UMEÅ UNIVERSITY
Department of Geography & Economic History
Spatial Planning & Development 2016
HOMEOWNERSHIP
&
UNEMPLOYMENT
A test of the Oswald Hypothesis in Sweden
Student
Oskar Bergkvist
Supervisor
Erika Sandow
Umeå 10’th of June 2016Acknowledgements
Many thanks to my superior supervisor Erika Sandow for her patience, dedication, good discussions and motivating attitude. Furthermore I want to thank my fellow students at the 1zero research center. Binary jokes doesn’t generally grow on trees but in the 1zero research center, they do… I also want to thank Umeå and Umeå University for peace and quiet and a motivating study environment.Abstract
The importance of a wellfunctioning housing market has been proposed for long within economics, economic geography and urban planning. A high mobility on the housing market most likely positively affects the dynamics of the labor market, a dynamic important for economic growth. Mobility defined as the link between the worker and the workplace in terms of transportation and housing are most likely essential components of a dynamic and wellfunctioning labor market. The Oswald hypothesis states that positive relationship between homeownership and unemployment exists, the lower mobility in the homeownership housing stock compared to the rental housing stock affects labor market mobility in a negative way which can be noted if European countries are compared. My thesis explores this relationship in a Swedish context by mobilizing a quantitative approach with aggregate data on municipal level ranging from 1998 to 2013. The Swedish housing market is in a deregulation process since 1992, a conversion process from public rental housing to homeownership coop apartments has taken place and public policies now favor homeownership over renting. Municipal data on unemployment, homeownership of apartment, rental tenant and control variables for economy and personal characteristics are applied in Pooled OLS, random effects and fixed effects regression models. The results from the Pooled OLS and the Random effects model confirms the positive relationship proposed by Oswald for homeownership of apartment but not for homeownership of detached housing. Also rental tenant show a positive relationship. The results from the fixed effect estimation rejects the hypothesis altogether and show a negative relationship.
Keywords:
Homeownership, Unemployment, Labor market, Oswald Hypothesis, Mobility, Swedish housing market
Contents
Acknowledgements 3 1 Introduction 4 1.2 Aim 7 2 Theory 7 2.1 The Oswald hypothesis 7 2.2 Earlier research and tests of the hypothesis 9 2.2.1 Research development 9 2.2.2 Support for the Hypothesis 11 2.2.3 Rejections of the Oswald hypothesis 15 2.2.4 The Swedish housing regime 20 3 Method 23 3.1 Data 24 Unemployment rate, dependent variable 24 Education 24 Homeownership, tenant status 25 Economics, GRP (Gross Regional Product) 26 Commuters 27 Sociodemographic forces and population 27 3.2 Regression analysis 28 3.2.1 Pooled OLS, Random and Fixed effects regression models 28 3.2.2 Dummie variables 29 3.2.3 Model diagnostics 29 3.3 Source criticism 30 4 Results 314.1 Descriptive statistics 31 4.1.1 Box plots of variables 33 4.2 Regression analysis 37 5 Discussion & Conclusions 39 5.1 Results 39 5.2 The Swedish context 40 5.3 Critique 43 6 References 45 Internet Sources 54 Appendix 1. SKL Municipality Groups 55 Appendix 2. 56 Classification of Swedish municipalities, 2011 56
1 Introduction
Few will argue that the relatively high unemployment levels experienced in Europe and other parts of the world the last decades is not a problem, politicians’ debate over the best solution daily. Unemployment bares many unwanted economic and social consequences, both on a national and individual level. The research on the topic has been vast, even though a century of research has not been able to fully understand the dynamics of unemployment. Besides grand macroeconomic theories, research has historically been focused on the relationship between unemployment and labor market characteristics, typically trade unionism, labor law and unemployment benefits (Blanchflower Oswald 2013). One of the most influential works in the field is Milton Friedman’s “The role of monetary policy” (1968) where he besides developing the grand macroeconomic ideas of earlier works by William Phillips (1958) and Samuelson and Solow (1960) and others he also propose cost of mobility as an important determinant of the functioning of the labor market. Mobility costs can be interpreted as the spatial relationship between worker and workplace, in terms of economic and time expenses, depending on mobility possibilities, mainly provided by infrastructure and housing (Friedman 1968).
In this view an important part of urban planning is to evaluate, plan and provide mobility possibilities. The functioning of the housing market is one aspect of the important spatial relationship between worker and workplace. Mobility possibilities and the general function of the housing market is therefore also related to economic performance. Regional housing policies affects regional economic growth according to the paper “Housing and the economy: policies for renovation” by Andrews et.al (2011) published by OECD. The idea is that housing policies and tenure status affects the moving rate or mobility, mobility costs and possibilities are related to the effectiveness of the labor market. If mobility bears low costs in money and time workers competence and labor are more effectively exploited on the labor market, accessibility of housing and infrastructure are therefore key factors (Andrews et. al 2011).
