DEBT AND DEPRESSION: CAUSAL LINKS AND SOCIAL NORM EFFECTS*

Transcription

1 The Economic Journal, 122 (September), Doi: /j x. Ó 2012TheAuthor(s). TheEconomicJournalÓ2012 Royal Economic Society. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. DEBT AND DEPRESSION: CAUSAL LINKS AND SOCIAL NORM EFFECTS* John Gathergood Individuals exhibiting problems repaying their debt obligations also exhibit much worse psychological health. Selection into problem debt on the basis of poor psychological health accounts for much of this difference. The causality between problem debt and psychological health may be two way. Using individual level UK panel data, local house price movements exogenous to individual households are used to establish the causality from problem mortgage debt to psychological health. In addition, the social norm effects of problem debt are investigated using local bankruptcy and repossession rates. Results indicate there are sizeable causal links and social norm effects in the debt psychological health relationship. This article examines the relationship between household problem debt, otherwise known as over-indebtedness, and psychological health, using the UKÕs household panel survey. Access to credit improves household welfare by facilitating consumption smoothing. However, inability to repay debts can result in drastic welfare losses arising from bankruptcy or the seizure of collateral such as housing. Psychiatrists commonly report problem debt as a source of severe anxiety and psychological distress (Fitch et al., 2007). In the medical literature small-scale studies based on individuals exhibiting poor mental health find problem debt to be a common correlate with depression, anxiety and even self-harm (Hatcher, 1994; Maciejewski et al., 2000; Reading and Reynolds, 2001). There is a well-documented statistical association between problem debt and poor psychological health at the individual level. Studies based on large samples of crosssectional survey data using self-reported health data show that the positive association between high levels of debt or usage of high-cost credit and poor psychological health is not readily explained by covariates such as demographic and related characteristics or other existing health conditions (Bartel and Taubman, 1986; Lea et al., 1995; Hamilton et al., 1997; Drentea, 2000; Brown et al., 2005; Lenton and Mosley, 2008). There is also evidence that high-debt households exhibit more prevalent adverse health behaviours which may be related to the formation of psychological health such as smoking and obesity (Grafova, 2007). What is more difficult to establish is causality between problem debt and psychological health. The positive relationship between the two might be explained by * Corresponding author: John Gathergood, School of Economics, Sir Clive Granger Building, University of Nottingham, Nottingham NG7 2RD, UK. john.gathergood@nottingham.ac.uk I thank the Economic and Social Research Council for providing funding for my post-doctoral research fellowship, under grant number PTA The data and tabulations used in this publication were made available through the ESRC Data Archive. The data were originally collected by the ESRC Research Centre on Micro-social Change at the University of Essex (now incorporated within the Institute for Social and Economic Research). Neither the original collectors of the data nor the Archive bear any responsibility for the analyses or interpretations presented here. [ 1094 ]

2 [ SEPTEMBER2012] CAUSAL LINKS BETWEEN DEBT AND unobserved factors not captured in cross-section analysis, or alternatively a two-way causality. 1 Also, an individualõs perception of the severity of their debt problems may be affected by their psychological health state. An individual with poor psychological health might be more, or less, inclined to subjectively report they are struggling with debts compared to an individual with good psychological health in the same financial situation. Bridges and Disney (2010) find evidence from UK household survey data that objective measures of debt problems correlate more weakly with subjective evaluations of poor psychological health than subjective measures of debt problems. This study investigates the relationship between problem debt and psychological health for the UK by using the British Household Panel Survey (BHPS) (previously used by Wildman, 2003 and Brown et al., 2005 in related studies). 2 This study makes the following contributions. First, it documents the large cross-sectional inequality in psychological health between those with and without problem debts. This is shown for both subjective and objective measures of psychological health. Households who report they face ÔdifficultyÕ meeting their housing payments (mortgage or rent), or are at least two months late on their housing payments, or who report that meeting their consumer credit repayments presents a Ôheavy burdenõ to their household, exhibit poorer psychological health. To establish this result is not driven by perceptions alone locallevel delinquency rates are used as an instrument for self-reported debt problems. We also present evidence of poorer psychological health among the spouses or partners of respondents reporting debt repayment difficulties, suggesting the reporting of debt difficulties is not driven by the psychological health state of the respondent. Second, this study shows that selection into problem debt on the basis of poor psychological health accounts for much of the observed cross-sectional variation in psychological health between those with and without problem debts. Individuals who move into arrears on their housing payments or into reporting a heavy burden of debts between two waves of data exhibited, on average, worse psychological health in the first wave of data compared with those not moving into debt problems. This positive selection into problem debt on the basis of poor psychological health accounts for most of the observed difference in health in the cross-section comparison between the two groups. Third, this study estimates the causal impact of worsening problem debt on psychological health by using movements in local-level house prices as exogenous variation in the severity of the consequences of inability to meet mortgage debt 1 A similar problem confronts the researcher seeking to understand the impact of income on health, as earned income is most likely endogenous to health status (Fritjers et al., 2005; Gardner and Oswald, 2007). 2 The key advantage of using the BHPS for this study is that it includes data on the geographical location of the household, not available in the Families and Children Survey (FACS) used in Bridges and Disney (2010). These data are crucial for the instrumental variable strategy used to identify the causal impact of problem debt (which uses local-level house price movements) and to establish reference group effects (which are defined at the local level) later in the article. The BHPS also has the advantage of including the General Health Questionnaire (GHQ) as an alternative to self-reported data on anxiety-related medical conditions. Also, the BHPS has the advantage of being more representative of the population as a whole. Whereas FACS surveys only family units with children and in the vast majority of cases interviews the mother (with women twice as likely to report depression compared with men in the UK), the BHPS is a representative sample of all UK households and adopts the typical convention of allowing the household to self-assign the household head and interviews all members of the household. Finally, the BHPS is a long-running household panel and so provides a usable number of observations of individuals with very severe debt problems.

