Employment and the Housing and Living Arrangements of Young Men: New Evidence from the Great Recession Gary V. Engelhardt Syracuse University Michael D. Eriksen University of Cincinnati Nadia Greenhalgh-Stanley Kent State University May 25, 2015 Abstract We investigate how extended families cope with adverse labor-market conditions and present new estimates of the loss of full-time employment for men during the Great Recession on coresidence with parents. Based on detailed data on sons of respondents in the Health and Retirement Study (HRS), our fixed-effect estimates indicate that on average a young man moving from full-time to either part-time to non-employment raises the likelihood of co-residing with a parent by approximately 10 percentage points. These effects are economically large, statistically significant, and suggest that the ability to move in with parents is an important way that young men adjust during economic downturns. We estimate the impact of employment loss during the recession on homeownership and mortgage distress: moving from full-time to nonemployment lowers the likelihood of owning a home by 8 percentage points and raises the likelihood of falling behind on payments and foreclose by 8 and 5 percentage points, respectively. Overall, we find that employment loss during the recession had a pronounced effect on the housing decisions of young men. Acknowledgements: The research reported herein was supported by a grant from the MacArthur Foundation program on How Housing Matters. The opinions and conclusions are solely those of the authors and should not be construed as representing the opinions or policy of the MacArthur Foundation, Syracuse University, Kent State University, or University of Cincinnati. All errors are our own.
1. Introduction How do extended families cope economically with adverse labor-market conditions such as unemployment? The extent to which families provide informal insurance arrangements among members against labor-market risk is a long-standing topic of interest in economics. In addition to private transfers of time and money (Altonji et al., 1997; Cox, 1987; Dynarski and Gruber, 1997),Taste families may provide insurance in the form of in-kind transfers such as housing (Bianchi et al., 2006). Recently, there has been substantial interest in whether parents use shared living arrangements as a form of risk-sharing in the face of labor-market uncertainty for their adult children (Kaplan, 2009, 2012; Wiemers, 2012). In this paper, we make three contributions to the literature. First, we present new estimates of the loss of full-time employment on co-residence between parents and adult children. We follow a large number of existing studies and focus on men, who often experience the most difficulty in adjusting to employment loss. As labor supply of prime-age men is very inelastic, most of the movements in and out of employment are likely to be from labor-demand rather than labor-supply shocks. Our estimates are based on the experience of these men in the Great Recession. In particular, we used detailed data on sons of respondents in the Health and Retirement Study (HRS) to construct a panel dataset of young men and their parents. We are able to track the men s employment and residential statuses before and during the recession and, therefore, see to what extent changes in labor-market attachment result in shared living arrangements with parents, and how this differs across socio-economic strata. A key empirical challenge is that there may be unobserved heterogeneity in the taste for leisure that may be correlated with the demand for shared living arrangements. If men with a high taste for leisure also prefer to live with their parents, then standard estimates examining the 2
effect of employment loss on parental co-residence would be biased upward. As a second contribution, we use a variety of fixed-effect estimation strategies that rely on the rich HRS panel data in an attempt to circumvent this concern. We also account for parents own health, income, employment, housing, and wealth trajectories in our estimation. Our fixed-effect estimates indicate that on average a young man moving from full-time to non-employment raises the likelihood of co-residing with a parent by 9 percentage points. These effects are economically large, statistically significant, and suggest that the ability to move in with parents is an important way that men adjust during economic downturns. We find similar estimates for moving from full- to part-time employment, which suggests that it is extensive margin of work that matters. Moreover, we find a pronounced effect of employment loss on marriage: moving from full-time to non-employment lowers the likelihood of being married by 7 percentage points. Although there are some differences by age, our estimates are uniform across other socio-economic strata, likely reflecting the fact that the depth and length of the recession left comparatively few families unscathed. During the recession, the national unemployment rate peaked at 10 percent, homeownership rates dropped significantly from historical highs, and mortgage distress rose sharply. As a third contribution, we estimate the impact of employment loss during the recession on homeownership and mortgage distress: moving from full-time to non-employment lowers the likelihood of owning a home by 8 percentage points and raises the likelihood of falling behind on payments and foreclose by 8 and 5 percentage points, respectively. Overall, we find that employment loss during the recession had a pronounced effect on the housing decisions of young men. 3
