credit report data (credit report attributes and credit scores in March 2002, March 2003 and March 2005) were appended to the client data in the initial sample by the national credit reporting agency Trans Union. All personally identifiable information was removed from client records before the data files were delivered to the research team. After dropping cases for which a match was not possible for all three years, the resulting analysis sample contained detailed credit report data and information from the initial counseling interview for 59,972 clients.<br><br> For this particular pooled sample, the most common form of the delivery channel is the telephone (67.7% of clients) followed by in-person delivery (22.6%) and Internet delivery (9.7%). Note that these proportions reflect the flow of business in early 2003 and may well be different at these same agencies today. Because Internet counseling was not consistently defined across agencies in 2003, the Phase 1 analysis focused only on telephone and in-person clients.<br><br> Data provided by the agencies for clients in the sample reflect information gathered during the first ccounseling interview d with the client. All of the agencies in our sample did a first interview lasting anywhere from 30 minutes to 75 minutes. The interview collected detailed budget and asset information and identified potential causes of the clients 9 financial problems.<br><br> Options were discussed and the counselor made a recommendation to the client. The counselor's recommendation and a written action plan were part of the product delivered to the consumer. Options for many clients included a recommendation to enroll in an agency-administered debt repayment program called a Debt Management Plan (DMP).<br><br> Consumers recommended for a DMP could choose not to enroll, but the DMP recommendation is a signal that the counselor thought the client was qualified. 2 Agencies provided data on the counselor 9s recommended action step for each client, as well as information on Debt Management Plans (DMPs) for those who started a plan. In this pooled sample, 62.1% of clients were recommended for a DMP, although that percentage varied widely across agencies from a minimum of 16.9% to a maximum of 77.1%.<br><br> However, the proportion of clients who actually started a DMP was substantially smaller (29.8% overall), as was the range across agencies (16.9% to 39.1%). Characteristics of Consumers at the Time of Counseling " The decision to seek counseling is an important signal of subsequent credit problems. Credit bureau information available at the time of counseling does not yet reflect private information that many counseled clients possess regarding recent changes in their financial condition.<br><br> Consequently, the act of seeking counseling is a valuable cearly warning d indicator that facilitates early intervention. o Counseled borrowers are much more likely to declare bankruptcy during the two years following counseling, relative to consumers in a nationally representative sample (17.6% vs. 1.2%), and are much more likely to have a serious public derogatory record item added to their credit report: 23.4% vs.<br><br> 2.6% (Table 17). o A counseled borrower 9s likelihood of having bankruptcy and negative public record items being added to their credit files during the two years after counseling is higher even after controlling for initial (2003) levels of a dozen credit bureau characteristics that are predictive of risk, including bankruptcy risk scores (Table 19). " The credit profile of borrowers who choose face-to-face counseling is different from borrowers who use telephone counseling.<br><br> It appears that consumers may be self-selecting into delivery channels based in part on their perception of the severity of their financial problems. o Consumers counseled in-person generally have a higher risk of future payment problems, as indicated by lower initial risk scores. They also have more accounts with positive balances, larger mortgage balances, fewer bank card accounts and less unsecured debt (Table 21).<br><br> Impact of Delivery Channel on Credit Experience after Counseling " Controlling for credit bureau information, counselor experience, and the information collected by the counselor during the counseling session (as reflected in the counselor's perception of the client 9s primary cause of financial difficulty and the counselor 9s recommended plan of action) the delivery channel for the initial counseling session appears to have little impact on two indicators of the client 9s creditworthiness measured two years after the initial counseling. 3 o Measured two years after counseling, there is no statistically significant difference between either the bankruptcy risk score or new account delinquency risk score of telephone and face-to-face clients (Table 27). o Face-to-face clients tend to be more likely to file for bankruptcy in the subsequent two years (Table 27).<br><br> o The counselor 9s recommendation conveys information about the client 9s future credit performance that is not otherwise observable through the credit report variables. Information obtained during the counseling interview regarding the primary cause of the client 9s financial difficulty is also predictive, especially when controlling for the full set of variables in the model (Table 27). o These results are robust across the entire sample, as well as subsets of clients from agencies that offer clients a choice between telephone and face-to-face counseling.<br><br> " Examination of two additional outcome measures 3 total non-mortgage dollar balances and the total number of accounts that were 30 days or more delinquent during the prior 18 months 3 revealed mixed findings with respect to whether delivery channel is associated with differences in post-counseling experience. o Total non-mortgage balances of clients measured two years after counseling do not seem to be affected by the counseling delivery channel (Table 26). o Consumers who experience face-to-face counseling have a reduced number of delinquencies, as measured two years after the initial counseling (Table 26).<br><br> Relationship of DMP to Credit Experience after Counseling " Consumers who were recommended for a DMP and chose to start payments had a significantly lower incidence of bankruptcy over the two years following counseling, and had higher bankruptcy and delinquency risk scores (signaling lower payment risk) at the end of the two-year period, holding other factors constant (Table 27). o Clients who were recommended for a DMP but chose not to start did not experience improved bankruptcy and delinquency scores. o Clients who were not recommended for a DMP but ended up starting one anyway (possibly because they addressed other problems in order to qualify for a DMP) experienced improved delinquency risk scores two years later, relative to DMP non-starters, but no change in bankruptcy risk scores.<br><br> " Further analysis indicates no significant difference in the positive effect of an engagement in a DMP across delivery channels. DMP clients in both the telephone and in-person delivery groups experienced improvement in risk scores (Table 28). 4 Caveats Several caveats to these findings should be noted.<br><br> First and foremost, because the sample of participating agencies was not selected to be representative of industry-wide practices, the results cannot be considered representative of the typical experience of counseled consumers nationwide. Instead, they reflect what is obtainable from a group of agencies that emphasize client education and identification of the underlying cause of financial problems. The fact that telephone and face-to-face delivery of counseling services appear to generate equivalent outcomes for consumers in this sample suggests that, when done well, the two delivery channels can be equally effective.