A Balanced View of Storefront Payday Borrowing Patterns Results From a Longitudinal Random Sample Over 4.5 Years

Transcription

1 Report 7-C A Balanced View of Storefront Payday Borrowing Patterns Results From a Longitudinal Random Sample Over 4.5 Years

2 A Balanced View of Storefront Payday Borrowing Patterns Results From a Longitudinal Random Sample Over 4.5 Years By: Rick Hackett 1 March 28, 2016 Introduction This is our final paper on storefront borrower use patterns and the proposed CFPB rule. Readers who have not done so should first review the synopsis of our Report 7-B, Searching for Harm in Storefront Payday Lending, in order to grasp the context of this final report. In prior work, we have attempted to grapple with the effects of sampling bias that occurs in any static pool analysis of payday loan use patterns - the oversampling of heavy users, as well as the inadequacy of a one-year sample used by the CFPB - to better understand the different patterns of use by different types of users. This report uses a statistically valid random sample of a constant pool of 1,000 users over 3.5 years, with a full year run-off period after the sampling period to avoid truncation effect. It answers the question: What does payday loan use look like if we observe a random sample of the constantly-evolving group of users over 4.5 years? Once we built our longitudinal sample group, we asked the same questions we answered in Report 7-B: How many loan sequences do borrowers use? How long are those sequences? What s the longest sequence per borrower? How long is a borrower s market life cycle (the time from first loan to last loan across all lenders)? We ask these questions in the context of the CFPB s hypothesis that payday borrowers are harmed by paying more in fees than they borrowed. At the median fee of $15 per $100 per pay period, this trigger would be hit by a sequence of more than six loans. 1. Rick Hackett is a Special Policy Consultant to nonprime101, and was formerly the Assistant Director for Installment and Liquidity Lending Markets at the CFPB. This report is based on statistical analysis by Amir Fekrazad, a PhD candidate at the University of Texas - Austin, and Heather Lamoureux, Research Associate for nonprime101. 1

3 Our method of building loan sequences, based on the precise pay period for each borrower, is explained in Searching for Harm in Storefront Payday Lending. Notably, we used multi-lender sequences for this analysis. That is, we counted as in sequence any loans that were taken out with a different lender if they were taken out within the same pay cycle as the last loan payoff with a prior lender. This will present the worst case data for debt trap behaviors, more negative than anything the CFPB can detect. We also added some new questions: If a borrower has a worst case that is a long sequence that may evidence CFPB harm, are the other sequences for that borrower likely to also suggest CFPB harm? Or are they shorter, less troublesome loan use patterns? What are the patterns of time off (i.e., periods out of debt) for borrowers who use loan sequences with varying intensity and duration? For borrowers who evidence shorter loan sequences and shorter overall life cycles in the product, what role does loan default play in their behavior? As explained in the section on building the sample, we think this longitudinal test group is more representative of the wide range of experiences in the storefront payday product - much more so than the static pool analyses that the CFPB has published. It also allows us to test precisely the predictions we made in Report 7-B on the relative size, over time, of the populations of heavier users versus lighter users of the product. We can now say precisely what the use pattern of every borrower in the sample was. Summary Results 1. Over a 3.5 year period, 1,211 new borrowers were added to the original 1,000 in order to maintain a constant population of 1,000 active borrowers per month. From the original 1,000, 302 persisted for the entire 3.5 years. borrowers coming into the product are the largest population in a longitudinal sample of payday borrowers. 2. New, replacement borrowers use the product much less intensely (fewer, shorter sequences of loans and shorter total time in product) than either the 302 persistent borrowers or the 698 replaced borrowers who were in the original sample (but ended use during 3.5 years). 3. The mean (average) sequence duration for replacement borrowers was five loans, well short of the CFPB definition of harm, in the sense of paying more in fees than the amount borrowed. For replaced borrowers, the mean was 7.66 loans and for persistent borrowers the mean was 8.41 loans. 2

