The relation between population aging and house prices

 Student number: 1741039, e-mail address: m.t.d.niezink@student.rug.nl I would like to thank my supervisor, Dr. L. Dam, for his helpful feedback.

The relation between population aging and house prices

Martine T.D. Niezink* MSc Finance Master’s thesis

Faculty of Economics and Business, University of Groningen

Supervisor: Dr. L. Dam

Date: January 10, 2014

Abstract

This study assesses the impact of population aging on house prices. Research is inconclusive about whether population aging puts downward or upward pressure on house prices. In addition, the influence of wealth within a country on this relation is studied. A macro level house price model is fitted to data of 23 OECD countries within differing time spans from 1971 to 2011. The main finding is that population aging puts downward pressure on house prices. The effect of country characteristics associated with wealth, on the relationship between population aging and house prices, suggest that wealth enhances the negative effect of population aging on house prices. However, these latter results are not significant in the model.

Keywords: population aging, house prices, OECD, wealth

1

1. Introduction

Purchasing a house is the largest single investment of most households. Therefore, the behavior of house prices has a big influence on household investments and thereby on housing expenses. Changes in house prices have large implications for household decisions, for example saving for retirement (Tsatsaronis and Zhu, 2004). Understanding how changes in house prices can be explained is of interest for financial planning of current and future households.

Population aging is one of the factors that might cause changes in house prices. This effect could increase rapidly the upcoming years because of the influence of the baby boom. Chen, Gibb, Wright and Leishman (2012) empirically address the question whether population aging is related to decreasing house prices. According to Marshall (1890), changes in prices and output of goods are determined by supply and demand, which can be represented by two curves. Where the demand and supply curve intersect, an equilibrium is established. As market conditions change, the curves shift and new equilibrium prices and outputs occur. Possibly, people sell their house as they age and find alternative living arrangements. Because supply of housing would increase, this would put downward pressure on house prices. The decision to move and choose for alternative living arrangements can be linked to the life cycle-permanent income theory. This theory describes the problem of dividing consumption between the present and the future. Consumers make estimates of their ability to consume in the long run and set current consumption to the appropriate fraction of that estimate (Hall, 1978). This estimate can be called wealth (Modigliani, 1971) or permanent income (Friedman, 1957). The level of wealth could influence housing decisions according to Toussaint, Quilgars, Jones and Elsinga (2011). People in the lower-priced segment of the housing market have fewer possibilities to cash in a significant amount of their housing wealth, while maintaining a reasonable standard of living.

This study uses panel ordinary least square (OLS) techniques in building a model to investigate the relation between population aging and house prices. Several analyses are performed to ensure robustness of the OLS results. Interaction effects are used to study the influence of a country’s wealth on the described relation.

2

main determinant of house prices. This study expands current literature by including a larger number of countries, which make the study more suitable for generalizations. Furthermore, exploring the effect of wealth on the relation between population aging and house prices is new in this line of research.

Data on 23 countries over differing time span from 1971-2011 are studied. All countries are member of the Organization for European Cooperation and Development (OECD) or in the transition phase of entering the OECD. The countries are separated into two main groups: nineteen developed OECD countries and four Central and Eastern European (CEE) transition countries. The OECD countries are divided in three sub groups: five large countries, fourteen small countries, and four catching-up OECD countries. In the models presented, the relation between population aging and house prices is studied including four control variables: GDP growth per capita (%), inflation CPI (%), long term interest rate (%) and population growth (%).

Results show a negative relation between population aging and house prices. So when countries age more rapidly because of the effect of the prior baby boom, a downward pressure on house prices can be expected. Furthermore, the effect of wealth within a country follows a trend. Wealthier countries experience a large effect of population aging on house prices. In the least wealthy Central and Eastern European (CCE) transition countries, population aging even makes house prices rise. However, these results are not significant.

2. Theoretical framework

2.1 Life cycle-permanent income theory

Housing is an important part of human consumption in life. Consumption behavior of individuals can be framed in the life cycle permanent income theory that describes the problem of dividing consumption between the present and the future. According to this theory, consumers make estimates of their ability to consume in the long run and set current consumption to the appropriate fraction of that estimate (Hall, 1978). This estimate can be called ‘wealth’ (Modigliani, 1971) or ‘permanent income’ (Friedman, 1957). Gianni (2007) combines the work of Modigliani (1966) and Friedman (1967) and comes to the following term: life cycle-permanent income model.

Deaton (1992) states that at any time in a consumer’s life span, optimal consumption is a constant fraction of discounted resources (human and financial) in current and future periods. This also includes housing resources. The life cycle-permanent income theory predicts that the baby boomers have saved for a proper retirement through building a pension or investing in financial or housing equity.

2.1.1 Financial strategies concerning housing decisions

3

countries1. There are three main ways in which house owners can profit from their housing wealth. First, when people become older, owner-occupiers become normally outright owners. This results in their housing cost being lower than that of tenants. In countries where mortgage markets play an important role in financing the purchase of a home, mortgage repayment usually ends before retirement. Therefore owner-occupiers will incur lower housing costs and they are aware of this. The financial advantage is even more pronounced in countries where rental markets cover a large part of the housing stock. Owner-occupiers in old age are therefore financially better off than tenants as result of the lower housing costs (Toussaint, Quilgars, Jones and Elsinga, 2011).

