Credit Smoothing. Sean Hundtofte and Michaela Pagel. February 10, Abstract

Credit Smoothing Sean Hundtofte and Michaela Pagel February 10, 2018 Abstract Economists believe that high-interest, unsecured, short-term borrowing, for instance via credit cards, helps individuals to smooth consumption in the event of transitory income shocks. This paper estimates that individuals do not appear to use such credit to smooth consumption when transitory income shocks such as unemployment occur. In contrast, it appears as if individuals smooth credit balances rather than consumption. Using a representative sample of U.S. households' credit reports, we instrument for local changes in employment using a (1991) style methodology, based on pre-period county-level industrial composition interacted with nationwide industry trends, and complement this methodology with an individual-level analysis to show that, on average, borrowing does not increase in response to unemployment shocks. This absence of increased borrowing occurs in spite of an excess supply of credit or nonbinding liquidity constraints as indicated by credit limits. These ndings are dicult to reconcile with theories of consumption smoothing, which would predict the demand for credit is countercyclical. Federal Reserve Bank New York, Sean.Hundtofte@ny.frb.org Columbia Business School, NBER, and CEPR; mpagel@columbia.edu The views expressed in this paper are solely those of the authors and do not necessarily reect those of the Federal Reserve Bank of New York or the Federal Reserve System. 1

1 Introduction How does high-interest, unsecured, short-term borrowing respond to transitory income shocks? Standard consumption models make a clear-cut prediction: if unsecured credit is ever used, then in response to adverse transitory income shocks. However, empirical evidence on borrowing in response to transitory income shocks remains scarce. Furthermore, disentangling credit demand and supply as well as transitory and permanent income shocks (the former may entice a borrowing response while the latter should not) is dicult. It could be that adverse income shocks increase default risk, making borrowing impossible, or that both borrowers and lenders have limited capacity to distinguish between transitory and permanent income shocks. In this paper, we seek to investigate and quantify how credit card borrowing responds to unemployment shocks during the period between 2000 and 2014 and separately (in case the crisis and subsequent recovery period is dierent) from 2008 to 2014. From 2008 to 2014, unemployment rates rose sharply but revolving credit outstanding in the United States fell 15% from its June 2008 (nominal, seasonally-adjusted) peak (Federal Reserve G19 Series). In contrast to our ndings, Keys et al. (2017) argue that the demand for credit was greater, but that this was outweighed by the supply response from lenders as employment fell. We follow Keys et al. (2017) in exploiting variation in county-level employment shocks, using a -style industry composition shift-share instrument, in order to produce estimates of the elasticity of equilibrium credit card account balances, limits, inquiries, and utilization with respect to income and employment. 1 In contrast to Keys et al, we look at individual-level outcomes 1 The estimation and interpretation of causal eects using -style instruments to isolate shocks to labor demand has been employed by a number of authors. An incomplete list of papers includes Blanchard et al. (1992); Gould et al. (2002); Aizer (2010); Nguyen et al. (2015); Chodorow-Reich et al. (2012); Maestas et al. (2016). 2

to show that individual credit-card borrowing does not typically increase in the event of unemployment shocks, even for those who have access to credit. These ndings are also consistent with those in Ganong and Noel (2016), who use transaction-level bank account data of one U.S. bank and show that, in their sample, borrowing increases by merely $23 two months after the onset of unemployment and by merely $45 two months after unemployment benet exhaustion even when individuals have substantial credit available. We thus conclude that private sources of consumer credit do not appear to facilitate private self insurance of individual consumption. Most economic models would suggest that credit demand should be countercyclical while credit supply is procyclical. However, we conclude that U.S. households allow their consumption to adjust rather than than their borrowing. [Such cyclicality of unsecured credit would amplify business cycles.] While an extensive literature has explored supply ampliers during the Great Recession, our paper suggests to examine demand ampliers via credit utilization of households during the initial expansionary and then contractionary period. Our ndings thus relate to the analysis in Herkenho et al. (2013) showing that access to unsecured credit appears to deepen business cycles. We use data on credit card accounts from the Federal Reserve Bank of New York/Equifax Consumer Credit Panel (CCP), covering the universe of accounts nationwide from 2000 to the present. We construct county-quarter measures of balances, limits, inquiries, and utilization. We nd that individuals in counties with adverse unemployment shocks do not appear to increase their overall outstanding revolving balances even with respect to available limitsnor do they increase their credit inquiries for new credit relative to counties with less adverse unemployment shocks. We thus show that on average, individuals do not increase credit card debt in response to unemployment even if they could 3

