Insured Privately? Wealth Stratification of Job Loss in the UK

Job loss is a significant income shock that can lead to declines in living standards and satisfaction. Wealth can provide a key resource in stratifying the risk and the consequences of such an event. In this article, I examine the extent to which wealth stratified the experience of job loss in the UK from 1991 to 2008. I distinguish between different wealth groups using information on homeownership and home value of primary residency, and then study whether these groups face different risks and/or consequences of job loss. The results show that renters were a significantly disadvantaged group compared to homeowners during the observation period. Not only did they faced a significantly higher risk of job loss, they also experienced greater declines in earnings, household income, and life satisfaction, and larger increases in income poverty in the year of job loss. Among homeowners, the risk and consequences of job loss were similar. In a country like the UK with minimal public insurance for unemployment, homeownership appears to provide a significant source of strat‐ ification for job loss.


Introduction
Wealth has recently gained wider recognition as a separate dimension of stratification and inequality.Wealth is more unequally distributed than income (Pfeffer & Waitkus, 2021) and is positively associated with a wide range of valued life outcomes: Those with higher wealth tend to have higher educational achievement, better health, and better subjective well-being (Killewald et al., 2017).Such wealth advantages are considered to arise from two main sources (Pfeffer, 2011).First, wealth provides a greater command over resources (e.g., through long-term use and cost savings) and can be used as collateral to access more resources and purchase goods, activities, and services.Second, wealth serves as both a psychological and a real private safety net against the consequences of risky behaviour and provides insurance against negative income shocks.Through these purchasing and insurance functions, wealth influences outcomes by shaping initial behaviour as well as its consequences.
These wealth advantages become especially critical around key life events.For anticipated and intentional events (e.g., various desired life transitions such as childbearing or marriage), wealth can provide the necessary resources, give individuals the opportunity to plan ahead, and possibly offer a choice on factors such as the timing, place, and type of such events.For unanticipated and unintentional events, such as unexpected income shocks, wealth can compensate for income and consumption losses and protect against negative consequences on living standards and satisfaction.
Job loss is an important life event that might be stratified by wealth.Evidence shows significant long-term losses in earnings, reductions in consumption, increased poverty risk, health deterioration, and declines in life satisfaction as a result of job loss (Brand, 2015;Jenkins, 2011;Kalleberg & von Wachter, 2017).Given the various advantages of wealth, not only the risk but the consequences of job loss as well may differ across wealth groups.While there is relatively little evidence on how the risk of job loss varies across wealth groups, evidence on the consequences of job loss across wealth groups is mixed.In a recent paper, Rodems and Pfeffer (2021) found strong stratification by wealth in the risk of material hardship following disruptive life events such as divorce, disability and income loss in the US.Kuhn and Brulé (2019), on the other hand, have argued that material resourcesincluding wealth-do not provide any buffer against the subjective well-being consequences of adverse events.André et al. (2019) recently showed differential effects of unemployment on subjective well-being between homeowners and renters in Australia, although they also found no stratification of effects among homeowners with different home equities.
In this article, I contribute to this literature by examining whether the risk and consequences of job loss vary across different wealth groups in the UK during 1991-2008.More specifically, I examine (a) how the risk of job loss differs across groups with different levels of wealth and (b) how the effects of job loss on valued life outcomes, such as poverty and life satisfaction, are moderated by wealth.To analyse the consequences of job loss, I use a dynamic event-study model (difference-indifferences) and focus on changes in the year of respondents' job loss relative to the year prior to the event.
To distinguish between wealth groups, I use information on housing wealth and separately look at (a) the differences between renters and homeowners and (b) the differences among homeowners with varying home values.The results reveal significant differences across wealth groups, specifically between renters and homeowners, in the consequences of job loss for earnings, poverty measured by income, and satisfaction with the use of leisure time.