The second way involves the more recently development of mortgage-equity release. This type of mortgage enables people to continue living in their house while housing wealth is released. This option follows the line of reason of the life cycle-permanent income theory, as it would change housing wealth from an illiquid to a liquid asset. Releasing housing equity is however not popular by retirees. There are two main reasons for this. First, people want to leave equity as a bequest. Home-owners with children appear less open to mortgage-equity release in older age than people without children. This is most noticeable in countries where the financial future of young adults is uncertain. As the welfare system does not provide back-up, the dwelling as a bequest is an important family strategy. In countries where children have more financially secure prospect, owner-occupiers with children do not want to become a burden to their children. People in these countries see their housing equity as a safety-net for future expenses (e.g. for health care), and as a result they prefer to keep their housing equity intact. The second reason why financial equity-release products are not popular has to do with trust in the providers of these products. People are afraid of losing control, running risks and becoming dependent. Even before the financial crises people have not fully trusted financial intuitions and this amount of trust has only decreased (Toussaint, Quilgars, Jones and Elsinga, 2011).

The third way by which house owners can profit from their housing wealth is by selling their house and thus cashing their housing equity and subsequently renting or buying a less expensive home. The most common strategy involves moving to a less expensive dwelling. In all countries studied by Toussaint, Quilgars, Jones and Elsinga (2011) the decision to move was not only based on the need for additional income, but also on the need to reduce the maintenance cost, which elderly consider to be a financial and physical burden. However, moving to a less expensive home is not an option for all owner-occupiers. This is only an option for owner-occupiers with a relatively expensive property. People in the lower-priced segment of the market have less possibilities for cashing in a significant amount of their housing wealth while maintain a reasonable standard of living.

1

4

Only in the Netherlands renting a home is an acceptable alternative. In other countries mortgage-equity release is seen as the only realistic option for releasing their housing equity (Toussaint, Quilgars, Jones and Elsinga, 2011). Homes represent a buffer that can be cashed-in in difficult times. Housing equity is also used in financial strategies when pension income would be too low (Toussaint, Quilgars, Jones and Elsinga, 2011). Similarly, Venti and Wise (2004) found that families in the United States seem to have saved enough to maintain their preretirement standard of living after retirement. Housing equity is thereby rather not used to support non-housing consumption. They state that it may be appropriate to think of housing as a reserve that can be used in catastrophic circumstances that result in a change of household structure. Note that this goes beyond the purpose of saving housing equity for retirement. Therefore, in the United States, when no pressure is available, elderly rather stay in their homes and not move and no relation between population aging and house prices would exist.

In Europe, the housing equity generally is considered a substantial reserve that could be used to meet future pension needs (Doling and Ronald, 2010). This applies especially to elderly, as population aging becomes a greater stress on European welfare states. As stated before, it is appropriate to think of housing as a reserve that can be used in catastrophic circumstances (Venti and Wise, 2004). However the financial crisis was not taken into account when the baby boom generation was planning for their retirement. Lusardi and Mitchell (2007) have compared wealth holding across two age cohorts: the early baby boom generation in 2004 and the same age group in 1992. They show levels and pattern of total net worth have changed relatively little over time. The median person from the baby boom generation has more wealth than his precursor. The baby boom generation also relies more on housing equity as key component of retirement assets. This makes them vulnerable to fluctuation in the housing market. Toussaint, Quilgars, Jones and Elsinga (2011) found that people will adjust their financial plans when problems arise. It is expected that as pension incomes decrease and care becomes more expensive, housing wealth will become a more relevant source of wealth in the financial strategies of people. This is also suggested by Dohling and Ronald (2010), which states housing could be a compliment to, or substitute for tax-funded pensions provisions.

In various countries, especially in Slovenia, elderly spend their housing equity to pay for their care in a residential setting. In countries where children play a crucial role in caring for parents, Hungary and Portugal, some people without children used or planned to use housing equity to receive care by leaving the dwelling as a gift to a person within their family (often someone within the extended family), who would take care of them (Toussaint, Quilgars, Jones and Elsinga, 2010).

2.1.2 Non- financial strategies concerning housing decisions

5

pleasant climates (Litwak and Longino, 1987). These migrants are commonly more healthy, young and wealthy (Walters, 2002). Secondly, there are assistance moves, where people are motivated by the need for support and assistance either by the community. Thirdly, and as a subtype of the assistance moves, there are the moves to institutional settings. These assistance movers are taking proactive decision to respond to health and functional status changes that possibly affect safety and well-being in current homes. Having a home symbolizes autonomy for many people (Wiles, 2005) and maintaining this independence in living arrangements can be a motivation to move for elderly (Oswald, Schilling, Wahl, and Gäng, 2002; Oswald and Wahl, 2004).

2.2 Population aging and house prices

Aging is a global phenomenon. Developed economies are aging fast, but emerging economies are following them (Takáts, 2010). In countries that are member of the Organization for European Cooperation and Development (OECD), population aging is influenced by the large age cohort born in the baby boom (in the period 1946-64) and the smaller age cohort born in the baby bust that followed (in the period 1965-76) (Chen, Gibb, Wright and Leishman, 2012). This cohort could have a big influence on the housing market. Mankiw and Weil (1989) were the first to study the influence of the baby boom on the demand for housing and how these changes in demand have translated into changes in residential investment and price of housing. They studied the case of the United States in the 1970s. In this context real per capita housing expenditures were directly related to age and observed changes in demand for housing, caused by the baby boom. A substantial positive impact in the price of housing in the 1970s was found. They forecasted that when the baby boom generation ages, there will be a substantial negative impact on the price of housing. Even an assets price meltdown could occur. Big aging economies like Germany, Japan, China and the United States account for the majority of global investable assets. Since economic theory suggest that population aging can drive asset prices down, there could also be an influence on house prices. Mankiw and Weil (1989) used this economic reasoning to motivate their empirical research.

