Full text
United States Government Accountability Office
Report to Congressional Committees
September 2023
CREDIT CARDS
Pandemic Assistance
Likely Helped Reduce
Balances, and Credit
Terms Varied among
Demographic Groups
Accessible Version
GAO-23-105269
September 2023
GAO Highlights CREDIT CARDS
Highlights of GAO-23-105269, a report to Pandemic Assistance Likely Helped Reduce
congressional committees
Balances, and Credit Terms Varied among
Demographic Groups
Why GAO Did This Study What GAO Found
Credit cards are the most common Approximately half of active credit card accounts carried a balance during the
consumer lending product by number period from June 2013 through 2019, according to GAO’s analysis of a
of users, with 82 percent of U.S. adults nongeneralizable sample of more than 650,000 credit card accounts from the
holding a credit card in 2022. However, Board of Governors of the Federal Reserve System. This included almost 35
credit card adoption rates vary by race, percent of all active accounts in the highest credit score category (720 and
ethnicity, and income. Consumers can above) and more than 30 percent of all active accounts in zip codes with median
use credit cards as a convenient
household incomes of $150,000 or more.
means of payment and source of
credit. Some consumers do not always After the onset of the COVID-19 pandemic in March 2020, cardholders in the
pay off their monthly credit card GAO sample generally paid down credit card balances and carried balances for
balances and can accumulate interest shorter periods, according to GAO’s analysis. Specifically, the share of all active
and fees over time, which can lead to accounts that carried a balance declined from 50 to 45 percent from April 2020 to
debt burden and affect their financial December 2021 (see figure). Federal pandemic assistance likely contributed to
health. In addition, the COVID-19 these improvements. GAO analysis suggests that after March 2020, cardholders
pandemic caused significant economic who carried balances increased their average credit card payments during the
disruptions and has affected months when pandemic assistance payments were disbursed.
consumers’ credit card usage.
GAO was asked to review consumer Percentage of Active Credit Card Accounts That Carried a Balance, 2019–2021
credit card usage. This report
examines (1) consumer credit card
usage from 2013 through 2019, (2)
how the COVID-19 pandemic and
related assistance affected credit card
usage from March 2020 through
December 2021, and (3) how credit
card costs and usage vary among
racial/ethnic groups.
GAO analyzed a nongeneralizable
sample of credit card data from the
Federal Reserve for June 2013–
December 2021 and used the Census
Bureau’s American Community
Surveys to estimate the median
household income and racial and Note: GAO’s analysis was based on a nongeneralizable sample of active general purpose credit card
accounts. This figure excludes the percentage of active credit card accounts that were seriously
ethnic composition in cardholders’ zip delinquent (90 or more days), which was less than 1 percent during this period.
codes. GAO also reviewed research
from the Consumer Financial Cardholder accounts in the sample that were in billing zip codes with a majority
Protection Bureau, Federal Reserve, of Black or African American or Hispanic or Latino residents likely had higher
and academics. Further, GAO interest rates and lower credit limits and carried balances longer compared with
interviewed representatives of federal accounts in predominantly White zip codes, as indicated by GAO analysis. For
agencies, six large credit card issuers, example, the difference in interest rates was on average about 1.3 percentage
three credit reporting agencies, a points. Cardholders in the sample that were in majority-Black or -Hispanic zip
banking association, and a consumer codes continued to face higher interest rates and lower credit limits as compared
advocacy organization. with cardholders in predominantly White zip codes who had the same credit
scores, zip-code income distribution, and revolving status. While accounts in the
View GAO-23-105269. For more information, sample that were in majority-Black or -Hispanic zip codes carried smaller
contact Alicia Puente Cackley at (202) 512- balances than accounts in predominantly White zip codes, higher interest rates
8678 or cackleya@gao.gov. combined with carrying balances longer can result in higher credit costs.
United States Government Accountability Office
Contents
GAO Highlights ii
Why GAO Did This Study ii
What GAO Found ii
Letter 1
Background 4
A Large Share of Active Cards in Our Sample Carried a Balance
before the Pandemic, and Many Continued to Do So for Long
Periods 9
Cardholders in Our Sample Generally Paid Down Balances during
the Pandemic 21
Cardholders in Our Sample in Areas with Majority-Black or
Hispanic Residents Likely Had Less Favorable Credit Terms
and Carried Balances Longer 32
Agency Comments 45
Appendix I: Objectives, Scope, and Methodology 46
Federal Reserve and Census Bureau Data 46
Analysis of Credit Card Account Status 47
Analysis of Revolving and Transacting Episode Durations 48
Analysis of Credit Cost Differences 49
Analysis of Credit Card Usage in Association with Pandemic-
Related Responses 50
Data Reliability Assessments 51
Appendix II: Data and Descriptive Statistics 53
Federal Reserve Credit Card Data 53
Census Bureau Income and Race/Ethnicity Data 58
Descriptive Statistics 59
Analysis of Credit Card Usage 61
Appendix III: Cluster Analysis of Race/Ethnicity in Cardholders’ Zip Codes 64
Data 64
Methodology 64
Results 66
Appendix IV: Econometric Analysis of Revolving and Transacting Episode Durations 67
Data and Definitions 67
Episode Durations 69
Page i GAO-23-105269 Credit Cards
Methodology of Analysis of Revolving Episodes under
Prepandemic and Pandemic Conditions 75
Assumptions and Limitations 80
Results 81
Methodology of Analysis of Revolving Episodes by Race/Ethnicity
in Cardholders’ Zip Codes 90
Appendix V: Econometric Analysis of Credit Card Payments 113
Methodology 114
Limitations 116
Results 117
Appendix VI: Econometric Analysis of Credit Card Terms and Revolving Balances by Race/Ethnicity in
Cardholders’ Zip Codes 121
Data and Variable Definitions 121
Methodology 122
Assumptions and Limitations 124
Results 125
Appendix VII: GAO Contact and Staff Acknowledgments 134
Tables
Table 1: Relationships Used to Impute Credit Card Data Variable
Observations with Missing Values 55
Table 2: Information on General Purpose Credit Card Account
Data Analyzed by GAO 57
Table 3: Characteristics of Consumer Credit Card Accounts, June
2013–December 2021 59
Table 4: Racial and Ethnic Composition of Zip Code Clusters,
September 2016 66
Table 5: Observed Revolving Episode Durations for June 2013–
December 2021 70
Table 6: Estimated Coefficients of Credit Card Payment Amounts
Made by Revolving Accounts 117
Table 7: Estimated Differences in Credit Card Interest Rates by
Race and Ethnicity in Cardholders’ Billing Zip Codes,
June 2013–December 2021 126
Table 8: Estimated Differences in Credit Limits by Race and
Ethnicity in Cardholders’ Billing Zip Codes, June 2013–
December 2021 128
Table 9: Estimated Differences in Revolving Balances by Race
and Ethnicity in Cardholders’ Billing Zip Codes, June
2013–December 2021 130
Page ii GAO-23-105269 Credit Cards
Table 10: Estimated Differences in Credit Terms and Revolving
Balances by Race and Ethnicity in Cardholders’ Billing Zip
Codes, June 2013–December 2021 133
Figures
Percentage of Active Credit Card Accounts That Carried a
Balance, 2019–2021 ii
Figure 1: Disbursement Timeline and Amounts of Federal
Pandemic Assistance 8
Data for Figure 1: Disbursement Timeline and Amounts of Federal
Pandemic Assistance 8
Figure 2: Percentage of Active Credit Card Accounts That Did and
Did Not Carry a Balance, June 2013–December 2019 10
Data for Figure 2: Percentage of Active Credit Card Accounts That
Did and Did Not Carry a Balance, June 2013–December
2019 10
Figure 3: Average Percentage of Active Credit Card Accounts
That Carried a Balance, by Credit Score of Account
Holder, June 2013–December 2019 14
Data for Figure 3: Average Percentage of Active Credit Card
Accounts That Carried a Balance, by Credit Score of
Account Holder, June 2013–December 2019 14
Figure 4: Average Percentage of Active Credit Card Accounts
That Carried a Balance, by Median Household Income of
Cardholder’s Zip Code, June 2013–December 2019 15
Data for Figure 4: Average Percentage of Active Credit Card
Accounts That Carried a Balance, by Median Household
Income of Cardholder’s Zip Code, June 2013–December
2019 15
Figure 5: Credit Card Median Nominal Revolving Balance, June
2013–December 2019 17
Data for Figure 5: Credit Card Median Nominal Revolving
Balance, June 2013–December 2019 17
Figure 6: Simulated Costs for a Cardholder Who Carries a
Balance for Different Lengths of Time 20
Figure 7: Percentage of Active Credit Card Accounts That Carried
a Balance, 2019–2021 22
Data for Figure 7: Percentage of Active Credit Card Accounts That
Carried a Balance, 2019–2021 22
Figure 8: Simulated Length of Credit Card Revolving Episodes
and Status after End of Revolving Episodes under
Prepandemic and Pandemic Conditions 26
Page iii GAO-23-105269 Credit Cards
Data for Figure 8: Simulated Length of Credit Card Revolving
Episodes and Status after End of Revolving Episodes
under Prepandemic and Pandemic Conditions 26
Figure 9: Median Nominal Credit Card Revolving Balances, 2020–
2021 28
Data for Figure 9: Median Nominal Credit Card Revolving
Balances, 2020–2021 28
Figure 10: Estimated Average Interest Rate Differences for
Cardholder Accounts by Race and Ethnicity of Billing Zip
Codes, June 2013–December 2021 35
Data for Figure 10: Estimated Average Interest Rate Differences
for Cardholder Accounts by Race and Ethnicity of Billing
Zip Codes, June 2013–December 2021 35
Figure 11: Estimated Average Credit Limit Differences for
Cardholder Accounts by Race and Ethnicity in Billing Zip
Codes, June 2013–December 2021 37
Data for Figure 11: Estimated Average Credit Limit Differences for
Cardholder Accounts by Race and Ethnicity in Billing Zip
Codes, June 2013–December 2021 37
Figure 12: Simulated Length of Credit Card Revolving Episodes
and Status after End of Revolving Episodes, by Race and
Ethnicity in Billing Zip Codes, June 2013–December 2019 41
Data for Figure 12: Simulated Length of Credit Card Revolving
Episodes and Status after End of Revolving Episodes, by
Race and Ethnicity in Billing Zip Codes, June 2013–
December 2019 42
Figure 13: Estimated Revolving Balance Differences for
Cardholder Accounts by Race and Ethnicity of Billing Zip
Codes, June 2013–December 2021 44
Data for Figure 13: Estimated Revolving Balance Differences for
Cardholder Accounts by Race and Ethnicity of Billing Zip
Codes, June 2013–December 2021 44
Figure 14: Examples of Revolving Episodes of Differing Lengths 62
Data for Figure 14: Examples of Revolving Episodes of Differing
Lengths 63
Figure 15: Illustrative Example of Credit Card Billing Cycle 68
Figure 16: Length of Median Revolving Episode for Accounts in
Revolving Status by Month 71
Data for Figure 16: Length of Median Revolving Episode for
Accounts in Revolving Status by Month 71
Page iv GAO-23-105269 Credit Cards
Figure 17: Multinomial Logit Combined Coefficients for Transitions
from Revolving Episode to Next Status, under
Prepandemic and Pandemic Conditions 83
Data for Figure 17: Multinomial Logit Combined Coefficients for
Transitions from Revolving Episode to Next Status, under
Prepandemic and Pandemic Conditions 84
Figure 18: Multinomial Logit Combined Coefficients for Transitions
from Revolving Episode to Next Status, by Race and
Ethnicity in Cardholders’ Billing Zip Codes 93
Data for Figure 18: Multinomial Logit Combined Coefficients for
Transitions from Revolving Episode to Next Status, by
Race and Ethnicity in Cardholders’ Billing Zip Codes 94
Abbreviations
ACS American Community Survey
CFPB Consumer Financial Protection Bureau
This is a work of the U.S. government and is not subject to copyright protection in the
United States. The published product may be reproduced and distributed in its entirety
without further permission from GAO. However, because this work may contain
copyrighted images or other material, permission from the copyright holder may be
necessary if you wish to reproduce this material separately.
Page v GAO-23-105269 Credit Cards
441 G St. N.W.
Washington, DC 20548
Letter
September 29, 2023
The Honorable Sheldon Whitehouse
Chairman
Committee on the Budget
United States Senate
The Honorable Bernard Sanders
Chair
Committee on Health, Education, Labor, and Pensions
United States Senate
Credit cards are the largest consumer lending product by number of
users, with 82 percent of U.S. adults holding a credit card in 2022.1
However, credit card adoption rates vary by income, race, and ethnicity.
Credit cards can provide consumers with a convenient means of paying
for goods and services and accessing credit. Some consumers carry
credit card balances from month to month. These consumers can
accumulate interest and fees over time, which could lead to debt burden
and affect their financial health. For example, the Consumer Financial
Protection Bureau (CFPB) estimated that Americans paid about $120
billion per year in credit card interest and fees from 2018 to 2020, which is
about $1,000 per year for each American household.
Since 2020, public policies, federal and private-sector assistance, and
restrictions prompted by the COVID-19 pandemic have affected
individuals’ financial behavior, including credit card usage. Early in the
pandemic, consumer reactions to the pandemic, public health policies
and restrictions such as stay-at-home orders, and temporary business
closures caused significant economic disruptions nationwide. In
response, the federal government and credit card issuers took actions to
provide financial assistance and debt relief (such as payment deferrals) to
consumers.
You asked us to examine credit card usage in the U.S. This report
examines (1) credit card usage from 2013 through 2019 and the
characteristics of cardholders who carried balances during this period, (2)
1See Consumer Financial Protection Bureau, The Consumer Credit Card Market
(September 2021) and Board of Governors of the Federal Reserve System, Economic
Well-Being of U.S. Households in 2022 (May 2023).
Page 1 GAO-23-105269 Credit Cards
Letter
how the COVID-19 pandemic and related assistance affected credit card
usage from March 2020 through December 2021, and (3) how credit card
costs and usage vary among racial and ethnic groups.
To address all three objectives, we analyzed a 0.1 percent
nongeneralizable sample of consumer credit card account data,
containing information on more than 650,000 individual credit card
accounts, from the Board of Governors of the Federal Reserve System’s
Capital Assessments and Stress Testing Report (FR Y-14M) for the
period from June 2013 through December 2021.2 Our analysis included
only general purpose credit cards, which are credit cards issued on
networks, such as Visa, Mastercard, American Express, and Discover,
which are accepted by a wide variety of merchants.3 The Federal Reserve
data allowed us to examine credit card usage by account, but the data did
not have information that would allow us to link multiple accounts held by
an individual cardholder. See appendixes I and II for additional detail on
the Federal Reserve data used in our analysis.
Additionally, because Federal Reserve data did not include updated
incomes or the race or ethnicity of the cardholders, we supplemented the
Federal Reserve data with data from the Census Bureau’s 5-Year
American Community Surveys for 2013–2020. We used census data on
household income and the shares of residents in each zip code who were
Asian, Black or African American (referred to as “Black” in this report),
Hispanic or Latino (referred to as “Hispanic” in this report), White, and
Other.4 We developed five scenarios representative of different,
commonly occurring racial and ethnic compositions of zip codes:
predominantly White (86 percent White), majority Black or African
2The Federal Reserve first started collecting the Y-14M data in June 2012. However, our
analysis used data beginning in June 2013 because the Federal Reserve did not collect
some data elements until that time. The data are collected from bank holding companies,
savings and loan holding companies, and intermediate holding companies with $100
billion or more assets (which generally include all of the largest credit card issuers,
according to Federal Reserve staff). As such, we cannot generalize our analysis results to
credit card accounts in all banks in the U.S.
3Another type of credit card, referred to as a private label card, can only be used at one
merchant or a small group of related merchants. American Express and Discover are both
payment networks and credit card issuers.
4For this report, we derived the race and ethnicity categories from those used in the
Census Bureau’s 5-Year American Community Surveys for 2013–2020. The Black, White,
and Asian race categories are all non-Hispanic. Census Bureau, American Community
Survey and Puerto Rico Community Survey Design and Methodology (November 2022).
Page 2 GAO-23-105269 Credit Cards
Letter
American (58 percent Black), majority Hispanic or Latino (62 percent
Hispanic), mixed race and ethnicity with majority White (59 percent
White), and mixed race and ethnicity with no majority. See appendix III for
additional detail on the zip-code-level race and ethnicity data and our
analysis.
Additionally, we used the Federal Reserve and census data to construct
econometric models to estimate the length of time accounts would have
carried a balance and how credit terms and revolving balances varied
with the racial and ethnic composition of the cardholders’ billing zip
codes, among other things.5 In our analysis of credit terms, we controlled
for cardholders’ credit scores and revolving status and for incomes in
cardholders’ zip codes.6 Similarly, in our analysis of revolving balances,
we controlled for cardholders’ credit scores and the incomes in
cardholders’ zip codes. Our review was not designed to examine whether
any differences in credit card terms and revolving patterns among racial
and ethnic groups resulted from fair lending disparities, which cannot be
measured as independent factors in our analysis. See appendixes IV
through VI for additional information on the methodology and results of
our econometric models.
We also reviewed research related to consumer credit card usage
conducted by CFPB, the Federal Reserve, and academics, including
information on consumers’ use of credit products in general and by race
and ethnicity. We interviewed staff from CFPB and the Federal Reserve
and representatives from six large credit card issuers, three credit
reporting agencies, the American Bankers Association, and the National
5To assess the reliability of the Federal Reserve and census data, we reviewed technical
documentation and interviewed agency staff knowledgeable about the data. We also
conducted electronic testing of the data to assess its reliability and limitations. We
determined that the data were sufficiently reliable for examining credit card usage,
describing household income and the racial and ethnic composition of cardholders’ zip
codes, and describing changes in cardholder payment behavior during the disbursements
of federal pandemic assistance.
6Because the Federal Reserve data did not include updated incomes of cardholders, we
controlled for the household incomes in cardholders’ zip codes by measuring the
percentage of households whose income fell into each of 16 income groups used in our
analysis.
Page 3 GAO-23-105269 Credit Cards
Letter
Consumer Law Center.7 To obtain information on how federal and private-
sector pandemic assistance contributed to changes in credit card usage
during the COVID-19 pandemic from March 2020 through December
2021, we reviewed reports from CFPB and the Federal Reserve. We also
examined data from the Census Household Pulse Survey conducted from
April 23, 2020, to October 11, 2021, and the 2020 annual filings from the
six issuers we interviewed.8 See appendix I for additional detail on our
objectives, scope, and methodology.
We conducted this performance audit from May 2021 to September 2023
in accordance with generally accepted government auditing standards.
Those standards require that we plan and perform the audit to obtain
sufficient, appropriate evidence to provide a reasonable basis for our
findings and conclusions based on our audit objectives. We believe that
the evidence obtained provides a reasonable basis for our findings and
conclusions based on our audit objectives.
Background
Overview of the Credit Card Market
In October 2021, credit cards, debit cards, and cash on hand were the
three most common ways for consumers to make payments for bills and
purchases, according to the Survey and Diary of Consumer Payment
7The six issuers were American Express, Bank of America, Capital One, Citibank,
Discover, and JPMorgan Chase. We selected issuers that had the largest purchase
volumes and outstanding balances in their credit card portfolios in 2020. The three credit
reporting agencies were Equifax, Experian, and TransUnion.
8The Household Pulse Survey, an experimental data product, is an interagency federal
statistical rapid response survey to measure household experiences during the COVID-19
pandemic. The Census Bureau conducts this survey in partnership with five other
agencies from the Federal Statistical System. Response rates over our period of analysis
ranged from 2.3 percent to 7.5 percent. Census applied weighting adjustments to mitigate
nonresponse bias, according to Census. All reported estimates have a relative margin of
error of 19 percent or less of the estimate, and all reported comparisons are statistically
significant at the 95 percent confidence level.
Page 4 GAO-23-105269 Credit Cards
Letter
Choice.9 In October 2021, 28 percent of payments were made with credit
cards and 29 percent with debit cards.
From 2013 through 2021, the number of consumer credit card accounts
steadily increased from about 383 million to about 532 million, and total
outstanding credit card balances were $856 billion at year-end 2021,
according to the Federal Reserve Bank of New York’s Consumer Credit
Panel.10 The top 10 credit card issuers held 82 percent of the outstanding
credit card balances in 2021, according to a Federal Reserve report.11
Total outstanding credit card balances were relatively small compared to
other types of consumer loans. Outstanding credit card balances
accounted for 5 percent of the dollar value of consumer loans at year-end
2021, while mortgages and auto loans accounted for 70 percent and 9
percent, respectively, according to the Consumer Credit Panel. However,
the interest rates of credit cards tend to be much higher than those of
other types of consumer loans. For example, the Federal Reserve
reported that in December 2021, the average interest rate for credit cards
was about 16 percent, compared with about 5 percent for new auto
loans.12
In addition, an estimated 84 percent of U.S. adults had a credit card
account in 2021, according to a Federal Reserve report.13 Half of
consumers had one or two credit cards, and 20 percent had five or more
in 2021, as estimated by the Survey and Diary of Consumer Payment
9Federal Reserve Bank of Atlanta, The 2021 Survey and Diary of Consumer Payment
Choice: Summary Results (Atlanta, GA: Sept. 17, 2022). The Survey and Diary of
Consumer Payment Choice is an annual survey, and the 2021 survey was the most
recently published survey at the time of our report.
10The Federal Reserve Bank of New York’s Consumer Credit Panel is a nationally
representative 5 percent random sample of all individuals whose credit file includes a
Social Security number (usually age 18 and over). According to the bank, the Consumer
Credit Panel does not contain lender names, and the data are anonymized.
11Board of Governors of the Federal Reserve System, Report to Congress: Profitability of
Credit Card Operations of Depository Institutions (July 2022).
12Board of Governors of the Federal Reserve System, G.19 Statistical Release,
“Consumer Credit: March 2022” (May 6, 2022).
13Board of Governors of the Federal Reserve System, Economic Well-Being of U.S.
Households in 2021 (May 2022).
Page 5 GAO-23-105269 Credit Cards
Letter
Choice.14 People with higher and middle-range incomes were more likely
to have a credit card, and people with lower incomes were less likely,
according to a 2021 Federal Reserve Survey.15
The survey also estimated that credit card ownership varied by race and
ethnicity: 93 percent for Asian adults, 88 percent for White adults, 77
percent for Hispanic adults, and 72 percent for Black adults. Lower rates
of credit card adoption among Black and Hispanic adults are consistent
with lower rates of bank account adoption among Black and Hispanic
households.16 In addition, unbanked households are much less likely to
have a credit card, according to the Federal Deposit Insurance
Corporation’s 2021 National Survey of Unbanked and Underbanked
Households.17
Key Definitions of Credit Card Use
In this report, we use the term “balance” to refer to the credit card balance
at the end of a billing cycle. We classify credit card accounts into six
groups based on how cardholders use the cards and whether interest is
charged during each month. See appendix II for additional information.
· Transacting accounts. If the cardholder pays all of the account
balance due for the billing cycle by the due date, we refer to the
cardholder as a “transactor” and to the account as a “transacting
account.”18 Balances for transacting accounts consist of purchases
made during the billing cycle. Repeat transactors are not charged
14Federal Reserve Bank of Atlanta, 2021 Survey and Diary.
15Board of Governors of the Federal Reserve System, Economic Well-Being of U.S.
Households in 2021. The CARD Act of 2009 requires credit card issuers to consider the
consumer’s ability to make required payments before opening a credit card account for the
consumer. Credit Card Accountability Responsibility and Disclosure Act of 2009, Pub. L.
No. 111-24, § 109, 123 Stat. 1734, 1743 (codified at 15 U.S.C § 1665e).
16See GAO, Banking Services: Regulators Have Taken Actions to Increase Access, but
Measurement of Actions’ Effectiveness Could Be Improved, GAO-22-104468
(Washington, D.C.: Feb. 14, 2022).
17Federal Deposit Insurance Corporation, 2021 FDIC National Survey of Unbanked and
Underbanked Households (October 2022).
18We assign this status primarily by identifying accounts with a positive starting balance
but no finance charges due. As a consequence of our assignment approach, any accounts
that are carrying a balance that has a zero percent promotional annual percentage rate
are also classified as transacting accounts.
Page 6 GAO-23-105269 Credit Cards
Letter
interest on the end-of-cycle balance or on new purchases during the
grace period (between the end of a billing cycle and the date payment
is due).
· Revolving accounts. If the cardholder pays less than the entire
account balance due for the billing cycle, or if the cardholder account
has been delinquent for less than 90 days, we refer to the cardholder
as a “revolver” and to the account as a “revolving account.”19 In
contrast with transactors, revolvers are charged interest on the
remaining unpaid balances from prior cycles (also called the revolving
balances), and immediately on any new purchases they make in the
current cycle.20
· Seriously delinquent accounts. If the cardholder does not make at
least the minimum payment by the due date, the cardholder is in
delinquency. For the purposes of this report, we classify accounts as
“revolving accounts” if they are less than 90 days delinquent, and as
“seriously delinquent” if they are 90 or more days delinquent.
· Inactive accounts. If the cardholder does not have a balance from
the previous billing cycle (i.e., did not make purchases prior to the
closing of the billing cycle) and has not made any payments in the
current month, we classify the account as “inactive.”
· Closed accounts. If an account can no longer be used because the
cardholder has closed it or because the cardholder has died, we
classify the account as “closed.”
· Charged-off accounts. If the account balance has been charged off
by the issuer because it is deemed unlikely to be collected, we
classify the account as “charged off.”
Credit cardholders can transition from month to month among any of the
first four groups.
19We assign this status primarily by identifying accounts with finance charges incurred in
the billing cycle that are not delinquent for more than 90 days. As a consequence of our
assignment approach, accounts that were previously carrying a balance and have started
to pay off their balance in full are not classified as transacting accounts until they are
eligible for their issuer’s grace period.
20Revolving balances can also include fees if cardholders fail to make payments by the
due date.
