The consumption expenditure response to unemployment: Evidence from Norwegian households

Fagereng, Andreas; Onshuus, Helene; Torstensen, Kjersti Næss Working Paper The consumption expenditure response to unemployment: Evidence from Norwegian households Working Paper, No. 6/2024 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Fagereng, Andreas; Onshuus, Helene; Torstensen, Kjersti Næss (2024) : The consumption expenditure response to unemployment: Evidence from Norwegian households, Working Paper, No. 6/2024, ISBN 978-82-8379-314-7, Norges Bank, Oslo, https://hdl.handle.net/11250/3172760 This Version is available at: https://hdl.handle.net/10419/310429 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ Working Paper The Consumption Expenditure Response to Unemployment: Evidence from Norwegian Households Norges Bank Research Authors: Andreas Fagereng Helene Onshuus Kjersti N. Torstensen Keywords: Unemployment, household finance, consumption expenditure, consumption smoothing, household heterogeneity 6 | 2024 Norges Bank Working Paper 1 Working papers fra Norges Bank, fra 1992/1 til 2009/2 kan bestilles på epost: [email protected] Fra 1999 og senere er publikasjonene tilgjengelige på www.norges-bank.no Working papers inneholder forskningsarbeider og utredninger som vanligvis ikke har fått sin endelige form. Hensikten er blant annet at forfatteren kan motta kommentarer fra kolleger og andre interesserte. Synspunkter og konklusjoner i arbeidene står for forfatternes regning. Working papers from Norges Bank, from 1992/1 to 2009/2 can be ordered by e-mail: [email protected] Working papers from 1999 onwards are available on www.norges-bank.no Norges Bank’s working papers present research projects and reports (not usually in their final form) and are intended inter alia to enable the author to benefit from the comments of colleagues and other interested parties. Views and conclusions expressed in working papers are the responsibility of the authors alone. ISSN 1502-8143 (online) ISBN 978-82-8379-314-7 (online) The Consumption Expenditure Response to Unemployment: Evidence from Norwegian Households* Andreas Fagereng, Helene Onshuus, and Kjersti N. Torstensen October 2023 Abstract We use detailed Norwegian administrative data to identify the income loss associated with the onset of unemployment and analyze the corresponding consumption expenditure response and the extent to which this response is related to household balance sheet components. Unemployment results in a significant, long-term decline in income. Consumption decreases by about one-third to one-half of the post-tax income reduction. This reduction is less pronounced for liquid households and more for indebted ones. Although both debt and liquidity impact consumption patterns, debt has a predominant influence, especially for households holding substantial amounts of both. These households, despite their liquidity, also reduce their consumption upon unemployment, while consistently dedicating a substantial part of their disposable income to mortgage commitments. Furthermore, we investigate heterogeneity along other important margins such as family composition and child age. Finally, the patterns of our spending responses (measured as the marginal propensity to consume, the MPC) are found to be more pronounced during recessions. JEL: D12, E21, E24 Keywords: unemployment, household finance, consumption expenditure, consumption smoothing, household heterogeneity *This paper should not be reported as representing the views of Norges Bank or the Norwegian Ministry of Finance. The views expressed are those of the authors and do not necessarily reflect those of Norges Bank or the Norwegian Ministry of Finance. Significant parts of this research project were carried out while Torstensen was working at Norges Bank. We are grateful to Kasper Roszbach for his help in facilitating the project and his continued support, and to Mette Ejrnæs, Moritz Kuhn, Gisle Natvik, and Kjetil Storesletten for their detailed feedback, guidance, and discussions. Fagereng acknowledges support from the ERC under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 851891). Fagereng: BI Norwegian Business School and Norges Bank, email: andreas.fager[email protected]. Onshuus: BI Norwegian Business School and Norges Bank, email: [email protected]. Torstensen: Norwegian Ministry of Finance, email: [email protected]. 1 1 Introduction Job displacement and subsequent unemployment represent a significant financial disruption for a↵ected households. Beyond the immediate loss of income, job displacement introduces uncertainty regarding both the length of time without income and the outlook for future wages. In light of this, households must decide how to adapt, ultimately choosing how much to adjust consumption. The response in consumption may depend on both the ability and willingness of the household to tap into their own or family savings, adjust household labor supply, or accrue debt. Understanding such household consumption decisions has long been a focus of economic research (Hall and Mishkin,1982;Dynarski and She↵rin,1987), and carries implications for a broad range of questions within economics.1While household liquid wealth has traditionally been associated with the magnitude of consumption responses (Deaton,1991;Carroll,1997;Kaplan, Violante and Weidner,2014), the Great Recession of 2007-2009 reemphasized the importance of balance sheet components: Elevated household leverage positions prior to the crisis contributed to depressing growth in household consumption in its aftermath, thereby prolonging the recovery (Dynan, Mian and Pence,2012;Eggertsson and Krugman,2012;Mian, Rao and Sufi,2013). Despite considerable progress especially among theoretical e↵orts to understand the importance of household heterogeneity in this regard (Krueger, Mitman and Perri,2016), our knowledge is often less developed about the micro-level facts stemming from the behavior of real-world households, often due to limitations of available data. Recent notable exceptions in the realm of consumption and unemployment (insurance) include work by Ganong and Noel (2019), Landais and Spinnewijn (2021), and Andersen, Jensen, Johannesen, Kreiner, Leth-Petersen and Sheridan (2023). Still, the empirical evidence to date on the relationship between heterogeneity in household balance sheet components and the consumption responses to income shocks remains incomplete. This article enhances current research by providing novel empirical evidence into household consumption and saving responses to unemployment, with a particular focus on heterogeneity in responses along the distribution of households’ prior balance sheet positions. We do so by 1These topics include research on the evaluation and design of optimal unemployment insurance (UI), the role of UI as an automatic stabilizer, the development of general equilibrium models more broadly, and macroprudential policies. An extensive literature review is provided at the end of the Introduction. 