Saving for a House:
The Effect of
Stricter Mortgage Regulations
*
MSc Economics & MSc Finance
Combined Master’s Thesis (EBM000A20)
Inge Krijgsheld (s2350548) Supervisor: Prof. dr. R.J.M. Alessie
23-06-2017
Abstract: This paper examines the effect of the recent introduction of stricter mortgage regulations on the
saving behaviour of Dutch households. The Dutch government has recently introduced the Temporary Regulation for Mortgage Credit, which imposed legal limits to Loan-to-Value and Payment-to-Income ratios. In addition, new restrictions to the mortgage interest deductibility scheme were introduced. Together, these regulations imply that households need to bring in more savings to buy a house, ceteris paribus. I use a difference-in-difference approach to examine whether households adjust their saving behaviour. As households intending to buy a house will be affected by the stricter regulations, I take these households as my treatment group, and investigate whether their saving behaviour changes relative to the control group of households that do not have any plans to buy a house. I find evidence that both the amount of savings and the probability that a household is saving increased after the introduction of the stricter mortgage regulations. This effect is most notable for (young) households currently renting accommodation.
JEL Classification: D14, D91
Keywords: household savings, mortgages
2
1. Introduction
During the 1990’s and the early 2000’s house prices were rising fast and did not appear to decrease any time soon. Many households that bought a house took out large mortgages, as the price was believed to increase in later years to eventually exceed the value of the mortgage. Consequently, the need to save for the purchase of a house was not great, as most costs could be covered by the mortgage. After the Global Financial Crisis, however, the consequences of these large mortgages became abundantly clear, as it is generally accepted that the crisis stems from the subprime mortgage market in the US. To prevent households from ending up with a mortgage they cannot afford, the Dutch government has implemented various restrictions to the supply of mortgages in recent years. These restrictions include maximum Loan-to-Value (LTV) and Payment-to-Income (PTI) limits, as well as stricter limitations for the Mortgage Interest Deductibility (MID) scheme. Ceteris paribus, this implies that households need to bring in more savings in order to buy the house of their choosing.
This raises the question as to how households are responding to these mortgage reforms. It is very important that households adjust their saving behaviour in response to the stricter mortgage regulations. A recent study by Hekwolter of Hekhuis et al. (2017) shows that the contribution of savings is becoming increasingly important in the Dutch housing market, especially in the large cities. If households do not save for a down payment, they may not be able to enter the housing market. According to Hekwolter of Hekhuis et al. (2017), this could be especially problematic for middle-income households, as there is a shortage of housing in the rental market, and their income is too high to be eligible for social housing.† Not being able to enter the housing market will therefore severely limit their housing options. On the other hand, instead of increasing the amount of savings, households could also rely on an increase in the amount of inter-vivos transfers from parents. This is not an option for all households, however. Therefore, the research question I will address in this paper is the following.
Do households adjust their saving behaviour in response to stricter mortgage regulations?
This paper investigates the impact of the stricter mortgage regulations on saving behaviour of Dutch households. Households that are planning on buying a house will be the households that are affected by the stricter mortgage regulations. Therefore, I estimate the relationship between having the intention to buy a house, either in the short run or long run, and the amount of savings. Additionally, I check whether
†
3
households that plan on buying a new house have a higher probability to save. After the stricter mortgage regulations have been implemented, households need to bring in more savings for the purchase of a house. Therefore, I hypothesize that the effect of intending to buy a house on the probability that a household is saving and on the amount of savings will be larger after these regulations have been introduced. Furthermore, as homeowners are able to use their housing wealth to buy a new house, I expect that the effect of planning to buy a house on the likelihood of saving and on the amount of savings will be largest for young households that are currently renting accommodation.
In order to conduct this analysis, I use data from the DNB Household Survey. In this survey, respondents are not only asked about their plans to buy a (new) house in the near future, but are also asked extensive questions on saving behaviour. I use responses to these survey questions to examine whether the relationship between the intention to buy a house and the amount of savings, as well as the relationship between the intention to buy a house and the probability that as household is saving has changed after the stricter mortgage regulations have been introduced.
I find evidence that the effect of intending to buy a house in the short run on the probability of saving and on the amount of savings is larger after the introduction of the stricter mortgage regulations. Moreover, I find that this effect is largest for (young) households that are currently renting accommodation. This increase in the likelihood of saving and in the amount of savings is mostly present for households that intend to buy a house in the short run. Few results support the hypothesis that households intending to buy a house in the long run have increased the amount of savings as a result of the stricter regulations.
This research will contribute to the existing literature as this is, to my knowledge, the first paper to use micro data on saving behaviour to examine the impact of stricter mortgage regulations on household saving behaviour. Previous research has been conducted on the effect of (stricter) mortgage regulations, but this research differs from previous studies in either one of two ways. First, many of these studies are cross-country studies (see e.g. Hayashi et al., 1988; Jappelli and Pagano, 1994) instead of longitudinal research, which is done in this paper. The use of longitudinal data allows me to research in what way households respond to the introduction of stricter mortgage regulations. Second, other studies such as Zeldes (1989) and Aron et al. (2012) use data on consumption in order to comment on saving behaviour, whereas I use microdata on savings, which allows me to observe household saving behaviour directly.
4
and will therefore contribute to the prevention of another crisis in the housing market. As young, renting households do adjust their saving behaviour, they are likely still able to enter the housing market. The stricter regulations do come at a cost, however. Households must delay the purchase of a house until their savings are sufficient for a down payment. Furthermore, these mortgage credit constraints cause a distortion of the optimal distribution of consumption over the households’ life cycle, as households need to limit consumption in order to save for a down payment. If the government is looking to boost consumption, it can therefore consider easing the mortgage regulations. As an alternative, a stimulation of intergenerational transfers of wealth may limit the distortion of the optimal distribution of consumption.
This paper is organized as follows. In section 2, I discuss the recent mortgage reforms that the Dutch government has implemented after the financial crisis. In section 3, I elaborate on the existing literature. I then go on to discuss the methodology in section 4 and the data used in section 5. The results are given in section 6, and section 7 concludes.
2. Recent Mortgage Reforms
The government has been concerned about excessive loan sizes in the Netherlands for many years. The government already encouraged banks and the AFM (Authority Financial Markets) to adopt a new code of conduct concerning mortgage loans in 2007, before the Global Financial Crisis had begun. However, in 2011 the residential mortgage debt was still extremely high at 106.2% of GDP (EMF, 2012), which gave rise to the implementation of new rules with regard to mortgage loans. This section will discuss several important regulatory changes in mortgage loans in response to the Global Financial Crisis. For an overview of all regulatory interventions in the Dutch housing market, see Scanlon and Elsinga (2014).
