Earnings Uncertainty, risk-Aversion and Homeownership

Earnings Uncertainty, Risk-Aversion and Homeownership♣ Luis Diaz-Serrano National University of Ireland Maynooth CREB and IZA Abstract: In this paper we investigate the effect of labour income uncertainty on the probability of homeownership in Germany and Spain. This study is motivated by two facts. Firstly, theoretical models tend to provide ambiguous results in this issue. Secondly, there is limited previous empirical evidence and the existing focuses exclusively on the US housing market. We claim that more international evidence is necessary in order to disentangle this puzzle. We develop a simple theoretical formula that highlights the pivotal role of risk attitudes in the housing tenure decisions that also allow us to introduce the concept of “skewness affection” as a relevant phenomenon. To carry out this test we propose an income uncertainty measure based on panel data labour income equations. We observe that households facing increasing income uncertainty display preference for renting while those located in a positively skewed income distribution show a greater propensity for homeownership. Income uncertainty analysis in housing decisions has important implications for the design of public housing policies and also for the design of private mortgage insurance products. Keywords: Risk aversion, skewness affection, homeownership, earnings uncertainty, transitory shocks in income, credit constraints. JEL classification: D1, R0, J0. Corresponding author: National University of Ireland, Maynooth Department of Economics Co. Kildare, Ireland. E-mail: [email protected] Tel: 353-1-7083728 | Fax: 353-1-7083934 ♣Comments from Donal O’Neil, Alexandrina P. Stoyanova and Olive Sweetman are gratefully acknowledged. 1. Introduction The determinants of the households’ housing tenure choice have been extensively analyzed in the economic literature. Among these determinants, from a theoretical point of view, uncertainties faced by the households during the planning period have received considerable attention. It is well known that housing is a consumption good, for which the amount purchased cannot be easily altered in response to fluctuations in income. As a consequence of this lack of flexibility in the response to income shocks, uncertainty in future income turns out to be one of the most relevant variable in the decision of homeownership. On one hand, borrowers facing greater income uncertainty may suffer severe restrictions to access to the mortgage market. On the other hand, even if they access to the credit market, it could be expected that risk-averse households will try to avoid mortgage downpayments. There are two main sources of labour income uncertainty: unemployment and fluctuations in income due to market forces. While these uncertainties are usually accounted for in most of the theoretical models, there are hardly any empirical tests. While theoretical studies tend to provide ambiguous predictions about the effect of income uncertainty on the housing tenure decision, the limited empirical evidence using US data, suggests that this effect is negative. However, empirical evidence based on one country only is not sufficient to disentangle the puzzle. First of all, both renting and property housing markets do not behave identically in all developed economies. Consequently, when taking their housing decisions, households face different institutional settings (housing, credit and labour market) in each country. These institutional differentials may generate not only different attitudes towards risk, but also differences in the individuals’ perception of risk in itself, i.e. what is perceived as risky in one country is not necessarily perceived as risky in another country. 1 The existing studies of the impact of income uncertainty on the probability of homeownership focus on the variance of future expected income1. In this paper, we question whether the single use of the variance is enough to draw individuals’ risk attitudes. Some recent labour economic literature (see e.g. Hartog and Vijverberg 2002 or Diaz-Serrano et al. 2003) shows that individuals appreciate positively skewed income distributions, even if they offer a smaller mean. This behavior is due to the fact that in these income distributions the chances of reaching high incomes are smaller, but the probability of a big loss is also smaller than in a more symmetric income distribution. In other words, great deviations from the mean are less likely in a more positively skewed distribution. This phenomenon is called “skewness affection”. The hypothesis exposed above is supported by prospect theory (Kahneman and Tversky, 1979, 1991), which states that the individual’s disutility caused by a loss is greater than the utility caused by a gain. According to this reasoning, risk-averse households would feel safer in a positively skewed income distribution than in a symmetric one, even if the more skewed distribution offers a smaller mean. Our conjecture is that risk-averse households exhibiting also “skewness affection” are more likely to purchase their dwelling, since their expectation of a mortgage downpayment is smaller. Hence, we consider that a suitable income uncertainty analysis of the housing tenure choices should account for this effect. In order to capture this behavior, besides using the first two moments of the income distribution (mean and variance), we propose to expand uncertainty analysis on tenure decisions up to the third moment (skewness). We derive this result from a simple theoretical formula and we test it empirically. We contribute to the existing literature in two ways. Firstly, we provide international evidence about the effect of income uncertainty on the probability of homeownership, which can be compared with the limited existing previous evidence from the US. Secondly, by testing 1 This literature will be reviewed in the next section. 