The remainder of the paper is organized as follows. Section 2 briefly reviews the previous literature in economics on living arrangements. Section 3 describes the basic econometric framework and HRS data. Section 4 presents the results on co-residence, as well as a large set of extensions and robustness checks. The homeownership results are discussed in Section 5. The paper concludes with a brief summary of findings and a discussion of caveats. 2. Related Studies Our analysis is most closely related to five interconnected strands in the existing literature in economics and demography. The first is centered in urban economics and has focused on the effect of household formation on homeownership rates. The homeownership rate is measured as the number of owner-occupier households divided by the total number of households. Economic conditions can affect both the number of owners versus renters via tenure choice (the numerator), as well as the number of households via household formation (the denominator). In an early, influential study, Börsch-Supan (1986) used data from the 1976-1977 waves of the American Housing Survey (AHS) and showed empirically that household formation was quite responsive to house prices and income. Other notable work in this area includes Haurin, Hendershott, and Kim (1993, 1994), Yelowitz (2007), Haurin and Rosenthal (2008), Lee and Painter (2013), and Paciorek (2013). A second strand has treated the work and household formation decisions of young adults as jointly determined. McElroy (1985) is the best-known study in this area. 1 In particular, she specified a structural model of labor supply and household formation, which she estimated using data on 203 never-married white men with completed schooling between the ages of 19 and 24 1 Haurin, Hendershott, and Kim (1993, 1994) also treat work decisions and income as endogenous. 4
from the 1971 National Longitudinal Survey of Young Men (NLS-YM) matched to data on their parents from the National Longitudinal Surveys of Mature Women (NLS-MW) and Men (NLS- MM), respectively. There are two important implications of her analysis. First, there is an ordered relationship between wage offers, employment, and residential independence. For wage offers below the adult child s reservation wage, the child does not work and co-reside with the parents; for offers above the reservation wage, the child works but still co-resides; for offers sufficiently above the reservation wage, the child works and lives independently. Second, the option of an adult child to co-reside with parents is a form of nonemployment insurance. McElroy s analysis abstracts from other forms of intergenerational support to young adults. A third strand of existing studies has treated co-residence as a type of intergenerational transfer that may be a substitute for (or complement to) financial transfers. This is a voluminous literature, whose focus is often on pinning down the motives for transfers. It spans both economics and demography, and is reviewed in detail in Bianchi et al. (2006). Ermisch (1999) is the best known paper in urban economics in this vein. He treats income as exogenous, but considers financial transfers in addition to co-residence. Rosenzweig and Wolpin (1993) treat human capital investments (and hence lifetime income), co-residence, and financial transfers as jointly determined. They estimate the parameters of a structural model based on a panel of 821 matched parent-son pairs for young men from the 1967 NLS-YM matched to parents in the NLS- MW and NLS-MM, and followed up to eleven waves between 1967 and 1981. They found that an additional week of unemployment of the son increased the likelihood of parental co-residence by 3 percentage points, based on fixed effects logit estimation. In economics, much of the existing work has examined the role of economic conditions on initial transitions out of the parental home and new household formation. In contrast, much 5
effort has been spent in the demography literature in making distinctions between different types of transitions. This includes those who initially transition out of the parental home, as well as their complement, i.e. those who fail to launch (Goldsheider and DaVanzo, 1985; Billari and Liefbroer, 2007; Bianchi et al., 2006; among many others). More recently, researchers have begun to examine the link between the economic status and residential patterns of young and middle-aged adults after they have had a spell living independently, with those who return to the parental home termed boomerang kids. Kaplan (2012) embedded the option of the adult child to move in and out of the parental home into an intergenerational dynamic, stochastic lifecycle framework. Building on McElroy (1985), co-residence in his model is a form of nonemployment insurance that allows adult children potentially to smooth non-housing consumption in the face of labor-market uncertainty. He estimated the structural model for 1,491 young men who did not attend college using monthly data from 1998 through 2002 on employment and co-residence drawn from the National Longitudinal Survey of Youth, 1997 Cohort (NLSY97) and found that labor market shocks affect the timing of moves into and out of the parental home. This type of insurance is particularly valuable for individuals with parents in the lower part of the income distribution, who have comparatively lower ability to supply financial transfers to the adult child in the face of an adverse labor-market shock. The option of moving back home also is associated with higher future earnings by allowing for more extensive search for jobs with higher earnings potential. Although Kaplan s period of study was 1998-2002, his work was particularly well-timed in that it appeared just when the United States was in the midst of the largest macroeconomic contraction since the Great Depression. This period beginning with the financial crisis in 2007, often termed the Great Recession, saw a historically high level of co-residence among American 6
individuals (Mykyta and Macartney, 2012; Mykyta, 2012). This has led to a number of very recent studies of the impact of the recession on living arrangements, all of which are closely related to our analysis below. Lee and Painter (2013) examined the impact of recessions on household formation. Specifically, they used data on young men and women from the Panel Study of Income Dynamics (PSID) from 1975-2009, which span four recessions as dated by the National Bureau of Economic Research (NBER), including the Great Recession. They found that being unemployed has a substantial negative impact on the transition out of the parental home for young adults. In particular, treating employment status as exogenous, their multinomial logit estimates indicate that unemployment lowers the likelihood of a transition by 50%. In addition, they found that, independent of the individual s unemployment status, more general macroeconomic and housing-market conditions matter for household formation, including the state unemployment rate, average wage, GDP growth, median gross rent, median house value, respectively, as well as whether the national economy is in a recession year. They also found much stronger effects for the Great Recession than earlier economic downturns. 2 Rogers and Winkler (2014) used American Community Survey (ACS) data from 2005 and 2011 and examined how the differential timing of the downturn in both labor and housing markets across metropolitan areas in the Great Recession affected household formation by young adults. Although their focus was on the impact of broader market conditions, treating employment status as exogenous, their logit estimates indicated that not being employed reduced the likelihood of living independently by 11%. 2 Wiemers and Bianchi (2014) found substantial differences in co-residence among women by birth cohort, primarily driven by changes in life expectancy across cohorts. 7