5
Figure 1: Maximum PTI ratio norm by income and interest rate in 2016
Source: Nibud
to the highest individual income plus 50% of the lowest income is used. This PTI ratio is then applied to total household income. To ensure robustness and manageability of the PTI ratios, individual households’ expenses are not considered, but standard expenses based on a household of two adults (with one income) and two children is used. The maximum mortgage that a household can obtain is the amount of a 30-year fixed interest annuity mortgage for which the annuity payments coincide with the maximum mortgage payment used in the PTI ratio. This formula is applied to all mortgages, regardless of the actual mortgage type. The interest rate of the actual mortgage is used to calculate interest payments, unless the fixed interest rate period is shorter than ten years. In that case a regulatory interest rate is applied, which is determined each quarter by the AFM. Figure 1 displays how PTI ratios are related to income and the mortgage interest rate. From Figure 1 it can be deduced that PTI ratios are positively related to both income and the interest rate on the mortgage. This is because basic consumption needs do not rise proportionally with income, giving the opportunity for higher mortgage payments. Furthermore, as the interest rate increases there are higher mortgage interest tax deductions, so that higher PTI ratios are possible. Mortgages with PTI ratios below the limit are called comply-mortgages, whereas mortgages with PTI ratios above the limit set by Nibud are referred to as explain-mortgages, which have to be documented and justified by the lender.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 24000 42000 64000 74000 84000 96000 P T I R atio
Annual Gross Income (in €)
6
Figure 2: Maximum mortgage Payment-to-Income (PTI) ratio norms on mortgages with 5%-5.5% interest rate from 2008-2016
Source: Nibud
In January 2011 legal limits to the PTI ratios were introduced (SEO, 2011). Adherence to the PTI limits was no longer voluntary under the CCM, but was now officially regulated. Moreover, the maximum PTI ratios were reduced, especially for households with relatively low incomes. In addition, five restrictive exception categories were defined, which made it more difficult for lenders to justify loans that exceeded the PTI limits. Explain-mortgages were only allowed under exceptional circumstances such as a committed wage increase, entrepreneurs without a fixed income (PTI ratios are then based on the average historical income), available liquid wealth, a mortgage rollover, or energy-saving investments in the house. Figure 2 shows the relationship between the maximum PTI ratio and income between 2008 and 2016. The decrease in maximum PTI limits in 2011 can be seen in Figure 2. For the lowest incomes there is a large decrease in the maximum PTI limit between 2010 and 2011. For the higher incomes, this effect is not as large. Nonetheless, the decrease in the maximum PTI limits for the higher incomes is clearly visible in Figure 2.
In addition to the tighter PTI rules, a new legal maximum LTV limit was established in 2011, on August 1st. The LTV represents the maximum mortgage that a household can obtain, based on the current value of the house for which the mortgage is taken out. The value of the house is either the price the household will pay or the value set by an independent appraiser, whichever is lower. The maximum LTV had been
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 17000 27000 37000 47000 57000 67000 77000 87000 P T I R atio
Annual Gross Income (in €)
7
112% before the new regulation (SEO, 2011). The new rule had only been announced a few months before by the Minister of Finance, which left prospective buyers with little time to adjust. The new LTV limit was set at 106%, whereas values above that limit occurred quite frequently before then. Furthermore, the part of the mortgage above 100% of the value should be repaid within seven years. Banks’ compliance with both the PTI and the LTV limit was to be monitored by the AFM regulator.
On January 1st 2013, the Temporary Regulation for Mortgage Credit (Tijdelijke Regeling Hypothecair
Krediet) was passed with further limitations to new mortgages. First of all, under the new regulation the
legal maximum PTI ratios were further reduced. However, this time the higher income households were affected by these reductions more than lower income households (see Figure 2). Moreover, the regulation imposed additional restrictions on the LTV limit. Starting January 1st 2013, the legal LTV limit would be 105%, and would gradually be reduced with one percentage point each year to 100% in 2018. In addition, on April 26th 2012, the government had announced more mortgage reforms, concerning the Mortgage Interest Deductibility (MID) scheme. These reforms were part of the EU stability pact to reduce the budgetary deficit to 3% of GDP. Starting from January 1st 2013, new mortgages had to be repaid within thirty years in annuities or linear payments in order to be eligible for the MID program. Moreover, from January 1st 2014 the highest tax rate that can be used to deduct taxes decreases with 0.5 percentage points each year, until it reaches 38% in 2044 (the highest tax rate is 52%). These new rules do not apply to mortgages that were closed before 2013. However, in the case of a mortgage refinancing, the amount that is added to the previous mortgage cannot benefit from the old MID rules. To limit the negative effects of these stricter mortgage regulations on the housing market, the Dutch government increased the amount of money exempt from taxation that parents are allowed to gift their children. Previously, parents were allowed to give their children approximately €52,000 tax-free. From October 2013 until the 1st
of January 2015 this amount was raised to €100,000. However, the gift had to be used for either the purchase of a house, repayment of mortgage debt, renovation of the house, or buying off the ground rent.
8
3. Literature Review
Although the purchase of a house is most commonly financed with a mortgage, this usually requires a down payment. Thus, to purchase a house, individuals need to have sufficient savings. This implies that young households need to delay the purchase of a house until their savings are high enough for a down payment (Chiuri and Jappelli, 2003). Among the first to discuss the theoretical implications of this down payment are Jackman and Sutton (1982) and Artle and Varaiya (1978). Jackman and Sutton (1982) present a lifetime consumption plan with the purchase of a long-lived physical asset (such as a house). An individual wishes to purchase a physical asset with the value 𝐴0. The individual can take out a loan for the purchase, using the physical asset as collateral. However, due to credit constraints, the individual must pay transaction costs 𝛾. Thus, the purchase will actually cost the individual (1 + 𝛾)𝐴0. In the case that the
physical asset is a house, 𝛾𝐴0 can be thought of as the size of the down payment. Individuals then face the
following optimisation problem:
max𝑐(𝑡),𝑎(𝑡)∫ 𝑢[𝑐(𝑡)]𝑑𝑡0𝑇 (1)
subject to
𝐾̇ = 𝑦(𝑡) − 𝑐(𝑡) + 𝑟(𝐾 − 𝛼𝐾0) + 𝛼(𝑅 − 𝑟)𝐴0 (2)
𝐾(𝑡) ≥ 𝛼𝐾0 (3)
This optimisation problem can be interpreted in the following manner. The individual maximizes consumption over his lifetime, until he dies at time T. The individual plans to buy a physical asset (e.g. a house) at some point in the future, denoted by the binary 𝛼, which takes a value of 1 if the individual has purchased the physical asset, and a value of 0 if he has not yet purchased the physical asset. The annual savings, 𝐾̇, at time 𝑡 are the individual’s income at that time (𝑦𝑡) minus the amount he consumes (𝑐𝑡), plus any interest income earned on previous savings (𝑟𝐾). If the individual purchases the physical asset, he must pay a down payment of the amount 𝐾0, after which 𝛼 takes a value of 1. Annual savings are then composed of the individual’s income minus consumption at time 𝑡 plus earnings on previous savings, which are decreased with the amount of the down payment. Additionally, the individual earns a yield 𝑅 on his physical asset 𝐴0, which is reduced by 𝑟𝐴0 (the interest he must pay on the borrowed amount). The optimal consumption plan that results from this optimisation problem is displayed in Figure 3. Until time 𝜏0 the individual is liquidity constrained, and will consume his entire income, without accumulating savings. From 𝜏0 onwards the individual’s income is high enough to save for the down payment, and the
9
Figure 3: The optimal consumption plan with credit constraints (heavy line)
Source: Jackman and Sutton (1982)
income to increase, but decreasing his net wealth to zero. The individual again consumes all his income, until 𝜏1, when he starts to save for old age. Within this model, if access to credit is limited, the individual
10
Evidence that higher down payment requirements lead to higher saving rates can be found in Hayashi et al. (1988). The authors use survey data from a study by the Japanese ministry of construction to show that the Japanese tend to buy houses later in the life-cycle with higher down payment ratios, which causes higher saving rates early in life. Deutsch et al. (2006) also explain that the Japanese credit restrictions are the cause of the high average age at which the Japanese buy a house. A more recent example of deferred homeownership is given in Xu et al. (2015), where the authors show that mortgage accessibility is an important constraint to homeownership for millennials in the US. In addition, Fetter (2013) shows that mortgage subsidies, i.e. lower down payment requirements, lead to an increase in homeownership by decreasing the average age for which households purchase a house. Moreover, Engelhardt (1996), Moriizumi (2003), and Aron et al. (2012) use consumption data to examine the effect of credit constraints on consumption. Engelhardt (1996) argues that the need to accumulate a down payment before qualifying for a mortgage imposes a binding credit constraint for many households wishing to own a home. The author uses real food consumption data from the Panel Study of Income Dynamics to show that households decrease consumption to save for a down payment, as evidenced by a 10% higher growth in real food consumption after the household no longer experiences this credit constraint. Similarly, Moriizumi (2003) estimates that Japanese households consume 30% to 40% less in the period before the purchase of a house. Moreover, Aron et al. (2012) demonstrate the importance of credit constraints for consumption, by examining the consumption behaviour of households in the UK, US and Japan. In the UK and US there have been large increases in the availability of credit, which has caused the consumption functions of these countries to shift upwards. In addition, the authors find that consumption-to-income ratios increased in the UK and US as mortgage down payment constraints were relaxed.