2 whether “skewness-affection” is a relevant phenomenon we shed some light in the understanding of how homeownership is planed and achieved by the households. To do so, we estimate a housing tenure choice model for Germany and Spain using different measures of income risk. The reason we choose these two countries for the comparative analysis is based on the fact that they represent the two opposite extremes in terms of labour, credit and housing market performance among the EU countries. Understanding to what extent income (un)certainty, via borrowers’ risk-aversion or lending constraints, is a barrier to homeownership has important implications for the design of public housing policies and also in the design of private mortgage insurance products. On one hand, we ask whether public institutions should subsidize households facing borrowing constraints or under a mortgage downpayment risk. And on the other hand, whether the existing private mortgage payment insurance market is efficient in first, covering the main households’ income risks, and second, offering affordable products to the households who face high rates of income risk The remainder of the paper is structured as follows. Section 2 summarizes the existing literature. In section 3 we briefly overview the institutional context in Germany and Spain. In section 4 we present the theoretical framework used as baseline for our empirical estimation. In section 5 we expose the empirical strategy. The data used is described in section 6. Section 7 reports the empirical results. Finally, section 8 summarizes and concludes. 2. Literature review There exist a vast literature that considers uncertainty in the relevant variables determining the probability of homeownership. Uncertainties in house prices, in future interest rates, in family composition, or in household income is the main focus of this literature. 3 Ioannides (1979) finds theoretical support to the negative relationship between uncertainty in housing prices and the propensity to own for risk-averse households. Henderson and Ioannides (1983) introduce randomness in the capital gains or loses generated by the investment in housing. In their theoretical model they predict that if households can invest in the capital market at a fixed rate, renting becomes more attractive than owning. Using macrodata Rosen et al. (1984) show that increasing uncertainty in the relative price of housing property with respect to the renting costs reduces the share of homeowners in the US. In Neuteboom (2003) a comparison of the costs and risks of mortgages for owner-occupiers is offered. He uses the loan-to-income and loanto-value ratios to show risks faced by households in the mortgage market among a selected group of EU countries. He shows that the risks faced by households among these countries differ markedly. The studies analyzing the effect of income uncertainty on the probability of homeownership are mainly theoretical. DeSalvo and Eeckhoudt (1982) found that the probability of unemployment exerts a negative effect on the homeownership decisions. Fu (1995) states that under the presence of liquidity constraints the theoretical relationship between income uncertainty and housing tenure is ambiguous. Turnbull et al. (1991) also observe an ambiguous relationship and point out that labour income uncertainty could have a non-negative effect if expected labour income embodies a compensating wage differentials for income risk2. Chung and Haurin (2002) specify a theoretical model that also provides ambiguous results. However their simulation study shows that uncertainty in future changes in the influential variables would make households rent. Haurin and Gill (1987), Haurin (1991) and Robst et al. (1999) find empirical 2 The existence of compensating wage differentials for income risk is well documented in the labor economics literature (e.g. King, 1974 and McGoldrick, 1995, for the US, Hartog et al., 2003, for Spain, Germany, The Netherlands and Portugal, and Diaz-Serrano et al., 2003, for Denmark). Al these works provide empirical evidence that reinforces the ambiguity in the relationship between income uncertainty and home ownership derived from their theoretical model. 4 evidence that housing consumption and the probability of homeownership in the US fall when income risk increases. The studies analyzing the effect of price uncertainty on homeownership, both theoretical and empirical in nature, reveal a clear negative effect in this relationship. However, as we show above, studies on the effect of income uncertainty on the probability of homeownership, that we are interested in, tend to display ambiguous results. Moreover, although unequivocal, empirical evidence is still very limited on focuses exclusively on the US housing market. We find this is a strong motivation for the present study. 3. The institutional context in Germany and Spain 3.1. The housing market The German housing market is characterized by the largest private rental sector among the EU countries (see table 1). The percentage of rented dwellings was about 58% in 1995 and 57% in 1999. The German housing legislation has historically supported the design of attractive policies to promote the renting market. Private landlords enjoy generous tax incentives if they offer their properties in the private rental sector, and can receive subsidized loans if they do so in the social rental sector. Local authorities also provide an important amount of urban land for constructing small social rented flats for the youngest and less favored population. This promotion of the rental market is combined with policies that regulate rental prices. Between 1995 and 1999 the housing cost index for privately rented dwellings increased by around 7% only. This scenario explains why in Germany the housing rental market is a strong alternative to homeownership. The efficient management of the rental market is translated