The effect of state legislation restricting payday lending on consumer credit delinquencies: An investigation of the debt trap hypothesis

Transcription

1 The effect of state legislation restricting payday lending on consumer credit delinquencies: An investigation of the debt trap hypothesis Chintal A. Desai a Virginia Commonwealth University Gregory Elliehausen Board of Governors of the Federal Reserve System March 31, 2014 JEL Codes: D12; G28 Keywords: Consumer Credit; Payday loans; Regulation; Fringe banking; Delinquency. The authors acknowledge the suggestions of Raluca Roman and seminar participants at Virginia Commonwealth University and Financial Management Association 2013 conference. The views expressed are the authors and not those of the Board of Governors of the Federal Reserve System or its staff. The authors are solely responsible for any errors. a Corresponding author: Department of Finance, Insurance and Real Estate, 301 W. Main Street, Richmond, VA 23284; Tel: (804) ; Fax: (804) ; cdesai2@vcu.edu.

2 The effect of state legislation restricting payday lending on consumer credit delinquencies: An investigation of the debt trap hypothesis Abstract The availability of payday loans is often implicated in financial distress. We test this hypothesis by analyzing delinquencies on revolving, retail, and installment credit in Georgia, North Carolina, and Oregon. These states have restricted the formerly easy availability of payday loans by either banning them outright or capping the fees charged by payday lenders. The results, based on difference-in-difference methodology, suggest that this legislation has had small, mostly positive, effects on delinquencies. The results do not support the debt trap hypothesis that payday loans exacerbate borrowers financial difficulties. With more states considering further restrictions on payday lending, our findings have policy implications.

3 Introduction A payday loan is a short-term loan of a small dollar amount. For a typical payday loan of $300, a borrower writes a postdated check for $345, which consists of the principal amount plus $45 fees ($15 per $100 borrowed). The borrower receives $300. The loan is due in the next pay period (commonly 14 days). On the due date, the borrower repays the loan amount plus fees or the lender recovers the principal and fees by depositing the check. In some states, the borrower can renew the loan by paying another $45 fee. In this example, the annual percentage rate (APR) is % (the periodic rate of 15% per period 365/14 periods in a year). The fee would be the same regardless of the term to maturity, but the APR increases as the time until the next payday becomes shorter. Demand for small short-term loans is strong. Since its emergence in the early 1990s, the payday lending industry s loan volume had grown to nearly $50 billion in 2012 (Hecht, 2013). This is more than the amount of consumer credit held by nonfinancial businesses, as reported in the Federal Reserve s G.19 release. The payday loan trade association estimates that about 19 million households use payday loans annually, which is about 17% of US households and 23% of households in states that allow payday lending. 1 The payday loan business is controversial; the triple-digit APR itself attracts criticism. 2 Perhaps more significant is the criticism that the single-payment structure of payday loans makes them difficult to repay. 3 Critics contend that payday loan consumers often find it necessary to 1 See The number of households may be higher if Internet lenders are included. 2 Because a large part of operating costs is fixed and the loan amount is small, payday loans are relatively costly to originate (Ernst & Young, 2009; Flannery and Samolyk, 2005). That small loans have relatively high costs is not unique to payday loans; the National Commission on Consumer Finance (1972) reported a similar finding for installment loans at consumer finance companies. Commission analyses showed that break-even APRs were inversely related to loan size. At very small loan sizes, break-even APRs exceeded 100%. Analyses also indicated that break-even APRs were inversely related to term to maturity. 3 For example, see Pew Charitable Trusts (2013). 1

4 renew their loans when they mature because they cannot repay the entire balance. Repeated renewals result in high levels of accumulated fees, which exacerbate customers financial difficulties. That a considerable number of payday-loan customers repeatedly use payday loans over a prolonged period of time is cited as evidence that a problem exists. This problem is sometimes characterized as the payday loan debt trap. The perception that a debt trap problem exists has stimulated action. Regulatory responses include limits on or prohibition of renewals, mandatory minimum intervals between successive payday loans, limits on the number of payday loans per year, and the requirement that payday lenders offer installment payment plans. A few states have eliminated the product, through either rate ceilings that make payday loans unprofitable or outright prohibition. The payday loan industry, through its trade organization, also addresses the issue in industry guidelines, which include a maximum of four renewals and an installment payment option for customers who have difficulty repaying their loans. The Consumer Financial Protection Bureau (2013) collected its own data on frequency of use and concluded that its findings raise substantial consumer protection concerns: The potential for consumer harm and the data gathered to date are persuasive that further attention is warranted (p. 45). The case for a payday debt trap has not been settled, however. Considerable shares of payday loan customers do not use payday loans often or for extended periods (Elliehausen, 2009). Moreover, frequent or extended use is not necessarily evidence of the problem. Consumers living from paycheck to paycheck may be vulnerable to even small shocks, and alternatives to a payday loan may be more costly than the payday loan. 4 Also, customers may 4 For example, not paying a utility bill risks a service disconnection. To restore service, a consumer has to pay the bill, a late fee, and a reconnection fee. In addition, utilities normally require a one- or two-month deposit and the consumer experiences an interruption in service until the service is reconnected. Writing a check to pay a utility bill without having sufficient funds in the account incurs overdraft fees charged by the bank and a nonsufficient funds 2

5 fully understand, when they take a payday loan, how long it will take them to repay; a recent study found that most customers of a large payday lender correctly estimated when they would repay their loan (Mann, 2013). These considerations suggest that for a many consumers, payday loans may not be a debt trap. While some consumers undoubtedly do have difficulty repaying payday loans, others likely benefit from payday lending. This leads Caskey (2010), a leading authority on fringe lending, to ask: Do payday lenders, on net, exacerbate or relieve customers financial difficulties? Several researchers have sought to collect empirical evidence to answer this question, but their findings are mixed. For example, Morse s (2011) evidence suggests that access to payday loans may enhance financial resilience following natural disasters; for instance, California communities with payday loan offices experienced fewer foreclosures in the aftermath of earthquakes than communities without payday loan offices. Morgan and Strain (2008) found an increase in returned checks, complaints about debt collection, and Chapter 7 bankruptcy filings following payday loan bans in Georgia and North Carolina. In contrast, Skiba and Tobacman (2011) found that marginally accepted payday loan applicants in Texas were subsequently more likely to file for bankruptcy than marginally rejected applicants. Zinman s (2010) analysis of consumer survey data showed that in the wake of a rate ceiling that effectively banned payday loans in Oregon, survey respondents in the state were more likely to report a deterioration in selfassessed financial situations but less likely to use short-term credit which included late payments on bills, returned checks, and overdrafts than respondents in neighboring Washington State. Melzer (2011) also analyzed consumer survey data, which showed that access to payday loans was associated with greater self-assessed difficulty in paying bills. In an fee charged by the utility. In addition, the utility may require payment in cash if a consumer repeatedly writes checks with insufficient funds. Frequent overdrafts may also cause the bank to close a consumer s checking account. 3