have done so and many of the unemployment shocks were arguably transitory. This study complements other work that has focused on the debt overhang of secured debt such as mortgages as in Mian et al. (2013), by showing that latent demand for unsecured credit is procyclical rather than countercyclical, potentially amplifying business cycles. We thus argue that previous results, such as those of Agarwal et al. (2015), who use a dierent identication approach, namely discontinuities in credit card oer algorithms to show a disconnect between credit supply and credit demand during the post-crisis recovery period, is relevant for individuals who are hit by unemployment. Sullivan (2008) also nds that very low asset households as well as wealthy households do not increase their debt in response to unemployment while the average eect for all other households is 11%. The author argues that low asset households are credit constrained, which we can directly address in this paper as we observe credit limits. More generally, our ndings underscore concerns regarding the value of - nance to society. High cost, unsecured credit access does not appear to be used to smooth consumption as many economists would believe. The self-insurance aspects of credit cards are limited if individuals misunderstand the high costs of credit in normal times or do not tap these lines in bad times. As such, credit demand could amplify business-cycle consumption volatility rather than mitigating it through consumption smoothing. In turn, government policy or education may have a role to play in aecting demand as well as the supply of credit. 2 Data We use quarterly data on credit card accounts from the Consumer Credit Panel (CCP) of the Federal Reserve Bank of New York \citep{leevanderklaauw2010}, 4

which has coverage of the universe of accounts nationwide for the period between 2000 and 2017 and separately for 2008 to 2014. The Consumer Credit Panel (CCP) is an anonymous longitudinal panel of individuals, comprising a 5\% random sample of all individuals who have a credit report with Equifax. The quarterly sample starts in 1999Q1 and currently ends in 2017Q3. The data is described in detail in \citet{leevanderklaauw2010}. We use the [1%] sample for purposes of the current analysis, which includes information on approximately 2,500,000 randomly selected individuals each quarter. The CCP provides credit registry data on all debts monitored by one of the three main credit bureaus, in addition to public records (bankruptcy, death) and mobility (address changes) for any individual that is visible to the credit registry, e.g. excluding the young and those without reported debts. This panel dataset allows the econometrician to track all aspects of individuals' nancial liabilities, including bankruptcy and foreclosure, mortgage status, detailed delinquencies, various types of debt, with number of accounts and balances. Address information is available at the census block level. Summary statistics for our sample are presented in Table [...] 3 Methodology Our main specications regress dollar changes in credit outcomes at the individual level on percentage changes in employment, where we instrument for the employment change using -style instruments. We instrument for the change in a county's employment with the interaction of the pre-period industry mix of employment in that local labor market and the national change in industry employment (exclusive of the given county). Our exclusion restriction to examine changes in debt balances in response to employment shocks is that the pre-period industrial mix interacted with the national industry trends does 5

not directly aect local credit card variables outside of its aect on employment. Of course, without further examining limits and other variables such as prices and quantities borrowed, we cannot draw further conclusions on supply versus demand responses to these employment shocks. More formally, our instrument is dened as follows: P redictedemployment c,t = i ( Employment i,t Employment i,t 1 1)EmploymentShare i,t,c where Employment i,t Employment i,t 1 1 is the change in national employment of industry i from time t 1 to t and EmploymentShare i,t,c is the share of employment in industry i at time t in county c. As credit outcomes we consider: the change in the credit card balance, the change in the credit card limit, the total balances of all revolving debts (including home equity), the number of credit inquiries (from any lender or application), and the credit utilization ratio. To calculate credit utilization ratios, we divide the outstanding balance on the category of revolving debt by the total appropriate credit limit. In turn, we regress the credit outcome on the contemporaneous or delayed employment instrument controlling for individual xed eects, thus controlling for all observable and unobservable time-invariant characteristics, and time xed eects. Furthermore, we may or may not control for the delayed FICO score, or only consider the period 2008 to 2014. We cluster standard errors at the county and time level as the employment instrument constitutes a county-level treatment. We take a random 1% subsample of the population and look at the average increase in borrowing in response to a shock-instrumented local change in employment. The standard deviation of local employment changes between 2000 and 2014 at the quarterly level is 1.82%. In turn, we follow Keys et al. (2017) and assume that the average worker earns $36,000 and, if faced by unemployment, a 6