3 1096 THE ECONOMIC JOURNAL [ SEPTEMBER commitments. It does so in the following manner: it is shown that mortgage holders who enter into arrears on their mortgage debts in localities where house prices are growing (and so their home equity ÔbufferÕ is increasing) suffer less deterioration in psychological health compared to individuals who enter into arrears in localities where house prices are falling (and so their home equity ÔbufferÕ is decreasing). This identification strategy also allows a natural comparison group renting households for whom the impact of rent arrears on psychological health is shown to be unaffected by local house price movements. This allows us to rule out the possibility that local house price movements simply act as a proxy for local economic conditions in these regressions. Home equity buffers have been shown to be important forms of consumption insurance for households facing adverse income shocks (Hurst et al. 2004; Benito, 2009). We show that households with mortgage payment problems who suffer house price falls are less likely to be able to refinance or extract housing equity and, furthermore, incur higher future mortgage payment costs. This primarily arises due to households who experience mortgage payment problems typically doing so in the early years of their mortgage contracts when they are most likely to seek to refinance to avoid reset rates arising after short-term mortgage deals elapse. House price falls reduce the option to do so and increase the chances of households facing the reset rate. Fourth, the impact of the onset of problem debt on psychological health is shown to demonstrate a Ôsocial normõ effect. 3 Individuals who exhibit the onset of problems repaying their unsecured debts in localities with a higher bankruptcy rate are shown to experience less deterioration in psychological health. In the context of a uniform bankruptcy law across localities (and so little reason to believe that non-payment on unsecured debts is more likely to result in a bankruptcy filing in one region compared with another) this result is interpreted as evidence of a reduced social stigma associated with problem debt in areas in which problem debt is more prevalent. Individuals who exhibit the onset of problems repaying their secured debts in localities with higher mortgage repossession rates also show less deterioration in psychological health. 1. Data 1.1. The BHPS The BHPS is a well-known long-running UK household panel survey which started in 1991 and has been conducted annually; the most recent available data is for the year All 18 available waves of data are used for this study. The BHPS is a general household survey which began with approximately 5,500 households with 10,000 individuals from England and Wales in 1991, interviewing adults in the household on a range of socioeconomic topics including their finances, labour market activity and health. As the health and debt data are central to this study, these are now considered in more detail. 3 Social norm effects have been widely studied in the labour supply and consumption literatures (Blinder and Pesaran, 1998; Lindbeck et al., 1999).

4 2012] CAUSAL LINKS BETWEEN DEBT AND DEPRESSION Psychological Health Data The BHPS includes two survey instruments related to psychological health. First, in the health module of the survey all adult respondents in the household are asked to identify the health problems or disabilities which they currently suffer from among those on a list, the most relevant of which for this analysis is ÔAnxiety, depression, bad nerves, psychiatric problemsõ. Respondents are asked to ignore temporary conditions when answering this question. We use answers to this question to construct an indicator variable which takes a dummy form with a value of 1 for yes and 0 for no. Second, the BHPS also includes the General Health Questionnaire (GHQ) in each wave. The GHQ comprises a series of 12 questions in which respondents are asked to identify how frequently they currently feel, relative to their normal state, depression, anxiety leading to insomnia, inability to cope and a number of related feelings (details of the particular questions asked are provided in Appendix A). Responses to the GHQ forms the basis for the ÔGHQ Caseness ScoreÕ, also known as the ÔCaseness GHQÕ, a wellknown scale measure of psychological health used in the medical and psychological literature and increasingly in the economics literature as a measure of ÔmentalÕ or ÔpsychologicalÕ health or ÔwellbeingÕ (such as Clark, 2003). 4 The GHQ Caseness score is ordered between 0 and 12, with 12 representing the poorest state of mental health Data on Household Finances and Problem Debts The BHPS asks respondents detailed questions on their household finances every five years (in waves 5, 10 and 15) so it is not possible to construct balance sheet information for each wave of the survey. 5 In addition, the head of household is asked the following questions about their householdõs financial situation. In all waves all households which either own a home via a mortgage or who rent a home are asked: ÔMany people these days are finding it difficult to keep up with their housing payments. In the last twelve months would you say you have had any difficulties paying for your accommodation?õ (Yes No) followed by the question ÔIn the last twelve months have you ever found yourself more than two months behind with your rent mortgageõ? (Yes No). In addition, in each wave since 1995 all households with outstanding unsecured credit on which they are making repayments are asked: ÔTo what extent is the repayment of such debts and the interest a financial burden on your household? Would you say it is... A heavy burden, Somewhat of a burden, Not a problemõ? 4 The GHQ Caseness Score is calculated by counting the number of cases in which an individual reports experiencing six negative feelings Ôrather more than usualõ or Ômuch more than usualõ, or experiencing six positive feelings Ôless so than usualõ or Ômuch less so than usualõ. Hence a score of 12 indicates the individual reported they feel each of the six negative feelings at least Ôrather more than usualõ plus each of the six positive feelings less or much less than usual, and a score of 0 indicates the individual feels each of the six negative feelings not more than Ôno more than usualõ and each of the positive feelings at least as much as usual. On this basis, a score of 12 represents the lowest level of psychological wellbeing (worst mental health) and a score of 0 represents the highest level of psychological wellbeing (best mental health). Some studies invert this 12-point score, known as the Ôinverted GHQÕ such that a higher value represents a better level of psychological health. 5 However, in each wave respondents are asked some questions relating to their finances: detailed questions about their income, the amount they save from their current income, an estimate of the value of their home and any debts secured against it together with the type of mortgage they currently hold and the cost of their monthly housing payment (mortgage or rent).