6 analysis similar to that of Melzer (2011), using credit bureau data, Bhutta (2012) found evidence that access to payday loans reduces the incidence of accounts going into collection, but has little effect on credit bureau scores. Edmiston (2011) examined county-level credit bureau data and found that average credit bureau scores were higher in states that allowed payday lending than in states that did not. Several factors may contribute to these disparate empirical results. For instance, some outcome variables in these studies, such as unemployment, bankruptcy, and foreclosure, are far removed from payday loan experiences; that payday loans could have a large effect on these outcomes seems unlikely. Other outcome variables are somewhat vague and subject to interpretation (e.g., self-assessed financial situation). Moreover, it is not clear that differences are due solely to payday loan availability. For example, states in the South and West tend to allow payday loans, but these regions have also had historically higher levels of delinquencies than New England and the Mid-Atlantic, which mostly do not allow payday lending. This difference in loan performance was true also before the emergence of the payday loan industry (Elliehausen, 1999). We examine changes in delinquency rates for consumer credit following payday loan bans in North Carolina, Georgia, and Oregon. Nearly all payday loan customers use consumer credit, and most types of consumer credit require monthly payments. Consumers might rationally use payday loans to avoid late payments on consumer debts, because fees for late payments, in many cases, would be greater than those for payday loans. We use a difference-in-difference specification comparing (1) restricted (North Carolina, Georgia, and Oregon) border counties with contiguous unrestricted (neighboring states that allow payday loans) border counties and (2) 4

7 restricted hinterland (non-border) counties with unrestricted contiguous border counties. 5 The second comparison addresses the possibility that effects are dampened because some consumers in restricted border counties will cross state lines to obtain payday loans. Following payday loan bans, we generally find only small increases in delinquencies of 30 to 59 days and 60 or more days in North Carolina, Georgia, and Oregon counties relative to counties in neighboring states. Our findings do not support the hypothesis that payday loans are a debt trap that exacerbates borrowers financial difficulties. That the increases are so small is reasonable, considering that the small dollar size of payday loans provides only limited relief from financial difficulties. Our evidence is similar to that of Bhutta (2012), who found little effect of payday loan access on credit bureau scores. Background and Research Design The payday loan industry emerged in the early 1990s when check cashers began accepting, for a fee, checks that had been postdated to the customer s next payday. As mentioned previously, the customer writes a check for the amount of the loan plus a fee and receives the loan amount. The lender receives payment of interest and principal when it later deposits the check. These transactions are loans and are subject to the federal Truth-in-Lending Act and state regulations for interest rates. Generally, payday loans are offered under state laws that allow higher rate ceilings than for small installment loans. These state payday loan laws require that lenders be licensed, regulate non-price terms and remedies, and, in some cases, restrict renewals. Sixteen states and the District of Columbia either prohibit payday lending outright or permit lenders to charge only small fees, thereby making the payday lending business 5 Throughout the paper, we use terms restricted and regulated interchangeably. 5

8 unprofitable. Among the states effectively prohibiting payday lending are Georgia (GA), North Carolina (NC), and Oregon (OR). These three states are notable because they had previously allowed payday lending. 6 All three states share border with states that allow payday lending. This permits a before/after comparison group quasi-experimental study design, which provides some assurance that the effects of payday-loan bans will be kept separate from the effects of other factors (Phillips and Calder, 1979, 1980). In this study, we analyze the impact of Georgia, North Carolina, and Oregon payday legislation on consumer credit delinquencies. As mentioned previously, nearly all payday loan customers (92%) use some form of consumer credit (Elliehausen, 2009). We analyze countylevel data on delinquencies of 30 to 59 days and 60 or more days for revolving, retail, and closed-end consumer installment (non-mortgage) accounts. 7 In 2007, 54% of payday consumers had credit cards, and 22% had retail cards. Closed-end consumer installment credit includes auto loans and other closed-end loans. Fifty-three percent of payday loan customers had auto loans, and 37% had other closed-end credit (Elliehausen, 2009). Consumer loans require servicing, typically on a monthly basis, and have penalties for late payment. Payday loans are commonly similar in size to payments on consumer debts, and penalties for late payment of the later are often greater than the fees for payday loans. It follows that banning payday loans might eliminate a less costly option for avoiding late-payments fee. Alternatively, the debt trap hypothesis suggests that the burden of finance charges from repeated payday-loan renewals would exacerbate customers problems with timely payment on other forms of debt. 6 Other states, such as New York and New Jersey, have strictly enforced usury laws and have thereby precluded the emergence of payday loans. 7 Morgan and Strain (2008) find a relationship between access to payday loans and Chapter 7 bankruptcy filings, and Skiba and Tobacman (2011) report that payday loan use is associated with higher Chapter 13 filings. However, as noted by Caskey (2010), filing for bankruptcy takes place several years after using payday loans. 6

9 We define the event date as the quarter in which a ban on payday loans went into effect. Following Huang (2008), we first identify border counties of Georgia, North Carolina, and Oregon, which we designate regulated border counties. Then, for each regulated (restricted) border county, we identify a matching contiguous county that share a border with it and is located in a neighboring state that allows payday lending. These counties are designated nonregulated contiguous counties. The main reason for identifying matching pairs based on geographic proximity is that the economic, demographic, cultural, and weather conditions will be similar in nonregulated contiguous and regulated border counties; if we observe any change in delinquency rates, it will likely be due to the enactment of legislation banning payday loans. 8 We then compute delinquencies for pre- and post-event periods. The pre-event period is defined as the four quarters preceding one year of the event quarter. The post-event period is the four quarters immediately after the event quarter. Our difference-in-difference approach uses both univariate and multivariate analyses. Our paper is most closely related to papers by Morgan and Strain (2008), Zinman (2010), and Edmiston (2011). Morgan and Strain (2008) analyze household-level financial stress after the ban on payday lending in Georgia and North Carolina. Their measures of financial stress include number of bounced checks, complaints to the Federal Trade Commission, and bankruptcy rates. Their main finding is the negative correlation between access to payday credit and level of financial difficulties. Our paper differs from theirs in methodology and dependent variable. Morgan and Strain also use difference-in-difference methodology with county-level data, but while they analyze all the counties in Georgia and North Carolina and compare them to the rest of the counties in the U.S., we compare only the border counties of Georgia, North 8 County pairs are listed in the Appendix. 7

10 Carolina, and Oregon with their matching contiguous counties. In the second paper related to our work, Zinman (2010) proactively surveys payday loan customers before and after legislation to limit the APR of payday loans to 36% went into effect in Oregon. He compares Oregon households subjective assessment of their financial condition with that of households in neighboring Washington State. The main finding is that regulation of access to payday loans has actually harmed borrowers. Our paper differs from Zinman s mainly by (1) analyzing actual data on negative financial events rather than responses to a survey of selfreported financial status, and (2) analyzing the impact of payday legislation on not only Oregon but also Georgia and North Carolina. The third paper closely linked to ours is by Edmiston (2011), who analyzes borrowers credit-standing variables such as credit score and delinquency rate at the county level. Using a simplified approach, he compares the average value of a given credit-standing variable in a county that allows payday lending to that of a county that does not. In a separate analysis, he includes only low-income counties. Overall, his results suggest that restrictions harm borrowers. Our paper differs from Edmiston s in methodology and sample selection. We use difference-indifference methodology, and we compare delinquency rates only for the border counties of Georgia, North Carolina, and Oregon and their matching contiguous counties from a nonregulated state. Data Collection The data of dependent (outcome) variables delinquencies on installment, revolving, and retail credit are from TrenData of TransUnion LLC. Two categories of delinquencies are considered, less serious and serious delinquencies. Less serious delinquency is defined as the 8