4 4. For replacement borrowers, nearly 80 percent of loan sequences did not meet the CFPB definition of harm. For the other two groups, more than 70 percent did not meet that definition. 5. Looking at the worst case per borrower (the longest loan sequence over 4.5 years) 49.8 percent of all borrowers never experienced the CFPB definition of harm. 6. Amongst the half of borrowers who have one sequence that hits the CFPB trigger for harm, there many shorter, less harmful sequences. For example, for the group of borrowers whose maximum sequence is ten loans in a row, the vast majority of other sequences contain fewer than seven loans. Most are two or three loans in duration. In other words, for every bad experience affecting the half of all borrowers who enter a debt trap at least once, there are many not-so-bad experiences. 7. At the mean, replacement borrowers (our largest group) were out of debt 60 percent of the time and in sequence 40 percent of the time. The ratios are reversed for persistent borrowers, and replaced borrowers are in debt about 50 percent of their life cycles. 8. For lighter users of the product (our replacement borrowers), the substantial amount of time between sequences suggests that the overall shorter period of product use (compared to other borrower groups) is not evidence of product aversion. In other words, the replacement borrowers appear to pay off their loans, stay out of debt for a period of time, and then return to the product. 9. borrowers do evidence a higher overall default rate than other borrower groups, which contributes to their shorter average duration of life cycle in the product. Building the Longitudinal Sample We started with a random sample of 1,000 borrowers who had a loan in July We sampled from over a million borrowers in that month. We followed each of those 1,000 borrowers for 3.5 years (though December 31, 2012) to build the sample and then continued to observe behavior for another year (to December 31, 2013). During the 3.5 year sampling period, each time a borrower permanently ended his use of the product, we added a new randomly sampled borrower who had a loan that month, but not in prior months, to maintain a constant population of 1,000 in-life-cycle borrowers each month. A borrower is in life cycle even if he has no loan in a month, if we can observe another loan to that borrower sometime in the future in the sampling period. We stopped replacing borrowers as of December 31, 2012, giving us a full year run-off period to mute truncation effect (i.e., to let those who joined the sample group very late in the sample go through at least a year of use). 3

5 This sampling method limits truncation bias effectively. For the original users who appeared in the product in July 2009 and stayed for a long time, we have a full 4.5 years of data. For replacement users added when an original user permanently drops out, our data shows they have a mean total days in the product of 266 days and a 75th percentile life cycle duration of 398 days, meaning that very few of the replacement borrowers added in the last year of the sample would have continued to use the product beyond the end of the additional year of observation period. Our statistics on use patterns should therefore be very representative. Using this method, we ended up with 2,211 total unique borrowers. That is, the original 1,000 pool required a 1,211 additional sample borrowers over time to maintain a level 1,000 active borrowers per month. Our sample breaks out into three groups: a. Borrowers: 302 unique borrowers had a continuous product life cycle from July 2009 through December At the extremes, this group could have a loan every month or one loan in July 2009 and one loan in December 2012 (the former being much closer to reality). b. Borrowers: 698 borrowers dropped out of the original sample and did not return through the end of the sampling period (December 2012). c. Borrowers: 1,211 new unique borrowers were added to the population at various times over the 3.5 years of the sampling period to replace a drop out who would never return. Some replacements were replacing earlier replacements. Each replacement was randomly selected from the pool of borrowers who were new to the dataset in the month the replacement was added to the sample. 2 We sampled various measures of intensity of product use for each of these groups. As in Report 7-B, our measures of use intensity were number of loan sequences per borrower, duration of loan sequences per borrower, and worst case or longest sequence per borrower. 2. We have intentionally selected replacement borrowers from the pool of borrowers who are new to the sampling period, because any other method would over-weight consumers who are already in a sequence. If we selected from all alternative IDs in the replacement month, we would inherently oversample heavy users, for the reasons discussed in Report 7-B. Once we decided to use some version of new borrower for replacement borrowers in the sample, it made sense to use truly new in the sense of not yet seen in the sampling period. We have seen in Report 7-B that use intensity measures, such as mean sequence duration, differ very little between screening for 30 days off and 180 days off from using a loan. In other words, borrowers who are new to the entire sampling period probably perform like borrowers who are new in the sense of alternative sampling screens, such as not borrowing in the last days. Finally, for reasons explained in Report 7-B, when defining a rule for the indefinite future, there are strong policy reasons to focus regulatory intervention on the experience of truly new borrowers because it is truly new borrowers who will be denied access as the proposed rule plays out over time. 4