If high mobility on the housing market is important, what controls this mobility? It’s suggested that government policy and especially legislation and regulation are important. Legislation and regulation concerning tenure status might be important factors controlling mobility possibilities on the housing market. The difference in mobility costs associated with different tenure statuses are evident (Dewilde & Decker 2016; Andrews et.al. 2011). Specific costs are connected to specific tenure statuses and are generally often to some degree controlled and affected by government policy. Governments tax, subsidize, support and favor development of homeownership and rental housing differently depending on ideology and economic outlook over time. This constellation of power relationships, ideology, historical and cultural patterns controlling the housing market can be referred to as “housing regime” as proposed by Kemeny (1981) and Dewilde and Decker (2016). The term housing regime basically includes the political and economic organization of the housing market, provision allocation and consumption of housing. In terms of mobility and different housing tenures international research suggest that rental housing tenure tends to favor ease of mobility the most in contrast to owned homes which tends to favor mobility the least (Andrews et. al. 2011; Dewilde & Decker 2016; Coulson and Fisher 2008; Isbaert 2013; Bauert et. al. 2014; Laamanen 2013; Green & Hendershott 2001a; Munch et al. 2006a; Van Leuvensteijn and Koning, 2004; Coulson and Fisher, 2002). Mobility costs are generally suggested to be the cause. In a regional context the combination of low market regulation and rental apartments seems to favor mobility the most, suggesting that rental apartments also yield the most positive labor market outcomes and economic growth. Even though renters tend to be the relatively most mobile of tenants other housing related factors might have greater impacts on residential mobility or labor market dynamics (Andrews et. al. 2011; Dewilde & Decker 2016)
Different sub groups of rental tenants and homeowners show different mobility behaviors. Besides suggesting that private renters are the most mobile of tenants, Andrews et. al. (2011) also propose that rental tenants living in social housing are the least mobile of all tenants, homeowners included. Rental tenants living in social housing or housing with reduced rent are less mobile than renters living in housing with market regulated rents, if rent subsidies are immobile. Further, homeowners can also be divided into two groups, homeowners with a mortgage and homeowners
without a mortgage. Homeowners with a mortgage are more mobile than homeowners without a mortgage (Andrews et. al. 2011)
If above holds true, that renters are more mobile than homeowners, homeownership rate is a possible determinant for unemployment (Blanchflower & Oswald 2013), since mobility is important for the effectiveness of the labor market.
This relationship was first noticed and described in a European context by Andrew J. Oswald in his paper ‘The Housing Market and Europe’s Unemployment; a non
Technical Paper’ (1999), which was important in the formation of the so called
Oswald hypothesis. The simple correlation that forms the basis of his hypothesis was the correlation between homeownership rate and unemployment rate, European countries with the highest unemployment rates also had the highest homeownership rates and countries with the lowest homeownership rates also had the lowest unemployment rates. As an example Spain has one of Europe’s highest unemployment rates, around 20 percent and also among the highest level of homeownership rates at about 80 percent. Switzerland on the other hand shows the opposite relationship, with an unemployment rate of around 3 percent and a homeownership rate of 30 percent (Blanchflower & Oswald 2013).
The Oswald hypothesis has been tested on many European countries by many scholars with varying results, but a more in depth analysis has not been made on Sweden. Nor on a regional level neither on county or municipal level. The Swedish context is interesting in many ways, low population density, long distances and a recent policy and regime shift from a housing regime with a substantial public rental housing stock to a housing regime where cooperative owned housing is dominating many urban areas (Holmqvist & Turner 2014). The spatial dimension of the relationship between unemployment and homeownership has not been taken into account by many scholars and the Swedish context is extra interesting since regional differences are so pronounced.
1.2 Aim
The aim of this thesis is to test the Oswald (1999) hypothesis in Sweden, to investigate the relationship between homeownership and unemployment. Does an increase in homeownership correlate with an increase in the unemployment rate in Sweden? As proposed by Oswald (1999), and also recently supported by Laamanen (2013) and Blanchflower and Oswald (2013). I will also explore spatial aspects of the relationship of homeownership rate and unemployment rate in Sweden.
2 Theory
In this section the general theoretical framework and earlier research is reviewed. First a review of the Oswald hypothesis and following reviews of other scholars tests of the hypothesis. Many scholars have tested and refined his methods and hypothesis for different locations around the world, with varying results. Some studies support his hypothesis and theory and some reject it. To be able to relate earlier research to the Swedish context the last decades development of the Swedish housing regime is also reviewed.
2.1 The Oswald hypothesis
In a series of papers (Oswald, A. J. 1996, 1997a, 1997b, 1998) Oswald investigates the relationship between homeownership and unemployment, of which the most influential paper is ‘The Housing Market and Europe’s Unemployment; a nonTechnical Paper’ Andrew J Oswald (1999), where Oswald argues and proposes that a
high share of homeownership leads to higher unemployment, Oswald could show a correlation between homeownership and unemployment both within and across countries, in his study a 5 percent unit higher homeownership roughly equals 1 percent unit higher unemployment in the investigated area.
Oswald (1999) argues that the global economy is dependent on workers ability to move around to find new jobs and to develop their careers. The mobility possibilities is also useful for employers who scout for workers with specific talents and skill sets, often hard to find in the local labor market. Private and public but not social rental
housing (Andrews et. al. 2011) offers the highest degree of ease of mobility. One of the European countries with the lowest unemployment rate also has the highest share of private rental housing. Switzerland’s wellfunctioning private rental housing market makes it easy for workers and employers to match respective requirements. Modern history reveals similar patterns at several points in time, most of Europe had low homeownership rates and low unemployment rates the period between 1950 and 1960 whereas the United States had a relatively high homeownership rate of 60% and also the highest unemployment rate in the industrialized world at the time. Since then and till the early 2000’s US homeownership rates and unemployment rates did not change much (Oswald A, J.1999), but since the financial crisis the homeownership rate has fallen to levels experienced in the 1960’s, just above 63%, compared to record high levels of around 69% in year 2005, the unemployment has simultaneously also fallen to precrisis levels (Howley, K. M. 2015). Among the industrialized nations only Japan and Switzerland has not experienced the strong increase in homeownership and unemployment, nations like Spain who experienced among the strongest growth in homeownership also experienced the most rapid increase in unemployment rate the last decades (Oswald A, J.1999). The correlation might involve many different processes but Oswald (1999) proposes five possible causational processes. First of all, there are unavoidable costs related to moving, selling a house or apartment is often expensive. In case of unemployment a homeowner is more likely to commute long distances for work or to take a local job that might not perfectly match one's competence, especially if there is a mortgage on the house or apartment. Because homeowners are less mobile than renters they are more vulnerable to regional economic downturns. In many European countries, like Spain or the UK, the affordable rental share of the housing market is almost nonexistent which makes it very hard for young people without capital to enter the housing market and move to where jobs are, instead they live with their parents. Second, this means that homeowners are not generally unemployed themselves but rather that unemployed people can't move to the right places, where jobs are. Third, as earlier mentioned, the immobility of homeowners, especially homeowners with a mortgage, makes for worse labor market matching between workers and employers. This inefficiency raises production costs and lowers real wages in a region, which hampers companies abilities to develop. The higher production costs both makes the
out from inefficiency. Fourth, strong homeownership communities are generally thought of as stable and desired in many ways and often have a greater influence on local politics. Homeowner’s engagement in local politics may however affect planning laws and regulations concerning land development, in such a way that it hampers development. Local lobbying groups of homeowners can have a powerful impact on local politics and might make it harder for startups, entrepreneurs and companies from setting up new operations in the neighborhood. Fifth and last, survey data on commuter patterns are clear, homeowners commute significantly more than renters, most often also longer distances, this might have undesired consequences in the form of traffic congestions, that makes it more expensive both economically and time wise to travel to work besides increased costs for the transport sector. It can also be argued that the higher commuter and transportation costs reduces the net gains of working compared to not working, acting like higher unemployment benefits, the attractiveness of not working is raised (Oswald A, J.1999).