11 percentage of consumers late in their payments by 30 to 59 days. Less serious delinquencies, if not frequent, probably would not seriously damage a credit score. Serious delinquency is defined as the percentage of borrowers whose payments are at least 60 days past due. Our information on state payday loan laws comes from websites of the Consumer Credit Research Foundation, Community Financial Services Association of America, National Consumer Law Center Inc., National Conference of State Legislatures, and Credit.com. Georgia The legislation in Georgia that bans payday lending was effective as of 5/1/2004. Therefore, the event time for Georgia is the 2 nd quarter of The pre-event (prior) period is four quarters starting from Quarter 3, The post-event (posterior) period is four quarters starting from Quarter 3, There are a total of 46 border counties in Georgia for our study. North Carolina Although North Carolina outlawed payday loans on 8/31/2001 (by allowing the existing law to sunset), payday lenders were still operating in the state under an agent model until 2006 (Melzer and Morgan, 2009). In this model, payday lenders form an alliance with federally insured depository institutions and serve as their agent in a state that prohibits payday lending. 9 We consider the effective date of legislation prohibiting payday lending in North Carolina as 3/1/2006; therefore the event date is the 1 st quarter of The pre-event period is four quarters starting from Quarter 2, The post-event period is four quarters starting from Quarter 2, Of the 39 border counties in North Carolina, we exclude two that neighbor on 9 See Fox and Mierzwinski (2001) for a detailed discussion of such practices. 10 Morgan and Strain (2008) consider December 2005 as the effective date in their study. 9

12 Georgia, as Georgia has not permitted payday loans since Oregon Payday lenders in Oregon are restricted by law to a maximum APR of 36%. Because of this, most of the payday lenders found their businesses to be unprofitable and left the state. The effective date of the legislation is 1/1/2007; therefore, the event date is the 1 st quarter of The pre-event period is four quarters starting from the 2 nd quarter of The post-event period is four quarters starting from the 2 nd quarter of We used the 18 border counties in Oregon for our study. Together, we have a sample of 103 border counties in which payday lending either is not allowed or is not profitable. For each regulated border county, we identify a corresponding contiguous border county of a neighboring state. For less serious delinquencies, we only have data up to the 4 th quarter of As a result, we could not analyze less serious delinquencies for Oregon. Results In this section, we first report the results of analyses using the regulated (restricted) border counties of Georgia, North Carolina, and Oregon and their matching contiguous nonregulated (unrestricted) border counties in states that allow payday lending. Then, for robustness checks, we report the results of similar analyses using the restricted hinterland counties of Georgia, North Carolina, and Oregon and the unrestricted contiguous border counties of neighboring states. 11 Zinman (2010) and Caskey (2010) also consider the effective date of Oregon payday legislation to be 1/1/

13 Table 1 reports the results of a univariate analysis comparing the mean values of revolving, retail, and installment (non-mortgage) delinquencies for the prior and posterior periods, for both the treatment (regulated border) and control (nonregulated contiguous) counties. For this analysis, we take the average of quarterly delinquencies for both prior and posterior periods and perform a paired t test. As reported in Panel A of Table 1, for the 46 border counties of Georgia, the average value of less serious delinquency on revolving credit is 3.10% for the pre-event period and 2.71% for the post-event period, resulting in a decline of 0.39%. For the sample of matched contiguous counties of neighboring states that permit payday lending, the average rate of less serious revolving delinquencies is 3.22% for the pre-event period and 2.64% for the posterior period, resulting in a decline of 0.58%. Therefore, in general, there is a decline in less serious delinquency on revolving credit for both treatment and control counties in the post-event period. However, the net decline in less serious delinquency on revolving credit between treatment and control groups is 0.20% [-0.39%-(-0.58%)]. It is different from zero at the 5% level of significance (paired t- value: 2.16). This suggests that legislation restricting payday lending in Georgia may have caused an increase in the level of less serious delinquencies on revolving credit in the post-event period for residents. For the 39 border counties of North Carolina, no significant change in the level of less serious delinquencies on revolving credit is found. For the most part, the effect of payday legislation is found to be insignificant. Because data on delinquencies in Oregon were not available for the post-event period, this analysis could not be performed for the 18 border counties of Oregon. The univariate analysis of serious delinquencies (payments 60 or more days past due) is reported in Panel B of Table 1. For the 46 border counties of Georgia, the average value of 11

14 serious revolving delinquency is 4.57% for the pre-event period and 3.04% for the post-event period, therefore a net decline of 1.53%. For the sample of matched contiguous counties, the average revolving delinquencies are 4.83% and 3.00% for the pre- and post-event periods, respectively. The decline in revolving delinquency rate, therefore, is 1.83% for the contiguous counties. Comparison of declines for treatment and control groups indicate that in the post-event period, the revolving delinquency in restricted border counties of Georgia (treatment group) is 0.30% higher than for the control group. This difference is statistically significant at the 5% level (paired t- value: 2.34). This result supports the earlier results for less serious delinquency on revolving credit. In the case of North Carolina and Oregon border counties, no notable change in the revolving delinquency rate is observed. Similarly, for other forms of consumer credit retail and installment the legislation seems to have little impact on the delinquencies. In summary, the results for North Carolina seem reasonable, with no notable effect of legislation on any form of delinquency. Oregon law has no effect on serious delinquencies in its border counties. For Georgia, the law has an increasing effect on delinquencies on revolving and retail credit. The large negative result for installment loans seems anomalous, since it is inconsistent with other results that show small positive or statistically insignificant effects for each type of credit. Next, we perform the pooled-ordinary Least Squares (OLS) on the sample involving data from both treatment and control group counties. Specifically, the following multivariate model is used to assess the effect of legislation prohibiting payday lending on county-level delinquency rates. Delinquency i,t = β 0 + β 1 TreatD i + β 2 TimeD t + δ Effect i,t + γ Z i,t + ε i,t, (1) where the dependent variable Delinquency measures county-level delinquency on 12

15 revolving/retail/installment credit for a given quarter. In one analysis, we use less serious delinquencies as the measure of Delinquency, and in the other analysis, we use serious delinquency as a measure of Delinquency. The dummy variable TreatD takes on a value of one if a given county i is in the treatment group that is, the border counties of GA, NC, and OR and a value of zero if a given county is one of the matched contiguous counties. The dummy variable TimeD takes on a value of one for the quarter t belonging to the post-event period and zero for the pre-event quarters. The vector Z includes other variables that might influence delinquency rates. The proxies measuring county-level economic conditions, demographics, and race/ethnicity are control variables. The continuous variables Unemp, 25to44, Black, and Hisp are the ratio of people seeking a job in the labor force, percentage of the population in the age group of 25 to 44, percentage of the population of African American origin, and percentage of the population of Hispanic ethnicity, respectively. The county-level data for demographics and race/ethnicity are from the U.S. Census Bureau, and the data on unemployment rate are from the Bureau of Labor Statistics. The black and Hispanic population data are included to account for differences in payment performance of these groups, which reflect differences in levels and variability of wealth and income of black and Hispanic borrowers relative to non-hispanic white and Asian borrowers (Avery, Brevoort, and Canner, 2010). The variable of interest is Effect, which is the interaction term of the dummy variables TreatD and TimeD. It has a value of one if a given county is in the treatment group (border counties of GA, NC, and OR) and the given time period (quarter) is in the post-event period, and zero otherwise. The positive value of its coefficient δ indicates that the level of delinquency on a given form of credit has increased in the counties of states that restrict payday legislation 13

