THE EFFECT OF HOUSE PURCHASES ON SPENDING BEHAVIOR IN THE NETHERLANDS
Leiden University, Governance and Global Affairs
Master Public Administration: Economics & Governance
Luuk de Haas S1372483
Supervisor: Dr. Max van Lent
Table of contents 1. Introduction 4 2. Relevance 6 2.1 Academic relevance 6 2.2 Policy relevance 6 3. Theoretical framework 7 3.1 Consumption 7
3.1.1 Life cycle theory 7
3.1.2 Wealth effect and collateral effect 8
3.2 Saving 10
3.2.1 Homeownership and saving 11
3.2.2 House price appreciation and saving 11
3.2.3 The effect of house price depreciation 12
3.3 Determining the value of a house 12
3.3.1 WOZ-value 13
4. Expectations and conceptual framework 15
4.1 Expectations 15
4.2 Conceptual framework 17
5. Research methodology 19
5.1 Data 19
5.1.1 Dutch National Bank Household Survey 19
5.1.2 Description of variables 20 5.2 Descriptive statistics 22 5.3 Empirical model 23 6. Results 29 7. Analysis 34 7.1 Analysis 34
7.2 Feedback on conceptual framework 37
8. Conclusion 38
Limitations and suggestions for further research 39
1. Introduction
Prices in general are mainly determined by the behavior of consumers (Salzman & Zwinkels, 2013). From time to time, this leads to what is called a pricing bubble. One of the first examples of this phenomenon can be found in the 17th century when tulips were brought from Turkey to the Netherlands. Over the course of two years, the prices of tulips rose to an astonishing high. Up to a point where traders were buying options on the harvest of tulips the next season, what would today be called futures. In 1637 one could exchange a tulip bulb for a canal house in Amsterdam. (Sooke, 2016) Until in February 1637 people could no longer afford the staggering price increase of the tulip. The demand disappeared, the tulip market crashed and a financial recession ensued. A similar event can be observed 370 years later. From 2000 until 2007, the United States Federal Reserve Bank maintained variable interest mortgage rates at a low level and the annual growth of household consumption in the United States was 3 percent on average. (Ministerie van Financiën, 2018; The World Bank, 2019) When the interest rate increased from 1 to 5 percent, an increasing number of homeowners were unable to pay their monthly mortgage costs and were forced to sell their houses. The supply of houses on the market skyrocketed. This led to a decrease in the value of collateral of many home owners, who were not able to pay off their mortgages. One bank after the other ended up in a state of financial distress and consequences carried over internationally, leading to a financial crisis. (Reuters, 2008; NOS, 2010; Van Ooijen & Van Rooij, 2016)
The focus of this thesis is to indicate if an increase of the value of houses has an effect on the spending behavior of individuals. The larger part of Dutch houses is purchased with the financial support of a mortgage, for which the property is used as collateral (Spiegelaar & Lennartz, 2018). If housing values experience an increase, this means that the collateral of mortgage-backed properties increases. The appreciation of a house leads to an increase of property tax, as will be explained in the theoretical framework of this paper. Additionally, it has been shown in various bodies of research that housing value appreciation has an effect on the level of household consumption and on saving behavior. However, little research has been conducted to explore this effect for Dutch households. This is elaborated on in the following paragraph. To determine if the degree or direction of this effect varies at different points in time, the central research question to this thesis is formulated accordingly: Does household spending behavior change around the purchase of a house?
Panel data from the Dutch National Bank Household Survey from 2004-2018 is used. The data set provides 10.498 unique observations of 3.597 individuals over the time span of 15 years (CentERdata, 2018). On average, this results in 2,92 years of data per individual. This study concentrates on the
spending behavior of households in terms of consumption and saving of one year before the purchase of a house through one year after the purchase.
The structure of this thesis is as follows: the second paragraph explains the academic relevance and the scientific relevance of this research. The third paragraph consists of the theoretical framework, which is based on an exploration of relevant literature. The expectations are formulated in the fourth paragraph. Additionally, a conceptual framework is presented. This is followed by a description of the data set and a comprehensive explanation of the empirical model. The results are presented in paragraph six. In the analysis, the findings of this study are related to earlier research which is described in the theoretical framework. After which the conceptual framework is reviewed based on the findings from the analysis. Finally, concluding remarks are presented in paragraph eight.
2. Relevance
2.1 Academic relevance
Earlier research has been conducted regarding the link between house prices and consumption behavior. The effect of housing wealth on consumption has been demonstrated based on macro-level research (Case, Quigley and Shiller, 2005; Case et al., 2013; Ahec Šonje, Čeh Časni & Vizek, 2014). This type of research revolves around research problems based on data on national level. Studies based on micro-level research, which is concentrated on individual-level data, have determined a similar effect of housing wealth on consumption (Campbell & Cocco, 2007; Spiegelaar & Lennartz, 2018). In addition, various studies suggest that an increase of housing value reduces the level of savings (Yoshikawa & Ohtake, 1989; Engelhardt, 1994; Case et al., 2005, Brounen, Koedijk & Pownall, 2017). The effects of housing wealth increase on consumption have been studied for different generations (Campbell & Cocco, 2007). As well as the effect on savings of younger and older generations (Krumm & Miller, 1989; Brounen, Koedijk & Pownall, 2017). Ahec Šonje et al. (2014) find no conclusive evidence for this effect in The Netherlands, contrary to the findings of Spiegelaar & Lennartz (2018). One recent study of the United States Bureau of Economic Research studies the effect of consumption around the time that individuals purchase a house (Benmelech, Guren & Melzer, 2017). This study is concentrated on the exploration of this effect in the United States. Since no previous research has been conducted on whether and to what extent this effect is present in The Netherlands, this thesis fills this gap of knowledge.
2.2 Policy relevance
Understanding the behavior of private consumption decisions is important for explaining and forecasting the fluctuations of economic activities on national level, and private consumption on an individual level (Ahec Šonje et al., 2014). Information of consumption levels and saving behavior around the purchase of a house is vital to organize financial institutions. For one, this provides policy makers with insight into household spending behavior at different points in time and across different generations. Policy makers can identify fluctuations of purchasing power around the time that individuals purchase a house. Additionally, it gives insight on the volatility of the purchasing power as a result of fluctuations of housing wealth. Furthermore, for policy makers on supranational level at the European Central Bank, a nuanced understanding of purchasing power on national level provides relevant information for the determination of interest rates of the euro.
3. Theoretical framework 3.1 Consumption
3.1.1 Life cycle theory
The basis of consumer behavior theory can be found in the life-cycle theory. This brings forward the idea that any form of wealth increase will not cause individuals to behave differently. According to this idea value increases in terms of current income, assets or future income are regarded as an increase of the total amount wealth throughout someone’s life. It assumes that individuals possess the self-control to do this without temporarily changing the level of consumption and that wealth is assumed to be distributed evenly (Ando & Modigliani, 1963). However, this is different in real life. Individuals are assumed to have different saving and consumption profiles when they are young, compared to when they are older. Thaler (1990) states that consumption is dependent on a change in income (Flavin, 1981; Barro, 1989; Wilcox, 1989). A small short-term increase in income is primarily regarded as current income, and an individual is more likely to want to consume this. Most of the time, a large short-term gain is added to an individuals’ assets account. And individuals are less likely to consume from the assets account (Thaler, 1990).