Page 7 GAO-23-105269 Credit Cards
Letter
Federal and PrivateSector Assistance Related to the
COVID19 Pandemic
In response to the COVID-19 pandemic, from April 2020 to December
2021 the federal government provided eligible consumers with direct
payments, including one-time economic impact payments, multiple
advance child tax credit payments, and expanded unemployment
insurance in varying amounts (see fig. 1).21 The amounts of the economic
impact payments and advance child tax credit payments were generally
based on an individual’s or household’s income or family status.
Figure 1: Disbursement Timeline and Amounts of Federal Pandemic Assistance
Data for Figure 1: Disbursement Timeline and Amounts of Federal Pandemic
Assistance
Type of Assistance Amounts
Economic impact payment 1 $1,200–$2,400, April 2020 – February 2021
Economic impact payment 2a $600–$1,200, January 2021 – February 2021
Economic impact payment 3 $1,400–$2,800, March 2021 – December 2021
Advance child tax credit $2,000–$3,600, July 2021 – December 2021
Expanded unemploymentb $600 per week, April 2020 – July 2021
$300 per week, January 2021 – August 2021
Source: GAO. | GAO-23-105269
Note: The amounts of the economic impact payments and the advance child tax credit payments
were based generally on an individual’s or household’s income. Individuals or households with
dependents could receive higher amounts of assistance if eligible.
a
The second economic impact payments began disbursements on Dec. 29, 2020.
21Some states terminated their participation in expanded unemployment insurance
programs starting in June 2021, while others participated through September 6, 2021,
when the expanded unemployment insurance programs ended.
Page 8 GAO-23-105269 Credit Cards
Letter
b
In response to the COVID-19 pandemic, the federal government created three temporary
unemployment insurance programs that expanded benefit eligibility and enhanced benefits.
Specifically, under the Federal Pandemic Unemployment Compensation program, weekly benefits of
$600 were available for weeks of unemployment from Mar. 29, 2020, through July 25, 2020, and
weekly benefits of $300 were available from Dec. 26, 2020, through Sept. 6, 2021. Some states
terminated their participation in the expanded unemployment insurance programs starting in June
2021, while others participated through Sept. 6, 2021.
The federal government also granted a temporary suspension of loan
payments for borrowers of certain mortgages and student loans and
enacted a moratorium on certain evictions during the pandemic.
Additionally, the Small Business Administration’s Paycheck Protection
Program supported employment by providing forgivable loans to small
businesses.22
Although not mandated to provide assistance, credit card issuers
generally offered voluntary debt assistance programs to cardholders who
requested them from March to December 2020, according to a CFPB
report and six credit card issuers.23 These programs included payment
deferrals and fee waivers. CFPB reported that about 25 million consumer
credit card accounts (with a total of $68 billion in balances) enrolled in
payment relief programs in 2020.
A Large Share of Active Cards in Our Sample
Carried a Balance before the Pandemic, and
Many Continued to Do So for Long Periods
About Half of Cards Carried a Balance from Month to
Month, Including Many Held by Cardholders with High
Credit Scores and Incomes
From June 2013 through December 2019, in any given month,
approximately half of the active credit card accounts were revolving
(carrying a balance), according to our analysis of a sample of more than
22For more information, see GAO, Paycheck Protection Program: Program Changes
Increased Lending to the Smallest Businesses and in Underserved Locations,
GAO-21-601 (Washington, D.C.: Sept. 21, 2021). For more information on federal
pandemic assistance, see GAO’s CARES Act work at https://www.gao.gov/coronavirus.
23Consumer Financial Protection Bureau, The Consumer Credit Card Market.
Page 9 GAO-23-105269 Credit Cards
Letter
650,000 individual credit card accounts from Federal Reserve data.24
These included many held by cardholders with high credit scores and
likely with relatively high incomes.25 The percentage of both revolving and
transacting accounts stayed stable year over year in this period (see fig.
2).26
Figure 2: Percentage of Active Credit Card Accounts That Did and Did Not Carry a Balance, June 2013–December 2019
Data for Figure 2: Percentage of Active Credit Card Accounts That Did and Did Not
Carry a Balance, June 2013–December 2019
Year Revolver Transactor
2013 53.5354 46.1583
24As previously discussed, our analysis was based on a nongeneralizable sample of
active general purpose credit card accounts. We discuss credit card usage during the
COVID-19 pandemic (2020–2021) later in this report.
25We approximated cardholders’ incomes by using the median household income in the
cardholder’s billing zip code.
26During this period, a substantial number of accounts were inactive each month. While
the share of total accounts that were inactive generally declined over time, it ranged from
27 percent to 36 percent of the total accounts each month. Additionally, accounts that
were seriously delinquent (90 or more days) ranged from 0.3 percent to 0.6 percent over
this period.
Page 10 GAO-23-105269 Credit Cards
Letter
Year Revolver Transactor
53.3373 46.3064
53.1632 46.4242
53.2722 46.298
53.4851 46.0451
53.2951 46.2354
53.0735 46.463
2014 53.1889 46.3683
53.3097 46.2697
52.7815 46.8174
51.8243 47.8193
51.5441 48.1214
51.7238 47.9584
51.8703 47.7638
51.8325 47.7389
52.0991 47.4675
52.1125 47.4382
51.8277 47.733
51.7356 47.7806
2015 52.0283 47.5007
52.2111 47.3497
51.7312 47.8642
50.9665 48.6309
50.7651 48.8403
51.0272 48.5764
51.3016 48.2771
51.0259 48.5175
51.2484 48.2583
51.5454 47.9278
51.3887 48.0906
51.2716 48.1987
2016 51.5733 47.9195
51.9864 47.5437
51.3267 48.2228
50.5998 48.9946
50.2236 49.3992
50.3242 49.3029
50.8904 48.6883
Page 11 GAO-23-105269 Credit Cards
Letter
Year Revolver Transactor
50.8964 48.6251
51.0137 48.4749
51.1839 48.2827
51.2838 48.1752
51.1045 48.3532
2017 51.5059 47.9317
51.8286 47.6534
51.5817 47.9364
50.7263 48.8317
50.5928 48.9842
50.8405 48.7475
51.2432 48.3214
51.304 48.2288
51.52 47.9634
51.6629 47.8022
51.5601 47.8914
51.3753 48.0639
2018 51.9553 47.4776
52.1045 47.3769
51.6904 47.8336
50.6503 48.9144
50.2415 49.324
50.5024 49.0804
50.9197 48.6498
51.0219 48.5401
51.0756 48.4484
51.1515 48.3432
51.1711 48.3115
51.0596 48.379
2019 51.2701 48.1578
51.284 48.1683
50.8711 48.6306
49.8781 49.6605
49.6276 49.9372
49.8768 49.7257
50.2576 49.3261
50.2413 49.3072
Page 12 GAO-23-105269 Credit Cards
Letter
Year Revolver Transactor
50.3532 49.1493
50.4044 49.0972
50.3681 49.1299
50.2541 49.2159
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of active general purpose credit card
accounts. This figure excludes the percentage of active credit card accounts that were seriously
delinquent (90 or more days), which was less than 1 percent during this period.
Credit Score Categories
Credit scores are used by lenders to determine consumers’ eligibility for credit, including credit
cards. Credit scores can be classified into categories based on the consumer’s creditworthiness. A
consumer with a higher credit score is considered more likely to repay a debt. For this report, we
applied the credit score categories used by the Consumer Financial Protection Bureau in their
credit card reporting:
· Deep subprime (below 580)
· Subprime (580–619)
· Near-prime (620–659)
· Prime (660–719)
· Super-prime (720 and above)
Source: GAO and Consumer Financial Protection Bureau. | GAO-23-105269
Active accounts held by cardholders with super-prime credit scores (720
and above) were less likely to carry a balance than those held by
cardholders with credit scores below 720, according to our analysis of
Federal Reserve data for June 2013–December 2019 (see fig. 3).27 About
35 percent of active accounts held by super-prime cardholders carried a
balance, compared with 74–86 percent of active accounts held by
cardholders in the lower credit score categories.
27The amount of unpaid debt, including credit card balances, is typically one of the factors
considered in determining a consumer’s credit score. There are no industry standard
credit score categories used by credit reporting agencies and credit card issuers. For this
report, we applied the credit score categories used by CFPB in its 2021 consumer credit
card market report.
Page 13 GAO-23-105269 Credit Cards
Letter
Figure 3: Average Percentage of Active Credit Card Accounts That Carried a
Balance, by Credit Score of Account Holder, June 2013–December 2019
Data for Figure 3: Average Percentage of Active Credit Card Accounts That Carried
a Balance, by Credit Score of Account Holder, June 2013–December 2019
Credit Score of Account Holder Percentage of active Percentage of
accounts that carried a total sample in
balance category
Deep subprime (below 580) 82.0865 7.34
Subprime (580-619) 85.5606 7.47
Near-prime (620-659) 82.1142 14.18
Prime (660-719) 73.7336 30.13
Super prime (720 and above) 34.5365 40.87
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of active general purpose credit card
accounts. For each credit score category, we first calculated the percentage of credit card accounts
that carried a balance out of the total number of active credit card accounts in that credit score
category for each month from June 2013 through Dec. 2019, and then took the average over all
months. Our credit score data included multiple types of credit scores, such as FICO and Vantage,
which were used and reported by credit reporting agencies and credit card issuers.
Furthermore, large percentages of active accounts, regardless of income
level, carried a balance from June 2013 through December 2019 (see fig.
Page 14 GAO-23-105269 Credit Cards
Letter
4). Accounts held by cardholders who lived in zip codes with a lower
median household income were more likely to carry a balance, according
to our analysis. However, many accounts in higher-income zip codes also
carried balances. For instance, about 33 percent of accounts in zip codes
with median household incomes of $150,000 or more carried a balance,
as did about 43 percent of accounts in zip codes with incomes between
$100,000 and $150,000.
Figure 4: Average Percentage of Active Credit Card Accounts That Carried a
Balance, by Median Household Income of Cardholder’s Zip Code, June 2013–
December 2019
Data for Figure 4: Average Percentage of Active Credit Card Accounts That Carried
a Balance, by Median Household Income of Cardholder’s Zip Code, June 2013–
December 2019
Median Household Income of Percentage of active Percentage of
Cardholder’s Zip Code accounts that carried a total sample in
balance category
Below $50,000 59.7566 20
$50,000-74,999 54.1459 40.29
$75,000-99,999 48.8519 23.62
Page 15 GAO-23-105269 Credit Cards
Letter
Median Household Income of Percentage of active Percentage of
Cardholder’s Zip Code accounts that carried a total sample in
balance category
$100,000-149,999 42.5256 14.76
$150,000 and above 33.4729 1.7
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and the Census Bureau’s American
Community Survey. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of active general purpose credit card
accounts. For each median household income category, we first calculated the percentage of credit
card accounts that carried a balance out of the total number of active credit card accounts in that
credit score category for each month from June 2013 through Dec. 2019, and then took the average
over all months.
Many Cardholders Carried a Balance for at Least a Year
Revolving Episode
A revolving episode is the period during which an account continuously carries a balance. For
example, an account that carries a balance from March to June has had a revolving episode of 4
months. For the purpose of our report, a revolving episode ends when a cardholder pays off the
full balance on the account, or the account transitions to seriously delinquent, inactive, or closed
status. Each time an account starts revolving again, the account begins a new revolving episode.
Source: GAO. | GAO-23-105269
Our analysis also found that most accounts that revolved carried a
revolving balance for at least a year. From June 2014 through December
2019, of accounts that were revolving in any given month, approximately
80 percent were in a long-term revolving episode (1 year or more).28
Further, the median nominal revolving balance grew from $1,530 in June
2013 to $1,720 in December 2019, or 12 percent (see fig. 5).29 After
28Because we defined long-term revolving episodes as 1 year or more, June 2014 was the
earliest month for which our data would allow us to make an accurate count of long-term
status. Similar to our definition of long-term status, a Federal Reserve study defines credit
card accounts that carried a balance for more than 6 of the past 12 months as “heavy
revolver accounts.” See Robert Adams and Vitaly Bord, “The Effects of the COVID-19
Shutdown on the Consumer Credit Card Market: Revolvers versus Transactors” (Board of
Governors of the Federal Reserve System, Oct. 21, 2020), accessed Jan. 24, 2022,
https://www.federalreserve.gov/econres/notes/feds-notes/the-effects-of-the-covid-19-
shutdown-on-the-consumer-credit-card-market-revolvers-versus-transactors-
20201021.htm.
29Cardholders are obligated to repay the nominal amount of their credit card balances.
Changes in nominal balances reflect changes in the price level and changes in the real
value of the balances.
Page 16 GAO-23-105269 Credit Cards
Letter
adjusting for inflation, the median revolving balance remained relatively
stable, growing by 2 percent during this period.30
Figure 5: Credit Card Median Nominal Revolving Balance, June 2013–December 2019
Data for Figure 5: Credit Card Median Nominal Revolving Balance, June 2013–
December 2019
Year Amount in dollars
2013 1529.92
1542.71
1569.41
1585.13
1575.38
1578.25
1592
2014 1596.47
1589.15
1559.37
1578.39
1581.68
30In December 2021 dollars, the median revolving balances were $1,827 in June 2013
and $1,867 in December 2019.
Page 17 GAO-23-105269 Credit Cards
Letter
1591
1585.53
1600.78
1625.90
1618.35
1630.33
1629.73
2015 1628.25
1618.56
1603.96
1618
1624.99
1618.67
1611.32
1619.63
1630.61
1624.92
1631.39
1632.71
2016 1627.92
1617.57
1600
1610.05
1616.27
1614.93
1616.37
1622.20
1654.10
1658.58
1658.13
1690.04
2017 1687.35
1677.19
1643.60
1651.74
1658.52
1650.17
1652.77
Page 18 GAO-23-105269 Credit Cards
Letter
1653.64
1661.94
1665.80
1677.79
1699.73
2018 1697.25
1698.05
1662.53
1679.01
1680.76
1662.10
1665.15
1668.90
1694.07
1676.68
1684.49
1689.96
2019 1705.92
1700.48
1679.65
1687.33
1685.70
1679.64
1707.82
1673.31
1686.58
1698.57
1712.90
1720.41
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of active general purpose credit card
accounts.
Cardholders with accounts that carried a balance from month to month
had different spending patterns than cardholders with accounts that did
not carry a balance. Our analysis of Federal Reserve data found that from
June 2013 through December 2019, fewer revolvers made purchases
with their credit cards and those who did generally spent less than
transactors. Specifically, on average, 55–64 percent of revolvers made
purchases in any given month during this period, compared with 79–84
Page 19 GAO-23-105269 Credit Cards
Letter
percent of transactors. In addition, among cardholders who made a
purchase, the monthly median purchase ranged from approximately
$164–$224 for revolvers and $436–$622 for transactors.
Cardholders can incur a relatively high cost of credit if they carry credit
card balances for a considerable amount of time (see fig. 6).
Figure 6: Simulated Costs for a Cardholder Who Carries a Balance for Different
Lengths of Time
Note: The average interest rate for credit cards that were assessed interest was 16.9 percent in Dec.
2019, according to the Board of Governors of the Federal Reserve System. We rounded the interest
rate to 17 percent and interest charges to whole dollar amounts. We assumed the cardholder did not
make new purchases on the given credit card while carrying the balance and paid off the balance in
equal monthly payments. For illustration purposes and ease of calculation, we developed these
examples by calculating the interest charges based on a monthly rate by dividing the annual interest
rate by 12. In practice, credit card interest charges are typically compounded on a daily basis.
According to our analysis of Federal Reserve data from June 2013
through December 2019, 5–11 percent of revolving accounts reached or
exceeded their credit limits in a given month. Credit cards offer
cardholders immediate access to credit, so cardholders with accounts
Page 20 GAO-23-105269 Credit Cards
Letter
that have exhausted their available credit may need to cut back or delay
spending, dip into their savings, or borrow from other sources. For
example, a 2021 CFPB survey estimated that half of consumers who had
difficulty paying bills or expenses borrowed either from a bank; from an
alternative financial service provider using an auto-title loan, a payday
loan, or through a pawn shop; or from friends and family.31 Of those who
borrowed, 21 percent used at least one form of alternative financial
services.
Some of these consumers likely faced higher borrowing costs. For
example, the average annual credit card interest rate was about 17
percent in December 2019.32 In contrast, the annual interest rate on a 2-
week payday loan of $100 with a typical $15 fee can be almost 400
percent.33 Furthermore, consumers who miss or delay paying a bill may
face delinquency on their credit records, which can increase the costs of
borrowing in the future, according to CFPB’s survey.
Cardholders in Our Sample Generally Paid
Down Balances during the Pandemic
After March 2020, More Cardholders Paid Off Their
Balances, and Revolvers Carried Balances for Shorter
Periods
An increasing share of cardholders paid off their balances after the onset
of the pandemic in March 2020. Our analysis of a sample of more than
650,000 individual credit card accounts from Federal Reserve data
showed that the share of active credit card accounts that were revolving
31Consumer Financial Protection Bureau, Consumer Use of Payday, Auto Title, and Pawn
Loans: Insights from the Making Ends Meet Survey, Research Brief No. 2021-1 (May
2021). The report was based on information from the first two waves of CFPB’s Making
Ends Meet survey, conducted in June 2019 and June 2020.
32Board of Governors of the Federal Reserve System, Report to the Congress on the
Profitability of Credit Card Operations of Depository Institutions (Washington, D.C.:
November 2020).
33Payday loans are a form of small-dollar, short-term consumer credit. Payday loans can
be rolled over to effectively last for longer than 2 weeks. See Consumer Financial
Protection Bureau, Market Snapshot: Consumer Use of State Payday Loan Extended
Payment Plans (April 2022).
Page 21 GAO-23-105269 Credit Cards
Letter
declined from 50 percent in April 2020 to 45 percent in December 2021
(see fig. 7).34 During the same period, the share of transacting accounts
increased from 50 percent to 54 percent. By comparison, our analysis of
data from June 2013 through March 2020 found that the shares of
revolving and transacting accounts stayed relatively stable year over
year, as previously discussed.
Figure 7: Percentage of Active Credit Card Accounts That Carried a Balance, 2019–2021
Data for Figure 7: Percentage of Active Credit Card Accounts That Carried a Balance, 2019–2021
Year Credit card accounts that carried a balance revolver Credit card accounts that did not carry a balance transactor
2019 51.2701 48.1578
51.284 48.1683
50.8711 48.6306
49.8781 49.6605
49.6276 49.9372
49.8768 49.7257
50.2576 49.3261
34During this period, less than 1 percent of active accounts were seriously delinquent.
Page 22 GAO-23-105269 Credit Cards
Letter
Year Credit card accounts that carried a balance revolver Credit card accounts that did not carry a balance transactor
50.2413 49.3072
50.3532 49.1493
50.4044 49.0972
50.3681 49.1299
50.2541 49.2159
2020 50.49 48.9436
50.5835 48.8678
50.3351 49.1543
49.863 49.6763
48.9334 50.6633
47.5818 52.0672
46.9627 52.7138
46.8637 52.8551
46.961 52.7789
47.1688 52.5749
47.098 52.6258
47.0823 52.6139
2021 47.2199 52.4841
46.8707 52.8266
46.0172 53.7033
44.6227 55.1356
44.0113 55.7495
44.2608 55.5222
44.6529 55.1425
44.9657 54.8094
44.994 54.758
45.3127 54.3929
45.1768 54.5142
45.2255 54.4179
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of active general purpose credit card
accounts. This figure excludes the percentage of active credit card accounts that were seriously
delinquent (90 or more days), which was less than 1 percent during this period.
To examine how the pandemic affected credit card usage, we analyzed
data from June 2013 through December 2021. We used multivariate
regression models to compare revolving episodes during the period April
2020–December 2021 (the first part of the pandemic) with those during
the period June 2013–March 2020. We made this comparison because
various conditions (related to public health, the economy, and public
Page 23 GAO-23-105269 Credit Cards
Letter
policy) prevailed during the pandemic that might have affected credit card
usage.35
We focused on two key characteristics of a cohort of new revolving
episodes: (1) durations of revolving episodes and (2) account statuses
after the end of revolving episodes. We simulated the distributions of
revolving episode durations under pandemic and prepandemic conditions.
These distributions provide insight into the proportion of revolving
episodes that lasted for specific numbers of months. We also simulated
the distributions of account statuses after the revolving episodes ended
under pandemic and prepandemic conditions. These distributions provide
insight into the proportion of revolving episodes that transitioned to
transacting, seriously delinquent and charged-off, inactive, or closed
status. Our analysis does not explain the reasons for differences in credit
card usage under prepandemic and pandemic conditions, and thus does
not establish a causal relationship between such conditions and credit
card usage. See appendix IV for additional information on the
methodology and results of our analysis.
Our analysis suggests that the conditions that prevailed during the
pandemic were associated with the following changes in revolvers’ credit
card usage:
· Shorter revolving episodes. Our analysis indicates that pandemic
conditions were associated with shorter revolving episodes. For
example, it indicates that under pandemic conditions, about 90
percent of revolving episodes would end sometime between 1 and 28
months, with 50 percent of the revolving episodes ending within 3
months (see fig. 8).36 In comparison, under prepandemic conditions,
35Some of the conditions of the pandemic that could have affected credit card usage
included public health conditions (e.g., social distancing), economic conditions (e.g.,
employment levels), and policy responses (e.g., pandemic-related government
assistance).
36An account can have multiple revolving episodes of varying durations. We report on the
90th percentile of the distribution of durations for a cohort of new revolving episodes
because it generally illustrates the experience of a typical revolving account in a given
month. For example, over the period of July 2016–July 2017—a year selected in the
middle of the time period covered by our data—the typical (median) revolving account in
our data was in a revolving episode of somewhere between 39 and 50 months. However,
in this time period, more than 90 percent of new revolving episodes in our data ended
within this time. The duration of a typical (median) new revolving episode was generally
between 4 and 5 months. For more information on our analysis methodology and results,
see app. IV.
Page 24 GAO-23-105269 Credit Cards
Letter
about 90 percent of revolving episodes would end sometime between
1 and 42 months—1.5 times longer than it would take under pandemic
conditions—with 50 percent of the revolving episodes ending within 4
months.
· More transitions to transacting status. For the 90 percent of
revolving episodes that would end sometime between 1 and 28
months under pandemic conditions, our analysis indicates that about
76 percent of these revolving episodes would transition to transacting
status (see fig. 8).37 In comparison, for the 90 percent of revolving
episodes that would end sometime between 1 and 42 months under
prepandemic conditions, a smaller share of revolving episodes (about
69 percent) would transition to transacting status.
· Fewer transitions to seriously delinquent and charged-off,
inactive, or closed status. Our analysis also indicates that under
pandemic conditions, revolving episodes would be less likely to
become seriously delinquent and charged off or to transition to
inactive or closed status, as shown in figure 8.
37We do not project the distribution of the statuses that follow all revolving episodes, given
that the time period for our data is not long enough to estimate with confidence the
distribution of transition destinations of the longest revolvers.
Page 25 GAO-23-105269 Credit Cards
Letter
Figure 8: Simulated Length of Credit Card Revolving Episodes and Status after End of Revolving Episodes under
Prepandemic and Pandemic Conditions
Data for Figure 8: Simulated Length of Credit Card Revolving Episodes and Status after End of Revolving Episodes under
Prepandemic and Pandemic Conditions
Simulated duration of revolving episodes
Month by which 50 percent Month by which 75 percent Month by which 90 percent
of revolving episodes would of revolving episodes of revolving episodes would
end would end end
Before the COVID-19 pandemic 4 12 42
(June 2013-March 2020)
During the COVID-19 pandemic 3 9 28
(April 2020-December 2021)
Simulated status to which 90 percent of revolving episodes would transition
Transactor Inactive Closed Seriously delinquent and charged-off
Before the COVID-19 pandemic 69 20 2 8
(June 2013-March 2020
Page 26 GAO-23-105269 Credit Cards
Letter
Simulated status to which 90 percent of revolving episodes would transition
Transactor Inactive Closed Seriously delinquent and charged-off
During the COVID-19 pandemic 76 19 1 5
(April 2020-December 2021
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of general purpose credit card
accounts. We used statistical models and simulations to compare revolving episodes under pandemic
and prepandemic conditions. However, our analysis does not explain the reasons for differences in
credit card usage under these conditions, and thus does not establish a causal relationship between
such conditions and credit card usage. See app. IV for more information on our methodology.
Pandemic Assistance Was Associated with Lower
Balances, Improved Credit Scores, and Fewer
Delinquencies
Reduced Balances
Pandemic-related assistance provided by the federal government and
credit card issuers likely contributed to cardholders paying down their
credit card balances, according to our analysis. Specifically, we estimated
that median nominal revolving balances declined from approximately
$1,737 in April 2020 to $1,529 in December 2021, or about 12 percent
(see fig. 9).38 By comparison, we estimated that median nominal revolving
balances generally increased from June 2013 through March 2020, as
previously discussed.
38In December 2021 dollars, the median revolving balance was $1,889 in April 2020,
which was 19 percent higher than that in December 2021. Changes in nominal balances
reflect changes in the price level and changes in the real value of the balances, as
previously discussed. A study found that credit card spending decreased from January to
April 2020, which contributed to the initial decline in credit card balances in April 2020.
The study also found that credit card revolving balances continued to decrease from April
2020 to early 2021, even though credit card spending slowly started to recover after April
2020. During this period, cardholders increased their credit card payments, which drove
the decline in revolving balances. See Robert M. Adams, Vitaly M. Bord, and Bradley
Katcher, “Why Did Credit Card Balances Decline So Much during the COVID-19
Pandemic” (Board of Governors of the Federal Reserve System, Dec. 3, 2021), accessed
Jan. 24, 2022, https://www.federalreserve.gov/econres/notes/feds-notes/why-did-credit-
card-balances-decline-so-much-during-the-covid-19-pandemic-20211203.html.
Page 27 GAO-23-105269 Credit Cards
Letter
Figure 9: Median Nominal Credit Card Revolving Balances, 2020–2021
Data for Figure 9: Median Nominal Credit Card Revolving Balances, 2020–2021
Year Amount in dollars
2020 1715.75
1718.18
1707.19
1736.93
1740.2
1744.81
1734.72
1715.61
1687.38
1647.32
1630.51
1582
2021 1538.99
1565.39
Page 28 GAO-23-105269 Credit Cards
Letter
Year Amount in dollars
1503.20
1505.01
1524.40
1507.62
1475.34
1466.01
1466.79
1476
1487.58
1528.75
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of active general purpose credit card
accounts.