into stable property prices. As a consequence is not surprising that the ratio of renters to owner-occupiers remains stable over time. 5 Contrary to Germany, the Spanish housing market displays the highest level of imbalance between tenures among the EU countries. The Spanish rental sector is the smallest in Europe (see table 1). In 1995, only 14% of the dwelling stock was rented, and this percentage fell to 11% in 1999. Unlike Germany, in the last decades, Spanish housing policies have been exclusively addressed to promote homeownership. Neither the supply nor the demand in the rental market has been encouraged in any way. Moreover, the existing 15% personal income tax-relief of the household’s renting costs was abolished with the 1998 Personal Income Tax Reform. As a result of the lack of incentives for landlords to rent Spain reports the second highest dwelling vacancy rate after Greece, slightly above 20%. These inefficiencies in the Spanish housing market have also lead to a dramatic increase in rental prices, between 1995 and 1999 the rent index of private rental dwellings has risen by 24%. In Spain the lack of supply and its low quality, added to the high costs, means that young households prefer to be owner-occupiers than renters, facing an aggressively high degree of indebtedness, even in the beginning of their working careers when their labour income may not be very stable. This scenario partially explains why the propensity to own is so high in Spain. Insert table 1 here 3.2. The credit market During the 1990s, decreasing interest rates and a high competition among lending institutions strongly relaxed the accessibility to the mortgage market in Spain. The 10-15 years of mortgage payment period was extended to 25-30 years, and the amount borrowed covering 80% of the value of the dwelling purchased was augmented to above 100%, covering the value of the dwelling, the derived taxes, and the transaction costs. The average pending time between the 6 demand and the concession of a mortgage was reduced from 3-6 months in the early 1990s to one week in 2000. The unique requirement the Spanish credit institutions impose on their borrowers is that they have an employment. The purchased dwelling serves by itself as a guarantee. These practices in the Spanish mortgage market give to credit impaired households access to a mortgage loan. Although this is a positive policy, there is no public provision of default mortgage protection, and private mortgage insurance products are neither affordable for those most at risk nor they offer a wide cover of the risks. Hence, most of these borrowers have a high probability of a mortgage downpayment if they experience a negative shock in income. Such strong relaxation in the accessibility to the credit market partially explains the “boom” in the Spanish housing market, the second greatest after Ireland among EU countries during the late 1990s. On the contrary, the German mortgage market is generally very conservative. The borrower still remains liable and the lender has recourse to other assets of the borrower, including future earnings. This allows the lender legal recourse to a proportion of the borrower’s salary and other income. The ability to seize a portion of the borrower’s earnings from his employer acts as a strong disincentive for those households with a relatively high probability of default. Given that there is no “exit” option, German borrowers are generally more reluctant to assume and aggressively high levels of indebtedness (in contrast with the Spanish ones). For instance, between 1994 and 1996 the percentage of households that experienced a mortgage downpayment changed from 0.8 to0.5% in Germany, whereas in Spain these numbers moved from 8.2 to 5.6%. 3.2. The labour market The poor performance of the Spanish housing market during the 1990s, in contrast with the German one, also coincides with a poor performance of the Spanish labour market. Although 7 the unemployment rate has fallen from 23% to 15% between 1995 and 1999, this figure still remains the highest in the EU. The consequence of this decrease of the unemployment rate is a dramatic increase of the share of fixed-term workers and the precariousness in wages and working conditions. On the contrary, Germany has reported relatively stable rate of unemployment moving from 8.2% in 1995 to 7.9% in 2000 (see table 1). Another dominant feature of the Spanish labour market is the high level of wage inequality. During the 1990s Spain reported similar levels of inequality to the US, which has been considered in the last decades as the paradigm of the unequal labour market, whereas Germany has reported one of the lowest levels of inequality among developed economies3. The marked differences in the tenure structures of the housing stock, in the housing market policies, in the labour market performance and in the accessibility to the credit market between Spain and Germany mean that uncertainty in labour income and attitudes towards risk might differ substantially between both countries. 4. Theoretical framework: a simple formula The decision of homeownership is usually modeled as a function of household income, the costs of owning compared to costs of renting, and a set of demographic variables (age, household size, etc.). In this section we develop a simple formula that accounts for the effects of labour income uncertainty in the housing tenure choice. We specify a household utility function that depends on owning and renting costs and labour income. Labour income is assumed to be additively decomposable between a deterministic and stochastic component. For the sake of 3 In a cross-country comparison using the Gini index, Bradbury (1993) reported a level of wage inequality of 0.3 for the US and of 0.2 for Germany during the late 1980s and early 1990s. Diaz-Serrano (2001) estimated for the same period a Gini value around 0.29 for Spain. 