In line with Thaler but in contrast to the ‘conventional’ life-cycle model, Levin (1998) introduced the behavioral life-cycle model. Originally designed by Shefrin and Thaler (1988), this model states that people’s self-control is based on three ideas: 1. In general, individuals experience the temptation to spend all their income on consumption, instead of saving for the future. 2. Individuals who do save, invest this in assets. 3. Individuals form ‘mental accounts’ and couple different levels of temptation to consume these assets (Levin, 1998). This temptation is expressed in the ‘marginal propensity to consume’ (MPC). In his study, Levin compared the applicability of both the conventional life-cycle model and the behavioral life-cycle model. The first test consists of a comparison of the MPC from income to the MPC from wealth. From the perspective of the conventional life-cycle model, the receipt of one “extra dollar of permanent income should change consumption as much as receiving an extra amount of wealth” (Levin, 1998). However, Levin found that the effect of income increase on consumption is 26 times larger than the effect of an increase in wealth. In the second test, Levin concentrated on the MPC from wealth to determine a different effect for different types of assets. In line with the behavioral life-cycle model, the second test suggests that the effect of housing wealth is negligible, thus smaller compared to the effect that liquid and future wealth have on consumption (Levin, 1998).
3.1.2 Wealth effect and collateral effect Studies based on macro level data
The ‘wealth effect’ can be described as the increase in the value of wealth which has an impact on the interpretation of current and future income and, in turn, affects consumption behavior (Case, Quigley and Shiller, 2005). Case et al. (2005) have done research on the effect of housing wealth and stock market wealth on household spending on consumption. The study is based on two sources of cross-sectional time-series data. First of which is an international panel of 14 OECD countries, observed annually from 1975 until 1999. The second set consists of panel data from the US with quarterly observations from 1982 until 1999. Case et al. (2005) have found that changes in house prices have a larger and more crucial effect on household consumption than changes in stock market prices. The international data shows that a 10 percent increase in housing wealth leads to an increase of household consumption of 1,1 percent. While virtually no effect was measured as a result of a 10 percent increase in stock market wealth. According to the data from the panel of the US states a 10 percent increase in housing wealth, as well as in stock market wealth, both lead to an estimate increase of 0,4 percent in consumption (Case et al., 2005).
A more recent version of this research was conducted in 2012 and published in 2013. The previous research dated from 2005 and was based on data from before the financial crisis of 2008. The high volatility of the economy at that time could have influenced the data. In this study, Case et al. (2013) have only focused on quarterly data of US states from 1975 until 2012 because they were unable to obtain comparable data from the panel of OECD countries. They have found that including data from the housing boom and the financial crisis caused a stronger relationship between housing market wealth and consumption, in comparison to the influence of stock market wealth on consumption. In addition to the wealth effect as described above, a second effect can be distinguished. An increase of housing market wealth leads to an increase in collateral, or home equity. The ‘collateral effect’ has enabled individuals to consume from their housing equity in the form of a second mortgage. In the housing boom of 2001 until 2005, home equity loans, second mortgages and cash-out refinance structures were good for an average of almost 700 billion dollars of equity extracted each year (Greenspan & Kennedy, 2008). The estimated elasticity of household spending on consumption to housing wealth ranges from 0,03 to 0,18 in a rising market and is estimated at 0,10 in a falling market. Which means that a 10 percent increase leads to an increased consumption between 0,3-1,8 percent. A decrease in housing wealth incurs a significant decrease in household consumption. For comparison, the estimated elasticity of household consumption to stock market wealth ranges from 0,014 to 0,027 (Case et al., 2013).
Ahec Šonje, Čeh Časni & Vizek (2014) have studied the impact of housing and stock market wealth effects on consumption. They have used a data set consisting a total of 30 developed and emerging countries, which have been categorized across financial structures. The years of the data ranges from 1970 to 2012, but this is dependent on the availability of data for each country. The research is concentrated on the discovery of a different effect on consumption in developed countries compared to emerging countries. Ahec Šonje et al. (2014) have divided the effects in long-run and short-run effects, where short run means that the housing bubble in the years before 2007 is incorporated in the results. They have found an indication that, in the short run, housing wealth positively affects private consumption in developed economies, except for The Netherlands and Singapore. While this positive relationship is not observed in emerging countries in the short run, Ahec Šonje et al. suggest that consumption is affected by housing wealth and stock market effects in the long run (Ahec Šonje et al., 2014).
Studies based on micro level data
Case et al. (2005, 2013), Ahec Šonje et al. (2014) have taken a macro-level approach and have studied the effect of housing wealth on consumption across multiple countries. Spiegelaar & Lennartz (2018) have focused the scope of their research specifically on The Netherlands. Their research is based on micro-level data of over 40.000 Dutch mortgage owners from April 2014 until December 2016. Spiegelaar and Lennartz (2018) indicate that an increase in housing market value of 1 euro relates to an increased MPC of 0,04 euros. Moreover, they have found that the effect on the marginal propensity to consume out of housing wealth is stronger in provinces where the increase of house prices was larger. A province in which house prices have increased stronger relative to other provinces is North Holland. Homeowners in North-Holland consume an average of 0,06 euros of every euro of housing wealth increase (Spiegelaar & Lennartz, 2018). This is 50 percent more than the national average.
Similar to Spiegelaar & Lennartz (2018) and contrary to the literature discussed in the previous section, Campbell and Cocco (2007) have studied the relation between a change in house prices and household consumption level with a focus on one specific country. Their research is based on data from the UK Family Expenditure Survey. To determine if age and homeownership have an impact on the relation of consumption and house prices, they have categorized the data in young homeowners, old homeowners, young renters and old renters. Campbell and Cocco (2007) state that the relation between housing wealth and consumption is positive and is that it is the strongest for old homeowners. Young homeowners, young renters and old renters experience a negative effect on household consumption as a result of a change in housing wealth. The magnitude of this effect is lower for young
homeowners than for old renters. The negative relationship is the strongest for the group of young renters (Campbell & Cocco, 2007).
In their study, Benmelech, Guren and Melzer (2017) suggest that the purchase of a house affects the level of consumption from three months before the purchase to one year after the purchase. Homebuyers are more inclined to invest in home durables or improvements on their new home. Home durables include furniture, household devices and consumer electronics (Fical Garone et al., 2019). Benmelech et al. (2017) have found that in the months leading up to the purchase, and one year after the purchase of a house, homebuyers spend 5.900 US dollars more compared to homeowners who did not buy a new house. Furthermore, this increase in spending is not affected by a change of spending in other categories of consumption. Their research concentrates on the time period between April 2001 and March 2013. The study is based on the Consumer Expenditure Survey , which collects monthly data of nearly 30.000 American households, complemented by a data base of building permits. The findings of Benmelech et al. (2017) suggest that the propensity to consume in terms of home durables increases with an average of 7,7 percent in the first nine months after the purchase of a house. With an increase of 13,4 percent in the first month and 16,1 percent in the second month. After the ninth month, the authors regard the effect of 0,3 percent as negligible. Over the course of the first twelve months after the purchase of a house, the average increase of spending on consumption is equal to 7,2 percent. Furthermore, Benmelech et al. (2017) have found that spending on consumption is reduced in the three months prior to the purchase of a house. The level of spending on consumption is equal to -13,8 percent three months before the purchase, -15,7 percent two months before the purchase and -10,7 percent one month before the purchase of a house (Benmelech et al., 2017).