11
Hypothesis 1: Households intending to buy a house in the (near) future increase the level of savings after the stricter mortgage regulations have been implemented, particularly young households currently renting accommodation.
Furthermore, Jackman and Sutton (1982) suggest that households facing a binding credit constraint can save more, or save longer. The findings by Deutsch et al. (2006) and Fetter (2013) imply that a higher down payment requirement leads to a higher average age at which households buy their first home. Thus, households need more time to save before their savings are sufficient for a down payment, which suggests that more households will be saving for a house at each point in time. This leads me to my second hypothesis.
Hypothesis 2: Households intending to buy a house in the (near) future have a higher probability to save after the stricter mortgage regulations have been implemented, particularly young households currently renting accommodation.
Furthermore, there are studies that focus on the effects of macro prudential policies, such as lowering LTV limits. Igan and Kang (2011) use microdata on variations in LTV and debt-to-income (DTI) limits across Korea and find that tighter LTV and DTI limits are associated with decreased transaction activity and fewer demand for homes, by decreasing the demand from households that already own a house. The authors find that the lower LTV and DTI limits are related to lower price expectations, which can play a key role in the formation of bubbles. Verbruggen et al. (2015) estimate that reducing the LTV limit in the Netherlands to 90% would lead to a long-run reduction in house prices of 10%, and decline consumption levels by approximately 1%. Moreover, the authors argue that the costs of lowering the LTV limit will mostly be felt during the transitional phase, during which the maximum LTV limit is gradually lowered, as individuals will need to increase savings to obtain the means to buy a house, which would lead to a reduction in private consumption and demand for owner-occupied housing. Although these effects would be mostly felt in the short run, financial stability would increase following the LTV reduction. The authors stress that the results need to be interpreted with caution, as the compulsory down payment represents a break with the past, and it is not known how existing homeowners will react and whether first-time buyers will anticipate this.
12
intending to buy a house will need to save more during the transitional phase. This leads to the following hypotheses:
Hypothesis 3: The effect of intending to buy a house on the amount of savings will increase each year starting from 2011, particularly for young households currently renting accommodation.
Hypothesis 4: The effect of intending to buy a house on the probability that a household is saving will increase each year starting from 2011, particularly for young households currently renting accommodation.
There is also a large strand of research that investigates the effects of housing-related fiscal policy, of which the Mortgage Interest Deductibility may be the most discussed in the Netherlands. The Mortgage Interest Deductibility can promote homeownership, which implies that eliminating this scheme, or limiting the tax benefits from this policy, can have a negative effect on housing transactions. Individuals may need to save more, as they are no longer able to afford a high mortgage. However, Hilber and Turner (2014) find that the MID merely promotes homeownership for the higher-income households. The effects of lowering the maximum interest rate will therefore not have large effects on lower- and middle-income households. Furthermore, although the MID reforms were not implemented until 2013, uncertainty around MID reforms had been present for years before that. Mastrogiacomo (2013) argues that policy uncertainty about the MID could increase savings. Thus, there may be limited effects on saving caused by the implementation of this policy. I therefore assume that any changes in saving behaviour caused by stricter regulations for taking out a mortgage will mostly be the result of reductions in the legal LTV and PTI limits.
4.1 Methodology
13
group is altered by the stricter mortgage regulations. I expect that the treatment group has more savings, and is more likely to save than the control group. If this difference is larger after the stricter mortgage regulations have been introduced, then I assume that these new regulations have caused the households that intend to buy a house to save more.
To test hypothesis 1 and 3, I take the amount of financial assets as the amount of savings. As financial assets are highly skewed due to income inequality, I take the natural logarithm of financial assets to transform the skewed distribution into one that is more approximately normal. The intention to buy a house can be both in the short run (within two years) and in the long run (after at least two years). Furthermore, I will control for the respondents’ sex, age class, education level, marital status, number of children, net income, and financial literacy. As the saving behaviour of the self-employed may significantly differ from other households (Hurst and Lusardi, 2004), a dummy for the self-employed will also be included. To capture the general trend in saving behaviour, I include time dummies. As mentioned above, most new mortgage regulations have been implemented in the year 2013, when the Dutch government introduced the Temporary Regulation for Mortgage Credit. I therefore examine whether the impact of intending to buy a house on financial assets is larger from 2013 onwards. To do so, I apply the difference-in-difference approach and include two interaction variables indicating that the household is intending to buy a house in the short run or long run from 2013 onwards. The equation that is used for estimation is (4). Again, data on income and financial assets can be highly skewed, which means that outliers can severely influence the results. I therefore estimate (4) using quantile regression, also known as the least-absolute-value model, with robust and clustered standard errors (see Parente and Santos Silva, 2016). Because this model minimizes the absolute values of the error terms (instead of the usual squared values of the error terms) it limits the impact of influential observations on the results.
ln(𝑊𝑖𝑡) = 𝛽1+ 𝛽2𝐼𝑆𝑅𝑖𝑡+ 𝛽3𝐼𝐿𝑅𝑖𝑡+ 𝛽4𝐼𝑆𝑅𝑖𝑡𝐷13𝑡+ 𝛽5𝐼𝐿𝑅𝑖𝑡𝐷13𝑡+ 𝑥𝑖𝑡′𝛾 + 𝑧
𝑖′𝛾 + 𝛾𝑡+ 𝜖𝑖𝑡 (4)
14
with the right hand side variables. As within-household correlation can cause the estimation of standard errors to be too small, I estimate (4) with standard errors corrected for clusters at the household level. ‡ The stricter mortgage regulations were not implemented all at once, but became stricter throughout the years. Thus, I expect that the relationship between intending to buy a house and financial assets will become larger over time, as more and more mortgage restrictions have been implemented. To test my third hypothesis, I estimate (5), where I replace the use of the interaction variables between the intention to buy a house and the year 2013, with yearly variables of the intention to buy a house. These yearly variables are created for the intention to buy a house in both the short run and the long run. I expect that the effect of intending to buy a house on financial assets will become larger over time, because the regulations for taking out a mortgage have become more and more strict. Therefore, I expect that there will be an increasing trend in 𝛽2 and 𝛽3. Any trend in these coefficients should take place between 2011
and 2016, as no new mortgage regulations were introduced between 2008 and 2010. According to hypothesis 3, the yearly effects from 2011-2016 should be different from the base year of 2008 and jointly significant, whereas the yearly effects of 2009-2010 should not be jointly significant. Moreover, I can use the estimation results of (5) to check whether I am correct in assuming a ‘jump’ in the relationship between financial assets and the intention to buy a house, as in specification (4). I do so by testing whether 𝛽2 is not significantly different from 2008 in the years 2009-2012, and whether the values of 𝛽2 for 2013-2016 are not significantly different from each other. I will conduct the same test for 𝛽3.
ln(𝑊𝑖𝑡) = 𝛽1+ 𝛽2𝑡𝐼𝑆𝑅𝑖𝑡+ 𝛽3𝑡𝐼𝐿𝑅𝑖𝑡+ 𝑥𝑖𝑡′ 𝛾 + 𝑧𝑖′𝛾 + 𝛾𝑡+ 𝜖𝑖𝑡 (5)
Again, I estimate (5) with the use of quantile regression to limit the impact of influential outliers. Furthermore, I expect that the effect of the intention to buy a house on savings may be different for households that currently rent accommodation. Homeowners may use their housing wealth to purchase a new house, whereas renting households must actively save for a down payment. Therefore, I will also estimate (4) and (5) for a subgroup of renters only. In addition, I use a second subset of young households (aged 45 or below) that are renting accommodation. Although I believe young households will mostly be affected by these stricter mortgage regulations, as I take each subgroup, the number of observations is significantly reduced. Therefore, I estimate (4) and (5) for these three different samples.