Risk of Job Loss
Many have been concerned with rising insecurity in the labour market in recent decades due to loosened employment protections and the proliferation of atypical contracts (e.g., Kalleberg, 2011Kalleberg, , 2018)).However, empirical evidence has not always been supportive of these arguments.For example, except for peaks during recessionary periods and some modest trends observed for specific groups (e.g., low-skilled men in France and Germany), the rate of involuntary job loss has been relatively stable-if anything, it has decreased over the last three decades in most OECD countries (Bergmann & Mertens, 2011;Davis, 2008;Farber, 2017;Givord & Maurin, 2004;Quintini & Venn, 2013).Manning and Mazeine (2022) further argue that the widely supposed trend towards rising insecurity is not observed even in subjective measures of job insecurity for the UK, the US, and Germany.In the UK, during the present study's observation period (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008), the rate of dismissals, redundancies, and job separations (i.e., flows from employ-ment to unemployment) also shows a relatively decreasing trend (Upward & Wright, 2019).
Trends aside, job loss affects a significant group of people every year.Quintini and Venn (2013) estimate that the rate of job loss among workers varies between 3% and 5% in OECD countries.These numbers tend to be significantly higher in recessionary periods; for example, Farber (2017) estimated that the rate of job loss doubled in the US after the 2008 recession.Upward and Wright (2019) estimate that the rate of redundancy in the UK was between 2% and 4.5% during 1991 and 2008.

Stratification of Job Loss Risk by Wealth
While we know little about the stratification of job loss risk across wealth groups, evidence shows a higher risk of job loss for those with lower educational attainment and social class (Brand, 2015;Farber, 2017;Hacker & Rehm, 2022;Quintini & Venn, 2013).For example, in the UK, although the gradient across groups is relatively diminished in the early 2000s, the rate of job loss is below 1% for those with a degree, while it is around 2% for those with a degree below the GSCE level (i.e., high school; see Gomes, 2012).Similarly, a clear social class gradient is also observed between blue-collar and white-collar workers in the UK concerning the risk of experiencing unemployment during the observation period (Goldthorpe & McKnight, 2006).
Whether similar patterns can be observed across wealth groups is an open question.Wealth is closely associated with education and thus might reflect similar patterns, whereby wealthier groups have a lower risk of experiencing a job loss.However, given the psychological safety net that wealth provides, wealthier individuals might follow riskier pathways and be more likely to leave their jobs for better opportunities.At the same time, the labour market in the UK is considered relatively open, external, and flexible.Specifically, this means that skills are generally transferable and progression is typically reached through job mobility; regulations for firing and hiring are not strict for either regular or temporary workers, and the divide between outsiders and insiders is not strong, such that those with stable and unstable jobs have similar chances of losing or finding employment (Ferragina & Filetti, 2022;Häusermann & Schwander, 2012).Considering the above, there is no clear theoretical expectation of the patterns of job loss risk across the distribution of wealth.