2 utilizing a unique data set sourced from Norwegian administrative records, covering over two decades of detailed annual income and balance sheet information for all Norwegian households. Originally gathered for wealthand income tax assessments, the data is highly reliable as financial institutions directly report each household´s asset and liability positions to the tax authorities. We can therefore, without concerns for measurement error, utilize the two key components of households’ balance sheets, liquid wealth and debt, in our analyses. Coupled with spell data from UI benefit applications this allows us to trace household balance sheet positions before, during, and after an unemployment spell. The data further contain employer-employee records, providing information about salaries, employment history and employer characteristics, demographic characteristics, information about education, and a registry of family relations. The richness and preciseness of the data allow us to construct a comprehensive measure of household consumption expenditure via the household budget constraint. We employ two methods to quantify the post-unemployment consumption response: First, an event study design is utilized, deploying a meticulously chosen control group and addressing selection and endogeneity concerns tied to unemployment through the incorporation of a robust set of control variables, which encapsulate employment characteristics and balance sheet components. Second, we utilize the control group to estimate the marginal propensity to consume (MPC) within the job-loss year, which not only enables a flexible interaction with covariates but also facilitates a nuanced comparison to the extant literature on MPC heterogeneity. We find that unemployment leads to a pronounced, persistent income drop, with earnings decreasing 20-30%, and only starting to recover two years from the initial job loss. Post-tax labor income decreases 10-15% instantly, maintaining a level below the control group for the following four years. Noteworthy is that households characterized by lower initial liquid assets or higher debt typically encounter milder long-term declines in both earnings and after-tax income, despite experiencing an initial earnings drop that is uniformly distributed across all ranges of debt and liquid assets. The income decline is accompanied by a notable decrease in consumption expenditures, amounting to between one-third and one-half of the after-tax income drop. The expenditure decrease is less drastic among households with higher liquidity, while more indebted households see a more severe drop. A considerable fraction of high-liquidity households also 3 bears substantial debt. Notably, even with substantial holdings of liquid assets at hand, these high-debt households markedly reduce their consumption upon unemployment. Within the event study framework, we also undertake heterogeneity analyses along other important household margins, such as family composition. While families with children tend to recover in terms of income from unemployment somewhat faster, we notice a particular divergence in response among households with younger children (below 5 years). These households’ consumption bounce back faster than for families with older children, which complements previous findings on investment in children in a partial insurance framework. When measuring the household response as an MPC out of unemployment, we find that, on average, a one-dollar income loss leads to a spending decline of about 40 cents. We observe significant MPC heterogeneity across both debt and liquid assets. Further, we uncover a U-shaped relationship between debt-to-income (DTI) and MPC—middle DTI tertile households have lower MPCs than those in both low and high DTI tertiles. Finally, we find that the U-shaped debt and MPC relationship persists across the liquid asset distribution. Lastly, we investigate how the household responses di↵er over the business cycle. On average, we find that income drops appear somewhat less severe during recessions. For the MPC out of unemployment, we find a modest increase during recessions, and results furthermore point in the direction of the U-shaped relationship we observe between the DTI ratio and the MPC as being slightly more pronounced during recessions. Related Literature The findings and analyses in this paper are related to several strands of the literature at the intersections of macroeconomics, labor, and public finance. Job displacement of high-tenured workers Our investigation leans on a well-established body of research on job displacement and its persistent e↵ects on income. Jacobson, LaLonde and Sullivan (1993) conducted seminal work in this area finding that in the United States, individuals who experienced job displacement faced average annual income losses of 25 percent over the long run. Numerous studies have followed since, 4 documenting the impact of job loss on workers earnings across time and space.2Our findings in this regard are in line with this evidence, especially from similar northern European countries as documented in Bertheau et al. (2023). A notable aspect of this literature is the frequent focus on samples consisting primarily of high-tenure workers, embedded in stable job relationships prior to job loss. The requirement is crucial in mitigating concerns regarding unobserved heterogeneity and selection into unemployment, ensuring that unemployment outcomes are compared to a prior, or counter-factual, state of employment (also discussed in Jarosch,2023). However, it also implies that the sample mainly consists of workers for whom a job loss represents a more persistent negative income shock, which is important to keep in mind when interpreting findings.3As detailed in Section 3, our approach aligns with the previous literature in imposing a job-stability requirement. This is done not only for the aforementioned reasons but also to examine a distinctly defined unemployment event, thereby increasing the probability that the observed consumption responses are attributed to a specific unemployment occurrence as opposed to one in a series.4 Consumption responses to job loss Our main analyses are related to a vast literature of the consumption responses to income changes (Hall and Mishkin,1982;Dynarski and She↵rin,1987;Blundell, Pistaferri and Preston,2008).5 An adjacent, and similarly relevant line of research has been concerned with the consumption response to job loss and the welfare e↵ects that arise when the income shock is (partly) o↵set by unemployment insurance (UI). Seminal research in this area includes Gruber (1997), who found a direct link between changes in food consumption during unemployment (using PSID data) and the level of UI benefit generosity. Browning and Crossley (2001), exploring the Canadian 2See for instance Couch and Placzek (2010), Davis, Von Wachter et al. (2011), Huttunen, Møen and Salvanes (2011), Kawano and LaLumia (2017), Krolikowski (2017), Flaaen, Shapiro and Sorkin (2019),Bertheau, Acabbi, Barceló, Gulyas, Lombardi and Saggio (2023), Lachowska, Mas and Woodbury (2020), Jarosch (2023), Schmieder, Von Wachter and Heining (2023). 3Indeed, Lachowska et al. (2020) find that loss of valuable worker-employer matches may explain half the wage loss for displaced workers. 4In Appendix Fwe also display the results for income paths when we reduce the strictness of this requirement and the findings are similar. 5For comprehensive literature reviews, refer to Browning and Lusardi (1996) and Jappelli and Pistaferri (2010). 