6 Number of Sequences per Borrower Table 1 shows the number of borrowers in each group, the number of loan sequences they used, and the number of loans included in those sequences. Table 1 Counts by Group Sample Characteristics Measure Number of Borrowers 1, Number of 13,173 23,807 18,623 Number of Sequences 2,606 3,107 2,213 Table 2 shows the percentage of borrowers in each group who had one loan sequence, two loan sequences, etc. Table 2 Percentage of Borrowers by Number of Sequences in Each of the Three Groups Sequences per Borrower N=2,606 Sequences N=3,107 Sequences N=2,213 Sequences Total

7 Table 2 gives us our first indication of relative intensity of product use by group. As we saw in Report 7-B, any static pool monthly group (here our original 1,000) is over weighted to heavy users already in a loan sequence. Thus, our replacement borrowers (by definition new users of the product) have a very high percentage of single-sequence use (56.81 percent). In contrast, the persistent borrowers are weighted to higher numbers of sequences, with percent having more than 10 sequences. The replaced borrowers look more like the heavy, persistent users than the new replacement users, in terms of total number of sequences used. 3 Table 3 presents summary statistics on number of sequences used over the 4.5 year observation period. The patterns observed in Table 2 are confirmed. borrowers have a mean of 2.15 sequences per borrower; the figure for replaced is 4.45 and for persistent it is Table 3 Summary Statistics of Borrowers by Number of Sequences in Each of the Three Groups Statistic on Sequences to Borrowers Sequences to Borrower N=1211 Sequences to Borrower N=698 Sequences to Borrower N=302 Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum Note, however, that number of sequences is only half of the equation. Duration of sequences is the other half. For example, the higher number of sequences used by replaced borrowers tend to be short (nearly 60 percent of their sequences involve 3 or fewer loans). See Table 5. 6

8 Duration of Sequences As noted in prior reports, the key metric for the CFPB definition of harm is the duration of loan sequences. A sequence in which fees eclipse the loan amount is held up as an indicia of harm. At the median national fee of $15 per $100 per loan, a sequence of seven loans would involve fees that exceed the loan amount. In Florida and Rhode Island, where the fee is capped at $10 per $100, a sequence of 10 loans in a row would hit the CFPB proposed trigger. Table 4 presents the summary statistics for each of our sample groups for number of loans per sequence. Table 4 Summary Statistics on All Sequences by Number of per Sequence Statistic on Number of per Sequence All Sequences N=2,606 All Sequences N=3,107 All Sequences N=2,213 Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum Observe, in Table 4, that the mean sequence duration for our largest group, replacement borrowers (5.05 loans), does not hit the CFPB trigger. The values for replaced borrowers (7.66 loans) and persistent borrowers (8.41 loans) are predictably higher. As we have observed in all prior samples, median behaviors are substantially lower than means, making it important to understand the distribution of behaviors at a more granular level. To this end, Table 5 shows the percentage of sequences with one loan, two or fewer loans, three or fewer loans, etc. 7

9 Table 5 Percentage Distribution of Number of per Sequence for All Sequences Number of Distribution by Number of per Sequence Percentage (All ) Percentage (All ) Percentage (All ) 1 Equal to 1 Loan Less Than 2 Less Than 3 Less Than 4 Less Than 5 Less Than 6 Less Than 7 Less Than 8 Less Than 9 Greater Than Two observations are of note. First, a large percentage of sequences never exceed the CFPB six-loan indicator of harm. Nearly 80 percent of replacement borrowers sequences, and nearly 74 percent for the other groups fall below the seven loan threshold. In other words, the CFPB could end all harmful sequences by truncating only percent of the loan sequences. Second, while sequences for replacement borrowers tend to be a bit shorter than the other groups, the distributions are fairly homogeneous overall, suggesting that within the behaviors of very-long-term users like our persistent borrowers, there may be some (or even many) loan patterns that do not constitute debt traps. We will return to this issue below. Figure 1 displays this homogeneity of sequence duration distribution across the three different groups in the sample. 8