2.2 Earlier research and tests of the hypothesis
2.2.1 Research development
The Oswald hypothesis has been tested for many countries and regions around the globe, with different methods and data, both with micro individual level and macro level aggregate data. The research does not give a unison answer, not surprising considering the political, social and economic differences between nations and regions, giving different housing regimes. Some research reject the hypothesis and some support it, most research support at least some of Oswald's many claims.
Housing regimes differ in character between countries which means that different rules and regulations affect the dynamics on the respective housing market, generally five housing tenure types can be distinguished, homeownership of house or apartment and private, public or social renter. All housing tenure types are not apparent in all countries and there are also differences within the respective types. The methodological development from the earliest to the most recent study is evident, Oswald’s earlier studies (Oswald, A. J. 1996, 1997a, 1997b, 1998, 1999) was mainly conducted with aggregate data and simple descriptive statistics and simple regressions, many following studies adopts either a micro (Green & Hendershott
2001b) (Coulson & Fisher 2002) or a macro (Coulson & Fisher 2008; Flatau, et. al. 2002; Isebaert 2013) approach for regression and survival analysis. Individual level or some aggregate level data is used in most models. In more recent studies it's common to model both individual and aggregate level data (Blanchflower & Oswald 2013; Coulson & Fisher 2008; Laamanen 2013).
L’Horty and Sari (2010) identify and sum up the divides within research development in the field, on a global level they point out the clear divide between the results produced by empirical analyses mobilizing micro and empirical analyses mobilizing macroeconomic approaches. Generally studies with a microeconomic approach reject the Oswald hypothesis whereas studies with a macroeconomic approach confirms it (Dietz & Haurin 2003; L’Horty & Sari 2010). Many scholars argue that the initial macroeconomic approach by Oswald (1999) suffers from spatial dependence (L’Horty & Sari 2010) between regional units and endogeneity bias. The endogeneity bias is dealt with in several micro approach studies (L’Horty & Sari 2010; Brunet & Lesueur, 2003; Brunet et al., 2007; Munch et al., 2008; Coulson & Fisher, 2008) but few macro approach studies do, there are a few though (L’Horty & Sari 2010; Blanchflower & Oswald 2013; Laamanen, J. P. 2013). The endogeneity bias problem is evident since unemployment rate and homeownership might be determined simultaneously (L’Horty & Sari 2010), there are however some model outcomes indicating that unemployment rate is more accurately determined by or correlated to homeownership with a time lag of approximately 5 years (Blanchflower & Oswald 2013). Spatial dependence occurs since the spatial units used, such as region, county, municipality or other administrative unit might be spatially dependent or correlated to each other. In the case of unemployment it means that the unemployment rate in one spatial unit might be dependent on or correlated to neighboring spatial units. L’Horty & Sari’s (2010) study is one of few studies that control for spatial dependence.
Short reviews of earlier research is presented below, divided by general support or rejection of the Oswald hypothesis and by region or country. In short method, specific data used and unique details are reviewed in relation to the Oswald hypothesis theoretical framework. The results are seldom unison, most studies support some of Oswald's claims and reject others.
2.2.3 Support for the Hypothesis
As mentioned earlier mostly studies including macroeconometric models find support for the Oswald Hypothesis, some scholars have used both micro and micro data and have sometimes also found contradicting results. Below, tests of the hypothesis in different countries is presented.
2.2.4 Research on the US
Many scholars have tried to reproduce Oswald's results on US states with a variety of models, Green and Hendershott, (2001b) produced one of the earliest studies, they generally agree with Oswald’s hypothesis. Their model show that unemployed homeowners experience longer unemployment spells than unemployed renters but the impact of homeownership on unemployment is according to them significantly weaker, only an eighth of that suggested by Oswald (1999). Unlike Oswald (1999) they wanted to test the hypothesis using micro data instead of aggregate data, to escape aggregate data bias. Green and Hendershott (2001b) use panel data roughly ranging from 1988 to 1992, they first model unemployment duration with a model design inspired by Lumsdaine et.al. (1995), essentially a value function weighting household's economic benefits of taking a job as fast as possible or to wait for a local job opportunity. To explain unemployment spell duration, the time between entering and leaving unemployment they use the Weibull distribution, a hazard function. During the work progress the dataset was extended with data from 1986 and 1987, this additional data made the model estimate a moderate mortgage “lock in” effect on mobility.
