Subjective Financial Situation and Overall Life Satisfaction: A Joint Modelling Approach

Subjective Financial Situation and Overall Life Satisfaction: A Joint Modelling Approach Daniel Gray: daniel.gray@sheffield.ac.uk University of Sheffield Abstract Analysing the German Socio-Economic Panel Survey (GSOEP), this study aims to ascertain the relationship between the household s financial position and overall life satisfaction. Existing studies which explore the relationship between household finances and well-being focus on the household s income. Income, however, is arguably an imperfect measure of a household s financial resources. As a result, this study explores the link between a variety of monetary and subjective financial measures and overall life satisfaction. Within a fixed effects framework, it is the subjective, opposed to monetary, financial measures which have a direct impact on overall life satisfaction. Once a recursive bivariate ordered probit specification is implemented to account for the potential endogeneity of the subjective financial measures, the results indicate that the subjective financial position mediates the effects between the household s monetary financial position and overall life satisfaction. Keywords: Well-Being, Household Finances, Ordered Logit, Bivariate Ordered Probit JEL codes: D14, I10 1

1 Introduction The topic of overall life satisfaction has received a large amount of attention from a variety of academic disciplines in the past three decades, including psychology and economic, but also from a wider public audience. Measures of overall life satisfaction are being proposed as a measure of an economies progress. It is argued that measures of overall life satisfaction capture information beyond that contained in more traditional measures of an economies progress, such as GDP, and should therefore be used in conjunction with these to inform and evaluate public policy. This idea is being replicated across the world and consequently, it is important to fully understand the determinants of overall life satisfaction. One area of an individual s life that could potentially have a dramatic impact on their level of well-being is their household s financial situation. In the past two decades, there has been a significant increase in the level of household debt within developed countries. Following significant reforms to the credit market in the late 1980s and early 1990s, household debt dramatically increased over the subsequent decades and until recently, this increase in debt levels was in conjunction with a period of sustained growth. Debt levels and the general financial position of the household, could potentially have a significant impact on the level of well-being of individual s. Hence, this study will explore the effects the household s financial situation has on individual well-being from an empirical perspective. The determinants of overall life satisfaction have been explored in a variety of contexts, see for example Dolan et al. (2008), Clark et al. (2008), MacKerron (2012) and Stutzer and Frey (2010) who provide a substantial review of the well-being literature. Within the existing literature a vast quantity of these studies explore the impact of income, with many using income as a proxy for an individuals level of financial resources. However, there remains a relative small number of studies which consider the impact of other monetary measures of the households financial resources on overall life satisfaction. The household s levels of assets, debt and net wealth arguably capture different aspects of the 2

household financial position to their income. Despite the dramatically changing composition of the households asset and debt portfolios in recent decades, the analysis of its effects on overall life satisfaction remains relatively sparse. Within the existing literature there exists a limited number of studies that control for the household s level of net wealth, debt levels or asset levels. For example, Headey and Wooden (2004), analysed the 2002 wave of the Household Income and Labour Dynamics in Australia (HILDA) survey and made a distinction between levels of well-being and ill-being. The authors found that household net wealth is as important as income in determining an individual s level of well-being and ill-being. Similarly, Headey et al. (2008), using a fixed effects linear model, found that net wealth is a statistically significant determinant of overall life satisfaction in the Netherlands and Hungary. Similarly, Brown et al. (2005) analysed the 2005 wave of the British Household Panel Survey (BHPS), via an ordered probit model, and found that it is levels of unsecured debt, opposed to secured debt, which influences an individuals level of psychological well-being, as measured by the GHQ12 score. In the USA, Drentea (2000) showed, using a sample of individuals from Ohio, that anxiety is positively related to debt levels and the debt to income ratio. Keese and Schmitz (2010) assess the effect of household indebtedness on a variety of different health measures for Germany using the GSOEP from between 1999 and 2009. The study considers the effect of indebtedness on three health measures, namely, general health satisfaction, mental health and obesity. The authors exploit three measures of indebtedness: the relative burden of loan repayments on the household budget; the income dedicated to loan repayments; and the ratio of credit card repayments to the household s net income. Once individual fixed effects are accounted for, household debt displays a strong negative relationship with overall health satisfaction and mental well-being. The level of debt, however, is not found to be a significant determinant of obesity. This once again outlines the importance of the accounting for individual fixed effects. Current studies which explore relationship between subjective financial measures and well-being, consistently find that subjective measures are important determinants of individual well-being. For example, Bridges and Disney (2010) explored the link between 3