5 context, found a generally small but heterogeneously large e↵ect of UI benefits on consumption, notably providing stronger smoothing for households with fewer liquid assets at the onset of unemployment. In recent years, this line of inquiry at the intersection of macro and public finance has spurred the works perhaps most closely related to our investigation:6 Ganong and Noel (2019) examine the impact of unemployment on consumer spending using de-identified bank data, and find that spending of the unemployed is highly responsive to the level of UI benefits, and drops sharply at both the onset of unemployment and at benefit exhaustion. While they do not observe liquid assets or liabilities directly, they apply an estimate of these positions and relate them to spending drops. In contrast, our work directly observes the household’s complete balance sheet position, thereby enabling a more detailed examination of and focus on, for instance, their interactions. Gerard and Naritomi (2021) investigate the degree of consumption smoothing among Brazilian households who receive a sizable severance pay, and find excess sensitivity with regards to the lump-sum transfer at unemployment onset. They do not explore the role of heterogeneity in initial balance-sheet positions. Landais and Spinnewijn (2021) explore di↵erent approaches to estimating the value of unemployment insurance using data from Sweden, in a setting comparable to ours. They propose two alternative approaches to infer the value of UI in addition to the traditional way of observing consumption responses to job loss, which involves considering the di↵erence in the MPC between the states of unemployed and employed. The second alternative approach (revealed preference) utilizes a kink in the UI system to gauge the value of insurance. In some of the supplementary analyses, they do detect a larger spending drop among households with more leverage (but the di↵erence is minor). The main aim of their article is to improve the understanding of the average valuation of UI, and while they undertake some analyses regarding heterogeneity, they largely leave the door open to further studies in this domain. Andersen et al. (2023) set out to quantify various mechanisms of consumption smoothing undertaken by unemployed households, and show that drawing on liquid assets is the most important way in which households reduce the impact of the income loss on spending, but undertake only limited heterogeneity analyses as to understand the magnitude 6See also some of the further literature on evaluating costs and benefits of social insurance and the design of optimal welfare policies (Baily,1978;Chetty,2006) studying the value of unemployment insurance (Engen and Gruber,2001;Chetty,2008;Hendren,2017;Kolsrud, Landais, Nilsson and Spinnewijn,2018). 6 3 Main empirical framework 3.1 A control group for the unemployed The ideal experiment for investigating consumption responses after job loss would involve random, unexpected employee terminations. Since such settings are impractical, one alternative is utilizing natural experiments where unemployment is quasi-randomly assigned. By conditioning on appropriate covariates, unemployment can be treated as if random. The literature, which often utilizes plant closures or mass layo↵s to assess unemployment e↵ects, bases its methodology on this notion, managing potential endogeneity issues up to a certain extent (Schwerdt,2011). This approach minimizes selection issues in unemployment, yet it does not necessarily account for potential anticipatory (consumption) adjustments by workers foreseeing such layo↵events, leading to possible bias in income and consumption correlations directly related to unemployment events. Moreover, utilizing mass-layo↵scenarios often leads to small sample sizes, making detailed analysis, such as relating portfolio composition to consumption responses, challenging. Rather than adopting a mass-layo↵approach, our strategy involves building a counterfactual through a control group method. Every worker who becomes unemployed is paired with a group of workers who, while similar in observables (utilizing extensive data to control for numerous variables influencing income growth and consumption), do not experience job loss during that period.13 However, a latent concern persists regarding potential unobserved disparities, such as variations in ability or diligence, between displaced workers and their employed counterparts. To navigate this potential influence of unobserved variables on a↵ecting income growth and consumption, we scrutinize the pre-trend of key variables. This aids in refining strategies to identify a control group that authentically mirrors the pre-trend of critical household income and balance-sheet components.14 (470,000) NOK or about 87,000 (76,000) USD. See https://www.ssb.no/arbeid-og-lonn/faktaside/ arslonn. 13To sidestep the potential pitfall of conditioning on post-treatment outcomes during the selection of treatment and control groups, as cautioned by Krolikowski (2018), we do not preclude workers in the control group from encountering unemployment in subsequent years. 14A related discussion and strategy for control group selection is found in Borusyak, Jaravel and Spiess (2017). See also Flaaen et al. (2019) for an alternative control group construction methodology. 13 3.2 High-dimensional near-neighbour matching To identify the suitable control group we return to the detailed administrative data sources. The starting point comes from imposing the same sample selection criteria as in Section 2on the full population of households, the only di↵erence being that these households do not become unemployed in that given year. This set of eligible households is termed "Possible controls". The matching procedure incorporates factors like age, education, and job tenure to comprehensively address labor market risks. To genuinely capture the likelihood of re-employment post-job loss, we select control group households that inhabit a labor market region similarly sized to their unemployed counterparts. Financial comparability between control and treatment groups is crucial, particularly when we want to study the spending response out of unemployment. Households are therefore selected for the control group based on similarity in liquid asset and debt levels on December 31st, two years before the job separation year, ensuring both groups are financially analogous. The households are also matched based on ownership of risky assets and housing, and income, measured at specific intervals before unemployment, sidestepping potential bias from strategic asset accumulation preceding unemployment episodes. Employing both exact matching for discrete variables like education and home ownership, and interval-based matching (±↵) for continuous variables such as age and income, our methodology enables each control to be matched to multiple unemployed households and vice versa (n-to-n matching).15 3.3 Comparing samples on observables and pre-trends From Section 2.3 we retrieve the sample of 11,497 households, which is our treatment group. In total our matching procedure selects 147,027 households from the set of Possible controls to be matched to these households. Table 1allows a comparison of key characteristics among samples of displaced workers, the control group, and