5 1098 THE ECONOMIC JOURNAL [ SEPTEMBER From answers to the first question we construct a (1 0) dummy variable for the respondentõs evaluation of their difficulty paying for their housing based on the yes no answer. We consider this answer to be a ÔsubjectiveÕ response as the interpretation of the term ÔdifficultÕ might vary between households. From the second question we construct a (1 0) dummy variable for the respondentõs objective housing arrears position based on their yes no answer. This is designated as an ÔobjectiveÕ measure because whether or not an individual is two months late on payments is not a matter of interpretation. From the third question we construct a (1 0) dummy variable for the respondentõs subjective evaluation of their difficulty meeting their unsecured debt payments which takes a value of 1 is the respondent reports ÔA heavy burdenõ or Ôsomewhat of a burdenõ and a value of 0 if they report Ônot a problemõ Summary Statistics Summary statistics for household demographic characteristics, education, employment and income are provided in Table 1. Individual characteristics and psychological health data are provided for the head of the household. As the housing payment questions are asked in every wave but the consumer credit payment question is asked only for 1995 onwards, summary statistics are shown for two periods separately and Comparing households in the whole sample (Column 1) with those with housing payment problems (Column 2, 9.7% of the sample) and those 2+ months late on housing payments (Column 3, 2.3% of the sample), households with payment problems of either type are typically younger, more likely to have a male household head, more likely to have children, be less educated, be in unemployment and have lower income. From Columns 4 and 5, those with consumer credit payment problems (16.2% of the sample) are typically more likely to have a male household head and have children, but there are less noticeable differences in other variables (particularly in household income). Comparing average measures of psychological health between heads of households with payment difficulties to those without payment difficulties: heads of households with housing payment problems exhibit GHQ scores which are on average 1.63% points higher and are 6% points more likely to report an anxiety-related illness. For the 2+ months late on housing payments indicator the differences are larger at 2.05% points and 12% points respectively and for the Ôheavy burdenõ consumer credit payments indicator the differences are 0.98% and 6% points. These differences in means are in each case statistically significantly different from zero at very high levels of confidence. By way of comparison, these differences are also larger than the average difference in GHQ score between individuals in employment and those unemployed, which are 1.6% points and 4% points respectively. 2. Results 2.1. Panel Evidence on Problem Debt and Psychological Health This subsection explores whether these observed differences in psychological health between those with and without housing and consumer credit payment problems

6 2012] CAUSAL LINKS BETWEEN DEBT AND DEPRESSION 1099 Table 1 Summary Statistics for BHPS Households Whole sample 2. Housing payment problems months late on housing payments 4. Whole sample 5. Consumer credit payments a heavy burden N 66,664 6,499 1,541 54,731 8,864 Percentage of sample Demographics Age Male = Married = Has children = Ethnic minority group = 1 Home owner = Highest educational qualification GCSE = A-level = Degree = Employment status Employed = Self-employed = Unemployed = Income Household income 2,100 1,400 1,200 2,100 2,000 (monthly, s) Psychological health GHQ12 Score (012) Suffers anxiety = Notes. Questions on housing payment problems asked in waves inclusive, question on consumer credit payment problems asked in waves inclusive only. Male, married, has children, ethnic minority group, GCSE, A-level, degree, employed, self-employed, unemployed are 1 0 dummy values taking the value 1 if ÔyesÕ and 0 otherwise. Individual characteristics are identified from the head of household. Monthly income is real household monthly gross labour income in 1995 prices (adjusted using the Retail Prices Index). arise due to covariates, selection or psychological health-driven biases in the perception of problem debt. First, to address the role of associated covariates and selection multivariate panel regression, estimates are presented. Second, to test whether the differences arise due to perceptions, the self-reported problem debt indicators (which may be biased by perception) are instrumented using local-level data on delinquency rates. First, those households with and without problem debts differ in a range of associated characteristics, as illustrated in Table 1. Also, the average differences between groups might arise due to selection into problem debt on the basis of poor psychological health, or alternatively selection out of problem debt on the basis of better psychological health. Table 2 presents a transition matrix of before and after average GHQ Caseness Scores and rates of reporting anxiety for individuals entering the