Some cardholders who received pandemic-related assistance reported
using it to pay down their credit card balances, according to the Census
Bureau’s Household Pulse Survey. Of the respondents who reported
receiving or expecting to receive the first economic impact payments, an
estimated 14–16 percent used or planned to use the funds to pay down
debts, including credit card debt.39 The estimated percentage increased to
50–54 percent in the surveys conducted after the distributions of the
second and third economic impact payments.40
The survey results are in line with our analysis of Federal Reserve data,
which estimated that cardholders increased their next credit card
payments by an average of $20 and $61 when the second and third
39Census conducted the Household Pulse Survey each week from April 23 to July 21,
2020, and every 2 weeks starting in August 2020. We reviewed the surveys conducted
from April 23, 2020, to October 11, 2021. In the surveys conducted from June 11 to July
21, 2020, from January 6 to March 29, 2021, and from April 14 to July 5, 2021,
respondents were asked if they received or expected to receive an economic impact
payment and on what they spent the payment, including paying down credit card
balances, student loans, or other debts. The 95 percent confidence intervals for the
estimates ranged from about 13 to 17 percent.
40The 95 percent confidence intervals for the estimates ranged from 49 to 54 percent
based on surveys conducted after the distribution of the second economic impact
payments, and from 48 to 59 percent based on surveys conducted after the third
economic impact payments.
Page 29 GAO-23-105269 Credit Cards
Letter
economic impact payments were disbursed, respectively.41 The average
payment amount also increased by $37 in each month when the advance
child tax credit payments were disbursed.42 These increases are in
addition to an overall average increased payment of $40 per month
toward credit card balances for revolving accounts from March 13, 2020,
through the end of December 2021. Given that the median monthly
payment by revolvers was $131 from June 2013 through December 2021,
the increases in the average credit card payment amounts during the
pandemic were substantial in general, and especially in the months
following each of the three federal economic impact payments and the
advance child tax credit payments. See appendix V for additional
information on our analysis.
While not all cardholders were eligible to receive federal pandemic cash
assistance, other factors may have provided additional cash flows to help
cardholders pay down their balances. For example, during the pandemic,
some cardholders may have benefited from appreciation of financial
assets or a refinancing of their mortgage. Other cardholders may have
been eligible for a pause on their student loan payments. As a result,
these cardholders may have had more cash to pay down balances,
according to the Federal Reserve’s research, CFPB staff, and some
industry participants with whom we spoke.43
41Our analysis also estimated that cardholders reduced their credit card payments by an
average of $49 in March 2020 when the COVID-19 pandemic was declared. When the
first economic impact payments were disbursed in April 2020, cardholders on average
increased their payments, largely reversing the March 2020 decrease. Another study
found that the distribution of the first economic impact payments was associated with a
substantial increase in payments on rents, mortgages, and credit cards. See Scott R.
Baker et al., “Income, Liquidity, and the Consumption Response to the 2020 Economic
Stimulus Payments” (NBER Working Paper No. 27097, Sept. 15, 2020)
https://www.nber.org/papers/w27097.pdf. A study also reported that cardholders eligible
for the economic impact payments increased their credit card payments substantially at
the start of 2021, which was likely supported by the second and third economic impact
payments. See Adams, Bord, and Katcher, “Why Did Credit Card Balances Decline.”
42Not all households were eligible to receive the economic impact payments or the
advance child tax credit since these types of assistance were generally based on an
individual’s or household’s income or family status. We did not have data that would have
allowed us to determine which cardholders were eligible for these pandemic assistance
payments. As a result, our estimates on the effect of pandemic assistance funds used
payment data for cardholders who both did and did not receive such funds.
43Adams, Bord, and Katcher, “Why Did Credit Card Balances Decline.” The industry
participants were two credit card issuers and two credit reporting agencies.
Page 30 GAO-23-105269 Credit Cards
Letter
Improved Credit Scores
Many cardholders’ credit scores improved during the pandemic. Our
analysis of Federal Reserve data found that revolvers on average had an
increase of 20 points in their credit scores from March 2020 through
December 2021. In comparison, revolvers on average had an increase of
7 points over a similar period from March 2018 through December 2019.
Additionally, 71 percent of revolvers saw an increase in their credit scores
from March 2020 through December 2021, compared with 65 percent of
revolvers from March 2018 through December 2019.
Consistent with our analysis, three credit card issuers and two credit
reporting agencies told us that credit scores increased for cardholders
across credit score categories during the pandemic. They attributed this
trend to cardholders paying down their credit card balances. Two of these
companies added that cardholders with subprime credit scores saw larger
increases in their credit scores compared with cardholders in other credit
score categories.44
Fewer Delinquencies
Federal pandemic assistance and credit card issuers’ voluntary payment
assistance programs likely provided support to some cardholders and
decreased delinquencies, according to CFPB and several industry
participants.45 As previously discussed, issuers’ payment assistance
programs generally included payment deferrals and fee waivers. Our
analysis of Federal Reserve data found that the percentage of seriously
44A study also found that credit scores for all borrowers generally increased after March
2020, especially for borrowers with lower credit scores. Also, credit card utilization of
borrowers with lower credit scores decreased early in the pandemic, which likely
contributed to increased credit scores. The decrease in credit card utilization could reflect
consumers reducing spending and using pandemic assistance to pay down debt. See
Sarena Goodman et al., “Developments in the Credit Score Distribution over 2020” (Board
of Governors of the Federal System, Apr. 30, 2021), accessed May 24, 2023,
https://www.federalreserve.gov/econres/notes/feds-notes/developments-in-the-credit-scor
e-distribution-over-2020-20210430.html.
45The industry participants were six credit card issuers, the American Bankers
Association, and two credit reporting agencies. All issuers surveyed by CFPB reported
offering pandemic-related relief programs. See Consumer Financial Protection Bureau,
The Consumer Credit Card Market.
Page 31 GAO-23-105269 Credit Cards
Letter
delinquent credit card accounts generally declined from April 2020 to
December 2021.46
A relatively small share of cardholders enrolled in the voluntary payment
assistance programs, according to four large issuers.47 For instance, two
of the issuers told us about 3–4 percent of their credit card customers
enrolled in a payment assistance program. CFPB found that cardholders
with lower incomes or credit scores or who were non-White were more
likely to enroll in payment assistance programs. Three credit card issuers
and a credit reporting agency also noted that many cardholders who were
enrolled in payment assistance programs continued to make payments or
only used the programs for a short period.
In an October 2020 comment letter to CFPB, the National Consumer Law
Center stated that unlike the mandatory mortgage assistance programs,
the voluntary credit card payment assistance programs may have
reached only consumers financially sophisticated enough to request
them.48 The center also noted that enrollment was subject to the
discretion of the issuer. It said that, as a result, these voluntary assistance
programs may not have reached some consumers in need of assistance.
Cardholders in Our Sample in Areas with
MajorityBlack or
46Our analysis also found that the share of seriously delinquent accounts started to
increase in the second half of 2021, although it largely remained below prepandemic
levels. Two credit card issuers and three credit reporting agencies reported a similar
increase in delinquencies after the end of the pandemic assistance in 2021. Additionally,
the Federal Reserve Bank of New York reported that the percentage of credit cardholders
who became 90 or more days delinquent during the fourth quarter of 2022 surpassed that
percentage from before the pandemic among younger cardholders. The percentage was
rising for older cardholders but had not yet reached prepandemic levels. The bank stated
that rising interest rates, inflation, and the end of various forms of pandemic assistance
may have been possible contributing factors to rising delinquency rates.
47The six credit card issuers we spoke with had terminated their pandemic-related
payment assistance programs or had only a small percentage of credit card accounts
remaining in these programs by the end of 2020, according to their 2020 annual filings.
See Consumer Financial Protection Bureau, The Consumer Credit Card Market.
48National Consumer Law Center, “Re: CARD Act Rules Review Pursuant to the
Regulatory Flexibility Act; Request for Information Regarding Consumer Credit Card
Market, Docket No. CFPB–2020–0027” (Oct. 27, 2020).
Page 32 GAO-23-105269 Credit Cards
Letter
Hispanic Residents Likely Had Less Favorable
Credit Terms and Carried Balances Longer
Credit Terms in MajorityBlack or Hispanic Zip Codes
Were Generally Worse Than Those in Predominantly
White Zip Codes
Our analysis of a sample of more than 650,000 individual credit card
accounts from Federal Reserve data suggests that, in comparison with
cardholder accounts in billing zip codes with predominantly White
residents, accounts in billing zip codes with more Black or Hispanic
residents on average had higher interest rates and lower credit limits.49
We found these differences both before and after controlling for
differences in other cardholder characteristics that can influence credit
card terms (credit scores, incomes in their zip codes, and revolving
status).50 Our analysis could not determine whether the racial and ethnic
disparities in credit terms resulted from fair lending disparities, which
cannot be measured as independent factors in our analysis.51
49In this report, we use the term “Black” to refer to “Black or African American” and
“Hispanic” to refer to “Hispanic or Latino.” Our analysis did not identify a cluster of zip
codes with a majority of Asian residents, and therefore we do not present results on
majority-Asian zip codes. We do not report on the “Other” race and ethnicity category
because of the small number of observations of credit card accounts in the Federal
Reserve data that were located in zip codes with residents in this category. As previously
discussed, the Federal Reserve collects data on credit card accounts from credit card
issuers. However, credit card issuers are not permitted to collect information on the race
or ethnicity of the cardholders. See apps. III and VI for additional information on our
econometric analysis of the association between credit terms and race and ethnicity in
cardholders’ zip codes.
50Because the Federal Reserve data did not include updated incomes of cardholders, we
controlled for the income distribution in cardholders’ zip codes, based on the household
incomes from Census’s American Community Surveys for 2013–2020. We measured the
distribution of income in cardholders’ zip codes as the percentage of households with
income in each of 16 income groups. See app. VI for additional information on our
methodology.
51Credit card issuers are prohibited from making lending decisions based on borrowers’
race or ethnicity. 12 U.S.C. § 1691(a).
Page 33 GAO-23-105269 Credit Cards
Letter
Interest Rates
Using data from June 2013 through December 2021, our analysis
suggests that accounts in billing zip codes with a greater share of Black
or Hispanic residents had higher interest rates on average than accounts
in billing zip codes with predominantly White residents (see fig. 10).52
Specifically, we estimated that interest rates were about 1.3 percentage
points higher on average for accounts in billing zip codes with a majority
of Black residents, and 1.4 percentage points higher for zip codes with a
majority of Hispanic residents. Among credit card accounts with the same
credit score, zip-code income distribution, and revolving status, these
differences persisted, although they were smaller (see fig. 10). For
instance, interest rates were 0.5 percentage points higher on average for
accounts in billing zip codes with a majority of Black residents, and 0.6
percentage points higher for accounts in billing zip codes with a majority
of Hispanic residents, relative to accounts in predominantly White zip
codes. See appendixes III and VI for additional information on our
methodology and the results of our analysis.
52For context, the median interest rate for accounts in predominantly White billing zip
codes was 17.5 percent in December 2019.
Page 34 GAO-23-105269 Credit Cards
Letter
Figure 10: Estimated Average Interest Rate Differences for Cardholder Accounts by
Race and Ethnicity of Billing Zip Codes, June 2013–December 2021
Data for Figure 10: Estimated Average Interest Rate Differences for Cardholder
Accounts by Race and Ethnicity of Billing Zip Codes, June 2013–December 2021
Before controlling for credit After controlling for credit
scores, zip-code incomes, scores, zip-code incomes,
and revolving status and revolving status
Majority Hispanic or Latino 1.4 0.6
Majority Black or African 1.3 0.5
American
Mixed race and ethnicity 0.4 0.2
with majority White
Mixed race and ethnicity 0.2 0.4
with no majority
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and the Census Bureau’s American
Community Survey. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of general purpose credit card
accounts. Because data from the Board of Governors of the Federal Reserve System do not provide
the race or ethnicity of cardholders, we supplemented with race and ethnicity data of the accounts’
billing zip codes, as reported by the Census Bureau’s 5-year American Community Surveys for 2013–
2020. We used econometric analysis to estimate the associations between credit card interest rates
and the zip-code clusters of varying racial and ethnic composition; zip-code clusters of predominantly
White residents served as the reference group. We do not account for all possible factors related to
differences in interest rates for cardholders in different racial and ethnic groups, which may result
from various unobservable factors. Also, the existence of differences in interest rates does not
establish whether fair lending disparities have occurred. We rounded the results of the interest rate
Page 35 GAO-23-105269 Credit Cards
Letter
differences to the nearest hundredth. See apps. III and VI for additional information on our
methodology and the results of our analysis.
Credit Limits
Using data from June 2013 through December 2021, our analysis
suggests that accounts in billing zip codes with a greater share of Black
or Hispanic residents had lower credit limits on average than accounts in
billing zip codes with predominantly White residents (see fig. 11).53
Specifically, credit limits for accounts in billing zip codes with a majority of
Black or Hispanic residents were on average about $3,412 and $4,285
lower, respectively. Further, as shown in figure 11, among credit card
accounts with the same credit score, zip-code income distribution, and
revolving status, these differences persisted, although they were smaller.
For instance, credit limits were on average $710 lower for accounts in
billing zip codes with a majority of Black residents, and $1,477 lower for
accounts in billing zip codes with a majority of Hispanic residents. See
appendixes III and VI for additional information on our methodology and
the results of our analysis.
53The median credit limit for accounts in predominantly White billing zip codes was $8,100
in December 2019.
Page 36 GAO-23-105269 Credit Cards
Letter
Figure 11: Estimated Average Credit Limit Differences for Cardholder Accounts by
Race and Ethnicity in Billing Zip Codes, June 2013–December 2021
Data for Figure 11: Estimated Average Credit Limit Differences for Cardholder
Accounts by Race and Ethnicity in Billing Zip Codes, June 2013–December 2021
Before controlling for credit After controlling for credit
scores, zip-code incomes, scores, zip-code incomes,
and revolving status and revolving status
Majority Hispanic or -4285 -1477
Latino
Majority Black or African -3412 -710
American
Mixed race and ethnicity -1330 -540
with majority White
Mixed race and ethnicity -1079 -1489
with no majority
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and the Census Bureau’s American
Community Survey. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of general purpose credit card
accounts. Because data from the Board of Governors of the Federal Reserve System do not provide
the race or ethnicity of cardholders, we supplemented with race and ethnicity data of the accounts’
billing zip codes, as reported by the Census Bureau’s 5-year American Community Surveys for 2013–
2020. We used econometric analysis to estimate the associations between credit card credit limits
and the zip-code clusters of varying racial and ethnic composition; zip-code clusters of predominantly
White residents served as the reference group. We do not account for all possible factors related to
differences in credit limits for cardholders in different racial and ethnic groups, which may result from
various unobservable factors. Also, the existence of differences in credit limits does not establish
whether fair lending disparities have occurred. We rounded the results for the credit limit differences
Page 37 GAO-23-105269 Credit Cards
Letter
to the nearest whole dollar amount. See apps. III and VI for additional information on our
methodology and the results of our analysis.
Our results are consistent with other studies that found meaningful racial
and ethnic differences in credit costs and the amount of credit approved
or the likelihood of receiving approval for other types of credit products,
including auto, mortgage, and small business loans.54 However, our
analysis does not on its own offer an explanation of what might be driving
our results. Our analysis was not designed to determine all the potential
reasons for racial and ethnic disparities in the interest rates cardholders
were charged or credit limits, which may be due to factors that either are
not captured in the data we analyzed or cannot be measured. Such
factors could include the extent to which cardholders compare credit
terms when choosing a card, competition among lenders, and lender-
54For information on auto loans, see Alexander W. Butler, Erik J. Mayer, and James P.
Weston, “Racial Disparities in the Auto Loan Market,” Review of Financial Studies, vol. 36,
no. 1 (2023): 1–41. For information on mortgage loans, see Robert Bartlett et al.,
“Consumer-Lending Discrimination in the FinTech Era,” Journal of Financial Economics,
vol. 143, no. 1 (2022): 30–56; Ping Cheng, Zhenguo Lin, and Yingchun Liu, “Racial
Discrepancy in Mortgage Interest Rates,” Journal of Real Estate Finance and Economics,
vol. 51, no. 1 (2015): 101–120; and Neil Bhutta, Aurel Hizmo, and Daniel Ringo, “How
Much Does Racial Bias Affect Mortgage Lending? Evidence from Human and Algorithmic
Credit Decisions” (working paper, Finance and Economics Discussion Series, Board of
Governors of the Federal Reserve System, Aug. 2, 2022),
https://www.federalreserve.gov/econres/feds/how-much-does-racial-bias-affect-mortgage-l
ending.htm. However, another study suggests that associations between borrowers’ race
and ethnicity and their mortgage rates are offset by associations between borrowers’ race
and ethnicity and their discount points, weakening associations between borrowers’ race
and ethnicity and the total price for their mortgage. See Neil Bhutta and Aurel Hizmo, “Do
Minorities Pay More for Mortgages?” Review of Financial Studies, vol. 34, no. 2 (2021):
763–789. For information on small business loans, see Mels de Zeeuw and Brett Barkley,
“Mind the Gap: Minority-Owned Small Businesses’ Financing Experiences in 2018,”
Consumer & Community Context, vol. 1, no. 2 (2019),
https://www.federalreserve.gov/publications/files/consumer-community-context-201911.pd
f.
Page 38 GAO-23-105269 Credit Cards
Letter
specific risk evaluation or loan approval processes.55 Differences in these
factors may be incidentally associated with race and ethnicity, or these
factors can themselves be influenced by race and ethnicity, and therefore
act as channels through which race and ethnicity influence interest rates
or credit limits.
Cardholders in MajorityBlack or Hispanic Zip Codes
Likely Carried Balances for Longer Than Those in
Predominantly White Zip Codes
We examined how the duration of revolving episodes (periods when an
account continuously carries a balance) varied among cardholders on the
basis of their billing zip code’s racial and ethnic mix. To do so, we used
multivariate regression models and simulations to compare the
distributions of revolving episode durations for cardholder accounts in five
clusters of billing zip codes with different racial and ethnic composition.
We made this comparison separately for revolving episodes during the
period before the pandemic (June 2013–March 2020) and during the
period of the pandemic (April 2020–December 2021). We compared
revolving episodes by racial composition of zip codes separately for the
prepandemic and pandemic periods because various conditions (e.g.,
those related to public health, the economy, and public policy) prevailed
during the pandemic that might have affected credit card usage and may
have done so differently for different groups.
55For information on borrowers comparing credit terms, see Victor Stango and Jonathan
Zinman, “Borrowing High versus Borrowing Higher: Price Dispersion and Shopping
Behavior in the U.S. Credit Card Market,” The Review of Financial Studies, vol. 29, no. 4
(2016): 979–1006, and Cheng, Lin, and Liu, “Racial Discrepancy in Mortgage Interest
Rates.” For information on lender competition, see Butler, Mayer, and Weston, “Racial
Disparities in the Auto Loan Market,” and Bhutta, Hizmo, and Ringo, “How Much Does
Racial Bias Affect Mortgage Lending?” The latter study also discusses how differences in
mortgage approval may be associated with lender-specific loan approval processes,
including stricter credit standards for all borrowers. Another study also stated that issuers’
internal risk-based pricing models may place different emphasis on risk factors such as
credit scores and late payments, and hence cardholders may receive different interest
rates from different issuers. See Stango and Zinman, “Borrowing High versus Borrowing
Higher.” As previously discussed, credit card issuers are prohibited from making lending
decisions, including decisions about credit terms offered, based on borrowers’ race and
ethnicity. 12 U.S.C. § 1691(a). Our analysis could not determine whether the racial and
ethnic disparities in credit terms faced by cardholders resulted from fair lending disparities,
which cannot be measured as independent factors in our analysis.
Page 39 GAO-23-105269 Credit Cards
Letter
Our analysis suggests that under prepandemic conditions, revolving
episodes of accounts in majority-Black or -Hispanic billing zip codes were
associated with the following differences when compared with revolving
episodes of accounts in predominantly White billing zip codes.
· Longer revolving episodes. Our analysis indicates that accounts in
majority-Black billing zip codes likely had longer revolving episodes
than those in predominantly White billing zip codes. For example, it
indicates that about 90 percent of revolving episodes in predominantly
White zip codes likely ended sometime between 1 and 39 months,
with 50 percent of the revolving episodes ending within 3 months (see
fig. 12).56 In comparison, about 90 percent of revolving episodes in
majority-Black zip codes likely ended sometime between 1 and 56
months, with 50 percent of the revolving episodes ending within 5
months. Cardholders in majority-Hispanic zip codes similarly had
longer revolving episodes (see fig. 12).
· Fewer transitions to transacting status. For the 90 percent of
revolving episodes of accounts in predominantly White billing zip
codes that ended sometime between 1 and 39 months, our analysis
indicates that about 71 percent of them likely transitioned to
transacting status. In comparison, a smaller share of revolving
episodes of accounts in majority-Black or -Hispanic billing zip codes
likely transitioned to transacting status. For example, of the 90 percent
of revolving episodes in majority-Black zip codes that ended within 56
months, 61 percent likely transitioned to transacting status. Results for
majority-Hispanic zip codes were similar (see fig. 12).
· More transitions to seriously delinquent and charged-off or
inactive status. Our analysis also indicates that revolving episodes of
accounts in majority-Black or -Hispanic billing zip codes were more
likely to transition to seriously delinquent and charged-off or inactive
status than those in predominantly White billing zip codes, as shown
in figure 12.
56As previously noted, we report on the 90th percentile of the results because this
percentile is more representative of accounts that were in a revolving episode in any given
month than is the median. See app. IV for additional information on our analysis.
Page 40 GAO-23-105269 Credit Cards
Letter
Figure 12: Simulated Length of Credit Card Revolving Episodes and Status after End of Revolving Episodes, by Race and
Ethnicity in Billing Zip Codes, June 2013–December 2019
Page 41 GAO-23-105269 Credit Cards
Letter
Data for Figure 12: Simulated Length of Credit Card Revolving Episodes and Status after End of Revolving Episodes, by Race
and Ethnicity in Billing Zip Codes, June 2013–December 2019
Simulated duration of revolving episodes for cardholders in various zipcode clusters
Month by which 50 percent Month by which 75 percent Month by which 90 percent
of revolving episodes would of revolving episodes of revolving episodes
end would end would end
Predominately White 3 11 39
Majority Black or African American 5 18 56
Majority Hispanic or Latino 5 16 48
Mixed race and ethnicity with majority 4 12 42
White zip codes
Mixed race and ethnicity with no 3 8 31
majority zip codes
Simulated status to which 90 percent of revolving episodes would transition
Transactor Inactive Closed Seriously delinquent and charged-off
Predominately White 71 20 2 7
Maj Black 61 22 2 15
Maj Hispanic 63 24 2 11
Mixed White and substantial shares 69 21 2 8
Mixed race 76 17 2 5
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and the Census Bureau. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of general purpose credit card
accounts. Because data from the Board of Governors of the Federal Reserve System do not provide
the race or ethnicity of cardholders, we supplemented with race and ethnicity of the accounts’ billing
zip codes, as reported by the Census Bureau’s 5-year American Community Surveys for 2013–2020.
We used statistical models and simulations to compare revolving episodes for each of the five zip-
code clusters of varying racial and ethnic composition; zip-code clusters of predominantly White
residents served as the reference group. However, our analysis does not explain the reasons for
differences in revolving episode durations among different racial and ethnic clusters of zip codes, and
thus does not establish a causal relationship between revolving episode duration and the racial and
ethnic composition in cardholders’ zip codes. See apps. III and IV for additional information on our
analysis.
Furthermore, consistent with our findings on credit card usage during the
pandemic, discussed earlier, our analysis indicates that under pandemic
conditions, revolving episodes for all accounts would likely shorten. In
particular, our analysis indicates that for accounts in majority-Black or -
Hispanic billing zip codes, about 90 percent of revolving episodes would
end within 39 months and 32 months, respectively. However, accounts in
billing zip codes with more Black or Hispanic residents would still have
longer revolving durations than accounts in predominantly White billing
zip codes under pandemic conditions. See appendixes III and IV for
additional information on our analysis.
Page 42 GAO-23-105269 Credit Cards
Letter
Our results are consistent with another study that found meaningful
differences in credit card revolving patterns on the basis of race and
ethnicity.57 However, similar to our findings on interest rates and credit
limits, our analysis does not on its own offer an explanation of what might
be driving our results. Our analysis was not designed to determine all the
potential reasons for racial and ethnic disparities in credit card revolving
patterns, which may be due to factors that either are not captured in the
data we analyzed or cannot be measured. Such factors could include
cardholders’ liquidity needs or their assessments of the characteristics of
credit cards compared with those of other payment methods.58
Differences in these factors may be incidentally associated with race and
ethnicity, or these factors can themselves be influenced by race and
ethnicity, and therefore act as channels through which race and ethnicity
influence revolving patterns.
Our analysis also suggests that, despite their likelihood of carrying
balances for longer, accounts in majority-Black or -Hispanic billing zip
codes carried lower revolving balances than accounts in billing zip codes
with predominantly White residents (see fig. 13).59 Specifically, revolving
balances carried by accounts in billing zip codes with a majority of Black
or Hispanic residents were $953 and $1,212 lower, respectively,
compared with accounts in billing zip codes with predominantly White
residents. Among credit card accounts with the same credit score and
zip-code income distribution, these differences persisted, although they
were smaller. For instance, revolving balances were $492 lower on
average for accounts in billing zip codes with a majority of Black
residents, and $711 lower for accounts in billing zip codes with a majority
of Hispanic residents. A lower revolving balance may reduce the absolute
credit costs for accounts in majority-Black or -Hispanic billing zip codes.
However, these accounts likely still paid more for each dollar of credit
57See Jae Min Lee and Yoon G. Lee, “Multidimensional Credit Attitude and Credit Card
Debt Behavior in the United States,” Review of Behavioral Finance, vol. 14, no. 2 (2022):
183–196.