In a closely related study, Wiemers (2012) examined the impact of unemployment on household composition and doubling up. She used data from the 1996, 2001, 2004, and 2008 Survey of Income and Program Participation (SIPP) panels, which span the period from December, 1998, through December, 2010. 3 With repeated observations on individuals, she estimated individual-specific fixed effects linear probability models and found that becoming unemployed raises the likelihood of living in a shared arrangement by 0.3%. While statistically significant, this an economically small effect. 4 3. Econometric Framework and Data Let i index the individual, j index the family, and t index the calendar year. Here, the individual refers to the adult child. We follow previous studies and specify a linear-inparameters econometric specification: Y U P δx. (1.1) ijt it it jt t ijt In the basic empirical models below, the dependent variable, Y, is an indicator that takes on a value of one if the individual co-resides at period t with a parent and zero otherwise. In extensions, we allow for dependent variables that measure finer gradations of household formation and housing choices. 5 Our primary interest is in the impact of the loss of full-time employment on co-residence. There are two focal explanatory variables. The first is U, an indicator variable that takes on a value of one if the individual is not employed and zero otherwise. The second is P, an indicator 3 There are two breaks, one in 2000 and one in 2007. 4 Dettling and Hsu (2014) examined a related topic, which is the extent to which increased debt accumulation by young adults results in increased co-residence with parents. 5 We interpret this specification as a reduced-form. We note, however, that most of the structural analyses in the literature specify indirect utility from residential choice as linear in parameters, yielding in effect a similar specification as the reduced-form we adopt above. 8
variable that takes on a value of one if the individual is part-time employed and zero otherwise. Those who are full-time employed are the excluded group in (1.1), and, therefore, represents the (conditional) mean fraction of men who are full-time employed and co-residing with a parent. The central objective is to obtain consistent estimates of and, which, for those who begin the previous period in full-time work, measure the impact of moving from full-time to nonemployment and part-time to non-employment on parental co-residence, respectively. We construct our analysis dataset from the HRS, a stratified random sample of over 25,000 individuals 50 and older, and their spouses (regardless of age). Individuals in the study are interviewed every even-numbered calendar year until they die, at which point an exit interview is conducted with their next of kin to gather information on the health and economic circumstances prior to and at the time of death. The study began in 1992, and every six years (e.g., 1998, 2004, 2010, 2016, etc.), a new birth cohort of individuals in their mid-50s enters the study, refreshing the panel. We use data from 2004-2012 waves that span both before and immediately after the financial crisis and recession. A key feature of the HRS is that it asks respondents in each wave about the economic, locational, and demographic status of their children. Given that the HRS sampling frame consists of individuals in their 50 s and older, the great majority of children of HRS respondents are adults. The implication then is that the HRS can be used to construct paired data on parents and adult children to analyze the impact of labor-market dynamics on co-residence. We follow the bulk of the previous literature in economics and focus on adult male children due to their inelastic labor supply (McElroy, 1985; Rosenzweig and Wolpin, 1993; Kaplan, 2009, 2012, among others). 9
In particular, we construct a panel dataset of adult men who are the children of HRS respondents. We restrict ourselves to men born between 1960 and 1982, who would have been between the ages of 23 and 52 during our sample period. To this, we merge economic, health, and demographic information about their parents, who are HRS respondents. This results in a panel dataset of paired data on adult men and their parents. Column 1 of Panel A of Table 1 presents basic summary statistics on the adult men in the sample. There are 45,958 person-year observations on 11,989 men. The average age was 40. Almost 64% were married. About 82% worked in full-time employment at the time of the interview; 5% worked part-time; and almost 12% did not work. The top row of Panel B shows summary statistics for the primary outcome in the empirical analysis: the fraction of men who co-reside with a biological parent or parent-in-law. About 9% of men in the sample resided with a parent. Panel C shows summary statistics for the parents of those men. The average age of fathers was 67, and 42% were employed and the time of their son s interview. The average age of mothers was 66, and 35% were employed. 4. Empirical Results for Parental Co-Residence Columns 2-4 of Table 1 show summary statistics for subsamples defined by the three employment categories: full-time, part-time, and not working, respectively. Reading across columns in the first row of Panel B, parental co-residence is lower for the full-time employed relative to the other two categories. In total, 6.5% of men who work full-time co-resided with a parent (column 2), whereas 22.9% of those not employed did (column 4). The raw difference in co-residence between the groups is 16.4 percentage points (i.e., 0.164 0.229 0.065). That is, 10