3.2 Saving
It has been demonstrated in earlier research that the appreciation of house prices leads to an increase of consumption of homeowners. This is however not an effect with an open end. Since young individuals who are planning to buy a house suffer from the increase of wealth from current homeowners (Skinner, 1989; Feldstein, 1977; Charnley & Wright, 1987). This means that future homeowners have to increase their level of savings, defined as sacrificing present wellbeing for future results (Brounen et al., 2016). An alternative option is that current homeowners leave the increase in housing wealth as a bequest to help their children with buying a house, instead of increasing their own consumption. This assumption is confirmed by research of Skinner (1989), which is based on data from the Panel Survey of Income Dynamics. This panel contains data from over 11.000 households in the United States. It shows that individuals who are concerned about their children have a lower marginal propensity to consume their capital gain and leave larger bequests. Skinner (1989) uses a fixed-effects
model and a cross-section time-series regression. Only the latter indicates an positive effect of housing value appreciation on saving in the 1970s.
3.2.1 Homeownership and saving
The impact of homeownership on savings is studied by Krumm and Miller (1986). This research is based on data from the Panel Survey of Income Dynamics taken from 1970 until 1979. They assumed that young households save for the down payment for the mortgage on their first house. After the purchase of their house, young households start saving again. By paying off their mortgage and through housing capital gain, they build up home equity. Krumm and Miller (1986) have found that homeowners between 1970 and 1979, had an estimated of 29.000 US dollar in home equity and 16.000 US dollars more in savings compared to non-homeowners.
3.2.2 House price appreciation and saving
To bring this in relation with a change in housing capital, Yoshikawa and Ohtake (1989) conducted research on the housing market in Japan. Their research uses data from the 1984 National Survey of Family Income and Expenditure. The dataset with 54.000 households contained virtually all households in Japan at the time of the study (Yoshikawa & Ohtake, 1989). They found that the aggregate savings rates for renter households which were planning to buy a house were lower with higher land prices. Meaning that the aggregate savings rate of 20 percent in 1984 reduced by about 0,25 percent. Renter households cancelled their plans to move to permanent residence and the total number of households planning to buy a house was lower in the situation of higher land prices. At the same time, consumption levels were higher among these renter households now that their housing purchase plans were out of the way. Research of Case et al. (2005) is described in paragraph 3.1.2. Similar to Yoshikawa and Ohtake (1989), Case et al. suggest that an increase in house prices has a negative effect on saving behavior. This study also confirmed that higher house prices reduced the probability that renter households saved for a down payment (Case et al, 2005).
In line with Yoshikawa and Ohtake (1989) and Case et al. (2005) are the findings of research done by Engelhardt (1994). Based on data from Canada it was found that an increase of 4000 Canadian dollars of house prices decreased the probability of saving by 1 percent point, and a reduction of 1200 Canadian dollars in accumulated assets (Engelhardt, 1994). Later research by Engelhardt (1996) suggests that households do not increase their savings and consumptions in a time of real capital gain on their current house. This is substantiated by research by Hoynes and McFadden (1997), who used micro-level data to examine a correlation between individual savings and residential capital gains. Engelhardt & Mayer (1998) substantiated Skinner’s (1989) assumption regarding housing capital gain that would be saved as a bequest. They have found that the appreciation of house prices increased the
reliability of first-home owners on transfers from family members to assist with the down payment (Engelhardt & Mayer, 1998). This further decreases the incentive for future homeowners to save.
Brounen, Koedijk and Pownall (2016) have conducted research on the saving behavior of different generations, based on a set of 1253 households in The Netherlands from The Dutch National Bank Household Survey of 2011. The generations are not divided in an old and young generation as in the study of Cocco and Campbell (2007), but in generations: born before 1945, born between 1945 and 1960 and born between 1960 and 1975. They suggest that the willingness to save decreases when individuals become older and that younger households have a higher propensity to save (MPS) than older households. At the same time, the MPS is higher among individuals which have a higher financial literacy than others. The study finds additional evidence on the impact of parental stimulation and a side job during youth years on the saving behavior of an individual later in life (Brounen et al., 2016).
3.2.3 The effect of house price depreciation
In the run-up to 2008, Dutch house prices experienced an increase. After 2008, we can observe a drop in house prices as well as in consumption. Previously described studies focused on the effect of house prices and housing wealth appreciation on the propensity to save. The Centraal Plan Bureau, the Dutch Bureau for Economic Policy Analysis published the research of Bijlsma and Mocking in 2017. Their study concentrates on the effect that the depreciation of house prices has on household savings. This is done for renter households and homeowners from 2006 until 2013. Bijlsma and Mocking (2017) suggest that the marginal propensity to save out of housing wealth can be categorized in a short-run effect as a shock response and a long-run effect. They have found that the MPS as a reaction on a decrease in housing wealth ranges from 0 to -0,05 in the short-run and from -0,02 to -0,013 in the long-run. The drop in house prices caused households to increase savings by 1,5 percent (Bijlsma & Mocking, 2017). The study of Hoynes and McFadden (1997) suggests that households who see the real value of their house decrease, do in fact reduce the level of consumption.
3.3 Determining the value of a house
In order to say something about consumer response on the change of house prices, it is important to explain factors that determine house prices. Salzman and Zwinkels (2013) have made a distinction between financial institutional factors and factors based on consumer behavior. Institutional factors that determine house prices are income, interest rates, the availability of credit, the tax structure and demographic factors. Factors based on consumer behavior consist of psychological biases that can exist among house buyers. Biases that are formulated by Salzman and Zwinkels (2013) are over-optimism and over-confidence, momentum effect and money illusion. The biases are explained in this paragraph. Over-optimism is regarded as “the most important psychological bias in real estate markets”
(Farlow, 2004). According to Farlow (2004), households believe that the annual increase of house prices is more than 11 percent. Furthermore, households make false judgments of future interest rates. Hand in hand with over-optimism goes over-confidence. This bias originates from the underestimation of risk, and the notion that an individual can predict what is going to happen in the real estate market. The magnitude of price changes at one point are extrapolated and used to make an assessment of future price changes. Research by Case et al. (2003) suggests that over 80 percent of households buy a house because of price increases. In this case we speak of the momentum effect (Salzman & Zwinkels, 2013). The fourth bias to be discussed is money illusion. This occurs when individuals are unable to correctly distinguish nominal values from real values in a time of inflation. Prices of houses are, more often than not, indicated in nominal value instead of real value. The price is thus not corrected for inflation. For many people, the purchase of a house is a large decision. Therefore the nominal purchase price from a long time ago is well remembered. The difference between the initial purchase price and the current value can give the impression that the value of the house has appreciated more than it had if the value had been corrected for inflation (Eichholtz, 1997; Salzman & Zwinkels, 2013). Farlow (2004) states that banks and media contribute to a distorted picture of the real estate market, since they often publish nominal rates instead of real rates.
3.3.1 WOZ-value
In 1994, The Netherlands adopted the ‘Wet waardering onroerende zaken’, or WOZ in short (CentERdata, 2018). This law determines the valuation of property in The Netherlands in terms of taxation purposes. The value of real estate property is re-evaluated by the responsible municipalities each year. An appraiser makes an assessment of the WOZ value based on information of the plot and the surface area of buildings on the property (Ministerie van Algemene Zaken, 2019). The WOZ-value through which municipal tax is calculated for one year, indicates the value of the property on the 1st of January one year before that. Appreciation in housing value can be expressed in an increase of the WOZ value. In addition to a second mortgage, in The Netherlands it is possible to receive an extra line of credit based on the surplus value of a house. The advantage of this type of credit, called WOZ credit, is that homeowners are not obligated to employ an appraiser. Because the appreciation of housing value is determined by the municipality.