90 day unemployment spell with an unemployment benet replacement rate of 2/3rds. In turn, that worker faces a quarterly uninsured reduction of $3,000 in income. Thus, a one standard deviation increase in unemployment should lead to an average increase in balances of $3,000*1.82% = $35 in the event of full self insurance. [check this under latest tables] 4 Results We estimate insignicant coecients of around $200 for the instrument on total revolving credit card balances, as can be seen in Table 1. To interpret the magnitudes of the coecients, note that a standard deviation of the instrument is 0.025. Thus, the insignicant credit card borrowing response to a one standard deviation increase in employment is approximately $5. The coecients are tightly estimated with standard errors of approximately 1.5 the size of the coecients. [Insert Table 1 about here] Furthermore, Table 13 displays the regression results for quantiles to see where the estimaton results are strongest or weakest and how linear they are. Even at higher ends of the conditional distribution, we do not see large changes in credit card balances in response to employment shocks. [Insert Table 13 about here] We also estimate insignicant coecients of around $1000 for the instrument on total credit card limits, as can be seen in Table 2. Thus, the insignicant credit card limit response to a one standard deviation increase in employment is approximately $25. Again, the coecients are tightly estimated with standard errors of approximately 1.2 the size of the coecients. 7

[Insert Table 2 about here] The picture changes to some extent when we look at total revolving credit, which includes mortgages and home equity loans. We estimate insignicant coecients for the instrument that vary and have relatively large standard errors. This holds true for balances as well as limits, as can be seen in Tables 3 and 4. We thus conclude that there exists a lot of heterogeneity in the way individuals respond to employment shocks when it comes to all forms of credit, not only unsecured high-interest credit such as credit cards. [Insert Tables 3 and 4 about here] Finally, we can look at two more credit outcomes, the credit utilization ratio and new inquiries. For credit utilization we again estimate very tight insignicant zero eects, as can be seen in Table 5. It appears as if individuals smooth credit utilization rates rather than consumption. For inquiries, we estimate a marginally signicant negative coecients for the crisis period from 2008 to 2014. Nevertheless, the coecients are not large. In response to a one standard deviation of the shock, individuals reduce their number of credit inquiries by 0.375, as can be seen in Table 6. [Insert Tables 5 and 6 about here] When we allow for heterogeneity by income, we again estimate mostly insignicant coecients for the instrument on total revolving credit card balances, as can be seen in Table 7. But more importantly, we nd a large degree of heterogeneity by a proxy for individual's baseline income. Debt balances in higher income areas respond to employment shocks more like the standard model would suggest they should: [net impact for highest quartile vs lowest quartile]. coecients on the interaction of employment with income are [positive and large, in the range of... and tightly estimated with standard errors of 8

approximately the size of the coecients.] [Insert Table 7 about here] We estimate somewhat larger, negative, and marginally signicant coecients of around $3000 for the instrument on total credit card limits over the whole period, as can be seen in Table 8. Thus, the negative credit card limit response to a one standard deviation increase in employment is approximately $75. Over the crisis period, however, the response is smaller and insignicant. Again, the coecients are tightly estimated with standard errors of approximately the size of the coecients. [Insert Table 8 about here] Again, the picture changes to some extent when we look at total revolving credit, which includes mortgages and home equity loans. We estimate marginally signicant and larger coecients for the instrument showing that better employment prospects increase borrowing, though somewhat less so in the crisis period. This holds true for balances as well as limits, as can be seen in Tables 9 and 10. We thus conclude that when we allow for income heterogeneity then employment has a positive eect on borrowing, which is more in line with a credit demand theory rather than consumption smoothing. [Insert Tables 9 and 10 about here] Finally, we can allow for heterogeneity in income and look at two more credit outcomes, the credit utilization ratio and new inquiries. For credit utilization we again estimate very tight insignicant zero eects, as can be seen in Table 11. It appears as if individuals smooth credit utilization rates rather than consumption. For inquiries, we estimate insignicant negative coecients, which are again small, as can be seen in Table 12. [Insert Tables 11 and 12 about here] 9

5 Conclusion Economists believe that high-interest, unsecured, short-term borrowing, for instance via credit cards, can help individuals to smooth consumption in the event of transitory income shocks (]cite]). We have shown that individuals do not appear to use such credit to smooth consumption. In contrast, it appears as if individuals smooth credit rather than consumption through transitory income shocks. We instrument for local changes in employment using a (1991) style methodology, based on pre-period county-level industrial composition interacted with nationwide industry trends and complement this methodology with an individual-level analysis to show that borrowing does not increase in response to unemployment shocks for most individuals. These ndings are dicult to reconcile with theories of consumption smoothing. Credit demand should be countercyclical, but may be procyclical in line with credit supply and amplify consumption volatility driven by the business cycle. 10