McFadden (1994) confirms the relation found by Mankiw and Weil (1989) for the 1940s, 1960s and 1980s. The shift in the period of 1970-85, when the baby boomers become over 23 years old, raised real per capita housing expenditures significantly. In this study it is expected that the shift of baby boomers becoming over 60 years old, will lower average real capita housing expenditures significantly. Both Mankiw and Weil (1989) and McFadden (1994) conclude that the cause of the 20% surge in real house prices in the 1969-1989 was the baby boomers entering the housing market and they expect a sharp decline in real house prices when they age.

6

Maniw and Weil (1989) for the 1980s. Also no impact of the forecasted declining real house price is expected.

Green and Hendershott (1996) have improved the research of Mankiw and Weil (1989) by separating expenditures into price and quantity components and by linking real prices of characteristics of numerous economic and demographic variables in addition to age. Willingness to pay for a constant-quality house varying only with age (partial derivative) and varying in a conjunction with other demographic and economic characteristics (a total derivative) are studied in his way. An analogue to Mankiw and Weil (1989) is found in that the willingness to pay for housing, of people in their late seventies, is 50% lower than that of people in their late fifties. This, however, did not have to do with age, but with the fact that 70 year olds in the 1980 had far less education and income than 50 year olds, caused by the surge in education after World War II. Mankiw and Weil (1989) even found a positive partial effect of age on housing demand. By only focusing on the total age derivative as a whole, the fact that the formal education of people in their fifties cannot decline and thus their real income is unlikely to fall significantly is ignored. Expectations for the following three decades predict changes in demographics have be modest in magnitude and not to have any significant influence on housing demand is expected and thereby real house prices.

The study of Mankiw and Weil (1989) suggest that the combined changes in housing demand caused by aging of the baby-boom and baby-bust generation could cause a decline in house prices in the 1990s. Levin, Montagnoli and Wright (2009) shows support of the hypothesis that population aging puts downward pressure on house prices real house prices increased substantially. They used a difference-in-difference method to study the effect of population decline and population aging in Scotland and England/Wales for 1968-2004, but not state supportive theories of their findings. Chen, Gibb, Leishman and Wright (2012) have found changes in age structure are not likely to be a main determinant of house prices. They used a micro simulation approach for Scotland, based on panel data from 1999-2008. Predictions were made until 2035. They present no theoretical support for their findings.

2.3 Economic foundation

Changes in price and output of a good are determined by supply and demand. Marshall (1890) describes this fundamental theory in economics, where a demand and a supply curve intersect in an equilibrium. This point determines the equilibrium price and output. The equilibrium is influenced by changes in supply and demand. They shift over time as market conditions change. Figure AI (Appendix A) shows how shifts in demand and supply curves cause an equilibrium to shift.

7 2.4 Developing hypotheses

There is no consensus on the relationship between population aging and house prices. Toussaint and Quilgars, Jones and Elsinga (2011) state that retirees see their housing equity as a buffer that can be cashed-in in difficult times. In the same line, Venti and Wise (2004) find housing equity to be a reserve that can be used in catastrophic circumstances that result in a change in household structure. This also suggests that no relationship between population aging and house prices exist in normal circumstances. However, Toussaint and Quilgars, Jones and Elsinga (2011) also state housing wealth is used to finance care in later life and housing equity is used as a reward for care takers (commonly close or extended family). In addition, non-financial motives, seeking for a more suitable environment for leisure or striving for independence late in life, are supportive of releasing housing equity by retirees (Litwak and Longino, 1987). These theories are in line with the life cycle-permanent income theory, since housing equity is made liquid. The life cycle-permanent income theory predicts the baby boomers will consume pension income, or investments in financial, or housing equity, to enjoy a proper retirement (Hall, 1978).

When elderly decide on selling their homes, this increases the supply of housing. According to the economic theory of Marshall (1890), the supply curve shifts right, which puts downward pressure on prices. So, when baby boomers age and population is aging, more houses come available on the market and house prices decline. The following hypothesis is formed:

H1: Population aging is negatively related to house prices.

Wealthy home owners are likely to have more options to move than less wealthy homeowners. This was already suggested by Toussaint, Quilgars, Jones and Elsinga (2011), who state that elderly with a less expensive home have smaller chance in selling their house and thereby releasing housing equity. Therefore a difference between the effect of aging of society in wealthier and less wealthy countries is expected. In wealthier countries more elderly are in the position to sell their house and find an alternative living arrangement. This increase in the supply of housing in wealthier countries is therefore expected to be more substantial than in less wealthy countries. When linked to the economic theory of Marshall (1890), the downward pressure on prices, as result of the more substantial increase of supply of housing, will be bigger in wealthier countries. The following hypothesis is formulated:

H2: Wealthier countries experience a larger influence of population aging on house prices.

8

3. Data and methodology 3.1 Data

3.1.1 Sample selection and data collection

The sample of this study is constructed following the study of Muellbauer and Murphy (2008), who model the main determinants of changes in house prices. Member countries of the OECD or countries in a transition phase of entering the OECD in 2007 are studied (Egert and Mihaljek, 2007). The sample consists of 23 countries2 covering differing time spans from 1971 to 2011 and comprises 341 measurements (see Appendix B, Table BI).

Data on house prices are retrieved from the website of the Bank for International Settlements (BIS). To create a sample that represents the whole housing market best, a four-step selection procedure is used to obtain house price data. First, the data was selected based on “property type”. Priority was given to: 1) all types of dwelling, followed by 2) single family housing, 3) single family houses- terraced and 4) flats. An exception to this procedure is Japan, where data on land for residential purposes are used to measure movements in house prices. The second selection criterion is: the “covered area”, where priority was given to: 1) whole country, 2) big cities, 3) capital cities and suburbs and 4) capital city. Third, the “priced unit” was preferably: 1) price per square meter, 2) the price per dwelling, 3) per unit and 4) per cubic meter. Lastly, data on “existing dwellings” was preferred over data on “new dwellings”, since the previous type covers more dwellings.

