JSTOR

A. Relevant Legislation and Industrial Organization

Two areas of legislation are relevant to the study of consumer credit after bankruptcy. Several aspects of personal‐bankruptcy law have potentially important implications for postbankruptcy consumers and their creditors, as do some of the restrictions placed on credit bureaus. This section covers each briefly and discusses the adaptation of bank‐card issuers to their legal environment.

Chapter 7 discharges existing debts, except those that cannot be discharged (e.g., student loans) and those that consumers reaffirm (e.g., car loans, so they can keep the cars). The discharge permanently enjoins the affected creditors from trying in any way to collect. In exchange, the consumer forfeits assets that exceed statutory exemptions, which vary by state. The alternative procedure is Chapter 13, where there is no forfeit or immediate discharge but instead a partial repayment on a 3–5 year schedule followed (if completed) by a discharge of the remainder.2 This study focuses on Chapter 7 because it is the more common choice (about 70% of personal filings) and because the discharge simplifies the relation between the filing and future credit access. With the affected debt no longer outstanding, the subsequent importance of the filing to potential creditors is simply through the fact that it occurred, and when it occurred. Apart from the removal of bankruptcy information, there is no particular significance to the postbankruptcy year.

The legal schedule aside, more time since filing is intuitively better news about a consumer, in that it indicates distance from the circumstances that precipitated filing. Empirical research identifies medical expenditures and debt as strongly related to the filing decision (Sullivan 1968; Stanley and Girth 1971; Sullivan, Warren, and Westbrook 1989; and Domowitz and Sartain 1999), and since the underlying medical problem is fatal in some cases and cured in others, it is likely that its lingering influence on survivors declines over time. The influence of other bankruptcy‐related events, such as job loss, business failure, or divorce, also decline. So an upward trend in credit access after bankruptcy makes sense from this perspective.

There is the potential for a countervailing force, however, in the law on repeat discharges. A filer is ineligible for a subsequent discharge until 6 years after a previous discharge; the discharge is the final event of the bankruptcy proceeding, so this means the filer cannot complete a second Chapter 7 filing until 6 years after completing the first. Therefore, recent Chapter 7 filers are different from the rest of the population in their delayed access to Chapter 7 as an escape from debt obligations. During the delay, impaired creditors may be able to garnishee wages or otherwise seize value, which means that the 6 year rule may boost the filer's creditworthiness, where this boost would decline to zero by 6 years. It would intuitively bottom out well before 6 years, since consumers' unsecured credit is generally bank cards, and the default‐risk difference between a bank‐card holder who can file in a few years and one who can file right away appears small. The one who can file in a few years could wait a while, charge his or her limit, maybe make a few minimum payments, and then file. So the likely path of creditworthiness after filing, given that the filing and its date are observable, is the sum of two trends: an upward trend due to generally changing circumstances and a downward trend that flattens some time before 6 years, due to the repeat‐discharge law.

The law that governs whether the filing and its date are observable is the Fair Credit Reporting Act (FCRA). Ten years after the date of adjudication of bankruptcy, the date shortly after filing when the court decrees that the filer is bankrupt, the FCRA requires credit bureaus to exclude the filing from reports (FCRA sec. 605(a)(1)). So the moment that the credit bureau carries out this order, the report that potential creditors can observe loses all reference to the filing. The moratorium for all other adverse items is 7 years (FCRA sec. 605(a)(2–6)), so all derogatory information about a filer's prepetition debt is gone by 7 years postpetition. Neither this nor any other law obliges creditors not to condition on bankruptcies more than 10 years before or other delinquencies more than 7 years before; the obligation is only for “consumer reporting agencies” not to report this information to third parties.3

In addition to restricting what credit reports contain, the FCRA also restricts the scenarios under which they may be furnished. The scenario important to the bank‐card industry, the one responsible for the 3 billion per year “pre‐approved” solicitations,4 is FCRA section 604(c)(1)(B), allowing bureaus to furnish reports “in connection with any credit or insurance transaction that is not initiated by the consumer,” provided that “the transaction consists of a firm offer of credit or insurance,” and also that the consumer has not expressly prohibited bureaus from doing this. A “firm offer of credit or insurance” is defined in FCRA section 603(l) as one where the consumer must be offered credit if the report meets “specific criteria” and if the information the consumer sends back to the creditor meets some additional limited standards. In this way, bank‐card issuers actively seek out consumers with attractive credit reports, quickly bringing the consumers' bank‐card access to their attention.