The model reveals several patterns regarding unemployment duration and individual characteristics. Age turns out to be an important determinant of unemployment
duration, younger homeowners under the age of 45 but especially under the age of 25 are reemployed notably faster than older homeowners, likely due to pressure from relatively larger mortgage payments, more substantial mortgage leverage. Also, maybe not so surprising, the model show that health, ethnicity and number of children are important determinants of unemployment in combination with housing tenure. Unemployed homeowners with children and unemployed renters with poor health tends to be reemployed slower than respective counterparts. Separated and widowed homeowners tend to be reemployed faster than renters with similar characteristics, also unemployed ethnic white renters are reemployed faster compared to ethnic minority renters (Green & Hendershott 2001b).
Coulson and Fisher (2008) made a more rigorous test of the hypothesis than in previous works (Coulson & Fisher 2002) when further investigating the US, this time also with aggregate data. Five different models were made, inspired both by Oswald (1999) and Munch et.al. (2006, 2008), the five regression models predict surprisingly different outcomes, the models use both micro level and aggregate level data. Models are designed with suitable instrumental variables, on personal characteristics in both models with micro and aggregate data, with both unemployment rate and wages as dependent variables. The general findings on an individual level are that homeowners are less likely to experience unemployment and that they have lower wages compared to renters, all else equal. On an aggregate level regional high homeownership rates are on the contrary associated with higher probability of individual workers being unemployed and wages are comparably higher, so generally complete opposite relationships from regressions on microdata. Since the models on micro and aggregate level data gives contradictory results, Coulson and Fisher (2008) argues that regional homeownership rates are likely not good instruments for individual tenure choice in regression models, and that even though aggregate model outcomes suggest that homeownership hampers the labor market the positive effects of homeownership might outweigh the negative labor market outcomes (Coulson & Fisher 2008).
Munch et.al. (2006, 2008) have researched the applicability of the Oswald hypothesis in Denmark, the first study (Munch et.al. 2006) generally rejects the hypothesis, homeowners are not more often unemployed than renters, in fact they find the opposite relationship. However, Danish homeowners proved to be less mobile than Danish renters, a key mechanism in the hypothesis as proposed by Oswald (1996, 1997a, 1997b, 1998 & 1999). According to Munch et.al. (2006) Homeownership hampers the propensity to move for job reasons but homeowners proves to be more successful than renters in finding jobs on the local labor market. Munch et.al. (2008) second study on Denmark give more support to Oswald's hypothesis than the first, even though they didn't find a positive correlation between unemployment duration and home ownership on an individual level, the results does show that homeowners are less geographically mobile when finding jobs than renters, both on the local labor market and outside the local labor market. Also, homeownership has a significant negative effect on the risk of unemployment and positive effect on wages, on an individual level. Also a study made by Ahn and Cuesta (2007) based on a bivariate probit model with individual household panel data from the European Community Household Panel (ECHP, 19952001) show that Danish homeowners are less geographically mobile on the labor market than renters. The study also includes data on France and Spain, the effect of homeownership on labor market mobility is noticeable in France but not in Spain. Ahn & Cuesta (2007) results also indicate that job mobility is driven by individual job satisfaction, individuals less satisfied with their jobs are more likely to change jobs and also housing. Job satisfaction is also related to job mobility in terms of commuting time, lower job satisfaction is related to longer commuting times and results in greater job mobility but not greater residential mobility (Ahn & Cuesta 2007).
2.2.6 Research on France, Paris
L’Horty and Sari (2010) tested Oswald’s hypothesis for some regions in Paris and found contradicting results, everything else being equal they found out that a high homeownership rate is associated with a low unemployment rate, lower than for renters in the controlled regions. The results presented might be a rehabilitation of the Oswald hypothesis since the model besides controlling for homeownership also
controls for percentage of private and public housing tenure and show that private renting is the most favorable housing tenure in terms of labor market outcomes.
Unlike many other studies this is made on a municipal level, a spatial level between micro and macro level, smaller than most regions used in other studies and not on an individual level, also one of very few studies made on France. The method used also controls for expected biases, spatial autocorrelation and endogeneity problems by mobilization of spatial regressions. Several regressions made supports the Oswald hypothesis, with homeownership as only explanatory variable and with several added municipality characteristics variables, homeownership turns out positively correlated to unemployment rate. The opposite relationship is found when homeownership and unemployment to work hazard rates are tested, which means that homeownership itself does not lead to unemployment more than other tenancies, rather even less so, but that some processes on a municipal level related to homeownership likely or might lead to unemployment. When compared to other tenancies which is also controlled for, homeownership is less favorable to employment than private renting but more favorable to employment than public renting. The higher mobility costs related to homeownership compared to private renting suggested by Oswald (1999) might be the divider. Lower mobility costs might make private renters to have more favorable labor market outcomes than homeowners, and also make public renters with stationary rent subsidies (social housing) have less favorable labor market outcomes, creating a “lock in” effect. This “lock in” effect is evident in most European countries with social housing, or stationary rent subsidies. It's argued that the relatively lower rents in social housing gives less financial incentives to move. A new job must make a greater impact on the individual's financial situation to compensate for the loss of rent subsidies when changing tenancy to private renting or homeownership when moving for a new employment (L’Horty & Sari 2010)
2.2.7 Blanchflower and Oswald applies macro and micro data
in US
Much like Coulson and Fisher (2008) also Blanchflower and Oswald (2013) mobilize both aggregate and micro level data, on unemployment rate, homeownership rate and education. Blanchflower & Oswald (2013) generally support the Oswald
hypothesis even though they find that homeowners are not more often unemployed than renters.
They argue that the dynamics on the housing market creates externalities, on the labor market and the economy, both locally and regionally. The hypothesis of “externalities” is also supported by recent research (Isbaert 2013; Bauert et. al. 2014; Laamanen 2013). Further Blanchflower & Oswald (2013) conclude that high homeownership in the US is associated with lower labor mobility, longer commutes and fewer new firms and establishments. Lower labor mobility is likely caused by the higher costs of moving related to homeownership compared to renting, also supported by Lux and Sunega (2014). The higher mobility costs are also likely a cause for longer commuting times so evident in areas with high homeownership rates. The fact that fewer new firms are established in areas with high homeownership rate is not fully understood, it can however be a consequence of NIMBY (not in my backyard) effects and zoning. (Blanchflower & Oswald 2013).