58For example, see Irina A. Telyukova, “Household Need for Liquidity and the Credit Card
Debt Puzzle,” Review of Economic Studies, vol. 80 (2013): 1148–1177, and Claire Greene
and Joanna Stavins, “Credit Card Debt Puzzle: Liquid Assets to Pay Bills” (Federal
Reserve Bank of Boston Working Papers, no. 22-8, June 2022), accessed Feb. 23, 2023,
https://www.bostonfed.org/publications/research-department-working-paper/2022/credit-ca
rd-debt-puzzle-liquid-assets-to-pay-bills.
59For context, the nominal median revolving balance for all accounts was $1,720 in
December 2019. See app. VI for additional information on our analysis.
Page 43 GAO-23-105269 Credit Cards
Letter
card use because they generally faced higher interest rates and carried
their balances for longer.
Figure 13: Estimated Revolving Balance Differences for Cardholder Accounts by
Race and Ethnicity of Billing Zip Codes, June 2013–December 2021
Data for Figure 13: Estimated Revolving Balance Differences for Cardholder
Accounts by Race and Ethnicity of Billing Zip Codes, June 2013–December 2021
Before controlling for credit After controlling for
scores and zip-code credit scores and zip-
incomes code incomes
Majority Hispanic or Latino -1212 -711
Majority Black or African -953 -492
American
Mixed race and ethnicity with -374 -258
majority White
Mixed race and ethnicity with -308 -564
no majority
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and the Census Bureau’s American
Community Survey. | GAO-23-105269
Note: Our analysis was based on a nongeneralizable sample of general purpose credit card
accounts. Because data from the Board of Governors of the Federal Reserve System do not provide
the race or ethnicity of cardholders, we supplemented with race and ethnicity data of the accounts’
billing zip codes, as reported by the Census Bureau’s 5-year American Community Surveys for 2013–
2020. We used econometric analysis to estimate the associations between credit balances carried by
cardholders and the zip-code clusters of varying racial and ethnic composition; zip-code clusters of
predominantly White residents served as the reference group. We do not account for all possible
Page 44 GAO-23-105269 Credit Cards
Letter
factors related to differences in revolving balances for cardholders in different racial and ethnic
groups, which may result from various unobservable factors. Also, the existence of differences in
revolving balances does not establish whether fair lending disparities have occurred. We rounded the
results for the revolving balance differences to the nearest whole dollar amount. See apps. III and VI
for additional information on our methodology and the results of our analysis.
Agency Comments
We provided a draft of this report to CFPB and the Federal Reserve for
review and comment. The agencies provided technical comments, which
we incorporated, as appropriate.
As agreed with your offices, unless you publicly announce the contents of
this report earlier, we plan no further distribution until 30 days from the
report date. At that time, we will send copies of this report to the
appropriate congressional committees, the Director of CFPB, the Chair of
the Federal Reserve, and other interested parties. In addition, the report
will be available at no charge on the GAO website at
https://www.gao.gov.
If you or your staff have any questions about this report, please contact
me at (202) 512-8678 or cackleya@gao.gov. Contact points for our
Offices of Congressional Relations and Public Affairs may be found on
the last page of this report. Key contributors to this report are listed in
appendix VII.
Alicia Puente Cackley
Director, Financial Markets and Community Investment
Page 45 GAO-23-105269 Credit Cards
Appendix I: Objectives, Scope, and
Methodology
Appendix I: Objectives, Scope,
and Methodology
This report examines (1) credit card usage from 2013 through 2019 and
the characteristics of cardholders who carried balances during this period,
(2) how the COVID-19 pandemic and related assistance affected credit
card usage from March 2020 through December 2021, and (3) how credit
card costs and usage vary among racial and ethnic groups.
Federal Reserve and Census Bureau Data
To address all three objectives, we obtained a 0.1 percent
nongeneralizable sample of consumer credit card account data from the
Board of Governors of the Federal Reserve System’s Capital
Assessments and Stress Testing Report (FR Y-14M) from June 2013
through December 2021.1 After cleaning and preparing the data, we had
30.5 million account-month observations of monthly billing cycle data on
more than 650,000 individual credit card accounts. The Federal Reserve
credit card data are collected from bank holding companies, savings and
loan holding companies, and intermediate holding companies with $100
billion or more in assets. As a result, we cannot generalize the results of
our analysis to credit card accounts in all banks in the U.S. The Federal
Reserve data allowed us to examine credit card usage by account, but
the data did not have information that would allow us to link multiple
accounts held by a given individual cardholder. We also cannot link
accounts to a specific issuer.
Our analysis included only general purpose credit cards issued on
networks, such as Visa, Mastercard, American Express, and Discover,
which are accepted by a wide variety of merchants.2 We removed
account-month observations with an address outside of the 50 U.S. states
and the District of Columbia—that is, excluding addresses in the U.S.
1The Federal Reserve first started collecting the Y-14M data in June 2012. However, our
analysis used data beginning in June 2013 because the Federal Reserve did not collect
some data elements until that time.
2Another type of credit card, referred to as a private label card, can only be used at one
merchant or a small group of related merchants. American Express and Discover are both
payment networks and credit card issuers.
Page 46 GAO-23-105269 Credit Cards
Appendix I: Objectives, Scope, and
Methodology
territories and outside of the U.S. See appendix II for more information on
the Federal Reserve data.
We supplemented the Federal Reserve data with data from the Census
Bureau’s 5-Year American Community Surveys for 2013–2020. We used
census data on household income and the shares of residents in each zip
code who were Asian, Black or African American (referred to as “Black” in
this report), Hispanic or Latino (referred to as “Hispanic” in this report),
White, and Other. See appendixes II and III for more information on the
census data.
Analysis of Credit Card Account Status
We analyzed the Federal Reserve data to describe different aspects of
credit card usage from June 2013 through December 2021, including by
cardholder characteristics such as credit score, estimated median
household income, and racial and ethnic composition of cardholders’
billing zip codes. We generally classify credit card accounts into six
groups on the basis of how cardholders use the cards and how the
accounts accrue interest charges during each month: transacting,
revolving, seriously delinquent, inactive, closed, and charged off (see
app. II for additional information on the definitions of the six account
statuses).
We examined the share of active accounts by status, including the share
of revolving accounts, transacting accounts, and seriously delinquent
accounts, and how the share of those accounts changed over time from
June 2013 through December 2021. We also calculated the percentage
of active accounts that were revolving from June 2013 through December
2019 by credit score category and zip-code-level median household
income, averaged over this period.
Additionally, we analyzed the amount of median nominal revolving
balance carried by revolving accounts and the percentage of revolving
accounts that reached or exceeded their credit limits from June 2013
through December 2021. We analyzed nominal balance amounts
because they represent the account balance that cardholders are
obligated to repay. While this approach focuses on the cardholder
experience, changes over time in nominal balances reflect changes in the
price level as well as changes in the real value of the balances.
Page 47 GAO-23-105269 Credit Cards
Appendix I: Objectives, Scope, and
Methodology
We also used econometric analysis to estimate the revolving balances
carried by cardholders in zip codes with different racial and ethnic
composition during the same period.3 Furthermore, we compared the
percentages of transacting accounts and revolving accounts that made
purchases and the purchase amounts. All dollar amounts are in nominal
terms unless otherwise noted.
Analysis of Revolving and Transacting Episode
Durations
We calculated the duration of revolving episodes—that is, the number of
consecutive months revolving accounts carried a balance—using Federal
Reserve data from June 2013 through December 2021. We assigned the
accounts to five categories based on the total duration of their revolving
episodes: short-term revolving accounts, which revolved for 3 months or
less; medium-term revolving accounts, which revolved for 4–11 months;
long-term revolving accounts, which revolved for 12 months or more; and
accounts we observed to be revolving for short or medium terms but for
which we did not observe the full episode because the observation
occurred at the start or end of the account’s appearance in our data. We
report on the fraction of long-term episodes from June 2014 through
December 2020 because too many of the observations in the first and last
years of our data are not complete episodes.
For accounts that were revolving in a given month, we also identified the
status of each revolving account in the following month in order to
construct econometric models that estimated the probability that revolvers
would either pay off their bill or become seriously delinquent. This
analysis allowed us to estimate the distribution of the number of months
revolving episodes would have been expected to last under conditions
that prevailed during the COVID-19 pandemic (April 2020–December
2021) and before the pandemic (June 2013–March 2020).4 We conducted
similar analysis for transacting episodes.
Our econometric models also allowed us to examine the distribution of
revolving and transacting episode durations under prepandemic
3See app. VI for additional information on the analysis and results.
4Some of the conditions could include public health conditions, economic conditions such
as employment levels, and effects of policy responses, such as restrictions and
government assistance in response to the pandemic.
Page 48 GAO-23-105269 Credit Cards
Appendix I: Objectives, Scope, and
Methodology
conditions (June 2013–March 2020) and pandemic conditions (April
2020–December 2021) by the racial and ethnic composition of
cardholders’ billing zip codes. Our analysis does not account for all
possible factors or establish causal relationships related to differences in
the revolving or transacting episode durations and account status under
prepandemic and pandemic conditions. Our analysis also does not
account for all possible factors or establish causal relationships related to
differences in the revolving or transacting episode durations and account
status for cardholders in zip-code clusters with different racial or ethnic
composition. See appendixes III and IV for additional information on the
methodology and results of our analysis.
Analysis of Credit Cost Differences
To estimate differences in credit costs for revolvers who carry balances
for different amounts of time, we used the 2019 average credit card
interest rate as published by the Federal Reserve and calculated interest
charges for two hypothetical cardholders carrying a balance over different
numbers of months.5 We also reviewed Consumer Financial Protection
Bureau (CFPB) reports for information on what consumers would be likely
to do when they had difficulty paying for a bill or expense, and how that
might affect their credit costs.6
Additionally, we constructed econometric models to examine the
relationships between cardholders’ credit card terms (interest rates and
credit limits) and the racial and ethnic composition of their zip codes. Our
results should be interpreted with caution. Our analysis was not designed
to determine all the potential reasons for racial and ethnic disparities in
credit terms, which may be due to factors that are either not captured in
the data we analyzed or cannot be measured. Differences in these factors
may be incidentally associated with race and ethnicity, or these factors
can themselves be influenced by race and ethnicity, and therefore act as
channels through which race and ethnicity influence credit terms. See
5Board of Governors of the Federal Reserve System, Report to Congress: Profitability of
Credit Card Operations of Depository Institutions (Washington, D.C.: November 2020).
6Consumer Financial Protection Bureau, Consumer Use of Payday, Auto Title, and Pawn
Loans: Insights from the Making Ends Meet Survey, Research Brief No. 2021-1 (May
2021) and Insights from the Making Ends Meet Survey, Research Brief No. 2020-1 (July
2020). The first CFPB report was based on information from the first two waves of the
agency’s Making Ends Meet survey, conducted in June 2019 and June 2020.
Page 49 GAO-23-105269 Credit Cards
Appendix I: Objectives, Scope, and
Methodology
appendixes III and VI for additional information on the methodology and
results of our analysis.
Analysis of Credit Card Usage in Association
with PandemicRelated Responses
To identify federal and private-sector COVID-19 pandemic assistance and
understand how the assistance may have contributed to changes in credit
card usage during the pandemic, we reviewed reports from GAO, the
Federal Reserve, and CFPB.7 We also reviewed the 2020 annual filings
from six large credit issuers: American Express, Bank of America, Capital
One, Citibank, Discover, and JPMorgan Chase. We selected these
issuers because they had the largest purchase volume and outstanding
balances in their credit card portfolios in 2020.
Additionally, using Federal Reserve data, we compared revolving
accounts’ credit scores on average during the pandemic (March 2020–
December 2021) and before the pandemic (March 2018–December
2019). Specifically, we estimated the average credit score point increase
and the share of revolvers that saw an increase in their credit score
during each of these two periods.
Further, we conducted a regression analysis to estimate the average
amount by which cardholders increased their payments when the three
waves of economic impact payments and advance child tax credit
7See for example, GAO, Stimulus Checks: Direct Payments to Individuals during the
COVID-19 Pandemic, GAO-22-106044 (Washington, D.C.: June 29, 2022) and COVID-
19: Opportunities to Improve Federal Response and Recovery Efforts, GAO-20-625
(Washington, D.C.: June 25, 2020); Robert M. Adams, Vitaly M. Bord, and Bradley
Katcher, “Why Did Credit Card Balances Decline So Much during the COVID-19
Pandemic” (Board of Governors of the Federal Reserve System, Dec. 3, 2021), accessed
Jan. 24, 2022, https://www.federalreserve.gov/econres/notes/feds-notes/why-did-credit-
card-balances-decline-so-much-during-the-covid-19-pandemic-20211203.html; Robert
Adams and Vitaly Bord, “The Effects of the COVID-19 Shutdown on the Consumer Credit
Card Market: Revolvers versus Transactors” (Board of Governors of the Federal Reserve
System, Oct. 21, 2020), accessed Jan. 24, 2022,
https://www.federalreserve.gov/econres/notes/feds-notes/the-effects-of-the-covid-19-
shutdown-on-the-consumer-credit-card-market-revolvers-versus-transactors-
20201021.html; and Consumer Financial Protection Bureau, The Consumer Credit Card
Market (September 2021) and Consumer Finances During the Pandemic: Insights from
the Making Ends Meet Survey (December 2021).
Page 50 GAO-23-105269 Credit Cards
Appendix I: Objectives, Scope, and
Methodology
payments were disbursed during the pandemic.8 In addition, we examined
data from the Census Bureau’s Household Pulse Survey conducted from
April 23, 2020, to October 11, 2021, related to how survey respondents
used or planned to use the federal economic impact payments, including
to pay down credit card and other debts.9
Data Reliability Assessments
To assess the reliability of the Federal Reserve’s credit card data, we
conducted electronic testing for missing data and obvious errors,
reviewed the Federal Reserve’s technical documentation, and interviewed
staff from the Federal Reserve and CFPB who were knowledgeable about
the data. We determined that the Federal Reserve data were sufficiently
reliable for the purpose of examining the distribution of credit card
accounts by account status, cardholder characteristics, and revolving
patterns for June 2013–December 2021.
Additionally, we reviewed technical documentation related to Census’s 5-
year American Community Surveys for 2013–2020, and determined that
the data were sufficiently reliable for describing the household income
and racial and ethnic composition in cardholders’ zip codes. Further, we
assessed the reliability of data from Census’s Household Pulse Survey
conducted from April 23, 2020, to October 11, 2021, related to how
survey respondents used or planned to use the federal economic impact
payments. We conducted data testing and reviewed Census’s technical
documentation, and we found these data to be sufficiently reliable for
describing how pandemic cash assistance contributed to changes in
credit card balances.
To assess the data reliability of the crosswalk of zip codes and zip-code
tabulation areas, we reviewed information from Census and UDS Mapper
(owner of the crosswalk data), including descriptions of the data and how
the data can be used. We determined the crosswalk data to be sufficiently
8See app. V for additional information on the regression analysis.
9The Household Pulse Survey, an experimental data product, is an interagency federal
statistical rapid response survey to measure household experiences during the COVID-19
pandemic. The Census Bureau conducts this survey in partnership with five other
agencies from the Federal Statistical System. Response rates over our period of analysis
ranged from 2.3 percent to 7.5 percent. Census applied weighting adjustments to mitigate
nonresponse bias, according to Census. All reported estimates have a relative margin of
error of 19 percent or less of the estimate, and all reported comparisons are statistically
significant at the 95 percent confidence level.
Page 51 GAO-23-105269 Credit Cards
Appendix I: Objectives, Scope, and
Methodology
reliable for our purpose of matching zip codes in the Federal Reserve’s Y-
14M data to zip-code tabulation areas in Census’s 5-year American
Community Surveys.
For all of our objectives, we also interviewed knowledgeable staff from
CFPB and the Federal Reserve and representatives from the six large
credit card issuers we selected, three credit reporting agencies (Equifax,
Experian, and TransUnion), the American Bankers Association, and the
National Consumer Law Center.
We conducted this performance audit from May 2021 to September 2023
in accordance with generally accepted government auditing standards.
Those standards require that we plan and perform the audit to obtain
sufficient, appropriate evidence to provide a reasonable basis for our
findings and conclusions based on our audit objectives. We believe that
the evidence obtained provides a reasonable basis for our findings and
conclusions based on our audit objectives.
Page 52 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
Appendix II: Data and Descriptive
Statistics
We used data from the Board of Governors of the Federal Reserve
System for all three of our objectives and from the Census Bureau for two
of them. This appendix describes the steps we took to clean the data and
create our analysis variables. It also describes the analysis dataset and
the descriptive statistics we analyzed.
Federal Reserve Credit Card Data
To examine consumer credit card usage, we obtained a 0.1 percent
nongeneralizable sample of consumer credit card account data from the
Federal Reserve’s Capital Assessments and Stress Testing Report (FR
Y-14M). For each account and month, the credit card account data
included monthly observations on purchase amounts, fees, interest
charges, cycle ending balances, payment amounts, whether an account
was closed, and the number of days a payment was past due. For each
account and month, the data also included observations on the
cardholder’s billing zip code and credit score, and the interest rate and
credit limit on the account, among other things. The data we analyzed are
an unbalanced panel with observations on each account for months
during the period from June 2013 through December 2021; the specific
months for which we have information vary from account to account.1
The Federal Reserve credit card data are collected from bank holding
companies, savings and loan holding companies, and intermediate
holding companies with $100 billion or more in assets.2 Therefore, we
cannot generalize the results of our analysis to credit card accounts in all
banks in the U.S. Additionally, the data sample excluded observations
with an address in a zip code that had only one credit card issuer in that
1The Federal Reserve first started collecting the Y-14M data in June 2012. However, our
analysis used data beginning in June 2013 because the Federal Reserve did not collect
some data elements until that time.
2The top 10 credit card issuers held 82 percent of the outstanding credit card balances as
of December 31, 2021, as reported by the Federal Reserve. See Board of Governors of
the Federal Reserve System, Report to Congress: Profitability of Credit Card Operations
of Depository Institutions (July 2022).
Page 53 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
month. The percentage of accounts excluded ranged from 0.3 percent of
accounts with billing zip codes in New Jersey to 12 percent of those in
Vermont. Our analysis included only general purpose credit cards issued
on networks, such as Visa, Mastercard, American Express, and Discover,
which are accepted by a wide variety of merchants. The Federal Reserve
data allowed us to examine credit card usage by account, but the data did
not have information that would allow us to link multiple accounts held by
a given individual cardholder. We also cannot link accounts to a specific
issuer.
To prepare the Federal Reserve data for our analysis, we performed the
following steps:
· We identified 358,626 fully duplicate observations, which we dropped.
· The reference numbers assigned to accounts were assigned by the
issuers, so while account reference numbers are unique within an
issuer, they may not be unique across issuers.3 We identified
accounts with the same reference numbers that were active in the
same month, which likely occurred because of this issue. To address
this issue, we separated accounts by first assigning unique
identification numbers to those duplicate accounts that had different
account origination dates. For the smaller number of duplicate
account reference numbers in a month that had the same origination
date, we separated them by billing zip code.
· We removed account-month observations with a billing address
outside of the 50 U.S. states and the District of Columbia—that is, we
excluded addresses in the U.S. territories and outside of the U.S.
To further prepare the Federal Reserve data for analysis, we imputed
values for variables with missing values where possible. Some accounts
in the data did not have information for one or more of the following
variables in some months: cycle ending balance, payment amount, and
fees and interest charges. To address this data limitation, we used values
of other relevant variables and imputed the missing values for these
variables to the extent possible. For example, for accounts missing cycle
ending balances, we replaced missing observations with the sum of
promotional balances, cash advance volume, penalty balances, and other
balances if those values were available. Table 1 describes the
3The account reference number is a unique identifier for each credit card account that is
different from the credit card’s account number and stays the same from month to month.
Page 54 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
relationships we used to impute these variable observations with missing
values.
Table 1: Relationships Used to Impute Credit Card Data Variable Observations with
Missing Values
Variable Description of imputation
Cycle ending Equals the sum of the promotional balances, cash advance
balance balances, penalty balances, and other balances at the end of the
billing cycle.
Payment amount Equals zero if account was newly opened.
Equals the sum of cycle beginning balance, new purchases,
balance transfers, cash advances, convenience checks, interest,
and fees, and less cycle ending balance and other credits.
Fees and interest Equals the sum of cycle ending balance and new purchases, and
charges less cycle beginning balance, payments, and credits.
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
For all account-month observations with the previous month’s cycle
ending balance available, we constructed a cycle beginning balance
equal to the previous month’s cycle ending balance. If the previous
month’s cycle ending balance was not available, we set the cycle
beginning balance to zero if the account was new (defined as an account
that was opened within 3 months of the observation). If the previous
month’s cycle ending balance was not available and the account was not
newly issued, we set the cycle beginning balance to be equal to the sum
of the cycle ending balance, payments, and credits, less new purchases,
balance transfers, cash advances, convenience checks, interest, and fees
if those values were available.
For each account-month observation, we created a new variable
describing the status of the account. To track the status of an account, we
created six categories for the account status for each month to classify
credit card accounts based on how cardholders use the cards and
whether interest is charged during each month.4
· Transacting accounts. If the cardholder pays all of the account
balance due for the billing cycle by the due date, we refer to the
cardholder as a “transactor” and to the account as a “transacting
account.” We primarily assign this status by identifying accounts with
4We assigned status according to the payment actions in the month, not the spending
decisions in that month.
Page 55 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
a positive starting balance but no finance charges due.5 Repeat
transactors are not charged interest on the end-of-cycle balance or on
new purchases during the grace period (between the end of a billing
cycle and the date payment is due). Accounts with 1 inactive month
(defined below) that is both preceded and followed by transacting
months are classified as transacting, even though they have no
balance to pay off in 1 month.
· Revolving accounts. If the cardholder pays less than the entire
account balance due for the billing cycle, or if the cardholder’s
account has been delinquent for less than 90 days, we refer to the
cardholder as a “revolver” and to the account as a “revolving account.”
We primarily assign this status by identifying accounts with interest
charges incurred in the billing cycle that are not delinquent for more
than 90 days.6 In contrast with transactors, revolvers are charged
interest on the remaining unpaid balances from prior cycles (also
called revolving balances), and immediately on any new purchases
they make in the current cycle.7
· Seriously delinquent accounts. If the cardholder does not pay at
least the minimum payment due for 90 days or more, we classify the
account as seriously delinquent. We use the issuer’s identification of
delinquency for this classification, and it overrides other designations
based on our account status definitions.
· Inactive accounts. If the cardholder has no starting balance or
payment activity in the billing cycle, we classify the account as
inactive. In this case, the cardholder would have chosen not to use
the card to make purchases in the prior month and therefore would
have no balance to pay, even though the account is still active in the
sense that the cardholder could continue to use it. As we have
designated statuses, technically, it is impossible for accounts to pass
directly from revolving status to inactive status. This is because before
an account could meet our criteria for inactive status, the cardholder
would have to pay off the revolving balance in the month before and
switch to transacting status for 1 month. However, it is not uncommon
5As a consequence of our assignment approach, any accounts that are carrying a balance
that has a zero percent promotional annual percentage rate are also classified as
transacting accounts.
6As a consequence of our assignment approach, accounts that were carrying a balance
previously and have started to pay off their balance in full are not classified as transacting
accounts until they are eligible for their issuer’s grace period.
7Revolving balances can also include fees if cardholders fail to make payments by the due
date.
Page 56 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
for an account to go inactive after the revolving balance is paid off. To
capture this movement in our transition model, for any single month of
transacting status that followed a revolving month and preceded an
inactive month, we reassigned the account to revolving status for that
month. This approach allowed us to catalogue these transitions as
movements from revolving status to inactive status.
· Closed accounts. We classify accounts as closed if the customer
has requested their closure or if the customer has died. Closed
accounts are flagged as such by the issuer and these flags override
other designations, including delinquency.
· Charged-off accounts. Charged-off accounts are flagged as such by
the issuing institution. These accounts have generally been seriously
delinquent and the institution has reason to believe the balance will
not be paid.
After we conducted the data cleaning steps described above, the final
dataset included 30.5 million account-month observations on 655,033
general purpose credit card accounts (see table 2).
Table 2: Information on General Purpose Credit Card Account Data Analyzed by GAO
Average number of Minimum number of Maximum number of
Total number of Total number of account- observations per observations per observations per
credit card accounts month observations account account account
655,033 30,469,820 47 1 103
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: General purpose credit cards are credit cards issued on networks, such as Visa, Mastercard,
American Express, and Discover, which are accepted by a wide variety of merchants.
In the models of status transitions (see app. IV), we used the next
month’s status, so we created a variable that captured the next month’s
status associated with the account. This variable was left missing when
the current month was the last observation included in the dataset. We
collapsed our six status categories into five because not all accounts
passed through the seriously delinquent status prior to being charged off.
We combined these two statuses here to account for the rare instances
when an account would be revolving or delinquent for less than 90 days
and transition directly to charged-off status. Further, it did not make
logical sense for a transacting account to transition directly to seriously
delinquent status, so in the few cases where that occurred, we dropped
the account-month observation as containing an error. Once we captured
the next status, we had no further need for closed or charged-off
accounts, so we dropped these from our data.
Page 57 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
Census Bureau Income and Race/Ethnicity
Data
To derive the median household income and the racial and ethnic
composition of cardholders’ zip codes, we used data from the Census
Bureau’s 5-Year American Community Surveys (ACS) for 2013–2020.
We matched ACS data at the zip-code tabulation area level to
cardholders’ zip codes in the credit card data.8 We matched the last year
of each 5-year ACS average with the year of the credit card billing cycle
for credit card data from June 2013 through December 2020. For
instance, credit card accounts with a billing cycle month in 2013 were
matched to the 2009–2013 5-year ACS. The only exception was for credit
card accounts with a billing cycle month in 2021, which we matched with
the 2016–2020 5-year ACS because the 2017–2021 ACS data were not
available at the time we created the dataset.9
We adjusted median household incomes for inflation and measured them
in constant 2021 dollars using the Consumer Price Index for All Urban
Consumers. Using the zip-code median household incomes, we created
six income groups: under $25,000, $25,000–$49,999, $50,000–$74,999,
$75,000–$99,999, $100,000–$149,999, and $150,000 and above. We
then assigned credit card accounts to one of these groups based on the
estimated median household income in the cardholders’ zip codes.