Consequences of Job Loss
Job loss is a costly event for several outcomes.Evidence shows significant losses in yearly earnings, both in the short and long term.The level of losses is dependent on the speed and conditions of re-employment.In the short term, losses account for between 20% and 50% of previous income across high-income countries (Bertheau et al., 2022;Couch & Placzek, 2010;Farber, 2017).While the level of these losses reduces over time, they are persistent in the long term.For example, Davis and von Wachter (2011) show that the negative effect of job displacement on yearly earnings is still observable after 20 years in the US.In the UK, losses in the year of job loss are estimated at around 40-50% of pre-displacement earnings, which decreases to 18% after five years and 10% after 10 years (Hijzen et al., 2010;Upward & Wright, 2019).
These losses might be compensated through multiple private and public mechanisms.Three private mechanisms are particularly relevant to wealth.The first is dissaving.Building a buffer against unexpected income shocks is likely an important motivation for saving (Carroll, 1997), especially in countries with inadequate public insurance, such as the UK (Banks et al., 2001;Lugilde et al., 2019).When households anticipate an income shock, they respond by reducing consumption, moving their investments to safer assets and increasing savings prior to the shock event (e.g., Barceló & Villanueva, 2016;Hendren, 2017).Dissaving might include using liquid assets (e.g., cash savings), converting illiquid assets to cash and borrowing (e.g., taking out loans, using credit).For example, for Denmark, Andersen et al. (2021) show that, during the first two years after a job loss, reduced saving in liquid assets accounts for around half of the total loss in household incomes.Braxton et al. (2020) show that in the early 2000s, in the US, around one-third of those who lost their jobs replaced a significant part of their lost earnings by borrowing, which, in turn, put households in debt and only delayed the consequences of the income shock (Kalleberg & von Wachter, 2017;Sullivan, 2008).The second mechanism is financial support from other households (e.g., parents or relatives), which has been shown to be particularly relevant during life course events to support children in need.However, their amount is usually small compared to the extent of losses (e.g., Karagiannaki, 2011;Leopold & Schneider, 2011;McGarry, 2016).The third mechanism is household labour supply.The existence of other earners in the household is a critical source for limiting losses in household income (Figari et al., 2010) and labour supply response (e.g., added worker effect) is an additional source, but significant mainly in recessionary periods and only for countries with weaker social security, such as the UK (Bredtmann et al., 2018;Bryan & Longhi, 2018).
Earnings losses can also be compensated publicly through social security programmes.In the UK, however, the generosity of unemployment insurance (UI) is particularly low.During the observation period (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008), the benefit level of UI was fixed at around 15-20% of average wages for singles (e.g., £73 in 2020), which amounts to an average replacement rate of approximately 20% of previous earnings.Eligibility is conditional on previous employment (at least six months) and certain behavioural requirements, such as being available for and actively seeking work, and the benefit is available for a maximum of six months.Minimum income schemes (MIS), including social assistance, child and housing benefits, and tax credits, are relatively more generous and were significantly increased during the 1999 New Labour welfare reform, providing on average around 50-60% of median income (OECD, 2022).Therefore, MIS might be particularly helpful for compensating earnings lossesalbeit only for households with low income and few assets, as these schemes are generally means-tested in terms of both income and assets (i.e., savings and property ownership).
If not compensated through these private or public mechanisms, earnings losses following a job loss might lead to reductions in household income and affect living standards and satisfaction.In the year of job loss, individuals are estimated to lose around 5-30% of their household income across OECD countries and around 20% in the UK (see, among others, Di Nallo & Oesch, 2021;Ehlert, 2012;Seim, 2019).Evidence also shows significant declines in consumption following a job loss in Canada, Denmark, and the US (Andersen et al., 2021;Browning & Crossley, 2008;Ganong & Noel, 2019).Employment events-more specifically reductions in earnings-are key trigger events for entering poverty as measured by income in the UK (Jenkins, 2011).Job loss is also associated with declines in life satisfaction through its effects on mental health, family disruption, and loss of psychosocial assets (e.g., self-confidence, goals, and meaning in life; see Brand, 2015;Paul & Moser, 2009).Non-pecuniary functions of employment, such as structuring time and fostering social relationships, are also expected to be affected by job losses (Jahoda, 1981).Evidence suggests that unemployment is associated with the degree to which individuals perceive their use of time as structured and useful (Wanberg et al., 1997), although unemployed individuals spend more time on leisure and enjoyable activities (Hoang & Knabe, 2021).

Stratification of Job Loss Consequences by Wealth
The consequences of job loss might vary significantly across wealth groups due to differences in initial earnings losses or the level of private and public compensation.First, initial earnings might differ if, for example, wealthier individuals return to employment more quickly and with conditions more similar to their previous job, compared with less wealthy individuals.For example, those with lower educational attainment remain unemployed for longer after a job loss (Quintini & Venn, 2013).
Second, the level of private compensation is likely to be higher for wealthier groups.Those with higher wealth have more savings (Rowlingson & McKay, 2011, pp. 53-80), are likely to have greater precautionary savings (Jappelli & Pistaferri, 2017), are more likely to borrow (Sullivan, 2008) and have higher incidences and levels of transfers from other households (Nolan et al., 2022).At the same time, most families do not have adequate liquid financial wealth to compensate for significant losses in earnings (Dickens et al., 2017;Gustafsson et al., 2021).For example, in the UK, around half of working-age households have less in savings than their monthly incomes, and around three-quarters have less than six months' monthly income (Gustafsson et al., 2021).In the US, those in the bottom decile of wealth do not generally borrow in response to income losses following a job loss, given their limited access to credit (Sullivan, 2008).
Third, in the UK, public compensation is likely to be significant only for those with low wealth, given its targeted design.All households might benefit from the UI scheme, as it does not include means testing, but given that it provides a relatively meagre and fixed amount, its value for wealthier households is likely to be insignificant.MIS benefits in the UK can only be received by those with low incomes and assets and hence are likely to compensate some losses for these groups.
As a result of the stratified process in initial earnings losses and the ability to compensate losses privately, the consequences of job loss in terms of poverty and life satisfaction are likely to be worse for less wealthy households.Public compensation can replace income losses only for those at the bottom of the distribution.For wealthier groups, job loss-even if it leads to significant losses in earnings-might have a rather limited effect on the risk of poverty and life satisfaction.