the set of possible controls from which the control group is chosen. The first panel of the table presents variables involved in the matching procedure. The set of possible controls is, on average (and at the median), slightly older, with higher income, and 15This also implies that all of our regressions and statistics presented below are weighted, using CEMweights, following Iacus, King and Porro 2012). Details on how the weights are constructed, as well as the full breadth of details on the matching procedure and ↵-values for each variable chosen are available in Appendix E 14 with di↵erent compositions when it comes to industry of occupation and education length. The balance sheet components of debt and liquid assets are more dispersed, both in terms of averages and median. The table shows that the matching procedure yields a sample closely resembling the sample of the unemployed in terms of observables, in most bases bringing both the average and median closer. We also observe that the samples are aligned with respect to education and industry of employment, particularly within the sectors of manufacturing and construction, and retail and services where the discrepancies were most noticeable. Although the di↵erences in labor income, debt, and financial assets means are still statistically significant between the unemployed sample and the chosen control group (see Appendix Table A3), they are now economically negligible.16 To further mitigate concerns regarding other unobserved di↵erences between the samples that could confound our results, all our plots that follow later will include periods prior to unemployment. In addition, we here study the pre-trends of other key characteristics in the years leading up to unemployment, specifically. We use a simple event study specification: Yi,t= 1 X k=4 kUk i,t⇥Ti+"i,t,(1) where idenotes household, calendar year is denoted by t, and time relative to the onset of unemployment is denoted by k. The dummy variable Tiindicates the individual belongs to the treatment group and does not change over time (i.e. the variable does not denote "treatment status"). Year relative to the onset of unemployment is indicated by Uk i,twhich takes the value one when kperiods have passed since the year of unemployment, and zero otherwise. The error term is "i,t. We run this regression on the sample of matched treatment and control households. Figure 1plots the development of income after tax, deposits, debt, the share owning risky assets, and housing in the pre-unemployment years. There are only minor observable di↵erences between the treatment group and the control group. We see that there are virtually no visible 16In Appendix Dwe develop a statistical measure for job loss risk, based on observables such as tenure, firm age, education, and sector. This prediction is available for all households, and we include it in Table 1 "Estimated probability of job loss" to further assess the sample selection procedure. As we would expect, also this likelihood is aligned through the matching procedure. Appendix Dalso compares the distribution of probabilities between the di↵erent samples. 15 di↵erences in the development of male labor income, spousal labor income, debt, and deposits. Table A4 in Appendix Fshows that although there are some statistically significant di↵erences in the development year-by-year, the numbers are vanishingly small and economically insignificant. Overall, the similarity of the unemployed and their matched group makes us confident in interpreting the chosen control group as a good proxy for the true counterfactual and we proceed using standard econometric techniques. 4 Event study of job loss In this section, we estimate the dynamic path of income, wealth, and expenditure in a four-year period after job loss. The regression specification is a simple di↵erence in di↵erence with staggered implementation: Yi,j,t=↵j+X k2{4:4} kUk i,t+X k2{4,...,2,0,...,4} kUk i,t⇥Tk i+"i,t,(2) where Yi,j,tdenotes outcome variable in year tfor household ibelonging to matching group j. The matching group includes one household in the treatment group and their chosen set of control households. The outcome variable is regressed on a matching group fixed e↵ect, ↵j, a set of dummy variables Uk i,tindicating year relative to registering as unemployed, with k2{4, ..., 0, ..., 4} denoting years passed since the onset of unemployment, a set of binary variables Tk iindicating that the household is in the treatment group, and "i,tis the error term which is assumed to be i.i.dnormally distributed. All standard errors are clustered at the matching groups, and the equation is estimated using weighted OLS.17 Figure 2displays the results for key variables. The left column depicts estimated relative time dummies for treated and control groups, while the right shows the average treatment e↵ect in 2014-USD and percentage deviation from pre-job loss averages.18 Earnings drop post-job loss, averaging a decrease of over 15,000 USD upon unemployment registration and an additional 5,000 17See Appendix Efor more details on the weights. 18The pre-job loss average is household-specific and is an average of the two years prior to registering as unemployed. 16 USD the following year. This second-year decline may stem from extended unemployment into the subsequent year, given the typical 24-month duration of UIB. With the income tax scheme’s progressivity, the after-tax labor income (including unemployment benefits) impact is mitigated, totaling just over a 5,000 USD drop, or 13 % of pre-job loss income, in the registration year. Figure 2reveals that after-tax labor income does not recover four years post-unemployment onset, remaining 10 % lower than the control group, mirroring findings from Jacobson et al. (1993) and similar studies. We identify a small statistically significant, but negligible in terms of economic significance, increase in spousal wage income in the year of job loss (third panel of Figure 2). This is in line with the findings by Hardoy and Schøne (2014), Andersen et al. (2023), and Halla, Schmieder and Weber (2020). Figures A4 and A5 in the appendix show how other household income measures and income sources develop over the event window. Although household income falls significantly both in the year of job loss and the year after, the last panel of Figure 2shows that the bulk of the fall in consumption expenditure happens in the first year of unemployment. Expenditure declines somewhat further in the subsequent year, but the bulk of the adjustment happens on impact, which is consistent with standard economic theory when unemployment is a mostly unforeseen and permanent shock to income. The average drop in consumption expenditure seems to be persistent, echoing the permanent drop in income. 4.1 The importance of liquidity and debt This section examines how income and spending responses to unemployment are associated with key household balance sheet components, centering on debt-to-income (DTI) and liquid-assetsto-income (LTI) ratios, indicative of financial constraints. The sample is divided into tertiles using DTI and LTI, derived from a two-year pre-unemployment household average, assessing respective income and consumption responses.19 Ultimately, households are grouped into four distinct DTI and LTI categories for further analysis. 