<br><br> We have presented results on four separate indicators of post-counseling outcomes for consumers, measured two years after the initial counseling visit. Two of these indicators (a commercially available bankruptcy risk score product; a commercially available new account delinquency risk score product) represent general measures of creditworthiness. Two indicators (total non-mortgage balances; number of accounts delinquent) reflect specific margins of credit usage.<br><br> In addition, we also provide results on the incidence of bankruptcy during the two years following counseling. While these indicators examine the consumer 9s credit experience from a variety of angles, other measures of the impact of the counseling experience would help to provide a more comprehensive picture. In particular, survey evidence on consumer attitudes, knowledge gained, and perceived financial stress, pre and post-counseling, would augment the objective measures of consumer credit performance and provide a more complete picture of counseling 9s impact.<br><br> It would also be helpful to extend the post-counseling observation period for the sample to see if differences emerge in the credit bureau data for telephone and face-to-face clients. For those clients for whom counseling does change their borrowing and payment behavior, two years may not be enough time for the change to be fully reflected in their credit reports and credit scores, especially if prior financial distress was severe and its impact was not yet reflected in the credit report at the time of counseling. There is some evidence that consumers 9 choice of delivery channel is associated with their credit usage patterns and resulting risk profile prior to counseling.<br><br> The analysis in this report accounted for these differences to the extent allowed by the available data. However, a more detail assessment of borrowers at the time of the initial counseling, either during the interview itself or through supplemental survey work, would allow for more precise controls for this self- selection. Controlling for self-selection would help determine if the equivalence in outcomes across the telephone and face-to-face delivery channels indicated equal effectiveness or was an artifact of initial client characteristics.<br><br> The results on the role of Debt Management Plans are particularly intriguing, but self-selection may be partly responsible. Clients who start DMPs outperform all other counseling clients on all of our outcome measures. Admittedly, clients who were recommended for DMPs are in better financial shape than clients who do not qualify.<br><br> But, the evidence also indicates that between two borrowers who are recommended for a DMP (i.e., borrowers for whom a DMP is both a 5 workable option and the best option), the borrower who actually starts payments in a DMP fares significantly better on all outcome measures at a point two years after counseling. Perhaps there is some residual self-selection effect driving this result (e.g., borrowers who make a commitment to start a DMP are more motivated to repay than borrowers that do not). Alternatively, perhaps the DMP experience itself (e.g., budgeting to make regular DMP payments; continued interaction with and reinforcement from the counseling agency) generates the improvement in the outcome indicators.<br><br> Given the significantly improved credit profiles for clients who do start DMPs, this phenomenon deserves closer study. 1 Evaluating the Impact of Delivery Channels for Credit Counseling Services by Michael Staten and John Barron Introduction In 2004 ten credit counseling agencies were selected through a competitive proposal process to participate in an empirical study of the effect of credit counseling on long-term borrower behavior. This multi-year project is jointly sponsored by the Consumer Federation of America and American Express with the objective of identifying best practices in the counseling industry and quantifying their impact on consumers.<br><br> The study will examine the effectiveness of face-to- face, telephone and Internet counseling across all types of consumer clients, including those who subsequently enroll in a DMP and those who receive financial counseling only or recommendations for legal and other assistance. Within each of the three major categories of counseling delivery channels, the study will consider how different approaches to the timing, duration and content of counseling influence client outcomes. Effectiveness of counseling will be gauged by using credit bureau data to examine the credit profile of counseled clients at the time of the initial counseling session and subsequently at points two or more years following counseling.<br><br> The study has been divided into two phases. Phase 1 focuses on an examination of the outcomes of prior counseling activities. The advantage of this retrospective analysis is that we need not wait several years to identify counseling outcomes.<br><br> A disadvantage is that archived data on individuals who received counseling in the past may not be sufficiently rich to clearly identify the effects of counseling, as opposed to other client characteristics that may also influence subsequent client behavior. A planned Phase 2 will modify the sample design to collect even more detailed information on incoming counseling clients and the treatment they receive, and follow these individuals over time. For Phase 1 of the project, participating agencies supplied detailed data from the counseling cintake d interview on more than 70,000 clients who were counseled during early 2003.<br><br> The following report presents some of the key findings of the Phase 1 analysis in terms of client credit profiles and payment experience over the two year period following the initial counseling. The findings suggest several issues that should be considered when designing Phase 2 of the study. The report is organized as follows.<br><br> Section I contains several tables that describe the data received from the ten counseling agencies and the composition of the pooled sample of clients 2 (e.g. distribution of clients across delivery channel, cause of financial problems, and counselor recommendation). Section II presents summary measures of the credit bureau profile of counseled clients in the sample, and how they compare (around the time of initial counseling) with a separate, nationally representative sample of all consumers with credit reports.<br><br> By pooling the sample of counseled consumers with the nationally representative sample, we examine how differences across individuals in key credit bureau variables (especially risk scores) can be used to predict which consumers will seek counseling services. Section III utilizes the credit bureau data to compare the counseled group vs. the national sample in terms of the incidence of bankruptcy and other derogatory public record information during the two years following counseling, as well as changes in risk scores.<br><br> Perhaps not surprisingly, we find that the very act of seeking out counseling is an important signal of subsequent credit problems. Individuals who seek counseling reveal themselves to be in financial stress, and that financial stress is often not fully captured by concurrent credit bureau data. Section IV examines observable differences across counseled clients that may lead them to choose one method of counseling delivery vs.<br><br> another. We use multivariate statistical (Probit) models to identify predictors of the choice of telephone versus face-to-face counseling. The results identify systematic differences in the clients who gravitate toward one channel vs.