Credit reports show creditors a long list of statistics relating to credit use. Creditors are free to process these statistics as they see fit, but they can also refer to commercially provided credit scores. Credit scores map credit reports to scalar estimates of creditworthiness, and while several scores are available, the FICO score of Fair Isaac and Company is by far the most widely referenced. Over 75% of mortgage loans use FICO scores,5 corporate credit‐rating agencies characterize the risk of consumer receivables in securitizations by reference to FICO scores,6 and underwriting standards often refer explicitly to FICO scores.7 So the FICO score is not only a respected estimate of a credit report's implied creditworthiness, it is the implied creditworthiness for some purposes.

B. Relevant Literature

With its exploration of credit access after Chapter 7, this study contributes to the growing literature on personal bankruptcy. The theoretical literature observes that the bankruptcy code balances the benefit of insuring against expensive misfortunes against the cost of facilitating creditor abuse (e.g., Rea 1984; Dye 1986; and Athreya 2001). One goal of the empirical literature is to establish where this balance is struck, with several studies taking the straightforward approach of quantifying the circumstances of those filing bankruptcy. Sullivan (1968), Stanley and Girth (1971), and Sullivan and colleagues (1989) all find substantial medical debt owed by bankrupts, and presumably discharged in bankruptcy, which indicates at least some amount of insurance benefit to bankruptcy. The incidence of abuse is harder to pin down. Culhane and White (1999) estimate much less repayment capacity among filers, and by implication much less evidence of abusive filing, than Barron and Staten (1997).

Another way to establish consumers’ use of bankruptcy is to fit a model of the filing decision. Domowitz and Sartain (1999) fit a model that embeds the choice between Chapters 7 and 13, and consistent with the earlier studies, they find high medical expenses provoking bankruptcy. Among the circumstances they associate with choosing Chapter 7 over Chapter 13 are low income, unemployment, and again high medical expenses. Gross and Souleles (2002) address the time series trend in aggregate bankruptcy filings, which was strongly upward from 1994 to 1998, and find by fitting a model predicting bankruptcy from household risk characteristics that credit‐card holders were 1% more likely to file in 1997 than in 1995, controlling for household risk. The implication is a simultaneous decrease in the perceived downside from filing, where this downside could be embarrassment, that is, “stigma”, or reduced access to postbankruptcy credit.

Staten (1993) directly addresses the role of postbankruptcy credit by documenting a significant fraction of recent filers with at least one new credit relationship—16% by 1 year postpetition and 53% by 5 years—and arguing that consumer creditors suffer a coordination problem. All creditors presumably want consumers to expect little credit after Chapter 7, since their historical recovery rate in Chapter 7 is so low,8 but they may also view recent filers as attractive customers and hence grant them credit. From this perspective, the upward trend in filings results from the upward trend in subprime consumer lending. Ausubel (1997) and Ellis (1998) also connect increasingly aggressive subprime lending to the filing rate, but their argument is that the extra loans simply increase the number of high‐debt households that benefit from filing.

A. Bankruptcy and Credit Scores

Let be the FICO percentile, as defined previously, for consumer c at the end of his or her yth postdischarge year, so that yth‐year change is . The empirical questions are whether this change is abnormally large for , compared to other values of y and whether it increases with . We address both questions at the same time by first sorting consumers two ways, by y into year groups 2 through 11 and by into initial‐score‐percentile ranges [0, 10) through [40, 50), taking means within each year/score‐range pair.11 The results are plotted as figure 1, which has initial score on the horizontal axis and the 1‐year FICO‐percentile change on the vertical axis. Each line shows the score‐percentile change as a function of the initial score percentile for a given postdischarge year.