7 1100 THE ECONOMIC JOURNAL [ SEPTEMBER Table 2 Transition Matrix: Entry into Debt Problems by Psychological Health Measures GHQ12 Score Anxiety-related illness Mean (SD) at t Mean (SD) at t + 1 % (SD) at t % (SD) at t +1 Housing payments No payment problems at t, payment problems at t +1 (N = 2,413) No payment problems at t, no payment problems at t +1 (N = 42,134) Not 2+ months late t, 2+ months late at t +1(N = 648) Not 2+ months late t, Not 2+ months late at t +1(N = 43,899) Consumer credit repayments Not a heavy burden at t, heavy burden at t +1(N = 3,561) Not a heavy burden at t, not a heavy burden at t +1 (N = 31,949) 2.97 (3.59) 3.40 (3.87) 0.14 (0.35) 0.16 (0.37) 1.78 (2.90) 1.80 (2.95) 0.07 (0.26) 0.07 (0.26) 3.48 (3.88) 4.06 (4.16) 0.19 (0.39) 0.24 (0.43) 1.93 (3.03) 1.94 (3.07) 0.08 (0.27) 0.08 (0.27) 2.42 (3.37) 2.64 (3.53) 0.12 (0.32) 0.12 (0.33) 1.78 (2.94) 1.78 (2.96) 0.08 (0.27) 0.08 (0.27) housing and consumer credit problem debt states compared with those not entering problem debt states. 6 A comparison suggests that, in each case most of the observed difference in psychological health by problem debt status in the cross-section arises due to positive selection into problem debt on the basis of poor psychological health. Those households exhibiting the onset of problem debts already had worse psychological health in the first wave. Panel regression estimates are presented in Table 3. The vector of control variables includes a broad range of demographic, education and employment, financial and related controls, details of which are provided in the notes accompanying the Table, plus regional dummies and year dummies. In each case, the dependent variable is the psychological health measure (the GHQ Caseness Score in Columns 1 and 2, the indicator variable of anxiety-related illness in Columns 3 and 4). Pooled panel estimates are presented alongside household fixed-effects estimates. Each model includes the three indicators of payment problems and is estimated for the 54,731 households present in at least two consecutive waves of the survey between 1995 (the first year in which the consumer credit question was asked) and In the pooled panel estimates in Columns 1 and 3, psychological health improves with employment and self-employment and is worse for men, the unemployed, those divorced and older individuals (nonlinearly). The coefficients on each of the problem debt measures are positive and significant at the 1% level. The coefficients on these variables are in each 6 For example, whereas in the cross-section those reporting difficulty paying for housing exhibited on average GHQ Caseness scores 1.63 points higher than those not reporting problems, in the transition data the deterioration in GHQ score among those developing difficulties paying for housing is 0.43 points. In the case of those 2+ months late with housing payments and those who face a Ôsomewhat or heavy burdenõ of consumer credit the difference in the transition is 0.58 (compared with 2.05 in the cross-section) and 0.24 (compared with 0.99 in the cross-section).

8 2012] CAUSAL LINKS BETWEEN DEBT AND DEPRESSION 1101 Table 3 Multivariate Estimates of Relationship Between Problem Debt and Psychological Outcomes Dependent variable: GHQ-12 Score (O.L.S.) Dependent variable: Anxiety-related illness (LPM) (1) (2) (3) (4) (1) Housing payment 1.15** (0.05) 0.62** (0.05) 0.05** 0.02** ÔproblemsÕ (2) 2+ months late housing 0.52** (0.11) 0.47** (0.11) 0.06** 0.03** payment (3) Consumer credit 0.75** (0.04) 0.33** (0.04) 0.04** 0.01** (0.003) Ôheavy burdenõ (4) Age (years) 0.11** 0.08** (0.02) 0.01** (0.00) 0.01** (0.00) (5) Age squared (years) 0.001** (0.00) 0.001** (0.00) 0.001** (0.00) 0.001** (0.00) (6) Male = ** (0.03) 0.04** (0.00) (7) Married = ** (0.05) 0.02 (0.09) (8) Divorced = ** (0.05) 0.43** (0.10) 0.04** 0.03** (9) Employed = ** (0.05) 0.85** (0.07) 0.16** 0.07** (10) Unemployed = ** (0.07) 0.31** (0.08) 0.11** 0.05** (11) Self-employed = ** (0.06) 0.86** (0.09) 0.17** 0.07** Regional dummies Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Fixed effects No Yes No Yes No. observations 54,731 54,731 54,731 54,731 F Prob>F R-squared No. groups 10,525 10,525 Mean in sample SD in sample Notes. Significant at *5% level, **1% level. Standard errors reported in parentheses. Control variables: age of head of household, age of head of household squared, dummy variables for male household head, employment status (employed, unemployed, self-employed), marital status (married, divorced), spouse educational and employment self-employment status, dummy variables for skill group (professional, skilled, semi-skilled), dummy variables for age of youngest child (03, 35, 512, 1216 years), dummy variables for whether member of occupational pension plan, whether moved home in last year, whether smokes, spouse smokes plus value of total outstanding mortgage debt in pounds; number of dependent children; household annual income, household annual income squared. case considerably smaller than the unconditional differences in means between those with and without problem debts provided in Table 2. In the models including household fixed effects (Columns 2 and 4) many of the covariates remain statistically significant. The coefficients on the problem debt variables are reduced in magnitude but remain statistically significant at the 1% level. The associations between problem debt and GHQ scores in Column 2 compared with the unconditional comparison (values given here in parenthesis) are: subjective difficulty paying for housing 0.62 (1.63), 2+ months late with housing payment 0.47 (2.05), subjective difficulty paying for consumer credit 0.33 (0.92). In Column 4, the equivalent values are 0.02 (0.08), 0.03 (0.12) and 0.01 (0.06) respectively. Therefore, in each case the magnitude of the association between problem debt and poor psychological health falls by at least two-thirds. These fixed-effects estimates establish there is a clear association between the onset of problem debt and the worsening of