Additionally, we merged the Federal Reserve credit card data with ACS’s
income distribution data, which include the number of households in each
zip-code tabulation area that fall into 16 more granular nominal income
groups, for our regression analyses related to credit terms and revolving
balances (as discussed in app. VI).
To describe the racial and ethnic composition of cardholders’ billing zip
codes, we used the shares of residents in each zip code that were non-
Hispanic White, Asian, Black or African American (referred to as “Black”
in this report), Hispanic or Latino of any race (referred to as “Hispanic” in
this report), and Other.
8Zip-code tabulation areas are generalized areal representations of U.S. Postal Service
zip code service areas, which are not areal features but a collection of mail delivery
routes.
9Census reported that it had to make special adjustments to the results of the 2020 1-year
ACS due to a higher-than-usual nonresponse rate during the COVID-19 pandemic.
Page 58 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
Descriptive Statistics
Table 3 presents descriptive statistics for the variables in our analysis.
Table 3: Characteristics of Consumer Credit Card Accounts, June 2013–December 2021
Standard
Variable Mean deviation Median Minimum Maximum
Data from the Board of Outstanding balance $1,749 $3,539 $312 -$335,497 $198,776
Governors of the Federal due at the end of a
Reserve System on billing cycle
credit card accounts Payment made by $540 $1,829 $70 -$341,867 $480,771
cardholders per
month
Annualized interest 17.3% 6.2% 17.0% 0% 32.7%
rate of the account
Credit limit $9,070 $8,457 $7,000 $0 $500,000
Credit score 751 80 768 1 991
Interest charges $51 $67 $29 -$21,162 $8,266
incurred
Data from the Census Median household $77,962 $30,989 $71,561 $3,948 $290,796
Bureau on distribution of income
household income in Percentage of 5.6% 3.9% 4.6% 0.0% 100.0%
cardholders’ zip codes households in a zip
code with household
income that was
under $10,000
Percentage of 4.1% 2.7% 3.5% 0.0% 100.0%
households in a zip
code with household
income of $10,000–
$14,999
Percentage of 4.1% 2.4% 3.7% 0.0% 100.0%
households in a zip
code with household
income of $15,000–
$19,999
Percentage of 4.3% 2.3% 4.1% 0.0% 100.0%
households in a zip
code with household
income of $20,000–
$24,999
Percentage of 4.2% 2.1% 4.0% 0.0% 100.0%
households in a zip
code with household
income of $25,000–
$29,999
Page 59 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
Standard
Variable Mean deviation Median Minimum Maximum
Percentage of 4.3% 2.0% 4.2% 0.0% 100.0%
households in a zip
code with household
income of $30,000–
$34,999
Percentage of 4.1% 1.8% 4.0% 0.0% 100.0%
households in a zip
code with household
income of $35,000–
$39,999
Percentage of 4.1% 1.8% 4.1% 0.0% 100.0%
households in a zip
code with household
income of $40,000–
44,999
Percentage of 3.7% 1.6% 3.7% 0.0% 100.0%
households in a zip
code with household
income of $45,000–
$49,999
Percentage of 7.4% 2.4% 7.4% 0.0% 100.0%
households in a zip
code with household
income of $50,000–
$59,999
Percentage of 9.7% 2.7% 9.7% 0.0% 100.0%
households in a zip
code with household
income of $60,000–
$74,999
Percentage of 12.8% 3.3% 12.8% 0.0% 100.0%
households in a zip
code with household
income of $75,000–
$99,999
Percentage of 9.4% 3.3% 9.5% 0.0% 100.0%
households in a zip
code with household
income of
$100,000–$124,999
Percentage of 6.3% 3.0% 6.2% 0.0% 100.0%
households in a zip
code with household
income of
$125,000–$149,999
Page 60 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
Standard
Variable Mean deviation Median Minimum Maximum
Percentage of 7.3% 4.5% 6.6% 0.0% 100.0%
households in a zip
code with household
income of
$150,000–$199,999
Percentage of 8.5% 8.7% 5.4% 0.0% 100.0%
households in a zip
code with household
income that was
$200,000 and above
Data from the Census Percentage of Asian 6.8% 10.1% 3.1% 0% 100%
Bureau on racial and residents
ethnic composition of Percentage of Black 10.0% 15.0% 4.3% 0.0% 100.0%
cardholders’ zip codes or African American
residents
Percentage of 16.6% 19.1% 9.2% 0.0% 100.0%
Hispanic or Latino
residents
Percentage of White 63.2% 26.1% 69.8% 0.0% 100.0%
residents
Percentage of 3.3% 3.5% 2.7% 0.0% 100.0%
residents who were
Other race
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and Census Bureau. | GAO-23-105269
Note: Negative values in outstanding balances and payment amounts could be due to data entry
errors or to cardholders being owed a credit at the end of the billing cycle. Additionally, the minimum
and maximum values of credit scores could be due to data errors since credit scores generally range
from 300 to 850.We used the Census Bureau’s 5-Year American Community Surveys for 2013–2020.
Analysis of Credit Card Usage
Account status. We calculated the share of active accounts (i.e.,
transacting, revolving, or seriously delinquent) by status for each month
for the period from June 2013 through December 2021. Additionally, we
calculated the average share of active accounts that were revolving, by
credit score category and zip-code-level median household income, for
the period from June 2013 through December 2019.
Revolving balances. For revolving accounts, we calculated the median
revolving balance amount for the period from June 2013 through
December 2021. Additionally, we calculated the percentage of accounts
that reached or exceeded their credit limit for each month for the period
from June 2013 through December 2019. Furthermore, we calculated the
shares of transacting accounts and revolving accounts that made
Page 61 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
purchases, and purchase amounts for those that did. All dollar amounts
are in nominal terms unless otherwise noted.
Durations of revolving and transacting episodes. We calculated the
duration of revolving episodes using Federal Reserve data from June
2013 through December 2021. We defined a revolving episode as a
period of 1 or more consecutive months in which a credit card account is
in revolving status (see fig. 14). A revolving episode ends when the status
of the account changes from revolving to transacting, seriously
delinquent, inactive, closed, or charged off. Each account can have
multiple revolving episodes. Similarly, we defined a transacting episode
as a period of 1 or more consecutive months in which the credit card
account is in transacting status.
Figure 14: Examples of Revolving Episodes of Differing Lengths
Page 62 GAO-23-105269 Credit Cards
Appendix II: Data and Descriptive Statistics
Data for Figure 14: Examples of Revolving Episodes of Differing Lengths
Number of 1 revolving episode 2 revolving 1 revolving episode
revolving episodes (1 month) episodes (3 (6 months)
(revolving episode months and 2
length) months
respectively)
Month Cardholder A Cardholder B Cardholder C
1 Transact Transact Transact
2 Revolve Revolve Revolve
3 Transact Revolve Revolve
4 Transact Revolve Revolve
5 Transact Transact Revolve
6 Transact Transact Revolve
7 Transact Revolve Revolve
8 Transact Revolve Transact
Source: GAO. | GAO-23-105269
We assigned the account-months that were in an episode for 12 months
or more the designation “long term” for the purposes of our descriptive
statistics, regardless of where they were in that episode.10
10Because we defined long-term revolving episodes as lasting 1 year or more, December
2020 was the last date for which we had data to make an accurate count of long-term
status.
Page 63 GAO-23-105269 Credit Cards
Appendix III: Cluster Analysis of
Race/Ethnicity in Cardholders’ Zip Codes
Appendix III: Cluster Analysis of
Race/Ethnicity in Cardholders’
Zip Codes
We conducted a statistical analysis to identify racial and ethnic
distributions that were typical for the zip codes of credit cardholders in our
analysis. This appendix presents the development, estimation, results,
and limitations of this analysis.
Data
We used a 0.1 percent nongeneralizable sample of credit card accounts
from the Board of Governors of the Federal Reserve System’s Y-14M
credit card data from June 2013 through December 2021, which included
zip-code information. Additionally, we used zip-code-level racial and
ethnic composition data from the Census Bureau’s 5-year American
Community Surveys from 2013 through 2020. See appendix II for
additional information on the data.
Methodology
Our report examined how credit terms, revolving episode durations, and
revolving balances of cardholders varied by race and ethnicity. We
estimated these differences using econometric models that accounted for
the racial and ethnic characteristics of cardholders’ zip codes. To illustrate
the implications of the models’ estimates, we used cluster analysis
methods to identify racial and ethnic distributions that were typical for the
zip codes in our population. We used the models to estimate how
outcomes varied across these typical zip codes.
To implement the cluster analysis, we identified a set of zip codes of
“typical” racial and ethnic composition in September 2016—the middle
month of our prepandemic period (June 2013–December 2019). This
approach helped address two aspects of our data. First, the mix of zip
codes represented in our data shifted moderately from June 2013 through
December 2021 because individual credit card accounts entered and
exited our data, and some changed their billing address to new zip codes
Page 64 GAO-23-105269 Credit Cards
Appendix III: Cluster Analysis of
Race/Ethnicity in Cardholders’ Zip Codes
during our period of analysis. Second, the racial and ethnic composition
of the zip codes also changed over this period.
To determine whether zip-code clusters using alternative time periods
would produce substantially different results, we also identified a set of
“typical” zip codes in September 2017 (the middle month of the entire
data period), as well as at the beginning and ending points of the period
in June 2013 and December 2021. While we found small shifts in the
racial and ethnic composition of the zip codes between June 2013 and
December 2021, the zip code clusters in September 2016 and September
2017 were similar. As a result, we used the clusters from September
2016 in all of our models on revolving episode durations, credit terms,
and revolving balances by race and ethnicity in cardholders’ zip codes
(discussed in subsequent appendixes). This approach helped ensure that
the variations in our model results are associated with variables included
in the models and not changes in the zip-code clusters.
We used a k-means cluster algorithm to identify clusters of zip codes with
similar racial and ethnic composition in our data. The algorithm grouped
zip codes such that those within a cluster were as similar as possible and
zip codes in different clusters were as different as possible. We selected
the k-means algorithm because it was computationally inexpensive and
produced a centroid vector of racial and ethnic percentages for each
cluster. We used Stata to identify the clusters, using Euclidean distance
as our measure of (dis)similarity. Our implementation of the algorithm
used a random selection of the race and ethnicity composition of five zip
codes as the starting centers for the five clusters and allowed for up to
10,000 iterations of the algorithm.
The pool of zip codes we used to generate clusters were those included
in the September 2016 month of our data, and each zip code was
repeated as often as the number of accounts appearing in that zip code.
This weighted our cluster centers toward the racial and ethnic
composition of zip codes that appeared more frequently in our data,
allowing our clusters to be a more typical representation of the racial and
ethnic characteristics of the accounts in our data, rather than of the zip
codes in our data.
A potential limitation of the k-means algorithm was that it required us to
specify the number of clusters for the algorithm to identify. We therefore
did not use the data to identify the best-fitting number of clusters, such as
with hierarchical clustering methods. To address this limitation and
identify an appropriate number of clusters, we ran the algorithm multiple
Page 65 GAO-23-105269 Credit Cards
Appendix III: Cluster Analysis of
Race/Ethnicity in Cardholders’ Zip Codes
times while varying the number of clusters from five to eight. Our
estimates using the five-cluster approach produced five clusters similar to
five of the clusters identified when identifying more than five clusters, so
we decided to use the more concise approach to develop our illustrative
racial and ethnic compositions.
Results
We identified the following five clusters:
· Zip codes of predominantly White residents
· Zip codes with a majority of Hispanic or Latino residents
· Zip codes with a majority of Black or African American residents
· Zip codes with a mix of all racial and ethnic groups with a majority of
White residents
· Zip codes with a mix of all racial and ethnic groups with no majority
(where the share of Asian residents is the highest among all clusters)
Our cluster analysis did not identify a cluster of zip codes with a majority
of Asian residents.
Table 4 outlines the racial and ethnic composition of the representative
zip codes that have the average characteristics of each of the clusters.
Table 4: Racial and Ethnic Composition of Zip Code Clusters, September 2016
Cluster name White Black or African American Hispanic or Latino Asian Other
Cluster 1: Predominantly White 86% 4% 5% 3% 3%
Cluster 2: Majority Hispanic or Latino 21 9 62 6 2
Cluster 3: Majority Black or African 25 58 12 3 3
American
Cluster 4: Mixed race and ethnicity 59 11 18 8 4
with majority White
Cluster 5: Mixed race and ethnicity 30 7 20 37 6
with no majority
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and Census Bureau. | GAO-23-105269
Note: Percentages do not sum to 100 percent due to rounding.
Page 66 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Appendix IV: Econometric
Analysis of Revolving and
Transacting Episode Durations
We conducted econometric analysis of the durations of revolving and
transacting episodes and transitions under pandemic and prepandemic
conditions, as well as revolving and transacting episodes for cardholders
in zip codes with different racial and ethnic composition. This appendix
presents the development, estimation, results, and limitations of these
analyses.
Data and Definitions
We used a 0.1 percent nongeneralizable sample of credit card accounts
from the Board of Governors of the Federal Reserve System’s Y-14M
credit card data from June 2013 through December 2021. We
supplemented these data with zip-code-level racial and ethnic
composition data from the Census Bureau’s 5-year American Community
Surveys from 2013 through 2020. See appendix III for full details on the
use of zip-code racial and ethnic composition. For our analysis of
revolving episodes, we limited our sample to all accounts revolving in the
month of the observation. Similarly, for our analysis of transacting
episodes, we limited the sample to all accounts transacting in the month
of the observation.
Our outcome variable in all of the analyses was the status of the credit
card account in the following month, for which we used five categories.
Transacting accounts. Accounts for which the cardholder paid all of the
balance due for the billing cycle by the due date, thereby not accruing any
interest charges on purchases (see fig. 15). Accounts with 1 inactive
month (defined below) that is both preceded and followed by transacting
months are classified as transacting, even though they have no balance
to pay off in 1 month. An implication of how we define transacting
accounts is that if a cardholder is taking advantage of a promotion that
provides zero percent interest on a revolving balance for a fixed period of
time, the card account will remain classified as transacting until any
Page 67 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
remaining balances are charged a positive interest rate at the end of the
promotional period.
Figure 15: Illustrative Example of Credit Card Billing Cycle
Revolving accounts. Accounts for which the cardholder paid less than
the entire balance due for the billing cycle, therefore incurring interest on
the outstanding balance and any new purchases. We included in this
category any accounts that were delinquent (i.e., the cardholder paid less
than the minimum balance due by the due date) for less than 90 days.
Seriously delinquent and charged-off accounts. Accounts for which
the cardholder did not pay at least the minimum balance due for 90 days
or more or for which the card issuer has charged off the account balance
due. Not all accounts passed through the seriously delinquent status prior
to being charged off, so we combined these two statuses to account for
the rare instances when an account would be revolving or delinquent for
less than 90 days and transition directly to charged-off status.
Inactive accounts. Accounts with no starting balance or payment activity
in the billing cycle. In these cases, the cardholders chose not to use the
card to make purchases in the prior month and therefore had no balance
to pay, even though the accounts were still active in the sense that the
cardholder could continue to use the card. Technically, it is impossible for
credit card accounts to pass directly from revolving status to inactive
Page 68 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
status. This is because before an account could meet our criteria for
inactive status, the cardholder would have to pay off the revolving
balance in the month before and switch to transacting status for 1 month.
However, it is not uncommon for an account to go inactive after a
revolving balance is paid off. To capture this movement in our
econometric models, for any single month of transacting status that
followed a revolving month and preceded an inactive month, we
reassigned the account to revolving status for that month. This approach
allowed us to catalogue these transitions as movements from revolving
status to inactive status.
Closed accounts. Accounts closed by the cardholder or the issuer. In
addition to customer requests, issuers close accounts for a variety of
reasons, including fraud or death of the cardholder.
The primary independent variable used in our models is the account’s
“time-in-status”—in other words, the number of months the account has
been in the same status as it is in the current observation. For example, if
an account was in transacting status in the previous month and is
revolving this month, its time-in-status is 1. If an account first started
revolving in January and has done so in every month since, the time-in-
status in June is 6 and the time-in-status in July is 7. The total length of
the revolving episode is equal to the time-in-status of the last month
before the account transitions into another status type.
Episode Durations
We analyzed two relevant populations of revolving accounts with differing
distributions of revolving episode durations:
· Revolving episodes of a cohort of revolving accounts. This
population provides insight into the overall probability that a new
revolver may revolve for different durations.
· Total revolving episode duration of the population of revolvers at
a given point in time. This population provides insight into the
probability of encountering a revolving account in the middle of a
revolving episode of a particular duration, out of the population of
revolving accounts in a given time period.
As demonstrated in table 5, these populations had very different
distributions. Overall, the median observed duration of all revolving
episodes pooled as one cohort was 4 months. However, the median
Page 69 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
account in revolving status across our entire sample was in a revolving
episode that we observed lasting 32 months. In other words, while the
median episode was short—4 months—the typical account currently
revolving was in a long term revolving episode, one that had more in
common with the 90th percentile of revolving episodes—32 months—than
with the median episode.
Table 5: Observed Revolving Episode Durations for June 2013–December 2021
10th 90th
percentile Median percentile
Observed duration of revolving episodes, 1 month 4 months 32 months
as a pooled cohort
Observed duration of revolving episodes 4 months 32 months 93 months
for all accounts in revolving status
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Furthermore, many of our revolving episodes were right-censored (we
could not observe when the episode ended because it ended after the
period of our dataset), left-censored (we could not observe when the
episode started because it started before the period of our dataset), or
both (we were unable to observe the entire span of the episode). Of the
observations of accounts in revolving status available to be used in our
analyses of revolving episodes, 3.2 million were left-censored and 3.3
million were right-censored. In contrast, 5.2 million account observations
were in revolving episodes whose start and finish we were able to
observe. Because we were more likely to observe the full length of a
shorter episode than a longer one, the total observed length of the
episodes of which those observations were a part was very different
between observations in uncensored episodes (average of 24 months)
and in censored episodes (average of 62 and 60 observed months for
left- and right-censored episodes, respectively).
We were able to capture the impact of censoring on our ability to directly
measure the durations of revolving episodes. For each revolving episode,
we looked at the observable length and, if the episode was censored, we
considered the possibility that its true length was longer than the longest
possible observable length. We could then calculate the median of the
revolving episode length in our data in two ways: (1) we could assume all
of these episodes in our data were fully observed, and (2) we could
assume that none of the censored revolving episodes in our data were
shorter than the longest observable episode (i.e., that they were at least
104 months long). These two approaches gave lower and upper bounds
Page 70 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
to the possible median statistic of revolving episode length. The range of
uncertainty about the median value du
+ and end (see fig. 16).
Figure 16: Length of Median Revolving Episode for Accounts in Revolving Status by Month
Data for Figure 16: Length of Median Revolving Episode for Accounts in Revolving Status by Month
Year Month Range for median duration of revolving Range for median duration of revolving
episodes with fully observed duration episodes with partially observed duration
2013 6/1/2013 16 104
2013 7/1/2013 19 104
2013 8/1/2013 20 104
2013 9/1/2013 22 104
2013 10/1/2013 22 104
2013 11/1/2013 23 104
2013 12/1/2013 24 104
2014 1/1/2014 24 104
2014 2/1/2014 25 104
2014 3/1/2014 26 104
2014 4/1/2014 27 104
Page 71 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Year Month Range for median duration of revolving Range for median duration of revolving
episodes with fully observed duration episodes with partially observed duration
2014 5/1/2014 28 104
2014 6/1/2014 29 104
2014 7/1/2014 29 104
2014 8/1/2014 30 104
2014 9/1/2014 30 104
2014 10/1/2014 30 104
2014 11/1/2014 31 104
2014 12/1/2014 31 104
2015 1/1/2015 31 104
2015 2/1/2015 32 104
2015 3/1/2015 33 104
2015 4/1/2015 34 104
2015 5/1/2015 35 93
2015 6/1/2015 35 78
2015 7/1/2015 35 71
2015 8/1/2015 36 69
2015 9/1/2015 36 63
2015 10/1/2015 36 59
2015 11/1/2015 36 57
2015 12/1/2015 37 55
2016 1/1/2016 37 52
2016 2/1/2016 37 51
2016 3/1/2016 38 51
2016 4/1/2016 39 52
2016 5/1/2016 40 52
2016 6/1/2016 40 51
2016 7/1/2016 40 50
2016 8/1/2016 40 48
2016 9/1/2016 41 47
2016 10/1/2016 41 46
2016 11/1/2016 41 46
2016 12/1/2016 39 49
2017 1/1/2017 39 47
2017 2/1/2017 39 46
2017 3/1/2017 39 47
2017 4/1/2017 40 48
Page 72 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Year Month Range for median duration of revolving Range for median duration of revolving
episodes with fully observed duration episodes with partially observed duration
2017 5/1/2017 41 48
2017 6/1/2017 40 47
2017 7/1/2017 40 46
2017 8/1/2017 40 46
2017 9/1/2017 39 45
2017 10/1/2017 39 46
2017 11/1/2017 39 45
2017 12/1/2017 39 45
2018 1/1/2018 38 44
2018 2/1/2018 38 44
2018 3/1/2018 39 44
2018 4/1/2018 40 45
2018 5/1/2018 40 46
2018 6/1/2018 40 45
2018 7/1/2018 39 45
2018 8/1/2018 39 44
2018 9/1/2018 39 44
2018 10/1/2018 39 44
2018 11/1/2018 38 44
2018 12/1/2018 38 44
2019 1/1/2019 37 42
2019 2/1/2019 36 42
2019 3/1/2019 36 43
2019 4/1/2019 37 45
2019 5/1/2019 36 46
2019 6/1/2019 36 46
2019 7/1/2019 35 46
2019 8/1/2019 34 46
2019 9/1/2019 34 46
2019 10/1/2019 33 48
2019 11/1/2019 32 50
2019 12/1/2019 32 50
2020 1/1/2020 31 50
2020 2/1/2020 30 51
2020 3/1/2020 30 54
2020 4/1/2020 30 59
Page 73 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Year Month Range for median duration of revolving Range for median duration of revolving
episodes with fully observed duration episodes with partially observed duration
2020 5/1/2020 31 76
2020 6/1/2020 31 104
2020 7/1/2020 31 104
2020 8/1/2020 30 104
2020 9/1/2020 30 104
2020 10/1/2020 29 104
2020 11/1/2020 28 104
2020 12/1/2020 27 104
2021 1/1/2021 26 104
2021 2/1/2021 26 104
2021 3/1/2021 26 104
2021 4/1/2021 26 104
2021 5/1/2021 25 104
2021 6/1/2021 24 104
2021 7/1/2021 22 104
2021 8/1/2021 20 104
2021 9/1/2021 18 104
2021 10/1/2021 17 104
2021 11/1/2021 16 104
2021 12/1/2021 14 104
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
To address challenges posed by censoring for measuring episode
duration directly, we took an indirect approach to modeling details about
revolving episode lengths, as described in detail later in this appendix.
This indirect approach allowed us to maximize the use of observations in
our sample, including censored episodes, to build simulated distributions
of the durations of a revolving episode cohort for revolving accounts
under prepandemic or pandemic conditions, as well as for revolving
accounts from billing zip codes of different racial and ethnic composition.
Because our descriptive results on revolving episode durations
highlighted the importance of longer-duration revolving accounts, we
report on the 90th percentile of the distribution of durations for a cohort of
new revolving episodes to provide an illustration of the experience of a
typical revolving account in a given month.
Page 74 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Methodology of Analysis of Revolving Episodes
under Prepandemic and Pandemic Conditions
We used a multinomial logit model to estimate relationships between the
number of months an account has been in revolving status and the
likelihood it would either continue revolving in the next month or transition
to a different account status, under pandemic and prepandemic
conditions. We used the coefficients to derive probabilities that an
account in revolving status would continue to revolve or transition
conditional on the number of months in revolving status, under pandemic
and prepandemic conditions. We then used these conditional probabilities
to construct two Markov transition matrices, one for the pandemic period
and one for the prepandemic period. Finally, we used the Markov
transition matrices to simulate the path a hypothetical cohort of revolving
episodes (i.e., all revolving episodes that start in the same month) would
take over time under pandemic and prepandemic conditions and to
estimate the share of the cohort that would exit revolving status in each
month and into which status.
Multinomial logit analysis. We used multinomial logit models to
estimate the probability that an account in revolving status would either
continue to revolve or transition to each of the other four account statuses
(transacting, inactive, seriously delinquent, or closed) as a function of the
account’s time-in-status under pandemic and prepandemic conditions.1
We assume that after a certain period of revolving, the time-in-status
matters less than it did originally, and we assume that the continuation
and exit probabilities are constant once an account has been revolving
long term (at least 18 months). We assume this constant transition risk
after 18 months of revolving.2 This allows us to balance the loss of long-
1Multinomial logit analysis has been used in a variety of contexts to measure transitions
from a baseline state to several outcome states. For example, it has been used to
measure teacher career paths, credit union decline, and labor market life cycles. See
GAO, National Credit Union Administration: Additional Actions Needed to Strengthen
Oversight, GAO-21-434 (Washington, D.C.: Sept. 23, 2021); Christian Duel, “Expanding
the Markov Chain Toolbox: Distributions of Occupation Times and Waiting Times,”
Sociological Methods & Research, vol. 50, no. I (2021): 401–428; and Cassandra M.
Guarino, Abigail B. Brown, and Adam E. Wyse, “Can Districts Keep Good Teachers in the
Schools That Need Them Most?,” Economics of Education Review, vol. 30 (2011): 962–
979.
2We conducted sensitivity analyses on our definition of long-term revolving accounts and
determined that a 24-month threshold gives qualitatively similar results.
Page 75 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
term revolving accounts from our dataset due to left censoring with the
restricted assumptions imposed by assuming a constant transition risk.
We specified our baseline model as follows:
where P(Next statusi) is the status of the revolving episode in the next
month. The other variables are as follows: is the constant and served
to capture the long-term likelihood of transition, along with any fixed
factors that affect the probability of transition; is a vector of indicator
variables to capture the individual account’s current time in revolving
status in the first 18 months of the status; and is an indicator for
whether the account-month observation was in April 2020–December
2021. We limited our sample to all account-months where we classified
the account as revolving in the current month and were able to identify
the status of the account in the next month.