10 Figure 1 Distribution of Sequences by Number of 100 Percentage of Sequences Equal to 1 Loan Less Than 2 Less Than 3 Less Than 4 Less Than 5 Less Than 6 Less Than 7 Number of in a Sequence Less Than 8 Less Than 9 Greater Than 10 The Worst Case Scenario: Longest 4 Sequence Per Borrower In our various studies, we have focused on the question: How many borrowers are harmed under the CFPB standard and how many never fall into a debt trap of more than six loans in a row. In a prior report, we estimated that 40 percent of borrowers were in a group that was unlikely to ever evidence harm by the worst-case sequence measurement. In the appendix, Table A-1 shows the summary statistics for longest sequence per borrower in this sample. In Table 6, we show the number of borrowers by maximum loan sequence. 4. We use the term longest sequence to refer to a borrower s sequence containing the greatest number of loans (and loan fees) for that borrower in the sample. The term may or may not overlap with temporal duration of sequences, depending on the actual number of days between loans in each of the borrower s sequences. On average, however, a sequence with seven loans will last longer than a sequence containing six loans. 9

11 Table 6 Percentage Distribution of Number of per Sequence for All Sequences Number of Distribution of Max Number of per Sequence Count of Borrowers N=1211 Count of Borrowers N=698 Count of Borrowers N=302 1 Equal to 1 Loan Less Than 2 Less Than 3 Less Than 4 Less Than 5 Less Than 6 Less Than 7 Less Than 8 Less Than 9 Greater Than IMPORTANT OBSERVATION: Observe that 1,101 ( ) borrowers never have a loan sequence of more than six loans. That is, 49.8 percent of all the borrowers in our 4.5 year statistically representative sample never hit the CFPB threshold for harm. It appears that our previous estimate of borrowers below the harm threshold, based on sampling borrowers once a year over four years, was too low. The distributions we see in Table 6 also confirm our findings in Report 7-B. The worst case behaviors for long-term users of the product are very different from newer, replacement users. The extreme behaviors of the persistent borrowers drive the averages for the product overall into ranges that far exceed median behavior - and provide the extreme examples put forth to support regulatory termination of the product. We see this starkly in Figure 2, which shows the heterogeneity of longest loan sequence by the different groups in our sample. 10

12 Figure 2 Distribution (Percentage) of Borrowers by Maximum Loan Sequence 90 Percentage of Borrowers or Sequences Equal to 1 Loan Less Than 2 Less Than 3 Less Than 4 Less Than 5 Less Than 6 Less Than 7 Number of per Sequence Less Than 8 Less Than 9 Greater Than 10 As Figure 2 shows, it is the persistent borrowers whose behavior is truly extreme, compared to other groups. Sixty-five percent of replacement borrowers never pay more in fees than they borrowed in any loan sequence. Not once. In contrast, over 80 percent of persistent borrowers have at least one sequence equal to or greater than 10 loans. These persistent borrowers make up more than 30 percent of the static pool from July 2009 (302/1000), but only 14 percent (302/2211) of the population in payday storefronts over a 3.5 year period. Thus, the static pool approach used by regulators will necessarily overstate the severity of the harm caused by the product. Is the Worst Case Representative of Overall Behavior? We have been solving for the relative sizes of the groups who never experience regulatory harm versus those who do. The next question is: Does a worst case sequence suggest multiple overly long sequences have occurred? Or are those who have at least one sequence that suggests regulatory harm also using the product in a less extensive way at other times in their product life cycle? In the appendix, Table A-2 through Table A-6 provide the answer to this question. Those tables show the distribution of shorter sequences for borrowers who have maximum sequence duration of three, six, seven, 10 and 16 loans. Figure 3 shows the results for borrowers who have a maximum sequence duration of 10 loans, a duration that regulatory studies have chosen to illustrate the large number of loans in long sequences. Note that amongst replacement borrowers, the single 10-loan sequence made up only three percent of all sequences. 11