<br><br> another. These results are useful for interpreting subsequent performance differences in the two groups during the years following counseling. Section V reports the multivariate analysis that identifies the effect of counseling channel on subsequent credit behavior, controlling for factors identified in Section IV that appear to affect the choice of counseling delivery method.<br><br> The analysis introduces various client performance measures drawn from the credit bureau data. It provides an initial look at how an individual's credit experience during the two years following counseling is related to variables that were observable at the time of counseling, including dimensions of their counseling experience. A concluding Section VI offers caveats and further discussion of the primary findings.<br><br> Section I: Counseling Agency Data A total of 34 agencies submitted grant proposals in response to the Request For Proposal (RFP) that was distributed by American Express to 600+ agencies in August 2004. A review committee consisting of representatives from American Express, Consumer Federation of America and Georgetown University evaluated the proposals and met in October 2004 to select finalists. From the group of 34 respondents, 10 agencies were selected as grant finalists.<br><br> Finalists were those agencies that were responsive in all of the required areas in the RFP, and received high ratings in the subjective areas of evaluation (data capture ability; description of the quality of their programs; evidence of innovative programs). All agency financial reports were screened for evidence of potential conflicts of interest or other problems of the type being investigated by the U.S. Internal Revenue Service.<br><br> Evidence of such problems (e.g., documented 3 outsourcing of back-office tasks to for-profit companies owned by agency Board members) negatively influenced the overall ranking of the agency. In addition, the proposal review committee used information on agency size and scope of operations (especially the percentage distribution of clients across counseling delivery channels) to select the finalists to achieve representation across several key dimensions of the sample design, as discussed in greater detail below. Site visits were conducted between November 1 and December 20, 2004 for each of the agency finalists.<br><br> Visits included opportunities to listen to ongoing counseling sessions and intake calls, either clive d or pre-recorded as part of the agency 9s quality assurance program. The ten agencies that emerged from this process as participants in the study were Auriton Solutions (Roseville, MN), ClearPoint Financial Solutions, Inc. (Richmond, VA), Consumer Credit Counseling Service of Greater Atlanta, Inc.<br><br> (Atlanta, GA), Consumer Credit Counseling Service of Los Angeles (Los Angeles, CA), Consumer Credit Counseling Service of Montana (Great Falls, MN), Consumer Credit Counseling Service of North Central Texas (McKinney, TX), InCharge Debt Solutions (Orlando, FL), LSS Financial Counseling Service (Duluth, MN), Novadebt (Freehold, NJ), and Money Management International (Houston, TX). Some are phone specialists, some do just face-to-face counseling, some do both. A number of agencies offer some form of counseling through the Internet.<br><br> At least one agency has a unique "crisis management" approach to the client; others utilize a somewhat more scripted approach. What they all have in common is an emphasis on client education and identification of the cause of underlying financial problems. In thinking about the sample, it is important to realize that the primary goal of Phase 1 is not to assess how a nationally representative sample of counseled clients performs over time.<br><br> Rather, the objective is to utilize the experience of agencies that appear to be high-quality providers in order to see if, when done well, counseling makes a difference, and to identify whether effectiveness is influenced by delivery method. Data in the sample reflect information gathered during the first ccounseling interview d with the client. All of the agencies in our sample conducted initial client interviews lasting anywhere from 30 minutes to 75 minutes.<br><br> The interview collected detailed budget and asset information and identified potential causes of the clients 9 financial problems. Options were discussed and the counselor made a recommendation to the client. The counselor's recommendation and a written action plan were part of the product delivered to the consumer.<br><br> This definition of the "first counseling session" or "intake interview" was communicated to all participating agencies. One option that was offered to many clients was enrollment in an agency-administered debt repayment program called a Debt Management Plan (DMP). Consumers recommended for a DMP could choose not to enroll, but the DMP recommendation is a signal that the counselor thought that customer was qualified.<br><br> Each agency was asked to provide information for all individuals who received a first-counseling session during March and April of 2003. A match of the counseling data to credit bureau data drawn from the second quarters of 2002, 2003, and 2005 was attempted. To maintain confidentiality, these matches were performed by one of the three major U.S.<br><br> consumer reporting agencies, Trans Union, LLC (TU). Each agency sent their files directly to TU; TU then matched 4 based on available personal identifiers and sent the matched data, stripped of all unique personal identifiers, to the project research team at the Credit Research Center (CRC), Georgetown University. For about 10,000 clients, a successful match across all three years could not be made.<br><br> Attempts to match failed for a variety of reasons, but the greatest dropout occurred when the agency did not collect the Social Security Number of the individual being counseled. Limited credit histories at the time of counseling also appeared to account for a number of dropouts. Using data collected by the agencies where available, the average size of unsecured debt and the number of unsecured creditors was substantially lower for individuals who could not be matched to credit bureau data Table 1 indicate the total number of observations within the 2003 time period that TU attempted to match and the percent that could not be matched to credit bureau data for all three years (2002, 2003, and 2005).<br><br> Overall, 61,476 clients (87%) were matched to credit bureau data for all three years. The extent of matching was similar across agencies with the notable exception of Agency B. For this agency, two-thirds of the clients could not be matched to credit bureau data.<br><br> The credit bureau provided a number of attributes for each individual. Of particular interest are two types of credit scores provided. Both scoring products are risk management tools that TU markets to creditors and other firms making credit-related decisions.<br><br> One reflects the risk of a serious delinquency on any account (equivalent in concept and roughly equivalent in scaling to the widely-reported FICO score product developed by Fair Isaac, Co.), while the second reflects the risk of bankruptcy. For the set of individuals that were matched to credit bureau data, the credit bureau data did not provide a complete set of risk scores (i.e., both scores for 2002, 2003 and 2005) for 2.4% of the clients. 1 The final column in Table 1 indicates the distribution across agencies of the 59,972 clients for whom a complete set of credit reports and credit risk scores were available for subsequent analysis.<br><br> 1 Note that there is no clear pattern to individuals missing such scores with respect to either the delivery channel or the counseling recommendation. 