the likelihood of reporting depression and a variety of objective and subjective debt measures in Britain using the Family and Children Survey. It is found that it is the subjective, rather than the objective, debt measures which have a direct impact on the likelihood of reporting depression. The study found, within a bivariate probit model, that the level of debt influences the likelihood of reporting being depressed through the subjective debt measures. Similarly, Reading and Reynolds (2001) find that self-reported debt problems are associated with higher levels of maternal depression. Using the BHPS, Wildman (2003), find that self-reported financial status, as well as expected financial position are positively related to self-reported health measures. Of the studies outline above which consider the household s level of assets, debt and net wealth, tend only to consider cross-sectional data and, therefore, do not account for individual heterogeneity 1. Following Ferrer-i Carbonell and Frijters (2004), when analysing subjective well-being measures, it is important to account for individual heterogeneity as this can have a significant impact on the parameter estimates. In addition, the analysis presented in this paper also accounts for the potential endogenety of the subjective financial position of the household in the determinants of overall life satisfaction. Within several studies, for example Wildman (2003), Reading and Reynolds (2001) and Bridges and Disney (2010), reported that the subjective, rather than the objective measures which influence a variety of well-being measures. This study, following the univariate analysis, attempts to account for the potential endogeneity by jointly modelling overall life satisfaction and subjective financial position. Analysing the 2002 and 2007 waves of the GSOEP the analysis presented in this paper builds on the existing literature in several ways. Initially, it provides a longitudinal analysis of overall life satisfaction whilst controlling for the household s level of net wealth, assets and debt. In addition, the analysis accounts for the head of household s subjective financial position. The study then attempts to account for the potential problem of endogeneity by jointly estimating the relationship between overall life satisfaction and subjective financial position by a bivariate recursive ordered probit model following Greene 1 One exception to this is Headey et al. (2008) who employed a fixed effects framework to analyse the impact of net wealth on overall life satisfaction in Hungary and the Netherlands 4

and Hensher (2010). Within this specification, individual heterogeneity is accounted for by employing a Mundlak transformation. The univariate results indicate that, the level of household assets and net wealth are positively related to overall life satisfaction. In line with Brown et al. (2005), we find that it is the unsecured, opposed to secured, debt which has an influence on overall life satisfaction. The subjective financial position has the expected impact, with individuals who are concerned with their economic situation reporting lower levels of overall life satisfaction. The joint modelling approach reveals that the subjective financial position appears to mediated the effects of the household s level of assets and debts, in addition, to the effects of unemployment and household income. The remainder of this paper is structured as follows. The next Section 2 outlines the data with Section 3 presenting the methodology employed in this paper. Section 4 outlines the results obtained and finally section 5 concludes. 2 Data This study draws upon the the German Socio-Economic Panel Survey (GSOEP). The GSOEP is a nationally representative panel survey of private households that commenced in West Germany in 1984 in which every household member above the age of 16 was interviewed. The survey was extended in 1990 to include East Germany. Wealth measures were included in the 2002 and 2007 waves of the GSOEP, and consequently are the focus of the study. The GSOEP asks respondents about the value of their property, financial assets, life insurance, business assets and tangible assets. The GSOEP also asks participating individuals about their outstanding debts which makes it possible to construct a variety of financial measures. Following Bertaut and Haliassos (2001) and subsequently Brown et al. (2005), the analysis focuses on the head of household, which is defined as the individual in the household who best knows how the household acts under general conditions, as a result these are likely to bear the consequences of the household s 5

financial position. This results in a balanced panel of 7,714 household heads. Following Dolan et al. (2008) and MacKerron (2012), this study analyses a single item measure of overall life satisfaction which is now widely used in the existing literature. The dependent variable is based upon the question, How satisfied are you with your life, all things considered? This is measured on an 11 point scale where 0 indicates completely dissatisfied and 10 represents completely satisfied. Clearly, this is ordinal in nature and so a fixed effects ordered logit model is employed to ascertain the determinants of overall life satisfaction. This approach accounts for the ordinal nature of the dependent variable and also controlling for individual heterogeneity. Table 1 and Figure 1 shows the summary statistics and distribution of the head of household s level of overall life satisfaction. In line with Dolan et al. (2008), the distribution of overall life satisfaction is highly skewed with the majority of individuals tending to report higher levels of overall life satisfaction. The subjective financial position for Germany, in line with Delken (2008) and Hofmann and Hohmeyer (2013), is based on the the question What is your attitude towards the following areas - are you concerned about them? Your own economic situation. The three possible responses to this question were not at all concerned, concerned and very concerned. This variable is measured on an ordinal scale where zero indicates not at all concerned and two represents very concerned. Once again the due to the dependent variable being defined on an ordinal scale, we apply a ordered choice model in order to take account of this. The subjective financial measures are initially included as explanatory variables in the univariate overall life satisfaction analysis. Within these specifications, not at all concerned is defined to be the omitted category, whilst concerned and very concerned are included in the analysis. The summary statistics and distributions of the subjective financial measures are presented in Table 1 and Figure 2. The average subjective financial position score is given to be 0.898, with 48.7% and 20.0% of household heads reporting concerned and very concerned respectively. 6