We included 18 indicator variables to control for the number of months an
account had been in revolving status, one for each of the first 18 months
of the revolving episode. We reserved the long-term revolving status as
the reference category, and therefore the constant of the model
effectively estimated the coefficient for those accounts in long-term
revolving status of 19 months or longer. This approach assumed that the
probabilities of continuing to revolve or transitioning to another status
were constant once a revolving episode lasted longer than 18 months
(“long-term revolving episode”).
We used this approach to address left censoring, as previously
discussed.3 We dropped all account-month observations that were in the
first measured 18 months of a left-censored revolving episode, but
retained the rest of the censored revolving episodes after the first 18
months since we no longer needed a precise measure of the time-in-
status.
We used the multinomial logit model to address the right censoring issue.
The model allowed each revolving episode to contribute the information
that was captured in our data: the revolving episodes either continued to
3Our approach to address the left censoring issue is similar to methodology in other
research. See Steven J. Haider and Jacob Alex Klerman, “Dynamic Properties of Welfare
Caseload,” Labour Economics, vol. 12 (2005): 629–648.
Page 76 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
revolve for an additional month or transitioned to one of the other four
account statuses in a particular month.
Additionally, we interacted all indicators (including the constant term) with
a COVID-19 pandemic indicator to allow for separate estimates of
transition probabilities during the pandemic period. We included all
months in the period April 2020–December 2021 in our COVID-19
indicator, since April was the first month when all payment decisions were
made after the pandemic was declared on March 13, 2020.
We did not control for calendar time, except for interacting an indicator for
months that occurred after the declaration of the pandemic in March
2020. We determined this approach was reasonable because our
descriptive findings suggested that the number of revolving accounts as a
percentage of active accounts had little or no change over time except
modest seasonal fluctuations and during the months when the pandemic
was declared or when government pandemic assistance payments were
disbursed.
We used a similar approach to model relationships between the number
of months an account had been in transacting status and the likelihood it
either continued to transact or transitioned to one of the other account
statuses during the pandemic and prepandemic periods. When analyzing
transacting episodes, we had only four possible states to transition to:
continuing to transact, revolving, inactive, and closed. Transacting
accounts cannot transition in 1 month to seriously delinquent status.
Markov chain analysis. To simulate the distributions of revolving
episode durations and transitions under pandemic and prepandemic
conditions, we used a Markov chain analysis. We first used the estimated
multinomial logit models described above to derive probabilities that an
account in revolving status would either continue to revolve or transition
to another status conditional on the number of months in revolving status,
before and during the pandemic. We then used the conditional
probabilities to construct pandemic and prepandemic Markov chain
transition matrices. Finally, we passed a state vector representing a
hypothetical cohort of revolving accounts through each of the transition
matrices repeatedly to simulate the share of the original cohort still
revolving each month over the course of a decade (i.e., 120 months)
along with the initial status of accounts that transitioned out of revolving
status.
Page 77 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Using the coefficients of the multinomial logit model, we calculated the
conditional probabilities using the following pair of formulas, for a
particular set of values for the vector :
and
where s = (0, 2, 3, 4) for the statuses other than revolver status (which is
s = 1). For our transactor models, s = 0 is our baseline outcome instead.
The vector is a vector of indicator variables for the time-in-status of the
conditional probabilities being calculated, where all are equal to zero
except for the element that is equal to the indicator for the time-in-status
being calculated, which is set to one. For example, if the time-in-status is
a vector of indicators, a card that has been revolving for 6 months will be
coded as a zero for all elements of the vector except for the sixth
element, which is coded as a one. If that account is in its 6th month after
March 2020, then the time-in-status interacted with the COVID indicator
will also have a one in the sixth element.
Generically, a Markov chain model describes how individuals move
between various states. Such a model can be represented as follows:
where t is the time period, S(t) is a vector that contains the share of
individuals in each of K states, and M(t) is the transition matrix between
the states, which translates the state vector to its state in the next
period: . The K possible states include all of the possible states:
revolving for each of the 18 months and long term, as well as the other
statuses a revolver can transition to. Because we were looking at one
cohort of revolving episodes, we treated each of the other four statuses
as absorbing states and did not model the transition back into revolving
status here. We kept the prepandemic and pandemic periods separate
and estimated two different Markov chains for the two conditions, even
though in practice, many revolving episodes bridged the two periods. As a
result, for our revolving episode models, our M(t) is a 23 by 23 transition
matrix.
Page 78 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
We allowed the probability of transitioning from revolving status to
another status to vary by time-in-status in S(t-1) for the first 18 months,
and then those probabilities would remain constant as the account
entered long-term time-in-status. Once a revolving episode transitioned to
a different status, we treated that status as an absorbing state for the
purpose of the revolving episode cohort’s distribution (though in real life,
those episodes may reenter revolving status at a future time).4 Therefore,
our state vector S(t) would be a 23 by 1 vector:
Our transition matrix M(t) would be a 23 by 23 matrix:
The transition probability matrix, M(t), remained constant over calendar
time, with the exception of a “permanent” change (within the scope of our
data) that occurred at the start of the pandemic.5 We therefore
constructed two separate transition probability M(t) matrices, one for
transitions under prepandemic conditions and one for transitions under
pandemic conditions. For revolving episodes that overlapped both the
prepandemic and pandemic periods, they contributed the information
about their transition risk in each month to the period that each month
occurs in: for example, for a revolving episode that lasted from January
2020–June 2020, its continuation as a revolver in January–March would
contribute to the estimation of the baseline transition risks for revolvers in
4We focused our analysis on the distribution of an episode cohort, so even though an
account can reenter revolving status in the future, it cannot reenter the same cohort of
revolving status.
5Another study also used a similar approach. See Haider and Klerman, “Dynamic
Properties of Welfare Caseload.”
Page 79 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
months 1–3, and it would contribute to both the baseline and the
pandemic interaction risk estimates for revolvers in months 4–6. When we
used those estimates for the Markov chain analyses, we were simulating
hypothetical scenarios where either the average prepandemic or the
average pandemic conditions persisted indefinitely. This facilitated an
understanding of the magnitude of the shift in cardholder behavior but
does not attempt to model the evolving distribution of revolver episodes
that begin prior to the pandemic and resolve at various points after the
pandemic began.
To simulate the predicted outcomes of a revolving episode cohort, our
S(1) is a vector with a 1 in the first row, followed by all zeros. Each time
we multiplied M(t) by S(t-1), we captured the resulting S(t) as the
distribution of the revolving episode cohort by months since the start of
the revolving episode.
We used a similar approach to simulate the distribution of transacting
episode durations and transitions.
Assumptions and Limitations
Our analysis of revolving episodes has limitations, and our results should
be interpreted with caution. The sample of credit card accounts we
analyzed may not represent the population of all credit card accounts,
and thus the sample of revolving episodes we analyzed may not
represent the population of all revolving episodes. It follows that our
results reflect the sample of revolving episodes we analyzed and may not
generalize to other revolving episodes.
In our multinomial logit model specification, we assumed that the
relationships between the number of months an account has spent in
revolving status and the likelihood it either continues to revolve or
transitions to another status were constant during the prepandemic
period. We also assumed that the relationships were constant, but
possibly different, during the pandemic period. As a result, the derived
Markov transition matrices for the pandemic and prepandemic periods
were also constant (but different). We assumed constant relationships
during the prepandemic period because the share of accounts in
revolving status varied little over time prior to the pandemic. We assumed
constant relationships during the pandemic period because that period is
short relative to the typical duration of revolving episodes in our data,
which limited our ability to reliably estimate more complex relationships.
Page 80 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
It follows from these assumptions that our simulated distributions of
revolving episode durations and transitions for the prepandemic and
pandemic periods are averages for each period. In both periods, we did
not attempt to parse factors other than time-in-status that might affect
risks of transitions, such as income, age, or other factors that might affect
credit card usage. For the pandemic period in particular, our results are
suggestive of revolving episode durations and transitions assuming the
state of the pandemic from April 2020 through December 2021 were to
continue for an extended period of time for the average credit cardholder.
However, the actual distribution of revolving episode durations and
transitions during some or all of the pandemic period might differ from our
simulated distributions because the conditions might not persist
sufficiently past our estimation period to have the simulated effect on all
revolving episodes, particularly those with longer durations. For example,
direct government pandemic assistance to individuals largely ended at
the end of 2021, and therefore we would expect that the conditions under
which revolvers are paying down their balances in 2022 might not be the
same as those we measured in 2020 and 2021. Further, neither our
results for the pandemic period nor our results for the prepandemic period
are necessarily generalizable to other periods.
In addition, while we allowed relationships between the number of months
an account spent in revolving status and the likelihood it either continued
to revolve or transitioned to another status to vary between the pandemic
and prepandemic periods, we cannot explain the reasons why those
relationships vary. The pandemic period differed from the prepandemic
period in a variety of ways that could have affected credit card usage,
including public health conditions (e.g., social distancing), economic
conditions (e.g., employment levels), and public policies (e.g., pandemic-
related government assistance). Our approach does not identify the
effects of specific pandemic conditions on credit card usage, and thus we
do not establish causal effects of any particular pandemic condition.
Similar limitations apply to our analysis of transacting episodes.
Results
Our analysis suggests that pandemic conditions were associated with
shorter revolving episodes. Also, under pandemic conditions, once the
revolving episodes ended, a larger share of accounts would have
transitioned to transacting status, and fewer would have transitioned to
seriously delinquent and charged-off, inactive, or closed status.
Page 81 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Figure 17 presents the estimated coefficients from the multinomial logit
model of credit card accounts in revolving status as a function of the time
spent in the current revolving episode, under prepandemic and pandemic
conditions, for each of the four transition outcomes modeled. In all cases,
a negative value of a baseline coefficient indicates that the outcome is
less likely than that of the reference category—remaining in revolver
status. A positive value of an interaction coefficient indicates that the
outcome is more likely than the baseline coefficient suggests, but not
necessarily more likely than the reference category outcome.
Page 82 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Figure 17: Multinomial Logit Combined Coefficients for Transitions from Revolving Episode to Next Status, under
Prepandemic and Pandemic Conditions
Page 83 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Data for Figure 17: Multinomial Logit Combined Coefficients for Transitions from Revolving Episode to Next Status, under
Prepandemic and Pandemic Conditions
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
long term closed status Baseline coefficient -6.42 -6.42 0.03 0.03
18 closed status Baseline coefficient -6.42 -6.28 0.15 0.15
17 closed status Baseline coefficient -6.42 -6.38 0.16 0.16
16 closed status Baseline coefficient -6.42 -6.26 0.14 0.14
15 closed status Baseline coefficient -6.42 -6.32 0.14 0.14
14 closed status Baseline coefficient -6.42 -6.37 0.14 0.14
13 closed status Baseline coefficient -6.42 -6.35 0.14 0.14
12 closed status Baseline coefficient -6.42 -6.35 0.13 0.13
11 closed status Baseline coefficient -6.42 -6.37 0.13 0.13
10 closed status Baseline coefficient -6.42 -6.31 0.12 0.12
9 closed status Baseline coefficient -6.42 -6.20 0.11 0.11
8 closed status Baseline coefficient -6.42 -6.26 0.11 0.11
7 closed status Baseline coefficient -6.42 -6.31 0.11 0.11
6 closed status Baseline coefficient -6.42 -6.32 0.10 0.10
5 closed status Baseline coefficient -6.42 -6.28 0.10 0.10
4 closed status Baseline coefficient -6.42 -6.18 0.09 0.09
3 closed status Baseline coefficient -6.42 -6.05 0.08 0.08
2 closed status Baseline coefficient -6.42 -5.55 0.06 0.06
1 closed status Baseline coefficient -6.42 -5.41 0.05 0.05
long term closed status Interaction with indicator of -0.47 -0.47 0.07 0.07
pandemic conditions
18 closed status Interaction with indicator of -0.47 -0.29 0.36 0.36
pandemic conditions
17 closed status Interaction with indicator of -0.47 -0.40 0.39 0.39
pandemic conditions
16 closed status Interaction with indicator of -0.47 -0.33 0.35 0.35
pandemic conditions
15 closed status Interaction with indicator of -0.47 -0.55 0.39 0.39
pandemic conditions
14 closed status Interaction with indicator of -0.47 -0.20 0.33 0.33
pandemic conditions
13 closed status Interaction with indicator of -0.47 -0.62 0.38 0.38
pandemic conditions
12 closed status Interaction with indicator of -0.47 -0.28 0.32 0.32
pandemic conditions
11 closed status Interaction with indicator of -0.47 -0.45 0.34 0.34
pandemic conditions
Page 84 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
10 closed status Interaction with indicator of -0.47 -0.75 0.37 0.37
pandemic conditions
9 closed status Interaction with indicator of -0.47 -0.88 0.36 0.36
pandemic conditions
8 closed status Interaction with indicator of -0.47 -0.51 0.31 0.31
pandemic conditions
7 closed status Interaction with indicator of -0.47 -0.39 0.29 0.29
pandemic conditions
6 closed status Interaction with indicator of -0.47 -0.62 0.30 0.30
pandemic conditions
5 closed status Interaction with indicator of -0.47 -0.34 0.25 0.25
pandemic conditions
4 closed status Interaction with indicator of -0.47 -0.40 0.23 0.23
pandemic conditions
3 closed status Interaction with indicator of -0.47 -0.44 0.21 0.21
pandemic conditions
2 closed status Interaction with indicator of -0.47 -0.26 0.14 0.14
pandemic conditions
1 closed status Interaction with indicator of -0.47 -0.40 0.12 0.12
pandemic conditions
long term seriously delinquent and charged- Baseline coefficient -5.11 -5.11 0.01 0.01
off statuses
18 seriously delinquent and charged- Baseline coefficient -5.11 -4.74 0.07 0.07
off statuses
17 seriously delinquent and charged- Baseline coefficient -5.11 -4.72 0.07 0.07
off statuses
16 seriously delinquent and charged- Baseline coefficient -5.11 -4.71 0.07 0.07
off statuses
15 seriously delinquent and charged- Baseline coefficient -5.11 -4.68 0.06 0.06
off statuses
14 seriously delinquent and charged- Baseline coefficient -5.11 -4.74 0.06 0.06
off statuses
13 seriously delinquent and charged- Baseline coefficient -5.11 -4.68 0.06 0.06
off statuses
12 seriously delinquent and charged- Baseline coefficient -5.11 -4.69 0.06 0.06
off statuses
11 seriously delinquent and charged- Baseline coefficient -5.11 -4.73 0.06 0.06
off statuses
10 seriously delinquent and charged- Baseline coefficient -5.11 -4.68 0.05 0.05
off statuses
9 seriously delinquent and charged- Baseline coefficient -5.11 -4.65 0.05 0.05
off statuses
Page 85 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
8 seriously delinquent and charged- Baseline coefficient -5.11 -4.66 0.05 0.05
off statuses
7 seriously delinquent and charged- Baseline coefficient -5.11 -4.65 0.05 0.05
off statuses
6 seriously delinquent and charged- Baseline coefficient -5.11 -4.69 0.05 0.05
off statuses
5 seriously delinquent and charged- Baseline coefficient -5.11 -4.61 0.04 0.04
off statuses
4 seriously delinquent and charged- Baseline coefficient -5.11 -4.58 0.04 0.04
off statuses
3 seriously delinquent and charged- Baseline coefficient -5.11 -4.30 0.03 0.03
off statuses
2 seriously delinquent and charged- Baseline coefficient -5.11 -4.97 0.04 0.04
off statuses
1 seriously delinquent and charged- Baseline coefficient -5.11 -5.53 0.05 0.05
off statuses
long term seriously delinquent and charged- Interaction with indicator of -0.30 -0.30 0.03 0.03
off statuses pandemic conditions
18 seriously delinquent and charged- Interaction with indicator of -0.30 -0.21 0.16 0.16
off statuses pandemic conditions
17 seriously delinquent and charged- Interaction with indicator of -0.30 -0.42 0.17 0.17
off statuses pandemic conditions
16 seriously delinquent and charged- Interaction with indicator of -0.30 -0.28 0.16 0.16
off statuses pandemic conditions
15 seriously delinquent and charged- Interaction with indicator of -0.30 -0.33 0.16 0.16
off statuses pandemic conditions
14 seriously delinquent and charged- Interaction with indicator of -0.30 -0.23 0.15 0.15
off statuses pandemic conditions
13 seriously delinquent and charged- Interaction with indicator of -0.30 -0.32 0.15 0.15
off statuses pandemic conditions
12 seriously delinquent and charged- Interaction with indicator of -0.30 -0.26 0.14 0.14
off statuses pandemic conditions
11 seriously delinquent and charged- Interaction with indicator of -0.30 -0.43 0.15 0.15
off statuses pandemic conditions
10 seriously delinquent and charged- Interaction with indicator of -0.30 -0.34 0.14 0.14
off statuses pandemic conditions
9 seriously delinquent and charged- Interaction with indicator of -0.30 -0.35 0.13 0.13
off statuses pandemic conditions
8 seriously delinquent and charged- Interaction with indicator of -0.30 -0.26 0.13 0.13
off statuses pandemic conditions
7 seriously delinquent and charged- Interaction with indicator of -0.30 -0.31 0.12 0.12
off statuses pandemic conditions
Page 86 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
6 seriously delinquent and charged- Interaction with indicator of -0.30 -0.24 0.12 0.12
off statuses pandemic conditions
5 seriously delinquent and charged- Interaction with indicator of -0.30 -0.21 0.10 0.10
off statuses pandemic conditions
4 seriously delinquent and charged- Interaction with indicator of -0.30 -0.31 0.10 0.10
off statuses pandemic conditions
3 seriously delinquent and charged- Interaction with indicator of -0.30 -0.21 0.08 0.08
off statuses pandemic conditions
2 seriously delinquent and charged- Interaction with indicator of -0.30 -0.03 0.09 0.09
off statuses pandemic conditions
1 seriously delinquent and charged- Interaction with indicator of -0.30 -0.10 0.10 0.10
off statuses pandemic conditions
long term inactive status Baseline coefficient -4.86 -4.86 0.01 0.01
18 inactive status Baseline coefficient -4.86 -4.46 0.06 0.06
17 inactive status Baseline coefficient -4.86 -4.52 0.06 0.06
16 inactive status Baseline coefficient -4.86 -4.49 0.06 0.06
15 inactive status Baseline coefficient -4.86 -4.47 0.06 0.06
14 inactive status Baseline coefficient -4.86 -4.40 0.05 0.05
13 inactive status Baseline coefficient -4.86 -4.35 0.05 0.05
12 inactive status Baseline coefficient -4.86 -4.30 0.05 0.05
11 inactive status Baseline coefficient -4.86 -4.21 0.05 0.05
10 inactive status Baseline coefficient -4.86 -4.13 0.04 0.04
9 inactive status Baseline coefficient -4.86 -4.08 0.04 0.04
8 inactive status Baseline coefficient -4.86 -4.02 0.04 0.04
7 inactive status Baseline coefficient -4.86 -3.90 0.03 0.03
6 inactive status Baseline coefficient -4.86 -3.80 0.03 0.03
5 inactive status Baseline coefficient -4.86 -3.60 0.03 0.03
4 inactive status Baseline coefficient -4.86 -3.32 0.02 0.02
3 inactive status Baseline coefficient -4.86 -2.87 0.02 0.02
2 inactive status Baseline coefficient -4.86 -2.65 0.02 0.02
1 inactive status Baseline coefficient -4.86 -4.49 0.03 0.03
long term inactive status Interaction with indicator of 0.22 0.22 0.02 0.02
pandemic conditions
18 inactive status Interaction with indicator of 0.22 0.25 0.12 0.12
pandemic conditions
17 inactive status Interaction with indicator of 0.22 0.31 0.12 0.12
pandemic conditions
16 inactive status Interaction with indicator of 0.22 0.28 0.11 0.11
pandemic conditions
Page 87 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
15 inactive status Interaction with indicator of 0.22 0.26 0.11 0.11
pandemic conditions
14 inactive status Interaction with indicator of 0.22 0.25 0.11 0.11
pandemic conditions
13 inactive status Interaction with indicator of 0.22 0.21 0.10 0.10
pandemic conditions
12 inactive status Interaction with indicator of 0.22 0.20 0.10 0.10
pandemic conditions
11 inactive status Interaction with indicator of 0.22 0.26 0.09 0.09
pandemic conditions
10 inactive status Interaction with indicator of 0.22 0.20 0.09 0.09
pandemic conditions
9 inactive status Interaction with indicator of 0.22 0.26 0.08 0.08
pandemic conditions
8 inactive status Interaction with indicator of 0.22 0.26 0.08 0.08
pandemic conditions
7 inactive status Interaction with indicator of 0.22 0.16 0.07 0.07
pandemic conditions
6 inactive status Interaction with indicator of 0.22 0.18 0.07 0.07
pandemic conditions
5 inactive status Interaction with indicator of 0.22 0.22 0.06 0.06
pandemic conditions
4 inactive status Interaction with indicator of 0.22 0.16 0.05 0.05
pandemic conditions
3 inactive status Interaction with indicator of 0.22 0.15 0.04 0.04
pandemic conditions
2 inactive status Interaction with indicator of 0.22 0.12 0.04 0.04
pandemic conditions
1 inactive status Interaction with indicator of 0.22 -1.47 0.11 0.11
pandemic conditions
long term transactor status Baseline coefficient -4.44 -4.44 0.01 0.01
18 transactor status Baseline coefficient -4.44 -3.81 0.05 0.05
17 transactor status Baseline coefficient -4.44 -3.81 0.04 0.04
16 transactor status Baseline coefficient -4.44 -3.79 0.04 0.04
15 transactor status Baseline coefficient -4.44 -3.76 0.04 0.04
14 transactor status Baseline coefficient -4.44 -3.73 0.04 0.04
13 transactor status Baseline coefficient -4.44 -3.67 0.04 0.04
12 transactor status Baseline coefficient -4.44 -3.56 0.03 0.03
11 transactor status Baseline coefficient -4.44 -3.32 0.03 0.03
10 transactor status Baseline coefficient -4.44 -3.26 0.03 0.03
9 transactor status Baseline coefficient -4.44 -3.19 0.03 0.03
Page 88 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
8 transactor status Baseline coefficient -4.44 -3.13 0.03 0.03
7 transactor status Baseline coefficient -4.44 -3.00 0.02 0.02
6 transactor status Baseline coefficient -4.44 -2.84 0.02 0.02
5 transactor status Baseline coefficient -4.44 -2.66 0.02 0.02
4 transactor status Baseline coefficient -4.44 -2.35 0.02 0.02
3 transactor status Baseline coefficient -4.44 -2.06 0.01 0.01
2 transactor status Baseline coefficient -4.44 -1.29 0.01 0.01
1 transactor status Baseline coefficient -4.44 -1.61 0.01 0.01
long term transactor status Interaction with indicator of 0.40 0.40 0.02 0.02
pandemic conditions
18 transactor status Interaction with indicator of 0.40 0.28 0.09 0.09
pandemic conditions
17 transactor status Interaction with indicator of 0.40 0.34 0.08 0.08
pandemic conditions
16 transactor status Interaction with indicator of 0.40 0.35 0.08 0.08
pandemic conditions
15 transactor status Interaction with indicator of 0.40 0.34 0.08 0.08
pandemic conditions
14 transactor status Interaction with indicator of 0.40 0.44 0.07 0.07
pandemic conditions
13 transactor status Interaction with indicator of 0.40 0.43 0.07 0.07
pandemic conditions
12 transactor status Interaction with indicator of 0.40 0.38 0.06 0.06
pandemic conditions
11 transactor status Interaction with indicator of 0.40 0.26 0.06 0.06
pandemic conditions
10 transactor status Interaction with indicator of 0.40 0.25 0.06 0.06
pandemic conditions
9 transactor status Interaction with indicator of 0.40 0.30 0.05 0.05
pandemic conditions
8 transactor status Interaction with indicator of 0.40 0.39 0.05 0.05
pandemic conditions
7 transactor status Interaction with indicator of 0.40 0.34 0.04 0.04
pandemic conditions
6 transactor status Interaction with indicator of 0.40 0.29 0.04 0.04
pandemic conditions
5 transactor status Interaction with indicator of 0.40 0.29 0.04 0.04
pandemic conditions
4 transactor status Interaction with indicator of 0.40 0.26 0.03 0.03
pandemic conditions
3 transactor status Interaction with indicator of 0.40 0.21 0.03 0.03
pandemic conditions
Page 89 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
2 transactor status Interaction with indicator of 0.40 0.19 0.02 0.02
pandemic conditions
1 transactor status Interaction with indicator of 0.40 0.11 0.02 0.02
pandemic conditions
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Baseline combined coefficient estimates are the sum of the constant term and each time-in-
status coefficient for each of the 1–18 time-in-status rows. The long-term row is the constant of the
regression. Interaction combined coefficient estimates are the sum of the overall coefficient of the
interacted variable and each time-in-status interacted coefficient for each of the 1–18 time-in-status
rows. The long-term row is the overall coefficient of the interacted variable. The error bars show the
95 percent confidence intervals for time-in-status. In all cases, the confidence intervals in the time-in-
status rows correspond with the confidence intervals of the time-in-status baseline coefficient or
interaction term. The long-term rows illustrate the confidence intervals for the constant or overall
coefficient of the interacted variable. As a consequence, each row in the figure contains all the
necessary coefficient information for estimating the probability of transition for an account in a
revolving episode in the given time-in-status, given values of the interaction term(s).
The Markov chain analysis produced a discrete probability distribution of
a hypothetical cohort of revolving episodes that started to revolve at the
same time. For reporting purposes, we focused on the distribution of
revolving episode durations as a statistic that illustrated the length of time
we would expect accounts to spend in a revolving episode. To summarize
the distribution, we selected the first month in which the percentage of the
revolving episode cohort that remained in revolving status was strictly
less than 0.5, 0.75, and 0.9. This allowed us to identify the month in which
the 50th, 75th and 90th percentiles of the cohort had left revolving status.
We also calculated the cumulative distribution of status destinations by
the time 90 percent of the cohort had exited revolving status.