(35 KB)

Fig. 1.— FICO‐percentile change over the yth postdischarge year

Open New Window

The graph makes several points. First, FICO‐percentile changes generally decline as the initial FICO percentile increases; in other words, FICO scores are mean‐reverting. As Section V shows, that holds for consumers in general, not just bankrupts. Second, there is a small seventh‐year effect that is stronger when the initial score is lower. Finally, there is a strong tenth‐year effect for the best initial credits. FICO percentiles in the 40–50 range at 9 years improve by 19, implying that these consumers move ahead of 19% of the nonfiler population in apparent creditworthiness when their flags are removed. This is far above the usual change for scores in this range, which appears negative. The 10th‐year effect decreases as decreases and is imperceptible for .

The 10th‐year effect suggests that the 10‐year reporting limit on bankruptcy information has a large effect on the perceived creditworthiness of borrowers, especially borrowers whose credit reports are otherwise positive. The 7th‐year effect, and its shape, is consistent with the regulation of nonbankruptcy derogatory information. Over this year, the last details of the filing (charge‐offs, repossessions, defaults, etc.) disappear, but the filing itself remains. This is good news for those whose details were relatively bad but not for those whose details were relatively good, so it fits that the 7th year boosts FICO percentiles for those with low initial scores but not high ones.

We establish the statistical significance of these patterns by regressing score changes on year dummies, along with a time trend, within each initial‐score interval. Let be the postdischarge year observed for consumer c, and and be indicator variables for and , respectively. We fit the model for each of the five ranges of . The results, collected in table 1, show a significantly positive 7th‐year effect for all but the highest initial scores, and a significantly positive 10th‐year effect for all but the lowest initial scores, thereby establishing the statistical significance of the results in figure 1. Reporting restrictions on adverse information deliver an immediate boost to apparent creditworthiness, as represented by FICO scores, especially that of the more‐creditworthy past filers 10 years later. To see how this affects the clearing of the credit market, we move the analysis from scores to actual credit.

Table 1 Seventh and Tenth‐Year Effects in Credit‐Score Percentiles

Open New Window

B. New Credit after Flag Removal

Does the effect of the flag‐removal regulation on credit scores translate to meaningful changes in access to unsecured credit? A straightforward way to find out is to fit a regression model similar to that of table 1 with the dependent variable changed to a measure of credit access. The credit reports offer many potential measures; we focus on the broadest measures of bank‐card access, the number and aggregate credit limit of a consumer's open bank cards.

The regression model is the same as in table 1, except that I7 is dropped and I11, an indicator for the eleventh year, is added. We drop I7 to focus on flag removal, and add I11 because there might be a lag before the credit access that flag removal brings appears on credit reports. For new credit to appear, an issuer must notice a consumer's score, mail that person the offer, receive the acceptance, and report back to the credit bureau. Even if this process begins soon after the flag goes down, sometime in the middle of the 10th year, it may not finish before the end of that year, so the 10th‐year effect could extend into an 11th‐year effect. Filers are sorted by into four ranges: [0, 16.67), or low score, [16.67, 33.33), or medium score, [33.33, 50), or high score, and no score, for those with missing values for .12 Of the panelists with , 28.1% have a low score, 34.4% have a medium score, 23.3% have a high score, 12.7% have no score, and 1.5% have and therefore excluded. Results are in panels A (Number of Open Cards) and B (Total Credit Limit of Open Cards) of table 2.