Additionally, to test hypothesis 2 and 4, I estimate the effect of the intention to buy a house on the likelihood of saving, by following Le Blanc et al. (2014). The measure of saving is derived from a
‡
15
question where the respondent was asked whether expenditures were higher, about equal, or lower than the income of the household over the past 12 months. I create a dummy variable that equals one if respondents indicate that expenditures were lower than income, suggesting that the household has saved money, and zero when expenditures were either about equal or lower than income, indicating that the household did not save. I then use equation (6), where 𝑆 represents the saving dummy, to estimate whether a household is more likely to save if it has the intention to buy a house. I expect that after the mortgage reforms a household is more likely to save if it has the intention to buy a house than before the reforms were implemented. I estimate (6) using a pooled probit model with standard errors corrected for clusters at the household level. Again, I estimate (6) for both the entire sample and the subsamples of renters and young renters. To test hypothesis 4, I check whether there is an increasing trend in the effect of the intention to buy a house on the likelihood of saving by estimating (7). To be in line with hypothesis 4, there should be an increasing trend in 𝛽2 and 𝛽3. The yearly effects of 2009-2010 should not be jointly significant, while the yearly effects of 2011-2016 should be larger and jointly significantly different from zero. Again, I use the results for specification (7) to test whether I am correct in assuming a ‘jump’ in 2013 in the relationship between the probability that a household is saving and the intention to buy a house in the short run or long run.
𝑃{𝑆𝑖𝑡 = 1|𝐼𝑆𝑅𝑖𝑡, 𝐼𝐿𝑅𝑖𝑡, 𝐷13𝑡, 𝑥𝑖𝑡, 𝑧𝑖𝑡} = ϕ(𝛽1+ 𝛽2𝐼𝑆𝑅𝑖𝑡+ 𝛽3𝐼𝐿𝑅𝑖𝑡+ 𝛽4𝐼𝑆𝑅𝑖𝑡𝐷13𝑡+ 𝛽5𝐼𝐿𝑅𝑖𝑡𝐷13𝑡+
𝑥𝑖𝑡′𝛾 + 𝑧𝑖′𝛾 + 𝛾𝑡) (6)
𝑃{𝑆𝑖𝑡 = 1|𝐼𝑆𝑅𝑖𝑡, 𝐼𝐿𝑅𝑖𝑡, 𝑥𝑖𝑡, 𝑧𝑖𝑡} = ϕ(𝛽1+ 𝛽2𝑡𝐼𝑆𝑅𝑖𝑡+ 𝛽3𝑡𝐼𝐿𝑅𝑖𝑡+ 𝑥𝑖𝑡′ 𝛾 + 𝑧𝑖′𝛾 + 𝛾𝑡) (7)
By estimating the effect of the intention to buy a house on both the amount of saving and the likelihood of saving, I will be able to examine how saving behaviour in general is influenced by the intention to buy a house, whether this differs after the most important limitations to mortgage credit have been imposed, and how this changes over the years as more and more mortgage reforms are implemented.
4.2 Robustness Check
16
households have put any money aside in the past 12 months. I then estimate the effect of the intending to buy a house in the short run and long run on the amount of money put aside using equation (8), where M represents the amount of money put aside in the past 12 months.
𝑀𝑖𝑡 = 𝛽1+ 𝛽2𝑡𝐼𝑆𝑅𝑖𝑡+ 𝛽3𝑡𝐼𝐿𝑅𝑖𝑡+ 𝛽4𝑡𝐼𝑆𝑅𝑖𝑡𝐷13𝑡+ 𝛽5𝑡𝐼𝐿𝑅𝑖𝑡𝐷13𝑡+ 𝑥𝑖𝑡′𝛾 + 𝑧𝑖′𝛾 + 𝛾𝑡+ 𝜖𝑖𝑡 (8)
I estimate (8) with the use of interval regression, which is a type of ordered probit model with known cut-off points. § Interval regression fits the linear model (8) to outcome values that are not observed directly but are known to fall within a certain interval with fixed endpoints. The parameters can then be interpreted in the same manner as OLS parameters. Again, I estimate (8) for both the entire sample and the subsamples of renters and young renters, and standard errors are corrected for clusters at the household level.
To check whether there is an increasing trend in the relationship between the intention to buy a house and the amount of money put aside, I also estimate (9), with yearly effects for the intention to buy a house in the short run and in the long run. There should be no trend from 2009-2010, as there were no new mortgage restrictions introduced in that period. An increasing trend should take place between 2011 and 2016, as the regulations for taking out a mortgage became more and more strict during that time period. Like I do with the results from the estimation of equations (5) and (7), I test whether there is a ‘jump’ in 2013 in the effect of intending to buy a house in the short run or long run on the amount of money put aside. If this holds, then 𝛽2 is not significantly different from 2008 in the years 2009-2012, and has the same values in 2013-2016. I also conduct this test for 𝛽3.
𝑀𝑖𝑡 = 𝛽1+ 𝛽2𝑡𝐼𝑆𝑅𝑖𝑡 + 𝛽3𝑡𝐼𝐿𝑅𝑖𝑡 + 𝑥𝑖𝑡′ 𝛾 + 𝑧𝑖′𝛾 + 𝛾𝑡+ 𝜖𝑖𝑡 (9) 5. Data
To conduct my analysis I make use of the DNB household survey (DHS). The survey is conducted by CentERdata, and has been issued yearly since 1993. The panel survey consists of six questionnaires, namely General Information on the Household, Household and Work, Accommodation and Mortgages, Health and Income, Assets and Liabilities, and Economic and Psychological Concepts. Due to significant
17
changes in the Accommodation and Mortgages questionnaire after 2007, I only use data from 2008-2016. Approximately 2,000 households participate in the survey each year, although response rates can differ per questionnaire. The surveys are filled out online. Not every household in the panel has access to a computer with an internet connection; these households are provided with a simple computer and access to the internet. All household members over the age of 16 within a participating household are interviewed. However, only data is retained from the household member that has filled out the Accommodation and Mortgages questionnaire. This is done for two important reasons. First, the Accommodation and Mortgages questionnaire is always filled out by only one household member. Discarding data from household members that have not filled out this questionnaire ensures that in every wave all households only enter the dataset once. Second, the most important explanatory variable for my analysis, the intention to buy a house, derives from the Accommodation and Mortgages questionnaire. There is no written rule on which household member fills out the Accommodation and Mortgages questionnaire, but the vast majority of responses comes from household heads (ranging from approximately 85% to 99% of responses). In other cases, the spouse or permanent partner usually fills out this questionnaire. In very few cases (less than 1% of all responses), the questionnaire is filled out by a parent, child, housemate or family member living in the house. As households within the panel are not entirely representative of the Dutch population in terms of homeownership and income, a set of weights is used to transform the panel into a more representative group of the Dutch population.
