subject’s probability of achieving homeownership at t conditional on having been a renter up until t, which can be expressed as follows:
ℎ(𝑡) = 𝑓(𝑡)/ 𝑆(𝑡) (1)
The numerator in this expression is defined as:
𝑓(𝑡) = lim∆𝑡→0Pr(𝑡 ≤ 𝑇 ≤ 𝑡 + ∆𝑡) / ∆𝑡 (2)
In this expression, T specifies the period in which the subject’s move into ownership takes place, while 𝑓(𝑡) can be interpreted as the probability of the subject moving from one state to the next within the interval [𝑡, 𝑡 + ∆𝑡]. Given 𝑓(𝑡), the failure function 𝐹(𝑡) can be derived:
The failure function is inextricably linked to the survival function 𝑆(𝑡), which constitutes the denominator of the dependent variable as defined in equation (1). The survival function 𝑆(𝑡) is equal to 1 − 𝐹(𝑡) and, in this thesis, represents the probability of renters not having moved into ownership up until some value of t. The survival function can be expressed as follows:
𝑆(𝑡) = Pr (𝑇 > 𝑡) (4)
To conduct a survival analysis, the used dataset needs to contain data on whether or not a subject has experienced the event of interest during some time interval, the period in which the subject became at risk of experiencing the event of interest and the period in which a subject became under observation, labeled as ‘entry time’. In Table II, the different components of a survival analysis, and how they are defined in this thesis, are summarized.
Table II
Survival Analysis’ Main Components
Table II displays the main components of this study’s survival analysis. In line with Andrew et al. (2006), a
household becomes ‘at risk’ of experiencing the failure event in the year the head of the household reaches adulthood. The time until the failure event is measured in years since the onset of the risk (analysis time), while the entry time marks the year in which a household entered the sample.
4.2
Model Choice
In line with Herbert and Tsen’s (2007) paper on the effect of down payment assistance on individual’s homeownership propensities, the hazard rate is modeled by use of a Proportional Hazards (PH) model. In a PH model, the hazard rate is assumed to satisfy a separability assumption that allows it to be expressed as follows (Jenkins, 2005):
ℎ(𝑡, 𝑋) = ℎ𝑜(𝑡) ∗ exp (𝑥′𝛽) (5)
Here, ℎ𝑜(𝑡) describes the pattern of duration dependence and is called the ‘baseline hazard’. This pattern of duration dependence is a function of analysis time and is assumed to be
common to all subjects.15 The second component of equation (5), exp (𝑥′𝛽) describes a
subject-specific function of covariates that scales relative to the baseline hazard (Jenkins, 2005). In a PH model, the two separate components are multiplicatively related (Hosmer, Lemeshow, & May, 2008) and each coefficient summarizes the proportional effect of absolute changes in the covariate of interest on the hazard rate. PH models function under the assumption that the included covariates are not duration-dependent, meaning that their effects are constant over (analysis) time. This assumption forms the foundation of PH models and is tested in section 6.
With regard to the baseline hazard, PH models allow for different specifications, of which the Parametric Proportional Hazard (PPH) and the Semi-parametric Proportional Hazard (SPH) are most frequently used in the financial literature (Hosmer, Lemeshow, & May, 2008). In a PPH specification, one imposes a structure on the distribution of the baseline hazard, while with the SPH specification, no structure on the baseline hazard is imposed. In this thesis, use if made of a Cox model, a parsimonious SPH model that is more flexible than its PPH counterparts and therefore can lead to a better representation of the hazard rate, as fewer restrictions are imposed (Hellman & Puri, 2002). In sum, a Cox model allows for the trend of renters becoming homeowners to be derived, rather than imposed.
4.3
Variables of Interest
As young aspiring homeowners are dependent on mortgage financing to fund the purchase of a home, the maximum mortgage they can obtain is likely to have a significant impact on first-time buyers’ homeownership propensities. This maximum obtainable mortgage can be approximated by one’s borrowing capacity as calculated by the Nibud:
𝐵𝑖𝜏 = 𝜅𝑖𝜏𝐼𝑖𝜏(1−(1+𝑅𝜏)
−L
𝑅𝜏 ) (6)
A household’s borrowing capacity in year 𝜏 is a function of household income 𝐼𝑖𝜏, the prevailing mortgage interest rate 𝑅𝜏, the length of the mortgage L (fixed at 30 years) and the
percentage of the income which is allowed to be spent on housing services according to the Nibud: 𝜅𝑖𝜏.