19Given that the control group is selected (among other variables) along debt, liquid assets, and income margins, it inherently exhibits similar distributions in DTI and LTI as the unemployed sample. Various splits of these distributions have been tested, including division into quartiles, yielding consistent results. Refer to Appendix Ffor details. 17 4.1.1 The separate importance of liquidity and debt Panel (a) of Figure 3shows findings from the LTI split, revealing a stable initial income drop across groups but a slightly stronger recovery for the low-liquidity group 3-4 years post-displacement. After-tax income trajectories underscore disparities, possibly reflecting tax system progressivity. The most liquid group encounters a minor consumption decline, while also seeing the most severe income drop. Panel (b) also shows responses stratified by the debt-to-income distribution, with the more indebted households seeing both a faster recovery of their income, but also experience a larger and more lasting drop in consumption. Taken together these results align well with evidence provided so far. In terms of liquidity, lowliquidity households, naturally curtail consumption more sharply in the face of income reductions (Carroll,1997), while the most liquid group, with assets of on average 77,600 USD (see Table 2), largely maintains consumption despite similar income drops. Analysis along the DTI split reveals that the most indebted also notably decreased consumption. Table 2shows that this group of households on average spend almost one-third of their income on mortgage commitments prior to job loss, emphasizing these commitments’ key role in household budget constraints. 4.1.2 The joint importance of liquidity and debt We move on to consider the consumption responses in the interaction of the distributions of the two balance sheet components. The bottom tertile of the LTI distribution, deemed closest to liquidity constraints, is defined as "low LTI", while the middle and top tertiles are grouped as "high LTI". Analogously, households on the upper spectrum of DTI distribution are likely closer to credit constraints, categorizing the top tertile as "high DTI" and the bottom two as "low DTI". The two first figures of Panel (c) reveal that high-LTI households generally experience a more gradual income recovery than their low-LTI counterparts, with the high-LTI low-DTI category revealing the most negative trajectory. Intriguingly, a reversal is noticeable in consumption response patterns. Despite experiencing a similar (or even slightly less negative) income decline, the high-LTI high-DTI group curtails consumption much more than the high-LTI low-DTI household. Again Table 2illuminates the importance of mortgage commitments for the high-DTI groups. Still, it might be somewhat surprising that the high-LTI high-DTI group exhibits a relatively strong re18 duction in consumption. Potential explanations might derive from a precautionary savings motive if households now view the future as generally more uncertain, or perhaps from a reassessment of their permanent income. They might also conserve their liquid bu↵er for other investment opportunities or potentially for entrepreneurship. Exploring behavioral explanations, particularly those related to mental accounting biases in preserving home ownership, could also serve to explain the findings. Given the complex interaction between liquid assets and debt, the observed income and consumption responses pose fascinating questions about household financial behavior and its further implications. Although it is beyond our scope to decipher all the nuances, it is compelling to ponder some implications of the findings for models and policy. 4.1.3 Implications for models and policy The findings stress the need to integrate household balance sheet dynamics into macroeconomic models, highlighting gaps in standard Bewley and other heterogeneous agent models’ predictions, especially in varied liquidity and debt situations. Unlike these models, which simplify the relationship between liquidity, consumption, and income shocks, incorporating mechanisms, akin to those in Kaplan et al. (2014), that accommodate real-world multifaceted financial decision-making is crucial. The nexus between assets and debt in household behavior necessitates careful policy crafting. Formulating policies attuned and potent in diverse financial contexts is vital. Our findings can inform policy design, such as Unemployment Insurance (UI) or post-crisis stimulus packages to stimulate aggregate demand, though application necessitates prudence to circumvent moral hazard, particularly when linked to financial ratios like DTI or LTI. Rather than being applied o↵ the shelf, our findings should be input as one component of a wider policy development approach. Another economic policy domain pertinent to our findings is macroprudential regulation. At first glance, our results suggest that DTI caps could soften consumption downturns for the unemployed. While such policies could temper the impacts identified in our study, it’s crucial to thoroughly examine their broader implications, if implemented. For instance, a potential adverse e↵ect would be if it led households to empty their liquid bu↵ers to overcome a strict DTI cap 19 when purchasing a home. Therefore, understanding both the immediate and the indirect impacts is essential when constraining households’ financial leeway. 4.2 Heterogeneity along other margins Navigating through further dimensions of heterogeneity, we explore the intersection of demographic variables and economic responses, specifically focusing on household compositions relative to child presence and age, as visualized in Figure 4. The upper panels of this figure categorize households into subsets—those with and without children and further into varying child age groups. One observation surfaces: households without children experience a notably sharper income decline and a subsequently sluggish recovery compared to those with children. Arguably however it is hard to draw comparisons between these groups as households with and without children may be vastly di↵erent. Particularly, families with children under 17 years exhibit a rapid income recuperation, hinting towards a potential correlation with either the higher employment quality or an elevated income necessity driven by parental responsibilities. A notable divergence in consumption response is evident among households with younger children (below 5 years of age), showcasing a robust resurgence in consumption, indicative of a parental perspective that prioritizes essential consumption during these initial formative years. This pattern echoes Carneiro, García, Salvanes and Tominey (2021), emphasizing the pivotal role of parental income in the early childhood phase (0-5 years) relative to later years (6-17).20 We also look into the di↵erence between couples that are married vs cohabiting or by age (Figure A7), but find no significant di↵erences in terms of the consumption responses. 5 The MPC out of Unemployment Every estimation of the marginal propensity to consume presupposes a level of income and consumption expenditure absent the income shock. Considering a framework of unobserved 20Carneiro and Ginja (2016) find that when they allow the parents’ reaction to vary with the age of the child (in a partial insurance framework) the permanent income shocks have statistically significant e↵ects only on inputs of children between ages 0 and 7. 