Not only does the survey contain information on many characteristics of the respondent, the survey also grants extensive information on different kinds of saving behaviour. If mortgage reforms affect households’ savings, then it should affect the savings of the households intending to buy a house. Therefore, I test whether the saving behaviour of this group has changed relative to households that do not intend to buy a house. I check whether households have any plans to buy a house in the future by using the question derived from the Accommodation and Mortgages questionnaire below.
Do you intend to buy a (another) house eventually? 1. No, I prefer to rent accommodation. 2. No, I cannot afford to buy accommodation.
3. Yes, preferably in the short-term (within two years). 4. Yes, in the long-term (more than two years from now). 5. I don’t have any intention to move.
18
Table 1: Distribution of responses to the question of whether the household intends to buy a (another) house eventually
Weighted Statistics
The variables ISR and ILR are constructed from the response to this question. If the respondent gave the third answer, the household is assumed to have an intention to buy a house in the short run. In that case, ISR will take a value of 1. If the respondent gave the fourth answer, the household is assumed to intend to buy a house in the long run, indicated with a value of 1 for the variable ILR. The distribution of responses to the question above can be found in Table 1. The percentage of people intending to buy a house does significantly vary over time. In 2008, the amount of people intending to buy a house is 16.3%. This drops to 12.36% in 2011, after which there is an increasing trend until 2016. The most notable change is from 2015 to 2016, when the percentage almost doubled from 13.51% to 24.81%. The intention to buy a house is related to the age class of the respondent. Respondents between the age of 25 and 35 are most likely to have an intention to buy a house. This large increase in the percentage of households intending to buy a house is therefore most likely caused by a change in the proportion of respondents aged 25-35. In 2016, this group is almost doubled. Nonetheless, households may have been impacted by the financial crisis, which could have left them unable to buy a house or made them lose interest in the purchase of a house. However, this could also be the result of the stricter mortgage rules, which according to Jackman and Sutton (1982) and Artle and Varaiya (1978) could lead to lower homeownership rates or a delay in homeownership.
Other control variables can also be constructed from responses to the survey and include age, sex, marital status, number of children, education level, net disposable income, financial literacy and a dummy for the self-employed. Financial literacy is proxied by asking respondents how knowledgeable they consider themselves to be with respect to financial matters. As this is an individual’s own estimate of financial literacy, this is far from a perfect measure of financial literacy. However, it does give an indication of how confident the individual is with handling financial matters. If they believe themselves to be very
2008 2009 2010 2011 2012 2013 2014 2015 2016
No, I prefer to rent 208 230 261 257 274 246 219 275 212
15.86% 19.60% 18.17% 18.90% 18.83% 16.47% 15.09% 16.31% 12.11%
No, I cannot afford to buy a house 162 152 200 184 182 207 196 222 197
12.36% 12.97% 13.92% 13.49% 12.54% 13.86% 13.51% 13.17% 11.26%
Yes, in the short-term 77 68 87 57 70 74 95 96 157
5.89% 5.80% 6.03% 4.18% 4.78% 4.95% 6.55% 5.68% 8.99%
Yes, in the long-term 136 112 124 111 115 150 148 132 277
10.38% 9.51% 8.65% 8.18% 7.94% 10.05% 10.23% 7.83% 15.82%
I have no intention to buy a house 648 571 697 712 762 758 730 888 815
49.36% 48.65% 48.54% 52.27% 52.35% 50.72% 50.34% 52.70% 46.57%
Other/Don't know 81 40 67 41 51 59 62 72 91
19
knowledgeable, they are likely to make more decisions regarding their financial affairs. In fact, Allgood and Walstad (2016) suggest that both actual and perceived financial literacy influence financial behaviour and that perceived financial literacy may be as important as actual financial literacy.
Detailed information on saving behaviour can be found in the questionnaire on assets and liabilities. The questionnaire introduces approximately forty asset and debt categories, a detailed description of which can be found in Alessie et al. (2002). Respondents are first asked whether they own any of the saving categories, and if so, what amount they have saved with each of them. If households indicate that they are owners, but do not give an amount, I follow the same methodology as Alessie et al. (2002) and impute the amounts, based on adjacent previous or future years and regression models. These assets and debt categories can then be summed to find various types of wealth. For this analysis I make use of total financial assets, which consists of checking, savings, and deposit accounts, employer-sponsored savings plans, deposit books, saving certificates, insurance policies, and different kinds of funds, stocks, bonds, and options. An important benefit of using financial assets is that it excludes housing wealth which homeowners may use for the purchase of a new house.
Table 2 displays the descriptive statistics of the dependent, explanatory, and control variables for the entire sample and for the subsamples of renters and renters below the age of 46. All variables are dummy variables, apart from the number of children, net income, and financial assets. Net income and financial assets are expressed in real terms. It is important to note that the distribution of financial assets is strongly non-normal. This is evident from the fact that the mean of financial assets is closer to the 75th percentile than the median. Therefore, I use the logarithm of financial assets in my estimations, in order to transform the distribution into a more normal one.
To estimate whether a household is more likely to save in response to the new mortgage regulations, I use responses from households on expenditure questions. Respondents are asked whether income was lower than, equal to, or higher than expenses within the last twelve months. If income exceeded expenses, the household has saved, and by contrast, should expenses have exceeded income, the household has dissaved. As visible in Table 2, 36.9% of all respondents indicated that income exceeded expenditures in the past 12 months.
20 Table 2: Descriptive Statistics
Weighted Statistics
have, they are asked to indicate how much they approximately put aside. The answer options consist of the following ranges:
- less than €1,500, - €1,500 to €5,000, Variable
Mean N Mean N Mean N
Female 0.314 13,135 0.446 3,616 0.611 1,036 Age class Age <=25 0.007 13,135 0.020 3,616 0.070 1,036 25<age<=35 0.107 13,135 0.127 3,616 0.444 1,036 35<age<=45 0.165 13,135 0.139 3,616 0.486 1,036 45<age<=55 0.183 13,135 0.166 3,616 0.000 1,036 55<age<=65 0.239 13,135 0.229 3,616 0.000 1,036 65<age<=75 0.203 13,135 0.200 3,616 0.000 1,036 Age>75 0.097 13,135 0.118 3,616 0.000 1,036 Married 0.606 11,556 0.375 3,134 0.228 870 Number of Children 0.561 13,135 0.300 3,616 0.625 1,036 Education
Lower intermediate and primary education 0.268 13,131 0.349 3,612 0.115 1,035 Intermediate vocational education 0.103 13,131 0.129 3,612 0.099 1,035 Secondary pre-university education 0.185 13,131 0.192 3,612 0.309 1,035 Higher vocational education 0.289 13,131 0.205 3,612 0.268 1,035
University education 0.155 13,131 0.125 3,612 0.210 1,035
Financial Literacy
Not knowledgeable 0.142 12,038 0.201 3,290 0.193 883
More or less knowledgeable 0.559 12,038 0.593 3,290 0.531 883
Knowledgeable 0.259 12,038 0.187 3,290 0.238 883
Very knowledgeable 0.039 12,038 0.019 3,290 0.039 883
Selfemployed 0.037 11,556 0.036 3,134 0.043 870
Intention to buy a (new) house in the short run 0.056 13,119 0.070 3,616 0.209 1,036 Intention to buy a (new) house in the long run 0.107 13,119 0.068 3,616 0.195 1,036 Income exceeds expenditures 0.369 12,067 0.255 3,307 0.257 886
Net income (x €1,000) 32.969 13,134 23.900 3,615 24.142 1,035
Financial assets (x €1,000) 53.677 11,761 27.611 3,247 14.137 875 Distribution of financial assets (x €1,000)
25th percentile 4.645 1.300 0.800
50th percentile 18.887 6.657 3.764
75th percentile 53.276 23.535 14.350
99th percentile 515.942 252.000 181.590
Whole sample Renters Young renters
21 Table 3: The amount of money put aside in the past 12 months
Weighted statistics - €5,000 to €12,500, - €12,500 to €20,000, - €20,000 to €37,500, - €37,500 to €75,000, - More than €75,000.