Housing Wealth and Wealth Stratification
I define wealth groups based on housing wealth.More specifically, I first distinguish between renters and homeowners.I then investigate how the effects of interest vary among homeowners by examining differences across the distribution of the gross value of their primary residence.
This choice can be justified on three main grounds.First, the wealth stratification patterns described above can equally apply to housing wealth.For example, housing wealth can provide an insurance function and prove to be a resource for smoothing consumption against transitory income shocks (Carroll, 1997).While one generally expects households to allocate precautionary savings to liquid assets with little cost (e.g., savings accounts), Carroll et al. (2003) find that precautionary wealth accumulated by the wealthier is mainly reflected in housing rather than other types of more liquid wealth.Similarly, a large literature has consistently found significant consumption responses to changes in housing wealth (Berger et al., 2018;Campbell & Cocco, 2007), which have sometimes been shown to be much higher than consumption responses to changes in financial wealth (e.g., Carroll et al., 2011).These housing effects are mainly explained by changes in households' perceived wealth or relaxed borrowing constraints (Campbell & Cocco, 2007).
Second, homeownership is not only a valued outcome as a marker of transition to adulthood, social status, and family formation (Bayrakdar et al., 2019;Coulter et al., 2020); it also represents the main-and often only-component of wealth for most households.For example, over the last three decades, around 60-65% of total net wealth (excluding private pensions) in the UK has been in housing (Office for National Statistics, 2022).Indeed, the gross value of a person's primary residence has been shown to highly correlate and closely proxy overall wealth in its associations with stratification outcomes, such as education and marriage (Blanden et al., 2021;Pfeffer & Killewald, 2017;Wagner et al., 2020).
Third, housing wealth is also the main source of wealth inequality within and between countries (Pfeffer & Waitkus, 2021).For example, substantial rises in home prices almost entirely explain the reductions in wealth inequality in the UK between 1996 and 2005 (Bastagli & Hills, 2012).In recent decades, homeownership has increased in tandem with rising housing prices, which has helped to slow the growth in wealth inequality (Holmans et al., 2007).However, the significant divide is now between renters and homeowners, which is usually a matter of having no wealth versus some wealth (Coulter, 2016).Variation in home value, on the other hand, reflects differences between those with low and high wealth.

Data
I used a sample from the British Household Panel Survey (BHPS) comprising data collected between 1991 and 2008.The initial BHPS sample included households from Britain; since 2001, with boost samples, Northern Ireland has also been included.I did not use data from Understanding Society (UKHLS), the successor of the BHPS for the more recent period, because yearly household income cannot be estimated based on the information collected in the UKHLS.

Target Population and Sample Selection
Given this study's focus on job loss, the target population consists of prime-age workers (i.e., between 25 and 55 years old) in the UK during the period 1991-2008.I used an unbalanced sample of 15,949 individuals with information from at least two years.I removed selfemployed individuals, full-time students, those who selfdescribed as long-term sick or disabled, retirees, and inactive from the sample, as their employment trajectories are likely to differ from the overall working-age population.To account for attrition, I calculated longitudinal weights for each outcome, accounting for the probability of dropping out in three-year periods and multiplying the inverse of this risk of attrition with cross-sectional weights of the middle year (i.e., the base year).