It is argued that the subjective financial position will capture information beyond that contained in the monetary financial measures. One potential explanation is that the head of households subjective financial position captures the relative position of the head of household. Equally, it could capture the level of control the individual feels they possess over their current financial position. As a result, the level of control the head of household feels maybe closely related to the head of households level of financial knowledge they possess. There is a vast literature which aims to explore the relationship between income and overall life satisfaction, see Ferrer-i Carbonell (2005) and Clark et al. (2008), however, income is arguably not the best indicator of the households financial resources. For example, household s with low incomes but high levels of net wealth could smooth their level of overall life satisfaction by drawing upon savings. Consequently, in line with Brown et al. (2005), Headey and Wooden (2004) and Headey et al. (2008), we control for a variety of monetary financial measures. These are namely the household s total assets, total debt, the level of unsecured and secured debt and the households level of net wealth. This division of net wealth into its constituent parts of total assets and total debt, and total debt further into secured and unsecured debt, will alow the exploration whether different assets and debts have differential impacts on overall life satisfaction. Following Brown and Taylor (2008), the level of total assets held by the household is given by the summation of the the household s financial assets, tangible assets and the current value of any property owned. The household s level of secured debt is generated from the question If you still have a loan taken out on your house/apartment, how high is the remaining debt (excluding interest)? and clearly refers to the value of any outstanding debt secured against any property owned. The level of unsecured debt is defined to be any outstanding debt, other than secured debt. This is generated from the question Leaving aside any mortgages on house or property or house-building loan: Do you currently still owe money on loans that you personally were granted by a bank, other organization, or private individual, and for which you personally are liable? How high are your outstanding debts? Total debt given by the summation of the unsecured and secured debt whilst the 7

household s level of net wealth is defined to be the the households total assets minus their total debt. In line with the existing literature the natural logarithm is taken in order to account for the skewed nature of the variables. All financial variable are inflated to the 2007 price levels. Figures 3 to 6 present the distributions of the household financial positions. Based upon the existing literature a variety of other socio-economic and demographic characteristics are controlled for in the analysis presented. This study where applicable, controls for the gender, age and age squared of the head of household. The natural logarithm of household net income is included in addition to the head of households highest level of education and the natural logarithm of household size. It also controls for the head of households labour market status, relationship status and self-assessed health status are all controlled for in both the overall life satisfaction equations and subjective financial position analysis. In addition to the variables outlined previously, in line with Joo and Grable (2004), the risk attitudes of the head of household are controlled for, as it is argued that the they potentially capture the level of financial knowledge possessed by the household head. In line with Ferrer-i Carbonell and Ramos (2010) and Brown et al. (2013), we assume that individual risk attitudes are time invariant. As a result, we match information contained in the 2004 wave of the GSOEP relating to the general risk attitude question to heads of households in the 2002 and 2007 waves of the GSOEP. Following Dohmen et al. (2005) we collapes this eleven point scale into a binary variable, where 1 indicates risk tolerance and 0 indicates risk aversion. Table 1 presents the summary statistics for dependent and independent variables. 8

3 Methodology 3.1 Fixed Effects Ordered Logit Model The univariate analysis presented in this study employs the methodology proposed by Baetschmann et al. (2011), that is the Fixed effects ordered logit model. Following Ferrer-i Carbonell and Frijters (2004) it is important to account for individual heterogeneity when analysing subjective well-being measures. The analysis presented in this study employs the method proposed by Baetschmann et al. (2011). The underlying model is based upon the latent variable model: y it = x itβ + α i + ɛ it, i = 1,..., N, t = 1,..., T (1) where yit is a latent measure of the i th head of household s overall life satisfaction in period t, x it is the vector of observable characteristics, and β is a vector of coefficients to be estimated. α i is a time invariant unobserved component and ɛ it in a white noise error term. What is observed is y it : y it = k if µ k < y it < µ k+1, k = 1,..., K. (2) Where the threshold parameters µ k are assumed to be strictly increasing for all values of k, and µ 1 = - and µ K+1 = +. It is assumed that the white noise error term, ɛ it is IID by the logistic distribution, it follow that the probability of observing outcome k for individual i in time period t is given to be: P r(y it = k x it, α i = Λ(µ k+1 x itβ α i ) Λ(µ k x itβ α i ) (3) where Λ(.) represents the cumulative logistic distribution. Baetschmann et al. (2011) outline two problems with direct maximum likelihood estimation of the model above. The first is a problem with identification. The component µ i,k 9

cannot be separated from α i, only µ i,k - α i = α i,k can be identified and, therefore, can be estimated consistently for instances where T tends towards infinity. Secondly, under the assumption of fixed-t asymptotics, due to an incidental parameter problem, cannot be estimated consistently. As a consequence, in short panels, this can result in substantially biased estimators of the coefficients, ˆβ, as stated in Greene and Zhang (2003). To consistently estimate the coefficients of β, it is required that the J levels of are required to be dichotomized, that is collapsed into binary outcomes. Within this chapter, the procedure outlined in Baetschmann et al. (2011) is used to estimate the coefficients. The method proposed in Baetschmann et al. (2011) jointly estimates all dichotomizations. A full proof is provided in Baetschmann et al. (2011). This estimation method is called the Blow-Up and Cluster (BUC) estimator. The estimator initially blows up the sample size by replacing every observation in the sample by K-1 copies of itself, and then dichotomises every K 1 copy of the individual at a different cut off point. The Conditional Maximum Likelihood logit is then estimated using the entire sample, giving the BUC estimates. The standard errors are clustered at the individual level as some individuals can potentially contribute several terms to the log-likelihood function. The BUC estimator avoids the problem of small sample sizes associated with cut off values. The fixed effects ordered logit model is implemented in STATA using the buclogit command proposed by Dickerson et al. (2012). Due to how the BUC estimator is implemented, it is not possible to obtain the corresponding marginal effects of the estimated coefficients. It is however possible to interpret the sign and significance of the parameter estimates. 3.2 Recursive Bivariate Ordered Probit Model The analysis presented in the previous chapter suggests that the head of household s subjective financial position is a significant determinant of overall life satisfaction. However, this relationship could potentially be biased if there are unobserved characteristics which influence both financial satisfaction and overall life satisfaction. These unobserved 10