16 (treatment group) after the enactment of the legislation compared to a similar county of states without such legislation. In other words, a positive sign of the coefficient on Effect suggests net increase in delinquency, thus the adverse effects of legislation that prohibits or restricts payday lending. In all regressions, standard errors are clustered at the county level to control for possible heteroscedasticity and/or autocorrelation. The results of multivariate analysis to assess the impact of legislation on less serious delinquencies, in samples from restricted border and unrestricted contiguous counties, are reported in Table 2. Dependent variables in regressions 1-4, 5-8, and 9-12 are less serious delinquencies on revolving, retail, and installment (non-mortgage) credit, respectively. In the analysis of revolving credit, the first two regressions use only the 46 Georgia border counties and their neighboring contiguous counties. We analyze each of these counties for eight quarters, four each for prior and posterior periods. In addition, we also analyze their 46 matching counties. Therefore, the total sample size is 46 x 2 x 8 = 736 observations for the first two regressions. Regressions 3 and 4 use the samples of Georgia and North Carolina border and matching contiguous counties. As mentioned previously, we could not use the subsample of Oregon border and matching counties, since less serious delinquency rates are not available. Also, we did not have less serious delinquency data for NC border and matching counties for the first quarter of Therefore, the total sample size is 46 x 2 x x 2 x 7 = 1,282 observations. In these regressions, to control for state effects, we use dummy variable GA_D, which takes on a value of one for those 736 observations involving Georgia and zero for the remaining 546 observations. In all the regressions, the standard errors are clustered at the county level to account for possible heteroscedasticity and/or autocorrelation. The first specification is the base specification, involving only the dummy variables TreatD and TimeD, and their interaction term Effect. The 14

17 second regression is an all-inclusive specification that includes the control variables that measure economic conditions, demographics, and race/ethnicity. As reported in the first regression of Table 2, the coefficient on Effect of suggests that the net increase in the average value of less serious revolving delinquencies in Georgia s border counties following payday loan legislation is of the magnitude 0.2%. (About 3% of Georgia consumers had less serious delinquencies in a quarter.) This change is statistically significant at the 5% level (robust t- value: 2.0). This result is similar to the earlier one reported in Panel A of Table 1. As shown in the second regression, even after accounting for other factors that might influence delinquencies, Georgia border counties experienced an increase in less serious delinquencies on revolving credit after the 2004 passage of legislation prohibiting payday lending. As reported in regressions 3 and 4, the coefficient on Effect for less serious revolving delinquencies for the border counties of Georgia and North Carolina is still positive, but not statistically significantly different from zero. The coefficient on Effect in regressions 5-12 of Table 2 suggest that the payday legislation did not have a significant effect on the level of less serious delinquencies for retail and non-mortgage installment loans in Georgia and North Carolina. In Table 3, we report the results of analyses using serious delinquencies for the three types of credit as the dependent variable. We could perform the analyses on the entire sample of 103 county pairs, resulting in a sample of 1,648 [103 x 2 x 8] observations. As reported in regression 2, the coefficient of on Effect suggests that after accounting for other factors, bans on payday loans caused an increase in revolving delinquencies by 0.31% in the border counties of Georgia; this change is statistically significant at the 10% level (robust t- value: 1.83). As shown in regression 10, the coefficient on Effect suggests that after the ban, the 15

18 level of serious installment delinquencies in border counties in Georgia fell by around 0.7% in the post-event period. (Sixty or more day delinquency rates for Georgia were less than 5% for installment credit.) This estimated effect is statistically significant at the 5% level (robust t- value: 2.24). Otherwise, bans on payday loans appear to have small or statistically insignificant effects on serious delinquencies. Overall, the results shown in Tables 2 and 3 tend to support the findings of univariate analysis as reported in Table 1, suggesting that payday loan bans had small, often statistically insignificant effects on consumer-credit delinquencies. For Georgia, increases in revolving delinquencies were small but statistically significant. The relatively large, statistically significant decline in serious delinquencies for installment loans in Georgia seems perverse: It is difficult to believe that loans averaging $300 to $400 could have such a large effect for 60-day or longer delinquencies but not for day delinquencies on less serious and on serious delinquencies for other types of credit, or for 60- day or longer delinquencies in North Carolina and Oregon. Robustness check: Analysis of restricted hinterland and unrestricted contiguous counties Payday borrowers from regulated (restricted) border counties may cross the state border to obtain payday loans in a neighboring state that allows payday lending. Therefore, for a robustness check, we also perform analyses comparing the delinquencies of regulated hinterland counties with those for nonregulated contiguous border counties of neighboring states. Following Huang (2008), we define a restricted hinterland county as one that is in the same state of the regulated border county. In addition, the hinterland county shares a border with the regulated border county, but not with the nonregulated contiguous county. For example, as shown in 16

19 Figure to Appendix, Grady County is a border county of Georgia, which does not permit payday lending. Grady County shares a border with Leon County of Florida, which permits payday lending. As per our definition, Grady County is a restricted border county and Leon County is an unrestricted contiguous county. Mitchell County is north of Grady County and farther from the border with Florida. We define Mitchell County as a restricted hinterland county. For 28 county pairs of 103 county pairs, we could not find a clean hinterland county as per our definition. In such cases, we select the hinterland county based on our judgment. (Regulated hinterland counties and regulated border and nonregulated contiguous counties are listed in the Appendix.) As shown in Table 4, data from hinterland counties do not support the debt trap hypothesis; payday loan bans had small or statistically insignificant effects on delinquency rates for most types of consumer credit in all three states. As reported in Panel B of Table 4, payday loan bans increased the serious delinquencies for retail credit by 0.31% in 46 border counties of Georgia. This increase is significant at the 10% level (robust t- value: 1.96). The results of Table 5 provide further evidence suggesting that payday bans have not significantly affected delinquency levels in the post-legislation period. Estimated effects of payday loan bans on serious delinquencies in Table 6 are mostly not statistically significant. As reported in regression 7 in Table 6, payday loan bans are statistically significant and positively related to retail delinquencies in the hinterland counties of Georgia, North Carolina, and Oregon. The estimated coefficient in regression 10 for installment credit in Georgia is positive and statistically significant. The estimated coefficient in the pooled regression with state fixed effects, however, is small, negative, and not statistically significant. Again, the relatively large size of the coefficient and the inconsistency with findings for other types of credit and other states experiences suggest that this result may be anomalous. 17

20 Overall, our results suggest that the impact of legislation on the level of financial distress is insignificant and do not support the hypothesis that access to payday loans contributes to the so-called debt trap. Concluding remarks Nearly one fifth of U.S. households use payday loans annually. These loans are intended to cover short-term needs for cash by credit-constrained consumers, but critics of the industry contend that the relatively high cost and single-payment feature are a debt trap for consumers. This paper investigates the effects of payday loan bans in Georgia, North Carolina, and Oregon on delinquencies for three types of consumer credit: revolving, retail, and installment accounts. Payday loan customers are heavy users of consumer credit, and payday loans may be a less costly option for addressing a cash shortfall than making late payments on consumer debts. We consider two forms of delinquencies, less serious delinquencies of days and serious delinquencies of 60 days or more. We identify 103 border counties of Georgia, North Carolina, and Oregon where payday lending was initially allowed without any restriction and later restricted, either through an outright ban or rate ceilings that made payday lending unprofitable. We then compare the delinquencies in these counties with those in contiguous counties of neighboring states where payday loans are permitted. The research design is based on eventstudy methodology, where the event date is the quarter in which the legislation banning payday lending went into effect; neighboring counties provide a control group. The results suggest that the effects of payday loan bans on delinquencies are generally small, positive, and not statistically significant. The results do not support the debt trap hypothesis, but they also suggest that payday loans may not be much help in avoiding late 18