on average moving from full-time to non-employment raises the likelihood of co-residing with a parent by 16.4 percentage points. The raw difference in co-residence between the full- and parttime employed is 14.7 percentage points (i.e., 0.147 0.212 0.065 ). These are large effects. However, as shown in the other rows of Panels A and C, the own and parental characteristics of men vary across employment groups, and these characteristics themselves might have independent effects on parental co-residence. To control for these other observed differences between men, we move in Table 2 to a regression framework. Specifically, the first column in this table shows the ordinary least squares (OLS) linear probability model estimates of and in (1.1). This specification includes a full set of calendar-year indicators,, and a vector of control variables, X, that consist of the man s age, plus the parental-level characteristics shown in Panel B of Table 1, all of which are time-varying. 6 The controls include parental age, income, wealth; indicator variables for employment, homeownership, and being married; and three measures of parental health status. The first measure is the number of limits to the parent s Activities of Daily Living (ADLs). The HRS collects information on five activities bathing, eating, dressing, walking across a room, and getting in and out of bed each designed to measure various dimensions of an individual s ability to function in his or her residential space. For each of the five tasks, the HRS records a 1 if the respondent had difficulty with that task and a zero otherwise. The scores are summed for the five tasks, so that the ADL variable ranges from 0 (no difficulties with any of the tasks) to 5 (difficulties with all of the tasks). The second measure is a count of the number of medical conditions a doctor had ever told the parent that he or she had. The eight conditions were high blood pressure, diabetes, 6 We control for family attributes of the biological parents separately to allow for parent s no longer residing together. Unfortunately, the HRS did not gather information on the biological parents of the spouses of married adult children of HRS respondents. So, we only have parental information for the parents of young men, which we control for the attributes of the biological mother and father separately. 11
cancer, lung disease, heart disease, stroke, psychiatric problems, and arthritis. The index ranges from 0 (the absence of all eight conditions) to 8 (the presence of all eight conditions) where, as before, a larger index value indicates poorer health. The third measure is a count of the number of limits to Instrumental Activities of Daily Living (IADLs), which are the ability to use a telephone, meal preparation, shopping, handle the household finances, and take medication. The OLS parameter estimates in column 1 of Table 2 show the regression-adjusted differences in co-residence between the three employment categories. 7 For example, controlling or adjusting for differences in age and parental demographic, economic, and health characteristics across men, ˆ OLS 0.144. That is, on average moving from full-time to nonemployment raises the likelihood of co-residing with a parent by 14.4 percentage points. Similarly, ˆ OLS 0.153. That is, on average moving from full- to part-time employment raises the likelihood of co-residing with a parent by 10.3 percentage points. These are large effects and similar to the unadjusted differences, suggesting that much of the variation in parental coresidence of men is not due to observable differences in family background. One empirical challenge with these findings is that there may be unobserved heterogeneity in a child s taste for co-residency with parents. Re-writing (1.1), this can be represented by, Y U P δx. (1.2) ijt it it jt t i ijt If labor-market transitions are correlated with this unobserved heterogeneity, then OLS estimates of and will be biased and inconsistent. This could occur, for example, if individuals with high taste for leisure also have a high taste for co-residency. Then it might appear that those who become unemployed are more likely subsequently to co-reside with a parent, but this would not 7 The estimates of the parameters associated with the control variables are not shown, but available upon request. 12
necessarily be a causal effect. To account for this, column 2 of Table 2 presents fixed-effects linear probability model estimates of and. Similar to Wiemers (2012), this allows us, in principle, to obtain consistent estimates of the parameters. With this estimator, controlling for calendar-year effects,, the estimates of and are identified by differences across time in the employment trajectories of sons. Our central identifying assumption is that these are (conditionally) uncorrelated with the error term,, in (1.2). Now, ˆ FE 0.093. That is, on average moving from full-time employment to being not employed raises the likelihood of co-residing with a parent by 9.3 percentage points. With a standard error of 0.017, this effect is statistically different from zero at conventional significance levels. Similarly, ˆ FE 0.103. That is, on average moving from full-time employment to being part-time employed raises the likelihood of co-residing with a parent by 10.3 percentage points. With a standard error of 0.018, this effect is statistically different from zero at conventional significance levels. a. Extensions Columns 3-4 show OLS and fixed-effect estimates, respectively, for the subsample of men who were co-residing in 2004, the baseline calendar year of the sample. Similarly, Columns 5-6 show OLS and fixed-effect estimates, respectively, for the subsample of men who were living independently at baseline in 2004. Those who return after a spell of independent living to co-reside with parents have been termed boomerang kids. The estimates in column 6 indicate that on average moving from full-time employment to being not employed for this group raises the likelihood of co-residing with a parent by 3 percentage points, whereas moving from fulltime employment to being part-time employed raises the likelihood of co-residing with a parent by 3.3 percentage points. 13