Table 1: Studies of wealth effects on household consumption
Authors Time span Predictor Sign
Case, Quigley and Shiller (2005) 1975-1999 (Int’l data) Housing wealth increase + 1982-1999 (US data) Housing wealth increase + Case, Quigley and Shiller (2013) 1975-2012 (US data) Housing wealth increase +
Campbell & Cocco (2007) 1988-2000 Housing wealth increase + (old homeowners) – (young homeowners) Ahec Šonje, Čeh Časni & Vizek (2014) 1970-2012 Housing wealth increase +
Benmelech, Guren & Melzer (2017) 2001-2013 Time 1 year before purchase – Time 1 year after purchase + Spiegelaar & Lennartz (2018) 2014-2016 Housing value increase +
Hoynes & McFadden (1997) 1968-1989 Housing value decrease –
Table 2: Studies of wealth effects on household savings
Authors Time span Predictor Sign
Skinner (1989) 1972-1980 Housing value increase + (homeowners leaving
a bequest)
Krumm & Miller (1986) 1970-1979 Homeownership +
Yoshikawa & Ohtake (1989) 1984 Housing value increase –
Case, Quigley and Shiller (2005) 1975-1999 (Int’l data) Housing value increase – (renter households) 1982-1999 (US data) Housing value increase – (renter households)
Engelhardt (1994) 1983-1984 Housing value increase –
Engelhardt (1996) 1984 & 1989 Housing value increase No effect Hoynes & McFadden (1997) 1968-1989 Housing value increase No effect
Engelhardt & Mayer (1998) 1988, 1990, 1993 Housing value increase + (homeowners leaving a bequest)
Brounen, Koedijk & Pownall (2016) 2011 Housing value increase – (old homeowners) + (young homeowners) Bijlsma & Mocking (2017) 2006-2013 Housing value decrease –
4. Expectations and conceptual framework 4.1 Expectations
The expectations of this study will be presented in this paragraph. Based on the expectations, we can formulate a set of hypotheses. Furthermore, this paragraph presents and explains the conceptual framework.
The literature explained in the theoretical framework in the previous paragraph suggests that multiple factors relate to household consumption. Higher household consumption can be the result of an increase in income. Scholars have indicated the positive effect that housing wealth appreciation has on consumption. An even stronger effect can be observed in developed countries and among older homeowners (Campbell & Cocco, 2007; Ahec Šonje et al., 2014). While it is not possible to assume that one automatically leads to the other, and vice versa, increased consumption can be brought in relation to a decrease in savings. Bijlsma & Mocking (2017) have used savings as a variable to calculate the consumption, which will be elaborated on in paragraph 5.1.2. Saving behavior of households is therefore included in this study. Studies that are reviewed in the theoretical framework concentrate on the relation between the fluctuations in housing value and individuals’ propensity to save. Earlier research suggests that increasing house prices have a negative effect on individuals’ saving behavior in anticipation of buying a house (Yoshikawa & Ohtake, 1989; Engelhardt, 1994; Case et al., 2005; Brounen et al., 2016). Where these studies fall short is to indicate whether the purchase of a house will result in a change in household consumption. This is the angle this thesis aspires to take. Accordingly, the following research question is formulated:
Does household spending behavior change around the time of purchase of a house?
‘Around the time of purchase’ is interpreted as one year before the purchase, and one year after the purchase of a house. This will be elaborated on in the paragraph on empirical methodology. The first hypothesis elaborates on the findings that housing wealth is positively related to household consumption (Case et al., 2003; Case et al., 2005; Campbell & Cocco, 2007; Ahec Šonje et al., 2014; Spiegelaar & Lennartz, 2018). This hypothesis is based on the sub-question:
Does the purchase of a house lead to a change in household consumption?
The expectation is the following: one year in anticipation of buying the house the household consumption level is lower. One year after the purchase, the consumption is expected to experience a slight increase. This expectation is based on the findings of the study that Benmelech et al. (2017) have conducted on households in the United States. This is expected to be the case for older homeowners
and not for young homeowners. This twofold expectation for the consumption level one year after buying a house is based on the findings of Campbell & Cocco (2007).
The findings of studies of wealth effects on savings suggest that an increase of house and land prices leads to a lower level of household savings (Yoshikawa & Ohtake, 1989; Engelhardt, 1994). In case of higher house prices, young households would be less inclined to buy a house. Moreover, house price appreciation compels them to increasingly rely on family support to provide for the down payment. The second hypothesis is based on the following sub-question: Does the purchase of a house lead to a change in household savings?
Based on the findings of Krumm and Miller (1986), this paper hypothesizes that future homeowners or current homeowners who are planning to move increase their savings before they buy a house. For example to be able to pay for a down payment. Furthermore, it is expected that one year after the purchase, savings are increase by younger households and reduced by older households. This idea is in line with Brounen et al. (2016).
To elaborate on the idea that higher housing value leads to a higher level of household consumption around the purchase of a house, this study uses the WOZ-value which is documented by the municipality. The difference in WOZ-value of the house in the first year and in the second year after the purchase will be studied. This way, the housing wealth effect on consumption can be tested. For the third hypothesis, the following sub-question is formulated: Does the purchase of a house lead to a change in household consumption, in the case of an increased WOZ-value?
An increase in WOZ-value, and therefore an increase in housing wealth is expected to have a positive effect on the level of consumption of old homeowners and a negative effect on the level of consumption of young homeowners. Furthermore, it is expected that savings of old homeowners will decrease and that savings of young homeowners will increase as a result of a WOZ-value increase (Brounen et al., 2016)
Table 3: Summary of hypotheses
Hypothesis Expectation
H1 1 year before purchase: consumption decrease
1 year after purchase: young homeowners decrease consumption old homeowners increase consumption
H2 1 year before purchase: savings increase
1 year after purchase: young homeowners increase savings old homeowners decrease savings
H3 Home wealth increase:
old homeowners increase consumption old homeowners decrease savings young homeowners decrease consumption young homeowners increase savings
4.2 Conceptual framework
The expectations mentioned in the previous paragraph are structured in a conceptual framework, presented below with the associated sign. The variables are discussed in order to explain the logic behind the framework. The main concepts, explained in the paragraph on the expectations, are consumption, savings, WOZ-value increase and the time of the purchase of a house. The twofold dependent variables in this case are the level of consumption and the level of savings. Independent variables are the time of the purchase of a house and a positive change of the WOZ-value. In the figure below, the dependent variables are the concepts with arrows directed towards them. For the independent variables the arrows run in the opposite direction. The direction of the arrow are based on the expectations, which are discussed in the previous paragraph 4.1. The consumption level is expected to be lower one year before buying a house. One year after the purchase, we expect a positive effect on consumption for old homeowners and a negative effect for young homeowners. The level of savings is expected to be positively impacted one year before the purchase of a house. The expectation is that younger households increase their saving behavior one year after they have purchased a house, and old households will experience a lower level of savings. Furthermore, the increase in WOZ-value is expressed in a twofold effect. For young homeowners, WOZ-value increase is expected to lead to an increase of the level of savings and a decrease of the level of household consumption. Old homeowners on the other hand, are expected to increase their consumption and decrease their level of savings in a situation of WOZ-value increase.
Fi gu re 1 : C on ce ptu al fr am ew or k
5. Research methodology
This study is based on a quantitative analysis. There are two main reasons for the choice of this type of research over other types of research. First, by using a quantitative type of analysis, we can explore whether there is a relationship between the purchase of a house and changes in consumption and saving behavior. The second reason is that the type of data that is used for this research consists of a large number of observations. A quantitative analysis is most suited for this type of data (Burrell & Gross, 2018). This chapter will explain the collection of the data and the construction of variables.