If first-time buyers would be allowed to utilize their pension contributions to make a down payment on a home, this reduction in pension savings is assumed to influence their
tenure choice behavior through their borrowing capacity, as the amount of income that can be spend on housing services increases. Focusing on this portion of income, rather than total income, fits the analysis well, as the newly freed-up pension capital is to be used for the purchase of a home only, not for something else. Since pension contributions are paid out as a percentage of gross income, rewriting equation (6) using a logarithmic transformation leads to the borrowing capacity formula suiting this study’s purpose:
ln (𝐵𝑖𝜏) = ln (𝜅𝑖𝜏𝐼𝑖𝜏(1−(1+𝑅𝜏)−L
𝑅𝜏 )) (7)
ln (𝐵𝑖𝜏) = ln(𝜅𝑖𝜏𝐼𝑖𝜏) + ln (1−(1+𝑅𝜏)−L
𝑅𝜏 ) (8)
This specification now allows for identification of two separate components: the (maximum) amount spendable on housing services based on subject i's income in year 𝜏 and an annuity factor. The first component is this study’s variable of interest, while the second components serves as a proxy for a household’s financing opportunities in year 𝜏.16
4.4
Control Variables
Beside a household’s income, the transition into ownership is linked to many events in, and characteristics of, the household. Here, several control variables are briefly discussed.
One of the most frequently included covariates in tenure choice studies relates to the head of the household’s education level. A subject’s education level is indicative of its income potential and also serves as a proxy for its knowledge of alternatives on the mortgage market (Helderman A. C., 2007). A subject’s education-level affects mortgage lenders’ willingness to supply credit as well, as Henderson and Ioannides (1987) found that young people with low incomes and a low education-level faced an increased probability of being denied a mortgage when compared to their more educated counterparts.
In addition to the head of the household’s education level, marital status has been found to be a primary demographic determinant of tenure choice behavior (Boehm & Schlottman, 2014). On the one hand, marriage serves as a proxy for union commitment, and more committed households are inclined to opt for owner-occupied housing (Feijten & Mulder, 2002). Moreover, marriage is likely to affect the maximum mortgage one can obtain,
as mortgage lenders deem married couples to be more stable. On the other hand, high-commitment households might be less inclined to move, as an increased probability exists that one or more of its household members is disadvantaged by the move (Mulder, 1993). A similar intuition applies to household size, as larger households are more likely to have one of its members disadvantaged by a move. Moreover, larger households tend have higher levels of non-housing expenditures, making it less likely that they can afford homeownership (Boehm & Schlottman, 2014).
Because of factors such as emotional place attachment and location-specific capital, people tend to move over short distances (Clark & Dieleman, 1996). This applies to the Netherland specifically, as three quarters of all moves are intra-municipal (Feijten & Visser, 2005). For this reason, it is likely that a household’s preference for a regional market sterns from their current residential location. Given the structure of the Dutch housing market, in which urban areas have a much smaller share of owner-occupied housing compared to the more rural areas (De Groot, Manting, & Mulder, 2013), the degree of urbanity of the region in which the household currently resides can serve as a proxy for the tightness of the housing market that it intends to move to.
Finally, the macroeconomic context in the form of financing costs and housing prices is important as well, as both have an impact on the affordability of owner-occupied housing. Higher housing prices and higher financing costs (typically modeled by ways of the mortgage interest rate) make first-time homeownership more expensive (Helderman A. C., 2007). However, house price appreciation has been found to have a positive effect on renters’ homeownership propensities as well, as recent housing returns attract aspiring homeowners (Andrew, Hauren, & Munasib, 2006).
4.5
Model Specification
Based on the discussion of this study’s research design, the specification by which the hazard rate is modeled in this thesis extends on equation (5) and can be expressed as follows:
ℎ(𝑡|𝑋𝑖𝑡, 𝑌𝑖𝑡, 𝑍𝑡) = ℎ𝑜(𝑡) ∗ exp (𝛼𝑋𝑖𝑡+ 𝛽𝑌𝑖𝑡+ 𝛾𝑍𝑡) (9)
Here, 𝑋𝑖𝑡 is the amount a household is allowed to spend on housing services according to the Nibud, 𝑌𝑖𝑡 are household-specific control variables (e.g. relative annuity factor, marital status,
currently resides in) and 𝑍𝑡 refers to a time-varying covariate that is equal for all households
(e.g. house price appreciation).
4.6
Limitations of the Model
While this study takes a novel approach in modeling the tenure choice behavior of young households, it suffers some shortcomings that need to be addressed.