Respondents could also indicate that they do not know, in which case their response is not used. Additionally, respondents who indicated that they did not put any money aside but did respond that expenditures were about equal to income over the past 12 months are given a value of zero, indicating the household neither saved nor dissaved. The distribution of these responses is given in Table 3. The majority of households either did not put any money aside or put €1,500 to €5,000 aside. A very small group put more than €12,500 aside. Respondents who indicated that they had not put any money aside and replied that expenditures were larger than income over the past 12 months are given a value of -1, indicating that the household dissaved. This is done to create a negative endpoint of −∞.
Responses to various questions present some prima facie evidence on the effects of the introduction of stricter mortgage regulations. Households that indicated in the survey on Accommodation and Mortgages that they had an intention to buy a (new) house were asked the follow-up question of whether they save money consciously for the future purchase or furnish of a house. Respondents were given the option of selecting multiple answers, which means that in total 25 answer combinations are possible. As not all
22 Table 4: Saving behaviour of households intending to buy a house
Weighted statistics
Data from 2009 is missing, as the question was significantly different in that year. 1 = Yes, save whatever we can afford. 2 = Yes, save a fixed amount per year. 4 = No, there is no possibility. 8 = No, by that time we will see how things are standing. 16 = No, we will pay all expenses by taking out loans. 32 = Otherwise.
options can be discussed here, and many are not too interesting, only a few salient responses are shown in Table 4. For example, the percentage of households that either save whatever they can afford or save a fixed amount per year has increased greatly in recent years. In 2008, this was 38.68%, whereas this is 46.87% in 2016, an increase of approximately 7 percentage points. The percentage of households that plan on paying all expenses related to the house by taking out loans has also undergone a significant change. The percentage of people who have solely given this response in their list of answers was 4.72% in 2008, but was a mere 1.88% in 2016. This result was first evidenced in 2011, when the amount was more than halved since the year before. The changes in responses given to this question are a first indication that savings are likely to be higher in more recent years for households intending to buy a house.
Furthermore, private transfers may also affect the amount of savings for the purchase of a house (see e.g. Engelhardt and Mayer, 1998; Guiso and Jappelli, 2002), so it is interesting to see whether this changes over time. In row 1 from Table 5 the amount of people responding that they have received a gift to finance their current accommodation is given, grouped by the year in which they replied. There is merely a small percentage of homeowners that have received a gift to finance their current accommodation. Nonetheless, there does appear to be an increasing trend from 2012 onwards, which could be the result of the stricter mortgage regulations. However, these responses are grouped by the year in which the responses are given, not by the year in which the house is bought. This implies that people responding in
2008 2010 2011 2012 2013 2014 2015 2016 1 60 56 54 42 77 90 99 145 27.54% 25.99% 31.62% 21.99% 33.47% 36.03% 42.72% 32.62% 2 24 39 28 30 20 31 17 63 11.14% 17.88% 16.51% 15.74% 8.82% 12.59% 7.24% 14.25% 4 26 34 22 40 42 42 34 61 11.71% 15.63% 13.06% 21.08% 18.39% 16.97% 14.48% 13.82% 8 66 46 42 45 48 47 53 110 30.11% 21.42% 24.43% 23.81% 21.05% 18.85% 22.87% 24.87% 16 10 10 4 5 8 5 7 8 4.72% 4.80% 2.20% 2.38% 3.55% 1.97% 3.05% 1.88% 32 10 13 9 9 13 10 5 16 4.67% 5.91% 5.47% 4.90% 5.76% 4.18% 2.06% 3.51%
More than one applies 17 13 9 14 16 18 13 31
23
Table 5: households that received a gift to finance the current accommodation
Weighted statistics
Row 1 displays the amount of households that have received a gift to finance their current accommodation sorted by the year in which they replied, row 2 displays the amount of households that have received a gift to finance their current accommodation sorted by the year in which their house was bought.
2015 could have received a gift for their accommodation in the year 1990 for example, which cannot possibly be the result of stricter mortgage regulations. Therefore, in row 2 the responses are grouped by the year in which the house is bought. Thus, as is visible in row 2 from Table 5, in 2008 there were 13 respondents that have bought a house and received a gift to help finance the purchase. Between 2009 and 2010 this percentage more than doubled, as well as between 2013 and 2014. This increase is most likely caused by the increase in the tax exempt amount that parents are allowed to gift their children, as explained in section 2. This increase was temporary, and ended in 2015. After 2014, the percentage of households that received a gift decreased again, from 27.21% to 8.92%. In 2016 this percentage increases to over 30%. However, it must be noted that the total amount of people that have bought a house in one of these years is much lower than in previous years. Even though the percentage of people receiving a gift is over 30% in 2016, this consisted of merely 4 respondents. Thus, these changes should be interpreted with some caution.
The changes in the distribution of the responses to the questions above represent some prima facie evidence on the effects of the stricter mortgage regulations. The next section will discuss the results of the estimations of equations (4) through (9), to formally test whether saving behaviour has been affected by the reforms of mortgage regulations.
6.1 Results
Table 6 displays the estimation results of (4) for the entire sample and the subsamples of renters and young renters. A first glance shows that all control variables have the expected sign and size, and most are significant in explaining the natural logarithm of the level of financial assets. The variables for age class are jointly significant, as well as the variables for education and financial literacy. The intention to buy a house both in the short run and in the long run indeed has a positive and significant effect on financial assets, as expected. More explicitly, the intention to buy a house in the short run increases financial assets
Year 2008 2009 2010 2011 2012 2013 2014 2015 2016
(1) 24 20 34 25 27 34 35 43 66
5.00% 5.14% 6.24% 4.77% 4.77% 5.44% 5.60% 5.77% 7.93%
(2) 13 9 24 19 12 9 21 6 4
24
Table 6: Estimation results for the logarithm of financial assets and the intention to buy a house
The table reports the estimation results of the quantile regression of equation (4) for the entire sample and the subsamples of renters and young renters (below the age of 46). The dependent variable is the natural logarithm of financial wealth. Standard errors are corrected for household clusters. ***, ** and * denote significance at the 1%, 5% and 10% level, respectively.
Subsample
Coef. S.E. Coef. S.E. Coef. S.E.