5.1 Data
5.1.1 Dutch National Bank Household Survey
The Dutch National Bank Household Survey (DHS) is used in this study. DHS is a panel survey and consists of cross-sectional time-series data of over 2000 households. The focus of the data set is to identify the saving behavior of households in terms of economic and psychological factors. DHS is representative of the Dutch population due to the variety of households in the data set and has been conducted annually since 1993 (Brounen et al., 2016; CentERdata, 2018). For this paper, we have used data from 2004 to 2018 because starting from 2004 respondents are asked to report the WOZ-value in the survey (CentERdata, 2018). This variable is used to test the third hypothesis, as is explained further in this paragraph. The survey questions, as well as the data are divided into sections. The first section contains general demographic information such as the composition of the household, the level of education, followed by a section with information on employment and pension. The third section is dedicated to the accommodation situation and mortgages. Followed by two sections with information about income and wealth. The last section with economic and psychological concepts presents information on the financial situation and the saving behavior of the household.
Information of survey respondents regarding different sources of income, such as welfare payments and subsidies, enable CentERdata to calculate aggregated gross and net incomes. A similar calculation is conducted on information on assets and liabilities of households, resulting in numbers of aggregated wealth. This part of the survey consists of information on e.g. mortgages, loans, savings or stocks. The survey consists of questions that can be answered with yes and no, questions that require a numeric answer and ‘interval answers’. If the survey provides answer options such as ‘between 500 and 1000 euros’ or ‘between 1000 and 2000 euros’ to the question how much an individual spends on consumption on a monthly basis, it is called an interval answer. CentERdata (2018) have taken the mean of the interval answers in the calculation of aggregated income and wealth numbers. If an answer option states ’50.000 or higher’, CentERdata have taken 50.000 as a reference point for their calculations.
5.1.2 Description of variables
The dependent variable throughout this study is household consumption. Unfortunately, DHS does not provide numbers on the household consumption. Because consumption data is not directly available from the survey, we have to calculate the consumption function for each individual for each year. This is done by focusing on the mirror of consumption: savings and mortgage payments (Bijlsma & Mocking, 2017). The consumption is calculated by taking the aggregated total annual net income which is provided by CentERdata and subtract annual savings and annual mortgage payoffs: C = (N − S − M ). To obtain a percentage value of a possible effect on consumption, the natural log of consumption is used. The implementation of variables in empirical models to answer the research questions and sub-questions is discussed in the next paragraph. In the description of the empirical model, the dependent variable consumption as a percentage change is referred to as ‘lnConsumption’. Table 4 provides descriptive statistics of the research population. High maximum values and low minimum values are observed for the level of consumption. This value is influenced respectively by individuals who have received an inheritance that year, or individuals who pay high levels of alimony.
Savings are reported in the dataset as the total amount of savings in the past 12 months. For variable ‘hoevspa’, DHS provided respondents with seven interval answers. The mean of these intervals is taken to determine a calculable value for the consumption estimation. This strategy is in line with the approach that CentERdata have used to calculate aggregate income and wealth (CentERdata, 2018). The values for the seven answer options are determined accordingly: ‘less than €1500’ becomes €750; ‘between €1500 - €5000’ becomes €3250; ‘between €5000 - €12.500’ becomes €8750; ‘between €12.500 - €20.000’ becomes €16.250; ‘between €20.000 - €37.500’ becomes €28.750; ‘between €37.500 - €75.000’ becomes €56.250; ‘€75.000 or more’ is maintained at €75.000. Similar to the natural log of the change in consumption, the effect on saving is expressed in a natural log function. The dependent variable is referred to as ‘lnSavings’.
Mortgage payments that individuals pay monthly are multiplied by 12 (months per year), in order to obtain correct information and incorporate this into the calculation of yearly household consumption. In ‘Descriptive statistics’ in table 4 on the next page, we can see that this calculation yields a number of negative values. A possible explanation for this could be that respondents have inaccurately answered the question if the mortgage payments are done monthly or yearly. Another explanation could be that households consist of two partners who are both paying a share of the mortgage payments. The survey reports the total amount of monthly mortgage payments per household. At the same time, DHS indicates whether individuals are part of a dual earner household. Based on this
information, the mortgage payments of individuals who are part of a dual earner household are divided by two.
One of the concepts suggested in the conceptual framework is the time of a house purchase. This study focuses on the exploration of an effect on consumption and savings one year before the purchase and one year after the purchase. To determine the year that a house has been purchased, we look at variable WOD35b in the data set of the DHS. WOZ-value reported in the data set corresponds to the value of the property on January 1st of the year prior to the year that individuals have completed the survey. Therefore, if respondents have participated in the survey for two consecutive years, the WOZ-value of the first year is used as the ‘current’ WOZ-WOZ-value of the second year. It is expected that an increase in WOZ-value holds a positive relationship to the effect of the household consumption and a negative relationship to saving behavior. This means that if the WOZ-value increases, the consumption is expected to increase and savings are expected to decrease. The relative percentage change of the WOZ-value is calculated as follows:
∆WOZ % = (𝑊𝑂𝑍𝑡 = 1 − 𝑊𝑂𝑍𝑡 = 0
𝑊𝑂𝑍𝑡 = 0 ) 100
To calculate property value increase for households that have bought a house one year before their participation in the survey, we have we have used the WOZ-value of the second year.
Consumption and saving behavior varies across ages. The effect of housing wealth on consumption and saving behavior is indicated to be different for younger households than for older households. Brounen et al. (2016) have used three generations: individuals born before 1945, born between 1945 and 1960 and born between 1960 and 1975. Campbell and Cocco (2007) have used two generations in their research, in which they draw the line at 40 years. Homeowners younger than 40 years old are seen as ‘young’ and over 40 years is indicated as ‘old’. This study takes the approach of two generations in order to specify a possible effect. The young generation consists of homeowners under 40 years old at the time of this study, which means that an individual is born after 1978. The old generation consists of homeowners over 60 years old at the time of this study, thus individuals that are born before 1960. Instead of opting for a young generation under 40 years old and an older generation over 40 years old, this study distinguishes between a generation ‘under 40 ‘and a generation ‘over 60’. In this case, we assume a greater difference between the two generations’ stages of life. On the other hand, this distinction results in a smaller population of people in the data set that are labeled as older. To observe the effect of ∆WOZ for the two different generations, the change in WOZ-value is interacted with the generation.