Methodology of Analysis of Revolving Episodes
by Race/Ethnicity in Cardholders’ Zip Codes
Methodology. To analyze revolving episode durations and transitions for
cardholders in zip codes with different racial and ethnic composition, we
used the same multinomial logit and Markov chain analysis approaches.
We expanded on our baseline multinomial logit models and interacted
them with the racial and ethnic composition of the cardholder’s zip code.
To more efficiently implement the expanded model’s computations, we
separately estimated the model parameters, first using all revolving
episodes for the period from June 2013 through March 2020 and then
using all revolving episodes for the period from April 2020 through
December 2021. Specifically, we estimated the parameters of the
following equation using observations of accounts in revolving status in
Page 90 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
each period:
In this specification, represent the
percentages of Black or African American residents, Hispanic or Latino
residents, Asian residents, and residents of Other race or ethnicity in the
cardholder’s zip code. The percentage of White residents in the zip code
serves as the reference category. We treated the time-in-status variables
as we did in the baseline models.
To make our results concrete and to effectively interpret our estimated
coefficients, we used our five zip-code clusters by racial and ethnic
composition as scenarios. We plugged in the cluster amounts for
( ) and calculated the conditional
probabilities for each of the five clusters. See appendix III for a discussion
of how we derived the five illustrative zip code clusters.
We then used the same process with a Markov transition matrix to
transform the conditional probabilities to distributions of revolving episode
durations and transitions, where each cluster scenario had a separate set
of values for M(t).
Assumptions and limitations. In addition to the limitations of our
baseline models discussed earlier, our expanded models have the
following additional limitation and should be interpreted with caution. Our
analysis does not on its own offer an explanation of what might be driving
our results, and we could not determine all the potential reasons for racial
and ethnic differences in credit card revolving patterns, which may be due
to factors that are either not captured in the data we analyzed or cannot
be measured. Such factors can include cardholders’ liquidity needs or
personal assessments of the characteristics of credit cards compared to
those of other payment methods.6 Differences in these factors may be
incidentally associated with race and ethnicity, or these factors
6For example, see Irina A. Telyukova, “Household Need for Liquidity and the Credit Card
Debt Puzzle,” Review of Economic Studies, vol. 80 (2013): 1148–1177, and Claire Greene
and Joanna Stavins, “Credit Card Debt Puzzle: Liquid Assets to Pay Bills” (Federal
Reserve Bank of Boston Working Papers, no. 22-8, June 2022), accessed Feb. 23, 2023,
https://www.bostonfed.org/publications/research-department-working-paper/2022/credit-
card-debt-puzzle-liquid-assets-to-pay-bills.
Page 91 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
themselves may be influenced by race and ethnicity, and therefore act as
channels through which race and ethnicity influence revolving patterns.
Results. Figure 18 presents the estimated coefficients from the
multinomial logit model of credit card accounts in revolving status as a
function of the time spent in the current revolving episode and the racial
and ethnic composition of the cardholder’s billing zip code, for each of the
four transition outcomes modeled. In all cases, a negative value of a
baseline coefficient indicates that the outcome is less likely than that of
the reference category—remaining in revolver status. A positive value of
an interaction coefficient indicates that the outcome is more likely than the
baseline coefficient suggests, but not necessarily more likely than the
reference category outcome.
Page 92 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Figure 18: Multinomial Logit Combined Coefficients for Transitions from Revolving Episode to Next Status, by Race and
Ethnicity in Cardholders’ Billing Zip Codes
Page 93 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
Data for Figure 18: Multinomial Logit Combined Coefficients for Transitions from Revolving Episode to Next Status, by Race
and Ethnicity in Cardholders’ Billing Zip Codes
Month Next status Coefficient for category Group Combined Combined Combined
constant coef lci uci
long closed status Baseline coefficient (includes the percentage -6.30 -6.30 0.05 0.05
term of White residents in cardholders' billing ZIP
codes as a reference category)
18 closed status Baseline coefficient (includes the percentage -6.30 -6.16 0.28 0.28
of White residents in cardholders' billing ZIP
codes as a reference category)
17 closed status Baseline coefficient (includes the percentage -6.30 -6.04 0.27 0.27
of White residents in cardholders' billing ZIP
codes as a reference category)
16 closed status Baseline coefficient (includes the percentage -6.30 -5.95 0.26 0.26
of White residents in cardholders' billing ZIP
codes as a reference category)
15 closed status Baseline coefficient (includes the percentage -6.30 -6.12 0.26 0.26
of White residents in cardholders' billing ZIP
codes as a reference category)
14 closed status Baseline coefficient (includes the percentage -6.30 -6.13 0.27 0.27
of White residents in cardholders' billing ZIP
codes as a reference category)
13 closed status Baseline coefficient (includes the percentage -6.30 -6.37 0.25 0.25
of White residents in cardholders' billing ZIP
codes as a reference category)
12 closed status Baseline coefficient (includes the percentage -6.30 -6.18 0.24 0.24
of White residents in cardholders' billing ZIP
codes as a reference category)
11 closed status Baseline coefficient (includes the percentage -6.30 -6.22 0.24 0.24
of White residents in cardholders' billing ZIP
codes as a reference category)
10 closed status Baseline coefficient (includes the percentage -6.30 -6.08 0.24 0.24
of White residents in cardholders' billing ZIP
codes as a reference category)
9 closed status Baseline coefficient (includes the percentage -6.30 -6.13 0.20 0.20
of White residents in cardholders' billing ZIP
codes as a reference category)
8 closed status Baseline coefficient (includes the percentage -6.30 -6.09 0.20 0.20
of White residents in cardholders' billing ZIP
codes as a reference category)
7 closed status Baseline coefficient (includes the percentage -6.30 -6.08 0.20 0.20
of White residents in cardholders' billing ZIP
codes as a reference category)
6 closed status Baseline coefficient (includes the percentage -6.30 -6.23 0.19 0.19
of White residents in cardholders' billing ZIP
codes as a reference category)
Page 94 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
5 closed status Baseline coefficient (includes the percentage -6.30 -6.11 0.18 0.18
of White residents in cardholders' billing ZIP
codes as a reference category)
4 closed status Baseline coefficient (includes the percentage -6.30 -6.07 0.16 0.16
of White residents in cardholders' billing ZIP
codes as a reference category)
3 closed status Baseline coefficient (includes the percentage -6.30 -5.88 0.14 0.14
of White residents in cardholders' billing ZIP
codes as a reference category)
2 closed status Baseline coefficient (includes the percentage -6.30 -5.50 0.11 0.11
of White residents in cardholders' billing ZIP
codes as a reference category)
1 closed status Baseline coefficient (includes the percentage -6.30 -5.25 0.09 0.09
of White residents in cardholders' billing ZIP
codes as a reference category)
long closed status Interaction with percentage of Asian residents -0.76 -0.76 0.37 0.37
term in cardholders' billing ZIP codes
18 closed status Interaction with percentage of Asian residents -0.76 -2.18 2.37 2.37
in cardholders' billing ZIP codes
17 closed status Interaction with percentage of Asian residents -0.76 -1.42 2.15 2.15
in cardholders' billing ZIP codes
16 closed status Interaction with percentage of Asian residents -0.76 -0.30 1.73 1.73
in cardholders' billing ZIP codes
15 closed status Interaction with percentage of Asian residents -0.76 -1.03 1.94 1.94
in cardholders' billing ZIP codes
14 closed status Interaction with percentage of Asian residents -0.76 -2.00 2.18 2.18
in cardholders' billing ZIP codes
13 closed status Interaction with percentage of Asian residents -0.76 -0.28 1.71 1.71
in cardholders' billing ZIP codes
12 closed status Interaction with percentage of Asian residents -0.76 -2.56 2.06 2.06
in cardholders' billing ZIP codes
11 closed status Interaction with percentage of Asian residents -0.76 -1.16 1.76 1.76
in cardholders' billing ZIP codes
10 closed status Interaction with percentage of Asian residents -0.76 -0.67 1.60 1.60
in cardholders' billing ZIP codes
9 closed status Interaction with percentage of Asian residents -0.76 -1.05 1.48 1.48
in cardholders' billing ZIP codes
8 closed status Interaction with percentage of Asian residents -0.76 -0.98 1.41 1.41
in cardholders' billing ZIP codes
7 closed status Interaction with percentage of Asian residents -0.76 -0.37 1.35 1.35
in cardholders' billing ZIP codes
6 closed status Interaction with percentage of Asian residents -0.76 -0.55 1.30 1.30
in cardholders' billing ZIP codes
5 closed status Interaction with percentage of Asian residents -0.76 -0.08 1.14 1.14
in cardholders' billing ZIP codes
Page 95 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
4 closed status Interaction with percentage of Asian residents -0.76 -0.34 1.05 1.05
in cardholders' billing ZIP codes
3 closed status Interaction with percentage of Asian residents -0.76 -0.33 0.95 0.95
in cardholders' billing ZIP codes
2 closed status Interaction with percentage of Asian residents -0.76 0.16 0.67 0.67
in cardholders' billing ZIP codes
1 closed status Interaction with percentage of Asian residents -0.76 0.20 0.57 0.57
in cardholders' billing ZIP codes
long closed status Interaction with percentage of Black or African -0.36 -0.36 0.17 0.17
term American residents in cardholders' billing ZIP
codes
18 closed status Interaction with percentage of Black or African -0.36 -0.05 0.89 0.89
American residents in cardholders' billing ZIP
codes
17 closed status Interaction with percentage of Black or African -0.36 -1.51 1.20 1.20
American residents in cardholders' billing ZIP
codes
16 closed status Interaction with percentage of Black or African -0.36 -1.70 1.16 1.16
American residents in cardholders' billing ZIP
codes
15 closed status Interaction with percentage of Black or African -0.36 -0.42 0.89 0.89
American residents in cardholders' billing ZIP
codes
14 closed status Interaction with percentage of Black or African -0.36 -0.57 0.91 0.91
American residents in cardholders' billing ZIP
codes
13 closed status Interaction with percentage of Black or African -0.36 0.66 0.70 0.70
American residents in cardholders' billing ZIP
codes
12 closed status Interaction with percentage of Black or African -0.36 -1.30 1.00 1.00
American residents in cardholders' billing ZIP
codes
11 closed status Interaction with percentage of Black or African -0.36 -0.53 0.84 0.84
American residents in cardholders' billing ZIP
codes
10 closed status Interaction with percentage of Black or African -0.36 -0.82 0.83 0.83
American residents in cardholders' billing ZIP
codes
9 closed status Interaction with percentage of Black or African -0.36 0.02 0.65 0.65
American residents in cardholders' billing ZIP
codes
8 closed status Interaction with percentage of Black or African -0.36 -2.05 0.96 0.96
American residents in cardholders' billing ZIP
codes
7 closed status Interaction with percentage of Black or African -0.36 -0.57 0.70 0.70
American residents in cardholders' billing ZIP
codes
Page 96 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
6 closed status Interaction with percentage of Black or African -0.36 -0.28 0.65 0.65
American residents in cardholders' billing ZIP
codes
5 closed status Interaction with percentage of Black or African -0.36 -0.97 0.69 0.69
American residents in cardholders' billing ZIP
codes
4 closed status Interaction with percentage of Black or African -0.36 -0.81 0.61 0.61
American residents in cardholders' billing ZIP
codes
3 closed status Interaction with percentage of Black or African -0.36 -0.63 0.52 0.52
American residents in cardholders' billing ZIP
codes
2 closed status Interaction with percentage of Black or African -0.36 -0.41 0.38 0.38
American residents in cardholders' billing ZIP
codes
1 closed status Interaction with percentage of Black or African -0.36 -0.69 0.34 0.34
American residents in cardholders' billing ZIP
codes
long closed status Interaction with percentage of Hispanic or -0.30 -0.30 0.15 0.15
term Latino residents in cardholders' billing ZIP
codes
18 closed status Interaction with percentage of Hispanic or -0.30 -0.14 0.75 0.75
Latino residents in cardholders' billing ZIP
codes
17 closed status Interaction with percentage of Hispanic or -0.30 -0.90 0.87 0.87
Latino residents in cardholders' billing ZIP
codes
16 closed status Interaction with percentage of Hispanic or -0.30 -0.64 0.76 0.76
Latino residents in cardholders' billing ZIP
codes
15 closed status Interaction with percentage of Hispanic or -0.30 -0.52 0.75 0.75
Latino residents in cardholders' billing ZIP
codes
14 closed status Interaction with percentage of Hispanic or -0.30 -0.15 0.68 0.68
Latino residents in cardholders' billing ZIP
codes
13 closed status Interaction with percentage of Hispanic or -0.30 -0.54 0.74 0.74
Latino residents in cardholders' billing ZIP
codes
12 closed status Interaction with percentage of Hispanic or -0.30 0.20 0.61 0.61
Latino residents in cardholders' billing ZIP
codes
11 closed status Interaction with percentage of Hispanic or -0.30 -0.26 0.65 0.65
Latino residents in cardholders' billing ZIP
codes
10 closed status Interaction with percentage of Hispanic or -0.30 -0.01 0.57 0.57
Latino residents in cardholders' billing ZIP
codes
Page 97 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
9 closed status Interaction with percentage of Hispanic or -0.30 -0.40 0.58 0.58
Latino residents in cardholders' billing ZIP
codes
8 closed status Interaction with percentage of Hispanic or -0.30 0.33 0.49 0.49
Latino residents in cardholders' billing ZIP
codes
7 closed status Interaction with percentage of Hispanic or -0.30 -0.66 0.58 0.58
Latino residents in cardholders' billing ZIP
codes
6 closed status Interaction with percentage of Hispanic or -0.30 -0.31 0.53 0.53
Latino residents in cardholders' billing ZIP
codes
5 closed status Interaction with percentage of Hispanic or -0.30 -0.25 0.48 0.48
Latino residents in cardholders' billing ZIP
codes
4 closed status Interaction with percentage of Hispanic or -0.30 -0.29 0.45 0.45
Latino residents in cardholders' billing ZIP
codes
3 closed status Interaction with percentage of Hispanic or -0.30 -0.36 0.40 0.40
Latino residents in cardholders' billing ZIP
codes
2 closed status Interaction with percentage of Hispanic or -0.30 -0.27 0.30 0.30
Latino residents in cardholders' billing ZIP
codes
1 closed status Interaction with percentage of Hispanic or -0.30 -0.47 0.26 0.26
Latino residents in cardholders' billing ZIP
codes
long seriously delinquent Baseline coefficient (includes the percentage -5.25 -5.25 0.03 0.03
term and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
18 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.76 0.14 0.14
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
17 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.93 0.13 0.13
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
16 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.82 0.13 0.13
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
15 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.83 0.12 0.12
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
14 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.87 0.13 0.13
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
13 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.87 0.11 0.11
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
Page 98 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
12 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.80 0.11 0.11
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
11 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.93 0.11 0.11
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
10 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.75 0.11 0.11
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
9 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.83 0.10 0.10
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
8 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.79 0.10 0.10
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
7 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.79 0.09 0.09
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
6 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.78 0.09 0.09
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
5 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.77 0.08 0.08
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
4 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.77 0.08 0.08
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
3 seriously delinquent Baseline coefficient (includes the percentage -5.25 -4.48 0.06 0.06
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
2 seriously delinquent Baseline coefficient (includes the percentage -5.25 -5.13 0.08 0.08
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
1 seriously delinquent Baseline coefficient (includes the percentage -5.25 -5.62 0.09 0.09
and charged-off of White residents in cardholders' billing ZIP
statuses codes as a reference category)
long seriously delinquent Interaction with percentage of Asian residents -0.28 -0.28 0.18 0.18
term and charged-off in cardholders' billing ZIP codes
statuses
18 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.72 0.94 0.94
and charged-off in cardholders' billing ZIP codes
statuses
17 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.82 0.92 0.92
and charged-off in cardholders' billing ZIP codes
statuses
16 seriously delinquent Interaction with percentage of Asian residents -0.28 -1.06 0.93 0.93
and charged-off in cardholders' billing ZIP codes
statuses
Page 99 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
15 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.93 0.84 0.84
and charged-off in cardholders' billing ZIP codes
statuses
14 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.59 0.84 0.84
and charged-off in cardholders' billing ZIP codes
statuses
13 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.42 0.75 0.75
and charged-off in cardholders' billing ZIP codes
statuses
12 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.86 0.78 0.78
and charged-off in cardholders' billing ZIP codes
statuses
11 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.42 0.72 0.72
and charged-off in cardholders' billing ZIP codes
statuses
10 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.04 0.66 0.66
and charged-off in cardholders' billing ZIP codes
statuses
9 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.52 0.66 0.66
and charged-off in cardholders' billing ZIP codes
statuses
8 seriously delinquent Interaction with percentage of Asian residents -0.28 -1.15 0.69 0.69
and charged-off in cardholders' billing ZIP codes
statuses
7 seriously delinquent Interaction with percentage of Asian residents -0.28 -1.26 0.67 0.67
and charged-off in cardholders' billing ZIP codes
statuses
6 seriously delinquent Interaction with percentage of Asian residents -0.28 -1.33 0.66 0.66
and charged-off in cardholders' billing ZIP codes
statuses
5 seriously delinquent Interaction with percentage of Asian residents -0.28 -1.07 0.58 0.58
and charged-off in cardholders' billing ZIP codes
statuses
4 seriously delinquent Interaction with percentage of Asian residents -0.28 -1.18 0.56 0.56
and charged-off in cardholders' billing ZIP codes
statuses
3 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.63 0.43 0.43
and charged-off in cardholders' billing ZIP codes
statuses
2 seriously delinquent Interaction with percentage of Asian residents -0.28 0.72 0.44 0.44
and charged-off in cardholders' billing ZIP codes
statuses
1 seriously delinquent Interaction with percentage of Asian residents -0.28 -0.03 0.52 0.52
and charged-off in cardholders' billing ZIP codes
statuses
long seriously delinquent Interaction with percentage of Black or African 0.41 0.41 0.08 0.08
term and charged-off American residents in cardholders' billing ZIP
statuses codes
Page 100 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
18 seriously delinquent Interaction with percentage of Black or African 0.41 0.44 0.38 0.38
and charged-off American residents in cardholders' billing ZIP
statuses codes
17 seriously delinquent Interaction with percentage of Black or African 0.41 0.82 0.35 0.35
and charged-off American residents in cardholders' billing ZIP
statuses codes
16 seriously delinquent Interaction with percentage of Black or African 0.41 0.63 0.35 0.35
and charged-off American residents in cardholders' billing ZIP
statuses codes
15 seriously delinquent Interaction with percentage of Black or African 0.41 0.52 0.34 0.34
and charged-off American residents in cardholders' billing ZIP
statuses codes
14 seriously delinquent Interaction with percentage of Black or African 0.41 0.66 0.33 0.33
and charged-off American residents in cardholders' billing ZIP
statuses codes
13 seriously delinquent Interaction with percentage of Black or African 0.41 0.44 0.33 0.33
and charged-off American residents in cardholders' billing ZIP
statuses codes
12 seriously delinquent Interaction with percentage of Black or African 0.41 0.60 0.31 0.31
and charged-off American residents in cardholders' billing ZIP
statuses codes
11 seriously delinquent Interaction with percentage of Black or African 0.41 0.70 0.30 0.30
and charged-off American residents in cardholders' billing ZIP
statuses codes
10 seriously delinquent Interaction with percentage of Black or African 0.41 0.56 0.29 0.29
and charged-off American residents in cardholders' billing ZIP
statuses codes
9 seriously delinquent Interaction with percentage of Black or African 0.41 0.85 0.26 0.26
and charged-off American residents in cardholders' billing ZIP
statuses codes
8 seriously delinquent Interaction with percentage of Black or African 0.41 0.61 0.27 0.27
and charged-off American residents in cardholders' billing ZIP
statuses codes
7 seriously delinquent Interaction with percentage of Black or African 0.41 0.72 0.25 0.25
and charged-off American residents in cardholders' billing ZIP
statuses codes
6 seriously delinquent Interaction with percentage of Black or African 0.41 0.69 0.24 0.24
and charged-off American residents in cardholders' billing ZIP
statuses codes
5 seriously delinquent Interaction with percentage of Black or African 0.41 0.85 0.22 0.22
and charged-off American residents in cardholders' billing ZIP
statuses codes
4 seriously delinquent Interaction with percentage of Black or African 0.41 1.11 0.20 0.20
and charged-off American residents in cardholders' billing ZIP
statuses codes
3 seriously delinquent Interaction with percentage of Black or African 0.41 0.94 0.17 0.17
and charged-off American residents in cardholders' billing ZIP
statuses codes
Page 101 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
2 seriously delinquent Interaction with percentage of Black or African 0.41 0.50 0.23 0.23
and charged-off American residents in cardholders' billing ZIP
statuses codes
1 seriously delinquent Interaction with percentage of Black or African 0.41 0.65 0.25 0.25
and charged-off American residents in cardholders' billing ZIP
statuses codes
long seriously delinquent Interaction with percentage of Hispanic or 0.41 0.41 0.07 0.07
term and charged-off Latino residents in cardholders' billing ZIP
statuses codes
18 seriously delinquent Interaction with percentage of Hispanic or 0.41 -0.08 0.35 0.35
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
17 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.47 0.32 0.32
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
16 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.44 0.30 0.30
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
15 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.34 0.30 0.30
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
14 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.41 0.29 0.29
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
13 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.51 0.27 0.27
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
12 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.32 0.27 0.27
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
11 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.41 0.27 0.27
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
10 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.02 0.26 0.26
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
9 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.20 0.25 0.25
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
8 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.43 0.23 0.23
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
7 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.38 0.22 0.22
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
6 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.37 0.21 0.21
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
Page 102 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
5 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.29 0.20 0.20
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
4 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.38 0.19 0.19
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
3 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.35 0.16 0.16
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
2 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.10 0.20 0.20
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
1 seriously delinquent Interaction with percentage of Hispanic or 0.41 0.13 0.22 0.22
and charged-off Latino residents in cardholders' billing ZIP
statuses codes
long inactive status Baseline coefficient (includes the percentage -4.81 -4.81 0.02 0.02
term of White residents in cardholders' billing ZIP
codes as a reference category)
18 inactive status Baseline coefficient (includes the percentage -4.81 -4.33 0.12 0.12
of White residents in cardholders' billing ZIP
codes as a reference category)
17 inactive status Baseline coefficient (includes the percentage -4.81 -4.46 0.12 0.12
of White residents in cardholders' billing ZIP
codes as a reference category)
16 inactive status Baseline coefficient (includes the percentage -4.81 -4.37 0.12 0.12
of White residents in cardholders' billing ZIP
codes as a reference category)
15 inactive status Baseline coefficient (includes the percentage -4.81 -4.44 0.11 0.11
of White residents in cardholders' billing ZIP
codes as a reference category)
14 inactive status Baseline coefficient (includes the percentage -4.81 -4.32 0.10 0.10
of White residents in cardholders' billing ZIP
codes as a reference category)
13 inactive status Baseline coefficient (includes the percentage -4.81 -4.30 0.10 0.10
of White residents in cardholders' billing ZIP
codes as a reference category)
12 inactive status Baseline coefficient (includes the percentage -4.81 -4.21 0.09 0.09
of White residents in cardholders' billing ZIP
codes as a reference category)
11 inactive status Baseline coefficient (includes the percentage -4.81 -4.19 0.09 0.09
of White residents in cardholders' billing ZIP
codes as a reference category)
10 inactive status Baseline coefficient (includes the percentage -4.81 -4.12 0.08 0.08
of White residents in cardholders' billing ZIP
codes as a reference category)
9 inactive status Baseline coefficient (includes the percentage -4.81 -4.08 0.08 0.08
of White residents in cardholders' billing ZIP
codes as a reference category)
Page 103 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
8 inactive status Baseline coefficient (includes the percentage -4.81 -3.99 0.07 0.07
of White residents in cardholders' billing ZIP
codes as a reference category)
7 inactive status Baseline coefficient (includes the percentage -4.81 -3.90 0.06 0.06
of White residents in cardholders' billing ZIP
codes as a reference category)
6 inactive status Baseline coefficient (includes the percentage -4.81 -3.76 0.06 0.06
of White residents in cardholders' billing ZIP
codes as a reference category)
5 inactive status Baseline coefficient (includes the percentage -4.81 -3.58 0.05 0.05
of White residents in cardholders' billing ZIP
codes as a reference category)
4 inactive status Baseline coefficient (includes the percentage -4.81 -3.32 0.05 0.05
of White residents in cardholders' billing ZIP
codes as a reference category)
3 inactive status Baseline coefficient (includes the percentage -4.81 -2.84 0.04 0.04
of White residents in cardholders' billing ZIP
codes as a reference category)
2 inactive status Baseline coefficient (includes the percentage -4.81 -2.58 0.03 0.03
of White residents in cardholders' billing ZIP
codes as a reference category)
1 inactive status Baseline coefficient (includes the percentage -4.81 -4.33 0.05 0.05