Constant 1.738 *** 0.635 2.064 1.665 2.962 2.732
Intention to buy in the short run 0.219 * 0.117 0.554 ** 0.271 0.332 0.299
Intention to buy in the long run 0.373 *** 0.089 0.758 ** 0.334 0.954 ** 0.485
Intention to buy in the short run from 2013 onwards -0.044 0.151 0.208 0.308 0.689 ** 0.323
Intention to buy in the long run from 2013 onwards -0.150 0.121 -0.099 0.345 -0.028 0.490
Age class 26-35 0.261 0.281 0.082 0.421 -0.222 0.521 36-45 0.606 ** 0.290 0.098 0.452 -0.107 0.570 46-55 1.109 *** 0.294 1.054 ** 0.467 56-65 1.306 *** 0.296 0.819 * 0.483 66-75 1.387 *** 0.302 1.050 ** 0.474 75+ 1.448 *** 0.307 1.506 *** 0.489 Female -0.306 *** 0.090 -0.513 *** 0.190 -0.852 *** 0.189 Married 0.139 0.085 0.172 0.185 -0.036 0.222 Number of Children -0.119 *** 0.042 -0.298 *** 0.098 -0.147 0.118 Selfemployed 0.194 0.156 -0.066 0.346 -1.030 1.145 Education
Intermediate vocational education 0.299 ** 0.133 0.251 0.326 0.979 ** 0.407
Secondary pre-university education 0.147 0.099 0.282 0.241 1.158 *** 0.356
Higher vocational education 0.452 *** 0.095 0.394 * 0.219 1.246 *** 0.387
University education 0.896 *** 0.114 1.194 *** 0.301 1.770 *** 0.565
Financial Literacy
More or less knowledgeable 0.305 *** 0.090 0.433 *** 0.161 0.523 ** 0.215
Knowledgeable 0.518 *** 0.103 0.603 *** 0.215 0.746 *** 0.280
Very knowledgeable 0.772 *** 0.152 0.668 0.446 0.792 0.718
Natural logarithm of Net Income 0.637 *** 0.057 0.581 *** 0.167 0.455 0.284
Year Fixed Effects
2009 0.087 0.057 0.141 0.137 0.182 0.368 2010 0.117 * 0.063 0.119 0.174 0.229 0.298 2011 0.012 0.068 0.007 0.151 0.224 0.348 2012 -0.020 0.066 0.071 0.156 0.186 0.310 2013 -0.073 0.073 -0.118 0.159 -0.299 0.316 2014 -0.099 0.071 -0.187 0.160 -0.385 0.326 2015 -0.041 0.071 -0.106 0.165 -0.412 0.293 2016 -0.058 0.074 -0.305 * 0.167 -0.498 0.318 N 9908 2578 655 R² 0.186 0.161 0.253
P-value test 2013 interaction variables = 0 0.462 0.738 0.073
P-value test age class = 0 0.000 0.000 0.792
P-value test education = 0 0.000 0.003 0.017
P-value test financial literacy = 0 0.000 0.024 0.044
Young Renters
(3) (2)
(1)
25
with 24.5%**, while intending to buy a house in the long run increases financial assets with 45.2% for the entire sample, whereas these numbers are 74% and 113.4% for renters. As homeowners have the possibility of using their housing wealth to finance their new accommodation, it is not surprising that the effect of intending to buy a house on financial assets is larger for the subsample of renters.
However, the parameters of interest are 𝛽4 and 𝛽5, indicating the change in this relationship from 2013 onwards. For the entire sample and the subsample of renters, this change is not significant. In fact, some estimations are negative. As I expected, young renting households do significantly save more when they intend to buy a house in 2013 and after. Young households renting accommodation save approximately 99.2% more when they intend to buy a house in the short run from 2013 onwards, which is a substantial amount. Young households intending to buy a house in the long run do not appear to save more as a result of the stricter mortgage regulations. Nonetheless, these results support hypothesis 1, as they suggest that household savings have increased for young renting households intending to buy a house.
Because the regulations for obtaining a mortgage have become stricter throughout the years, I expect a positive trend in the effect of having an intention to buy a house on financial assets. I therefore estimate (5), with yearly interaction variables replacing the dummy interaction variables indicating that households intend to buy a house from 2013 onwards. These results can be found in Table 7, and are similar to the results in Table 6. When looking at the results for the entire sample, there is no clear trend in the effect of intending to buy a house in the short run on financial assets. Nevertheless, the relationship is largest and significantly different from 2008 in 2012 and 2013. As this is around the time that many new restrictions for taking out a mortgage were introduced, this suggests that households were affected by the new mortgage regulations. However, looking at the short run intention to buy a house, the yearly effects are not jointly significant between 2011 and 2016. In contrast, the yearly effects for the long run intention to buy a house are jointly significant between 2011 and 2016. There is a positive trend in these effects between 2013 and 2015, around the time that many new mortgage restrictions were introduced. Moreover, these yearly effects are not significant from 2009-2010, which is as expected. The latter results provide evidence for a trend from 2011-2016. Although there is a positive trend in the relationship between financial assets and the intention to buy a house in the short run from 2012-2015 for young renters, both the short run and long run yearly interaction variables are not jointly significant, nor are they jointly significant for the subsample of renters. Therefore, these results provide mixed evidence for hypothesis 3. Although a positive trend is visible for young renters, it is not significant. Nonetheless, for all households there is a positive trend in the yearly effects of intending to buy a house in the long run.
26
Table 7: Estimation results for the logarithm of financial assets and the intention to buy a house with yearly effects
The table reports the estimation results of the quantile regression of equation (5) for the entire sample and the subsamples of renters and young renters (below the age of 46), with yearly interaction variables that replace the dummy variables indicating the years 2013-2016. The dependent variable is the natural logarithm of financial wealth. Standard errors are corrected for household clusters. ***, ** and * denote significance at the 1%, 5% and 10% level, respectively.
Subsample
Coef. S.E. Coef. S.E. Coef. S.E.
Constant 1.795 *** 0.658 2.029 1.696 2.563 2.607
Intention to buy in the short run -0.086 0.161 0.540 0.495 0.520 1.644
Intention to buy in the long run 0.063 0.183 0.383 0.582 0.567 1.606
Age class 26-35 0.270 0.282 0.133 0.392 -0.203 0.545 36-45 0.586 ** 0.292 0.188 0.428 -0.200 0.576 46-55 1.097 *** 0.294 1.111 ** 0.451 56-65 1.282 *** 0.298 0.849 * 0.464 66-75 1.380 *** 0.303 1.088 ** 0.455 75+ 1.433 *** 0.310 1.540 *** 0.473 Female -0.305 *** 0.089 -0.505 *** 0.187 -0.838 *** 0.201 Married 0.143 * 0.084 0.200 0.191 -0.054 0.286 Number of Children -0.125 *** 0.042 -0.335 *** 0.106 -0.212 0.143 Selfemployed 0.181 0.182 0.006 0.387 -0.826 0.749 Education
Intermediate vocational education 0.303 ** 0.136 0.291 0.324 0.709 0.483 Secondary pre-university education 0.162 * 0.099 0.287 0.243 1.089 *** 0.345 Higher vocational education 0.451 *** 0.098 0.398 * 0.228 1.193 *** 0.376
University education 0.894 *** 0.113 1.139 *** 0.342 1.624 *** 0.544
Financial Literacy
More or less knowledgeable 0.305 *** 0.091 0.445 *** 0.173 0.432 * 0.246
Knowledgeable 0.532 *** 0.105 0.635 *** 0.236 0.603 * 0.315
Very knowledgeable 0.759 *** 0.154 0.646 0.543 0.376 0.595
Natural logarithm of Net Income 0.640 *** 0.058 0.583 *** 0.178 0.521 * 0.278 Year Fixed Effects
2009 0.009 0.067 0.096 0.153 0.314 0.486 2010 0.040 0.075 0.066 0.188 0.009 0.358 2011 -0.102 0.081 -0.051 0.160 0.261 0.481 2012 -0.100 0.076 -0.016 0.165 0.267 0.366 2013 -0.160 * 0.083 -0.126 0.169 -0.106 0.370 2014 -0.171 * 0.080 -0.278 0.173 -0.223 0.397 2015 -0.118 0.081 -0.192 0.184 -0.570 * 0.344 2016 -0.132 0.084 -0.358 ** 0.182 -0.636 0.458
Cross effects intention to buy in the short run
Intention to buy in the short run 2009 0.323 0.245 -0.493 0.620 -0.556 1.598 Intention to buy in the short run 2010 0.459 0.282 0.268 0.657 0.336 1.696 Intention to buy in the short run 2011 0.395 0.313 0.082 0.628 -0.385 1.709 Intention to buy in the short run 2012 0.457 * 0.242 0.273 0.821 -0.492 1.669 Intention to buy in the short run 2013 0.555 ** 0.223 0.170 0.590 0.031 1.701 Intention to buy in the short run 2014 0.229 0.249 0.354 0.520 0.268 1.632 Intention to buy in the short run 2015 0.155 0.201 0.276 0.557 0.812 1.636 Intention to buy in the short run 2016 0.173 0.205 -0.036 0.541 0.648 1.651 Cross effects intention to buy in the long run
Intention to buy in the long run 2009 0.330 0.257 0.085 0.587 -0.633 1.700 Intention to buy in the long run 2010 0.291 0.221 0.154 0.717 1.059 1.798 Intention to buy in the long run 2011 0.630 *** 0.222 -0.449 0.753 -0.619 1.803 Intention to buy in the long run 2012 0.201 0.250 0.522 0.715 0.071 1.655 Intention to buy in the long run 2013 -0.039 0.245 -0.242 0.690 -0.094 1.632 Intention to buy in the long run 2014 0.127 0.257 0.340 0.659 0.115 1.668 Intention to buy in the long run 2015 0.345 0.229 0.630 0.606 0.853 1.569 Intention to buy in the long run 2016 0.184 0.214 0.304 0.646 0.588 1.643
N 9908 2578 655
R² 0.186 0.163 0.264
P-value test short run cross effects 2009-2010=0 0.186 0.463 0.397 P-value test short run cross effects 2011-2016=0 0.226 0.932 0.335 P-value test long run cross effects 2009-2010=0 0.320 0.977 0.104 P-value test long run cross effects 2011-2016=0 0.035 0.353 0.621 P-value test jump in intention to buy in the short run from 2013 onwards 0.260 0.848 0.664 P-value test jump in intention to buy in the long run from 2013 onwards 0.074 0.452 0.177
(1) (2)
Renters
27
Table 8: Estimation results for the probability that a household is saving and the intention to buy a house
The table reports the probit estimates of equation (6) for the entire sample and the subsamples of renters and young renters (below the age of 46). The dependent variable is a dummy variable that takes a value of 1 if expenditures were lower than income over the past 12 months, and 0 if not. Standard errors are corrected for household clusters. ***, ** and * denote significance at the 1%, 5%, and 10% level, respectively.
