Information and credit access: using bankruptcy as a signal

Abstract

Legally, a bankruptcy flag can appear on an individual's credit report for up to 10 years after the filing. The flag affects an individual's credit score, and in turn, an individual's access to credit. In this article, we investigate how the bankruptcy flag affects access to credit along three dimensions–loan acceptance, the price of the loan as is determined by the interest rate, and the amount of credit the household receives. Using the Panel Study of Income Dynamics and the Survey of Consumer Finances, we estimate a series of two-stage models corrected for sample selection and adjusted to account for the household's level of creditworthiness. We find that the bankruptcy flag increases the probability of being denied access to a loan. The flag also increases interest rates for unsecured loans and lowers the credit limits available to households. The findings have important implications with respect to current bankruptcy code and the impact that information, such as the bankruptcy flag, can have on the efficiency of the credit markets.

Notes

¹ In fact, Stavins (2000) finds that net revenues from credit cards are higher for banks that charge higher fees and interest rates despite the higher delinquency rates on these credit cards.

² The FCRA only applies to credit reporting agencies; it does not apply to creditors that obtained information about the bankruptcy filing during the 10-year window.

³ The 2005 Bankruptcy Abuse Prevention and Consumer Protection Act increased the waiting time between filings from 6 to 8 years.

⁴ There may be concern that households that are discouraged are different from those that are turned down for a loan. In our sample, discouraged households have lower income and net worth, and they are more likely to be single, female, black, unemployed and renters. In results not presented here but available upon request, we exclude the discouraged households from the sample, and the results are quantitatively similar. Thus, we leave the discouraged households in the sample.

⁵ For the interest rate model, there are two possible types of selection–whether the household applies for credit and whether the credit application is accepted. In the SCF, we are unable to separate out these two effects for the interest rate models. For this reason, we correct for selection related to having an automobile loan or having a credit card. To identify our selection equation, we include factors that control for the potential application bias as well as the possible bias due to loan acceptance or rejection.

⁶ One might argue that current and past employers are listed on an individual's credit report. However, we do not know the duration of time the individual has spent with each employer and whether it was full-time or part-time work. Also, it is has been documented that information on employers is one of the most inaccurate pieces of information on an individual's credit report (Avery et al., 2003, 2004).

⁷ Because most households respond to the PSID between March and June of the survey year, a household could file for bankruptcy in the summer of 1996 but not be classified as such in the 1996 wave. Consequently, the last year in our sample is 1995.

⁸ In the PSID, we have information that indicates whether the household filed under Chapter 7 or Chapter 13. This information, however, is not included in the SCF. Under Chapter 7, the bankruptcy flag appears on a credit report for 10 years. Under Chapter 13, the flag also can remain on a credit report for up to 10 years, but is rarely reported for more than 7 years. Since the SCF does not include information on the chapter the household filed, we need to assume that, regardless of the chapter, the bankruptcy flag is removed after the tenth year. Using the PSID, we excluded households that filed for Chapter 13 and included only those that filed for Chapter 7. We found that the results were stronger when we used only Chapter 7 filers from the PSID. Thus, our assumption that the flag remains on the credit report for 10 years may understate our SCF findings.

⁹ Unfortunately, the 1998 and 2001 SCF do not include information on a household's state of residence and so comparisons cannot be made between the PSID and SCF. For literature on how state bankruptcy exemptions affect the supply and demand of credit, see Gropp et al. (1997), Berkowitz and Hynes (1999), Lin and White (2001), Fan and White (2003) and Berkowitz and White (2004).

¹⁰ Recall that in the SCF respondents are asked about whether they applied for a loan within the last 5 years, without specifying an exact year. Since we are relying on the timing of the loan application relative to the bankruptcy filing, the robustness of our results to changes in the definition of the flag variable is important. We conducted sensitivity analysis regarding the timing of the bankruptcy flag relative to being denied access, assuming three different years for the year the household was denied a loan. (1) The household was denied in the year of the survey (1998 or 2001); (2) the household was denied 2 years before the survey and (3) the household was denied 5 years before the survey. The results were similar for these different definitions. Further, the SCF results are likely to be reliable, because we know the actual year in the PSID and the SCF results are similar to those found for the PSID.

¹¹ The Fair Credit Reporting Act also allows potential creditors to view past bankruptcy filings if the loan exceeds $150 000 (FCRA 605(b)(1)), which means that most creditors should know whether a household applying for a mortgage has ever filed for bankruptcy, regardless of how long ago it occurred.

¹² Table 4 shows that those with less education are less likely to apply for a loan. In turn, those with less education that actually do apply for a loan are more likely to be turned down for a loan. Thus, one reason the PSID results on education may be counterintuitive is that we were unable to control for this type of selection.

¹³ Berkowitz and White (2004) found that a previous bankruptcy filing tripled the probability of being turned down for a loan for owners of unincorporated businesses. Their estimates are likely to be higher than ours, because they did not control for sample selection or bankruptcy filers’ poor credit histories. Using a similar specification as theirs, we find that a previous bankruptcy filing more than doubles the probability of being turned down for a loan.

¹⁴ In our sample, 59% of households do not have a credit card balance. Because these households are not paying interest, it can be argued that they may not be affected by the interest rate. For our purposes though, we are interested in the terms of credit available, regardless of whether the household actually pays interest. Thus, we include in our sample those that carry no balance on their credit card. In results not presented here, we excluded these households from the credit card interest rate specification. The results did not change significantly.

¹⁵ Appendix presents the interest rate results ignoring selection, using Ordinary Least Squares (OLS) on the sample of households with an automobile loan or with a credit card. The coefficients on the unbiased bankruptcy flag effect are similar to those reported in Table 5.

¹⁶ Musto (2004) addresses a similar issue and finds that the credit limit increases by $1800 in the year the flag is removed.

¹⁷ Table A1 presents the OLS estimates for the credit limit model. The results are consistent with those found in Table 6.

¹⁸ The Administrative Office of the US Courts (2005) describes the primary purposes of bankruptcy as twofold, to ‘give an honest debtor a ‘fresh start’ in life, and to repay creditors in an orderly manner to the extent that the debtor has property available for payment.’