Table 3 shows estimates for further sample splits. Many existing studies have documented a higher frequency of co-residence after employment loss for men from families with low lifetime socio-economic status (SES). Columns 1-8 show splits for a variety of proxies for parental SES: educational attainment, wealth, the number of functional limitations, and homeownership, respectively. 8 Interestingly, there are few differences by parental characteristics in the co-residence response to employment. Men whose parents were homeowners were more likely to co-reside after employment loss. Columns 9-10 show separate estimates by the son s age. The co-residence of younger men under 40 is much more responsive to employment. b. Robustness Checks The empirical results thus far show evidence that loss of full-time employment generates significant increases in co-residence for young men. A potential empirical challenge to these findings, however, is that there might be unobserved factors that vary across families and are correlated with child labor-market dynamics. To represent this, in (1.2) can be decomposed as, Y U P δx, (1.3) ijt it it jt i j t ijt where represents time-invariant unobserved heterogeneity at the family level. For example, across extended families, parents may differ in permanent income (Lee and Painter, 2012), which is typically difficult to measure, but might be correlated with child labor-market attachment. 9 Alternatively, across extended families, parents might differ in their provision of informal insurance via co-residence in a way that is correlated with child labor-market attachment. In the 8 For educational attainment and functional status, which are measured for both parents in the HRS, we use the higher of the two values to define the sample split. 9 Or in a lambda-constant dynamic model of private income transfers, the unobserved heterogeneity could represent the parents marginal utility of (full) income. 14
presence of such an unobserved family fixed effect, estimates of and, would be biased and inconsistent. To address this, we construct a sibling panel dataset on sons for HRS respondents with more than one son, as the HRS gathers child information for all children. With a matched sibling-pair-to-parent dataset, we track residential and employment status of brothers and their parents over time. Hence, we can control both for time-variant unobserved heterogeneity at the family level and at the individual level with a two-way fixed-effect estimator: within a son over time, and between sons. With this estimator, controlling for calendar-year effects,, the estimates of and are identified by differences across time in the employment trajectories between sons in the same family. Our central identifying assumption is that these are (conditionally) uncorrelated with the error term,, in (1.3). In principle, to ensure consistent estimates, the sibling differences in employment trajectories should due to differential shocks to labor demand between sons across time, and not to differential shocks to labor supply. Since most estimates in the literature indicate that prime-age male labor supply is quite inelastic, we restrict our sample to brothers, a group whose changes in employment is most likely due to shocks to labor demand across the recession. These results are presented in Table 4. In total, there are 3,382 HRS respondents who have at least 2 biological sons represented in the survey. Those families form the sample for this estimation. In all specifications in the table, we include the same vector of control variables for the son s age and the time-varying parental characteristics, X, that were used in Tables 2-3. Column 1 shows the estimates of and for this sample using the same individual fixed effects estimator as for the full sample in Table 2. The estimates indicate that on average moving from full-time employment to being not employed for this sample of brothers raises the 15
likelihood of co-residing with a parent by 4.2 percentage points, whereas moving from full-time employment to being part-time employed raises the likelihood of co-residing with a parent by 4.3 percentage points. Column 2 shows estimates with both individual and family fixed effects. To the third decimal place, the results are identical to those with just individual fixed effects. One potential concern about the identification in column 2 is that it depends on the exogeneity of differential shocks to labor demand between sons across time. These shocks could be correlated if brothers were exposed to the same local economic conditions (in the extreme, for example, laid off by the same employer). Column 3 provides a further robustness check by controlling for a sibling-pair locational fixed effect. Specifically, for each child, the HRS asks the respondent (i.e., the parent) whether the child lives within a 10-mile radius of the respondent. The sibling-pair locational effect takes on a value of one if both sons in the sibling pair live within 10 miles of the parent and zero otherwise. Since even within metropolitan areas, much of the recession s effect was localized (e.g., unemployment, foreclosures, etc.), this fixed effect accounts for whether the sons were subject to the same local economic conditions. As the results in column 3 indicate, the estimated impacts of employment loss are virtually unchanged. Finally, one time-varying factor that does not appear as a control variable in specifications (1.1)-(1.3) is the marital status of the son. As Lee and Painter (2013) and others have pointed out, marriage is an endogenous household formation outcome for young men that itself most likely is heavily influenced by employment status. To explore this, Table 5 presents individual fixed-effect estimates for the full sample of specifications isomorphic to those in (1.2), but where the dependent variable is an indicator for whether the son is married or not. In column 1, the estimates indicate that on average moving from full-time to non-employment decreases the likelihood of being married by 4.5 percentage points, whereas moving from full- to part-time 16