8 - Individual’s transitory labour income: () 2000 1994 1ˆ exp 7 ii t y ε t ε = =∑ (10) - Individual’s labour income uncertainty: {} 2000 2 2 1994 1ˆ ˆexp( ) 7 iit t i y εε σε = =− ∑ (11) - Individual’s labour income skewness: {} 2000 3 3 1994 1ˆ ˆexp( ) 7 iit t y εε κε = =− ∑i , (12) where the exponential transformation is applied in order to transfer back to money metric the different components of the ln(wit). 6. Data and variable construction 6.1. The dataset The data comes from the European Community Household Panel (ECHP). In this paper we use the waves covering the period 1994-2000 for Germany and Spain. For the first three waves (1994-1996), the ECHP files for Germany contain information coming from both the German Socio Economic Panel (GSOEP) and the ECHP, whereas for the period (1997-2000) the whole sample comes from the GSOEP. For Spain, all the sample period (1994-2000) comes from the original ECHP. The number of surveyed households for Germany ranges from 11,175 in 1994 to 5,693 in 2000, and from 7,206 in 1994 to 5,132 in 2000 for Spain. Table 2 displays the different sample sizes for each wave. The ECHP and the GSOEP contain information not only at household, but also at individual level. The household characteristics that we consider relevant for the present study are the household size and composition, demographic characteristics, income and accommodation. The accommodation questions provide information about the type of dwelling, the year when the household moved there, renting costs and mortgage payments. 15 Besides household information, we also use personal information (age, gender, etc.) and socioeconomic characteristics (employment status, earnings, education, etc.). Insert table 2 here 6.2. Sample restrictions In order to estimate labour income uncertainty, we restrict our sample to the household heads. There are two reasons to impose this restriction. Firstly, as we are interested in income uncertainty arising from labour market forces, to calculate income risk we need to keep out other sources of household income that tend to display a very transitory nature and that have nothing to do with market forces. Secondly, household heads’ labour income is the most important source of income for most of the households. Thus, the role it plays in the owner-occupation decision is more important than that played by other sources of household income. In table 3 we report the share of household head labour income in the overall household income for Germany and Spain. Insert table 3 here Our endogenous variable in the tenure choice equation (7) is a dummy variable that takes 1 if the household is owner-occupier and 0 if the household is a renter. In order to avoid possible bias in the estimated effects of the income variables on the probability of homeownership, we impose two further restrictions. Firstly, we keep out of the sample the households that have purchased the dwelling before the initial survey year, 1994. And second, we do not include the households that are recent owners but do not outstand a mortgage. The aim of these two restrictions is twofold. On one hand, we do not know the households’ characteristics when de 16 tenure decision is made for those who bought their dwelling before 1994. Hence, restricting transitions from renting to owning during the sample period allows matching the observed household characteristics with the tenure status decision. On the other hand, the tenure status of recent owners without a mortgage outstanding has probably nothing to do with the fundamentals and the influential variables assumed to affect the homeownership decision. They may well have inherited the dwelling or have received a free allowance out-of-pocket. Although this is a small fraction of our sample, keeping these households in the sample could obscure the relationship between the income variables and the probability of homeownership. 6.3. Variables The matrix Xi used to estimate household head’s labour income equation (8) contains the following set of variables; gender, education, labour experience and its squared, tenure and its squared, weekly working hours and the type of labour contract. To estimate renting and owning costs used in the housing tenure choice equation (7) we use the real average monthly rent and the real monthly mortgage payments8 computed for each region (NUTS)9. Permanent income (ypi), transitory income (y ε I), income uncertainty ( σ ε i2), and income skewness ( κ ε 3) are estimated as defined in expressions (9) to (12). The remaining variables in the choice equation (7) are defined in table 4. Among them, we account for household head’s unemployment history, household size, household’s capital income, and some household head’s demographic characteristics such age 8 The literature provides alternative ways to compute the owner costs based on subjective appreciation involving mortgage rates, property tax, loan-to-value ratios or house price inflation among others (see e.g. Rosen, 1979; Henderson and Ioannides, 1987; Haurin et al., 1994). Because our data lack most of these variables, we proxy the owning costs by the monthly mortgage payments. 9 The NUTS (nomenclature of territorial units for statistics) classification for Germany is Bader-Wuntterberg, Bayern, Berlin, Brandenburg, Bremen, Hamburg, Hessen, Mecklenburg-Vorpommem, Niedersachsen, NordrheinWestfalen, Rheinland-Pfalz, Saarlan, Sachsen, Sachsen-Anhalt, Schleswig-Holstein, and Thuringen. The Spanish NUTS are North-West (Galicia, Cantabria, Asturias), North-East (Pais Vasco, Navarra, Aragon), Center (CastillaLeon, Castilla-La Mancha, Extremadura), East (Catalunya, Valencia, Baleares), South (Andalucia, Murcia), Canarias and Madrid. 