References Agarwal, S., S. Chomsisengphet, N. Mahoney, and J. Stroebel (2015). Do banks pass through credit expansions to consumers who want to borrow? Technical report, National Bureau of Economic Research. Aizer, A. (2010). The gender wage gap and domestic violence. The American economic review 100 (4), 1847., T. J. (1991). Boon or boondoggle? the debate over state and local economic development policies. Blanchard, O. J., L. F. Katz, R. E. Hall, and B. Eichengreen (1992). Regional evolutions. Brookings papers on economic activity 1992 (1), 175. Chodorow-Reich, G., L. Feiveson, Z. Liscow, and W. G. Woolston (2012). Does state scal relief during recessions increase employment? evidence from the american recovery and reinvestment act. American Economic Journal: Economic Policy 4 (3), 118145. Ganong, P. and P. Noel (2016). How does unemployment aect consumer spending? Technical report, Working paper. Gould, E. D., B. A. Weinberg, and D. B. Mustard (2002). Crime rates and local labor market opportunities in the united states: 19791997. The Review of Economics and Statistics 84 (1), 4561. Herkenho, K. F. et al. (2013). The impact of consumer credit access on unemployment. Work. Keys, B., J. Tobacman, and J. Wang (2017). Rainy day credit? unsecured credit and local employment shocks. Working Paper. 11

Maestas, N., K. J. Mullen, and D. Powell (2016). The eect of population aging on economic growth, the labor force and productivity. Technical report, National Bureau of Economic Research. Mian, A., K. Rao, and A. Su (2013). Household balance sheets, consumption, and the economic slump. The Quarterly Journal of Economics 128 (4), 1687 1726. Nguyen, H.-L., M. Greenstone, and A. Mas (2015). Essays on Banking and Local Credit Markets. Ph. D. thesis, Massachusetts Institute of Technology. Sullivan, J. X. (2008). Borrowing during unemployment unsecured debt as a safety net. Journal of human resources 43 (2), 383412. 12

Figures and Tables 2000 to 2014 2008 to 2014 2000 to 2014 2008 to 2014 Credit card balance employment 181.6 190.3 229.2 329.8 (294.3) (291.8) (396.9) (402.1) Lagged risk score Lagged employment 8.47 *** 10.2 *** 8.36 *** 10.2 *** (0.38) (0.41) (0.39) (0.41) 133.7 255.3 234.9 392.0 (281.9) (292.8) (383.7) (373.2) Observations 9297351 87811762 41673576 40904083 89383330 89281858 40912518 40899108 2 R 2 0.029 0.032 0.053 0.056 0.029 0.032 0.053 0.056 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 1: Linear regression of the change in individual credit card balances on the (lagged) employment shock at the county level controlling for individual and time xed eects as well individual risk scores. Standard errors are double clustered at the county and time level. 13

2000 to 2014 2008 to 2014 2000 to 2014 2008 to 2014 Credit card limits employment Lagged risk score 1065.9 1061.1 627.4 680.7 (1179. 3) (1203.3) (1621.6) (1637.0) 7.47 *** 7.17 *** 7.37 *** 7.17 *** (0.43) (0.62) (0.43) (0.62) Lagged employment 454.7 557.1-250.7-146.2 (1038.6) (1049.6) (1232.2) (1244.6) 87811762 41673576 40904083 89383330 89281858 40912518 40899108 Observations 92973 512 R 2 0.025 0.026 0.068 0.069 0.026 0.026 0.068 0.069 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 2: Linear regression of the change in individual total credit card limits on the (lagged) employment shock at the county level controlling for individual and time xed eects as well individual risk scores. Standard errors are double clustered at the county and time level. 14