13 In contrast, for the same borrowers, single-loan sequences (loans that were paid and stayed paid) made up over 30 percent of the borrower experiences, and percent of all sequences for such borrowers were less than seven loans in duration (and thus presumably not harmful ). Comparable figures for replaced borrowers are 6.36 percent 10-loan sequences, percent single loans, and percent of sequences less than seven loans in duration. Only the persistent borrowers have a distribution that is not weighted to shorter sequences. Figure 3 Distribution of All Sequences for Borrowers With Maximum Sequence of Percentage of Other Sequences Loan Other Loan Sequences The takeaway is that our 49.8 percent of borrowers who never have a sequence longer than six loans is only one tranche of non-harmful use. Many other borrowers who have a single sequence that hits the trigger for harmful have many more loan sequences or loan experiences that are not. With these facts, we argue again (as we did in Report 7-B) that a rule banning the product will harm more consumers than it hurts. 12

14 Is the Less Intense Use by Borrowers Indicative of Product Aversion? We have observed in Report 7-B that replacement or newer borrowers tend to have shorter product life cycles. In Report 7-B and in this report, we have observed that these borrowers tend to have shorter loan sequences. These are both signs of use of the product as it was intended: a short-term solution to a short-term problem. These observations prompt the question: Do newer or replacement borrowers get in and out quickly because they experience distress? Are they running away? In this section, we attempt to answer this question by looking at two metrics. First, do newer borrowers tend to have an initial intense use, with few pay cycles out of debt, and then leave? We approach this by measuring the number of pay cycles in sequence versus the number of pay cycles out of sequence between the first and last loan in the data set. Second, we explore the role of defaults in affecting the duration of loan sequences for the different groups of borrowers in this sample. Time In and Out of a Loan Sequence We begin with context for this metric. Table 7 shows summary statistics on the market life cycle durations of the three groups in our sample. We define market life cycle as the time elapsed between the first loan and the last loan in our data set. Table 7 Summary Statistics on Life Cycles (By Days) Statistic on Life Cycles Life Cycle N=1211 Life Cycle N=698 Life Cycle N=302 Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum

15 As noted above, the mean life cycle of a replacement borrower is relatively brief at 266 days. borrowers are in the product for more than three times as long. borrowers, by definition, are in the product for the entire sampling period. Are replacement borrowers using the product relatively continuously during their short life cycles? Are they leaving quickly because of a bad experience? The data says not. Table 8 compares the mean time in a sequence and the mean time between sequences for each of the three groups in our sample. Table 8 Mean Days In and Days Between Sequences Mean Days Borrowers Borrowers Borrowers In Sequence Between Sequences Complete summary statistics for days in sequence and between sequences appear in the appendix on Tables A-7 and A-8. We can see from Table 8 that replacement borrowers spend more time between sequences than in sequences. Given the relatively short duration of sequences for replacement borrowers, this can only occur if there are significant periods between sequences spread across the overall life cycle. If the contrary were true, then we would see the majority of life-cycle days taken up within sequences, followed by a quick exit from the product. In short, the data does not support a hypothesis that the short life cycles of replacement borrowers suggest product aversion. Figure 4 presents this comparison as counts of pay periods in sequence and between sequences. We can observe in Figure 4 the same relationship of periods of product use and periods between product uses - the replacement borrower s product life cycle is spent 40 percent of the time in a sequence and 60 percent of the time out of a sequence. 14