17 and current marital status. In our choice equation we also control for the type of dwelling purchased, and the reason for changing the dwelling if any. Insert table 4 here 7. Estimation results In table 5 we report the sample means for the explanatory variables used in the earnings equation (8) and in the tenure choice equation (7). The sample statistics reveals that household head’s characteristics determining permanent labour income (schooling, experience, tenure, weekly hours worked, and the type of contract) are systematically greater for owners than for renters. Therefore, it seams plausible to expect that household head’s labour income profiles will display different patterns between both tenures status. To test this assumption, we estimate the earnings equation (8) separately for owners and renters. Results are reported in table 6. All coefficients are significant at 1% level and with the expected signs. We also carry out a Chow test of structural change between both tenures. With F-statistic of 127 for Germany we strongly reject the null hypothesis that owners and renters income profiles are equal. In the case of Spain, the F-statistic is only 2.52, but high enough to reject the null hypothesis at 5% significance level. Hence, for both Spain and Germany, we estimate separated measures of permanent and transitory income, income uncertainty and skewness for owners and renters. Insert table 5 here Insert table 6 here 18 Table 7 and 8 contain a descriptive analysis of the intertemporal income uncertainty and skewness measures. We include both types of measures, the ones calculated from labour income equation (8) and also from the standard formula of the CV and skewness (hereafter K) over yearly labour income10. The summary statistics reveal that 2 ˆi ε σ , 3 ˆi ε κ posses enough variability to look for effects in the choice equation (7). As expected, income uncertainty is markedly higher in Spain than in Germany. Time income distributions also tend to be substantially more positively skewed in Spain. This result is indicative of the fact that the labour market performance in both countries is very different. We also find marked intra-country differences between tenures, i.e. systematic lesser income uncertainty and greater positive skewness for owners than for renters. In Germany, income uncertainty is about 35% greater for renters than for owners, whereas for Spain it is about 14% greater. Differences in income skewness between tenures are even more important, in Spain estimated skewness is six times greater for owners than for renters. The same proportion in absolute value is just 2.5 in Germany. Comparisons across population groups also behave according to expectations. As we predict in our theoretical formula (6), these results suggest that income uncertainty and skewness play a substantial role in the house tenure choices in both countries. Insert table 7 here Insert table 8 here 10 The standard formula of the coefficient of variation is CV=σy/µy, and for skewness K=Σ(y-µy)3/Tσy 3, where µy is the time average of yearly income, and σy is the standard deviation. 19 In the second step, intertemporal measures of income uncertainty and skewness are used to estimate the effect of earnings uncertainty and skewness on the discrete choice between renting and owning, we use a pooled cross-section probit11. Since our data set is a panel, in order to get unbiased estimates we select just the last wave the household has participated, and the tenure status is determined in that wave. Results of the probit estimation of the choice equation (7) are reported in table 9. We focus on the results in columns (1) and (2) picking up the effects of the income variables based on the residuals of the income equation (8). In both countries we find a strong negative relationship between labour income uncertainty and the probability of homeownership. As we predict in our theoretical formula, the relationship with earnings skewness is also strongly significant and positive. In table 9 we also distinguish between negative and positive shocks. The variable called TI2 is a dummy variable that takes 1 if during the sample period (1994-2000) the absolute value of the average of the negative shocks is greater than the average of the positive ones. This variable exerts a significant negative effect on the probability of homeownership in both countries, whereas the effect of transitory income is positive. The average marginal effects reported in table 9 reveal that variables related to transitory income (average transitory income, uncertainty and skewness) have a markedly stronger effect in Spain than in Germany, whereas this is reversed for the effect of permanent income. As a control we also report the estimates using the CV as uncertainty measure and K as skewness, both variables are significant and with the expected signs. The remainder variables in the choice equations also behave according to expectations. Employment variables are very important for the tenure status in both countries. Job mobility, 11 The long run nature of housing purchases means that during our sample period households do not experience more than one transition from renting to owning. Therefore we discard the use of panel data estimation, since the potential improvements provided by this technique do not apply to our case. 20 previous unemployment status and long run unemployment (12 or more months) tend to reduce significantly the probability of homeownership. The effect of these variables is, once more, stronger in Spain than in Germany. Public employees are more like to own in Germany, but do not in Spain. Households possessing greater capital income, and those where the household head’s spouse is a wage earner also display a greater propensity to own. User costs are very significant and with the expected signs. Consistent with our theoretical formula, the greater the owner-occupancy cost, the lower the probability of homeownership. The opposite holds for the renting costs. However, this effect is more important across the German regions than in the Spanish ones. Concerning the type of dwelling, homeowners show a strong preference for detached and semi-detached houses. 