Data on age cohorts are retrieved from the website of the OECD iLibrary. All cohorts, that include people of 65 years and older, are combined. In this study this group is called the elderly. “Elderly” is measured by the percentage of the population that is 65 years old or older. Data for “GDP per capita”, “inflation”, measured as Consumer Price Index (CPI of all items), “long term interest rate” and “population totals” are also retrieved from OECD iLibrary.

To transform the stationary index data into non-stationary data, that can be used to perform a reliable regression analyses, growth rates of the variables: elderly, the house price index, GDP per capita, and total population numbers, are used. Furthermore, elderly growth is centered around a new point to decrease influences of multicollinearity. I will elaborate on this later.

3.1.2 Data description

Table 1 presents means and standard deviations of all 23 OECD countries for all variables used in this study: the dependent variable, house price growth (%) and the independent variables: centered elderly growth (%), GDP growth per capita (%), inflation CPI (%), long term interest rate (%) and total population growth (%). Notable is that

2

9

average house prices are generally increasing, except in Hungary, Ireland, Japan, and Slovenia. Average GDP per capita is increasing for all countries and lies between 0.29% in Slovenia and 6.84% in Poland. The inflation rate is on average 2.78% and the long term interest rate is on average 6.30%. On average populations are growing in all countries, except in Hungary and Poland.

For good understanding of the data on population aging the variable ‘elderly growth’ (%) is described instead of the variable used in the model, ‘centered elderly growth’. The average percentage growth in the percentage elderly within the OECD countries used in this study is 0.01%. All average elderly growth percentages lie between -0.001% (Austria) and 0.033% (Japan) (see Appendix B, Table BII). Austria is the only country in which on average the percentage elderly is declining. In all other countries the percentage elderly is on average increasing and this increase is most rapidly in Japan. Figure 1 shows the average percentage of elderly growth within the sample countries. The figure shows that the populations are aging. The percentage of elderly is growing for all years, except for the 1980’s. The average percentage elderly is declining in those years. The trends of the elderly growth in the 23 individual countries are comparable, no systematic differences between countries appear to be present3.

Figure 2 gives a graphical overview of average house price in the countries studied over time. The house price index shows a positive trend until 2007, when house prices stabilize and thereafter sharply drop in 2010.

Figure 1 Average elderly growth (%) for

23 countries from 1960 to 2011

This figure shows the average percentage of elderly in the total population of the 23 countries included in this study from 1960 to 2011.

Figure 2 Average house price index for 23

countries from 1960 to 2011

This figure shows the average house price index of 23 countries included in this study from 1960 to 2011.

3

Figures are available on demand.

10

Table 1 Descriptive statistics

This table presents the mean and standard deviation of all continuous variables used in the model: house price growth (%), centered elderly growth (%), GDP growth per capita (%), inflation CPI (%), long term interest rate (%), and population growth (%).

Country House price

growth (%) Centered elderly growth (%) GDP growth (%) Inflation CPI (%) Long term interest rate (%) Population growth (%)

Mean St.dev. Mean St.dev. Mean St.dev. Mean St.dev. Mean St.dev. Mean St.dev.

11 3.2 Methodology

In this study 23 countries at separate points in time within each country are analyzed. These data have a multilevel structure. The number of countries is denoted C; the total number of years in each country is denoted tj for country j (j = 1, 2,..., C). The total

number of country-year combinations studied is denoted by Y = Ʃjtj;

3.2.1 Grouping countries by level of wealth

To study the effect of country characteristics associated with wealth on the relation between population aging and house prices, the 23 countries are categorized. Based on the study of Egert and Mihaljek (2007) the countries are grouped into two main groups: nineteen developed non-transition OECD countries and four Central and Eastern Europe (CEE) transition countries (Czech Republic, Hungary, Poland and Slovenia). In the past years, the specified transition countries have joined the OECD, but this study follows the initial grouping of the countries and thereby keeps the group of Central and Eastern Europe (CEE) transition countries separate. A further split up of the developed non-transition OECD countries is made based on size of the economy and growth rates of GDP per country, which is taken as the measure of wealth within a country. Three subgroups are created, with: five large countries, fourteen small countries, and four catching-up OECD countries4. Large OECD non-transition countries have the largest sized economies and growth rates of GDP per country, followed by small OECD non-transition countries, and thereafter catching-up OECD countries. CEE non-transition countries have the smallest sized economies and GDP growth rates. The sub-panels large and small OECD countries are mutually exclusive. Small and catching-up countries overlap.

Three dummies are created to form the four groups: Developed OECD countriesj

taking value 1 if country j is a developed non-transition OECD country and 0 if country j is a CEE transition country, Small OECD countriesj taking value 1 if it is a small country

and 0 if it is a large country and Catching-up OECD countriesj taking value 1 if it is a

catching-up country and value 0 if it is not. The groups and the matching dummy combination are presented in Table 2.

4

12

Table 2 Groups and related dummy variables

This table shows the dummy combinations for every group of countries: Central and Eastern European (CEE) transition countries, large OECD countries, small OECD countries and catching-up OECD countries. Three dummies used to produce these groups indicate whether the groups are: developed OECD countries, small OECD countries and/or catching-up OECD countries.

Groups \ Dummy variables Developed OECD countries

Small OECD countries

Catching-up OECD countries

CEE transition countries 0 0 0

Large OECD countries 1 0 0

Small OECD countries 1 1 0

Catching-up OECD countries 1 1 1

3.2.2 Random effects models

The relation between population aging and house prices will be studied using a random effects model and parameters are estimates using the ordinary least squares method.