characteristics within the overall life satisfaction equation could be correlated with independent variables which capture the head of household s financial satisfaction. This could lead to the estimates capturing both the effect of the head of household s subjective financial satisfaction and, in addition, the impact of the unobserved characteristics on overall life satisfaction. Generally, when an endogenous variable is encountered, an instrumental variable approach is often implemented in order to ascertain the causal unbiased relationship. Consequently, instrumental variable techniques have become a common approach in the applied micro-econometrics literature. However, problems arise when both the dependent and potentially endogenous variable are discrete and ordered in nature. In this situation, standard instrumental variable (IV) techniques, such as two stage least squares, frequently fail and consequently, a different econometric technique is required. Greene and Hensher (2010) assert that an IV approach is not applicable in a non-linear model, such as the ordered probit model, as a traditional IV approach is based upon the moments of the data. As the ordered probit specification uses maximum likelihood estimation approach, opposed to OLS, it is not obvious how an IV method would apply. Following, Greene and Hensher (2010), when both the dependent variable and endogenous variable are ordered, the preferable estimation technique is a bivariate ordered probit model. Following Ferrer-i Carbonell and Frijters (2004), it is also important to control for unobserved heterogeneity when exploring the determinants of subjective satisfaction measures. Consequently, in line with the univariate ordered probit analysis discussed above and as suggested in Greene and Hensher (2010), Mundlak fixed effects are implemented, that is the inclusion of the group means of the time varying variables. For the two dependent variables, y i1 and y i2 which indicate subjective financial position and overall life satisfaction, respectively, the bivariate ordered probit specification is defined as: yi1 = β 1 x i1 + γ 1 x i1 + ɛ i1, y i1 = j if µ j 1 < yi1 < µ j, j = 0,..., J. (4) y i2 = δ 1 y i1 + β 2 x i2 + γ 2 x i2 + ɛ i2, y i2 = k if µ k 1 < y i2 < µ k, k = 0,..., K. (5) 11

where β 1 and β 2 are vectors of parameters to be estimated, δ 1 is an unknown scalar, x i1 and x i2 are vectors of observable characteristics whilst x i1 and x i2 are the group means and provide the Mundlak correction. µ j and µ k represent the threshold parameters which are to be estimated, whilst the error terms ɛ i1 and ɛ i2 are identically distributed, with a bivariate normal distribution, with a mean of zero and unit variance and correlation coefficient. That is: ɛ i,1 N 0, 1 ρ (6) 0 ρ 1 ɛ i,2 where the covariance term is defined to be Corr(ɛ i1, ɛ i2 ) = ρ 1,2. All standard errors are clustered at the individual level to allow for repeated observations over time. In the instance that ρ is equal to zero; the bivariate model becomes a pair of univariate models. If ρ is found to be statistically different from zero, then this implies a correlation between the unobservable characteristics of the two equations, and so a joint modelling estimation is preferred as it accounts for the endogeneity of subjective financial in the overall life satisfaction equation. In a bivariate specification, failure to reject the null hypothesis (ρ = 0) suggests that endogeneity is not a problem and therefore the coefficients estimated in a univariate specification do not suffer from bias. Should there be sufficient evidence to reject the null hypothesis; this suggests that subjective financial situation is not exogenous and consequently the results are biased. In the instance where is positive, it follows that unobserved characteristics increase both financial satisfaction and overall life satisfaction. If ρ is negative, then the opposite applies. 12

4 Results 4.1 Univariate Analysis Table 2 presents the coefficients of the determinants of overall life satisfaction. As previously stated, it is not possible to comment on the magnitude so we focus on the sign and significance of the parameter estimates. The results are generally in line with those obtained in the existing literature. Compared to being married, being divorced or separated is inversely related to overall life satisfaction. Similarly, in line with Winkelmann and Winkelmann (2003), Headey and Wooden (2004) and Ferrer-i Carbonell and Frijters (2004), unemployment has a detrimental impact on the level of overall life satisfaction. Consistent with the existing literature, self-assessed health status is positively related with overall life satisfaction, that is better health is associated with higher levels of overall life satisfaction. In line with Dolan et al. (2008), the natural logarithm of household income exerts a positive and statistically significant impact on overall life satisfaction, suggesting diminishing marginal utility of income. Once individual heterogeneity is accounted for, education fails to display a statistically significant impact with overall life satisfaction. Focussing on the monetary variables reveals that the level of total assets held by the household is associated with higher levels of overall life satisfaction and, in line with Brown et al. (2005), the separation of total debt reveals that it is unsecured opposed to secured debt which is inversely related to overall life satisfaction. Similarly, in line with prior expectations and Headey and Wooden (2004) higher levels household net wealth are associated with reporting higher levels of overall life satisfaction. The variables which capture the head of household subjective financial position have the expected impact on the level of overall life satisfaction. That is, compared to reporting not being worried, both being concerned and very concerned are detrimentally related to the level of overall life satisfaction. This supports Wildman (2003) and Bridges and Disney (2010) who both report the subjective financial measures are significant de- 13