9 1102 THE ECONOMIC JOURNAL [ SEPTEMBER psychological health at the household level controlling for time-invariant heterogeneity but that the estimated effects are weaker than in the unconditional comparisons. Of course, these results do not establish the direction of causality between these contemporaneous changes. Secondly, one difficulty with using the subjective measures of problem debt is the possibility that self-reported measures of problem debt or arrears might themselves be biased by an individualõs mental health state. If a respondentõs mental health state impacts upon their perception of their problem debt state, subjective measures of ÔproblemsÕ and ÔburdensÕ reported by respondents could be unreliable. Ideally, we would use lender-provided debt data, or data externally validated by some other means, which is not possible in the BHPS. Instead we present evidence against this bias by using two alternative approaches. First, we instrument subjective responses using lender-provided measures of locallevel mortgage and consumer credit delinquency, exploiting geographic variation in non-payment of debts. These local-level measures will correlate with actual debt problems but not purely perceived problems; although there is the possibility that poor mental health may affect the perceptions of individuals such that they report their own debt state to be something closer to the problematic state they observe in others in their locality. Results show the geographic concentration of housing and consumer credit payment problems are strong instruments for self-reported payment problems. Table 4 presents IV estimates in which these local-level rates are used as instruments alongside OLS estimates. 7 In each case the IV procedure is implemented using twostage least squares. 8 In each case the instruments are precisely defined at the 0.001% level. Results from the second stage regressions return coefficients of very similar magnitude to those in Table 3. Second, we examine the relationship between the payment ÔproblemsÕ and ÔburdensÕ responses given by the household head and the psychological health of the household headõs spouse or partner. If the household headõs perception of a payment difficulty arises due to his or her mental health state and not due to an actual difficulty, we would not expect to find a positive relationship between the head of householdõs answers to the payment difficulty questions and the psychological health of the household headõs spouse or partner. If the payment difficulty is an actual problem and not purely a perception, we would expect to observe the psychological health effects for a partner or spouse who shares in the householdõs financial situation. Of course, in the latter case we would not expect the household headõs psychological health to correlate perfectly with the psychological health of a spouse or partner as observed psychological health arises due to combinations of genetic, historical and environmental factors not all shared by household members. As we have individual level data for households in our sample, we can examine this relationship. 7 Measures of the local-level mortgage and consumer credit delinquency rate at available at the county level from the Council of Mortgage Lenders (CML) and Experian. CML provide data on the proportion of outstanding home loans at least three months in arrears. Experian provide data on the proportion of consumer credit products at least three months delinquent. 8 The dependent variables are as before. Separate models are estimated which include housing payment problems and consumer credit a heavy burden. In the housing payment regressions the sample is comprised of all years for which the housing payment question was included in the BHPS, in the case of the consumer credit payment regressions the sample is the years only as before.

10 2012] CAUSAL LINKS BETWEEN DEBT AND DEPRESSION 1103 Table 4 IV Estimates for Relationship Between Problem Debt and Psychological Health Outcomes GHQ12 Score Anxiety-related illness (1) Housing payment problems (2) Credit burden (3) Housing payment problems (4) Credit burden Dependent variable: GHQ12 Score (012) (1) Housing payment ÔproblemsÕ (2) Consumer credit Ôheavy burdenõ (3) Local mortgage arrears rate (4) Consumer credit delinquency rate OLS IV 2nd stage 0.77** (0.04) 0.76** (0.10) 0.67** (0.04) IV 1st stage 0.85** (0.04) OLS IV 2nd stage 0.02** 0.57** (0.06) IV 1st stage 0.90** (0.04) LPM IV 2nd stage 0.02** (0.003) 0.01** IV 1st stage 0.85** (0.04) LPM IV 2nd stage 0.01** (0.002) IV 1st stage 0.90** (0.04) Regional dummies Yes Yes Yes Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes Yes Yes Yes Fixed effects Yes Yes Yes Yes Yes Yes Yes Yes F No. observations 66,664 66,664 54,731 54,731 66,664 66,664 54,731 54,731 No. groups 11,936 11,936 10,525 10,525 11,936 11,936 10,525 10,525 Mean in sample SD in sample Notes. Significant at *5% level, **1% level. Standard errors reported in parentheses. Additional control variables as in Table 3.