8. Conclusions and discussion In this paper, we expand the usual income uncertainty analysis (mean-variance) of the housing tenure decisions by deriving the effect of the asymmetry (skewness) in the income distribution on the probability of homeownership. We use a simple theoretical formula that allows determine the crucial role that both variance and skewness in income play on the housing tenure choices. In order to test empirically our theoretical results we perform reduced form equations estimates of the probability of homeownership in Germany and Spain. Our empirical results confirm the predictions derived from the theoretical formula, significant negative effect for risk and positive for skewness. This evidence suggests that households are risk averse and also exhibit “skewness affection” when they plan their home tenure status. Moreover, our results concerning income risk are consistent with previous empirical evidence from the US. The alternative measures of income risk and skewness based on the estimated residuals of earnings panel data equations have also shown a good performance. This confirms our 21 assumptions about the degree of knowledge that individuals have about their future earnings. We also find that unemployment history of the household head exerts a significant effect on the probability of homeownership. Household heads that have been unemployed at least once after 1989 are less likely to own. This effect is augmented if they have experienced long duration unemployment (above 12 months). The fact that our empirical estimations hold in both countries endows our theoretical and empirical results with a good consistency. It is remarkable that even in a so flexible credit market as the Spanish, income uncertainty is still a barrier to homeownership. This circumstance has significant implications for both public and private institutions. Recently there exist the perception that the state pension fund system will experience severe restrictions in the long-run in Germany and Spain. To be more specific, the forecast about future demographic evolution and on future social security contributions, predict a crash of the Spanish pension system about the year 2020. Undoubtedly, homeownership is one of the best hedging tools against such a pessimistic scenario. In this context, we claim that more active public housing welfare policies, like for instance the provision of public mortgage downpayment assistance, are necessary in order to promote homeownership among households facing higher rates of income uncertainty. In 1999 the German government approved a new insolvency law that relaxed the borrower’s degree of liability by allowing himself to declare in “private bankruptcy”. This situation permits to an overly indebted borrower solve the outstanding financial obligations by a court settlement in case it is not possible do it out of the court. It represents a first step for relaxing the extremely conservative German credit market that can encourage to German borrowers to assume higher levels indebtedness. Undoubtedly, it is a small door open in order to promote accessibility to mortgage loans for those with more volatile income, but maybe not 22 enough to reach a more desirable levels of homeownership that allow most of the households to face up to the expected future restrictions in the German pension system. The corollary of our results also applies to the existing private mortgage payment insurance policies offered by credit institutions. Pryce and Keoghan (2002) find empirical evidence for the UK that mortgage borrowers most at risk are less likely to undertake such private mortgage insurance products. According to this, it could be that credit institutions need to redesign their mortgage protection instruments in order to make it more affordable and cover a wider range of households’ financial risks. The understanding to what extent the (in)accessibility to a mortgage loan is driven by credit constraints or by households’ aversion to the risk of a mortgage downpayment is still a relevant question to be answered. This exercise would require suitable data on households’ credit quality constraints (see e.g. Rosenthal 2002, or Barakova et al. 2003). It is plausible to expect a positive relationship between constraints in the accessibility to mortgage loans and households’ income uncertainty in a conservative credit market like the German one, however, but not necessarily in a flexible credit market like the Spanish one. Although we find this could be a very interesting extension of this work, it surpasses the goals of this paper. Hence, further research on this issue is encouraged. 23 References Barakova, I.; R.W. Bostic; P.S. Calem and S.M. Wachter (2003), Does credit quality matter for homeownership?, Journal of Housing Economics 12, pp. 318-336. Bradbury, B. (1993): Male pre and tax wage inequality: a six country comparison, The Luxembourg Income Study working paper 90. Luxembourg: CEPS INSTEAD. Chung, E.C. and D.R. Hauring (2002), Housing choices and uncertainty: the impact of stochastic events, Journal of Urban Economics 52, 193-216. CIRIEC (2003), Housing statistics in the EU 2002, International Centre for Research and Information on the Public and Cooperative Economy, University of Liege, Belgium. DeSalvo, J.S. and L.R. Eeckhoudt (1982), The Effect of Unemployment Risk on Consumption Behavior, Zeitschrift fur Nationalokonomie 42, 411-418. Diaz-Serrano, L.; J. Hartog and H.S. Nielsen (2003), Compensating wage differentials for schooling risk in Denmark, IZA discussion papers #963, Bonn, Germany. Fu, Y. (1995), Uncertainty, liquidity, and housing choices, Regional Science and Urban Economics 25, 223-236. Green, G.; J. Coder and P. Ryscavage (1992), International Comparisons of Earnings Inequality for Men in the 1980s, Review of Income and Wealth 38, pp. 1-15. Hartog, J. and W. Vijverberg (2002), Do wages really compensate for risk aversion and skewness affection, IZA discussion papers #426, Bonn, Germany. Hartog; J.; E.J.S. Plug, L. Diaz-Serrano and A.J.C. Vieira, (2003), Risk compensation in wages: a replication, Empirical Economics 28, pp. 639-647. Haurin, D.R. and H.L. Gill (1987), Effects of income variability on the demand for owneroccupied housing, Journal of Urban Economics 22, 136-150. 