A random effects model consists of a fixed part, the regression coefficients, and a random part, consisting of a random effect at the group level and a random effect at the individual level. When effects of group level variables are tested, as this study does, the random effects model should be used. In fixed effects models all differences between countries would already be explained by the fixed effects and there would be no explainable between-group variability left (Snijders and Bosker, 1999).

In order for the random effects model to be reliable, four assumptions need to be considered. Note that these assumptions also apply to ordinary linear regression models. These models are a special case of the random effects models, where the amount of randomness on group level is zero. The first assumption states that all continuous independent variables should have a linear relation with the dependent variable. Second, all error terms should be independent; no auto-correlation should exist. Third, there should be constant variance of the error term, homoskedastic variance. Fourth, the error term should have a normal distribution (Snijders and Bosker, 1999).

3.2.3 Model specification

Hereafter, three models are presented. First, the baseline model is shown. This model consists of all direct effects of covariates, excluding any effect of country characteristics per groups. The baseline model is described in equation 1:

HPGij = β0 + β1 EldGCij + β2 GDPij + β3 CPIij + β4 LTIij + β5 TPoPGij + εij (1) where, HPGij represents the percentage growth in the house price index, EldGCij

represents the centered percentage growth in the elderly rate, GDPGij represents the

percentage growth in the GDP (per capita), CPIij represents consumer price index (all

13

population growth, and εij represents the error term, for country i at year j.

To decrease influences of multicollinearity, the average elderly rate is subtracted from every data point per country. The variable “centered elderly growth” is used in the model to estimate the effect of change in the percentage elderly on house prices.

The second model considers all direct effect including the effect of country characteristics per group. This model is described in equation 2:

HPGij = β0 + β1 EldGCij + β2 GDPij + β3 CPIij + β4 LTIij + β5 TPoPGij + β6 Developed OECD countriesj + β7 Small OECD countriesj + β8 Catching-up OECD countriesj + εij (2) The description of the dummy variables in this model can be found in Section 3.2.1. The third model, the interaction model, includes all direct effects and in addition it includes the interaction effects. This model is shown in equation 3:

HPGij = β0 + β1 EldGCij + β2 GDPij + β3 CPIij + β4 LTIij + β5 TPoPGij + β6 Developed OECD countriesj + β7 Small OECD countriesj + β8 Catching-up OECD countriesj + β9 (EldGCij * Developed

OECD countriesj) + β10 (EldGCij * Small OECD countriesj) + β11 (EldGCij * Catching-up OECD

countriesj) + εij (3)

non-transition countries have compared to CEE non-transition countries, (EldGCij × Small OECD

countries) is the extra effect small developed non-transition countries have compared to large developed non-transition countries and (EldGCij × Catching-up OECD countries) is

the extra effect catching-up developed non-transition countries have compared to small developed non-transition countries.

3.3 Outlier detection

Outliers in the sample are detected using a method of Hadinvo van Hadi (1992, 1994). Normally, outliers are determined based on the most comprehensive model estimated. In this study this would be the model including the interaction effects. However, the interaction effects create collinearity and make the proposed method unsuitable. Yet, the occurrence of collinearity is expected when adding interaction effects. Since no new data are added to the baseline model and data are only re-categorised, this model will suffice for detecting outliers. Outliers are excluded on a 1% significance level. Twenty-four observations are excluded from the sample.

4. Results

4.1 Checking regression assumptions

14

random coefficients model includes the extra assumption of independent and identically distributed group effects. However, both are included in the error term and this assumption is therefore in line with the regression assumption mentioned and met. White's test for heteroskedasticity is based on H0 that the error terms variance is constant. The White test is significant. This indicates there is that residual variance is not constant, heteroskedastic (see Appendix C, Table CII). Fourth, the error term should have a normal distribution. For all three models tested, the Skewness/Kurtosis tests for normality is supporting H1 of non-normality (χ2(2) < 0.05), as is shown in Table CIII (Appendix C).

This indicates the error term is not normally distributed.

In sum, several but not all assumptions of the random effects model are satisfied, so the results of the analyses should be interpreted with care.

4.2 Regression estimations

The Hausman test, with H0 stating difference in coefficients not to be systematic, is accepted (χ2

(5) = 11.93, p < 0.05). Despite this finding a random effects model is used.

When fixed effects would be applied, problems would arise with explaining between group variability, which is relevant in determining the effect of country characteristics associated with wealth.

The result related to the all OLS estimations are displayed in Table 3. OLS standard errors are unbiased in case of an identically and independently distributed error term. They may be biased if this is not the case. In the previous section, White's test for heteroskedasticity shows the residual variance is not constant. The Wooldrich test for autocorrelation within the error term also shows strong evidence for first order autocorrelation. Therefore, the parameters in all models are estimated using cluster robust standard errors, such that hetroskedasity and autocorrelation are taken into account.

First, I will solely focus on the control variables used in this study. In the baseline model, effects are directed as expected. GDP growth per capita (%) and population growth are positively associated with house price growth. Both relations are highly significant. The relation between both control variables: inflation CPI (%), long term interest rate (%) and house price growth are directed as expected, but are not significant. The results discussed above remain valid in all models presented hereafter. Only the effect of the long term interest rate (%) starts of to be an insignificant effect in the baseline model, but when the country effects and the interaction effects are added to the model, the size of this effect increases a little and becomes significant.

15

Table 3 Random effects models studying the relation between population aging and

house prices

This table presents three models. First the baseline model is presented. In this model the following variables are included to estimate house price growth: centered elderly growth (%) GDP growth (%), inflation CPI (%), long term interest rate (%) and population growth (%). In the second model presented three dummies are added to include country effects of the four country groups researched in this study. The last model presented also includes interaction effects.