Table 2 Tenth and Eleventh‐Year Effects on the Number and Credit Limit of Bank Cards

Open New Window

For the past filers in the high score and categories, the 10th‐year effect implies a substantial boost in credit access. In panel A, the point estimates for both groups are around one extra card per four consumers in the 10th year, and close to one per two consumers in the 11th. Similarly, in panel B, credit limits rise by over $1,000 on average in the 10th year for both groups, and the medium score filers see a similar increase in the 11th.13 There is little perceptible effect for filers with low or missing scores. These results demonstrate that the apparent increase in creditworthiness does affect actual allocation of credit. Lenders appear to depend on the scores, or similar functions of credit reports, and removing this information from reports affects their decisions. The average changes for each initial‐score/discharge‐year combination are plotted as figures 2 (Number of Open Cards) and 3 (Total Credit Limit of Open Cards), where the postdischarge year is on the horizontal axis, the credit‐access change is on the vertical axis, and there is one line for each initial‐score range. Both graphs show little change for low‐ and missing‐score consumers in each year, whereas for the higher‐score consumers, there is a pronounced spike at year 10. There is little perceptible trend around 6 years, which is consistent with the repeat‐discharge moratorium having little effect, though—owing to the reasons outlined previously—the effect might be economically significant but too diffused across postdischarge years to detect.

(31 KB)

Fig. 2.— Change in number of open bank cards over yth postdischarge year

Open New Window

(35 KB)

Fig. 3.— Change in credit limit of open bank cards over yth postdischarge year ($)

Open New Window

These results from the Bankruptcy Panel show significant new credit access for the relatively more‐creditworthy filers following the flag removal. They cannot show the arrival of this credit in much detail, however, because the observations are quarterly and do not directly show the flag removal. By contrast, the Geographic Panel is monthly and does show the flag removal, so for panelists that lose flags, we know exactly when they lose them and can track their credit access over the months following. The cost of using the Geographic Panel is that the population of filers losing flags is small, and the data on credit limits are unreliable.

Our sample is all 134 panelists who lose Chapter 7 bankruptcy flags in 8/97 through 9/98, months 6 through 19 of our sample. This allows us to track their credit access from 5 months before to 18 months after the month they lose their flags, which we call month 0. We sort them by their month −1 FICO scores into three groups, with cutoffs chosen to distribute the 134 panelists relatively evenly: the lower group, with scores below 585 (48 panelists), the middle group, with scores from 586 to 670 (45 panelists), and the upper group, with scores above 670 (41 panelists). We also track all the filers who do not lose their flags over the period (1,119 panelists), to reflect general trends. For this contrast group, month 0 is 6/98. The statistic tracked is “Total number of bank cards opened within 6 months of the profile date,” the broadest and highest‐frequency measure of recent bank‐card acquisition; the result, with one line per group, is plotted as figure 4.

(35 KB)

Fig. 4.— New bank cards in past 6 months, by months since flag removal (Geographic Panel).

Open New Window

What figure 4 adds to figure 2 is that the new bank cards arrive soon after the regulatory event, arriving on reports in the second half of the first postflag year. The new, above‐trend credit starts appearing on reports around 6 months post removal, and the 6‐month measure appears to start trailing off around 12 months.

This section shows that the 10th‐year withdrawal of bankruptcy information from credit reports delivers an immediate boost in apparent creditworthiness for the filers with at least medium creditworthiness going into the year. It further shows that the boost translates to significant new credit access for these filers over the ensuing year. These results illustrate the efficacy of manipulating consumer credit by regulating credit bureaus. By the same token, they show that the filers received credit they would not have received under full information, which begs the question of whether this credit is too much. Excess credit implies excess household risk and delinquency, eventually; the remainder of the article addresses this question of whether this occurs by following filers’ scores and delinquency as far into the postflag experience as the data allow.

A. Fico Scores in the Longer Run

The empirical question is what happens to the scores after their initial rise. If the scoring algorithm makes efficient use of bankruptcy information, we would expect the ex‐flag scores to return eventually to their cum‐flag levels, as newly arriving information makes up for the information lost to the flag removal, if the consumers’ true creditworthiness remained constant. However, excess credit suggests a decline in true creditworthiness, so that the net effect of the regulation may be that household risk is higher down the road than it was in the first place. In this case, we should find ex‐flag scores to be abnormally low compared to cum‐flag scores. To run a meaningful comparison we must account for the mean reversion, apparent in figure 1, that is a general feature of FICO scores.