Subsample
Coef. S.E. Coef. S.E. Coef. S.E.
Constant -4.765 *** 0.442 -4.332 *** 0.767 -5.132 *** 1.098
Intention to buy in the short run 0.038 0.101 0.212 0.220 0.344 0.285
Intention to buy in the long run 0.083 0.075 0.372 * 0.200 0.151 0.266
Intention to buy in the short run from 2013 onwards 0.235 * 0.124 0.441 * 0.235 0.254 0.307
Intention to buy in the long run from 2013 onwards -0.034 0.091 0.220 0.220 0.498 * 0.297
age class 26-35 0.254 0.286 -0.004 0.302 -0.044 0.297 36-45 0.269 0.292 -0.038 0.339 -0.015 0.341 46-55 0.203 0.293 0.179 0.343 56-65 0.240 0.293 0.058 0.347 66-75 0.298 0.293 0.286 0.349 75+ 0.130 0.297 0.010 0.358 Female -0.142 *** 0.052 -0.201 ** 0.103 -0.360 ** 0.142 Married 0.020 0.052 -0.011 0.106 0.036 0.191 Number of Children -0.187 *** 0.028 -0.210 *** 0.073 -0.273 *** 0.093 Selfemployed -0.226 ** 0.106 0.121 0.221 -0.570 0.376 Education
Intermediate vocational education 0.152 * 0.085 -0.037 0.153 -0.042 0.365
Secondary pre-university education 0.134 ** 0.068 0.200 0.125 0.333 0.264
Higher vocational education 0.224 *** 0.062 0.249 * 0.133 0.246 0.281
University education 0.252 *** 0.072 0.151 0.151 0.415 0.275
Financial Literacy
More or less knowledgeable 0.208 *** 0.057 0.112 0.099 0.024 0.184
Knowledgeable 0.354 *** 0.066 0.324 *** 0.123 0.294 0.215
Very knowledgeable 0.489 *** 0.102 0.127 0.232 0.318 0.304
Natural logarithm of Net Income 0.383 *** 0.034 0.347 *** 0.073 0.432 *** 0.111
Year Fixed Effects
2009 0.015 0.048 -0.009 0.096 -0.079 0.210 2010 0.037 0.047 0.061 0.097 0.104 0.238 2011 -0.007 0.049 0.075 0.099 0.088 0.228 2012 0.022 0.049 0.052 0.096 0.068 0.210 2013 0.007 0.051 -0.079 0.100 -0.104 0.230 2014 -0.025 0.050 -0.144 0.104 -0.195 0.230 2015 0.007 0.050 -0.029 0.098 0.120 0.225 2016 0.021 0.052 -0.179 * 0.103 -0.021 0.237 N 10906 2936 775 Pseudo R² 0.056 0.068 0.136 Wald Chi² 334.420 129.500 100.300
P-value test 2013 interaction variables = 0 0.144 0.132 0.243
P-value test age class = 0 0.313 0.208 0.987
P-value test education = 0 0.002 0.233 0.398
P-value test financial literacy = 0 0.000 0.044 0.227
(1) (2) (3)
28
Table 9: Probit marginal effects for the intention to buy a house on the probability that a household is saving
The table reports the probit marginal effects of the intention to buy a house both in the short and long run before and from 2013 onwards on the likelihood of saving for the entire sample and the subsamples of renters and young renters (below the age of 46). The dependent variable is a dummy variable that takes value 1 if expenditures were lower than income over the past 12 months, and 0 if not. Standard errors are corrected for household clusters. ***, ** and * denote significance at the 1%, 5%, and 10% level, respectively.
Finally, I test whether I am correct in assuming a ‘jump’ in the relationship between financial assets and the intention to buy a house from 2013 onwards, as I do in specification (4). In the subsamples of renters and young renters, I fail to reject the hypothesis that there is a ‘jump’ from 2013 onwards in the financial assets of households intending to buy a house. I can therefore base my conclusions on the estimation results of (4). For the entire sample, I reject the assumption that there is a jump in the relationship between financial asset and the intention to buy a house in the long run.
The results for the estimation of equation (6) can be found in Table 8. Again, all control variables have the expected sign, although the age class does not significantly influence the likelihood of a household’s income exceeding expenditures. The marginal effects for intending to buy a house before and from 2013 onwards are displayed in Table 9. Again, the parameters of interest are 𝛽4 and 𝛽5, which denote the
change in the relationship between the intention to buy a house and the likelihood to save. Households that intend to buy a house in the short run are 8.4 percentage points more likely to save after the implementation of stricter mortgage regulations. There is no significant change in the relationship between the intention to buy a house in the long run and financial assets. Furthermore, households renting accommodation and intending to buy a house in the short run are 13.3 percentage points more likely to save from 2013 onwards. Renting households also are not more likely to save if they intend to buy a house in the long run from 2013 onwards. Young renters with the intention to buy a house in the short run are not significantly more likely to save from 2013 onwards. Although this may seem strange, it might be the case that these households were already highly likely to save before 2013. However, young renters intending to buy a house in the long run are 14 percentage points more likely to save from 2013 onwards. These results are in support of hypothesis 2 as households that wish to buy a house in the (near) future are more likely to save after the stricter mortgage regulations have been introduced.
Subsample
Coef. S.E. Coef. S.E. Coef. S.E.
Intention to buy in the short run 0.014 0.036 0.068 0.074 0.104 0.090
Intention to buy in the short run from 2013 onwards 0.084 * 0.044 0.133 * 0.071 0.071 0.086
Intention to buy in the long run 0.030 0.027 0.122 * 0.070 0.044 0.078
Intention to buy in the long run from 2013 onwards -0.012 0.032 0.066 0.066 0.140 * 0.083
(3)
Renters Young Renters