Measures
The main event-job loss-was defined based on individual information on monthly unemployment.I define job loss based on three conditions: (a) being unemployed for at least three months in the current year, (b) being employed for at least 4.5 months in the previous year, and (c) not being unemployed for more than three months in the previous year.This definition captures more substantive shocks while leaving out temporary movements in and out of employment that would be expected to have limited influence on yearly outcomes.Moreover, it includes both employer-initiated events (e.g., termination, lay-offs due to downsizing, closure, other business operations) and employee-initiated events (e.g., health problems, care responsibilities, dissatisfaction with work or career).
The variables used to measure homeownership and home value are self-reported.I used the question on housing tenure to define homeownership.Those who owned their house (all or share), either through a mortgage or outright payment, were defined as homeowners and all others as renters.Home value is the gross value of respondents' primary residence and reflects respondents' answers to the question: "How much would you expect to get for your home if you sold it today?"This variable is transformed into a percentile rank.
I used multiple outcome measures (namely earnings, net household income, poverty based on income, and poverty based on deprivation) and various satisfaction measures (namely satisfaction with life overall, social life, amount of leisure time, and use of leisure time).For earnings and net household income, I used the imputed variables provided by Levy and Jenkins (2012), which subsequently show gross yearly earnings considering usual pay from main and second jobs (including income from self-employment) and net household income (including earnings, private transfers, investment income, taxes, and transfers).I equivalised household income using the modified OECD scale.I used two poverty measures: one based on income, with a threshold of 60% of median household income, and the other based on deprivation, using six available deprivation items and with a threshold of one (i.e., having more than one deprivation was considered poverty).These six items are the ability to afford (a) keeping one's home warm, (b) paying for an annual holiday, (c) replacing old furniture, (d) buying new clothes, (e) eating meat on alternate days, and (f) having visitors once a month.Lastly, satisfaction indicators were measured with a seven-point Likert scale following the question: "In general, are you satisfied with your life?" Answers ranged from completely satisfied to not satisfied at all.
As control variables, I included other types of critical events that might be associated with job loss and affect outcomes, such as partnership dissolution (formal divorce of marriage or civil partnership, widowhood, separation) and number of children.I used age, calendar year, and gender to residualise outcome variables.
Data for all these variables are available for the period 1990-2008, except material deprivation and life satisfaction, for which data are only available from 1996.
In Table 1, descriptive statistics for the target population are presented.These statistics are presented separately for renters versus homeowners and treatment versus control groups in the Supplementary Material.

Effects of Interests
I am interested in the change in outcomes (Y) in the year of job loss event (E) compared to the level just prior to the event.This is a descriptive quantity: The group of individuals who experience a job loss is likely to be selective and the interest here is in describing the outcomes of this group, not identifying the causal effect of job loss (in Section 4.5, I discuss further what this means for estimation).
My questions involve whether and how much this effect varies (a) between renters and homeowners and (b) among homeowners with different home values.These can be formally defined as follows: a. Renters vs. homeowners: b.Among homeowners with different home values: In these definitions, t is event time (where t = 1 is the year of the event), r is homeownership status (where r = 1 is renter), and q is percentiles of home value.

Estimation
To estimate these effects of interest, I used a dynamic event-study model (difference-in-differences) with fixed effects, defined as follows: Prior to the analysis, Y was outcome residualised for year, age and gender.The first term is event time t dummies, including two lags and four lead years (i.e., t = −2, −1, 0, 1, 2, 3, 4) relative to the timing of job loss.The remaining lags and leads are bunched together.C is controls, such as partnership and number of children, and  i is the individual fixed effects.I am interested on the effect in the year of job loss-that is,  i1 , which is an estimate of  1rseparately for renters and homeowners.This is a difference-in-differences specification, where the control group is those who never experience a job loss.Therefore, the effects show the difference between those who experienced a job loss (treatment) and those who did not experience a job loss (control).Although the effect of interest here is purely descriptive, I apply this model to account for general trends in the population, more specifically in workers' earnings.For example, part of the loss in earnings after a job loss Notes: Estimates are weighted using survey design weights; education shows the highest qualification measured based on five categories, namely, degree (5), other high degree, A-level, GCSE, other qualification, and no qualification (1).comes from foregone earnings: the possible growth in earnings if the job loss had not occurred.By including a control group who have never experienced a job loss, such growth trajectories can be accounted for.
In the Supplementary File, I compare the characteristics of the treatment and control groups separately for renters and homeowners.Overall, the groups appear to be very similar except that the control group has slightly higher homeownership and value, more children, higher earnings and household income, and lower deprivation.Still, the differences are relatively small.Individual fixed effects also account for differences among these groups that are constant over time.
For the comparison among owners, I apply a smoothvarying coefficient model (Rios-Avila, 2020), using the same model as in Equation 1.The idea is to estimate a linear relationship between two variables, such as job loss and income, conditional on a non-linear smoothing variable, such as home value percentiles.Similar to nonparametric regressions, a smoothing function is first estimated based on a kernel method.Then, the coefficients from that model are used in the main model where the main relationship is estimated for each chosen bandwidth (e.g., 20 groups, one for each five-percentiles).Thus, it is a semi-parametric model that relaxes the linearity assumption of interaction models and allows for estimating interaction effects flexibly across a continuous variable.
Before estimation, outcome variables were residualised for year, age, and gender fixed effects.Such flexible residualisation aims to average out period, life cycle, and gender effects.I used log transformation for earnings and household income and present the results as semi-elasticities (i.e., percent change in outcome in the year of job loss relative to the previous year).Average marginal effects are presented for poverty, showing percentage changes in the rate of poverty, while standardised coefficients are used for the satisfaction measures.The first graph shows the rate of job loss across the observation period and how it varies between renters and owners; the second graph shows how the rate of job loss among homeowners across the distribution of home value in the pooled data; quantiles are defined based on the gross home value of primary residency in the year before the job loss; the estimates of the first graph show grouped averages across years, while the grey areas represent 95% confidence intervals of the point estimate; the second graph shows estimates from pooled data based on binned scatters (using 100 bins) and quadratic fit, calculated using binscatterplot command in Stata (see Stepner, 2013); for definitions of outcomes see Section 7.1.