employed decreases the likelihood of being married by 3.4 percentage points. Columns 2-3 show splits by age. Men who were older had greater responsiveness of their marriage decisions to employment. 5. Results for Homeownership Next, we turn to impacts on homeownership. The HRS asks respondents (parents) about the homeownership status of their children only every other wave, yielding data on the homeownership of sons only for 2004, 2008, and 2012. Table 6 shows estimates of the impacts of non- and part-time employment on homeownership using the following model, isomorphic to (1.2), H U P δx, (1.4) ijt it it jt i t ijt where the dependent variable, H, is an indicator that is one if the son owned a home and zero otherwise. For the purposes of comparison, column 1 presents estimates for the specification (1.2) with parental co-residence as the outcome, but estimated just for the subsample of observations from 2004, 2008, and 2012. Those estimates indicate that on average moving from full-time to non-employment raises the likelihood of co-residing with a parent by 7 percentage points, whereas moving from full- to part-time employment raises the likelihood of co-residing with a parent by 6.1 percentage points. The second column in the table presents individual fixed-effect estimates for a specification isomorphic to (1.2), but where the dependent variable is an indicator for whether the son is a homeowner or not. The estimates indicate that on average moving from full-time to non-employment decreases the likelihood of being a homeowner by 2.9 percentage points; moving from full- to part-time employment decreases the likelihood of being a homeowner by 17
3.4 percentage points. 10 When the sample is split by age, almost of this reduction in homeownership occurs for men under 40 (columns 3-4). 11 For the purposes of comparison, column 5 shows estimates for what is often the sample in many studies of homeownership, which is men who live independently. As Lee and Painter (2013) among others pointed out, this is an endogenous sample, because independent living is itself responding to employment loss, as evidenced in the previous analysis in the current paper. Comparing columns 2 (full sample) and 5 (endogenous sample), the estimated impacts of employment loss do not differ significantly qualitatively nor statistically. 6. Impact of Employment on Mortgage Distress The recession was characterized by historically high levels of mortgage distress, brought on by the financial crisis and the real estate bust, as well as the consequent rise in unemployment and decline in incomes. We end our analysis by presenting in Table 7 estimates the impact of employment on mortgage distress using the following model, isomorphic to (1.2), M U P δx, (1.5) ijt it it jt i t ijt where the dependent variable, M, is an indicator of mortgage stress for the son. In 2008, 2010, and 2012, the HRS asked respondents (i.e., parents) whether they or any of their children had fallen behind in payments or were in foreclosure. Unfortunately, the HRS did not ask which child was affected. Therefore, in panel A of column 1 the dependent variable is an indicator that takes on a value of one if the son is in a family in which someone had fallen behind in payments, and zero otherwise. The parameter in (1.5), therefore, measures the impact of the son moving 10 These findings are robust to sibling-pair fixed effects as were done in Table 4 for co-residence, and sibling-part locational fixed effects. These results are shown in Appendix Table 1. 11 Appendix Table 2 shows homeownership results for a variety of other sample splits. 18
from full-time to non-employment on the likelihood of being in a family in which someone had fallen behind in payments. Estimates in column 1 are restricted to only those sons who were reported to previously be homeowners in earlier waves, and thus at risk for mortgage distress. In panel A, moving from full-time to non-employment increases the likelihood of falling behind in payments by 7.8 percentage points. The distress measure in panel B is an indicator for having experienced a foreclosure. Moving from full-time to non-employment increases the likelihood of foreclosure by 5 percentage points. As explained above, the HRS did not ask parents to separately identify which child was in mortgage distress, so the estimates in column 1 represent a weighted average of the impact of loss of full-time employment for the focal son plus possibly other sons and daughters who experienced distress. As a robustness check, we report in column 2 estimates where we alternatively restrict the sample to families in which only one son was a homeowner in 2004 to better identify the treated child. Those estimates are qualitatively similar and not statistically different from those in column 1. Overall, we conclude that full-time employment loss had a substantial effect on increasing mortgage stress for men during the recession. 7. Summary We present new estimates of the impact of the loss of full-time employment on coresidence between parents and adult children. We follow a large number of existing studies and focus on young men, who often experience the most difficulty in adjusting to unemployment. Our estimates are based on the experience of these men in the Great Recession. We use detailed data on sons of respondents in the Health and Retirement Study (HRS) to construct a panel 19
dataset of young men and their parents that tracks employment and residential statuses before and during the recession. Our fixed-effect estimates indicate that on average a young man moving from full-time employment to being not employed raises the likelihood of co-residing with a parent by 9.3 percentage points. Similarly, on average moving from full-time employment to being part-time employed raises the likelihood of co-residing with a parent by 10.3 percentage points. These effects are economically large, statistically significant, and suggest that the ability to move in with parents is an important way that young men adjust during economic downturns. Interestingly, these effects are quite uniform across socio-economic strata, likely reflecting the fact that the depth and length of the recession left comparatively few families unscathed. One methodological contribution of our work is that we assess the robustness of our main individual fixed-effect estimates with a set of two-way fixed effect estimates, based on both individual- and family-level effects. Specifically, in each wave, the HRS asks respondents, who are in their early 50 s and older, about the economic, locational, and demographic status of their children. From these data, we construct a panel of adult men and their brothers, who are the children of HRS respondents, for which we can track residential and employment status. This allows us to track sibling pairs across time, both before and during the recession, to account for time-invariant unobserved heterogeneity via individual-specific fixed effects, as well as family fixed effects. These estimates, therefore, are identified by the differential time pattern of employment losses across brothers within the same family. Using this new approach, we find very similar effects. Overall, our results are very robust. We temper these conclusions with a couple of caveats. First, we limited our analysis to young men, primarily because men traditionally have had more difficulty adjusting to economic 20