2000 to 2014 2008 to 2014 2000 to 2014 2008 to 2014 Total revolving balance employment Lagged risk score -130.3 1833.7-2211.1-2642.8 (6401.5) (6923.4) (3544.2) (3853.7) -28.4 *** -21.0 *** -28.2 *** -21.0 *** (0.99) (1.12) (0.99) (1.12) Lagged employment -1066.0-944.4-2011.1-2462.9 (6263.0) (6889.0) (3093.6) (3314.0) 124216070 66831259 60298057 138877688 126396063 66207453 60291545 Observations 1406049 61 R 2 0.497 0.505 0.749 0.751 0.502 0.504 0.750 0.751 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 3: Linear regression of individual total revolving balances on the (lagged) employment shock at the county level controlling for individual and time xed eects as well individual risk scores. Standard errors are double clustered at the county and time level. 15

2000 to 2014 2008 to 2014 2000 to 2014 2008 to 2014 Total credit limit employment Lagged risk score -8631.8-4138.6-8402.1-10360.4 (14788.0) (15691.8) (5498.2) (5715.9) 25.4 *** 23.4 *** 25.7 *** 23.4 *** (1.84) (3.96) (1.81) (3.96) Lagged -11829.4-11600.2-11334.6 * -13295.5 * employment (14462.7) (15670.6) (4919.9) (5029.1) Observations 1406049 124216070 66831259 60298057 138877688 126396063 66207453 60291545 61 R 2 0.629 0.637 0.836 0.837 0.636 0.636 0.836 0.837 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 4: Linear regression of the change in individual total credit limits on the (lagged) employment shock at the county level controlling for individual and time xed eects as well individual risk scores. Standard errors are double clustered at the county and time level. 16

2000 to 2014 2008 to 2014 2000 to 2014 2008 to 2014 Utilization ratio employme nt shock Lagged risk score 2.40 2.38 1.37 1.36 (2.40) (2.41) (1.72) (1.71) -0.0035 *** -0.0026 ** -0.0035 *** -0.0026 ** (0.00082) (0.00072) (0.00081) (0.00072) Lagged employme nt shock 2.59 2.54 1.41 1.37 (2.43) (2.41) (1.75) (1.73) 90349327 88388736 41565357 41369729 89959171 89875688 41395108 41364333 Observati ons R 2 0.070 0.071 0.127 0.127 0.071 0.071 0.127 0.127 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 5: Linear regression of individual utilization ratios on the (lagged) employment shock at the county level controlling for individual and time xed eects as well individual risk scores. Standard errors are double clustered at the county and time level. 17

2000 to 2014 2008 to 2014 2000 to 2014 2008 to 2014 # Credit inquiries employment Lagged risk score 0.87 1.61-14.8 * -15.7 ** (5.74) (5.69) (5.54) (5.59) 0.080 *** 0.085 *** 0.080 *** 0.085 *** (0.0012) (0.0024) (0.0012) (0.0024) Lagged employment 4.21 5.10-11.9 * -12.1 * (5.65) (5.85) (5.15) (5.21) 123763265 62583446 60057191 130609260 125931434 62090363 60050681 Observations 1319078 38 R 2 0.439 0.441 0.500 0.503 0.436 0.439 0.499 0.503 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 6: Linear regression of the number of individual credit inquiries on the (lagged) employment shock at the county level controlling for individual and time xed eects as well individual risk scores. Standard errors are double clustered at the county and time level. 18

2000 to 2014 2008 to 2014 Credit card balance *Initial income Employment Per capita income 0.019 0.010-0.024 * -0.024 * (0.019) (0.017) (0.0097) (0.011) -237.1-34.7 593.5 683.7 (368.7) (348.3) (457.7) (471.3) -0.00067 * -0.0014 ** 0.0013 0.0010 (0.00032) (0.00041) (0.00065) (0.00062) Lagged risk 8.48 *** 10.1 *** score (0.38) (0.41) N 91846243 86738474 41151976 40391367 R 2 0.029 0.033 0.056 0.059 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 7: Linear regression of the change in individual credit card balances on the (lagged) employment shock at the county level (interacted with initial income in year 2000) controlling for individual and time xed eects as well individual risk scores and per capita income. Standard errors are double clustered at the county and time level. 19

2000 to 2014 2008 to 2014 Credit card limits *Initial income Employment Per capita income 0.21 *** 0.21 *** 0.20 ** 0.21 ** (0.040) (0.040) (0.063) (0.064) -3454.5 * -3387.1 * -2876.3-2861.5 (1594.0) (1591.3) (2169.8) (2184.0) -0.0013-0.0020 * 0.0015 ** 0.0014 ** (0.00075) (0.00081) (0.00047) (0.00048) Lagged risk 7.48 *** 7.17 *** score (0.43) (0.62) N 91846243 86738474 41151976 40391367 R 2 0.025 0.027 0.070 0.071 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 8: Linear regression of the change in individual credit card limits on the (lagged) employment shock at the county level (interacted with initial income in year 2000) controlling for individual and time xed eects as well individual risk scores and per capita income. Standard errors are double clustered at the county and time level. 20