Risk of Job Loss
estimates from a binned scatter plot of the pooled data (i.e., for all years), where the quantiles are defined separately for each year based on the respondent's gross home value in the previous year (i.e., the year before the job loss), then averaged across years.The results for the former comparison show grouped mean estimates across years with 95% confidence intervals, indicated in grey.
The results show significant variation between renters and homeowners.While in general there is a downward trend in the risk of job loss, renters have a consistently higher rate of job loss than homeowners over the years.On average, the rate of job loss is around around 3% for renters and 1-1.5% for homeowners.Among homeowners, however, we do not see much variation.Across the distribution, the rate of job loss is around 1-1.5%.Considering these results, the main divide appears to be between renters and homeowners, while the risk of job loss is relatively equally distributed among homeowners.

Consequences of Job Loss for Renters Versus Homeowners
Figure 2 shows how various outcomes differ in the year of job loss for renters compared with homeowners.The estimates are presented as percentages, rates, or standard deviation changes relative to the base year (t − 1).
Starting with yearly gross earnings and yearly net household income, both renters and homeowners lost a significant percentage of their earnings and income in the year of job loss.However, renters lost considerably more than homeowners: 89% versus 56% of earnings and 29% versus 22% of household income.These differences are statistically significant for earnings but not for household income.The losses in household income are lower, possibly due to household labour supply or transfers from other households.
Similar differences are also observed for poverty as measured by income.In the year of job loss, the risk of poverty increases by 20% among renters compared with 8% among homeowners.Both changes are statistically different from both each other and zero.These differences might be a result of the different losses experienced for household income but also the position of renters versus homeowners in the income distribution: renters are more likely to be closer to the poverty threshold, which may lead to more people falling below this threshold after job loss.Even if similar relative losses are observed for renters and homeowners, the risk of poverty rises more for renters given their initial position.
There are some increases in the risk of poverty measured in terms of deprivation, around 4% for both groups.Given the differences in income poverty risk The graphs show changes in different outcomes in the year of job loss and how this effect varies between renters and homeowners; results for income variables show percentage losses, poverty variables show increases in the risk of poverty, and satisfaction variables show changes in standard deviation; the estimates are based on a difference-in-differences model.The shaded areas show a 95% confidence interval of estimates; the models include other risk events such as partnership dissolution and childbirth as controls; outcomes are residualized for year, age, and gender; for definitions of outcomes see Section 7.1.between renters and homeowners, this might be surprising.However, first, the losses in household income are very similar between the two groups, and second, material deprivation shows lack of consumption.Hence, both groups seem to manage to smooth consumption to the extent that they do not face significant consequences for material deprivation.
Lastly, significant differences between renters and homeowners can also be observed for the satisfaction measures, specifically regarding use of leisure time.Homeowners experience slight increases in their satisfaction with life overall and with their amount and use of leisure time, as well as slight decreases in satisfaction with their social lives (around 0.12 standard deviation).None of these changes, however, is statistically significant.On the other hand, renters experience statistically and substantively significant declines in their satisfaction with their social lives and use of leisure time (by 0.3 and 0.42 standard deviation, respectively).Therefore, although homeowners experience significant declines in earnings and household income following a job loss, the effect of these losses on other life outcomes, such as poverty and life satisfaction, is relatively limited.