2.2.8 Research on Finland
Laamanen’s (2013) results are generally in line with Blanchflower & Oswald’s (2013) results and concludes that on an individual level homeowners often have more positive labor market outcomes than do renters but on an aggregate geographical level high homeownership rates creates negative externalities, both on the labor market and in the economy. It is argued that the negative externalities are likely a consequence of individual debts associated with home purchases, the debts lead to a reduction in consumption and demand. A decrease in demand for both locally and regionally produced goods and services. The homeownership financed by debt also likely leads to increased local and commuter distance to job market competition. Laamanen’s study has a unique feature, it exploits the rental deregulation reform implemented in Finland, the reform created exogenous variation in homeownership across regions, thereby avoiding the endogeneity problem that earlier studies dealt with.
As mentioned earlier, the microeconometric approach applied by many scholars generally leads to the conclusion that homeowners are less often unemployed than renters and therefore the hypothesis is rejected. Below tests of the hypothesis in different countries is presented.
2.2.10 Research on US, both micro and macro data
Coulson and Fisher (2002) tested the Oswald hypothesis using individual micro data from the US, the hypothesis is generally rejected, the model suggest that owners, all else equal, have better labor market outcomes, meaning fewer and shorter spells of unemployment and higher wages than renters. The models by Coulson and Fisher (2008) designed on micro data generally agrees with their earlier work (Coulson & Fisher 2002), besides the models made on aggregate data.
Green and Hendershott (2001a), the second study conducted in 2001, rejects the hypothesis, they deny the correlation between unemployment and homeownership across US states as proposed by Oswald (1999). The model is one of the few with a macroeconometric approach, made with aggregate data that rejects the hypothesis, they show that the correlation disappears when age classes are introduced and variables are weighted with ageing population and number of households in the model. Green and Hendershott (2001a) have used data ranging from 1970 to 1990 and analyzed crosssectional variation in homeownership and unemployment rates. The development of the model and especially the introduction of age classes in the model gives the unique finding that the correlation between unemployment and homeownership doesn’t exist for younger and older households but is evident for middleaged households. They argue that young households compared to middle aged households have had little or no possibility and time to save enough capital to buy a home and get emotionally attached to a geographical area, so young households are more likely to relocate as a response to unemployment than middleaged households. Older households are most often not affected by unemployment or labor market dynamics since they are most often not a part of the work force but retired
(Green & Hendershott 2001a). These findings are in line with and give some explanation to earlier research on homeowners and renters mobility by Henderson and Ioannides (1989), who showed that homeowners on average wait 14 years between moves while renters only wait 4 years. The data mobilized by Green and Hendershott (2001a) show that by 1995 as little as 27 percent of the 1991 homeowners had moved, while in the case of renters, as many as 85 percent of the 1991 renters had moved (Green & Hendershott 2001a).
2.2.11 Research on Australia
Flatau et.al. (2002) tested and generally rejected the Oswald hypothesis, after testing it in Australia, and proposed that mortgage leverage effects might be a viable reason for the correlation between homeownership and unemployment that Oswald (1999) propose. They tried to replicate Oswald’s (1999) method by making a simple OLS model but they also developed several more models with three different geographical levels and more independent variables controlling for population characteristics, such as education, age, race and so on. Besides adding variables controlling for population characteristics they also expand the tenure status types by dividing homeownership in two categories, homeowners with mortgage and outright owners without mortgage and by dividing tenants in private and public tenants. The following study by Flatau et.al. (2003) further develops the models with micro data, individual survey data of income and housing costs 199497, about 13500 persons living in private residential dwellings, in working age they investigated possible mortgage leverage effects. A for the time unique feature of the study is that it includes many states of ownership and tenancies, both ownership with and without outstanding mortgage and private, public and rentfree renters. The main conclusions are that individual homeowner’s experience fewer and shorter unemployment spells than renters, in general. The model outcomes also show a significant difference between leveraged and nonleveraged homeowners, homeowners with and homeowners without outstanding mortgage, homeowners with outstanding mortgage show even fewer and shorter unemployment spells than homeowners without mortgage, outright owners. Flatau et.al. (2003) argue that the greater incentive to remain employed and to become reemployed rapidly is created by the will to pay the mortgage and keep the house is the cause of this pattern. The argumentation is similar for the patterns revealed between different rental tenanciesand unemployment, those paying below market rents in rentfree or social housing tenancies has a lower incentive to avoid unemployment or to work hard to get reemployed than private renters paying market rents. The one exception that support the Oswald hypothesis is that out right female owners have longer unemployment spells than similar renters, of all tenancies (Flatau, et. al. 2003).
2.2.12 Research on the Netherlands
Van Leuvensteijn and Koning (2004) generally rejects Oswald's hypothesis after testing it in the Netherlands, their results point in the opposite direction, renters are more often unemployed than homeowners. The results stem from models on micro panel data stretching from 1989 to 1998. A dataset with histories from individuals on labor and housing market outcomes as individual job mobility, duration and probability of being a homeowner, variables are modelled simultaneously in a non parametric model. The data includes 595 individual job changes, and there is no correlation to be found between job changes and housing status.
They argue, in opposition to other scholars that the housing market is affected by the labor market rather than the other way around, which is a unique idea among the studies made on the subject. They also state that their empirics suggests that job commitment is one of the most important determinants of housing decisions, if job commitment is controlled for the impact of homeownership decreases significantly. It is especially evident that no evidence for homeownership affecting the risk of job changes is found, as well as the risk of nonparticipation in the labor market, in fact homeownership shows a negative effect on unemployment (Leuvensteijn & Koning 2004).