Variable Baseline model + country effects + interaction effects

Elderly growth, centered (%)

-92.493* -88.590* 32.362

(41.738) (42.022) (186.842)

GDP growth per capita (%) 1.034** 1.028** 1.045**

(0.110) (0.110) (0.112)

Inflation CPI (%) 0.194 0.267 0.391

(0.231) (0.236) (0.239)

Long term interest rate (%) -0.233 -0.305 -0.316

(0.165) (0.171) (0.174)

Population growth (%) 2.759** 2.543** 2.107**

(0.853) (0.897) (0.807)

Developed OECD countries 0.983 1.277

(1 = yes, 0 = no) _{(2.020) (1.812)}

Small OECD countries 0.984 0.877

(1 = yes, 0 = no) _{(1.128) (0.831)}

Catching-up OECD countries -1.860 -1.548

(1 = yes, 0 = no) _{(1.402) (1.154)}

Interaction centered elderly -281.465

growth & developed OECD c. _(205.578)

Interaction centered elderly 158.061

growth & small OECD _(104.649)

Interaction centered elderly 59.154

growth & catching-up OECD (118.185)

Constant -0.618 -1.535 -1.823

(1.075) (1.931) (1.808)

N 431 431 431

Overall R-squared 0.216 0.221 0.230

Adjusted R-squared 0.207 0.206 0.210

16

Findings concerning the relation between population aging and house price growth (%) are graphically presented in Figure 3. The country characteristics of the groups specified determine the intercept of the lines. All parameters in this model are positive. Thus, the intercepts of all groups that are member of the OECD, lie higher than that of CEE transition countries. The parameter for large OECD countries is 0.983, for small OECD countries 1.966 and for catching-up OECD countries it is 0.106. However, none of these differences between groups are significant.

When the found parameters are combined into Figure 3, the intercept of small countries is the highest at 5.280, thereafter large countries are intercepting at 4.403, catching-up countries intercept at 3.732 and Central and Eastern European (CEE) countries intercept at 3.126. The interaction effects presented in Table 3 can be combined to derive the slope of the relation between population aging and house price growth for each group of countries. The slope of the effect of large OECD countries is -249.103, of the small OECD countries it is -91.042, of the catching-up countries it is -31.887 and of the CEE transition countries the slope is 32.362. Wealthier countries seem to have a steeper negative slope than less wealthy countries. CCE transition countries have a positive slope.

Whether the model is improved by adding the country characteristics of the groups and interaction effects to the baseline model is tested with a partial test. The partial F-test is neither significant when country effects are added to the base line model (χ2(3) = 2.48, p > 0.05), nor when infection effects are added and the model is completed (χ2(3) = 3.95, p > 0.05). The model including the country effects and the model including the interaction effects are no improvement of the baseline model.

Figure 3 Relation between population aging and house price growth

This graph shows the relationship between population aging, measured as the centered elderly growth (%), and house price growth (%). The only varying parameter is centered elderly growth. The averages of the control variables are taken as a constant.

-10 -5 0 5 10 15 20 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 Ho us e pr ic e gr o wt h ( % )

Centered elderly growth(%)

Relation population aging and house price growth

Large OECD countries Small OECD countries Catching-up OECD countries

17 4.3 Robustness checks

In multiple steps a model with only main effects, excluding the country effects, is constructed, by adding the control variables: GDP per capita (%), inflation CPI (%), long term interest rate (%) and population growth (%) to the relation between elderly growth (%) and house price growth (%). Five equations are presented in Table DI (Appendix D). Population aging has a significant negative relation with house prices in all five equations. Furthermore, the coefficient remains stable within all equations presented. In sum, when controlling for relevant influences, the relation between population aging and house prices remains significant and stable.

An alternative measure for population aging: the centered percentage of elderly within a country, is used in constructing a model including interaction effects. The relation between house price growth and population aging appears to stay negative, but is no longer significant (see Appendix D, Table DII). The interaction effect of small countries and population aging is significant in this model. So aging in small countries, which are less wealthy than larger countries, has a smaller negative influence on house prices in this model. The signs of the control variables are directed as expected.

5. Conclusion

In this study the impact of population aging on house prices is explored by examining macro-economic relationships in 23 countries over different time spans from 1971 to 2011. In addition, the effect of wealth within a country on this relation is studied.

Results show that population aging puts downward pressure on house prices. This supports hypothesis 1, stating that population aging is negatively related to house prices. This result is in line with the economic theory, an increase in housing supply, because elderly choose for alternative living arrangements, decreases house prices (Marshall, 1890). Furthermore it confirms initial predictions of Mankiw and Weil (1989) and McFadden (1994) and findings of Levin, Montagnoli and Wright (2009), who found that population aging puts downward pressure on house prices in Scotland and England/Wales in the period 1968-2004. The results contradict the study of Green and Hendershott (1996), who state changes in demographics will be modest and will not have any significant influence on housing demand and thereby real house prices. Moreover the finding contradict Chen, Gibb, Leishman and Wright (2012), who found changes in age structure are not likely to be a main determinant of house prices.

18 prices”, is found.

There are several limitations to this study that need to be mentioned. Since not all regression assumptions are met: error terms are not independent, variance of the error term is heteroskedastic and the error term is not normally distributed, results should be interpreted with care. However, the effects found are strong and consistent, so it is reasonable to assume the relation exists. There are also limitations with regard to the data. When comparing measures of house prices challenges are faced since the measure of housing is heterogeneous. This makes the house price measure inconsistent and less comparable. Furthermore, effort was made to find sufficient measurements for the different country categories. However, the categorization resulted in groups of differing sizes, with for example fourteen developed OECD countries and four CEE transition countries. The low number of observations for some groups result in less power for observing relations in this part of the model, which may be the reason why no significant effect was observed here. This should be taken into account in future research.