21 payments on consumer debts. Our results are similar to those of Bhutta (2012), who found that payday loans had little effect on credit scores and do not appear, on net, to exacerbate consumers debt problems. This contradicts the debt trap hypothesis that underlies many current regulatory initiatives on payday lending. The analyses of objective data on more proximate difficulties due to cash shortfalls would be a possible extension of our study. Morgan and Strain s (2008) analyses of overdrafts and complaints are examples of such difficulties. One to 29 day delinquencies, late utility payments, and service disconnections would be other difficulties that could be analyzed using our approach. 19

22 References: Avery, R. B., Brevoort, K. P., & Canner, G. B. (2010). Does Credit Scoring Produce a Disparate Impact? Finance and Economics Discussion Series. Divisions of Research and Statistics and Monetary Affairs. Federal Reserve Board. Washington. Bhutta, N. (2012). Payday Credit Access and Household Financial Health: Evidence from Consumer Credit Records. Available at SSRN: Caskey, J. (2010). Payday Lending: New Research and the Big Question. FRB of Philadelphia Working Paper No Federal Reserve Board of Philadelphia. Philadelphia. Consumer Financial Protection Bureau. (2013). Payday Loans and Deposit Advance Products: A White Paper of Initial Data Findings. Edmiston, K. D. (2011). Could Restrictions on Payday Lending Hurt Consumers? Economic Review - Federal Reserve Bank of Kansas City, First Quarter, Elliehausen, G. (1999). Trends in Regional Bankcard Delinquencies. Monthly Statements: A profile of Consumer Borrowing & Payment Behavior, TransUnion LLC and Georgetown University - Credit Research Center,(June) 6-7. Elliehausen, G. (2009). An analysis of consumers' use of payday loans. Financial Services Research Program, The George Washington University. Washington, DC. Ernst & Young. (2009). The Cost of Providing Payday Loans in a US Multiline Operator Environment. In Study prepared on behalf of the Financial Service Centers of America. Flannery, M., & Samolyk, K. (2005). Payday Lending: Do the Costs Justify the Price? FDIC Center for Financial Research Working Paper. Washington, DC. Fox, J. A., & Mierzwinski, E. (2001). RENT-A-BANK PAYDAY LENDING How Banks Help Payday Lenders Evade State Consumer Protections: Consumer Federation of America (CFA) and the U. S. Public Interest Research Group (U.S. PIRG). Hecht, J. (2013). Forging Ahead: Growth, Opportunity and the Direction of the Alternative Financial Services Sector, Stephens Inc. Huang, R. R. (2008). Evaluating the real effect of bank branching deregulation: Comparing contiguous counties across US state borders. Journal of Financial Economics, 87, Mann, R. (2013). Assessing the Optimism of Payday Loan Borrowers. Center for Law and Economic Studies. Columbia University School of Law. Melzer, B. T. (2011). The real costs of credit access: Evidence from the payday lending market. The Quarterly Journal of Economics, 126, Melzer, B. T., & Morgan, D. P. (2009). Competition and Adverse Selection in the Small-Dollar Loan Market: Overdraft versus Payday Credit. Federal Reserve Bank of New York Staff Reports. Working paper. New York. Morgan, D. P., & Strain, M. R. (2008). Payday Holiday: How Households Fare after Payday Credit Bans. Federal Reserve Bank of New York Staff Reports. Federal Reserve Bank of New York. New York. Morse, A. (2011). Payday lenders: Heroes or villains? Journal of Financial Economics, 102, National Commission on Consumer Finance. (1972). Consumer Credit in the United States. Washington: US Government Printing Office. Pew Charitable Trusts. (2013). How Borrowers Choose and Repay Payday Loans. Phillips, L. W., & Calder, B. J. (1979). Evaluating consumer protection programs: Part I. Weak 20

23 but commonly used research designs. Journal of Consumer Affairs, 13, Phillips, L. W., & Calder, B. J. (1980). Evaluating consumer protection laws: II. Promising methods. Journal of Consumer Affairs, 14, Skiba, P. M., & Tobacman, J. (2011). Do Payday Loans Cause Bankruptcy? Vanderbilt University and University of Pennsylvania, Working paper. Zinman, J. (2010). Restricting consumer credit access: Household survey evidence on effects around the Oregon rate cap. Journal of Banking & Finance, 34,

24 Table 1: Univariate analysis for border and contiguous counties Panel A: Less-serious delinquencies (payments due 30 to 59 days) (1) Revolving Credit Regulated border (treatment) counties 22 Nonregulated contiguous (control) counties Difference-in- Difference N pre period post period pre period post period Value t-stat Georgia % 2.714% 3.221% 2.637% 0.202% 2.16** North Carolina % 2.297% 2.421% 2.383% % Oregon % N/A 1.80% N/A - - Total w/o OR % 2.523% 2.85% 2.520% 0.092% 1.30 (2) Retail Credit Georgia % 1.321% 1.842% 1.358% 0.109% 1.31 North Carolina % 1.226% 1.342% 1.273% % Oregon % N/A 0.82% N/A - - Total w/o OR % 1.277% 1.613% 1.319% 0.037% 0.62 (3) Installment (non-mortgage) Credit Georgia % 4.286% 4.251% 3.765% % North Carolina % 3.534% 3.445% 3.539% % Oregon % N/A 1.92% N/A - - Total w/o OR % 3.941% 3.882% 3.661% % Panel B: Serious delinquencies (payments due above 60 days) (1) Revolving Credit Regulated border (treatment) counties Nonregulated contiguous (control) counties Difference-in- Difference N pre period post period pre period post period value t-stat Georgia % 3.042% 4.833% 2.999% 0.306% 2.34** North Carolina % 2.323% 3.189% 2.560% 0.009% 0.09 Oregon % 1.891% 2.026% 1.829% % Total % 2.568% 3.720% 2.628% 0.139% 1.88* (2) Retail Credit Georgia % 1.956% 2.764% 1.797% 0.113% 0.86 North Carolina % 0.957% 2.098% 1.043% 0.176% 1.54 Oregon % 0.498% 0.930% 0.433% 0.004% 0.04 Total % 1.323% 2.191% 1.273% 0.118% 1.57 (3) Installment (non-mortgage) Credit Georgia % 6.326% 7.715% 6.203% % North Carolina % 3.621% 8.082% 4.363% 0.013% 0.03 Oregon % 2.356% 2.643% 2.488% % Total % 4.608% 6.968% 4.857% % Notes: N is the number of counties. N/A denotes that data are not available to us. The t-statistics are based on paired t test for the difference in the change in delinquency rate for the treatment group and that for the control group. The symbols * and ** indicate the statistical significance at the 10% and 5% levels, respectively.