Consequences of Job Loss Among Homeowners
In Figure 3, I show how various outcomes differ in the year of job loss and whether these changes vary among homeowners.The estimates are presented as percentages, rates or standard deviation changes relative to the base year (t − 1).
The results show significant losses in earnings (around 40-50% of pre-job loss earnings) among all homeowners.These losses are relatively similar across the distribution.Likewise, losses in household income are significant (around 20-25%) and generally similar across the distribution.On the other hand, the changes in the risk of poverty as measured by income follow a clear gradient, high at the bottom (e.g., around 18%) and low-almost zero-at the top.This may be due to the varying positions of these groups in the income distribution: Those with lower home values are more likely to fall into poverty because their initial incomes are more likely to be closer to the poverty threshold.
Despite such a clear gradient in poverty as measured by income, poverty as measured by deprivation does not appear to significantly rise in the year of job loss for any homeowners across the distribution.Similarly, there are no differences in satisfaction with overall life, social life, or amount and use of leisure time among households with different home values.Therefore, despite significant losses in earnings and household income, the consequences of job loss for deprivation and life satisfaction appear to be muted for homeowners, even those with the lowest home values.This is possibly due to the private insurance provided by wealth, specifically through  2020), a semiparametric kernel regression, where the effect of job loss varies as a "smooth" function of quantiles; the shape of the function is estimated using multiple thresholds (i.e., 20) and certain bandwidths, which are optimally estimated by the program (using vc_pack Stata package by Rios-Avila, 2020); the shaded areas show a 95% confidence interval of estimates; the models include other risk events such as partnership dissolution and number of children as controls; outcomes are residualized for year, age and gender; for definitions of outcomes see Section 7.1.dissaving, which seems to be equally relevant across the distribution of housing wealth.

Robustness Checks
I tested the sensitivity of results to the definitions of job loss and renter status.The results are presented in the Supplementary File.Regarding the definition of job loss, increasing or decreasing the condition of unemployment duration following the job loss event did not significantly affect the results.Using a longer unemployment condition (at least five months) decreased the risk of job loss, while using a shorter unemployment condition (at least two months) increased the risk of job loss specifically for renters and those with lower home values, reflecting a gradient across wealth groups (Supplementary File, Figures S2-S3).However, these changes in the definition of job loss did not alter the findings on the consequences (Supplementary File, Figures S4-S7).
Regarding the definition of a renter, I tested whether social renters (e.g., those who rented through social housing or housing associations) differed from private renters, as the latter might be relatively more disadvantaged in terms of housing cost and security.Around half of all renters are social renters.The consequences for social renters are substantively similar and only slightly (but not significantly) worse for poverty as measured by deprivation and for life satisfaction (Supplementary File, Figure S1).I also ran the analysis separately for women and men.The results showed generally greater effects for women, especially for renters' life satisfaction measures (Supplementary File, Figures S12-S15).Lastly, I ran the analysis using the longitudinal weights provided by the survey, with substantively similar results (Supplementary File, Figures S8-S11).