Results show when countries age more rapidly, because of the effect of the prior baby boom, a downward pressure on house prices can be expected. However, the data used covers a time span which only partly includes the financial crises which started in 2007. The conclusions and future house price trends could be influenced by the way elderly are affected during the crisis. Housing equity could become a compliment to, or substitute for, tax-funded pension provisions (Dohling and Ronald, 2010). Besides, it is expected that as pension incomes decrease and care becomes more expensive, housing equity will become more relevant as a source of wealth in the financial strategies of people (Toussaint, Quilgars, Jones and Elsinga, 2011). This should be taken into account when expectations are made about future house price trends. Further research should consider the effect of the financial crisis on financial strategies of current and future elderly.

19

Appendix A

Figure AI Supply and demand curves (Marshall, 1890)

This figure shows two demand and supply curves. The equilibrium price and quantity is established in the point the curves intersect. Market condition can change this equilibrium point through the supply and/or the demand side. When supply increases, the supply curve shifts right and the market clears at a lower price and a larger quantity (equilibrium of demand and supply new curve). When demand increases, the demand curve shifts right, the market clears at a higher price and a larger quantity (equilibrium of demand new and supply curve).

P

ri

ce

Quantity

New equilibrium following shifts in supply and demand

20

Appendix B

Table BI Data coverage per country in years

This table presents an overview of the years covered by the dependent and independent variables retrieved: house price growth (%), centered elderly growth (%), total population growth (%), GDP growth per capita (%), inflation CPI (%) and long term interest rate, for all 23 OECD countries used in this study. “Min.” and “Max.”, indicate the first and last year covered. “Range year (N)” indicates the number of years for which all variables used in this study are covered.

21

Appendix B (continued)

Table BII Descriptive statistics of growth of percentage elderly (%), per country This table shows the mean, standard deviation, minimum and maximum values and skewness and kurtosis of growth of the percentage of elderly for the 23 countries used in this study.

Country Mean St. dev. Min. Max. Skewness Kurtosis

22

Appendix C

Figure CI Scatter plots checking for linearity

Scatter plots showing relation between continuous independent variables: population aging, measured by the centered elderly growth (%) , GDP growth per capita (%), inflation CPI (%), long term interest rate (%). population growth (%) and the dependent variable and house price growth (%).

-2 0 0 20 40 Ho u se p ri c e g ro wth ( % ) -1 0 1 2 3 Population growth (%) -2 0 0 20 40 Ho u se p ri c e g ro wth ( % ) -.04 -.02 0 .02 .04

Centered elderly growth (%)

-2 0 0 20 40 Ho u se p ri c e g ro wth ( % ) -5 0 5 10 15 Inflation CPI (%) -2 0 0 20 40 Ho u se p ri c e g ro wth ( % ) 0 5 10 15

Long term interest rate (%)

-2 0 0 20 40 Ho u se p ri c e g ro wth ( % ) -10 -5 0 5 10 15

23

Appendix C (continued)

Table BII White's test for heteroskedasticity

The White's test for heteroskedasticity tests H0 of homoskedasity. The test statistic shows a significant result and the alterative hypothesis H1 is supported. This indicates there is that residual variance is not constant.

χ2

(53) = 130.41

Prob > χ2

= 0.0000

Table BIII Skewness/Kurtosis tests for normality

The Skewness/Kurtosis tests for normality tests H0 of normality. The test statistic shows a significant result and the alterative hypothesis H1 is supported. This indicates there is that the error term is not normally distributed.

Model 0: Prob > χ2 = 0.0000 Model 1: Prob > χ2 = 0.0000 Model 1: Prob > χ2 = 0.0000

Table CI Wooldridge test for autocorrelation in panel data The Wooldridge test for autocorrelation tests H0 of no first-order autocorrelation. The test statistic shows a significant result and therefore H1, which indicates there is first order autocorrelation, is supported.

F(1,22) = 77.996

24

Appendix D

Table DI Panel regressions testing robustness of variable: Elderly growth (%)

This table presents five equations in which the control variables are added to the regression between House prices growth (%) and elderly growth (%) in the following order: GDP growth (%), inflation CPI (%), long term interest rate (%), population growth (%).

Eq.(1) Eq.(2) Eq.(3) Eq.(4) Eq.(5)

Elderly growth (%) -124.788* -112.000* -110.322* -110.034* -121.34**

(46.189) (41.719) (41.874) (41.844) (36.313)

GDP growth per capita (%) 0.100** 0.992** 1.006** 1.022**

(0.107) (0.111) (0.111) (0.110)

Inflation CPI (%) 0.031 0.270 0.240

(0.178) (0.232) (0.229)

Long term interest rate (%) -0.272 -0.295

(0.167) (0.165) Population growth (%) 1.940* (0.851) Constant 4.419** 0.247 0.184 1.141 1.536 (0.599) (0.633) (0.739) (0.943) (1.237) N(countries) 23 23 23 23 23 N 431 431 431 431 431 Overall R-squared Adj. R-squared 0.0268 0.1951 0.1955 0.1966 0.2339

25

Appendix D (continued)

Table DII Random effects models studying the relation between population aging and

house prices, using the centered percentage of elderly within a country

This table presents multiple equations adding control variables to the regression between house prices growth (%) and the centered percentage elderly (%) in the following order: GDP growth (%), inflation CPI (%), long term interest rate (%), population growth (%).