5 Percent Dropped Because Visit Occurred Outside Early 2003 Period Percent Dropped Because Not Fully Matched to Credit Bureau Data Percent Dropped Because Of Missing Risk Score Agency A 0.0% 6.0% 2.3% B 0.0% 66.1% 3.7% C 14.4% 12.8% 3.2% D 6.5% 6.2% 2.1% E 0.0% 4.2% 2.2% F 5.0% 18.9% 2.8% G 0.0% 7.7% 2.3% H 23.3% 9.0% 7.5% I 10.7% 24.5% 3.4% J 2.6% 20.0% 3.2% Starting Number of Observations 72,253 Starting Number of Observations 71,003 Starting Number of Observations 61,476 Of These, Number Within 2003 Time Period 71,003 Of These, Number Matched to Credit Bureau Data 61,476 Of These, Number With Risk Score 59,972 Percent Dropped 1.7% 13.4% 2.4% Table 1: Counseling Observations Dropped From Analysis Due to Incorrect Time Period, Inability to Fully Match to Credit Bureau Data, Or Missing Risk Score The sample drops 1.7% of the cases when the date of the initial counseling did not occur as specificed by the sampling instructions. For 5 of the 10 agencies, these were cases that were not initiated during the March 2003 to April 2003 period.<br><br> For one agency, these were cases that were not initiated in March, April, or May 2003. The other four agencies had no observations outside the designated time span. We did retain the observations when the counseling date was not available.<br><br> For two agencies, this was a sizable proportion of their total observations (i.e., 28% and 62% of their observations). Note that 22 of the 59,972 observations did not provide information on the delivery channel - these observations are omitted from subsequent analysis, leaving 59,950 observations. 6 Table 2 divides counseling clients by the delivery channel for the initial interview.<br><br> Two agencies provided only telephone counseling, and a third was predominately telephone with less than 5 percent of clients counseled face-to-face. At the other end of the delivery spectrum, two agencies offered primarily face-to-face or internet counseling sessions. The remaining five agencies offered a combination of delivery channels for counseling.<br><br> For this particular pooled sample, the most common form of the delivery channel is the telephone. However, recall that our sample is not a representative sample of individuals who seek counseling. In particular, the sample is dominated by large volumes of clients from two agencies, one of which does telephone counseling exclusively.<br><br> 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Overall Table 2: Counseling Delivery Channels Across Agencies In Sample (10 Agencies: 59,950 Observations) Face-to-face Internet Telephone Table 3 indicates differences across the agencies in the recommendations that arose from the counseling session. Note that one agency (E) not only did not distinguish between referral to other services and referral to legal assistance but also did not distinguish between either type of referral and financial counseling only. Table 3 indicates large differences in the extent to which counseling resulted in a debt management plan being proposed.<br><br> However, when one considers the percent of clients who actually started a DMP (see the last column in Table 3), the differences across agencies are not as pronounced. Table 4 reports the recommendation of the counselor by type of delivery channel. Note that as a percent of all counseling sessions, DMP recommendations were made least frequently for face- to-face counseling clients.<br><br> Nevertheless, face-to-face sessions display the highest cconversion rate d in terms of percent of clients who start a DMP (see last row in Table 4). 7 Self Manage/ Client Can Handl DMP recommended Financial Counseling Onl Referral to Other Agency/ Servic Referral to Legal Assistanc Not Available Some DMP payments mad Agenc A 3.7% 26.7% 64.8% 0.6% 4.2% 0.0% 16.9% B 10.3% 59.6% 12.6% 2.2% 4.3% 11.1% 20.2% C 14.2% 29.8% 41.4% 6.8% 5.6% 2.2% 31.1% D 3.9% 51.7% 22.2% 15.6% 4.0% 2.7% 35.9% E 5.3% 64.4% 0.0% 30.1% 0.0% 0.2% 36.3% F 4.5% 43.2% 26.1% 6.9% 7.8% 11.5% 25.0% G 5.2% 77.1% 1.1% 7.8% 4.8% 3.9% 27.4% H 1.5% 32.8% 61.3% 0.0% 1.1% 3.3% 39.1% I 0.0% 16.9% 83.1% 0.0% 0.0% 0.0% 16.9% J 0.0% 34.4% 0.2% 0.0% 0.0% 65.5% 25.6% Total 5.2% 62.1% 9.9% 13.9% 3.0% 5.9% 29.8% Table 3: Counselor Recommendations Across Agencies and Percent of DMPs Started Total number of observations is 59,950. With the exception of agency I , the category "DMP recommended" includes individuals who were offerred but chose not to enroll in a debt management plan.<br><br> For two agencies ( A and J ), the variable indicating the number of DMP payments made was missing, so the existence of payments was infered from a variable indicating DMP status. The "self manage" category includes what various agencies refer to as "client can handle", "choose to self manage", or "self administer". "Not available" can be due to an incomplete session.<br><br> 8 Face-to-face Internet Telephone Overall Counselor Recommendation Self Manage/Client Can Handl 6.6% 3.3% 5.1% 5.2% DMP recommended 53.4% 71.9% 63.5% 62.1% Financial Counseling Only 19.7% 17.8% 5.5% 9.9% Referal to Other Agency/Service 10.7% 2.6% 16.6% 13.9% Referral to Legal Assistance 4.9% 2.1% 2.5% 3.0% Not Available 4.7% 2.3% 6.8% 5.9% Overall 100.0% 100.0% 100.0% 100.0% Some DMP payments made 32.9% 21.5% 29.4% 29.4% Table 4: Delivery Channel By Counselor Recommendation Percentages are for 59,950 sample of counseling agency clients. Table 5 summarizes the primary reason identified by the counselor for the credit difficulties of the individual seeking counseling, using nine broad categories. Note that one agency, agency B , provided no reasons.<br><br> For the other agencies, there is largely agreement that the two most likely reasons are related to either a disruption in income or poor money management. While of value, these responses are subjective. It is not clear that different agencies, or even counselors within the same agency, use the same criteria in selecting a primary cause of credit difficulties.<br><br> The Appendix indicates the various responses that were collected within each of the nine broad categories. For the various cprimary d reasons identified for credit difficulties, Tables 6A, 6B, and 6C indicate how those primary reasons distribute across counselor recommendations. In Table 6A, the percentages by reason sum to 100% across recommendations.<br><br> In Table 6B, the percentages by recommendation sum to 100% across reasons. Table 6C displays the percent of individuals who start a DMP by primary cause of financial difficulty, subdivided by whether the client was recommended for a DMP. Note that 22,749 clients were not recommended for a DMP, but 3.6% of this group ended up starting a DMP.<br><br> This can occur when a client 9s financial circumstances change subsequent to the initial counseling session, perhaps even as a result of the session (e.g., client sells an asset or obtains a part-time job to supplement income). The first column displays more detail on the cDMP start-rate d for this group, by primary cause of the client 9s financial distress. Similarly, the second column shows that 37,201 clients were recommended for a DMP, and 45.2% of this group actually started a DMP.<br><br> 9 Table 7 indicates how the primary causes distribute across delivery channels. Interestingly, the primary cause for a client's financial problems does not appear to be critical in determining the type of delivery channel chosen. Table 8 provides information on the housing situation of those seeking counseling by delivery channel.