Abusing slightly the notation of the previous section, let be the FICO score of consumer c on observation date t. The mean reversion is apparent if, using just the contrast group of nonfilers in the Bankruptcy Panel, we regress the change of FICO scores from the beginning to the end of the sample period, that is, , on the initial score . This is the first regression in panel A of table 3. The coefficient on is significantly negative, consistent with mean reversion. To avoid confusing this predictable drift with the time series dynamics of interest, all the tests for FICO‐score changes include the initial FICO as an additional regressor.

Table 3 Longer‐Run Scores and Delinquency in the Bankruptcy Panel

Open New Window

To establish the longer‐run dynamics of FICO scores, we append to the contrast group those filers who lost their flags by 4/96. As explained in Section III, these are the filers with discharge months between 11/84 and 7/86. The indicator variable is 1 for these 2,888 panelists and 0 for the 20,000 contrast‐group panelists. We run two regressions in which and are the independent variables. In the first regression, the dependent variable is , whereas in the second regression, it is . The first regression measures the excess change to an early date in the postflag experience, and the second regression measures the excess change to a later date. From table 1, we expect a positive loading on LOSEFLAG in the first regression, reflecting the initial rise; the loading in the second regression shows the longer‐run effect. Results are in the second and third rows of table 3, panel A.

As expected, the excess FICO change of the flag‐losing panelists to 4/96 is positive. But tracking them down the road another 17 months, we find the opposite. Despite the intervening boost from the FCRA, the 9/97 scores of these panelists are significantly lower than their 7/94 scores predict. That is, the excess credit documented previously appears to lead to a long‐run net increase in credit risk. For a better perspective on the timing of this reversal, we turn to the Geographic Panel, where we can track the panelists who lose their flags between 4/97 and 3/98 from exactly the month the flags disappear to exactly 24 months later. The downside is that there are only 96 such panelists. To these 96, we append, analogously to the preceding, all the nonfiler panelists, defined as those panelists without bankruptcy flags as of 12/97. For the 96 panelists who lose flags, we define month 0 to be the first month without a flag, and for the nonfiler panelists, we define month 0 to be 1/98. We run five regressions that predict the change in FICO scores 0, 6, 12, 18, and 24 months after the flag removal, with the same explanatory variables from panel A of table 3: the cum‐flag FICO score from month −1, and the indicator LOSEFLAG for the flag‐losing filers. The results are in panel A of table 4.

Table 4 Longer‐Run Scores and Delinquency in the Geographic Panel

Open New Window

We find that the loading on LOSEFLAG turns from significantly positive in months 0 and 6 to significantly negative by month 18. The 24‐month loading is also negative, but the confidence interval is wide. Putting together these longer‐run FICO score results, the boost in apparent creditworthiness delivered by flag removal is seen to be short lived, gone in a year and soon replaced by a score deficit. By this measure, removal of bankruptcy information increases the credit risk of the affected households.

While the credit‐score reversal has direct implications for the households’ ongoing access to credit, due to creditors’ extensive reliance on FICO scores, they do not completely settle the question of whether the households actually become less truly creditworthy. This is because FICO scores automatically decline when consumers accept or inquire after credit.14 We need a more direct view of true credit risk, which we get by predicting actual delinquency. This is the task of the last group of tests.

B. Delinquency in the Longer Run

Is ex‐flag delinquency more likely than cum‐flag credit scores predict? That is, using the full‐information scores from before the flag removal, what is the effect of the postremoval increase in credit on actual delinquency? To find out, we predict whether a consumer is delinquent on a post‐flag‐removal date, using three predictors: the cum‐flag score, the lagged value of the same measure of delinquency, and the indicator for flag‐losing filers. The empirical question is whether losing the flag leads to greater delinquency down the road.