Variable Baseline model + country effects + interaction effects

Elderly centered (%) -0.110 -0.048 -0.789

(0.237) (0.246) (1.499)

GDP growth per capita (%) 1.004** 1.040** 1.013**

(0.111) (0.111) (0.113)

Inflation CPI (%) -.276 0.228 0.237

(0.233) (0.238) (0.240)

Long term interest rate (%) -0.276 -0.321 -0.340

(0.184) (0.188) (0.184)

Population growth (%) 2.983* 2.860* 1.850*

(1.665) (1.022) (0.836)

Developed OECD countries 0.799 0.705

(1 = yes, 0 = no) (2.165 (5.049)

Small OECD countries -1.929 -0.855

(1 = yes, 0 = no) _{(1.597) (1.064)}

Catching-up OECD countries -1.929 -4.611

(1 = yes, 0 = no) _{(2.384) (2.856)}

Interaction elderly centered -0.098

& developed OECD countries _(1.525)

Interaction elderly centered 1.013

& small OECD countries _(0.364)*

Interaction elderly centered 0.749

& catching-up OECD countries _(0.767)

Constant -0.116 -1.313 1.404 (1.665) (2.384) (5.061) N 431 431 431 Countries 23 23 23 Overall R-squared 0.203 0.207 0.231 Adjusted R-squared 0.193 0.192 0.211

26

References

Blake, Kevin S., and Aleksandra Simic, 2005, Elderly housing consumption: Historical patterns and projected trends, Washington, DC: Department of Housing and Urban Development.

Chen, Yu, Kenneth Gibb, Robert Wright and Chris Leishman, 2012, The impact of population ageing on house prices: a micro-simulation approach, Scottish Journal of Political Economy 59, 523-542.

Deaton, Angus, 1992, Understanding consumption, Oxford: Oxford University Press. Doling, John, and Richard Ronald, 2010, Property-based welfare and European

homeowners: how would housing perform as a pension?, Journal of Housing and the Built Environment 25, 227-241.

Friedman, Milton A., 1957, Theory of Consumption Function, Princeton, NJ: Princeton University Press.

Green, Richard K., and Patric H. Hendershott, 1996, Age, demographics and real house prices, Regional Science and Urban Economics 26, 465-80.

Hall, Robert E., 1978, Stochastic implication of the life cycle-permanent income hypothesis: theory and evidence, Journal of Political Economy 86, 971-987. Hadi, Ali S., 1992, Identifying multiple outliers in multivariate data, Journal of the Royal

Statistics Society 54, 761-771.

Hadi, Ali S., 1994, A modification of a method for detection of outliers in multivariate samples, Journal of the Royal Statistics Society 56, 393-396.

Levin, Eric J., Alberto Montagnoli and Robert E. Wright, 2009, Demographic Change and the Housing Market, Urban Studies 46, 27-43.

Litwak, Eugenee and Charles. F. Longino, 1987, Migration patterns among the elderly: A developmental perspective, The Gerontologist 27, 266-272.

Lusardi, Annamaria, and Olivia S. Mitchell, 2007, Baby Boomer retirement security: the role of financial literacy and housing wealth, Journal of Monetary Economics 54, 205-214.

Mankiw, Gregory N. and David N. Weil, 1989, The baby boom , the baby bust, and the housing market, Regional Science and Urban Economics 19, 235-258.

Marshall, Alfred, 1890, Principles of Political Economy.On a tenté récemment une démonstration peu convaincante pour prouver que la problématique de lévolution économique n'était pas étrangère à Marshall, (Glassburner B., 1955, Alfred Marshall on Economic History and Historical Development, Quarterly Journal of Economies 69, 577-595.)

McFadden, Daniel L., 1994, Demographics, the housing market, and the welfare of the elderly. In David A. Wise, Studies in the economics of aging, 225-285, Chicago: The University of Chicago Press and NBER.

27

Modigliani, Franco, 1971, Monetary Policy and Consumption, Consumer Spending and Monetary Policy: The Linkages, Boston: Federal Reserve Bank of Boston, 9-84. MuellBauer, John, and Anthony Murphy, 2008, Housing markets and the economy: the

assessment, Oxford Review of Monetary Policy 24, 1-33.

Oswald, Frank, and Graham. D. Rowles, 2006, Beyond the relocation trauma in old age: new trends in today’s elders’ residential decisions, New Dynamics in Old Age: Environmental and Societal Perspectives, 127-152.

Oswald, Frank, Oliver Schilling, , Hans, W. Wahl, and Karin Gäng, 2002, Trouble in paradise? Reasons to relocate and objective environmental changes among well-off older adults, Journal of Environmental Psychology 22, 272-288.

Oswald, Frank, and Hans W. Wahl, 2004, Housing and health in later life, Reviews on Environmental Health 19, 223-252.

Peek, Joe and James A. Wilcox, 1991, The measurement and determinants of single- family house prices, Real Estate Economics 19, 353-382.

Poterba, James M., David N. Weil and Robert Shiller, 1991, House price dynamics: The role of tax policy and demography, Brookings Papers on Economic Activity 2, 143-183.

Snijders, Tom and Roel Bosker, 1999, Multilevel analysis: an introduction to basic and advanced multilevel modeling, London: Sage Publications.

Takáts, Előd , 2010, Ageing and asset prices, Basel: Bank for International Settlements, Monetary and Economic department.

Toussaint, Janneke, Deborah Quilgars, and Anwen Jones, 2011, Comparing eight countries: The role of housing equity in retirement. In Housing wealth in

retirement strategies: Towards understanding and new hypotheses, 105-130,

Amsterdam: IOS Press.

Tsatsaronis, Kontas, Haibin Zhu, 2004, What drives house prices dynamics: cross country evidence, Bank of International Settlements Quarterly Review. Venti, Steven, F., and David, A Wise, 2004, Aging and housing equity: another look,

Perspectives on the economics of aging, University of Chicago Press 127-180 Walters, William H., 2002, Later-life migration in the United States: A review of recent