First, as pension contributions are paid out as a percentage of gross income, freed-up pension capital in the form of useable pension contributions affect households’ disposable income, not gross income. The amount a household is allowed to spend on housing is based on gross income however, which complicates the ‘income-shock’ households experience when they are allowed to utilize their pension contributions for the purchase of a home. With the current modeling procedure, it is thus implicitly assumed that the Nibud will adjust the percentage of income households are allowed to spend on housing upwards, since disposable income will increase. Second, this approach assumes that young households see their first home as a perfect substitute to their pension as a vehicle for wealth accumulation, so that their risk profile remains unaffected. Third, given the way in which the variable of interest is constructed, allowing young households to use their pension contributions to fund a home purchase will raise the amount they could maximally borrow for the length of the entire mortgage-term. It is unrealistic that policymakers would allow households to do so, as households would then be shifting most of their savings from a diversified pool of investments (pension) into one specific asset: their home.
In sum, the approach taken in this study suffers some noteworthy pitfalls. However, if young households would be allowed to use their pension contributions for the purchase of a home, this modeling approach is likely to better reflect policymakers’ intended effects. Given the structure of the variable of interest and the employed methodology, it will therefore provide insights with regard to the proposal’s maximum effect on young households’ homeownership propensities.
5
Data and Descriptive Statistics
In this section, the used data is discussed. First, the different data sources are described. Second, the process of constructing the used dataset is discussed. Third, descriptive statistics are provided. Fourth, the size of pension contributions as a portion of income is provided for different age cohorts.
5.1 Data Sources
The data used in this thesis comes from four sources: the DNB Household Surveys (DHS), the Dutch Association of Real Estate Brokers and Real Estate Experts (NVM), the Central Bureau of Statistics (CBS) and the Nibud. An overview of these data sources, and what is extracted from them is provided in Table III. Here, the sources most important for this thesis, the DHS and the Nibud, are briefly discussed.
Table III
Overview of the Used Data Sources
This table states this study’s key data sources and what is extracted from them. For the mortgage interest rates, quarterly rates were converted to annual ratesusing simple averages. *All variables, except for the ‘Maximum Portion of Income Allowed to be Spend on Housing’ cover the 1994 – 2014 period. In order to maximize the size of the studied sample, the ‘Maximum Portion of Income Allowed to be Spend on Housing’ of 1995 was used for 1994 as well.
The DHS comprises longitudinal data on economic and psychological aspects of households’ financial behavior. It is sponsored by De Nederlandsche Bank (DNB) and constitutes the main project of the CentERpanel, (a panel) representative of the Dutch-speaking population. Recruitment for CentERpanel takes place based on a random sample drawn from a private postal address file. In the CentERpanel, households are compensated for completed questionnaires and, in case of attrition, replaced by households with similar characteristics with regard to income level, region and head of the household’s age. This way, the CentERpanel attempts to preserve its continuity.17
The DHS essentially consists of three components. The first relates to the tracked household members’ personal characteristics. The second consists of five questionnaires
17 A completed questionnaire leads to a reward of about €0.25. The respondent can decide to have this amount
related to work and pensions, housing, income and health, assets and debt and economic and psychological concepts. The third component consists of two data-files that are derived based on the five questionnaires and contains aggregate information on wealth and income. The questionnaire related to housing forms the foundation of this thesis, while the first and third components serve as its supplements. The CentERpanel mirrors the Dutch population on most demographic and financial characteristics, and sample weights can used for the characteristics that is does not (Teppa & Vis, 2012)18.The qualities of the DHS have led it to be frequently
employed as data source in academic research and policy-oriented studies focused on households’ financial behavior. These studies relate to a variety of subjects such as wealth accumulation (van Rooij, Lusardi, & Alessie, 2012), risk aversion (Kapteyn & Teppa, 2011), stock market participation (van Rooij, Lusardi, & Alessie, 2011) and consumer’s choice of payment (Bolt, Jonker, & Renselaar, 2010). Given its characteristics, the DHS provides a good foundation to study the policy implications of the proposed link between pensions and housing.
The second (main) source employed in this thesis is the Nibud. The Nibud is a foundation established in 1979 concerned with providing information on the planning of family finances. Each year it provides guidelines to families, financial intermediaries and the government on ‘rational’ financial planning (Francke, Van de Minne, & Verbruggen, 2014). In this thesis, the focus is on the portion of a household’s income one can spend on housing. This portion is expressed as a percentage of the main wage earner’s income and is derived based on a residual method in which income is adjusted for non-housing costs.Assuming that non-housing costs do not increase proportionally with income, the Nibud ‘allows’ households with higher income levels to spend a larger portion of their income on housing services.