We say that consumer c is delinquent on date t if the consumer has an open trade whose present status is 30 or more days delinquent. In that case, the value of is 1; otherwise it is 0. The prediction model is a Probit, where the dependent variable is and the explanatory variables are the FICO score and the value of D as of the initial date 7/94 and the previous indicator LOSEFLAG. We fit this model for the sample used in the regressions of table 3, panel A, and report the results in table 3, panel B. The first model predicts delinquency as of 4/96, and the second model predicts delinquency as of 9/97, both using 7/94 information. A positive loading means that the probability of delinquency goes down as the variable goes up.

The initial FICO scores come in significantly positive, as expected. Higher scores imply a lower probability of delinquency 21 and 38 months later. The lagged value of D is significantly negative, which is also expected. There is no separate influence of LOSEFLAG as of 4/96, but by 9/97 it predicts extra delinquency. Controlling for a consumer's score and delinquency at 7/94, we find greater delinquency by 9/97 if this person's bankruptcy information disappeared by 4/96. That is, dropping the flag increases the longer‐run incidence of delinquency. Running the analogous test on the Geographic Panel, reported in panel B of table 4, we find that, after 2 years, LOSEFLAG relates to greater delinquency, with a p‐value of 10.5%.

Panel B of tables 3 and 4 show a greater incidence of delinquency among ex‐flag consumers than cum‐flag scores predict. The underlying cause suggested by tables 1 and 2 is that flag removal allows consumers to take down more credit than cum‐flag scores predict. We can test for this effect in the Bankruptcy Panel with regressions predicting ex‐flag credit in 4/96 and 9/97 with the predictive variables from panel B of table 3: the FICO score and the lagged dependent variable in 7/94, and the indicator LOSEFLAG. The fitted value of LOSEFLAG shows ex‐flag credit net of the credit predicted by cum‐flag credit and scores. Three measures of credit are considered: total outstanding balance on all open trades, total outstanding balance on all open bank‐card trades, and total credit limit on all open bank‐card trades, denoted , , and , respectively, for consumer c on date t. Results are in panel C of table 3.

The predictive regressions show substantially more ex‐flag borrowing than cum‐flag scores and borrowing predict. Total borrowing is $4,956 more in the short term, $8,208 more in the longer term, and about a quarter of this extra borrowing is with bank cards.15 The extra bank‐card borrowing is about two‐thirds of the extra bank‐card limit. So the excess probability that postflag consumers go delinquent coincides with an accumulation of excess debt, consistent with the latter causing the former.

The final test is for the influence of state exemptions. States with higher exemptions allow consumers to keep more property from creditors, so strategic consumers (in the sense of Fay et al. 2002) see more benefit from delinquency if they live in these states. Sorting consumers by their states’ exemptions, we should expect to see the excess credit provoke relatively more delinquency among the high‐exemption filers. This would be consistent with the finding of Gropp, Scholz, and White (1997) that car‐loan interest rates are higher for debtors in high‐exemption states. Only the Geographic Panel indicates state of residence, so we can test for this effect only in its small sample.

The first step is to classify states as high‐ and low‐exemption. For each state and the District of Columbia we add up the homestead and nonhomestead exemptions, catalogued by Gropp et al. (1997), using the federal exemption wherever it is higher. If consumer c resides in a state at or above the median total exemption, the value of is set to one, and otherwise it is zero. We test whether the effect of flag removal on future delinquency is stronger in the higher‐exemption states by starting with the Probit model of panel B of table 4, and adding as additional explanatory variables both and the interaction . The empirical question is whether the interaction enters negatively, that is, whether it predicts delinquency. The results are in panel C of table 4.

The Probit model never finds LOSEFLAG to be significant by itself, but the interaction with the exemption level enters negatively at the latest date, 24 months post flag.16 While the sample is small and the potential for interference from other cross‐state effects cannot be ruled out, this supports the strategic view of consumers’ use of credit in which they impute their legal rights to keep value into their credit usage.