16 Figure 4 Mean Pay Periods in Sequence and Between Sequences During Lifecycle 100% 90% 80% Mean Pay Periods in Sequences Mean Pay Periods Between Sequences Percentage of Life Cycle 70% 60% 50% 40% 30% 20% 10% 0% Sample Summary statistics supporting Figure 4 appear in the appendix in Tables A-9 and A-10. What Role Do Defaults Play in the Longitudinal Sample? We also asked the question whether differences in default rates and the distribution of defaults across sequences might account for some of the differences in use patterns of our different borrower groups. We find that replacement borrowers do have a significantly higher percentage of loan sequences that end in default than our other groups. Table 9 shows the differences. Table 9 Default Behavior Default by Sequence N=2,606 N=3,107 N=2,213 Frequency Percentage Frequency Percentage Frequency Percentage No Default 1, % 2, % 2, % Default % % % Total 2, % 3, % 2, % 15

17 Table 9 supports an argument that part of the reason our replacement borrowers show lesser intensity of use is that they have a greater tendency to fail in the product and exit through failure. However, this higher failure rate does not indicate a higher cost to the borrower of failure. Most of those failures happen at the first loan (single loan sequence), as shown in Table Table 10 Frequency of Default by Number of in Sequence Number of in a Frequency Percentage Frequency Percentage Frequency Percentage Sequence One pre-publication reviewer of this report queried the extent to which the lighter overall use patterns of replacement borrowers were attributable to a higher frequency of firstloan defaults (i.e., single-loan sequences that ended in default). We have tested the sensitivity of our analysis to this phenomenon. We present the key analyses reported in this paper, but excluding first-loan defaulters, in Appendix B. We do not find a material change in the data that would change our conclusions. For example, the exclusion of first-loan defaulters reduces the percentage of borrowers who are never harmed from 49.8% to 46.4%, a minor difference that does not change our overall conclusions. 16

18 We can see from Table 10 that all groups tend to concentrate defaulted sequences in one or two loan sequences. If a borrower is going to fail in a sequence, it tends to happen early. Notably, the replacement borrowers have significantly higher rates of default than the other groups in two and three loan sequences, supporting the idea that some of the preponderance of shorter sequences in this group is attributable to failure. This last point raises a further question. Does an early exit from the product (long before harm from the amount of fees paid has occurred) suggest a different form of harm? Is a default harmful? Industry points out that payday loans are not reported to prime credit bureaus (at present) and thus do not harm general creditworthiness when they default. But do they harm access to payday credit? Our data suggests that there is a restriction on further payday credit when a consumer defaults, as shown in Table 11. Table 11 Percentage of Defaulted Borrowers Given Second Chance and Outcomes Group Default Count Second Chance Count Second Chance Percentage Second Chance Defaulting Again % 53.30% % 52.90% % 35.00% Table 11 suggests that replacement borrowers may also have a shorter overall experience because as a group they are less likely to have the strong relationships with lenders or other characteristics that lead a lender to make the product accessible after a default. At least in the case of persistent borrowers, the lenders decision appears economically rational - borrowers who have a history of performing payday loans with relatively high frequency are more likely to perform again, even following a default. To summarize our review of patterns that may help explain the different use intensities of different borrower groups, we do not find evidence of product aversion in the use patterns of replacement borrowers. We do, however, find some evidence that they are more likely to default early in a sequence than the more intense users like the persistent borrower group. That tendency may contribute to their pattern of lighter use. 17

19 Conclusions The most significant take away from this study is that 49.8 percent of consumers in the sample never experience harm as defined by the CFPB. For readers who have followed all of our reports in this series, this should not be surprising, because our sampling method was designed to remove the sampling bias of the single-month static pool method used in the regulatory studies. It also removes the truncation error, studying too short a time period. For the 50.2 percent who experience CFPB harm at least once, the vast majority of these borrowers loan sequences are shorter, not qualifying as harm. Thus, our analysis suggests there is insufficient harm to justify a rule that would effectively ban the product, and the CFPB should consider a less severe regulatory intervention. 18