24 Table 5: Sample means (1994-2000) Germany Spain Renters Owners Renters Owners Household head demographics Female 0.352 0.219 0.154 0.088 (Age) 40.149 45.773 41.493 46.670 Years of schooling 12.630 13.265 10.610 10.337 (Married) 0.584 0.851 0.687 0.870 Household head's spouse working (Wspouse) 0.490 0.623 0.319 0.343 Household size (Hsize) 2.630 3.200 3.266 3.736 Housdehold head employment Labour experience 21.432 27.131 24.115 29.941 Firm tenure 8.390 11.819 9.550 13.778 No fixed-term contract 0.553 0.569 0.356 0.433 Weekly hours worked 41.070 43.218 43.717 45.049 Public worker (Public) 0.203 0.253 0.211 0.188 Household head unemployment At least once unemployed after 1989 (Unemp89) 0.319 0.174 0.482 0.337 At least once long unemployed after 1989 (Longun89) 0.142 0.066 0.248 0.151 Reason for the last change of dwelling Job motives (Jobmobil) 0.045 0.007 0.058 0.010 House motives 0.184 0.151 0.139 0.076 Personal motives (Permobil) 0.108 0.039 0.142 0.044 Type of dwelling Detached or semi-detached house 0.171 0.659 0.146 0.338 (Flat) 0.671 0.166 0.845 0.657 Occupancy costs Owner-occupancy costs 619 € 602 € 331 € 326 € Renting costs 417 € 418 € 170 € 168 € Income Household head's net yearly labour income 19,456 € 25,553 € 14,338 € 15,533 € Total household net income 27,847 € 36,648 € 17,225 € 20,689 € Household capital income 417 € 854 € 286 € 589 € # of individuals 5,668 3,950 832 4,044 # of observations 20,505 14,770 2,323 16,658 Source: Own computations based on the ECHP. Notes: The names in parenthesis refer to the variables defined in table 4. 31 Table 6: Panel data random effects estimates of labour income equation (7) Germany Spain Owners Renters Pooled sample Owners Renters Pooled sample Constant 8.2875 (68.91) 7.7986 (81.57) 7.9459 (107.04) 12.9485 (103.46) 12.0887 (41.04) 12.8522 (111.69) Years of schooling 0.0399 (11.92) 0.0354 (12.93) 0.0375 (17.29) 0.0518 (21.22) 0.0618 (10.18) 0.0518 (22.47) Experience 0.0152 (3.90) 0.0236 (9.23) 0.0242 (11.55) 0.0253 (6.68) 0.0120 (1.93) 0.0241 (7.10) Experience squared -0.0003 (-4.95) -0.0005 (-9.33) -0.0005 (-11.78) -0.0004 (-7.09) -0.0003 (-1.95) -0.0004 (-7.48) Firm tenure 0.0456 (12.00) 0.0773 (24.5) 0.0635 (26.21) 0.0578 (13.07) 0.0831 (7.59) 0.0593 (14.44) Firm tenure squared -0.0013 (-7.75) -0.0025 (-18.18) -0.0020 (-18.53) -0.0018 (-9.68) -0.0024 (-5.09) -0.0018 (-10.46) log(weekly hours) 0.3774 (14.05) 0.4270 (18.44) 0.4039 (23.05) 0.0805 (2.83) 0.2905 (4.04) 0.1050 (3.97) No fixed-term contract 0.0483 (5.44) 0.0502 (6.34) 0.0465 (7.90) 0.1394 (11.31) 0.0912 (2.65) 0.1311 (11.34) Female -0.5136 (-19.23) -0.2774 (-15.46) -0.3775 (-24.17) -0.3512 (-9.64) -0.2210 (-3.25) -0.3320 (-9.99) # of individuals 3,270 4,242 6,949 3,355 674 3,740 # of observations 11,439 13,490 25,379 13,252 1,784 15,036 Source: Own computations based on the ECHP. 32 Table 7: Estimates of household head’s labour income uncertainty by selected groups Income uncertainty and skewness based on residuals equation (7) Income uncertainty based on CV and K on yearly income Germany Spain Germany Spain Owners Renters Owners Renters Owners Renters Owners Renters Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Total 0.188 0.253 0.255 0.402 0.336 0.330 0.384 0.402 0.222 0.239 0.261 0.270 0.290 0.258 0.364 0.311 Age group 25 to 35 0.172 0.192 0.298 0.481 0.342 0.348 0.476 0.535 0.261 0.268 0.293 0.285 0.292 0.287 0.333 0.314 35 to 45 0.204 0.265 0.254 0.372 0.338 0.316 0.295 0.238 0.210 0.216 0.239 0.254 0.291 0.240 0.351 0.322 45 to 55 0.189 0.327 0.210 0.287 0.354 0.354 0.398 0.391 0.178 0.232 0.208 0.239 0.294 0.255 0.372 0.277 55 to 65 0.151 0.219 0.196 0.375 0.214 0.147 0.263 0.266 0.203 0.237 0.254 0.249 0.185 0.133 0.417 0.324 Sector of employment Private 0.205 0.267 0.265 0.384 0.383 0.361 0.437 0.443 0.245 0.249 0.276 0.274 0.334 0.270 0.418 0.319 Public 0.144 0.209 0.223 0.407 0.197 0.147 0.232 0.184 0.153 0.196 0.203 0.249 0.148 0.142 0.162 0.167 Unemployed. after 1989 No 0.172 0.245 0.214 0.346 0.294 0.295 0.319 0.412 0.186 0.212 0.209 0.249 0.241 0.238 0.252 0.258 Yes 0.239 0.273 0.329 0.480 0.392 0.366 0.462 0.379 0.327 0.281 0.349 0.283 0.349 0.270 0.456 0.322 Marital status Not married 0.166 0.187 0.264 0.352 0.334 0.339 0.413 0.510 0.198 0.203 0.280 0.283 0.307 0.287 0.399 0.355 Married 0.193 0.266 0.248 0.433 0.336 0.330 0.366 0.322 0.227 0.246 0.248 0.260 0.287 0.253 0.341 0.277 Household type Single 0.168 0.208 0.239 0.348 0.281 0.210 0.346 0.487 0.191 0.211 0.246 0.262 0.243 0.195 0.403 0.399 Single w. kids 0.122 0.067 0.299 0.372 0.368 0.289 0.304 0.195 0.183 0.222 0.346 0.325 0.291 0.231 0.341 0.284 Couple 0.177 0.226 0.251 0.346 0.365 0.456 0.408 0.500 0.163 0.179 0.266 0.274 0.311 0.326 0.293 0.256 Couple w. kids 0.196 0.272 0.258 0.466 0.331 0.309 0.418 0.364 0.244 0.257 0.250 0.262 0.290 0.253 0.372 0.293 Source: Own computations based on the ECHP. 33 Table 8: Estimates of household head’s labour income uncertainty by selected groups Income uncertainty and skewness based on residuals equation (7) Income uncertainty based on CV and K on yearly income Germany Spain Germany Spain Owners Renters Owners Renters Owners Renters Owners Renters Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. Total 0.010 0.198 -0.004 0.260 0,144 0,850 0,022 0,600 -0.010 0.714 -0.091 0.639 0.139 0.704 0.026 0.580 Age group 25 to 35 -0.004 0.102 -0.004 0.288 0,125 0,890 0,091 0,602 -0.018 0.665 -0.069 0.610 0.135 0.731 0.035 0.495 35 to 45 0.022 0.288 -0.006 0.308 0,171 0,886 0,028 0,757 0.025 0.751 -0.069 0.657 0.110 0.660 0.023 0.671 45 to 55 0.008 0.048 -0.002 0.188 