20 counterfactuals, define the outcome for individual (or household) iin year tas: yi,t=Ti,t·yi,t(Ti,t=1) +(1 Ti,t)·yi,t(Ti,t=0) (3) Where Tiis 1 if the individual is unemployed, the observed outcome is yi,t(Ti=1), and the unobserved counterfactual is yi,t(Ti=0). The genuine treatment e↵ect—i.e the job loss e↵ect on income or spending (not the marginal propensity to spend) - is, ⌧y i,t=yi,t(Ti=1) yi,t(Ti=0). To estimate the unobserved counterfactual, we employ the household-specific control groups from Section 3.1, estimating income and spending growth using the control group’s average growth rates, assuming that without job loss, a household would parallel the control group’s average income and spending growth. The income shock and the consumption response is constructed in the following way: IncomeShocki,t=INCT i,t(1 +˜ gINC i,t)⇤INCT i,t1,Ci,t=CT i,t(1 +˜ gC i,t)⇤CT i,t1(4) where INCT i,tis the observed after-tax labor income of the unemployed in household iin year t, and Ci,tis observed consumption expenditure. Further, gINC and gCare the estimated growth rates of income and consumption expenditure in the control groups: gINC i,t=PJi j=1 1 Ji INCC j,i,t INCC j,i,t1!, where INCC j,i,tis the income of the male in household jin year tin the control group of household i, and Jiis the number of controls chosen for a treated household i. The growth rate of consumption is constructed in the same manner. The fundamental (although untestable) assumption is that, had the treated household not encountered unemployment, their income and spending would have evolved akin to the control group’s average, a presumption substantiated by observed pre-trend similarities between treatment and control groups (refer to Section 3). 5.1 The average marginal propensity to consume To estimate the average MPC in the sample, we regress the consumption response f Cit on the income shock, a constant, and a set of controls. Our baseline estimation equation is f Ci=0+1IncomeShocki+X0 i+"i,(5) 21 where Xiis a vector of control variables including a fourth-order polynomial in age, a secondorder polynomial in the number of children below age 18, a dummy for whether the household is married or cohabiting, and a set of calendar year dummy variables. The first row of Table 3reports the MPC out of an additional dollar lost due to unemployment within the year of job loss. The table presents results from six distinct specifications of the estimation equation, progressively and variably augmenting the control variables set. Controlling for pre-job loss income level, in columns IV-VI, the coefficient shifts minimally across specifications. Column VI, our preferred specification, controls for pre-job loss debt, real assets, deposits, and risky assets.21 We observe that for every dollar lost in the unemployment onset year, households, on average, curtail their spending by 41 cents, broadly consistent with existing literature. 5.2 Heterogeneity in the MPC To scrutinize the heterogeneity in the MPC, we do two key explorations: (i) segmenting the sample into tertiles based on distinct balance sheet components and worker characteristics, then reestimating Equation 5, integrating an interaction between the income shock and tertile-dummies, and (ii) re-enacting (i) within each LTI-tertile, introducing controls for the tertiles of the DTI distribution, interacting with the income shock.22 The insights, presented in Figure 5, exhibit constrained variation in the MPC across male labor income (Panel 5a) and household income distributions (Panel 5b), and age (Panel 5c). Nevertheless, palpable heterogeneity in MPC comes to light upon examining net wealth, liquid assets, and debt. Figures 5e to 5h uncover a compelling, nearly linear distinction in MPCs among households with varying liquid assets and a U-shaped relationship in the distribution of debt-to-income. The non-monotonicity, though not dissected in detail, implies variations in credit access and refinancing options across debt levels. The second exercise, encapsulated in Table 6, reinforces a persistent U-shaped relationship between DTI and the MPC across LTI distributions, with more elevated MPC evident in low-DTI and low-LTI households. 21We control for each variable by dividing the sample into quartiles of the distribution and including a dummy variable for each quartile. 22The estimation equation is given by f Ci=P3 j=1⇣jIncomeShocki·I(Zi=j)+↵jI(Zi=j)⌘+X0 i+"i, where I(Zi=j) is an indicator function taking the value 1 if household ibelong to tertile jof the variable of interest. 22 Gruber, Jonathan, “The Consumption Smoothing Benefits of Unemployment Insurance,” The American Economic Review, 1997, 87 (1), 192–205. Hall, Robert E and Frederic S Mishkin, “The Sensitivity of Consumption to Transitory Income: Estimates from Panel Data on Households,” Econometrica, 1982, 50 (2), 461–481. Halla, Martin, Julia Schmieder, and Andrea Weber, “Job displacement, family dynamics, and spousal labor supply,” American Economic Journal: Applied Economics, 2020, 12 (4), 253–87. Hardoy, Inés and Pål Schøne, “Displacement and household adaptation: insured by the spouse or the state?,” Journal of Population Economics, 2014, 27 (3), 683–703. Hendren, Nathaniel, “Knowledge of future job loss and implications for unemployment insurance,” American Economic Review, 2017, 107 (7), 1778–1823. Huttunen, Kristiina, Jarle Møen, and Kjell G Salvanes, “How destructive is creative destruction? E↵ects of job loss on job mobility, withdrawal and income,” Journal of the European Economic Association, 2011, 9(5), 840–870. Iacus, Stefano M, Gary King, and Giuseppe Porro, “Causal inference without balance checking: Coarsened exact matching,” Political Analysis, 2012, 20 (1), 1–24. Jacobson, Louis S, Robert J LaLonde, and Daniel G Sullivan, “Earnings losses of displaced workers,” The American Economic Review, 1993, 83 (4), 685–709. Jappelli, Tullio and Luigi Pistaferri, “The consumption response to income changes,” Annu. Rev. Econ., 2010, 2(1), 479–506. and , “Fiscal policy and MPC heterogeneity,” American Economic Journal: Macroeconomics, 2014, 6(4), 107–36. Jarosch, Gregor, “Searching for job security and the consequences of job loss,” Econometrica, 2023, 91 (3), 903–942. Johnson, David S, Jonathan A Parker, and Nicholas S Souleles, “Household expenditure and the income tax rebates of 2001,” American Economic Review, 2006, 96 (5), 1589–1610. Kaplan, Greg and Giovanni L. Violante, “A Model of the Consumption Response to Fiscal Stimulus Payments,” Econometrica, 2014, 82 (4), 1199–1239. and Giovanni L Violante, “Microeconomic heterogeneity and macroeconomic shocks,” Journal of Economic Perspectives, 2018, 32 (3), 167–194. , , and Justin Weidner, “The wealthy hand-to-mouth,” Brookings Papers on Economic Activity, 2014, pp. 77–138. Kawano, Laura and Sara LaLumia, “How Income Changes During Unemployment Evidence from Tax Return Data,” Journal of Human Resources, 2017, 52 (2), 418–456. Kekre, Rohan, “Unemployment insurance in macroeconomic stabilization,” Review of Economic Studies, 2022, p. rdac080. Koijen, Ralph, Stijn Van Nieuwerburgh, and