downturns, but also so that we could more readily compare our results to previous finds in the literature. Naturally, it would important to expand the analysis to women, especially given women s secular increase in labor force participation over the last three decades. Second, our sample only runs through 2012. This means that our findings should be thought of the shortterm effects of employment loss on housing and living arrangements. Once more recent data become available, we will be able to estimate longer-run impacts. This will be one of many avenues of future research. 21
References Altonji, J., Hayashi, F., Kotlikoff, L. 1997. Parental altruism and inter vivos transfers: Theory and evidence. Journal of Political Economy 105 (6), 1121-1166. Bianchi, S., Hotz, J., McGarry, K., Seltzer, J. 2006. Intergenerational ties: Alternative theories, empirical findings and trends, and remaining challenges. California Center for Population Research Working Paper No. 024-06. Billari, F., Liefbroer, A. 2007. Should I stay or should I go? The impact of age norms on leaving home. Demography, 44 (1), 181-198. Börsch-Supan, A. 1986. Household formation, housing prices, and public policy impacts. Journal of Public Economics 30, 145-164. Cox, D. 1987. Motives for private income transfers. Journal of Political Economy, 95 (3), 508-546. Dettling, L. Hsu, J. 2014. Returning to the nest: Debt and parental co-residence among young adults. Federal Reserve Board Finance and Economics Discussion Series Paper No. 2014-80. Dynarski, S., Gruber, J. 1997. Can families smooth variable earnings? Brookings Papers on Economic Activity 1, 229-303. Ermisch J. 1999. Prices, parents, and young people s household formation. Journal of Urban Economics 45, 47-71. Goldscheider, F., DaVanzo, J. 1985. Living arrangements and the transition to adulthood. Demography 22 94), 545-563. Haurin, D., Hendershott, P., Kim, D. 1993. The impact of real rents and wages on household formation. Review of Economics and Statistics 75, 284-293. Haurin, D., Rosenthal, S. 2008. The influence of household formation on homeownership rates across time and space. Real Estate Economics 35 (4), 411-450. Kahn, J., Goldscheider, F. 2013. Growing parental economic power in parent-adult child households: Co-residence and financial dependency in the United States, 1960-2010. Demography 50, 1449-1475. Kaplan, G. 2009. Boomerang kids: Labor market dynamics and moving back home. Federal Reserve Bank of Minneapolis Research Department Working Paper No. 675. Kaplan, G. 2012. Moving back home: Insurance against labor market risk. Journal of Political Economy 120 (3), 446-512. 22
Lee, K., Painter, G. 2013. What happens to household formation in a recession? Journal of Urban Economics 76, 93-109. Mykyta, L. 2012. Economic downturns and the failure to launch: The living arrangements of young adults in the U.S., 1995-2011. U.S. Census Bureau SEHSD Working Paper No. 2012-24. Mykyta, L. Macartney, S. 2011. The effects of recession on household composition: Doubling up. U.S. Census Bureau SEHSD Working Paper No. 2011-4. Mykyta, L. Macartney, S. 2012. Sharing a household: Household composition and economic well-being, 2007-2010. Current Population Report P60-242. Washington, DC: U.S. Census Bureau. Rogers, W., Winkler, A. 2014. How did the housing and labor market crises affect young adults living arrangements? IZA Discussion Paper No. 8586. Rosenzweig, M., Wolpin, K. 1993. Intergenerational support and the life-cycle incomes of young men and their parents: Human capital investments, coresidence, and intergenerational financial transfers. Journal of Labor Economics 11 (1), 84-112. Wiemers, E. 2011. The effect of unemployment on household composition and doubling up. National Poverty Center Working Paper No. 11-12. Wiemers, E. Bianchi, S. 2014. Sandwiched between aging parents and boomerang kids in two cohorts of American women. University of Massachusetts Boston Department of Economics working Paper no. 2014-06. Yelowitz, A. 2007. Young adults leaving the nest: The role of the cost of living. In Danziger, S., Rouse, C. (Eds.) The Price of Independence: The Economics of Early Adulthood. Russell Sage Press: New York, pp. 170-206. 23
Table 1. Sample Means for Variables Used in Analysis (1) (2) (3) (4) Subsample: Working Working Full-Time Part-Time Full Sample Not Working Characteristics A. Individual Characteristics Age 39.9 39.9 38.6 40.2 Working full-time 0.824 1 0 0 Working part-time 0.054 0 1 0 Not working 0.123 0 0 1 B. Outcomes Co-resides with a parent 0.092 0.065 0.229 0.212 Married 0.647 0.701 0.405 0.391 Homeowner 0.350 0.401 0.123 0.121 Behind on mortgage payments 0.097 0.090 0.143 0.124 Home foreclosure 0.037 0.033 0.055 0.054 C. Parental Characteristics Father is working 0.420 0.427 0.452 0.346 Mother is working 0.346 0.354 0.366 0.279 Father is homeowner 0.850 0.860 0.829 0.774 Mother is homeowner 0.780 0.800 0.727 0.661 Father s age 67.594 67.634 66.564 67.771 Mother s age 65.839 65.864 64.715 66.141 Father s ADL limits 0.277 0.348 0.348 0.439 Mother s ADL limits 0.344 0.402 0.402 0.614 Father s IADL limits 0.129 0.117 0.160 0.218 Mother s IADL limits 0.114 0.098 0.138 0.208 Number of father s medical conditions 2.089 2.072 1.991 2.276 Number of mother s medical conditions 2.156 2.090 2.160 2.596 Father's family income 78,750 82,188 69,494 53,892 Mother's family income 58,259 61,729 51,193 37,876 Father's family wealth 545,764 578,378 429,353 324,339 Mother's family wealth 415,177 454,112 306,934 199,852 Number of person-year observations 45,958 37,879 2,466 5,637 Note: Authors calculations based on HRS sample described in text. 24