<br><br> 10 Primary Reason A B C D E F G H I J Overall DEATH/ILLNESS/INJURY/DISABILITY 9.6% 8.1% 7.9% 10.9% 6.8% 7.6% 15.5% 5.9% 13.5% 8.8% FRAUD/LEGAL PROBLEMS/EXPENSES 0.5% 0.3% 0.5% 1.1% 0.3% HOME/AUTO EXPENSES 3.8% 1.9% 0.4% 0.9% 0.7% 0.7% INCOME LOSS/INSTABILITY 32.6% 27.5% 26.2% 40.1% 27.3% 33.5% 10.3% 27.7% 49.1% 33.9% MARITAL/DOMESTIC PROBLEMS/SUPPORT 5.1% 4.7% 5.9% 4.9% 4.5% 4.8% 5.9% 0.1% 6.3% 4.8% OTHER/NA 8.9% 100.0% 22.1% 20.6% 7.6% 12.1% 7.0% 45.8% 50.0% 20.4% 13.9% POOR MONEY MGMT 43.5% 32.4% 36.1% 36.4% 48.1% 44.0% 21.0% 12.7% 10.6% 36.8% SCHOOL EXPENSES 0.3% 0.7% 0.9% 0.1% 1.6% 3.0% 0.8% SUBSTANCE ABUSE/GAMBLING 0.4% 0.2% 0.8% 0.1% 0.4% 0.1% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% Number of observations 59,950 Table 5: Primary Causes of Credit Difficulties, As Identified by Counselor The various specific primary causes are summarized in terms of the above nine broad categories. See Appendix A for detail on c ategory components Agency 11 Primary Reason Self Manage/ Client Can Handle DMP recommended Financial Counseling Only Referal to Other Agency/ Service Referral to Legal Assistance Not Available Total Number of obs. DEATH/ILLNESS/INJURY/DISABILITY 3.7% 64.1% 8.1% 15.1% 3.7% 5.3% 100.0% 5,274 FRAUD/LEGAL PROBLEMS/EXPENSES 9.3% 70.4% 6.2% 6.8% 5.6% 1.9% 100.0% 162 HOME/AUTO EXPENSES 16.0% 54.9% 11.5% 12.2% 4.1% 1.4% 100.0% 419 INCOME LOSS/INSTABILITY 4.1% 65.7% 8.3% 14.2% 2.7% 5.1% 100.0% 20,312 MARITAL/DOMESTIC PROBLEMS/SUPPORT 4.8% 63.6% 9.9% 13.5% 3.2% 5.1% 100.0% 2,855 OTHER/NA 10.6% 30.9% 18.6% 16.3% 2.9% 20.7% 100.0% 8,345 POOR MONEY MGMT 4.4% 69.6% 8.6% 12.9% 3.1% 1.5% 100.0% 22,043 SCHOOL EXPENSES 7.6% 77.1% 9.3% 4.3% 1.4% 0.2% 100.0% 485 SUBSTANCE ABUSE/GAMBLING 9.1% 47.3% 16.4% 12.7% 14.5% 0.0% 100.0% 55 Overall 5.2% 62.1% 9.9% 13.9% 3.0% 5.9% 100.0% 59,950 Table 6A: Primary Cause of Financial Difficulty, by Counselor Recommendation The various specific primary reasons (causes) are summarized in terms of the above nine broad categories.<br><br> See Appendix A for d etails Counselor Recommendation 12 Primary Reason Self Manage/ Client Can Handle DMP recommended Financial Counseling Only Referal to Other Agency/ Service Referral to Legal Assistance Not Available Overall DEATH/ILLNESS/INJURY/DISABILITY 6.2% 9.1% 7.2% 9.5% 10.8% 8.0% 8.8% FRAUD/LEGAL PROBLEMS/EXPENSES 0.5% 0.3% 0.2% 0.1% 0.5% 0.1% 0.3% HOME/AUTO EXPENSES 2.1% 0.6% 0.8% 0.6% 0.9% 0.2% 0.7% INCOME LOSS/INSTABILITY 26.5% 35.8% 28.2% 34.5% 30.9% 29.3% 33.9% MARITAL/DOMESTIC PROBLEMS/SUPPORT 4.4% 4.9% 4.8% 4.6% 5.0% 4.1% 4.8% OTHER/NA 28.3% 6.9% 26.1% 16.3% 13.5% 48.9% 13.9% POOR MONEY MGMT 30.8% 41.3% 31.8% 34.0% 37.5% 9.4% 36.8% SCHOOL EXPENSES 1.2% 1.0% 0.8% 0.3% 0.4% 0.0% 0.8% SUBSTANCE ABUSE/GAMBLING 0.2% 0.1% 0.2% 0.1% 0.4% 0.0% 0.1% Total 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% Number of observations 3,139 37,201 5,938 8,345 1,798 3,529 59,950 Table 6B: Primary Cause of Financial Difficulty, by Counselor Recommendation The various specific primary reasons (causes) are summarized in terms of the above nine broad categories. See Appendix A for d etails. Counselor Recommendation 13 Primary Reason DMP not recommended DMP recommended Overall Number of obs.<br><br> DEATH/ILLNESS/INJURY/DISABILITY 3.1% 49.3% 32.7% 5,274 FRAUD/LEGAL PROBLEMS/EXPENSES 10.4% 36.0% 28.4% 162 HOME/AUTO EXPENSES 3.7% 45.2% 26.5% 419 INCOME LOSS/INSTABILITY 3.5% 46.2% 31.5% 20,312 MARITAL/DOMESTIC PROBLEMS/SUPPORT 4.1% 44.0% 29.5% 2,855 OTHER/NA 1.9% 34.5% 12.0% 8,345 POOR MONEY MGMT 5.2% 45.6% 33.3% 22,043 SCHOOL EXPENSES 6.3% 36.9% 29.9% 485 SUBSTANCE ABUSE/GAMBLING 3.4% 46.2% 23.6% 55 Overall 3.6% 45.2% 29.4% Number of observations 22,749 37,201 59,950 Table 6C: Percent Starting DMP by Primary Reasons Identified by Counselor For Credit Difficulties The various specific primary reasons (causes) are summarized in terms of the above nine broad categories. DMP Started 14 Primary Cause Face-to- face Internet Telephone Overall DEATH/ILLNESS/INJURY/DISABILITY 8.2% 6.0% 9.4% 8.8% FRAUD/LEGAL PROBLEMS/EXPENSES 0.4% 0.2% 0.2% 0.3% HOME/AUTO EXPENSES 1.5% 1.0% 0.4% 0.7% INCOME LOSS/INSTABILITY 30.4% 30.2% 35.6% 33.9% MARITAL/DOMESTIC PROBLEMS/SUPPORT 4.9% 4.7% 4.7% 4.8% OTHER/NA 15.8% 11.7% 13.6% 13.9% POOR MONEY MGMT 37.5% 44.0% 35.5% 36.8% SCHOOL EXPENSES 1.0% 2.1% 0.6% 0.8% SUBSTANCE ABUSE/GAMBLING 0.3% 0.0% 0.0% 0.1% Total 100.0% 100.0% 100.0% 100.0% Number of observations 13,567 5,791 40,592 59,950 Table 7: Primary Causes Identified by Counselor For Credit Difficulties Across Agencies in the Sample The various specific primary causes are summarized in terms of the above nine broad categories. Delivery Channel Face-to-face Internet Telephone Total Number Housing Situation Own 41.4% 33.5% 41.1% 24,235 Rent 34.1% 37.6% 41.1% 23,511 Live with Parents / Other 18.3% 0.5% 7.3% 5,467 Not Available 6.2% 28.4% 10.5% 6,737 Overall 100.0% 100.0% 100.0% 59,950 Table 8: Housing Situtation By Channel of Delivery for Sample Matched to Credit Bureau Files 15 II.<br><br> Credit Bureau Data Credit bureau data provide an objective assessment of the client before and after the credit counseling experience. Based on credit data provided by TU, Table 9 provides a comparison between a nationally representative random sample of consumers ("national sample") and individuals in the counseling sample in terms of the total number of trades (accounts) with a positive balance and total balances across all trades. Not surprisingly, the average number of trades for individuals in the national sample is lower.<br><br> Note that those seeking counseling have both higher mean and median levels of total balances. Percentile National Sample Counseling Sample National Sample Counseling Sample 1% 0 0 $0 $0 5% 0 1 $0 $0 10% 0 1 $0 $2 25% 1 3 $0 $10 50% 3 6 $12 $31 75% 5 9 $76 $92 90% 8 12 $169 $161 95% 10 15 $240 $208 99% 14 21 $464 $334 Mean 3 7 $57 $62 Standard Deviation 3 5 $116 $80 Number of Observations 263,937 59,950 263,937 59,950 Percent of individuals with active/verified bankcard 71.6% 82.3% 71.6% 82.3% Table 9: Credit Bureau Variables for National Representative Sample Versus Credit Counseling Sample: Trades and Balances Number of trades with positive balance Total balance across all trades (in 1,000s) Credit bureau variable names are AT29 and AT33; National sample credit bureau data from June, 2003. Counseling sample credit bureau data from March, 2003.<br><br> 16 If we focus on consumers who have one or more active/verified bank card accounts, such accounts are more likely to exist for consumers in the counseling sample. Among individuals with active/verified bank card accounts, Table 10 indicates that the distribution regarding the number of accounts is similar across the two groups (national sample and counseled group). However, Table 10 also indicates that bank card account utilization (balance as a percent of credit limit) is clearly higher for the counseled group.<br><br> Percentile National Sample Counseling Sample National Sample Counseling Sample 1% 1 1 0 0 5% 1 2 0 0 10% 2 3 0 33.3 25% 3 5 0 66.7 50% 6 7 0 100 75% 9 11 50 100 90% 14 15 100 100 95% 17 18 100 100 99% 25 25 100 100 Mean 7 8 27 79 Standard Deviation 5 5 38 29 Number of Observations 189,008 49,340 189,008 49,340 Percent of individuals with active/verified bankcard 100.0% 100.0% 100.0% 100.0% Table 10: Credit Bureau Variables for National Representative Sample Versus Credit Counseling Sample: Bankcard Use Number of active or active/verified bank cards Percent of bankcard trades greater than 50% of limit Credit Bureau variable names are BC01 and BC30; National sample credit bureau data from June, 2003. Counseling sample credit bureau data from March, 2003. Two risk scores were provided by the credit bureau.<br><br> One measures the likelihood that a consumer will file for bankruptcy. The second score measures the likelihood of serious delinquency on a newly opened account. For both scores, a higher score indicates lower risk (i.e., either less likely to file for bankruptcy or less likely to become seriously delinquent on a newly opened account).