11 1104 THE ECONOMIC JOURNAL [ SEPTEMBER Table 5 Multivariate Estimates of Relationship Between Problem Debt and Spouse Psychological Outcomes Dependent variable: Spouse GHQ-12 Score (O.L.S.) Dependent variable: Spouse anxiety-related illness (LPM) (1) (2) (3) (4) (1) Housing payment ÔproblemsÕ 0.90** (0.09) 0.59** (0.10) 0.06** 0.02** (2) 2+ months late housing payment 0.60** (0.20) 0.71** (0.21) 0.03** 0.02** (3) Consumer credit Ôheavy burdenõ 0.57** (0.06) 0.25** (0.05) 0.04** 0.01** (0.004) Regional dummies Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Fixed effects No Yes No Yes No. observations 43,016 43,016 43,016 43,016 F Prob>F R-squared No. groups 11,263 11,263 Mean in sample SD in sample Notes. Significant at *5% level, **1% level. Standard errors reported in parentheses. Additional control variables (for household head) as in Table 3, plus controls for spouse characteristics: age, age squared, dummy variables for male, employment status (employed, unemployed, self-employed), educational and employment self-employment status, dummy variables for skill group (professional, skilled, semi-skilled), dummy variables for whether member of occupational pension plan. Table 5 presents estimates from models in which the partner spouseõs psychological health data are the dependent variable and the household headõs responses to the payment difficulty questions enter as a dependent variable. For completeness, we include the full set of household head control variables together with control variables for partner spouse characteristics. As can be seen from Table 5, in each case the coefficients on the payment problem and burden variables are positive, statistically significant and have magnitudes similar to those in Table 3. Hence head of household reported payment difficulties are associated with poorer psychological health on the part of his or her spouse or partner. On this basis, we conclude that the self-reported data on payment difficulties is not severely affected by a perception-bias Evidence From House Price Changes The results from the previous subsection document that the onset of problem debt is associated with deterioration in psychological health but the causality between these two might run in either direction. An obvious identification strategy is to exploit exogenous variation in sources of psychological health or problem debt, that is, a variable correlated with psychological health which is exogenous to changes in 9 On this basis, we continue to use the self-reported measures of problem debt in the subsequent analysis.

12 2012] CAUSAL LINKS BETWEEN DEBT AND DEPRESSION 1105 individual indebtedness or, conversely, a variable correlated with problem debt which is exogenous to individual changes in psychological health. This study uses a source of exogenous source of variation in the severity of an individualõs problem debt: housing equity shocks arising from movements in local-level house prices which make the consequences of arrears on mortgage payments more or less severe. The rationale for this is as follows. Unlike mortgage debt, movements in the value of an individualõs property are largely exogenous to the actions of the individual household. However, house price movements do impact upon the severity of late or nonpayment of mortgage debts via their effect on the housing equity a homeowner owns in their home. If faced with difficulty paying a mortgage it is unambiguously better for an individual to face such a scenario with more rather than less housing equity. 10 The null hypothesis under such an exercise is that individuals who exhibit the onset of problems paying for their housing but contemporaneously benefit from a positive housing equity shock will see less deterioration in their psychological health as the effects of their payment problems are mitigated in part by their equity gain. Using local-level house price shocks as an instrument for the severity of problem mortgage debt also has the attraction of presenting a natural comparison group: renters, who experience late payment of their housing payments but do not benefit from increases in the value of the home in which they are resident. Of course, assignment into housing tenure is not exogenous to the individual household (unlike the value of house price changes in the locality). Renters are typically younger, with lower incomes and less likely to have children, so these and related covariates need to be included as additional controls in the econometric model. There is also an added advantage to the homeowners renters comparison: one objection to interpreting house price shocks as a proxy for housing equity movements is that positive house price shocks might also reflect positive local income shocks (which increase housing demand and so cause house values in the locality to increase). Comparing the outcomes for renters with homeowners allows us to exploit renters as a comparison group who experience the effects of local income shocks correlated with house price movements but not the shocks in home equity. To show the relevance of house price movements to homeownerõs mortgage activity and the particular relevance of changes in housing equity for homeowners with mortgage arrears, a series of models are first estimated to quantify the impact of house price movements on household payment problems, mortgage refinancing and equity extraction plus mortgage costs. This is done to substantiate the idea that house price movements are relevant for the psychological health of households, particularly those with payment difficulties. Local-level house price data are obtained from the Halifax Building Society (now Lloyds-Halifax Bank of Scotland) Mix-Adjusted House Price Index (2011), which is available at the county level. 10 Households who experience equity falls will suffer increases in their leverage and have less scope to refinance (and so face higher future payments), withdraw equity or sell their home without incurring a capital loss. More housing equity increases the likelihood of being able to refinance a mortgage onto more favourable terms, and increases the equity buffer if an individual is forced to sell their home. Hurst and Stafford (2004) present evidence that households use housing equity as a source of insurance when faced with income shocks.

13 1106 THE ECONOMIC JOURNAL [ SEPTEMBER Table 6 House Price Changes, Problem Debt and Future Mortgage Activity Whole sample Mortgage holders only (1) 2+ months late on housing payment t +1 (1) Refinances mortgage t +1 (2) Withdraws equity t +1 (3) Change in monthly mortgage payments (%) t +1 (1) 2+ months late 0.09** 0.07** (0.02) 0.02** (0.003) 0.09** (0.03) (2) D house price ( Õ0,000s) ** (0.002) 0.01** (0.006) 0.01 (3) 2+ months late D house price ( Õ0,000s) 0.02** (0.004) 0.02** (0.004) 0.01** (0.002) 0.03** (0.003) Regional dummies Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Fixed effects Yes Yes Yes Yes No. observations 66,664 35,949 35,949 35,949 F Prob>F No. groups 11,936 6,248 6,248 6,248 Mean in sample SD in sample Notes. Significant at *5% level, **1% level. Standard errors reported in parentheses. Additional control variables as in Table 4. Table 6 presents estimates from a number of panel data models for household mortgage activity. In the first column, the dependent variable is whether the household is 2+ months late on mortgage payments in the next wave. The coefficient on the house price term is statistically insignificant. Local house price movements have no effect on the likelihood of future payment arrears. However, the coefficient on the interaction term (in the third row) implies that households who are 2+ months late on their housing payments at t and experience a subsequent decline in local-level house prices of 10,000 are, compared to the baseline predicted probability, twice as likely to be 2+ months late on their payment at t + 1. The estimates in the subsequent columns show that those households who are 2+ months late on their mortgage payments at t and experience house price falls between t and t+1 are less likely to refinance (Column 3), less likely to withdraw home equity (Column 4) and, crucially, experience higher future mortgage payments (Column 5). Therefore, households with mortgage arrears who experience price falls are more likely to face future arrears, higher mortgage costs and less scope to refinance, including equity withdrawal. We would, therefore, expect such households to suffer increased psychological stress. 11 Following on from these results, we now present estimates of the impact of house price movements on psychological health for households with and without payment 11 It is perhaps not surprising that we find house price falls for those in arrears lead to negative outcomes. The majority of households who experience mortgage arrears do so in the early years since purchase (for an examination of the dynamics of mortgage arrears in the BHPS see Gathergood (2009)). At this stage in the life-cycle households are likely to be highly leveraged, so house price movements can have substantial effects on refinancing opportunities.