5.2 Descriptive statistics Table 4: Descriptive statistics
N of unique observations = 3.597 Mean SD Minimum Maximum
Individual demographics
Year of birth 1956,12 14,81 1915 1997
Sex (1 = Male, 2 = Female) 1,28 0,45 1 2
Highest level of education (0=no education,
7=university) 5,01 1,50 0 7
Monthly net income (in euros) 2597,04 1828,84 -400 57.475,34
Annual consumption level 26647,95 21845,56 -247720,10 689704,10
Generations
Generation under 40 years old (born after 1978) 1982,85 3,31 1979 1997 Generation over 60 years old (born before 1960) 1945,89 8,59 1915 1959
Composition of the household
Number of household members 2,48 1,25 1 9
Number of children 0,68 1,07 0 7
Partner (0 = no, 1 = yes) 0,78 0,41 0 1
Occupation 2,37 1,46 0 4
1 = employed, 2 = own business, 3 = freelance 4 = unemployed
Degree of Urbanization (1 = very high, 5 = very low) 3,11 1,28 1 5
Region (1 = Three largest cities, 2 = West, 3,18 1,40 1 5
5.3 Empirical model
This section of the methodology paragraph explains the methods for answering the main research question and each sub-question. It presents the empirical model for each sub-question. The data from the DNB Household Survey consists of panel data, or cross sectional time-series data. To analyze panel data this study employs fixed effects (FE) and ordinary least squares (OLS) regression models. A fixed effects model keeps individuals in the panel ‘fixed’ and removes the effect of time-invariant characteristics. (Torres-Reyna, 2018) Because the model only concentrates on varying variables over time, it allows us to study the impact of potential changes of the independent variables within individuals in the panel data over a period of time. An ordinary least squares (OLS) model is used in addition, in order to estimate the effect of time-invariant characteristics that the FE model removes. And to explore a possible overall effect in the panel data population, instead of solely focusing on an effect within individuals. In the previous paragraph 5.1.2, different variables are discussed, which are implemented in different types of models. To estimate the effect on consumption the OLS regression in model 1 is as follows:
Model 1 (OLS)
𝐶𝑖 = α + 𝛽1𝐵𝐸𝐹𝑂𝑅𝐸𝑖+ 𝛽2𝐶𝑈𝑅𝑅𝐸𝑁𝑇𝑖+ 𝛽3𝐴𝐹𝑇𝐸𝑅𝑖+ 𝜀𝑖
Where the outcome variable Ci is the natural log of the level of consumption, expressed in a percentage change. is the estimate of the intercept. 1BEFOREi is the year dummy variable for individual i. This indicates if an individual has purchased a house in the year preceding the year that he or she participated in the DHS. 2CURRENTi is the year dummy variable for individual i that indicates whether an individual has purchased a house in the same year of participating in the DHS. 3AFTERi is the year dummy variable for individual i that indicates if an individual has purchased a house in the year after he or she has participated in the DHS. If 3AFTERi is equal to one, then 2CURRENTi and 3AFTERi are equal to zero. And vice versa.I is the error term for the ith participant (Field, 2013).
Model 2 (FE)
𝐶𝑖𝑡 = 𝛼𝑖+ 𝛽1𝑖𝐵𝐸𝐹𝑂𝑅𝐸𝑖𝑡+ 𝛽2𝑖𝐶𝑈𝑅𝑅𝐸𝑁𝑇𝑖𝑡+ 𝛽3𝑖𝐴𝐹𝑇𝐸𝑅𝑖𝑡+ 𝜀𝑖𝑡 𝛼𝑖 = 𝛼 + 𝑈𝑖
In the fixed effects model (2) above, Cit is the outcome variable consumption as a natural log, which indicates a percentage change. i is the unknown intercept for each individual. is the intercept, and Ui is the random effect on the intercept for that individual (Field, 2013). In this case, ni is the unknown coefficient for that independent variable (Torres-Reyna, 2007). 1BEFOREit is the year dummy variable for individual i at time (year) t to indicate whether an individual has purchased a house the year prior to their participation in the DHS. 2CURRENTit is the year dummy variable for individual i at time (year) t. This predictor variable is a year dummy variable which indicates that individuals have purchased a house in the same year that they have participated in the survey. 3AFTERit is the year dummy for individual i at time (year) t which is equal to one if respondents have purchased a house in the year after they have first participated in the DHS. I is the error term for the ith participant at time (year) t (Field, 2013).
Model 3 (OLS)
𝐶𝑖 = α + 𝛽1𝑖𝐵𝐸𝐹𝑂𝑅𝐸𝑖+ 𝛽2𝑖𝐶𝑈𝑅𝑅𝐸𝑁𝑇𝑖+ 𝛽3𝑖𝐴𝐹𝑇𝐸𝑅𝑖+ 𝛽4𝑖𝑈40𝑖+ 𝛽5𝑖𝑂60𝑖+ 𝜀𝑖
Model 3 is based on an OLS regression which accounts for two generation groups. Discussed in the previous paragraph is that the panel data population has been distinguished in a generation under 40 years old and a generation over 60 years old at the time of this study. As with our previous OLS model in model 1 for the natural log change in consumption, is the estimate of the intercept. 1BEFOREi is the dummy variable for individual i which indicates if respondents have purchased their house one year prior to the year of their participation in the survey. 2CURRENTi indicates if respondents have purchased their house in the same year as the year of the survey. And 3AFTERi indicates if respondents have purchased a house one year after their participation in the survey. 4U40i is the dummy variable for respondents born after 1978 and 5O60i is the dummy variable for respondents born before 1960. As in model 1, I is the error term for the ith participant (Field, 2013).
Model 4 (OLS)
𝐶𝑖 = α + 𝛽1𝑖𝐴𝐹𝑇𝐸𝑅𝑈40𝑖+ 𝛽2𝑖𝐴𝐹𝑇𝐸𝑅𝑂60𝑖+ 𝜀𝑖 𝛽1𝐴𝐹𝑇𝐸𝑅𝑈40𝑖 = (𝛽n𝐴𝐹𝑇𝐸𝑅𝑖× 𝛽n𝑈40𝑖) 𝛽2𝐴𝐹𝑇𝐸𝑅𝑂60𝑖 = (𝛽n𝐴𝐹𝑇𝐸𝑅𝑖× 𝛽n𝑂60𝑖)
Model 4 is used to focus on a general effect on consumption for individuals in either of the generations, who have bought a house one year after their participation in the survey. In which Ci is the natural log consumption level, is the intercept and I is the error term. 1AFTERi is the dummy variable for the latter effect. nU40i and nO60i are the dummy variable for respondents born respectively after 1978 and before 1960. 1iAFTERU40i is the interacted term of variables nU40i and1AFTERi. 2iAFTERO60i is constructed through the interaction of the variables nO60i and1AFTERi. Simply copying model 3 and excluding variables 1BEFOREi and 2CURRENTi leads to omitted variable bias. Therefore the estimation through interaction terms is necessary to show a possible effect on consumption of individuals in either of the two generations, in the year after the purchase of a house.
Model 5 (FE) 𝐶𝑖𝑡 = α𝑖𝑡 + 𝛽1𝑖∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖𝑡 + 𝛽2𝑖∆𝑊𝑂𝑍𝑈40𝑖𝑡 + 𝛽3∆𝑊𝑂𝑍𝑂60𝑖𝑡 + 𝜀𝑖𝑡 𝛼𝑖 = 𝛼 + 𝑈𝑖 𝑛𝑖= 𝛽𝑛+ 𝛾𝑛𝑖 (n = 1 … 3) 𝛽1∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖 = (𝑛∆𝑊𝑂𝑍𝑖 × 𝑛𝐴𝐹𝑇𝐸𝑅𝑖) 𝛽2∆𝑊𝑂𝑍𝑈40𝑖 = (𝑛𝑈40𝑖 × 𝑛𝐴𝐹𝑇𝐸𝑅𝑖) × 1∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖 3∆𝑊𝑂𝑍𝑂60𝑖 = (𝑛𝑂60𝑖 × 𝑛𝐴𝐹𝑇𝐸𝑅𝑖) × 1∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖
The concentration on the effect that is sought to be indicated by the use of model 4 is used in model 5 above. This is to indicate a collateral effect in percentage change of consumption for individuals who have purchased a house in the year after participating in the survey and who have experienced a change in the WOZ-value in that year. Where is the intercept and Ui is the random effect on the intercept for that individual (Field, 2013). ni is the unknown coefficient for each independent variable. 1∆WOZAFTERi is the function of WOZ change for individual i at time t, which is interacted with the dummy for individuals who have purchased a house one year after their participation in the survey. For variable 2∆WOZU40i we have interacted a change in WOZ-value with the generation of
respondents under 40 who have purchased a house the year after their participation in the survey. 3∆WOZO60i is calculated in a similar way but for individuals in the generation over 60 years old. I is the error term for the ith participant at time (year) t (Field, 2013).