20 Appendices: Table of Contents Appendix A: Data Supporting Main Report Table A-1: Summary Statistics on Maximum Loan Sequences 20 Table A-2: Distribution All Sequences Where the Maximum Loan Sequence Is Equal to Three 20 Table A-3: Distribution All Sequences Where the Maximum Loan Sequence Is Six 21 Table A-4: Distribution All Sequences Where Maximum Loan Sequence Is Seven 21 Table A-5: Distribution All Sequences Where Maximum Loan Sequence Is Table A-6: Distribution All Sequences Where the Maximum Loan Sequence Is Equal to Table A-7: Summary Statistics on Days in a Sequence 23 Table A-8: Summary Statistics on Days Between a Sequence 23 Table A-9: Summary Statistics on Pay Periods in Debt (In a Loan Sequence) 24 Table A-10: Summary Statistics on Pay Periods Out of Debt (Between Loan Sequences) 24 Appendix B: Recomputation Of Principal Findings, Excluding First-Loan Defaults Table B-1: Sample Counts 25 Table B-2: Summary Statistics Sequences 25 Table B-3: Distribution of Sequences 26 Table B-4: Summary Statistics on Number of per Sequence 26 Table B-5: Summary Statistics on Maximum Loan Sequence Comparison With First Payment Default Excluded 27 Table B-6: Distribution of Maximum Loan Sequence Comparison with First Payment Default Excluded 27 Table B-7: Frequency Distribution of Number of per Sequence on Maximum Loan Sequences No First Payment Default Comparison 28 Table B-8: Distribution of Number of per Sequence on Maximum Loan Sequence No First Payment Default Comparison 28 19

21 Appendix A Data Supporting Main Report Table A-1 Summary Statistics on Maximum Loan Sequences Statistic on Maximum per Sequence Maximum N=1,211 Maximum N=698 Maximum N=302 Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum Table A-2 Distribution All Sequences Where the Maximum Loan Sequence Is Equal to Three If Borrower's Maximum Loan Sequence Is 3 Sequences (By Number of ) Frequency Percent Frequency Percent Frequency Percent 1 Loan Sequence Loan Sequences Loan Sequences

22 Table A-3 Distribution All Sequences Where the Maximum Loan Sequence Is Six If Borrower's Maximum Loan Sequence Is 6 Distribution of All Other Sequences (By Number of ) Frequency Percentage Frequency Percentage Frequency Percentage 1 Loan Sequence Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Table A-4 Distribution All Sequences Where the Maximum Loan Sequence Is Seven If Borrower's Maximum Loan Sequence Is 7 Distribution of All Other Sequences (By Number Frequency Percentage Frequency Percentage Frequency Percentage of ) 1 Loan Sequence Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences

23 Table A-5 Distribution All Sequences Where the Maximum Loan Sequence Is 10 If Borrower's Maximum Loan Sequence Is 10 Distribution of All Other Sequences (By Number of ) Frequency Percentage Frequency Percentage Frequency Percentage 1 Loan Sequence Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Table A-6 Distribution All Sequences Where Maximum Loan Sequence Is 16 If Borrower's Maximum Loan Sequence Is 16 Distribution of All Other Sequences (By Number of ) Frequency Percentage Frequency Percentage Frequency Percentage 1 Loan Sequence Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences Loan Sequences

24 Table A-7 Summary Statistics on Days in a Sequence Statistic on Days in Sequences by Borrower Days in a Loan Sequence by Borrower N=1,211 Days in a Loan Sequence by Borrower N=698 Days in Loan a Sequence by Borrower N=302 Minimum Maximum Mean Standard Deviation Q1-25% Median - 50% Q3-75% % % % % Maximum Table A-8 Summary Statistics on Days Between a Sequence Statistic on Days Between Loan Sequences Days Between Loan Sequence Days Between Loan Sequence Days Between Loan Sequence Minimum Maximum Mean Standard Deviation Q1-25% Median - 50% Q3-75% % % % % Maximum

25 Table A-9 Summary Statistics on Pay Periods in Debt (In a Loan Sequence) Statistic on Pay Periods in Loan Sequences Pay Periods in a Loan Sequence Pay Periods in a Loan Sequence Pay Periods in a Loan Sequence Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum *Rounded to the nearest Pay Period Table A-10 Summary Statistics on Pay Periods Out of Debt (Between Loan Sequences) Statistic on Pay Periods Between Loan Sequences Pay Periods Between Loan Sequence Pay Periods Between Loan Sequence Pay Periods Between Loan Sequence Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum *Rounded to the nearest Pay Period 24