14 2012] CAUSAL LINKS BETWEEN DEBT AND DEPRESSION 1107 Table 7 Exogenous House Price Changes, Problem Debt and GHQ12 Scores (O.L.S. Estimates) Dependent variable: GHQ12 Score (1) Owners only (2) Renters comparison group (3) Consumer credit problems (1) 2+ months late 1.18** 1.24** (0.12) (0.21) (2) D house price ( Õ0,000s) (3) 2+ months late D house 0.42** 0.11 price ( Õ0,000s) (0.11) (0.12) (4) 2+ months late owner 0.21** (0.03) (5) D house price ( Õ0,000s) owner 0.05 (0.03) (6) 2+months late D house price 0.64** ( Õ0,000s) owner (0.18) (7) Heavy burden 0.47** (0.07) (8) Heavy burden D house price ( Õ0,000s) 0.01 (0.04) (9) Heavy burden owner 0.15 (0.09) (10) D house price ( Õ0,000s) owner 0.05 (0.03) (11) Heavy burden D house price ( Õ0,000s) owner 0.21 (0.12) Regional dummies Yes Yes Yes Year dummies Yes Yes Yes Fixed effects Yes Yes Yes No. observations 46,776 66,664 54,731 F Prob>F No. groups 8,713 11,936 10,525 Average in sample SD in sample Notes. Significant at *5% level, **1% level. Standard errors reported in parentheses. Additional control variables as in Table 4. problems. Results are presented in Table 7 (in which the GHQ score is the dependent variable) and Table 8 (in which anxiety as a medical condition is the dependent variable). In each case results are presented firstly for the sample of owners only (Column 1), then in a model including the renters comparison group (Column 2) and finally also for the sample in which consumer credit payments are observed and interacted with the house price movement (Column 3). In Table 7, Column 1 the coefficient on the change in the county-level house price is statistically insignificant. Hence house price changes do not affect the psychological health of homeowners. The interpretation of the coefficient on the interaction term is that an individual who experiences the onset of housing payment arrears but a simultaneous positive increase in local-level house prices of 10,000 experiences a deterioration in their GHQ score of 0.42 points less than an individual who does not experience a positive house price gain. In Column 2, the renters comparison group is introduced into the model. Results indicate the negative impact of falling house prices on GHQ scores is specific to

15 1108 THE ECONOMIC JOURNAL [ SEPTEMBER homeowners only. The interaction term between late payment and the homeowner dummy implies homeowners with late payments suffer worse GHQ scores compared with renters. The interaction term between late payment, homeownership and the house price change indicates homeowners with late payments who suffer house price falls experience an increase (worsening) of their GHQ score. This effect is limited to homeowners only, with no impact for the renter comparison group. The magnitude of the interaction term implies a homeowner in late payment who suffers a 10,000 fall in house price experiences an increase in their GHQ score of 0.64 points. Column 3 presents estimates for the same model as Column 2 but with consumer credit payment problems as the problem debt variable and is estimated over the sample of households. Homeowners with consumer credit payment problems will also see their financial situation worsen if house prices fall as their scope for extracting home equity to repay outstanding consumer credit will diminish. The direction and magnitudes of the homeowner interaction terms are similar to before: homeowners who are late with payments have worse GHQ scores; homeowners late with payments who experience house price falls also have worse GHQ scores. However, the coefficients on these interactions are not statistically significantly different from zero. Table 8 presents results from models with the (1 0) dummy variable for whether the individual suffers from anxiety or a related condition as the dependent variable. Results here reveal the same pattern as in Table 7. The pattern in the coefficients in the model for homeowners only (Column 1) shows negative local house price movements which accompany the onset of housing payment problems result in increased likelihood of suffering an anxiety-related medical condition for homeowners who are late on their housing payments. In Column 2, the relationship to the reference renters group is the same as before with no effect of house price movements on renters. The model for consumer credit in Column 3 returns expected signs on the homeowner and consumer credit burden interaction terms but these are again not statistically significantly different from zero. Taken together, these results show that exogenous variation in the severity of arrears on housing payment arising from local-level house price movements causally impact the extent of deterioration in psychological health (by either of the measures used). This effect is stronger for homeowners with late payments on their housing debts than those who experience a heavy burden of consumer credit. An explanation for this difference is that the financial characteristics of households late on housing payments appear much worse than those households facing a heavy burden of consumer credit such that the housing equity buffer is more important for the latter group than for the former (see Table 1) Social Norm Effects in the DebtDepression Relationship This final subsection in the analysis investigates the existence of social norm effects in the relationship between problem debt and psychological health. To the authorõs knowledge, such effects have not been investigated elsewhere. This is perhaps surprising: a growing empirical literature in economics finds that individual perceptions