<br><br> Table 11 indicates that, on average across nearly 60,000 observations, individuals counseled in person have higher risk (i.e., have lower scores) of bankruptcy and serious delinquency. Closer examination reveals that among the five agencies (A, C, D, G, and H) that had significant numbers of clients counseled through in-person and telephone channels (as opposed to specializing in one or the other), bankruptcy risk scores for in-person clients were significantly lower (relative to telephone clients) for three of these five agencies and significantly higher for one. For delinquency risk scores, two of the five agencies had scores significantly 17 lower for clients counseled in person.<br><br> Thus, there is some evidence that in-person clients at these five agencies were, on average, higher risk than their telephone counterparts. [Note: Agency C has a substantially higher average risk score for its telephone clients, relative to any other client subgroup in the sample. Possibly this is due to a targeted marketing effort or affiliation that attracts a group of clients who appear to be lower risk, at least as measured by the credit report around the time of initial counseling.] Face-to- face Internet Telephone Overall Agency A Bankruptcy Risk Score 227 260 252 248 Delinquency Risk Score 595 607 596 600 B Bankruptcy Risk Score 232 232 Delinquency Risk Score 580 580 C Bankruptcy Risk Score 260 250 419 333 Delinquency Risk Score 594 611 671 630 Bankruptcy Risk Scor 220 210 219 Delinquency Risk Scor 580 573 579 E Bankruptcy Risk Scor 235 235 Delinquency Risk Scor 592 592 Bankruptcy Risk Scor 268 212 207 255 Delinquency Risk Scor 603 590 589 600 Bankruptcy Risk Scor 243 256 253 251 Delinquency Risk Scor 590 600 603 599 Bankruptcy Risk Scor 278 252 273 Delinquency Risk Scor 597 590 595 I Bankruptcy Risk Scor 245 245 Delinquency Risk Scor 589 589 Bankruptcy Risk Scor 261 233 234 Delinquency Risk Scor 596 589 589 Overal Bankruptcy Risk Scor 239 256 246 245 Delinquency Risk Scor 588 602 597 595 Table 11: Average 2003 Credit Bureau Bankruptcy and Delinquency Risk Scores By Delivery Channel and Agency Sample includes only observations with credit bureau risk scores.<br><br> Note that higher score indicates lower risk of bankruptcy. Total number of observations is 59,950 18 Table 12 indicates the differences in the risk scores between the counseling sample and the nationally representative sample of consumers. Table 12 indicates that the counseling sample has substantially lower bankruptcy and new account delinquency risk scores than the national sample.<br><br> For instance, whereas around 50% of the national sample has a delinquency risk score over 740, less than 10% of the counseling sample has a delinquency risk score this high or higher. And, this difference occurs even though the national sample includes some individuals very similar to those in the counseling group. Percentile National Sample Counseling Sample National Sample Counseling Sample 1% 15 5 424 419 5% 128 15 460 445 10% 143 37 498 470 25% 329 134 614 531 50% 693 180 742 588 75% 835 333 841 658 90% 959 522 880 723 95% 976 648 894 759 99% 987 870 908 843 Mean 599 245 718 595 Standard Deviation 294 190 139 95 Number of Observations 263,937 59,950 263,937 59,950 Percent of individuals with risk score 85.7% 97.6% 85.7% 97.6% Table 12: Credit Bureau Variables for National Representative Sample Versus Credit Counseling Sample: Risk Scores Bankruptcy Risk Score New Account Risk Score National sample credit bureau data from June, 2003.<br><br> Counseling sample credit bureau data from March, 2003. 19 Credit bureau variables as predictors of decision to seek counseling Individuals in the counseling group are clearly different from individuals in a nationally representative sample of consumers with credit reports. We now consider a wider array of factors that are associated with seeking credit counseling assistance.<br><br> To do so, we combine the national credit bureau sample with the counseling sample to create a pooled dataset, and define a variable that equals one if the individual is from the counseling sample and zero otherwise. We then estimate a Probit model with the dependent variable indicating whether the individual is from the counseling sample. For the pooled sample, 18.5 percent are directly identified as having sought counseling at approximately the same time as the credit bureau data was pulled.<br><br> Of course, there are surely individuals within the national sample who also sought counseling during this period that are not identified. Their presence introduces measurement error that, among other things, can lead to a bias toward zero in our estimated coefficients. That is, an observed relationship between credit bureau variables and the choice of counseling must be stronger in this sample (relative to one in which there are no unidentified counseled consumers) to achieve statistical significance.<br><br> Table 13 presents the results for the Probit model estimation that includes as explanatory variables selected credit bureau variables as measures of credit profile and creditworthiness. Coefficients on the explanatory variables indicate the marginal effect of a change in the variable on the probability of being in the counseled sample. Table 13 indicates that individuals with lower risk scores, a higher number of trades with positive balances (both overall and bankcard), a greater utilization rate for bank cards, and larger numbers of delinquencies and inquiries are more likely to seek out counseling.<br><br> Note that in Table 13, the coefficients indicate the marginal effect of a one unit change in the independent variable. For instance, an increase by 100 in the bankruptcy risk score (one unit) is predicted to decrease the likelihood of an individual seeking counseling by .0247 or 2.47%. Table 13 also reports averages for the explanatory variables for the two samples (control and counseling) to provide the reader with a better idea of how the counseling sample differs from the nationally representative sample across various credit bureau variables.<br><br> 20 Coefficient in Probit Model (z-statistic) Mean of variables for control sample (standard deviation) Mean of variables for counseling sample (standard deviation) Bankruptcy risk score (in 100s) -0.0247*** 5.99 2.45 (65.13) (2.94) (1.9) New account risk score (in 100s) -0.0131*** 7.18 5.95 (14.26) (1.39) (0.95) Total number of trades with balance > 0 0.0096*** 3.39 6.61 (35.65) (3.24) (4.53) Total balance, non-mortgage trades (in 10,000s) -0.0014*** 1.74 2.92 (7.41) (4.16) (3.43) Total balance, mortgages (in 10,000s) -0.0021*** 3.99 3.28 (23.88) (10.07) (6.56) Number of bankcard trades 0.0004*** 5.44 7.44 (2.97) (5.27) (5.07) Proportion of bankcard trades with balance over 50% of limit 0.1202*** 0.19 0.65 (53.04) (0.34) (0.4) Number of non-installment trades over 50 % of limit 0.0106*** 0.92 3.59 (24.62) (1.62) (3.1) Number of trades 30 or more days past due in last 18 mths 0.0077*** 0.73 3.14 (25.06) (1.72) (3.24) Number of currently past due balances 0.0152*** 0.18 1.19 (24.58) (0.66) (1.77) Number of inquiries in last 6 months 0.0041*** 1.06 2.16 (16.69) (1.8) (2.72) Highest retail credit limit (in 1,000s) 0.0018*** 1.47 1.29 (5.73) (2.28) (1.89) Number of observations 323,887 263,937 59,950 Table 13: Probit Model of Factors Affecting Likelihood that an Individual in Credit Bureau Pooled Sample is From Counseling Agency Sample Not reported is a variable indicating missing values for the variable measuring the percent of bankcard trades with balance over 50%. This variable is not defined if an individual has no bank cards. Coefficients indicate marginal effects; * significant at 10%; ** significant at 5%; *** significant at 1% Table 14 suggests that another predictor of an individual seeking out counseling is a decline in the risk score over the preceding 12 months.