Model 6 (OLS)
𝑆𝑖 = α + 𝛽1𝐵𝐸𝐹𝑂𝑅𝐸𝑖+ 𝛽2𝐶𝑈𝑅𝑅𝐸𝑁𝑇𝑖+ 𝛽3𝐴𝐹𝑇𝐸𝑅𝑖+ 𝜀𝑖
Model 6 is constructed in a similar fashion as model 1. For this reason the predictor variables are explained more concisely. The outcome variable in this case is Si, which is the percentage change of savings. Obtained by taking the natural log of the change in saving behavior. is the intercept, I is the error term for the ith participant (Field, 2013). 1BEFOREi is the dummy variable for individuals who have purchased a house the year before participating in the survey. 2CURRENTi indicates whether an individual has purchased a house in the year of participating in the DHS. 3AFTERi indicates if an individual has purchased a house a year after their participation in the DHS.
Model 7 (FE)
𝑆𝑖𝑡 = 𝛼𝑖+ 𝛽1𝑖𝐵𝐸𝐹𝑂𝑅𝐸𝑖𝑡+ 𝛽2𝑖𝐶𝑈𝑅𝑅𝐸𝑁𝑇𝑖𝑡+ 𝛽3𝑖𝐴𝐹𝑇𝐸𝑅𝑖𝑡+ 𝜀𝑖𝑡 𝛼𝑖 = 𝛼 + 𝑈𝑖
𝑛𝑖= 𝛽𝑛+ 𝛾𝑛𝑖 (n = 1 … 3)
Model 7 above is based on model 2. This fixed effects model has Sit as the outcome variable, which is the percentage change of saving behavior. i is the unknown intercept for each individual. is the intercept, and Ui is the random effect on the intercept for that individual (Field, 2013). ni is the unknown coefficient for that independent variable. 1BEFOREit is the dummy variable for individuals who have purchased a house the year before participating in the survey in year t. 2CURRENTit corresponds to an individual that has purchased a house in the year of participating in the DHS in year t. 3AFTERit indicates in year t whether an individual has purchased a house a year after their participation in the DHS. It is the error term for individual i in year t.
Model 8 (OLS)
𝑆𝑖 = α + 𝛽1𝑖𝐵𝐸𝐹𝑂𝑅𝐸𝑖+ 𝛽2𝑖𝐶𝑈𝑅𝑅𝐸𝑁𝑇𝑖+ 𝛽3𝑖𝐴𝐹𝑇𝐸𝑅𝑖+ 𝛽4𝑖𝑈40𝑖+ 𝛽5𝑂60𝑖+ 𝜀𝑖
Model 8 is based on the OLS regression of model 3, although it has the percentage change of savings as the outcome variable. is the estimate of the intercept. 1BEFOREi is the dummy variable for individual i to indicate if respondents have purchased their house one year prior to the year of their participation in the survey. 2CURRENTi indicates if respondents have purchased their house in the same year as the year of the survey and 3AFTERi indicates if respondents have purchased a house one year after their participation in the survey. 4U40i and 5O60i are dummy variables to indicate whether an individual in the panel belongs to the generation under 40 years or over 60 years old. I is the error term for individual i.
Model 9 (OLS)
𝑆𝑖 = α + 𝛽1𝑖𝐴𝐹𝑇𝐸𝑅𝑈40𝑖+ 𝛽2𝑖𝐴𝐹𝑇𝐸𝑅𝑂60𝑖+ 𝜀𝑖 𝛽1𝐴𝐹𝑇𝐸𝑅𝑈40𝑖 = (𝛽n𝐴𝐹𝑇𝐸𝑅𝑖× 𝛽n𝑈40𝑖) 𝛽2𝐴𝐹𝑇𝐸𝑅𝑂60𝑖 = (𝛽n𝐴𝐹𝑇𝐸𝑅𝑖× 𝛽n𝑂60𝑖)
Model 9 takes a similar approach as model 4. Model 4 is used to focus on a general effect on consumption for individuals in either of the generations, who have bought a house one year after their participation in the survey in which Ci is the natural log consumption level, is the intercept and I is the error term. 1AFTERi is the dummy variable for the latter effect. nU40i and nO60i are the dummy variable for respondents born respectively after 1978 and before 1960.
Model 9 is focused on a possible effect on saving for individuals in either of the two generations. Given that they have purchased a house one year after participation in the survey, which is indicated by variable 1AFTERi. The dummy variable for individual i in the generation under 40 is specified as nU40i. The dummy variable for individual i in the generation over 60 is nO60i.. 1iAFTERU40i is the interacted term of variables nU40i and1AFTERi. On the other hand, the interaction of nO60i andnAFTERi leads to variable 2iAFTERO60i.The approach of interacting variables controls for omitted variable bias. Si is the change in savings in percentage, is the intercept and I is the error term.
Model 10 (FE) 𝑆𝑖𝑡 = α𝑖𝑡+ 𝛽1𝑖∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖𝑡+ 𝛽2𝑖∆𝑊𝑂𝑍𝑈40𝑖𝑡 + 𝛽3∆𝑊𝑂𝑍𝑂60𝑖𝑡 + 𝜀𝑖𝑡 𝛼𝑖 = 𝛼 + 𝑈𝑖 𝑛𝑖= 𝛽𝑛+ 𝛾𝑛𝑖 (n = 1 … 3) 𝛽1∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖 = (𝑛∆𝑊𝑂𝑍𝑖 × 𝑛𝐴𝐹𝑇𝐸𝑅𝑖) 𝛽2∆𝑊𝑂𝑍𝑈40𝑖 = (𝑛𝑈40𝑖 × 𝑛𝐴𝐹𝑇𝐸𝑅𝑖) × 1∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖 3∆𝑊𝑂𝑍𝑂60𝑖 = (𝑛𝑂60𝑖 × 𝑛𝐴𝐹𝑇𝐸𝑅𝑖) × 1∆𝑊𝑂𝑍𝐴𝐹𝑇𝐸𝑅𝑖
In the fixed effects regression of model 10, the calculation of variables 1∆WOZAFTERi, 2∆WOZU40i, and 3∆WOZO60i is done in the same way as described in model 5. This model is used to explore a possible collateral effect on household savings within individuals in the panel data either under 40 years or over 60 years old, after they have purchased a house. Si is the outcome variable, which is the natural log of a change in savings expressed as a percentage change. is the intercept and Ui is the random effect on the intercept for that individual. ni is the unknown coefficient for each independent variable. I is the error term for individual i in year t (Field, 2013).
6. Results
This paragraph explains the estimates from the empirical model, which is explained in paragraph 5.3. In order to formulate an answer to the research question and sub-questions, potential effects have been divided in an effect on the level of consumption and an effect on the level of savings. The natural log of consumption and savings is used to indicate a percentage change. Therefore the dependent variables are expressed as lnConsumption and lnSavings. Effects on lnConsumption are shown in model 1 to 5 in table 5, and the effects on lnSavings in model 6 to 10 in table 6. For models 2, 5, 8 and 10 we have used a fixed effects model to calculate estimates. The other models are based on an OLS regression.