2000 to 2014 2008 to 2014 Total revolving balance *Initial income Employment Per capita income -1.31 ** -1.26 ** -0.81 *** -0.85 *** (0.43) (0.45) (0.16) (0.17) 28090.4 ** 28588.4 ** 11922.4 * 11914.3 * (8444.0) (9003.8) (4700.4) (4991.2) 0.028 * 0.030 * 0.022 *** 0.024 *** (0.011) (0.011) (0.0056) (0.0061) Lagged risk -28.3 *** -20.9 *** score (0.99) (1.11) N 139017863 122789311 66063659 59597005 R 2 0.499 0.507 0.751 0.753 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 9: Linear regression of individual total revolving balances on the (lagged) employment shock at the county level (interacted with initial income in year 2000) controlling for individual and time xed eects as well individual risk scores and per capita income. Standard errors are double clustered at the county and time level. 21

2000 to 2014 2008 to 2014 Total revolving limits *Initial income Employment Per capita income -2.41 * -2.16-0.86 ** -0.80 * (1.04) (1.10) (0.30) (0.30) 43483.0 * 41882.6 6939.7 3700.2 (19501.9) (20925.3) (7773.9) (7810.7) 0.068 * 0.065 * 0.044 *** 0.046 *** (0.025) (0.025) (0.011) (0.012) Lagged risk 25.3 *** 23.4 *** score (1.84) (3.95) N 139017863 122789311 66063659 59597005 R 2 0.630 0.639 0.838 0.839 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 10: Linear regression of the change in individual total credit limits on the (lagged) employment shock at the county level (interacted with initial income in year 2000) controlling for individual and time xed eects as well individual risk scores and per capita income. Standard errors are double clustered at the county and time level. 22

2000 to 2008 2008 to 2014 Utilization ratio *Initial income Employment Per capita income -0.00000051-0.0000044-0.000014-0.000014 (0.000019) (0.000019) (0.000022) (0.000022) 2.45 2.51 1.60 1.59 (2.34) (2.35) (1.62) (1.62) 0.000000096 0.00000051 0.0000043 0.0000044 (0.00000062) (0.00000063) (0.0000040) (0.0000043) Lagged risk -0.0035 *** -0.0026 ** score (0.00083) (0.00073) N 89247047 87310024 41045090 40852162 R 2 0.070 0.071 0.127 0.127 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 11: Linear regression of individual utilization ratios on the (lagged) employment shock at the county level (interacted with initial income in year 2000) controlling for individual and time xed eects as well individual risk scores and per capita income. Standard errors are double clustered at the county and time level. 23

2000 to 2014 2008 to 2014 # Credit inquiries * Initial income Employment Per capita income 0.00038 0.00027-0.00014-0.00019 (0.00020) (0.00019) (0.00019) (0.00018) -7.29-4.12-12.8-12.7 (6.06) (5.79) (6.76) (6.61) 0.000015 0.000012 0.0000051 0.0000036 (0.0000090) (0.0000096) (0.0000068) (0.0000068) Lagged risk 0.080 *** 0.085 *** score (0.0012) (0.0024) N 130399700 122340816 61856581 59358534 R 2 0.440 0.442 0.500 0.503 Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 12: Linear regression of the number of individual credit inquiries on the (lagged) employment shock at the county level (interacted with initial income in year 2000) controlling for individual and time xed eects as well individual risk scores and per capita income. Standard errors are double clustered at the county and time level. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Quantile 2 5 10 15 25 50 75 90 95 98 Credit card balance Employment -218.9 29.9 41.1 47.0 13.3 5.80 *** 2174.6 *** 9371.2 8714.7 140.4 (16232.0) (55548.0) (3316.7) (9365.4) (446.5) (1.10) (96.7) (17327.9) (12062.3) (52476.2) N 1371636 1371636 1371636 1371636 1371636 1371636 1371636 1371636 1371636 1371636 R 2 Table 13: Quantile splits of the linear regression of the change in individual credit card balances on the (lagged) employment shock at the county level (interacted with initial income in year 2000) controlling for individual and time xed eects as well individual risk scores and per capita income. Standard errors are double clustered at the county and time level. 24