terminants of overall life satisfaction. 4.2 Bivariate Analysis The results presented in Table 3 show the recursive bivariate probit specifications. The results advocate the uses of a joint modelling technique as across all four specifications, there is a rejection of the null hypothesis that the correlation between the unobservable characteristics is equal to zero. This suggests that the results presented in the univariate specifications are biased due to endogeneity. Furthermore, a positive correlation between the unobservable characteristics of both overall life satisfaction and subjective financial position, that is there are some unobservable characteristics which cause heads of households to report higher levels of overall life satisfaction and being more worried with their current economic situation. Focussing on the financial worries equation reveals that females report being more concerns relating to their economic situation compared to their male counter parts. Being divorced or separated is associated with higher levels of financial worry compared to household heads who are married or in a relationship. In line with Headey and Wooden (2004), Plagnol (2011) and Hansen et al. (2008), better health status is inversely related to financial worries. As expected the variables closely related to an individuals financial position are statistically significant determinants of the head of household s level of financial concerns. For example, household income serves to reduce the level of financial concerns experienced by the head of the household, whilst being unemployed is associated with higher levels of financial worries. In addition, the risk attitudes of the head of household is found to be a statistically significant determinant subjective financial position, with more risk tolerant household heads reporting level of financial worry. Following, Joo and Grable (2004) this could be due to the risk attitudes of the head of household capturing the the level of financial knowledge of the head of the households. 14

The monetary financial measures indicate that all types of debt considered, total, unsecured and secured, are positively related to the level of financial worry reported. Interestingly the level of assets held by the household does not influence the concerns the individual possesses about there economic situation. This lack of relationship is attributed to the wording of the question, which may cause individuals to focus on negative aspects of their financial position, rather than positive aspects such as their levels of savings and assets. Focussing on the determinants of overall life satisfaction in the joint modelling approach, indicate some differences compared to the univariate analysis presented in section 4.1. In contrast to the univariate analysis, it is found that the level of household income is not a statistically significantly determinant of overall life satisfaction. Similarly, unemployment is not a statistically significant determinant of overall life satisfaction once a bivariate specification is implemented. Self- assessed health status maintains a positive relationship with overall life satisfaction, with better health being associated with higher levels of overall life satisfaction. The results show that the subjective financial position is statistically significant determinant of overall life satisfaction with, higher levels of financial concerns being associated with lower levels of overall life satisfaction. Once a joint modelling approach is implemented, it shows that, at the 5% level, the monetary financial measures fail to have a statistically significant impact on the level of overall life satisfaction. The results presented in this study support the findings of Bridges and Disney (2010), who find that the subjective debt burden mediates the effects of debt levels to the likelihood or reporting depression. Within this study appears that the effects of the household monetary financial position on overall life satisfaction is mediated through the subjective financial position. In addition, variables closely related to the households financial position, such as income and employment status, also have a limited direct effect on overall life satisfaction. They are found however to have a large indirect impact through the subjective financial measures. 15

5 Conclusion Overall life satisfaction has received an increased a large amount of interesting from a variety of academic disciplines, including psychology and economics, and also from a wider public audience. In addition, levels of well-being are being increasingly proposed to inform and evaluate public policy and an economies progress. Within the existing literature, a large amount of attention has been placed on the influence of income on well-being and happiness, however, this is arguably not the best measure of an individuals financial resources. Consequently, this study controls for a variety of monetary financial measures. This study aims to explore the relationship between the household s financial position and overall life satisfaction in Germany. Analysing the 2002 and 2007 waves of the GSOEP, this study controls for a variety of monetary and subjective financial measures. Initially, the study provides a longitudinal analysis of the determinants of overall life satisfaction. It then goes on to account for the potential endogeneity of the subjective financial position in the overall life satisfaction equation by employing a recursive bivariate ordered probit model. The univariate results indicate that the subjective financial position of the head of household is a statistically significant determinant of overall life satisfaction, with more concerns relating to current economic status being inversely related with the level of overall life satisfaction. The level of assets and unsecured debt are positively and inversely related to overall life satisfaction, respectively. In line with the existing literature, unemployment and divorce are inversely related to overall life satisfaction, whist both self-assessed health status and household income are positively related with overall life satisfaction. The results from the bivariate ordered probit model advocates the joint modelling approach, suggesting the results presented in the univariate model are biased. The results indicate that the level of debt held by the household increases the level of financial concerns. Also, unemployment and the level of household income are found to increase and 16

reduce the level of financial worry respectively. Within a joint modelling approach, it appears that the subjective financial position mediates the effects of the variables closely related to the households financial position. That is, the level of debt, income and the employment status of the head of household, influence the subjective financial position, but not directly the level of overall life satisfaction. This result supports the findings of Bridges and Disney (2010) who report that the subjective debt burden mediates the effects between the level of debt and the likelihood of reporting depression. 17