In table 5 we observe that, in general, the annual household consumption increases in the year after the purchase of a house. Model 1 indicates a statistically significant and positive effect. The estimate suggests that consumption increases by 8,3 percent one year after households have purchased a house. The estimates for the other two variables are not statistically significant. However this does not rule out the presence of an effect. In model 2 we used a fixed effects model to observe individuals who have purchased a house in the years that we have available in the data. We can suggest that, in these years, individuals that have changed from ‘going to buy a house next year’ or ‘purchased a house that same year’ to ‘purchased a house last year’, have a significantly higher level of consumption. Using the fixed effects model, the consumption level increase is equal to 8,8 percent. The increase of the consumption level is relative to the consumption in the year before households were planning to buy a house, and relative to the consumption in the year that a house was purchased. However, a lower percentage change cannot be determined in the table, given the low statistical significance of the coefficients. This can be due to a low number of observations for individuals who have bought a house in the same year that they have participated in the survey (n = 138), and the number of individuals who have participated in the survey the year before they have bought a house (n = 39).
As we have observed in model 1 and 2, the household consumption in general increases in the year after buying a house. This relation is present in model 3 as well. The coefficient for the increase of the consumption level is statistically significant with 99,6%. In the research of Campbell and Cocco (2007), which is discussed in the theoretical framework, the relation between housing wealth and consumption is divided in the effect on older and younger homeowners. As discussed in the previous paragraph on research methodology this study takes a similar approach in model 3, by dividing respondents in two generations: 1, The generation born after 1978. 2. The generation before 1960. The estimate for young homeowners suggests a negative effect, indicating a decrease of the consumption level of 16,3 percent in the year before purchasing a house. Given the lack of statistical
significance of the estimate for old homeowners, the effect indicated by this coefficient is not conclusive, however it cannot be ruled out that there is no effect at all. Important to note is that the R-squared value of all models is quite low, and that this value in model 3 is higher than that of the other models. In this case, it means that the independent variables explain 0,37 percent of the variation in the dependent variable. A reason for the low values can be the low number of observations for individuals who have bought a house in the same year of their participation in the survey, or individuals who participated in the survey one year before they have bought a house. Model 4 is only focused on the level of consumption after having purchased a house. We can observe a clear indication that the level of consumption one year after the purchase of a house is significantly higher for old homeowners. In general, old homeowners increase their consumption levels with 20,5 percent. This coefficient is statistically significant at 1 percent level. For model 5, a similar focus is used. This model is concentrated on a possible relation between consumption and the change in WOZ-value one year after the purchase of a house. This effect is divided into the previously described generations of individuals under 40 and individuals over 60 years old. The coefficients indicate that there is an effect that young homeowners reduce their level of consumption more than old homeowners. Because the estimates are not statistically significant it is impossible to confirm that this effect is a result of a change in WOZ-value.
Table 5: Results for time-effects and wealth effects on consumption
lnConsumption I (OLS) II (FE) III (OLS) IV (OLS) V (FE) Purchase house (t = 0) -0,078 -0,068 -0,050 current year (0,076) (0,066) (0,077) Purchase house (t = 1) -0,198 -0,144 -0,184 next year (0,123) (0,998) (0,123) Purchased house (t = -1) last year 0,083** (0,0389) 0,088** (0,346) 0,112*** (0,040) Young homeowners -0,163*** (0,029) Old homeowners -0,022 (0,015)
Purchased house last year -0,054
young homeowners (0,071)
Purchased house last year 0,205***
old homeowners (0,079) ∆WOZ (t = -1) Last year 0,005 (0,008) ∆WOZ young homeowner -0,164 (0,010)
∆WOZ old homeowner -0,003
(0,009)
R-squared 0,0008 0,0008 0,0037 0,0007 0,0005
N 10.498 10.498 10.498 10.498 8.150
Table 6 gives an overview of the estimates from the different regressions of the independent variables on the percentage change of the level of savings. A positive relationship between time variables and the level of savings can be indicated in model 6. In general, individuals increase their level of savings with 57,1 percent in the same year that they purchase a house. Individuals who have purchased a house one year before they have participated in the DHS increase household savings by 43,2 percent. These positive effects are statistically significant at 1 percent level. This means that the probability that this value is derived by chance is equal to or smaller than 1 percent (Lacey, n.d.) By the use of a fixed effects model in model 7 we measure a possible effect on the saving behavior for individuals over time. This model focuses on households which have changed from ‘going to buy a house next year’ to ‘purchased a house that same year’ or to ‘purchased a house last year’ during years that they have completed the DHS. Similar to the estimates in model 6, we find that households have increased levels of savings in the years that they have purchased a house and one year after the purchase of a house. The statistically significant estimates for these effects are respectively 60,8 percent and 29,2 percent relative to other years. The estimate of the level of savings one year before the purchase of a house is negative, however not statistically significant. Important to notice is that the standard errors for this time variable are quite high throughout model 6-8, meaning that the values are less concentrated around the mean (Lacey, n.d.).
In order to see if possible changes in saving behavior in varying time periods are different across generations, we have incorporated variables for a generation of individuals born after 1978 and a generation born before 1960 in model 8. Similar to models 6 and 7, we find that buying a house is positively related to savings. Individuals who have purchased a house in the same year as their participation in the DHS, increase their savings by 50,9 percent. This effect is statistically significant at a level of 1 percent. Savings one year after the purchase of a house do not appear to decrease, since the statistically significant estimate for this effect is equal to 36,6 percent. Furthermore, the OLS regression in model 8 suggests that in general, young homeowners are more inclined to save than old homeowners. The coefficients for both generations are respectively 8,8 percent and -8,7 percent. If we concentrate on the level of savings for different generations one year after the purchase of a house, we observe the opposite effect. While young homeowners increase the level of savings by 43,1 percent, old homeowners increase this level by 60,6 percent relative to other years. An explanation for this difference between the two generations could be that old homeowners take a shorter period to pay off their mortgage, or the will to leave a bequest. This is elaborated on in the next paragraph. The fixed effects regression in model 10 explores a relationship between savings and the change in WOZ-value for individuals over time, who have purchased a house one year after their participation in the DHS. Evidence for an effect cannot be indicated nor dismissed based on the estimates in the model, because
the lack of statistical significance. Moreover, the R-squared value of 0,11 percent for model 10 is quite a low. This can be due to a low number of observations in the sample to which the variables apply.
Table 6: Results for time-effects and wealth effects on savings
lnSavings VI (OLS) VII (FE) VIII (OLS) IX (OLS) X (FE) Purchase house (t = 0) 0,571*** 0,608*** 0,509*** current year (0,152) (0,150) (0,153) Purchase house (t = 1) -0,175 -0,083 -0,203 next year (0,255) (0,216) (0,254) Purchased house (t = -1) last year 0,432*** (0,083) 0,292*** (0,084) 0,366*** (0,085) Young homeowners 0,087* (0,049) Old homeowners -0,088** (0,034)
Purchased house last year 0,431***
young homeowners (0,114)
Purchased house last year 0,606***
old homeowners (0,189) ∆WOZ (t = -1) Last year -0,017 (0,992) ∆WOZ young homeowner 0,027 (0,099)
∆WOZ old homeowner 0,000
(0,101)
R-squared 0,0089 0,0084 0,0124 0,0053 0,0011
N 4.563 4.563 4.563 4.563 2.751