Van Leuvensteijn and Koning (2004) also propose some explanations for these patterns, within and outside of Oswald's explanatory framework. First, the Netherlands are one of the most densely populated countries in Europe, commuting distances are also generally short since the country is relatively small. For these reasons people mostly tend to change jobs without changing residence. Second, like many European countries also the Netherlands have had a strong positive
related to homeownership (Van Leuvensteijn & Koning 2004)proposed as a mobility hampering mechanism by Oswald (1999). Third, also the Netherlands have a not insignificant social renting sector as part of the housing stock, with regulated and subsidized rents. The regulation of the social renting sector may result in higher moving costs for some tenants, tenants who move and become private renters or homeowners. Like other scholars in the field modelling with micro data they propose that homeowners are more sensitive to the decrease in income that unemployment bears with it because they often have mortgages and also in the case of the Netherlands, homeowners are not generally eligible for social assistance but have to break into their housing equity if they can’t afford the mortgage. Tenants on the other hand have access to the rent subsidy system in case of severe loss of income, they can be seen as partly insured against loss of income. In this prospective homeowners have a greater incentive to not become unemployed and to invest more in job specific capital and ask for higher wages. (Van Leuvensteijn & Koning 2004) The study on Australia by Flatau et.al. (2003) could also show mortgage leverage effects, homeowners with mortgages had significantly shorter unemployment spells than homeowners without a mortgage and even more so compared to renters. However, a study with micro data on homeowners labor market outcomes and job search behaviors by Rouwendal and Nijkamp (2010) on the Netherlands show a slightly different pattern. In the model, homeowners have a higher job search intensity and also shorter unemployment spells all else equal compared to renters, even more so for homeowners with higher mortgages and general higher housing costs. Dutch homeowners do generally have higher housing costs than renters with similar individual characteristics. This explains the earlier empirical findings that homeowners experience fewer and shorter unemployment spells in general compared to renters, overriding the evident hampering effect of homeownership on geographical mobility on the labor market. An interesting finding is that the Oswald hypothesis is supported when homeowners with lower housing costs than renters, as is often the case with outright owners are compared to renters. Outright owners experience more and longer unemployment spells than renters, and hampers geographical mobility even more than homeowners with a mortgage (Rouwendal & Nijkamp 2010)
2.2.13 Research on the UK
Battu and Phimister (2008) tested the Oswald hypothesis using UK micro data and they generally reject it, they do however add some interesting insights related to the relationship between homeownership and unemployment, and spatial mobility patterns associated with different tenures. Battu and Phimister (2008) study job to job transitions and also transitions from unemployment, they control for spatial mobility by distinguishing whether or not a new employment was associated with a nonlocal residential move or if a new job was attained in the local job market. The model also control for tenure endogeneity and unobserved heterogeneity. The model outcomes suggest that homeownership acts hindering for the employed and public renting is more hindering for the unemployed, homeowners with an employment more seldom than other tenants move long distances for a new job and unemployed public renters are less likely to move long distances for a new job in a nonlocal labor market. However, unemployed private renters do show a higher probability than public renters and homeowners of attaining a new employment in a distant labor market, private renters prove to be the most mobile. An interesting finding is also that there is a clear difference within different educational or labor groups, homeowners that are skilled manual/nonmanual workers prove less mobile than managerial/professional workers. The general conclusions are in line with other similar studies (Munch et al. 2006a; Van Leuvensteijn & Koning, 2004; Coulson & Fisher, 2002) regarding the rejection of the Oswald hypothesis based on micro data but it is also concluded that homeownership in general negatively affects job mobility which might induce noticeable negative labor market outcomes on an aggregate macro level (Battu & Phimister 2008)
2.3 The Swedish housing regime
2.3.1 The Grand Restructuring, a deregulation process
Hedenmo & von Platen (2007) have reviewed the last 130 years of Sweden's housing market politics. They among others (Holmqvist & Turner 2013; Clark & Johnson 2009; Lind & Lundström 2007; Turner & Whitehead 2002) concludes that since 1991 the housing regime in Sweden has undergone a Neoliberalisation process or a “grand
restructuring”, as named by Turner and Whitehead (2002). The grand restructuring has resulted in a similar situation experienced in the beginning of the 1900’s, before the 1930’s when housing policies were completely reworked by the social democratic party (Hedenmo & von Platen 2007). Today as then almost all municipalities experience a shortage of housing and a steep increase in prices, the housing market is viewed as close to overheat, many warn a housing bubble is just around the corner, especially in the three major cities, as discussed by Holmqvist and Turner (2013). The restructuring or deregulation process meant a decrease of the rental sector by roughly 11% in favor of ownership of coop apartments, detached housing and speculation. Housing policies dominated by subsidies designed to favor construction of rental apartments was redesigned to instead favor ownership, also policies opened the possibility for conversion of existing public housing to form coop ownerships, cooperatively owned housing. The remaining public housing stock is still under heavy regulation, especially rent regulation. Christophers et.al. (2013) therefore argue that the housing regime in Sweden is a hybrid system rather than a strict neoliberal totally deregulated system. A combination of a heavily regulated rental sector and a deregulated ownership sector. As part of the deregulation process also condominiums ownership apartments are introduced as an alternative for new housing production, since 2009. The idea is that condominiums apartments can more easily be sub leased since there is no democratic process involved and the owner actually owns the apartment and not like in the case of coops where the owner only owns the right to live in the apartment and also must live in the apartment (Holmqvist & Turner 2013). The European Union has also put pressure on Sweden to deregulate and reform the housing market, convert it to a more market oriented housing market more like the EU norm. The pressure from EU has resulted in a change of the business structure of the public housing companies and of the rent regulation system. As a consequence of an appeal from representatives of the private rental sector to the EU commission the public housing companies are since 2009 (SOU 2008:38; Prop. 2009/10:185) obliged to be run business like and not as previously, run as nonprofits (Christophers et. al. 2013). In 2010 the government passed an act (Prop 2009/10:185) that also put an end to the rent setting role of public rental housing over private rental housing, rents based on utility is since 1940’s negotiated between the public rental sector companies and tenant unions. This change opened up for
market rents, only in the production of new rental housing has it made a notable difference so far (Holmqvist & Turner 2013).