25 Table 2: Analysis on less serious delinquencies (30 to 59 days due) for restricted border and unrestricted contiguous counties Revolving Credit Retail Credit Installment Credit GA only GA and NC GA only GA and NC GA only GA and NC (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) TreatD * 0.007** 0.004* 0.005*** [-0.58] [-1.29] [-0.66] [-0.42] [-1.18] [-1.64] [-0.95] [-0.22] [1.87] [2.06] [1.74] [2.73] TimeD *** *** *** *** *** *** *** *** *** *** *** ** [-8.44] [-7.76] [-5.44] [-4.88] [-7.64] [-7.31] [-5.61] [-5.02] [-4.28] [-4.38] [-3.20] [-2.06] Effect ** ** [2.00] [2.12] [1.24] [1.00] [1.15] [1.07] [0.51] [0.29] [-0.72] [-0.65] [-0.54] [-0.93] Unemp ** [-0.63] [0.75] [0.15] [1.32] [-0.73] [2.50] 25to *** 0.038** 0.067*** 0.041*** [3.22] [2.38] [5.02] [4.16] [-0.91] [-0.44] Black 0.031*** 0.025*** 0.015*** 0.013*** 0.042*** 0.030*** [9.01] [10.05] [6.79] [7.71] [5.24] [5.18] Hisp * [1.62] [1.73] [0.33] [0.62] [0.41] [0.43] GA_D 0.006*** 0.005*** 0.003*** 0.002*** 0.009*** 0.008*** [5.22] [6.74] [4.17] [4.84] [4.62] [6.01] Const *** *** 0.008* 0.018*** *** *** 0.054** 0.034*** 0.025* [17.91] [0.93] [26.08] [1.74] [18.39] [-0.90] [22.56] [-0.42] [25.30] [2.54] [22.65] [1.85] N R Notes: The dependent variable is the county-level less serious delinquencies which is defined as the ratio of the number of consumers who are delinquent (30 to 59 days due) on their given credit account to the total number of consumers for that account. The sample consists of 46 and 39 border counties of GA and NC, respectively, and their matching contiguous counties from neighboring states that have unrestricted payday access. The list of county pairs is provided in the Appendix. The dummy variable TreatD equals one for a border county of GA and NC (treatment group), and zero for the control group counties. For each county, TimeD takes on a value of one for the posterior period, and zero for the prior period. The posterior period consists of four quarters immediately after the effective date. The prior period includes four preceding quarters starting from first year prior to the effective quarter. The effective quarter in the quarter in which the state legislation restricting payday lending went into the effect. The variable of interest is Effect. It is the interaction dummy variable of TreatD and TimeD. It takes on a value of one for a posterior period of a treatment county, and zero otherwise. The control variables 25to44, Black, Hisp, and Unemp are continuous and measure a county s percentage of the population in the age group 25 to 44, percentage of the population of African-American origin, percentage of the population of Hispanic ethnicity, and the percentage of the labor force seeking a job, respectively. GA_D is a dummy variable that takes on a value of one for a county-pair if its treatment county is a GA border county, and it takes a value of zero for a county-pair if the treatment county is a NC border county. N is the number of countyquarter observations. In all the regressions, standard errors are clustered at the county-level to control for heteroscedasticity and/or autocorrelation. The symbols *, **, and *** indicate the statistical significance at the 10%, 5%, and 1% levels, respectively. 23

26 Table 3: Analysis on serious delinquencies (60+ days due) for border and contiguous counties Revolving Credit Retail Credit Installment Credit GA only GA, NC, and OR GA only GA, NC, and OR GA only GA, NC, and OR (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) TreatD * [-0.89] [-1.46] [-1.23] [-0.79] [0.18] [0.54] [-0.49] [0.47] [0.68] [1.78] [-0.18] [0.96] TimeD *** *** *** *** *** *** *** *** *** *** *** *** [-14.01] [-13.69] [-10.34] [-9.90] [-7.94] [-7.62] [-12.21] [-11.83] [-8.42] [-7.29] [-7.94] [-7.57] Effect * * ** [1.82] [1.83] [1.06] [1.08] [0.71] [0.41] [1.20] [1.20] [-1.18] [-2.24] [-0.46] [-0.72] Unemp ** 0.425** 0.357*** [-0.03] [1.38] [0.97] [2.30] [2.04] [4.56] 25to *** 0.061*** 0.113*** 0.059*** [5.63] [3.84] [5.00] [4.65] [0.72] [1.43] Black 0.042*** 0.033*** 0.034*** 0.029*** 0.103*** 0.103*** [10.97] [11.17] [10.88] [12.22] [6.43] [10.23] Hisp * ** [-0.68] [1.18] [1.22] [1.77] [1.63] [2.22] GA_D 0.019*** 0.010*** 0.016*** 0.009*** 0.046*** 0.028*** [14.96] [6.93] [15.42] [8.12] [13.21] [9.19] NC_D 0.008*** *** *** 0.016*** [7.93] [0.12] [8.88] [1.04] [10.19] [5.91] Const *** *** 0.007* 0.028*** ** 0.012*** ** 0.077*** *** [19.72] [0.46] [22.77] [1.74] [14.14] [-2.13] [13.36] [-2.17] [13.77] [0.36] [12.08] [-0.73] N R Notes: The dependent variable is the county-level serious delinquency which is defined as the ratio of the number of consumers who are delinquent (60 + days due) on a given credit account to the total number of consumers for that credit account. The sample is of 103 county pairs that includes the border counties of GA, NC, and OR and their contiguous counties from neighboring state with unrestricted payday access. The list of county pairs is provided in the Appendix. The dummy variable TreatD equals one for a border county of GA and NC (treatment group), and zero for the control group counties. For each county, TimeD takes on a value of one for the posterior period, and zero for the prior period. The posterior period consists of four quarters immediately after the effective date. The prior period includes four preceding quarters starting from first year prior to the effective quarter. The effective quarter in the quarter in which the state legislation restricting payday lending went into the effect. The variable of interest is Effect. It is the interaction dummy variable of TreatD and TimeD. It takes on a value of one for a posterior period of a treatment county, and zero otherwise. The control variables 25to44, Black, Hisp, and Unemp are continuous and measure a county s percentage of the population in the age group 25 to 44, percentage of the population of African-American origin, percentage of the population of Hispanic ethnicity, and the percentage of the labor force seeking a job, respectively. GA_D / NC_D is a dummy variable that takes on a value of one for a county-pair if its treatment county is a GA / NC border county, and it takes a value of zero otherwise. N is the number of county-quarter observations. In all the regressions, standard errors are clustered at the county-level to control for heteroscedasticity and/or autocorrelation. The symbols *, **, and *** indicate the statistical significance at the 10%, 5%, and 1% levels, respectively. 24

27 Table 4: Robustness Check: Univariate analysis for hinterland and contiguous counties Panel A: Less-serious delinquencies (payments due 30 to 59 days) (1) Revolving Credit Regulated hinterland (treatment) counties Nonregulated contiguous (control) counties Difference-in- Difference N pre period post period pre period post period value t-stat Georgia % 2.698% 3.221% 2.637% % North Carolina % 2.277% 2.421% 2.383% 0.024% 0.26 Oregon % N/A 1.80% N/A - - Total w/o OR % 2.505% 2.85% 2.520% % (2) Retail Credit Georgia % 1.444% 1.842% 1.358% % North Carolina % 1.171% 1.342% 1.273% % Oregon % N/A 0.82% N/A - - Total w/o OR % 1.319% 1.613% 1.319% % (3) Installment (non-mortgage) Credit Georgia % 4.303% 4.251% 3.765% % North Carolina % 3.477% 3.445% 3.539% 0.067% 0.86 Oregon % N/A 1.92% N/A - - Total w/o OR % 3.924% 3.882% 3.661% % Panel B: Serious delinquencies (payments due above 60 days) (1) Revolving Credit Regulated hinterland (treatment) counties 25 Nonregulated contiguous (control) counties Difference-in- Difference N pre period post period pre period post period value t-stat Georgia % 3.124% 4.833% 2.999% 0.122% 0.82 North Carolina % 2.359% 3.189% 2.560% 0.108% 1.19 Oregon % 1.887% 2.026% 1.829% 0.008% 0.10 Total % 2.618% 3.720% 2.628% 0.097% 1.29 (2) Retail Credit Georgia % 2.070% 2.764% 1.797% 0.306% 1.96* North Carolina % 0.938% 2.098% 1.043% 0.128% 1.13 Oregon % 0.405% 0.930% 0.433% % Total % 1.350% 2.191% 1.273% 0.180% 2.14** (3) Installment (non-mortgage) Credit Georgia % 5.807% 7.715% 6.203% % North Carolina % 3.824% 8.082% 4.363% % Oregon % 2.834% 2.643% 2.488% 0.432% 2.64** Total % 4.536% 6.968% 4.857% % Notes: N is the number of counties. N/A denotes that data are not available to us. The t-statistics are based on paired t test for the difference in the change in delinquency rate for the treatment group and that for the control group. The symbols * and ** indicate the statistical significance at the 10% and 5% levels, respectively.