References Baetschmann, G., K. E. Staub, and R. Winkelmann (2011). Consistent estimation of the fixed effects ordered logit model. IZA Discussion Paper No. 5443. Bertaut, C. and M. Haliassos (2001). Debt revolvers for self control. HERMES Center Working Paper, 01 11. Bridges, S. and R. Disney (2010). Debt and depression. Journal of health economics 29 (3), 388 403. Brown, S., G. Garino, and K. Taylor (2013). Household debt and attitudes toward risk. Review of Income and Wealth 59 (2), 283 304. Brown, S. and K. Taylor (2008). Household debt and financial assets: evidence from germany, great britain and the usa. Journal of the Royal Statistical Society: Series A (Statistics in Society) 171 (3), 615 643. Brown, S., K. Taylor, and S. Wheatley Price (2005). Debt and distress: Evaluating the psychological cost of credit. Journal of Economic Psychology 26 (5), 642 663. Clark, A. E., P. Frijters, and M. A. Shields (2008). Relative income, happiness, and utility: An explanation for the easterlin paradox and other puzzles. Journal of Economic Literature, 95 144. Delken, E. (2008). Happiness in shrinking cities in germany. Journal of Happiness Studies 9 (2), 213 218. Dickerson, A., A. Risa Hole, and L. Munford (2012). The relationship between well-being and commuting re-visited: Does the choice of methodology matter? Sheffield Economic Research Paper Series. Dohmen, T., A. Falk, D. Huffman, U. Sunde, J. Schupp, and G. Wagner (2005). Individual risk attitudes: New evidence from a large, representative, experimentally-validated survey. IZA Discussion Paper No. 1730. 18

Dolan, P., T. Peasgood, and M. White (2008). Do we really know what makes us happy? a review of the economic literature on the factors associated with subjective well-being. Journal of Economic Psychology 29 (1), 94 122. Drentea, P. (2000). Age, debt and anxiety. Journal of Health and Social Behavior, 437 450. Ferrer-i Carbonell, A. (2005). Income and well-being: an empirical analysis of the comparison income effect. Journal of Public Economics 89 (5), 997 1019. Ferrer-i Carbonell, A. and P. Frijters (2004). How important is methodology for the estimates of the determinants of happiness?*. The Economic Journal 114 (497), 641 659. Ferrer-i Carbonell, A. and X. Ramos (2010). Inequality aversion and risk attitudes. Greene, W. H. and D. A. Hensher (2010). Modeling ordered choices: A primer. Cambridge University Press. Greene, W. H. and C. Zhang (2003). Econometric analysis (Vol. 5). Prentice hall Upper Saddle River, NJ. Hansen, T., B. Slagsvold, and T. Moum (2008). Financial satisfaction in old age: a satisfaction paradox or a result of accumulated wealth? Social indicators research 89 (2), 323 347. Headey, B., R. Muffels, and M. Wooden (2008). Money does not buy happiness: Or does it? a reassessment based on the combined effects of wealth, income and consumption. Social Indicators Research 87 (1), 65 82. Headey, B. and M. Wooden (2004). The effects of wealth and income on subjective well-being and ill-being*. Economic Record 80 (s1), S24 S33. Hofmann, B. and K. Hohmeyer (2013). Perceived economic uncertainty and fertility: Evidence from a labor market reform. Journal of Marriage and Family 75 (2), 503 521. 19

Joo, S.-h. and J. E. Grable (2004). An exploratory framework of the determinants of financial satisfaction. Journal of Family and Economic Issues 25 (1), 25 50. MacKerron, G. (2012). Happiness economics from 35 000 feet. Journal of Economic Surveys. Plagnol, A. C. (2011). Financial satisfaction over the life course: The influence of assets and liabilities. Journal of Economic Psychology 32 (1), 45 64. Reading, R. and S. Reynolds (2001). Debt, social disadvantage and maternal depression. Social Science & Medicine 53 (4), 441 453. Stutzer, A. and B. S. Frey (2010). Recent advances in the economics of individual subjective well-being. Social Research: An International Quarterly 77 (2), 679 714. Wildman, J. (2003). Income related inequalities in mental health in great britain: analysing the causes of health inequality over time. Journal of health economics 22 (2), 295. Winkelmann, L. and R. Winkelmann (2003). Why are the unemployed so unhappy? evidence from panel data. Economica 65 (257), 1 15. 20

6 Appendix Table 1: Summary statistics Variable Mean Std. Dev. Min. Max. Overall Life Satisfaction 6.901 1.748 0 10 Financial Concerns 0.898 0.709 0 2 Female 0.371 0.483 0 1 Age 51.992 15.227 18 97 Age Squared/100 29.35 16.522 3.24 94.09 Ln(Household Size) 0.775 0.516 0 2.565 Ln(Household Income) 10.315 0.657 4.681 13.833 Never Married 0.155 0.362 0 1 Widow 0.091 0.287 0 1 Divorced 0.137 0.344 0 1 Not in Labour Force 0.139 0.346 0 1 Retired 0.203 0.402 0 1 Unemployed 0.051 0.219 0 1 Education 1 0.021 0.142 0 1 Education 2 0.018 0.131 0 1 Education 3 0.152 0.359 0 1 Education 4 0.169 0.375 0 1 Poor Health 0.144 0.351 0 1 Satisfactory Health 0.354 0.478 0 1 Good Health 0.396 0.489 0 1 Very Good HEalth 0.07 0.255 0 1 Ln(Net Wealth) 7.286 6.409-15.35 16.799 Ln(Total Assets) 8.153 5.373 0 16.927 Ln(Total Debt) 3.551 5.153 0 16.032 Ln(Unsecured Debt) 1.783 3.698 0 15.501 Ln(Secured Debt) 3.01 4.985 0 16.032 Concerned 0.488 0.5 0 1 Very Concerned 0.205 0.404 0 1 N 15424 21