Table 2. OLS and Fixed-Effect Estimates of the Impact of Employment Status on Parental Co-Residence for All Sons, Robust Standard Errors in Parentheses (1) (2) (3) (4) (5) (6) All Sons Sample and Estimator: Sons Co-Residing at Baseline in 2004 Sons Living Independently at Baseline in 2004 Explanatory Variable OLS Individual Fixed Effects OLS Individual Fixed Effects OLS Individual Fixed Effects Not employed 0.144 0.093 0.294 0.095 0.069 0.030 (0.008) (0.017) (0.020) (0.046) (0.006) (0.006) Part-time employed 0.153 0.103 0.261 0.093 0.075 0.033 (0.010) (0.018) (0.023) (0.017) (0.009) (0.007) Number of men 11,898 11,898 2,535 2,535 9,488 9,488 Note: The dependent variable for all regressions is an indicator that takes on a value of one if the son coresided with the parent and zero otherwise. Each column in the table shows the parameter estimates of beta and phi in equation (1) in the text from a separate regression, with the associated sample and estimator shown in the column heading. There are three employment statuses: full-time, part-time, and not employed. Therefore, the excluded group is comprised of those who are full-time employed. All regressions included a set of controls for the calendar year and the variables for the parent: age, number of ADLs, number of IADLs, number of medical conditions, income, wealth, and indicators for homeownership, employment, and marital status. All standard errors are heteroscedasticity-robust and clustered at the individual-level. 25
Table 3. Individual Fixed-Effect Estimates of the Impact of Employment Status on Parental Co-Residence for All Sons by Selected Subgroups, Robust Standard Errors in Parentheses (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Sample: All Sons, by Highest Education Level Achieved by Parents Parental Wealth Parental Limitations to ADL or IADL Parental Homeownership Son s Age Explanatory Variable High School or Less More than High School Below Median Above Median At Least 1 None Own Rent 23-39 Years Old 40-52 Years old Not employed 0.047 0.055 0.045 0.058 0.057 0.045 0.056 0.027 0.070 0.023 (0.008) (0.011) (0.008) (0.011) (0.013) (0.008) (0.008) (0.011) (0.011) (0.008) Part-time employed 0.048 0.060 0.048 0.051 0.046 0.054 0.056 0.047 0.079 0.025 (0.010) (0.012) (0.010) (0.012) (0.017) (0.009) (0.009) (0.013) (0.012) (0.009) Number of men 7,420 5,393 7,532 6,904 5,186 10,654 9,682 3,716 7,124 8,013 Note: The dependent variable for all regressions is an indicator that takes on a value of one if the son co-resided with the parent and zero otherwise. Each column in the table shows the parameter estimates of beta and phi in equation (1) in the text from a separate regression, with the associated sample and estimator shown in the column heading. There are three employment statuses: full-time, part-time, and not employed. Therefore, the excluded group is comprised of those who are full-time employed. All regressions included a set of controls for the calendar year and the variables for the parent: age, number of ADLs, number of IADLs, number of medical conditions, income, wealth, and indicators for homeownership, employment, and marital status. All standard errors are heteroscedasticity-robust and clustered at the individual-level. 26
Table 4. Individual and Family Fixed-Effect Estimates of the Impact of Employment Status on Parental Co-Residence for All Sons with Brothers, Robust Standard Errors in Parentheses (1) (2) (3) (4) (5) (6) Sample and Estimator: Explanatory Variable Individual Fixed Effects All Sons Sons Living Independently at Baseline in 2004 Individual, Family, and Individual and Sibling-Location Individual Family Fixed Fixed Effects Fixed Effects Effects Individual and Family Fixed Effects Individual, Family, and Sibling-Location Fixed Effects Not employed 0.042 0.042 0.043 0.026 0.026 0.026 (0.008) (0.009) (0.030) (0.007) (0.009) (0.009) Part-time employed 0.043 0.043 0.043 0.027 0.027 0.027 (0.009) (0.030) (0.010) (0.009) (0.010) (0.010) Number of brothers 8,033 8,033 8,033 6,453 6,453 6,453 Note: The dependent variable for all regressions is an indicator that takes on a value of one if the son co-resided with the parent and zero otherwise. Each column in the table shows the parameter estimates of beta and phi in equation (1) in the text from a separate regression, with the associated sample and estimator shown in the column heading. There are three employment statuses: full-time, part-time, and not employed. Therefore, the excluded group is comprised of those who are full-time employed. All regressions included a set of controls for the calendar year and the variables for each biological the parent: age, number of ADLs, number of IADLs, number of medical conditions, income, wealth, and indicators for homeownership, employment, and marital status. All standard errors are heteroscedasticity-robust and clustered at the individual-level in columns 1 and 4, and at the family level in columns 2-3 and 5-6. 27
Table 5. Individual Fixed-Effect Estimates of the Impact of Employment Status on Being Married for All Sons by Selected Subgroups, Robust Standard Errors in Parentheses (1) (2) (3) Sample Explanatory Variable All Sons Under Age 40 Age 40 and Older Not employed -0.045-0.033-0.095 (0.008) (0.029) (0.029) Part-time employed -0.034-0.026-0.042 (0.009) (0.027) (0.085) Number of men 11,898 3,428 10,554 Note: The dependent variable for all regressions is an indicator that takes on a value of one if the son was married and zero otherwise. Each column in the table shows the parameter estimates of beta and phi in equation (1) in the text from a separate regression, with the associated sample and estimator shown in the column heading. There are three employment statuses: full-time, part-time, and not employed. Therefore, the excluded group is comprised of those who are full-time employed. All regressions included a set of controls for the calendar year and the variables for each biological parent: age, number of ADLs, number of IADLs, number of medical conditions, income, wealth, and indicators for homeownership, employment, and marital status. All standard errors are heteroscedasticity-robust and clustered at the individual-level. 28