28 Table 5: Robustness checks: Analysis on less-serious delinquencies (30 to 59 days due) for hinterland and contiguous counties Revolving Credit Retail Credit Installment Credit GA only GA and NC GA only GA and NC GA only GA and NC (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) TreatD ** 0.006*** 0.003* 0.004*** [0.42] [0.08] [-0.05] [0.05] [0.50] [0.54] [0.22] [0.67] [2.45] [2.67] [1.71] [2.84] TimeD *** *** *** *** *** *** *** *** *** *** *** [-8.44] [-7.71] [-5.54] [-4.63] [-7.64] [-7.09] [-5.77] [-4.83] [-4.28] [-4.12] [-3.25] [-1.44] Effect [-0.21] [-0.18] [0.03] [-0.14] [-0.01] [-0.11] [-0.10] [-0.32] [-0.50] [-0.70] [0.18] [-0.64] Unemp *** [-0.22] [1.14] [0.41] [1.43] [0.78] [4.37] 25to * [0.82] [1.79] [-0.05] [0.94] [0.17] [1.03] Black 0.031*** 0.024*** 0.019*** 0.016*** 0.030*** 0.019*** [8.20] [8.76] [5.96] [7.08] [5.41] [4.58] Hisp [-0.36] [-0.43] [-0.32] [-0.99] [0.20] [0.04] GA_d 0.006*** 0.006*** 0.004*** 0.004*** 0.009*** 0.009*** [6.23] [7.73] [5.39] [6.12] [5.85] [7.56] Const *** 0.020** 0.025*** 0.010* 0.018*** 0.013* 0.014*** *** 0.030** 0.034*** [17.91] [2.37] [26.28] [1.84] [18.39] [1.74] [21.44] [0.94] [25.30] [2.15] [24.52] [0.80] N R Notes: The dependent variable is the county-level less serious delinquencies which is defined as the ratio of the number of consumers who are delinquent (30 to 59 days due) on their given credit account to the total number of consumers for that account. The sample consists of 46 and 39 hinterland counties of GA and NC, respectively, and the contiguous counties from neighboring states that have unrestricted payday access. The list of county pairs is provided in the Appendix. The dummy variable TreatD equals one for a hinterland county of GA and NC (treatment group), and zero for the control group counties. For each county, TimeD takes on a value of one for the posterior period, and zero for the prior period. The posterior period consists of four quarters immediately after the effective date. The prior period includes four preceding quarters starting from first year prior to the effective quarter. The effective quarter in the quarter in which the state legislation restricting payday lending went into the effect. The variable of interest is Effect. It is the interaction dummy variable of TreatD and TimeD. It takes on a value of one for a posterior period of a treatment county, and zero otherwise. The control variables 25to44, Black, Hisp, and Unemp are continuous and measure a county s percentage of the population in the age group 25 to 44, percentage of the population of African-American origin, percentage of the population of Hispanic ethnicity, and the percentage of the labor force seeking a job, respectively. GA_D is a dummy variable that takes on a value of one for a county-pair if its treatment county is a GA hinterland county, and it takes a value of zero for a county-pair if the treatment county is a NC hinterland county. N is the number of county-quarter observations. In all the regressions, standard errors are clustered at the county-level to control for heteroscedasticity and/or autocorrelation. The symbols *, **, and *** indicate the statistical significance at the 10%, 5%, and 1% levels, respectively. 26

29 Table 6: Robustness checks: Analysis on serious delinquencies (60+ days due) for hinterland and contiguous counties Revolving Credit Retail Credit Installment Credit GA only GA, NC, and OR GA only GA, NC, and OR GA only GA, NC, and OR (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) TreatD [0.12] [-0.44] [-0.55] [-0.86] [-0.08] [-0.46] [-0.69] [-0.93] [0.02] [0.71] [-0.44] [0.34] TimeD *** *** *** *** *** *** *** *** *** *** *** *** [-14.01] [-14.14] [-10.34] [-9.89] [-7.94] [-8.15] [-12.21] [-11.92] [-8.42] [-7.71] [-7.94] [-7.63] Effect * * [0.57] [0.70] [0.65] [0.58] [1.60] [1.53] [1.73] [1.56] [-0.96] [-1.89] [-0.21] [-0.65] Unemp * 0.389* 0.399*** [-0.89] [1.13] [-0.32] [1.67] [1.99] [4.74] 25to * ** [-0.42] [1.52] [-1.87] [0.41] [-2.40] [-1.24] Black 0.045*** 0.034*** 0.034*** 0.030*** 0.093*** 0.094*** [11.40] [11.13] [8.46] [10.17] [6.50] [9.26] Hisp * ** 0.050** [0.75] [1.08] [1.75] [1.40] [2.05] [2.28] GA_D 0.020*** 0.012*** 0.017*** 0.011*** 0.043*** 0.029*** [15.76] [8.15] [15.72] [6.86] [12.95] [9.02] NC_D 0.008*** *** 0.003** 0.033*** 0.020*** [9.26] [0.63] [10.75] [2.29] [11.46] [6.08] Const *** 0.044*** 0.025*** 0.016*** 0.028*** 0.036*** 0.011*** *** 0.082*** 0.038*** [19.72] [4.36] [23.55] [3.36] [14.14] [3.54] [13.38] [1.11] [13.77] [3.02] [13.00] [1.56] N R Notes: The dependent variable is the county-level serious delinquency rate which is defined as the ratio of the number of consumers who are delinquent (60 + days due) on a given credit account to the total number of consumers for that credit account. The sample is of 103 county pairs that includes the hinterland counties of GA, NC, and OR and their contiguous counties from neighboring state with unrestricted payday access. The list of county pairs is provided in the Appendix. The dummy variable TreatD equals one for a hinterland county of GA and NC (treatment group), and zero for the control group counties. For each county, TimeD takes on a value of one for the posterior period, and zero for the prior period. The posterior period consists of four quarters immediately after the effective date. The prior period includes four preceding quarters starting from first year prior to the effective quarter. The effective quarter in the quarter in which the state legislation restricting payday lending went into the effect. The variable of interest is Effect. It is the interaction dummy variable of TreatD and TimeD. It takes on a value of one for a posterior period of a treatment county, and zero otherwise. The control variables 25to44, Black, Hisp, and Unemp are continuous and measure a county s percentage of the population in the age group 25 to 44, percentage of the population of African-American origin, percentage of the population of Hispanic ethnicity, and the percentage of the labor force seeking a job, respectively. GA_D / NC_D is a dummy variable that takes on a value of one for a county-pair if its treatment county is a GA / NC hinterland county, and it takes a value of zero otherwise. N is the number of county-quarter observations. In all the regressions, standard errors are clustered at the county-level to control for heteroscedasticity and/or autocorrelation. The symbols *, **, and *** indicate the statistical significance at the 10%, 5%, and 1% levels, respectively. 27