Table 2: Fixed Effects ordered Logit Model: Determinants of Overall Life Satisfaction VARIABLES 1 2 3 4 Age 0.0358 0.0201 0.0259 0.0344 (0.0269) (0.0273) (0.0272) (0.0272) Age Squared -0.0750*** -0.0631** -0.0675*** -0.0719*** (0.0251) (0.0254) (0.0253) (0.0253) Household Size 0.0401 0.0254 0.0340 0.0118 (0.130) (0.130) (0.130) (0.131) Household Income 0.352*** 0.341*** 0.339*** 0.342*** (0.0921) (0.0920) (0.0925) (0.0928) Never Married -0.261-0.254-0.267-0.260 (0.191) (0.193) (0.192) (0.192) Widow 0.0740 0.0547 0.0489 0.0353 (0.280) (0.282) (0.283) (0.284) Divorced -0.391** -0.375** -0.386** -0.388** (0.167) (0.169) (0.168) (0.170) Not in Labour Force 0.166 0.172* 0.165 0.166 (0.103) (0.104) (0.104) (0.103) Retired 0.218 0.220 0.211 0.209 (0.145) (0.146) (0.146) (0.145) Unemployed -0.372*** -0.373*** -0.374*** -0.376*** (0.134) (0.134) (0.135) (0.134) Education 1-0.361-0.362-0.358-0.331 (0.566) (0.559) (0.569) (0.554) Education 2-0.183-0.192-0.183-0.211 (0.676) (0.672) (0.672) (0.687) Education 3 0.350 0.338 0.332 0.309 (0.510) (0.511) (0.511) (0.525) Education 4-0.0424-0.0681-0.0875-0.0865 (0.688) (0.692) (0.695) (0.713) Poor Health 1.114*** 1.119*** 1.113*** 1.105*** (0.168) (0.167) (0.168) (0.167) Satisfactory Health 1.809*** 1.814*** 1.810*** 1.801*** (0.177) (0.176) (0.176) (0.176) Good Health 2.529*** 2.530*** 2.528*** 2.521*** (0.185) (0.184) (0.184) (0.184) Very Good Health 2.935*** 2.939*** 2.935*** 2.933*** (0.216) (0.216) (0.216) (0.216) Net Wealth 0.0172*** (0.00578) Total Assets 0.0193** 0.0156* (0.00802) (0.00805) Total Debt -0.0103 (0.00754) Unsecured Debt -0.0232*** (0.00856) Secured Debt 0.00518 (0.00849) Concerned -0.549*** -0.547*** -0.549*** -0.549*** (0.0739) (0.0741) (0.0741) (0.0741) Very Concerned -1.384*** -1.380*** -1.378*** -1.379*** (0.104) (0.104) (0.104) (0.104) Observations 18,402 18,402 18,402 18,402 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 22