5.2
Dataset Construction
Based on unique individual identifiers the different DHS waves were first merged per component.19 Subsequently, these different components were merged with one another,
resulting in an unbalanced panel dataset. As stated in section 4, survival analysis requires a specific data structure and therefore, several transformations took place before the used dataset was finalized.
Focusing on a sample of households that were renters at the start of the observation period, the event of interest, a move into owner-occupancy, was identified based on the year
18 As sample weights were only available as of 2001, no sample weights were used in this study. 19 Derived based on unique household identifiers and its members’ identification number.
since which a household has lived in their current home. The reported year is assumed to be the year in which the adjustment in housing consumption through moving takes place. Based on the tenure status after the move, the nature of the housing consumption adjustment is identified. The year in which this event took place marks the event of interest. On several occasions, a household reported that it had moved during the observation period (e.g. 1999) but dropped out for two year (e.g. 1999 and 2000) and then re-entered the sample. For these cases, data was generated based on an inter- and extrapolation procedure described in Table
XV.
In addition to the event of interest, the period in which the subject became at risk of experiencing the failure event had to be identified. In this study, the onset of the risk is marked as the year in which the head of the household reached adulthood. The studied sample regards potential first-time buyers, defined as households of which the head was 35 or younger at the start of the observation period.20 Clark et al. (1994) were the first to observe
that first-time buyers are concentrated in this age cohort, and Francke et al. (2014) and Fernandez-Coregudo and Muelbauer (2006) used this cohort to proxy for (actual) first-time buyers in their studies.
To enable the study of young renters’ homeownership propensities, the NVM, CBS and Nibud data were merged with the DHS data on an annual basis. Then, transformations were executed to properly define the variable of interest and control variables. In Table IV, the model’s used covariates and their specification are described, while in Table V, the covariates’ expected signs, based on sections 2 and 3, are stated.
20 In this thesis, the head of the household is defined as the household member that accumulates the most income
Table IV
Covariates’ Employed Specifications
In this table, the main model’s covariates and their employed specification are described. All demographic variables (‘Education Level’, ‘Marital Status’, ‘Urbanity’ and ‘Household Size’) are directly extracted from the DHS and are modeled as dummies. The income variable is defined as the logarithm of the ‘Amount Allowed to be Spend on Housing’ according to the Nibud. Financing costs are included by ways of the difference in the annuity factor relative to the first year a household came under observation (in logs; negative differences were replaced with zeros). Housing price appreciation is accounted for employing CBS’ housing price index.
Table V
Covariates’ Expected Coefficients
In this table, the main model’s covariates, expected signs and the rationale behind these expected signs are stated.
5.3 Descriptive Statistics
Now that the data has the appropriate structure, descriptive statistics can be provided. In
Table VI a condensed life-table of young households transitioning from renting into
Table VI Condensed Life Table
In this table, an overview of the used data is presented by ways of a life table. Here, ‘Analysis Time’ marks the number of years since the head of the household turned 18. The survival function (reported in the full life table in
Table XIV) is estimated based on the risk pool and the number of households transitioning into ownership.
As Table VI shows, a large number of observations got censored over the observation period. Most of the subjects first enter the sample when the head of the household is between 25 and 36 years old. In this sample, none of the households become homeowners before the age of 22. Thereafter, the portion of households becoming homeowners gradually increases up until the head of the household turns forty. Then, the number of renting households transitioning into homeownership decreases. This trend in households’ homeownership propensities is consistent with the empirical literature (De Groot, Manting, & Mulder, 2013), as households that are in the process of family-formation are likely to accommodate this process by looking for housing that will suit their (future) family’s needs. Upon realizing household stability, it becomes less likely that renters will transition into ownership, as the current dwelling is likely to accommodate the household’s housing needs already.
In Table VII, descriptive statistics for the studied sample are provided. Variables’ means, standard deviations and ranges are displayed for two subsamples: renters becoming homeowners during the observations and renters that do not become homeowners during the observation period.
Table VII
Descriptive Statistics of the Studied Sample
This table shows the means, standard deviations (S.D.), minimums (Min.) and maximums (Max.) for this study’s fundamental variables. Moreover, the numbers of observations these statistics are based on are reported in the upper row.