0,181 0,907 0,002 0,505 -0.072 0.715 -0.086 0.662 0.187 0.770 -0.007 0.571 55 to 65 0.026 0.147 -0.002 0.256 0,016 0,059 -0,080 0,521 -0.003 0.753 -0.189 0.686 0.250 0.732 0.136 0.586 Sector of employment Private 0.017 0.220 -0.002 0.268 0,175 0,948 0,025 0,657 -0.021 0.706 -0.098 0.643 0.145 0.719 0.013 0.590 Public -0.010 0.102 -0.012 0.224 0,044 0,353 0,006 0,028 0.023 0.737 -0.067 0.625 0.121 0.656 0.085 0.538 Unemployed after 1989 No 0.012 0.204 -0.004 0.204 0,132 0,748 0,007 0,694 0.034 0.694 -0.059 0.637 0.239 0.711 0.074 0.618 Yes 0.006 0.178 -0.005 0.335 0,159 0,963 0,038 0,489 -0.139 0.758 -0.146 0.639 0.017 0.677 -0.016 0.543 Marital status Not married 0.043 0.354 -0.003 0.279 0,145 0,872 0,060 0,610 -0.013 0.680 -0.087 0.630 0.102 0.706 0.072 0.643 Married 0.003 0.140 -0.005 0.245 0,144 0,848 0,001 0,593 -0.009 0.722 -0.094 0.646 0.146 0.704 0.002 0.545 Household type Single 0.000 0.040 0.001 0.274 0,202 1,025 0,057 0,671 0.087 0.717 -0.093 0.635 0.021 0.771 0.045 0.676 Single (kids) 0.003 0.008 -0.012 0.302 0,003 0,172 0,149 0,790 0.041 0.656 -0.067 0.643 0.085 0.730 0.284 0.670 Couple 0.021 0.164 -0.003 0.265 0,060 0,635 -0,077 0,517 -0.039 0.696 -0.105 0.621 0.248 0.702 0.020 0.506 Couple (kids) 0.008 0.221 -0.006 0.252 0,161 0,897 0,036 0,570 -0.006 0.723 -0.083 0.651 0.152 0.700 -0.018 0.516 Source: Own computations based on the ECHP. 34 Table 9: Probit with robust standard error estimates of the housing tenure choice equation (6). Germany Spain (1) (2) (3) (4) (1) (2) (3) (4) Cosntant -5.1233 (-6.97) -5.0171 (-6.86) -3.7191 (-6.83) -3.7048 (-6.66) -1.4037 (-1.44) -1.6173 (-1.56) 0.5580 (0.55) 0.4789 (0.46) H. Head labour income Income uncertainty -0.6955 (-3.76) -0.1162 -0.9487 (-4.31) -0.1565 -0.1815 (-1.65) -0.0284 -0.1981 (-1.75) -0.0309 -0.6775 (-2.99) -0.2634 -0.9163 (-3.35) -0.3571 -0.4526 (-2.43) -0.1795 -0.5161 (-2.76) -0.2047 Income skewness 0.2232 (3.04) 0.0368 0.0751 (1.87) 0.011 7 0.2238 (1.87) 0.0872 0.1906 (2.48) 0.0756 Transitory income 0.9384 (3.14) 0.156 7 1.4999 (2.91) 0.2474 0.8909 (1.72) 0.3464 1.2747 (2.10) 0.4968 Permanent income/1000 0.0308 (8.05) 0.0051 0.0301 (7.79) 0.0050 0.0001 (1.77) 2.5·10-5 0.0001 (1.65) 2.3·10-5 Average negative shocks greater than average positive shocks (TI2) -0.2065 (-3.07) -0.032 6 -0.2177 (-3.11) -0.0338 -0.5518 (-2.05) -0.2173 -0.6205 (-2.27) -0.2432 Yearly income/1000 0.0061 (6.70) 0.0009 0.0060 (4.95) 0.0009 0.0001 (2.86) 4.0·10-5 0.0001 (2.74) 3.9·10-5 H. Head characteristics (Married) 0.3313 (3.81) 0.0525 0.3352 (3.85) 0.0525 0.4114 (5.38) 0.0605 0.4117 (5.40) 0.0603 0.5238 (3.37) 0.2059 0.5314 (3.37) 0.2091 0.5658 (4.09) 0.222 6 0.5871 (3.66) 0.2307 (Age) 0.0542 (1.84) 0.0091 0.0506 (1.70) 0.0083 0.0704 (3.07) 0.0110 0.0698 (2.92) 0.0109 0.0801 (1.65) 0.0311 0.0730 (1.49) 0.0285 0.0179 (0.42) 0.0071 0.0192 (0.45) 0.0076 Age squared -0.0007 (-2.20) -0.0001 -0.0007 (-2.1) -0.0001 -0.0009* (-3.38) -0.0001 -0.0009 (-3.17) -0.0001 -0.0011 (-2.07) -0.0004 -0.0010 (-1.89) -0.0004 0.0179 (0.42) -0.0002 -0.0004 (-0.93) -0.0002 H. Head employment Unemployed after 1989 (Unemp89) -0.1863 (-2.65) -0.0299 -0.1682 (-2.37) -0.026 7 -0.2732 (-4.41) -0.0405 -0.2634 (-4.21) -0.0390 -0.0647 (-1.80) -0.0250 -0.0607 (-1.69) -0.023 7 -0.2671 (-2.26) -0.105 7 -0.2396 (-2.01) -0.0948 Long term unemp. After 1989 (Longun89) -0.2740 (-3.06) -0.040 6 -0.2862 (-3.16) -0.041 6 -0.2051 (-2.52) -0.0291 -0.2119 (-2.62) -0.0299 -0.2802 (-1.89) -0.1103 -0.2743 (-1.87) -0.125 6 -0.2734 (-1.84) -0.108 7 -0.2938 (-1.98) -0.1168 Public worker (Public) 0.1271 (1.86) 0.0221 0.1296 (1.88) 0.0223 0.1547 (2.46) 0.025 7 0.1538 (2.47) 0.0255 0.0726 (0.56) 0.0281 0.0749 (0.57) 0.0290 0.0992 (0.78) 0.0392 0.0903 (0.71) 0.0357 35 36 Table 9: Continuation Germany Spain (1) (2) (3) (4) (1) (2) (3) (4) Household variables Capital ncome/1000 0.0185 (2.89) 0.0031 0.0181 (2.79) 0.0030 0.0189 (2.77) 0.0030 0.0187 (3.04) 0.0029 0.0010 (1.86) 0.0004 0.0009 (1.84) 0.0004 0.0011 (1.72) 0.0004 0.0011 (1.71) 0.0004 Household size (Hsize) 0.0504 (1.82) 0.0084 0.0492 (1.78) 0.0081 0.0599 (2.38) 0.0094 0.0597 (2.52) 0.0093 0.0247 (0.51) 0.009 6 0.0271 (0.54) 0.0105 0.0213 (0.51) 0.0084 0.0240 (0.57) 0.0095 H. Head's spouse working (Wspouse) 0.1322 (1.98) 0.0221 0.1379 (2.06) 0.022 7 0.1527 (2.61) 0.0240 0.1536 (2.61) 0.0093 0.3141 (2.74) 0.1208 0.3289 (2.83) 0.1268 0.2946 (2.81) 0.1159 0.3018 (2.87) 0.1187 Costs Owner-occupancy costs -0.0011 (-5.14) -0.0002 -0.0181 (-5.12) -0.0002 -0.0010 (-5.67) -0.0002 -0.0010 (-5.72) -0.0002 -0.0001 (-3.68) 9.5·10-5 -4.9·10-5 (-3.60) -1.9·10-5 -4.1·10-5 (-3.19) -1.6·10-5 -4.1·10-5 (-3.26) -1.6·10-5 Renting costs 0.0034 (6.16) 0.000 6 0.0034 (6.18) 0.000 6 0.0031 (6.25) 0.0005 0.0031 (6.41) 0.0005 0.0001 (4.35) 2.9·10-5 0.0001 (4.26) 2.8·10-5 0.0001 (3.67) 2.4·10-5 0.0001 (3.67) 2.4·10-5 New dwelling is a flat (flat) -1.0844 (-18.18) -0.2211 -1.0791 (-17.79) -0.2179 -1.0658 (-19.74) -0.2082 -1.0664 (-20.17) -0.2080 -0.7419 (-5.60) -0.2663 -0.7483 (-5.52) -0.2694 -0.7298 (-6.01) -0.2731 -0.7309 (-5.98) -0.2734 Motive of last change of dwelling Job motives (Jobmobil) -0.4189 (-3.17) -0.054 6 -0.4280 (-3.20) -0.0548 -0.3993 (-3.15) -0.0489 -0.4069 (-3.31) -0.0494 -0.4391 (-2.12) -0.173 6 -0.4214 (-2.00) -0.1668 -0.4534 (-2.32) -0.178 7 -0.4500 (-2.33) -0.1774 Personal motives (Permobil) 0.2605 (2.75) 0.0495 0.2263 (2.35) 0.0418 0.2467 (2.97) 0.0438 0.2502 (2.89) 0.0444 0.0956 (0.66) 0.0369 0.1087 (0.75) 0.0420 0.2239 (1.69) 0.0878 0.2352 (1.78) 0.0921 Sample size 3,630 4,749 668 774 Source: Own computations based on the ECHP. Notes: Endogenous variable: 1 if owner, 0 if renter. Columns (1) and (2): Uncertainty, skewness, transitory and permanent labour income estimated from income equation (7), and measured as (8) to (11). Columns (3) and (4): Uncertainty and skewness measured as the CV and K. Z-values in parenthesis; elasticities in italic font. The names in parenthesis refer to the variables defined in table 4.