Table 3: Recursive Bivariate Ordered Probit Model: Overall Life Satisfaction and Subjective Financial Position Specification Independent Variables 1 2 3 4 Financial Worries Overall Life Satisfaction Financial Worries Overall Life Satisfaction Financial Worries Overall Life Satisfaction Financial Worries Overall Life Satisfaction Female 0.0851*** 0.0854*** 0.0774*** 0.0860*** 0.0776*** 0.0851*** 0.0764*** 0.0855*** -0.0254-0.0233-0.0254-0.0226-0.0254-0.0226-0.0254-0.0226 Age Age Squared Ln(Household Size) Ln(Household Income) Never Married Widowed Divorced Not in Labour Force Retired Unemployed Education 1 Education 2 Education 3 Education 4 Poor Health Satisfactory Health Good Health Very Good Health Risk Tolerant Concerned Very Concerned Ln(Total Assets) Ln(Total Debt) Ln(Unsecured Debt) Ln(Secured Debt) 0.0402*** 0.0264** 0.0459*** 0.0259** 0.0351*** 0.0246** 0.0457*** 0.0238** -0.012-0.0115-0.0123-0.0115-0.0123-0.0113-0.0124-0.0116-0.0118-0.0317*** -0.0145-0.0301*** -0.00783-0.0291*** -0.0157-0.0289*** -0.0115-0.0104-0.0117-0.0105-0.0117-0.0105-0.0117-0.0105 0.104* 0.0298 0.100* 0.0293 0.106* 0.0243 0.110* 0.0307-0.0557-0.0527-0.0565-0.0521-0.0567-0.0522-0.0565-0.052-0.215*** 0.0645-0.222*** 0.0506-0.226*** 0.0515-0.216*** 0.053-0.04-0.0457-0.0406-0.0419-0.0407-0.0418-0.0404-0.0415 0.0397-0.115 0.0515-0.107 0.0552-0.101 0.0364-0.109-0.0758-0.075-0.0769-0.0746-0.0772-0.0746-0.0769-0.0744-0.0192 0.039-0.0192 0.0339-0.0158 0.0313-0.0163 0.0353-0.128-0.13-0.131-0.13-0.131-0.13-0.131-0.13 0.160** -0.135* 0.170** -0.121* 0.175** -0.118 0.153** -0.123* -0.0748-0.0753-0.0758-0.0733-0.0761-0.0734-0.0758-0.0729 0.0252 0.0593 0.0273 0.0603 0.0305 0.0609 0.0238 0.061-0.043-0.0396-0.0435-0.0396-0.0436-0.0396-0.0435-0.0396 0.0625 0.0903 0.0703 0.0917 0.076 0.0944* 0.0628 0.0928* -0.0619-0.0559-0.0628-0.0559-0.0631-0.056-0.0627-0.0559 0.557*** 0.0181 0.569*** 0.0355 0.575*** 0.0346 0.565*** 0.0366-0.0602-0.0784-0.061-0.0676-0.0611-0.0671-0.0609-0.0669-0.219-0.154-0.214-0.157-0.228-0.159-0.214-0.161-0.289-0.266-0.289-0.266-0.292-0.264-0.289-0.266-0.409* -0.00144-0.402-0.0098-0.416* -0.0213-0.411* -0.0181-0.246-0.289-0.247-0.287-0.248-0.287-0.248-0.287-0.399** 0.104-0.395** 0.0833-0.397** 0.0791-0.398** 0.0823-0.201-0.24-0.2-0.237-0.202-0.236-0.201-0.236-0.373 0.043-0.368 0.0263-0.366 0.0267-0.375 0.0282-0.232-0.275-0.231-0.273-0.232-0.272-0.233-0.273-0.0823 0.511*** -0.0778 0.501*** -0.0783 0.501*** -0.083 0.501*** -0.0778-0.079-0.0788-0.0758-0.079-0.0755-0.0789-0.0757-0.222*** 0.745*** -0.220*** 0.728*** -0.220*** 0.728*** -0.225*** 0.726*** -0.0808-0.0968-0.0818-0.0879-0.082-0.0869-0.0819-0.0876-0.372*** 0.978*** -0.372*** 0.952*** -0.372*** 0.952*** -0.375*** 0.949*** -0.084-0.12-0.085-0.104-0.0852-0.102-0.0852-0.103-0.485*** 1.168*** -0.484*** 1.134*** -0.488*** 1.135*** -0.486*** 1.131*** -0.0973-0.147-0.0984-0.126-0.0986-0.123-0.0986-0.125-0.0720*** -0.0811*** -0.0900*** -0.0801*** -0.023-0.0219-0.0218-0.0218-0.870*** -0.921*** -0.923*** -0.927*** -0.175-0.128-0.125-0.126-1.833*** -1.935*** -1.937*** -1.947*** -0.342-0.249-0.241-0.244-0.00562 0.00547* -0.00593* 0.00429-0.00343-0.00328-0.00347-0.0033 0.0129*** 0.00139-0.00304-0.00297 0.0138*** -0.00291-0.00343-0.00336 0.0132*** 0.00605* -0.00345-0.00323 Ln(Net Wealth) -0.00542** 0.00440* -0.00241-0.00233 Cut 1,1 Cut 1,2 Cut 2,1 Cut 2,2 Cut 2,3 Cut 2,4 Cut 2,5 Cut 2,6 Cut 2,7 Cut 2,8 Cut 2,9 Cut 2,10-8.096*** -6.863*** -6.844*** -6.627*** -0.344-0.36-0.359-0.35-6.520*** -5.263*** -5.238*** -5.026*** -0.341-0.358-0.357-0.348-4.181*** -4.222*** -4.210*** -4.159*** -0.907-0.593-0.577-0.569-3.829*** -3.876*** -3.863*** -3.813*** -0.924-0.604-0.588-0.58-3.379*** -3.432*** -3.420*** -3.371*** -0.947-0.621-0.604-0.597-2.928*** -2.987*** -2.974*** -2.926*** -0.97-0.638-0.62-0.614-2.565*** -2.629*** -2.616*** -2.569*** -0.989-0.652-0.633-0.628-1.913* -1.986*** -1.972*** -1.927*** -1.023-0.677-0.658-0.653-1.458-1.537** -1.523** -1.479** -1.047-0.695-0.675-0.67-0.77-0.859-0.844-0.801-1.083-0.721-0.7-0.697 0.304 0.2 0.215 0.255-1.139-0.763-0.741-0.738 1.09 0.975 0.99 1.028-1.18-0.795-0.771-0.77 0.494*** 0.542*** 0.542*** 0.548*** ρ -0.169-0.128-0.124-0.126 Wald Chi Squared 8.58 p-value = 0.0034 17.92 p-value = 0.0000 19.16 p-value = 0.0000 18.82 p-value = 0.0000 Observations 15,424 15,424 15,424 15,424 Robust standard errors in parentheses *** p 0.01, ** p 0.05, * p 0.1 23

Figure 1: Overall life satisfaction Figure 2: Concerns with economic situation 24

assets.png Figure 3: Natural Logarithm of Household Total Assets Debt.png Figure 4: Natural Logarithm of Household Total Debt 25

debt.png Figure 5: Natural Logarithm of Household Unsecured Debt debt.png Figure 6: Natural Logarithm of Household Secured Debt 26

wealth.png Figure 7: Natural Logarithm of Household Net Wealth 27