Table VI shows that on average, the renters that do not become homeowners during the
observation period enjoy higher incomes than the ones that do make the transition into owner-occupancy. Moreover, the interest rates prevailing in the mortgage market appear to be lower for those that remain renters, indicative of lower financing costs.21 This appears to be
counterintuitive, as lower mortgage interest rates and higher incomes are associated with better affordability of owner-occupied housing. However, the portion of income individuals are allowed to spend on housing services is higher for the renters that become homeowners compared to those that do not.
In addition to the relevant financial determinants, demographic variables show that (on average) renters that transition into owner-occupancy during the observation period are more educated, with only 8% having enjoyed education up to lower secondary schooling. Moreover, households that make the transition into homeownership are marked by relatively
21 This finding is sample-specific since there would be no difference observed if the studied sample was more
few married couples and smaller household sizes, indicative of higher mobility relative to the renters that remain renters. House price appreciation is lower during the observation period for those that remain renters, and a large share of the renters remaining renters reside in the more rural areas.
In addition to the dataset based on which the model is estimated, the magnitude of the pension contributions as a portion of income needs to be acknowledged. In Table VIII, pension contributions expressed as a percentage of income for the cohorts of interest are stated.
Table VIII
Pension Contributions as a Fraction of Income
In this table, pension contributions made out of an individual’s income are displayed as a function of gross income for different age cohorts.
The percentages reported mark the ‘income-shock’ young households will experience when the proposed policy goes into effect. Based on these shocks, policy simulations are conducted in section 6.
6
Results
In this section, the results of this study’s empirical analysis are presented. First, the outcomes of the conducted survival analyses are described. Subsequently, the outcomes of the proportional hazards assumption test are discussed. Finally, the results of various policy simulations are provided.
6.1
Model Outcomes
The estimation results of the Cox model used to analyze the time to first-time homeownership are reported in Table IX. To extend on the existing literature and test H2, several Cox-regressions were performed. Coefficients, hazard ratios, Z-scores and significance levels are reported for all models.22 A positive coefficient indicates that a covariate has an accelerating
effect on the time until a renting household moves into ownership, or alternatively, a lower likelihood of remaining a renter. Hazard ratios on the other hand represent the ratio of the hazard associated with a one-unit change in an independent variable relative to the hazard rate before this change. A hazard ratio greater than one indicates that an increase in the independent variable raises the probability that a subject moves into ownership. Since hazard ratios are more easily interpreted, the focus in the remainder of this section is on the reported hazard ratios. As the studied sample includes multiple observations per household, household-clustered standard errors were used to account for within-household correlation.
Model I follows the literature prior to Clark et al. (1994) and focuses on households’
life-course characteristics exclusively. In this model, income is included directly using a log-specification. Model II extends on Model I by including covariates related to the macroeconomic context and (regional) housing market characteristics. In Model III, the effects of income and financing costs on households’ homeownership propensities are jointly modeled by use of households’ borrowing capacity. Model IV builds upon Model III, but isolates the effects of income and financing costs by decomposing households’ borrowing capacity into the maximum amount it is allowed to annually spend on housing and its relative annuity factor (both using a log-specification).
Table IX
Results Survival Analysis
This table displays the outcomes of this study’s survival analysis. Model I focuses on households’ life-course characteristics. Model II extends on Model I by including covariates related to the macroeconomic- and regional context. In these two models, ‘Income’ is included directly using a log-specification, while ‘Financing Costs’ are modeled as changes in the mortgage interest relative to the year a household entered the sample.
Model III models the effects of income and financing costs jointly through households’ borrowing capacity. Model IV separates households’ borrowing capacity into an income and financing costs component. The income
component comprises the amount a household can maximally spend on housing services (annually) as determined by the Nibud, while financing costs are modeled by the changes in the annuity factor relative to the year a household entered the sample (both components have a log specification). The Efron method is used for handling ties. *** indicates significance at 1%, ** indicates significance at 5%, and * indicates significance at 10%.
In all models, income (through different specifications) significantly impacts households’ homeownership propensities. In Model I, the impact of income is largest, as a percentage-point increase in gross income raises the hazard rate by 111% (ceteris paribus). When adding covariates related to the macroeconomic context and housing market characteristics, the income coefficient in Model II decreases slightly, but still increases the hazard of making the move from renting to owning by 89%. When the effect of income is modeled jointly with the effect of financing costs using households’ borrowing capacity, a percentage point increase in the household’s borrowing capacity increases the hazard of transition by 62%. In Model IV, where the effects of financing costs and income are isolated, the coefficient of the amount of income households are allowed to spend on housing services shows that for every