16 2012] CAUSAL LINKS BETWEEN DEBT AND DEPRESSION 1109 Table 8 Exogenous House Price Changes, Problem Debt and Suffering Anxiety (L.P.M. Estimates) Dependent variable: Whether suffers anxiety (1 0) (1) Owners only (2) Renters comparison group (3) Consumer credit problems (1) 2+ months late 0.03** 0.02** (0.006) (0.008) (2) D house price ( Õ0,000s) (3) 2+ months late D house 0.004** 0.01 price ( Õ0,000s) (0.02) (4) 2+ months late owner 0.02** (0.007) (5) D house price ( Õ0,000s) owner 0.01 (6) 2+months late D house price 0.005** ( Õ0,000s) owner (7) Heavy burden 0.01** (0.003) (8) Heavy burden D house price ( Õ0,000s) (0.003) (9) Heavy burden owner 0.02 (10) D house price ( Õ0,000s) owner (0.006) (11) Heavy burden D house price ( Õ0,000s) owner 0.01 (0.006) Regional dummies Yes Yes Yes Year dummies Yes Yes Yes Fixed effects Yes Yes Yes No. observations 46,776 66,664 54,731 F Prob>F R-squared No. groups 8,713 11,936 10,525 Average in sample SD in sample Notes. Significant at *5% level, **1% level. Standard errors reported in parentheses. Additional control variables as in Table 4. and choices are influenced by those of others. 12 This raises the prospect that social norm effects might be present in the relationship between problem debt and psychological health. The particular social norm effect hypothesised here is the impact of social stigma arising from problem debt. The literature on the social stigma of individual indebtedness and adverse debt outcomes such as bankruptcy presents evidence that higher reference group bankruptcy rates diminish the social stigma associated with being declared bankrupt (Fay et al., 2002; Cohen-Cole and Duygan-Bopp, 2008). The falling 12 For example, Clark (2003) shows the impact of unemployment on psychological health is less severe for individuals who live in localities in which the unemployment rate is higher, and hence is more of a Ôsocial normõ among the population. This finding is contrary to a standard labour market analysis in which higher local unemployment is indicative of fewer job opportunities and would result in increased psychological stress.

17 1110 THE ECONOMIC JOURNAL [ SEPTEMBER stigma of bankruptcy has been widely cited as a reason why the bankruptcy filing rate increased rapidly in both the UK and US during the early 2000s, despite little change in the number of individuals who might benefit financially from filing. As with unemployment, the negative psychological effect of high debt might arise in large part due to the perceived stigma of problem debts rather than the material losses incurred by overindebtedness and bankruptcy. The existence of reference group effects is investigated in the following manner. Two contexts for problem debts are considered: problem housing debt in the context of the prevailing local housing repossession rate; and problem consumer credit debts in the context of the prevailing local personal insolvency rate. County-level repossessions data are provided by the Council for Mortgage Lenders. For bankruptcy data, we use data on the bankruptcy orders issues by courts in England and Wales provided by the Insolvency Service. So in both cases the Ôreference groupõ level of bankruptcy is defined at a relatively broad ÔlocalÕ definition. Tables 9 and 10 present results for models in which these reference group rates of mortgage repossessions among mortgage holders and individual bankruptcies (cases per 100) are included in the specification in an interaction term to capture the impact of the local bankruptcy repossession rate on the psychological health of those with problem debt. In Table 9 the GHQ score is the dependent variable, in Table 10 anxiety as a medical condition is the dependent variable. Column 1 presents estimates for a model estimated on a sample of all individuals. The reference group effect is captured by interacting the dummy variable for individuals exhibiting 2+ months late on payments with the local repossessions rate. Results firstly reveal the rate of local repossession rate has no impact on wellbeing independent of late payments (row 2). However, the coefficient on the interaction term (row 6) implies that individuals experiencing the onset of mortgage arrears in regions in which mortgage arrears are more prevalent see less deterioration in their psychological health scores compared with individuals who exhibit an onset of mortgage arrears in regions with lower mortgage arrears rates. The coefficient value implies this effect is small. The mean repossession rate across all region-year observations is 0.89% and range from the 25th to the 75th percentile is 0.53%. The coefficient value of implies a 0.5% point increase in the repossession rate is associated with a point reduction in the GHQ Caseness Score. On this basis, a very high regional repossession rate of 10% would be required to offset approximately one quarter of the negative effect of late payment on the GHQ Caseness Score (a 10% rate would reduce the GHQ Score by 0.24 points, the coefficient on the late payment variable is 1.06 (row1)). In Column 2, a similar exercise is undertaken for the case of the subjective consumer credit payments burden question and the regional bankruptcy rate. The interaction term between the two is again statistically significant (at the 1% level) both for renters (row 9) and owners (row 12), and stronger for owners. The magnitudes imply the onset of consumer credit problem debt in a region with a bankruptcy rate of 10% leads to approximately half the deterioration in psychological health which would be experienced at a bankruptcy rate of 0%. Table 10 repeats the exercise from Table 9 with the objective psychological stress measure as the dependent variable. In these specifications the interaction terms on reference-level mortgage arrears and the bankruptcy rate are both negative but are much less statistically significant.