Since the 1940’s the welfare state has focused on public housing built for the masses, whereas now the shift in policy has redirected the support to owning but also to low income households and deprived neighborhoods, a step towards the kind of social housing so common in Europe and US. Harloe (1995) divides social housing provision in two different models, the grand restructuring is converting the social housing provision model from a general mass model to a residual model, from a wide to a more selective social housing regime. Imbedded in this model framework is also the support for ownership and the dismantling of the public housing sector. Since the conversion was made legal in 1992 a set of policy changes has been made to further favor and promote ownership. The prices for conversion of rental apartments to co op apartments has been systematically set substantially lower than market price when offered for sale to tenants. Also, in 2008 the real estate tax was abolished and replaced with a lower flat municipal fee (Holmqvist & Turner 2013).
2.3.2 Housing cost development
The grand restructuring of the housing market has also likely affected the poverty levels in Sweden, the poverty has risen steadily since 2001. The link between income and housing outcome is evident. One reason is that housing allowances has not been adjusted to price index and change in policy has also led to fewer people being entitled to allowances. The most vulnerable groups are single, single parent and young households, the absolutely most vulnerable are single women with children. The government hardly fulfil the welfare goal of housing equality, as a consequence of the grand restructuring. Besides lower affordability, a consequence of lower allowances and higher unemployment, raised rents and constantly rising house prices, the high mortgage level of Swedish households has increased significantly (Holmqvist & Turner 2013). In 2011 Swedish households were higher mortgaged than ever before, the household liability was 170% (IMF 2012a) of the yearly disposable income compared to 50% in 1995 (BKN 2011). Only in Hong Kong and South Africa the prices have increased faster, mortgage debt have naturally increased in the same pace as the housing prices. After some regulation adjustments of the mortgage
market the mortgage debt subsided to 165% in 2012 (SCB 2013), the report also argue that the high and increasing housing price is partly a consequence of low housing production (SCB 2013). The prices for rental apartments are also constantly increasing, right after the deregulation reform in 1991 the prices increased 30%. If the price for a rental apartment is compared to cost of owning, it is revealed that it is the most expensive form of tenancy if compared by living space as number of rooms. Public rental housing tenants also have lower disposable incomes compared to ownership tenants (SCB 2015). This pattern is likely a consequence of the deregulation reform, the shift in policy from tax subsidized rental apartments to mortgage tax subsidies on ownership.
2.3.3 Housing stock Mobility
Research regarding mobility dynamics in rented housing stock and owned housing stock by Andrews (2011) concludes that the mobility is generally higher in the rental housing stock in European housing markets, especially among private rental tenants. According to Brandén and Pistol (2013) this is also the case on the Swedish housing market, on a national level year 2012 the mobility in the rental housing stock was 22 percent and in the owned housing stock 9 percent. For metropolitan, suburban and large cities the difference is smaller but still evident (Brandén & Pistol 2013).
3 Method
In this section I will describe and motivate the applied methods, also in relation to methods used in similar research. First a general description of the methods and then a more in detail description of methodological aspects and choices of variables.
The goal from the start was to apply a method like the ones applied by Blanchflower & Oswald (2013) and Laamanen (2013), they apply both aggregate and individual level panel data. Very few studies besides those two just mentioned have applied both aggregate and individual level data, therefore, in sake of comparability I’ve only applied aggregate level data. The choice of variables are mainly motivated by works of Blanchflower & Oswald (2013), Laamanen (2013) and L’Horty and Sari (2010).
Panel data on numerous variables ranging from 1998 to 2013 was attained from SCB and descriptive statistics as box plots was made, descriptive statistics in combination with earlier research mainly by Blanchflower & Oswald (2013), Laamanen (2013), L’Horty and Sari (2010), Coulson and Fisher (2002) and Flatau et.al. (2002) was used as a guide in the choice of dependent, independent and control variables. In order to further investigate the relationship between unemployment and homeownership, beyond the capacity of descriptive statistics, Pooled OLS (ordinary least squares) regression and GLS (generalized least squares) random effect regression models was estimated. To improve the models, year dummies, spatial dummies and personal characteristics control variables were added. Year dummy variables are added to control for temporal or time variation effects in the dependent
variable, which is unemployment rate. Spatial dummy variables are added to control for spatial effects in the dependent variable, in this case county and SKL municipality group effects were tested for, in the final models SKL municipality groups is used. The personal and regional characteristics control variables used in the final model are variables for, education, ethnicity (foreign born), age, young workforce, gender (female workforce), population density, commuters and GRP per capita.
3.1 Data
This section explains the data, namely the variables included and related to the model, some variables mentioned were tested but not included in the model. Most of the variables fit in the concept of labor market mobility, as defined by Friedman (1968). Most of the variables are represented in earlier research by for example Coulson and Fisher (2002), of Blanchflower & Oswald (2013), Laamanen (2013), L’Horty and Sari (2010) and Flatau et.al. (2002). The data set was obtained from statistics Sweden (SCB) and contains data on all variables from 1998 to 2013.
Dependent variable
Unemployment rate
Independent variables
Homeownership rate Apartment
Tenant
Education
Foreign‐born
Age
Young workforce
Female workforce
Population density
Commuters
GRP/capita
3.1.1 Unemployment rate, dependent variable
The dependent variable in the panel data regression model is unemployment rate, unemployment rate within the group 2064 year olds. Unemployed are defined as openly unemployed, defined by Statistics Sweden.