<br><br> Consumers who visited a counselor in 2003 experienced a median reduction in their bankruptcy risk score of 55 points, and a median reduction in their delinquency risk score of 26 points during the 12 months prior to their visit. By comparison, the median change in risk scores was at or near zero for consumers in the nationally representative sample over a different (longer) period, June 2003 to December 2004. 21 Percentile Counseling Sample, March 2002 to March 2003 National Sample, June 2003 to Dec.<br><br> 2004 Counseling Sample, Mar. 2002 to March 2003 National Sample, June 2003 to Dec. 2004 1% -647 -503 -201 -161 5% -455 -294 -144 -92 10% -350 -204 -117 -60 25% -189 -54 -72 -20 50% -55 0 -26 2 75% 4 47 8 31 90% 81 158 40 67 95% 146 239 60 91 99% 317 424 101 141 Mean -98 -10 -33 3 Standard Deviation 184 158 63 55 Number of Observations 59,950 263,937 59,950 263,937 Table 14: Change in Credit Bureau Risk Scores During 12 Months Prior to Counseling Change in Bankruptcy Risk Score Change in New Account Risk Score Table 15 indicates how the changes in risk scores are related to both the counseling delivery channel and the subsequent counselor recommendation.<br><br> Note that clients who are referred for legal assistance experienced the largest average decreases in risk scores during the prior year, while those deemed able to self-manage had the lowest average reduction in risk scores. Both relationships seem consistent with expectations. With respect to the channel of delivery, the average magnitude of the drop in risk score over the year prior to counseling was similar across channels.<br><br> 22 Face-to-face Internet Telephone Overall Self Manage/Client Can Handle Change in Bankruptcy Risk Score -50 -126 -96 -84 Change in Delinquency Risk Score -15 -22 -31 -26 Number of Observations 897 191 2,051 3,139 DMP recommended Change in Bankruptcy Risk Score -99 -100 -102 -101 Change in Delinquency Risk Score -35 -32 -34 -34 Number of Observations 7,250 4,165 25,786 37,201 Financial Counseling Change in Bankruptcy Risk Score -96 -93 -88 -93 Change in Delinquency Risk Score -34 -28 -30 -31 Number of Observations 2,669 1,028 2,241 5,938 Referal to Other Agency/Service Change in Bankruptcy Risk Score -95 -99 -94 -94 Change in Delinquency Risk Score -34 -26 -32 -33 Number of Observations 1,449 151 6,745 8,345 Referral to Legal Assistance Change in Bankruptcy Risk Score -116 -105 -117 -116 Change in Delinquency Risk Score -49 -48 -43 -46 Number of Observations 664 121 1,013 1,798 Not Available Change in Bankruptcy Risk Score -56 -105 -92 -86 Change in Delinquency Risk Score -24 -40 -34 -32 Number of Observations 638 135 2,756 3,529 Overall Change in Bankruptcy Risk Score -94 -100 -99 -98 Change in Delinquency Risk Score -34 -31 -33 -33 Number of Observations 13,567 5,791 40,592 59,950 Table 15: Change in Credit Bureau Risk Scores During 12 Months Prior to Counseling Visit, by Subsequent Counselor Recommendation And by Delivery Channel 23 Section III: A Comparison of Outcomes: Counseled Clients vs. Nationally Representative Sample In this section, we examine post-counseling changes in the incidence of negative public record items (including bankruptcy) of the counseled group, relative to a nationally representative sample of consumers. We also examine the post-counseling change in credit risk scores.<br><br> The results must be viewed only as suggestive, however, because the time period used to measure performance for the counseled group (the 24-month period from March 2003 to March 2005) differs from the time period available for measuring performance of the national sample (the 18- month period from June 2003 to December 2004). Table 16 indicates the changes in bankruptcy risk scores and new account delinquency risk scores for the counseling sample between March 2003 and March 2005, side-by-side with the comparable change for the national sample between June 2003 and December 2004. To account for the fact that those seeking counseling begin the period with significantly lower risk scores compared to consumers in the nationally representative sample, we divide individuals according to their initial score values in 2003.<br><br> In particular, we separate the combined sample into deciles based on their risk score, and then measure the average subsequent change in the risk score for each decile. The summary statistics reported in Table 16 indicate that among individuals with high initial risk scores, the subsequent drop in risk scores was especially large for individuals who sought counseling. In other words, controlling for the risk score of an individual as of 2003, the extent of the reduction in the risk score is greater for the counseling group over the following period as compared to the national sample.<br><br> A simple regression analysis (not shown) indicates that, controlling for 2003 bankruptcy risk scores, the subsequent bankruptcy score for individuals in the counseling sample fell by approximately 59 points compared to the control group. Similarly, controlling for 2003 delinquency risk scores, the subsequent delinquency score for individuals in the counseling sample fell by 22 points compared to the control group. Though the differences are slightly less, similar results hold even if one excludes from the analysis those who subsequently filed for bankruptcy.<br><br> 24 Decile for 2003 Bankruptcy Risk Score (combined sample) Minimum Score in Decile (2003) Maximum Score in Decile (2003) National Sample, June 2003 to Dec. 2004 National Sample Number of Observations Counseling Sample, March 2003 to March 2005 Counseling Sample Number of Observations 1 0 135 56.8 18,342 62.6 16,332 2 136 169 45.0 18,442 35.5 11,692 3 170 269 61.6 20,408 15.7 12,019 4 270 427 37.8 22,145 -48.4 10,268 5 428 596 -4.2 26,901 -134.3 5,620 6 597 705 -26.1 30,496 -198.8 1,938 7 706 752 -18.7 31,649 -200.1 687 8 753 847 -34.2 31,503 -272.2 705 9 848 950 -64.0 32,159 -278.2 473 10 951 998 -59.6 31,892 -139.9 216 Mean/No. Obs.<br><br> -9.7 323,887 -8.3 59,950 Decile for 2003 Deliquency Risk Score (combined sample) Minimum Score in Decile (2003) Maximum Score in Decile (2003) National Sample, June 2003 to Dec. 2004 National Sample Number of Observations Counseling Sample, March 2003 to March 2005 Counseling Sample Number of Observations 1 402 490 24.7 23,891 33.9 8,579 2 491 552 12.0 20,338 6.5 12,337 3 553 606 9.0 19,023 -0.9 13,645 4 607 658 7.7 21,742 -17.0 10,497 5 659 702 6.3 25,400 -33.4 6,628 6 703 751 5.8 27,958 -47.1 4,759 7 752 805 4.0 29,710 -56.1 2,309 8 806 843 -3.2 32,135 -51.1 599 9 844 874 -7.1 31,869 -54.8 373 10 875 925 -12.8 31,871 -46.2 224 Mean/No. Obs.<br><br> 3.3 323,887 -7.6 59,950 Table 16: Change in Credit Bureau Risk Scores, 2003 - 2005 Change in Bankruptcy Risk Score Change in New Account Risk Score Table 17 displays results regarding change in public record items. For this analysis, we identified those consumers that had no public records evident in their 2003 credit report, either with respect to the broader range of derogatory public record items or more specifically bankruptcy. As Table 17 indicates, 93.6% of the national sample and 90.6% of the counseling sample had no public record of bankruptcy as of 2003.<br><br> With respect to all types of derogatory 25 public records, 86.4% of the national sample and 81.3% of the counseling sample had no public derogatory reports in their credit file. Separate inspection revealed that, for the counseling group, the consumers who had no public records were similarly represented across delivery channels and counselors 9 recommendation. For individuals without derogatory public records (either bankruptcy or the more inclusive category) at the outset of the period, we then identified those who had one or more derogatory public record items (or bankruptcy filing) in March of 2005 for the counseling sample and in December of 2004 for the national sample.