26 Appendix B Recomputation Of Principal Findings, Excluding First-Loan Defaults Note: For comparison purposes, the tables in Appendix B will show original computations (full sample) and the same computations excluding the first payment default observations. (Bold is sample excluding first payment defaults). Table B-2 Table B-1 Sample Counts Summary Statistics Sequences Statistic on Sequences to Borrowers Sample Characteristics Measure Number of Borrowers 1, * Number of 13,173 13,020 23,807 23,770 18,623 18,623 Number of Sequences 2,606 2,479 3,107 3,078 2,213 2,213 *Bold Numbers Are Counts Excluding First Payment Default Sequences to Borrower N=1,211 Sequences to Borrower N=1095 Sequences to Borrower N=698 Sequences to Borrower N=674 Sequences to Borrower N=302 Sequences to Borrower N=302 Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum

27 Table B-3 Distribution of Sequences Sequences per Borrower N=2,606 N=2,480 N=3,107 N=3,079 N=2,213 N=2, Total Table B-4 Summary Statistics on Number of per Sequence Statistic on Number of per Sequence No First Default No First Default No First Default Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum

28 Table B-5 Summary Statistics on Maximum Loan Sequence Comparison With First Payment Default Excluded Statistic on Maximum per Sequence No First Default No First Default No First Default Minimum Maximum Mean Standard Deviation Q1-25% Median-50% Q3-75% % % % % Maximum Table B-6 Distribution of Maximum Loan Sequence Comparison with First Payment Default Excluded Distribution Of Maximum Loan Sequences No First Default No First Default No First Default Equal to 1 Loan Equal to 2 or Less Equal to 3 or Less Equal to 4 or Less Equal to 5 or Less Equal to 6 or Less Equal to 7 or Less Equal to 8 or Less Equal to 9 or Less Equal to 10 or Greater

29 Table B-7 Frequency Distribution of Number of per Sequence on Maximum Sequences No First Payment Default Comparison Distribution No First Default No First Default No First Default Equal to 1 Loan Equal to 2 or Less Equal to 3 or Less Equal to 4 or Less Equal to 5 or Less Equal to 6 or Less Equal to 7 or Less Equal to 8 or Less Equal to 9 or Less Equal to 10 or Greater % Maximum Table B-8 Distribution of Number of per Sequence on Maximum Loan Sequence No First Payment Default Comparison Distribution Maximum Loan Counts No First Default No First Default Equal to 1 Loan No First Default Equal to 2 or Less Equal to 3 or Less Equal to 4 or Less Equal to 5 or Less Equal to 6 or Less Equal to 7 or Less Equal to 8 or Less Equal to 9 or Less Equal to 10 or Greater % Maximum

30 nonprime101.com provides research studies and articles about non-prime consumer behavior to help the public and researchers better understand them. The rate of non-prime consumers, which include thin-file, no-file and prior prime consumers, continues to rapidly grow and nonprime101.com provides unbiased and empirical studies that show the credit usage behaviors, activities and needs of non-prime consumers as a whole. Clarity Services, Inc. provides powerful real-time fraud detection and credit risk management solutions for Middle America. By leveraging unique data assets and scores, Clarity Services suite of FCRA regulated reports and scores empower providers with visibility into critical consumer information not available on traditional bureau reports. For more information, visit clarityservices.com. A Balanced View of Storefront Payday Borrowing Patterns Results From a Longitudinal Random Sample Over 4.5 Years 6. We note that mean fee levels for the data are higher than the Main Report and mean fee levels for early reporters to the Clarity system are lower. The standard deviation for fee levels for all three samples suggests significant heterogeneity in fee levels in all three samples. We conclude that while different, the fee levels are statistically similar. We also note a hypothesis that our subsample of early adopters of the Clarity system included a higher percentage of state-compliant lenders than exists in the later years, which would tend to depress the mean value of fee levels. NPRP- 7-C