3.1.2 Education
The Education variable used in the model is defined as share of population with university studies of 3 years or more, amongst 1664 year olds, the rest of the data set includes 2064 year olds but it's not likely that any person under the age of 22 has studied 3 years or more at university level. Education level likely has a strong impact on unemployment, most researchers use it as a control (Coulson and Fisher 2002; Blanchflower and Oswald 2013; Laamanen 2013; L’Horty and Sari 2010; Flatau et. al. 2002)
I tested many different educational variables for different educational levels and also combined variables, for example share of homeowners with higher education.
Controlling for higher education with respect to tenant status is good to control, so that the variables for homeownership will accurately represent homeownership, since relatively many homeowners are highly educated. Descriptive statistics show that in some municipalities as many as fifty percent of homeowners have studied 3 years or more on university level compared to the municipality average ranging from five to thirty percent. Also the variable might show more spatial differentiation. Besides the above described variables other education variables were tested, representing high school and grammar school level and also dummies for percentage of municipal education levels but diagnostics indicated too high multicollinearity in most cases. In the final model only one educational variable is used.
3.1.3 Homeownership, tenant status
The variables expressing homeownership or tenant status in the model are homeownership apartment and tenant, tenant refers to tenants in both the public and the private rental sector. The difference in spatial distribution between homeownership of detached house and homeownership of apartment seen in the descriptive statistics (Fig. 510) makes it interesting to investigate further but the change in the share of homeownership of detached housing is next to none over the years or across municipalities, about 1 percent nationwide. The variable representing homeownership of detached house is therefore only used as a reference in the model. Metropolitan, large cities, suburban and more densely populated areas are expected to have more apartments than sparsely populated areas, as shown in descriptive statistics (Fig.710) Metropolitan areas and suburban in Sweden has also experienced a shift from more rental to more ownership apartments the last decades due to political reforms (Dewilde & Decker 2016; Homquist & Turner 2014; ). The variable
homeownership apartment is also interesting considering the conversion reform of
rental to ownership apartments that has taken effect in Sweden the last decades. The model made by Laamanen, J. P. (2013) also incorporates a policy reform effect in the model, a governmental experiment investigating the possible societal outcomes of a market rent reform in Finland, in the investigation market rents were implemented in parts of the country. Diagnostics indicated to high multicollinearity when more than one homeownership variable was present in the model and when testing for the
combined Homeownership variable it was not significant when spatial dummies was added, likely as a consequence of spatial differentiation, which can also be seen in descriptive statistics (Fig.710).
Some scholars like Coulson and Fisher (2002), Blanchflower and Oswald (2013), L’Horty and Sari (2010) and Flatau et.al. (2002) also control for rental housing tenant, in most cases also divided into two variables, one for private and one for public tenant. I didn’t have the possibility to separate them, limited by the data set. In many cases public rental housing is also referred to as social housing and is heavily subsidized, often located in less attractive areas and targeted towards economically weak individuals, whereas private rental housing is not subsidized and often found in more attractive locations. For these reasons it makes a lot of sense to make two variables, since the relationship to unemployment likely differs in character. In Sweden there is no division between the two groups of tenants but as a consequence of the deregulation reform the old public and private rental housing stock is more rigorously regulated whereas newly produced rental apartments, private or public basically have market rents (presumptive rents), which are generally higher (Homquist & Turner 2014). So there is a divide on the rental housing market and two variables instead of one would’ve been preferable.
3.1.4 Economics, GRP (Gross Regional Product)
GDP and GRP affect unemployment according to numerous scholars, exactly how is likely contextual but Okun’s law from 1962 is still considered valid as fundamental starting point of discussion. Okun's law is simple and basically states that a decrease in unemployment leads to an increase in productivity and as a result economic growth, the opposite reasoning is also valid, an increase in unemployment leads to decreased productivity and therefore negative economic growth. Okun's law is valid in a continuously growing economy. Okun's law has been tested empirically for many countries and regions during long periods of time and is proven to still have substantial validity in explaining the relationship between unemployment and GDP, even though the exact relationship has proven numerically different from country to country and region to region (Farsio & Quade 2003) (Freeman 2000, 2001).
The model in this paper includes GRP per capita in absolute values rather than GRP growth, both has been tested and it seems GRP in absolute figures better show regional differences, since the growth is only positive or negative, it also has an inverted relationships to unemployment. GDP in absolute figures have a negative relationship and GDP growth a positive relationship, which in this case actually violates Okun's law. The GDP per capita in absolute figures are converted from county to municipal level so the resolution is unfortunately not as high as the rest of the data set. It is important to include a GDP variable since GDP is also related to inflation and national interest rates, which affects the mortgage market and therefore also the housing prices and housing market dynamics (Bjørnland, H. C., & Jacobsen, D. H. 2010). A variable for GDP is also included in models by Laamanen, J. P. (2013) and Blanchflower and Oswald (2013). The variable was logged in order to make a model in which marginal changes are interpreted in terms of multiplicative (percentage) changes in the dependent variable.
3.1.5 Commuters
The commuters variable is defined as people who commute for work in another municipality, divided by total working population 1665, so commuters expressed as share of municipality working population. In accordance with the old Accounting identity model commuting patterns are important to control for when modelling on unemployment (Elhorst 2003). Also Coulson and Fisher (2002), L’Horty and Sari (2010) and Flatau et.al. (2002) has variables controlling for commuting or equivalent. Commuting is also related to mobility which is one of the key external causal effects suggested by Oswald (1999), Blanchflower and Oswald (2013) and Laamanen, (2013).
3.1.6 Sociodemographic forces and population
Sociodemographic and population related factors affect economic activities and also the housing market. In order to control for some of these factors a number of variables are used, namely; female workforce participation rate, foreign born workforce participation rate, young (2030 year olds) workforce participation rate, population density and age. Female, foreign born and young work force