Discussion
I asked whether there is any variation in how significant job loss events were experienced across different wealth groups in the UK between 1991 and 2008.Various advantages of wealth, mainly compensating income losses and smoothing consumption, provide the basis for stratification of risks and consequences of job loss.I examined outcomes related to income, consumption and life satisfaction and how the effect of a job loss on these outcomes varies across wealth groups, which are defined based on housing wealth.
The results reveal that the main divide is between renters and homeowners.Risk of job loss was significantly higher for renters than homeowners throughout the observation period, while it did not vary among homeowners with different home values.Similarly, the short-term consequences of job loss for earnings, income-based poverty and satisfaction with life and use of leisure time differ significantly for renters and homeowners, but less so among homeowners with different home values.Among homeowners, both the risk and consequences of job loss are relatively similar across the distribution (with one exception of income-based poverty possibly due to varying income positions relative to the poverty threshold of those with different home values).
What might explain homeowners' advantage over renters when it comes to the consequences of job loss?There might be mechanisms directly or indirectly related to wealth.The first is the private compensation mechanisms that are directly related to greater wealth, such as dissaving for smoothing consumption.This is a reasonable explanation, as we find similar losses in household income between renters and homeowners but significantly different outcomes for poverty as measured by income and satisfaction measures.Therefore, although homeowners may experience significant losses in earnings and household income in the year of job loss, they may compensate for these losses through their savings and limit the negative effects of the job loss on life satisfaction.The second is the mechanisms that are indirectly related to wealth, more specifically selection into homeownership (e.g., Lersch & Dewilde, 2019).Homeowners might be advantaged in observed characteristics (e.g., education) or unobserved characteristics (e.g., personality, social skills, networks) that are conducive to faster re-employment with better conditions, higher savings, and better life satisfaction.Similarly, given assortative mating on education and employment, homeowners might be more likely to have other earners in their household, which limits losses in household income.Alternatively, given their advantaged position, they might be more likely to receive support from other households, such as parents or relatives.
Why do we not find similar differences among homeowners with different home values in terms of the risk and consequences of job loss?Regarding the risk of job loss, we did not have a clear expectation considering the flexible and open labour markets in the UK, and the results show no clear difference in the risk of job loss among homeowners.Regarding the consequences of job loss, similar outcomes for earnings and household income shows that the level of income shock was similar among homeowners.Given that, outcomes for deprivation-based poverty and life satisfaction does not vary among homeowners possibly because all homeowners (so not only the richer ones) had enough resources to compensate for income losses through dissaving and prevented any declines in consumption and life satisfaction.This supports arguments that the renters have become a particularly disadvantaged group in the UK, which has significant implications for the life chances of future generations (Coulter, 2016).
The analysis is limited in several respects.First, I only examined the consequences in the year of job loss, not how these losses develop over time and are affected by anticipation prior to the event.Second, individuals who experience job loss are likely to be a selective group, and certain characteristics that lead to job loss might also be the source of disadvantaged outcomes.Similarly, homeowners' advantages might not be related to their wealth per se but rather to other observed or unobserved characteristics.Third, although the analysis reveals important insights into what compensation mechanisms may be responsible for differences in consequences among wealth groups, they are not directly observed in the analysis.Lastly, this analysis is based on the UK, a country with relatively meagre social security; hence, the results might not be generalisable to areas with more generous UI schemes.
Especially in countries without adequate social insurance systems (and efforts to suppress the risk of negative life events) such as the UK, private insurance is a crucial factor for maintaining living standards and satisfaction after individuals are exposed to incomedisrupting life events.This study's results show stratification in the risk and consequences of job loss, specifically between renters and homeowners.although it is still a relatively rare event to have a significant impact in the short run, job loss risk might lead to widening income and wealth inequality, in the long run, considering the stratification in its economic consequences.Adapting a generous UI scheme could help reduce layers of inequality and serve as a crucial strategy for dealing with the risk of job loss (Kalleberg & von Wachter, 2017).

Figure 1 Figure 1 .
Figure 1 shows the risk of job loss over two decades, between 1991 and 2008, (a) for renters and homeowners and (b) among homeowners across the distribution of home values.The results for the latter comparison show

Figure 2 .
Figure2.Outcomes in the year of job loss: Renters vs. homeowners.Notes: The graphs show changes in different outcomes in the year of job loss and how this effect varies between renters and homeowners; results for income variables show percentage losses, poverty variables show increases in the risk of poverty, and satisfaction variables show changes in standard deviation; the estimates are based on a difference-in-differences model.The shaded areas show a 95% confidence interval of estimates; the models include other risk events such as partnership dissolution and childbirth as controls; outcomes are residualized for year, age, and gender; for definitions of outcomes see Section 7.1.

Figure 3 .
Figure3.Outcomes in the year of job loss among owners across the distribution.Notes: The graphs show changes in different outcomes in the year of job loss and how this effect varies across the distribution of home value; quantiles are constructed based on home value in the year before the job loss.Results for income variables show percentage losses, poverty variables show increases in the risk of poverty, and satisfaction variables show changes in standard deviation; the estimates are based on a smooth-varying coefficient model proposed by Rios-Avila (2020), a semiparametric kernel regression, where the effect of job loss varies as a "smooth" function of quantiles; the shape of the function is estimated using multiple thresholds (i.e., 20) and certain bandwidths, which are optimally estimated by the program (using vc_pack Stata package by Rios-Avila, 2020); the shaded areas show a 95% confidence interval of estimates; the models include other risk events such as partnership dissolution and number of children as controls; outcomes are residualized for year, age and gender; for definitions of outcomes see Section 7.1.