30 Appendix : List of county-pairs Countypair # Restricted border county, State Unrestricted contiguous county, State Restricted hinterland county, State 1 Seminole, GA Jackson, FL Miller, GA Yes 2 Decatur, GA Gadsden, FL Baker, GA Yes 3 Grady, GA Leon, FL Mitchell, GA Yes 4 Thomas, GA Jefferson, FL Colquitt, GA Yes 5 Brooks, GA Madison, FL Cook, GA Yes 6 Lowndes, GA Madison, FL Berrien, GA Yes 7 Echols, GA Hamilton, FL Lanier, GA Yes 8 Clinch, GA Columbia, FL Atkinson, GA Yes 9 Ware, GA Baker, FL Pierce, GA Yes 10 Charlton, GA Nassau, FL Brantley, GA Yes 11 Camden, GA Nassau, FL Glynn, GA Yes 12 Chatham, GA Jasper, SC Bryan, GA Yes 13 Effingham, GA Jasper, SC Bulloch, GA Yes 14 Screven, GA Allendale, SC Jenkins, GA Yes 15 Burke, GA Barnwell, SC Emanuel, GA Yes 16 Richmond, GA Aiken, SC Jefferson, GA Yes 17 Columbia, GA Edgefield, SC McDuffie, GA Yes 18 Lincoln, GA McCormick, SC Wilkes, GA Yes 19 Elbert, GA Abbeville, SC Oglethorpe, GA Yes 20 Hart, GA Anderson, SC Madison, GA Yes 21 Franklin, GA Oconee, SC Jackson, GA No 22 Stephens, GA Oconee, SC Banks, GA Yes 23 Habersham, GA Oconee, SC Hall, GA Yes 24 Rabun, GA Macon, NC Clarke, GA No 25 Towns, GA Clay, NC White, GA Yes 26 Union, GA Cherokee, NC Lumpkin, GA Yes 27 Fannin, GA Polk, TN Dawson, GA Yes 28 Murray, GA Polk, TN Gilmer, GA Yes 29 Whitfield, GA Bradley, TN Gordon, GA Yes 30 Catoosa, GA Hamilton, TN Pickens, GA No 31 Walker, GA Hamilton, TN Cherokee, GA No 32 Dade, GA Dekalb, AL Forsyth, GA No 33 Chattooga, GA Cherokee, AL DeKalb, GA No 34 Floyd, GA Cherokee, AL Bartow, GA Yes 35 Polk, GA Cherokee, AL Paulding, GA Yes 36 Haralson, GA Cleburne, AL Cobb, GA No 37 Carroll, GA Cleburne, AL Douglas, GA Yes 38 Heard, GA Randolph, AL Coweta, GA Yes 39 Troup, GA Chambers, AL Meriwether, GA Yes 40 Harris, GA Lee, AL Talbot, GA Yes 41 Muscogee, GA Lee, AL Taylor, GA No 42 Chattahoochee, GA Russell, AL Marion, GA Yes 43 Stewart, GA Russell, AL Webster, GA Yes 44 Quitman, GA Barbour, AL Randolph, GA Yes 45 Clay, GA Henry, AL Calhoun, GA Yes 46 Early, GA Houston, AL Dougherty, GA No 47 Jackson, NC Oconee, SC Rowan, NC No 48 Transylvania, NC Pickens, SC Alexander, NC No 49 Henderson, NC Greenville, SC Buncombe, NC Yes 50 Polk, NC Spartanburg, SC Catawba, NC No 51 Rutherford, NC Spartanburg, SC McDowell, NC Yes 52 Cleveland, NC Cherokee, SC Burke, NC Yes 28 Clean hinterland?

31 53 Gaston, NC York, SC Lincoln, NC Yes 54 Mecklenburg, NC York, SC Iredell, NC Yes 55 Union, NC Lancaster, SC Cabarrus, NC Yes 56 Anson, NC Chesterfield, SC Stanly, NC Yes 57 Richmond, NC Marlboro, SC Montgomery, NC Yes 58 Scotland, NC Marlboro, SC Hoke, NC Yes 59 Robeson, NC Dillon, SC Cumberland, NC Yes 60 Columbus, NC Horry, SC Bladen, NC Yes 61 Brunswick, NC Horry, SC Pender, NC Yes 62 Currituck, NC Virginia Beach City, VA Washington, NC No 63 Camden, NC Chesapeake City, VA Pasquotank, NC Yes 64 Gates, NC Suffolk City, VA Perquimans, NC Yes 65 Hertford, NC Southampton, VA Chowan, NC Yes 66 Northampton, NC Greensville, VA Bertie, NC Yes 67 Warren, NC Brunswick, VA Halifax, NC Yes 68 Vance, NC Mecklenburg, VA Franklin, NC Yes 69 Granville, NC Mecklenburg, VA Wake, NC Yes 70 Person, NC Halifax, VA Durham, NC Yes 71 Caswell, NC Danville City, VA Alamance, NC Yes 72 Rockingham, NC Henry, VA Guilford, NC Yes 73 Stokes, NC Patrick, VA Forsyth, NC Yes 74 Surry, NC Carroll, VA Yadkin, NC Yes 75 Alleghany, NC Grayson, VA Wilkes, NC Yes 76 Ashe, NC Grayson, VA Orange, NC No 77 Watauga, NC Johnson, TN Caldwell, NC Yes 78 Avery, NC Carter, TN Davie, NC No 79 Mitchell, NC Unicoi, TN Davidson, NC No 80 Yancey, NC Unicoi, TN Randolph, NC No 81 Madison, NC Madison, TN Chatham, NC No 82 Haywood, NC Cocke, TN Moore, NC No 83 Swain, NC Sevier, TN Lee, NC No 84 Graham, NC Monroe, TN Harnett, NC No 85 Cherokee, NC Polk, TN Johnston, NC No 86 Clatsop, OR Wahkiakum, WA Tillamook, OR Yes 87 Columbia, OR Cowlitz, WA Washington, OR Yes 88 Multnomah, OR Clark, WA Clackamas, OR Yes 89 Hood River, OR Skamania, WA Yamhill, OR No 90 Wasco, OR Klickitat, WA Jefferson, OR Yes 91 Sherman, OR Klickitat, WA Marion, OR No 92 Gilliam, OR Klickitat, WA Wheeler, OR Yes 93 Morrow, OR Benton, WA Grant, OR Yes 94 Umatilla, OR Walla walla, WA Union, OR Yes 95 Wallowa, OR Idaho, ID Linn, OR No 96 Baker, OR Washington, ID Benton, OR No 97 Malheur, OR Owyhee, ID Polk, OR No 98 Harney, OR Humboldt, NV Crook, OR Yes 99 Lake, OR Washoe, NV Deschutes, OR Yes 100 Klamath, OR Modoc, CA Lane, OR Yes 101 Jackson, OR Siskiyou, CA Douglas, OR Yes 102 Josephine, OR Siskiyou, CA Lincoln, OR No 103 Curry, OR Del Norte, CA Coos, OR Yes 29

32 Figure to Appendix: Restricted hinterland county, GA Restricted border county, GA Unrestricted contiguous county, FL 30