Roine Vestman, “11. Judging the Quality of Survey Data by Comparison with “Truth” as Measured by Administrative Records,” in “Improving the Measurement of Consumer Expenditures,” University of Chicago Press, 2015, pp. 308–346. Kolsrud, Jonas, Camille Landais, and Johannes Spinnewijn, “The value of registry data for consumption analysis: An application to health shocks,” Journal of Public Economics, 2020, 189, 104088. , , Peter Nilsson, and Johannes Spinnewijn, “The optimal timing of unemployment benefits: 29 Theory and evidence from Sweden,” American Economic Review, 2018, 108 (4-5), 985–1033. Korinek, Anton and Alp Simsek, “Liquidity Trap and Excessive Leverage,” American Economic Review, 2016, 106 (3), 699–738. Kreiner, Claus Thustrup, David Dreyer Lassen, and Soren Leth-Petersen, “Measuring the Accuracy of Survey Responses Using Administrative Register Data: Evidence from Denmark,” in Christopher D. Carroll, Thomas F. Crossley, and John Sabelhaus, eds., Improving the Measurement of Consumer Expenditures, University of Chicago Press, February 2014, pp. 289–307. Krolikowski, Pawel, “Job ladders and earnings of displaced workers,” American Economic Journal: Macroeconomics, 2017, 9(2), 1–31. , “Choosing a Control Group for Displaced Workers,” Industrial and Labor Relations Review, 2018, 71 (5), 1232–1254. Krueger, Dirk, Kurt Mitman, and Fabrizio Perri, “Macroeconomics and household heterogeneity,” in “Handbook of macroeconomics,” Vol. 2, Elsevier, 2016, pp. 843–921. Kueng, Lorenz, “Excess sensitivity of high-income consumers,” The Quarterly Journal of Economics, 2018, 133 (4), 1693–1751. Kyyrä, Tomi and Ralf A Wilke, “Reduction in the long-term unemployment of the elderly: A success story from Finland,” Journal of the European Economic Association, 2007, 5(1), 154–182. Lachowska, Marta, Alexandre Mas, and Stephen A Woodbury, “Sources of displaced workers’ long-term earnings losses,” American Economic Review, 2020, 110 (10), 3231–3266. Landais, Camille and Johannes Spinnewijn, “The value of unemployment insurance,” Review of Economic Studies, November 2021, 88 (6), 3041–3085. McKay, Alisdair and Ricardo Reis, “Optimal automatic stabilizers,” The Review of Economic Studies, 2021, 88 (5), 2375–2406. Meyer, Bruce D, Wallace KC Mok, and James X Sullivan, “Household Surveys in Crisis,” Journal of Economic Perspectives, 2015, 29 (4), 199–226. Mian, Atif, Kamalesh Rao, and Amir Sufi, “Household Balance Sheets, Consumption, and the Economic Slump,” The Quarterly Journal of Economics, 2013, 128 (4), 1687–1726. Misra, Kanishka and Paolo Surico, “Consumption, income changes, and heterogeneity: Evidence from two fiscal stimulus programs,” American Economic Journal: Macroeconomics, 2014, 6(4), 84– 106. Modigliani, Franco and Richard Brumberg, “Utility analysis and the consumption function: An interpretation of cross-section data,” Franco Modigliani, 1954, 1(1), 388–436. Nakamura, Emi and Jón Steinsson, “Identification in macroeconomics,” Journal of Economic Perspectives, 2018, 32 (3), 59–86. Olafsson, Arna and Michaela Pagel, “The liquid hand-to-mouth: Evidence from personal finance management software,” The Review of Financial Studies, 2018, 31 (11), 4398–4446. Schmieder, Johannes F, Till Von Wachter, and Jörg Heining, “The costs of job displacement over the business cycle and its sources: Evidence from Germany,” American Economic Review, 2023, 113 (5), 1208–1254. Schwerdt, Guido, “Labor turnover before plant closure: ‘Leaving the sinking ship’ vs. ‘Captain throwing ballast overboard’,” Labour Economics, 2011, 18 (1), 93–101. 30 Tables and Figures Figure 1: Development of key observables before job loss (a) Labor income after tax     8 8 8 8 7UHDWHG &RQWURO (b) Spousal labor income after tax      8 8 8 8 7UHDWHG &RQWURO (c) Safe assets      8 8 8 8 7UHDWHG &RQWURO (d) Debt      8 8 8 8 7UHDWHG &RQWURO (e) Share owning risky assets      8 8 8 8 7UHDWHG &RQWURO (f) Housing      8 8 8 8 7UHDWHG &RQWURO Notes: Vertical lines show 95% confidence intervals, where standard errors are clustered at the matching group level. Observations are weighted using CEM-weight, described in Appendix E. Monetary values are CPI-adjusted with 2014 as base year, and measured in USD (NOK/USD =6.3019). 31 Figure 2: Income loss and consumption expenditure responses after job loss       8 8 8 8 8 8 8 8 8 7UHDWHG &RQWURO          8 8 8 8 8 8 $EVROXWHOHIW 3HUFHQWULJKW (a) Earnings, pre-tax     8 8 8 8 8 8 8 8 8 7UHDWHG &RQWURO           8 8 8 8 8 8 $EVROXWHOHIW 3HUFHQWULJKW (b) After-tax labor income     8 8 8 8 8 8 8 8 8 7UHDWHG &RQWURO           8 8 8 8 8 8 $EVROXWHULJKW 3HUFHQWOHIW (c) Spousal earnings, pre-tax     8 8 8 8 8 8 8 8 8 7UHDWHG &RQWURO         8 8 8 8 8 8 $EVROXWHOHIW 3HUFHQWULJKW (d) Consumption expenditure Notes The left column shows the dynamic path of each outcome variable (k+k⇥Ti), whereas the right column show the di↵erence between the treated and control group (k), measured both in absolute terms (left-hand side axes) and percentage change relative to pre-job loss average (right-hand side axes). Top and bottom 1% of observations are censored when estimating percentage change. Vertical lines show 95% confidence intervals, and standard errors are clustered at the level of the matching group. Observations are weighted using CEM-weight, described in Appendix E. Monetary values are CPIadjusted with 2014 as base year, and measured in USD (NOK/USD =6.3019). 32 Figure 3: Income loss and consumption expenditure responses across LTI and DTI groups (a) By LTI (a) Earnings, pre-tax      8 8 8 8 8 8 8 8 7HUWLOH 7HUWLOH 7HUWLOH (b) After-tax labor income      8 8 8 8 8 8 8 8 7HUWLOH 7HUWLOH 7HUWLOH (c) Consumption expenditure     8 8 8 8 8 8 8 8 7HUWLOH 7HUWLOH 7HUWLOH (b) By DTI (d) Earnings, pre-tax      8 8 8 8 8 8 8 8 7HUWLOH 7HUWLOH 7HUWLOH (e) After-tax labor income      8 8 8 8 8 8 8 8 7HUWLOH 7HUWLOH 7HUWLOH (f) Consumption expenditure     8 8 8 8 8 8 8 8 7HUWLOH 7HUWLOH 7HUWLOH (c) By LTI-DTI (g) Earnings, pre-tax      8 8 8 8 8 8 8 8 /RZ/7,ORZ'7, /RZ/7,KLJK'7, +LJK/7,ORZ'7, +LJK/7,KLJK'7, (h) After-tax labor income   8 8 8 8 8 8 8 8 /RZ/7,ORZ'7, /RZ/7,KLJK'7, +LJK/7,ORZ'7, +LJK/7,KLJK'7, (i) Consumption expenditure     8 8 8 8 8 8 8 8 /RZ/7,ORZ'7, /RZ/7,KLJK'7, +LJK/7,ORZ'7, +LJK/7,KLJK'7, Notes: Low (high) LTI refers to tertile 1 (2,3) of the distribution of liquid-assets-to-income. A high (low) DTI refers to tertile three (1,2) of the distribution of debt-to-income. Tertiles are computed for each calendar-year cohort of job losers using the mean of LTI or DTI in the two years preceding job loss. All variables are measured as percentage change relative to the pre-job loss average of the treatment group (year U1). Top and bottom 1% of observations are censored. 95% confidence intervals. Standard errors are clustered at the matching group level. 