<br><br> Percentile National Sample Counseling Sample National Sample Counseling Sample Percent with No Public Bankrupcty/Derogatory Records in June 2003 93.6% 86.4% Percent with No Public Bankrupcty/Derogatory Records in March 2003 90.6% 81.3% Of Those with No Public Bankruptcy Record, Percent With Public Bankruptcy/Derogatory Record 18 months later 1.2% 2.6% Of Those with No Public Bankruptcy Record, Percent With Public Bankruptcy/Derogatory Record 24 months later 17.6% 23.4% Table 17: Change in Public Record Items, 2003 - 2005 Public Bankruptcy Record Public Derogatory Record Note that 17.6% and 23.4 % of counseled clients, respectively, either subsequently filed for bankruptcy or had a derogatory public record (that included bankruptcy filing) within 24 months after the counseling experience. In contrast, for the national sample, 1.2% and 2.6 %, respectively, either subsequently filed for bankruptcy or had a derogatory public record (that included bankruptcy filing) 24 months after the counseling experience. Table 12 indicated a substantial difference in risk scores for the counseling sample compared to the national sample.<br><br> Clearly this can be one contributing factor that explains the differences in bankruptcy filing rates between the two groups reported in Table 17. However, even after controlling for risk scores at the outset, the regression model estimates displayed in Table 18 indicate that those who visited a counseling agency had an increased likelihood of a subsequent bankruptcy or derogatory public record. In Table 18, we consider specifications that include either one or both risk scores.<br><br> In addition, recognizing that the likelihood of a bankruptcy filing may be non-linear in risk scores, and in particular that individuals with extreme values may be 26 different from others, we include as control variables two additional dummy variables for each risk score that equal one if the individual's risk score is in the lowest or highest decile. The first two columns of Table 18 indicate that an increase in either an individual's bankruptcy risk score or an individual's delinquency risk score reduces the likelihood of subsequent bankruptcy, as expected. Recall that these commercial scores are scaled so that higher scores indicate lower risk.<br><br> However, those with the highest scores (top decile) and those with the lowest scores (bottom decile) are even less likely to subsequently file for bankruptcy than predicted by a simple linear specification. The fourth and fifth columns of Table 18 indicate a similar finding for a derogatory event. Note that for both the bankruptcy and derogatory public record events, an individual in the counseling sample is more likely to experience a "bad" outcome, even after controlling for risk scores.<br><br> The third and sixth columns in Table 18 include both bankruptcy and delinquency risk score variables. There are two interesting features to note regarding these results. First, it remains the case that individuals in the counseling sample are more likely to experience either a subsequent bankruptcy or derogatory event.<br><br> Second, with both bankruptcy and delinquency risk scores in the specification, an increase in the delinquency risk score, holding constant the bankruptcy risk score, is associated with an increase in the likelihood of a bankruptcy for those not in the bottom decile. Tables 19 and 20 further illustrate that, controlling for initial risk scores, counseling is an important signal of subsequent credit problems even after adding a number of additional control variables that have been identified as important components of risk scores. Table 19 examines the link between the decision to seek counseling and a subsequent bankruptcy event or a public derogatory event.<br><br> Table 20 examines the link between the decision to seek counseling and subsequent risk scores. 2 Individuals in the counseling sample reveal themselves to be in financial distress. Tables 19 and 20 show that financial stress is not fully captured by the credit bureau data at the time of counseling.<br><br> Among other implications, this characteristic of counseled clients makes it difficult to isolate the value of counseling itself because contemporaneous credit bureau information often doesn 9t immediately capture the private information that induces individuals to seek counseling. Because creditors typically only report updates on account activity to the credit bureau once per month, there is almost always a lag time between a credit event and the time it is first reflected on the credit report. New financial distress takes time to manifest in credit behavior (e.g., delinquency), and additional time to appear on a credit report.<br><br> 2 Table 19 considers restricted samples of individuals with no bankruptcy filing event (column 1) or no public derogatory event (column 2) in 2003. Table 20 provides predictions of subsequent risk scores for restricted and full samples. For the counseling sample, two years separate the 2003 independent variables and the subsequent event.<br><br> For the national sample, one and one-half years separate the 2003 independent variables and the subsequent event. 27 Bankruptcy scores only Delinquency scores only Bankruptcy and Delinquency scores Bankruptcy scores only Delinquency scores only Bankruptcy and Delinquency scores Counseling group indicator 0.0663*** 0.1052*** 0.0566*** 0.0697*** 0.1109*** 0.0664*** (92.39) (111.89) (85.44) (77.98) (104.20) (75.15) Bankruptcy risk score: 2003 (in 100s) -0.0049*** -0.0060*** -0.0107*** -0.0091*** (55.17) (50.96) (79.04) (50.86) Bankruptcy score in bottom 10% of combined sample: 2003 -0.0036*** -0.0031*** -0.0041*** -0.0027*** (8.86) (8.06) (5.72) (3.84) Bankruptcy score in top 10% of combined sample: 2003 -0.0103*** -0.0076*** -0.0084*** -0.0053** (5.36) (3.92) (3.67) (2.30) Delinquency risk score: 2003 (in 100s) -0.0083*** 0.0035*** -0.0206*** -0.0028*** (36.35) (14.79) (63.84) (7.30) Delinquency score in bottom 10% of combined sample: 2003 -0.0103*** -0.0038*** -0.0113*** -0.0049*** (18.74) (8.04) (13.23) (6.01) Delinquency score in top 10% of combined sample: 2003 -0.0164*** -0.0114*** -0.0177*** -0.0159*** (9.97) (9.47) (8.20) (8.73) Observations 301492 301492 301492 276865 276865 276865 Coefficients indicate marginal effects. * significant at 10%; ** significant at 5%; *** significant at 1% Table 18: Predicting Likelihood of Actual Bankruptcy Filings and Public Derogatory Events Based on Credit Bureau Risk Scores and Counseling Event (Probit Model) Sample resticted to clients that had no record of a public bankruptcy filing in 2003 (first three columns) or no record of a pu blic derogatory event (second three columns).<br><br> The proportion of individuals who subsequently had a record of a public bankruptcy filing is .042 and the proportion of individuals who subsequently had a record of a public derogatory event is .063. Bankruptcy Event Derogatory Event 28 Bankruptcy Event Derogatory Event Counseling group indicator 0.0414*** 0.0510*** (67.94) (59.35) Bankruptcy risk score (in 100s) -0.0041*** -0.0064*** (34.41) (33.79) Bankruptcy score in bottom 10% of combined sample -0.0038*** -0.0051*** (10.20) (7.31) Bankruptcy score in top 10% of combined sample -0.0072*** -0.0052** (3.84) (2.29) Delinquency risk score (in 100s) 0.0012*** -0.0061*** (4.27) (12.77) Delinquency score in bottom 10% of combined sample -0.0011** -0.0011 (2.10) (1.24) Delinquency score in top 10% of combined sample -0.0099*** -0.0139*** (8.16) (7.55) Total number of trades with balance > 0 0.0005*** 0.0005*** (8.45) (4.71) Total balance, non-mortgage trades (in 10,000s) 0.0003*** 0.0005*** (9.44) (9.21) Total balance, mortgages (in 10,000s) -0.0000* 0.0001*** (1.73) (3.80) Number of bankcard trades 0.0002*** 0.0005*** (5.97) (8.05) Proportion of bankcard trades with balance over 50% of limit 0.0045*** 0.0064*** (7.20)

less