33 Figure 4: Income loss and consumption expenditure responses by various margins (a) Households with or without children, all ages (a) Earnings, pre-tax      8 8 8 8 8 8 8 8 +DVFKLOGUHQ 1RFKLOGUHQ (b) After-tax labor income      8 8 8 8 8 8 8 8 +DVFKLOGUHQ 1RFKLOGUHQ (c) Consumption exp.     8 8 8 8 8 8 8 8 +DVFKLOGUHQ 1RFKLOGUHQ (b) Households with or without children, 0-17 (d) Earnings, pre-tax     8 8 8 8 8 8 8 8 +DVFKLOGUHQ 1RFKLOGUHQ (e) After-tax labor income      8 8 8 8 8 8 8 8 +DVFKLOGUHQ 1RFKLOGUHQ (f) Consumption exp.       8 8 8 8 8 8 8 8 +DVFKLOGUHQ 1RFKLOGUHQ (c) Households with young children (0-5), older children (6-17) or none (g) Earnings, pre-tax      8 8 8 8 8 8 8 8 &KLOGUHQ 2OGHUFKLOGUHQRQO\ 1RFKLOGUHQ (h) After-tax labor income      8 8 8 8 8 8 8 8 &KLOGUHQ 2OGHUFKLOGUHQRQO\ 1RFKLOGUHQ (i) Consumption exp.       8 8 8 8 8 8 8 8 &KLOGUHQ 2OGHUFKLOGUHQRQO\ 1RFKLOGUHQ (d) Married or cohabiting couples (j) Earnings, pre-tax     8 8 8 8 8 8 8 8 0DUULHG &RKDELWLQJ (k) After-tax labor income     8 8 8 8 8 8 8 8 0DUULHG &RKDELWLQJ (l) Consumption exp.       8 8 8 8 8 8 8 8 0DUULHG &RKDELWLQJ Notes: All category variables are measured in the year of job loss, and groups are kept constant throughout the event window. All variables are measured as percentage change relative to the pre-job loss average of the treatment group (year U1). Top and bottom 1% of observations are censored. 95% confidence intervals. 34 Figure 5: The marginal propensity to consume across the distribution of income, age and wealth (a) Labor income a.t. (t-1)            (b) Household income a.t. (t-1)              (c) Age (t)             (d) Net wealth (t-1)              (e) Liquid assets (t-1)              (f) Debt (t-1)             (g) Liquid assets-to-income (t-1)             (h) Debt-to-income (t-1)             Notes The coefficients are obtained from regressions where the income shock is interacted with dummies for tertiles of the variable interest, including as controls a fourth-order polynomial in age, a second-order polynomial of no. of children, a dummy for marital status, a dummy for belonging to each tertile of the distribution, lagged household income, and lagged net wealth. The y-axis plots the mean of the variable within each tertile. All monetary values are CPI-adjusted with 2014 as base year, and measured in 1000 USD (NOK/USD=6.3019). Standard errors are clustered at the industry level, and vertical lines indicate 90% confidence intervals. 35 Figure 6: Income loss and consumption expenditure responses over the business cycle Recessions as defined by Aastveit et al. (2016) (a) Earnings, pre-tax      8 8 8 8 8 8 8 8 5HFHVVLRQ 1RUHFHVVLRQ (b) After-tax labor income      8 8 8 8 8 8 8 8 5HFHVVLRQ 1RUHFHVVLRQ (c) Consumption exp.     8 8 8 8 8 8 8 8 5HFHVVLRQ 1RUHFHVVLRQ Recessions as defined by OECD (d) Earnings, pre-tax     8 8 8 8 8 8 8 8 5HFHVVLRQ 1RUHFHVVLRQ (e) After-tax labor income      8 8 8 8 8 8 8 8 5HFHVVLRQ 1RUHFHVVLRQ (f) Consumption exp.     8 8 8 8 8 8 8 8 5HFHVVLRQ 1RUHFHVVLRQ Notes: Aastveit et al. (2016) define the following periods as recessions: 2001Q2-2001Q3, 2002Q32003Q1, 2008Q3-2010Q2. All catergory variables are measured in the year of job loss, and groups are kept constant throughout the event window. Top and bottom 1% of observations are censored. 95% confidence intervals. 36 Table 1: Summary statistics for targeted and non-targeted variables. Mean Median Unemployed Control group Possible controls Unemployed Control group Possible controls Targeted variables Demographics Age 44.7 44.7 46.1 45.0 45.0 46.0 Balance-sheet Male labor income 77,040 75,819 82,384 70,535 69,577 73,564 Debt 198,963 191,221 197,839 174,344 168,440 149,615 Liquid assets 35,828 34,543 124,112 16,882 16,460 27,824 Share homeowner (%) 0.94 0.95 0.89 Share w/risky assets (%) 53.71 53.70 59.32 Education Low education 34.67 34.60 27.60 High School Education 43.09 43.18 36.56 Higher Education 22.24 22.22 35.39 Industry composition Agriculture 0.60 0.44 0.82 Education 2.74 2.29 4.99 Health and social services 2.92 2.05 4.22 Manufacturing and construction 37.46 39.36 27.74 Other services 1.36 0.97 2.04 Public admin. and defence 2.24 1.95 6.34 Retail and services 47.19 48.67 31.93 Unknown 5.49 4.27 21.91 Employment Firm tenure 6.9 7.1 7.1 7.0 7.0 6.0 Non-targeted variables Demographics Share with children (%) 66.94 67.88 55.08 Number of children 1.4 1.5 1.2 1.0 1.0 1.0 Balance-sheet Consumption 101,157 100,622 111,452 93,098 93,344 95,650 Spouse’s labor income 47,198 46,885 45,644 48,020 47,724 45,932 Safe assets 25,576 24,098 49,389 12,616 12,145 19,098 Risky assets 10,252 10,445 74,723 192 220 1,117 Share receiving sickness benefits (%) 10.74 10.48 9.63 Share receiving disability benefits (%) 0.22 0.33 3.41 Employment Share public employer 0.08 0.06 0.22 Estimated probability (%) 1.0 0.9 0.6 0.8 0.7 0.4 Days Unemployed 234 91 Year unemployed 2007 2006 N 11,497 147,027 251,618 11,497 147,027 251,618 Notes: Monetary values are CPI-adjusted with 2014 as base year, and measured in USD (NOK/USD =6.3019). Possible controls include the full set of households that satisfy all sample selection criteria except actual job loss. Mean and median of the control group are weighted using CEM-weights, see Online Appendix E. All variables are measured two years prior to the year of job loss, except tenure and the probability of unemployment which is measured one year prior to job loss, and age which is measured in the year of job loss. The estimated probability of job loss is based on a probit regression with the following controls: public employer, tenure, education, industry, firm age, and firm size (see Online Appendix Dfor details). 37 Table 2: Summary statistics (averages) by LTI-DTI-groups Panel (a): By the separate distributions of LTI and DTI Low LTI Mid LTI High LTI Low DTI Mid DTI High DTI Demographics and unemployment Year 2006 2006 2006 2006 2006 2005 Age 42.5 43.2 45.4 47.9 43.4 39.9 Days unemployed 233 234 235 273 219 211 Balance sheet Debt 232,015 208,970 160,592 79,369 197,611 322,557 Debt-to-income 2.38 2.05 1.50 0.77 1.93 3.21 Liquid assets 5,372 19,467 77,639 54,210 27,487 20,838 Liquid assets-to-income 0.06 0.20 0.76 0.54 0.27 0.20 Interest paid 13,954 11,395 82,27 4,380 10,983 18,104 Interest paid /hh. income a.t. 0.15 0.11 0.08 0.04 0.11 0.18 Mortgage ammortization /hh. income a.t.* 0.27 0.24 0.21 0.15 0.23 0.32 N 3,848 3,827 3,822 3,804 3,818 3,875 Panel (b): By the joint distribution of LTI and DTI Low LTI, Low LTI, High LTI, High LTI, low DTI high DTI low DTI high DTI Demographics and unemployment Year 2006 2005 2005 2006 Age 44.7 40.1 46.0 39.8 Days unemployed 254 211 243 211 Balance sheet Debt 158,861 312,515 131,312 331,561 Debt-to-income 1.63 3.21 1.25 3.21 Liquid assets 5,628 5,091 53,481 34,960 Liquid assets-to-income 0.06 0.05 0.53 0.34 Interest paid 9,609 18,736 6,996 17,538 Interest paid /hh. income a.t. 0.10 0.20 0.07 0.17 Mortgage amortization /hh. income a.t.* 0.22 0.32 0.19 0.32 N 2,016 1,832 5,606 2,043 Notes: Monetary values are CPI-adjusted with 2014 as base year, and measured in USD (NOK/USD =6.3019). Low (high) LTI refers to tertile 1 (2,3) of the distribution of liquid-assets-to-income. A high (low) DTI refers to tertile three (1,2) of the distribution of debt-to-income. Tertiles are computed for each calendar-year cohort of job losers using the mean of LTI or DTI in the two years preceeding job loss. All variables are measured one year before job loss, except age and year of job loss, which is measured in the year of job loss. Mortgage amortization is a back-of-the-envelope calculation assuming 25 years downpayment plans starting at 30 years old and 5% interest rate, using group-averages for age and debt to calculate monthly payments: amortization =interest rate·debt 1(1+interestrate)55age . 38