of White residents in cardholders' billing ZIP
codes as a reference category)
long inactive status Interaction with percentage of Asian residents 0.06 0.06 0.15 0.15
term in cardholders' billing ZIP codes
18 inactive status Interaction with percentage of Asian residents 0.06 -1.28 0.85 0.85
in cardholders' billing ZIP codes
17 inactive status Interaction with percentage of Asian residents 0.06 -0.58 0.79 0.79
in cardholders' billing ZIP codes
16 inactive status Interaction with percentage of Asian residents 0.06 0.24 0.69 0.69
in cardholders' billing ZIP codes
15 inactive status Interaction with percentage of Asian residents 0.06 0.31 0.65 0.65
in cardholders' billing ZIP codes
14 inactive status Interaction with percentage of Asian residents 0.06 -0.25 0.65 0.65
in cardholders' billing ZIP codes
13 inactive status Interaction with percentage of Asian residents 0.06 0.26 0.58 0.58
in cardholders' billing ZIP codes
12 inactive status Interaction with percentage of Asian residents 0.06 -0.56 0.62 0.62
in cardholders' billing ZIP codes
11 inactive status Interaction with percentage of Asian residents 0.06 -0.30 0.55 0.55
in cardholders' billing ZIP codes
10 inactive status Interaction with percentage of Asian residents 0.06 -0.06 0.50 0.50
in cardholders' billing ZIP codes
9 inactive status Interaction with percentage of Asian residents 0.06 -0.58 0.50 0.50
in cardholders' billing ZIP codes
Page 104 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
8 inactive status Interaction with percentage of Asian residents 0.06 -0.43 0.46 0.46
in cardholders' billing ZIP codes
7 inactive status Interaction with percentage of Asian residents 0.06 0.05 0.40 0.40
in cardholders' billing ZIP codes
6 inactive status Interaction with percentage of Asian residents 0.06 -0.00 0.37 0.37
in cardholders' billing ZIP codes
5 inactive status Interaction with percentage of Asian residents 0.06 0.23 0.32 0.32
in cardholders' billing ZIP codes
4 inactive status Interaction with percentage of Asian residents 0.06 0.31 0.27 0.27
in cardholders' billing ZIP codes
3 inactive status Interaction with percentage of Asian residents 0.06 0.64 0.22 0.22
in cardholders' billing ZIP codes
2 inactive status Interaction with percentage of Asian residents 0.06 0.52 0.20 0.20
in cardholders' billing ZIP codes
1 inactive status Interaction with percentage of Asian residents 0.06 0.56 0.30 0.30
in cardholders' billing ZIP codes
long inactive status Interaction with percentage of Black or African -0.45 -0.45 0.08 0.08
term American residents in cardholders' billing ZIP
codes
18 inactive status Interaction with percentage of Black or African -0.45 -0.79 0.43 0.43
American residents in cardholders' billing ZIP
codes
17 inactive status Interaction with percentage of Black or African -0.45 -0.62 0.42 0.42
American residents in cardholders' billing ZIP
codes
16 inactive status Interaction with percentage of Black or African -0.45 -0.75 0.41 0.41
American residents in cardholders' billing ZIP
codes
15 inactive status Interaction with percentage of Black or African -0.45 -0.11 0.35 0.35
American residents in cardholders' billing ZIP
codes
14 inactive status Interaction with percentage of Black or African -0.45 -0.52 0.35 0.35
American residents in cardholders' billing ZIP
codes
13 inactive status Interaction with percentage of Black or African -0.45 -0.49 0.34 0.34
American residents in cardholders' billing ZIP
codes
12 inactive status Interaction with percentage of Black or African -0.45 -0.22 0.30 0.30
American residents in cardholders' billing ZIP
codes
11 inactive status Interaction with percentage of Black or African -0.45 -0.24 0.28 0.28
American residents in cardholders' billing ZIP
codes
10 inactive status Interaction with percentage of Black or African -0.45 -0.32 0.27 0.27
American residents in cardholders' billing ZIP
codes
Page 105 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
9 inactive status Interaction with percentage of Black or African -0.45 -0.10 0.24 0.24
American residents in cardholders' billing ZIP
codes
8 inactive status Interaction with percentage of Black or African -0.45 -0.38 0.24 0.24
American residents in cardholders' billing ZIP
codes
7 inactive status Interaction with percentage of Black or African -0.45 -0.29 0.22 0.22
American residents in cardholders' billing ZIP
codes
6 inactive status Interaction with percentage of Black or African -0.45 -0.27 0.20 0.20
American residents in cardholders' billing ZIP
codes
5 inactive status Interaction with percentage of Black or African -0.45 -0.17 0.17 0.17
American residents in cardholders' billing ZIP
codes
4 inactive status Interaction with percentage of Black or African -0.45 -0.27 0.15 0.15
American residents in cardholders' billing ZIP
codes
3 inactive status Interaction with percentage of Black or African -0.45 -0.41 0.13 0.13
American residents in cardholders' billing ZIP
codes
2 inactive status Interaction with percentage of Black or African -0.45 -0.37 0.12 0.12
American residents in cardholders' billing ZIP
codes
1 inactive status Interaction with percentage of Black or African -0.45 -0.69 0.20 0.20
American residents in cardholders' billing ZIP
codes
long inactive status Interaction with percentage of Hispanic or 0.07 0.07 0.06 0.06
term Latino residents in cardholders' billing ZIP
codes
18 inactive status Interaction with percentage of Hispanic or 0.07 0.11 0.29 0.29
Latino residents in cardholders' billing ZIP
codes
17 inactive status Interaction with percentage of Hispanic or 0.07 0.23 0.29 0.29
Latino residents in cardholders' billing ZIP
codes
16 inactive status Interaction with percentage of Hispanic or 0.07 0.02 0.28 0.28
Latino residents in cardholders' billing ZIP
codes
15 inactive status Interaction with percentage of Hispanic or 0.07 -0.12 0.28 0.28
Latino residents in cardholders' billing ZIP
codes
14 inactive status Interaction with percentage of Hispanic or 0.07 -0.09 0.26 0.26
Latino residents in cardholders' billing ZIP
codes
13 inactive status Interaction with percentage of Hispanic or 0.07 -0.10 0.25 0.25
Latino residents in cardholders' billing ZIP
codes
Page 106 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
12 inactive status Interaction with percentage of Hispanic or 0.07 -0.18 0.24 0.24
Latino residents in cardholders' billing ZIP
codes
11 inactive status Interaction with percentage of Hispanic or 0.07 0.17 0.21 0.21
Latino residents in cardholders' billing ZIP
codes
10 inactive status Interaction with percentage of Hispanic or 0.07 0.11 0.20 0.20
Latino residents in cardholders' billing ZIP
codes
9 inactive status Interaction with percentage of Hispanic or 0.07 0.21 0.18 0.18
Latino residents in cardholders' billing ZIP
codes
8 inactive status Interaction with percentage of Hispanic or 0.07 0.18 0.17 0.17
Latino residents in cardholders' billing ZIP
codes
7 inactive status Interaction with percentage of Hispanic or 0.07 0.13 0.16 0.16
Latino residents in cardholders' billing ZIP
codes
6 inactive status Interaction with percentage of Hispanic or 0.07 0.12 0.15 0.15
Latino residents in cardholders' billing ZIP
codes
5 inactive status Interaction with percentage of Hispanic or 0.07 0.08 0.13 0.13
Latino residents in cardholders' billing ZIP
codes
4 inactive status Interaction with percentage of Hispanic or 0.07 0.04 0.12 0.12
Latino residents in cardholders' billing ZIP
codes
3 inactive status Interaction with percentage of Hispanic or 0.07 -0.10 0.10 0.10
Latino residents in cardholders' billing ZIP
codes
2 inactive status Interaction with percentage of Hispanic or 0.07 -0.27 0.09 0.09
Latino residents in cardholders' billing ZIP
codes
1 inactive status Interaction with percentage of Hispanic or 0.07 -0.49 0.15 0.15
Latino residents in cardholders' billing ZIP
codes
long transactor status Baseline coefficient (includes the percentage -4.38 -4.38 0.02 0.02
term of White residents in cardholders' billing ZIP
codes as a reference category)
18 transactor status Baseline coefficient (includes the percentage -4.38 -3.74 0.09 0.09
of White residents in cardholders' billing ZIP
codes as a reference category)
17 transactor status Baseline coefficient (includes the percentage -4.38 -3.74 0.08 0.08
of White residents in cardholders' billing ZIP
codes as a reference category)
16 transactor status Baseline coefficient (includes the percentage -4.38 -3.77 0.08 0.08
of White residents in cardholders' billing ZIP
codes as a reference category)
Page 107 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
15 transactor status Baseline coefficient (includes the percentage -4.38 -3.67 0.08 0.08
of White residents in cardholders' billing ZIP
codes as a reference category)
14 transactor status Baseline coefficient (includes the percentage -4.38 -3.63 0.07 0.07
of White residents in cardholders' billing ZIP
codes as a reference category)
13 transactor status Baseline coefficient (includes the percentage -4.38 -3.56 0.07 0.07
of White residents in cardholders' billing ZIP
codes as a reference category)
12 transactor status Baseline coefficient (includes the percentage -4.38 -3.46 0.06 0.06
of White residents in cardholders' billing ZIP
codes as a reference category)
11 transactor status Baseline coefficient (includes the percentage -4.38 -3.21 0.06 0.06
of White residents in cardholders' billing ZIP
codes as a reference category)
10 transactor status Baseline coefficient (includes the percentage -4.38 -3.18 0.05 0.05
of White residents in cardholders' billing ZIP
codes as a reference category)
9 transactor status Baseline coefficient (includes the percentage -4.38 -3.13 0.05 0.05
of White residents in cardholders' billing ZIP
codes as a reference category)
8 transactor status Baseline coefficient (includes the percentage -4.38 -3.03 0.05 0.05
of White residents in cardholders' billing ZIP
codes as a reference category)
7 transactor status Baseline coefficient (includes the percentage -4.38 -2.95 0.04 0.04
of White residents in cardholders' billing ZIP
codes as a reference category)
6 transactor status Baseline coefficient (includes the percentage -4.38 -2.77 0.04 0.04
of White residents in cardholders' billing ZIP
codes as a reference category)
5 transactor status Baseline coefficient (includes the percentage -4.38 -2.56 0.03 0.03
of White residents in cardholders' billing ZIP
codes as a reference category)
4 transactor status Baseline coefficient (includes the percentage -4.38 -2.26 0.03 0.03
of White residents in cardholders' billing ZIP
codes as a reference category)
3 transactor status Baseline coefficient (includes the percentage -4.38 -1.95 0.03 0.03
of White residents in cardholders' billing ZIP
codes as a reference category)
2 transactor status Baseline coefficient (includes the percentage -4.38 -1.15 0.02 0.02
of White residents in cardholders' billing ZIP
codes as a reference category)
1 transactor status Baseline coefficient (includes the percentage -4.38 -1.45 0.02 0.02
of White residents in cardholders' billing ZIP
codes as a reference category)
long transactor status Interaction with percentage of Asian residents 0.56 0.56 0.12 0.12
term in cardholders' billing ZIP codes
Page 108 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
18 transactor status Interaction with percentage of Asian residents 0.56 0.63 0.50 0.50
in cardholders' billing ZIP codes
17 transactor status Interaction with percentage of Asian residents 0.56 0.33 0.49 0.49
in cardholders' billing ZIP codes
16 transactor status Interaction with percentage of Asian residents 0.56 0.31 0.48 0.48
in cardholders' billing ZIP codes
15 transactor status Interaction with percentage of Asian residents 0.56 0.38 0.46 0.46
in cardholders' billing ZIP codes
14 transactor status Interaction with percentage of Asian residents 0.56 0.41 0.43 0.43
in cardholders' billing ZIP codes
13 transactor status Interaction with percentage of Asian residents 0.56 0.37 0.41 0.41
in cardholders' billing ZIP codes
12 transactor status Interaction with percentage of Asian residents 0.56 0.57 0.37 0.37
in cardholders' billing ZIP codes
11 transactor status Interaction with percentage of Asian residents 0.56 0.47 0.33 0.33
in cardholders' billing ZIP codes
10 transactor status Interaction with percentage of Asian residents 0.56 0.74 0.30 0.30
in cardholders' billing ZIP codes
9 transactor status Interaction with percentage of Asian residents 0.56 0.60 0.29 0.29
in cardholders' billing ZIP codes
8 transactor status Interaction with percentage of Asian residents 0.56 0.46 0.27 0.27
in cardholders' billing ZIP codes
7 transactor status Interaction with percentage of Asian residents 0.56 0.85 0.24 0.24
in cardholders' billing ZIP codes
6 transactor status Interaction with percentage of Asian residents 0.56 0.79 0.22 0.22
in cardholders' billing ZIP codes
5 transactor status Interaction with percentage of Asian residents 0.56 0.77 0.20 0.20
in cardholders' billing ZIP codes
4 transactor status Interaction with percentage of Asian residents 0.56 1.00 0.18 0.18
in cardholders' billing ZIP codes
3 transactor status Interaction with percentage of Asian residents 0.56 1.01 0.16 0.16
in cardholders' billing ZIP codes
2 transactor status Interaction with percentage of Asian residents 0.56 1.33 0.13 0.13
in cardholders' billing ZIP codes
1 transactor status Interaction with percentage of Asian residents 0.56 0.77 0.13 0.13
in cardholders' billing ZIP codes
long transactor status Interaction with percentage of Black or African -0.43 -0.43 0.07 0.07
term American residents in cardholders' billing ZIP
codes
18 transactor status Interaction with percentage of Black or African -0.43 -0.30 0.29 0.29
American residents in cardholders' billing ZIP
codes
17 transactor status Interaction with percentage of Black or African -0.43 -0.47 0.29 0.29
American residents in cardholders' billing ZIP
codes
Page 109 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
16 transactor status Interaction with percentage of Black or African -0.43 -0.34 0.27 0.27
American residents in cardholders' billing ZIP
codes
15 transactor status Interaction with percentage of Black or African -0.43 -0.31 0.26 0.26
American residents in cardholders' billing ZIP
codes
14 transactor status Interaction with percentage of Black or African -0.43 -0.53 0.26 0.26
American residents in cardholders' billing ZIP
codes
13 transactor status Interaction with percentage of Black or African -0.43 -0.51 0.24 0.24
American residents in cardholders' billing ZIP
codes
12 transactor status Interaction with percentage of Black or African -0.43 -0.45 0.22 0.22
American residents in cardholders' billing ZIP
codes
11 transactor status Interaction with percentage of Black or African -0.43 -0.58 0.20 0.20
American residents in cardholders' billing ZIP
codes
10 transactor status Interaction with percentage of Black or African -0.43 -0.56 0.19 0.19
American residents in cardholders' billing ZIP
codes
9 transactor status Interaction with percentage of Black or African -0.43 -0.47 0.17 0.17
American residents in cardholders' billing ZIP
codes
8 transactor status Interaction with percentage of Black or African -0.43 -0.54 0.16 0.16
American residents in cardholders' billing ZIP
codes
7 transactor status Interaction with percentage of Black or African -0.43 -0.36 0.15 0.15
American residents in cardholders' billing ZIP
codes
6 transactor status Interaction with percentage of Black or African -0.43 -0.41 0.13 0.13
American residents in cardholders' billing ZIP
codes
5 transactor status Interaction with percentage of Black or African -0.43 -0.59 0.12 0.12
American residents in cardholders' billing ZIP
codes
4 transactor status Interaction with percentage of Black or African -0.43 -0.49 0.11 0.11
American residents in cardholders' billing ZIP
codes
3 transactor status Interaction with percentage of Black or African -0.43 -0.65 0.10 0.10
American residents in cardholders' billing ZIP
codes
2 transactor status Interaction with percentage of Black or African -0.43 -0.81 0.08 0.08
American residents in cardholders' billing ZIP
codes
1 transactor status Interaction with percentage of Black or African -0.43 -0.75 0.08 0.08
American residents in cardholders' billing ZIP
codes
Page 110 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
long transactor status Interaction with percentage of Hispanic or -0.17 -0.17 0.05 0.05
term Latino residents in cardholders' billing ZIP
codes
18 transactor status Interaction with percentage of Hispanic or -0.17 -0.22 0.23 0.23
Latino residents in cardholders' billing ZIP
codes
17 transactor status Interaction with percentage of Hispanic or -0.17 -0.28 0.22 0.22
Latino residents in cardholders' billing ZIP
codes
16 transactor status Interaction with percentage of Hispanic or -0.17 -0.09 0.21 0.21
Latino residents in cardholders' billing ZIP
codes
15 transactor status Interaction with percentage of Hispanic or -0.17 -0.25 0.20 0.20
Latino residents in cardholders' billing ZIP
codes
14 transactor status Interaction with percentage of Hispanic or -0.17 -0.34 0.20 0.20
Latino residents in cardholders' billing ZIP
codes
13 transactor status Interaction with percentage of Hispanic or -0.17 -0.37 0.19 0.19
Latino residents in cardholders' billing ZIP
codes
12 transactor status Interaction with percentage of Hispanic or -0.17 -0.41 0.18 0.18
Latino residents in cardholders' billing ZIP
codes
11 transactor status Interaction with percentage of Hispanic or -0.17 -0.34 0.15 0.15
Latino residents in cardholders' billing ZIP
codes
10 transactor status Interaction with percentage of Hispanic or -0.17 -0.25 0.14 0.14
Latino residents in cardholders' billing ZIP
codes
9 transactor status Interaction with percentage of Hispanic or -0.17 -0.23 0.13 0.13
Latino residents in cardholders' billing ZIP
codes
8 transactor status Interaction with percentage of Hispanic or -0.17 -0.32 0.13 0.13
Latino residents in cardholders' billing ZIP
codes
7 transactor status Interaction with percentage of Hispanic or -0.17 -0.21 0.11 0.11
Latino residents in cardholders' billing ZIP
codes
6 transactor status Interaction with percentage of Hispanic or -0.17 -0.35 0.11 0.11
Latino residents in cardholders' billing ZIP
codes
5 transactor status Interaction with percentage of Hispanic or -0.17 -0.37 0.10 0.10
Latino residents in cardholders' billing ZIP
codes
4 transactor status Interaction with percentage of Hispanic or -0.17 -0.44 0.09 0.09
Latino residents in cardholders' billing ZIP
codes
Page 111 GAO-23-105269 Credit Cards
Appendix IV: Econometric Analysis of
Revolving and Transacting Episode Durations
3 transactor status Interaction with percentage of Hispanic or -0.17 -0.44 0.08 0.08
Latino residents in cardholders' billing ZIP
codes
2 transactor status Interaction with percentage of Hispanic or -0.17 -0.76 0.07 0.07
Latino residents in cardholders' billing ZIP
codes
1 transactor status Interaction with percentage of Hispanic or -0.17 -0.67 0.06 0.06
Latino residents in cardholders' billing ZIP
codes
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
Note: Baseline combined coefficient estimates are the sum of the constant term and each time-in-
status coefficient for each of the 1–18 time-in-status rows. The long-term row is the constant of the
regression. Interaction combined coefficient estimates are the sum of the overall coefficient of the
interacted variable and each time-in-status interacted coefficient for each of the 1–18 time-in-status
rows. The long-term row is the overall coefficient of the interacted variable. The error bars show the
95 percent confidence intervals for time-in-status. In all cases, the confidence intervals in the time-in-
status rows correspond with the confidence intervals of the time-in-status baseline coefficient or
interaction term. The long-term rows illustrate the confidence intervals for the constant or overall
coefficient of the interacted variable. As a consequence, each row in the graphic contains all the
necessary coefficient information for estimating the probability of transition for an account in a
revolving episode in the given time-in-status, given values of the interaction term(s). We omit
reporting the results for the Other racial group category because the group is small and not
necessarily cohesive.
Page 112 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
Appendix V: Econometric
Analysis of Credit Card Payments
This appendix presents the development, estimation, results, and
limitations of our econometric analysis of credit cardholders’ payment
amounts toward revolving balances before and during the COVID-19
pandemic.
Generally, cardholders can decide to pay any amount between zero and
their total balance, though if they pay less than their minimum payment
amount, they will enter delinquency and be subject to additional fees.
Each cardholder’s decisions are driven by a multitude of idiosyncratic
factors, such as their earnings that month, the timing of their pay, other
irregular expenses, and the relative interest rates of different debt
obligations.
However, we might expect that certain factors observable in our data,
such as the payment amount in the previous months, would influence the
amount a cardholder chooses to pay. For example, we found that 23
percent of cardholders with revolving balances paid the same amount
toward their balance as they did in the previous month, compared with
only 9 percent of transacting cardholders. We might also expect that
cardholders would reconsider the amounts they pay toward their balance
if their balance is increasing. Also, according to research by JPMorgan
Chase Institute, their sample of checking account holders who also had a
Chase credit card and who received tax refunds made, on average,
substantially larger payments toward their credit card balances when their
tax refund arrived and then slowly added to their credit card debt over the
year as the extra funds ran out.1
Furthermore, payment decisions might have changed during the COVID-
19 pandemic—for example, in response to substantial increases in
uncertainty from the pandemic or to pandemic-related government
assistance that started to flow to individuals and their employers
beginning in April 2020.
1Diana Farrell, Fiona Greig, and Amar Hamoudi, Tax Time: How Families Manage Tax
Refunds and Payments (JPMorgan Chase Institute, 2019).
Page 113 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
Methodology
Using the Board of Governors of the Federal Reserve’s Y-14M credit card
data from June 2013 through December 2021, we modeled the decisions
revolvers made about credit card payment amounts, including the
potential association between various pandemic-related events and
cardholders’ payment amounts.2 To examine these dynamics, we
developed a baseline model and an extended model to estimate the
amount paid each month by cardholders who had revolved in the prior
month.
Our baseline model is as follows:
where Payment amount is the amount paid by the cardholder in a given
month.
The other variables are as follows:
· Cycle beginning balance is the total balance at the start of the month.
· Usual payment is the average of the amount paid over the previous 3
months.
· Change in revolving balance is the difference between the revolving
balance of the prior month and that from 3 months prior. We defined
revolving balance as the difference between the cycle beginning
balance and the amount paid in that cycle. A positive amount in the
difference indicates that the revolving balance has grown, and a
negative amount indicates it has decreased.
· COVID is an indicator for whether or not the estimated payment due
date occurred after March 13, 2020 (when the COVID-19 pandemic
was declared). Our data did not have payment due dates, so we
estimated the payment due dates by adding 21 days to the date of the
previous cycle ending date, which is the minimum allowed timing for
payment due dates.
2For more information on the Federal Reserve data and the definition of a “revolver,” see
app. II.
Page 114 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
· is a vector of indicators for the months of the year (with September
as the reference month).
We used a fixed-effects ordinary least squares modeling approach, with
the fixed effects at the account ID level. We further used robust standard
errors to allow for heteroscedasticity between accounts and serial
correlation within each account.
Because we used lagged information about cardholders in our model, we
dropped the first 3 months that each account appears in the data. We
further dropped any observations with negative cycle beginning balances,
negative payments, or negative average payment estimates.
We then extended the baseline model to estimate the association
between payment choices and the COVID-19 pandemic initial shutdown
in March 2020, as well as the four federal-level disbursements of cash
assistance directly to individuals. Our extended model is as follows:
In addition to our indicator for the entire pandemic period (COVIDit), we
constructed five indicator variables to capture potential differences in
credit card payments made in the month immediately following the
shutdown and in the months when disbursements of each of the
pandemic assistance payments were made.3 The advance child tax credit
payments were disbursed six times in 2021, so we flagged each of those
payment cycles as potentially influenced by the credit. We defined
indicators for each of these five different pandemic-related events as
follows:
· Indicator for initial shutdown month for observations with an estimated
payment due date that occurred after March 13, 2020, and on or
before April 13, 2020 (shutdown)
3The federal government declared the COVID-19 pandemic on March 13, 2020. The first
disbursements of the three economic impact payments occurred on April 10, 2020,
December 29, 2020, and March 17, 2021. The first disbursement of the advance child tax
credit payments occurred on July 15, 2021.
Page 115 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
· Indicator for the first economic impact payment if the observation’s
estimated payment was after April 10, 2020, and on or before May 10,
2020 (eip 1)
· Indicator for the second economic impact payment if the observation’s
estimated payment was after December 29, 2020, and on or before
January 30, 2021 (eip 2)
· Indicator for the third economic impact payment if the observation’s
estimated payment was after March 17, 2021, and on or before April
17, 2021 (eip 3)
· An indicator for the advance child tax credit payments is assigned to
all observations where the observation’s estimated payment was after
July 15, 2021, ending with the last month of our data in December
2021 (CTC)
We again used fixed effects and robust standard errors to allow for
heteroscedasticity between accounts and serial correlation within each
account.
Limitations
Our analysis has limitations and our results should be interpreted with
caution. In all cases, our estimates of effect sizes were average effect
sizes of the entire population of revolving accounts in our sample of the
Y-14M data. We were not able to determine which cardholders
associated with the accounts in our sample received pandemic
assistance. As a result, our estimates of the effect of pandemic
assistance funds used payment data for cardholders who both did and did
not receive such funds.
Additionally, the effect sizes of the economic impact payments and the
advance child tax credit were not directly comparable because a much
smaller population would have been eligible to receive the advance child
tax credits than was eligible to receive the economic impact payments. In
both cases, we did not know what fraction of cardholders in our sample
received either type of payment. Further, we were not able to determine
that these increased credit card payment amounts were due to the federal
payments or to other events that occurred with similar timing that would
have facilitated some cardholders’ average increase in payments.
Page 116 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
Results
Table 6 lists the results of our model estimates for both the baseline and
extended models.
Table 6: Estimated Coefficients of Credit Card Payment Amounts Made by
Revolving Accounts
Baseline model coefficients Extended model coefficients
Variable (t-statistic) (t-statistic)
Average recent 0.22*** 0.22***
payments (25.34) (25.30)
Account balance 0.09*** 0.09***
(79.12) (79.11)
Change in revolving 0.05*** 0.05***
balance (25.36) (25.37)
COVID-19-era 42.89*** 39.53***
(25.19) (21.38)
Shutdown — -88.42***
(-22.33)
Economic impact — -44.23***
payment I (-11.22)
Economic impact — 20.07***
payment II (4.51)
Economic impact — 60.79***
payment III (13.36)
Advance child tax — 36.51***
credit (14.55)
Month indicator
January 5.46*** 6.65***
(2.73) (3.23)
February 15.37*** 19.00***
(7.42) (9.15)
March 42.25*** 48.80***
(20.74) (23.91)
April 28.20*** 37.37***
(13.94) (17.62)
May 18.92*** 25.48***
(9.77) (13.02)
Page 117 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
Baseline model coefficients Extended model coefficients
Variable (t-statistic) (t-statistic)
June 10.70*** 15.04***
(5.42) (7.55)
July 12.39*** 14.82***
(6.33) (7.53)
August 13.80*** 13.72***
(6.96) (6.92)
October -4.16** -3.86**
(-2.2) (-2.04)
November -5.74*** -5.47***
(-3.04) (-2.89)
December -1.93 -0.81
(-0.99) (-0.41)
Constant -39.24*** -42.63***
(-6.02) (-6.54)
Fixed effects at the Yes Yes
account level:
N 8.2 million 8.2 million
R2: Within 0.04 0.04
Between 0.44 0.44
Overall 0.18 0.18
Legend: ** = significantly different from zero at the 95 percent confidence level; *** = significantly
different from zero at the 99 percent confidence level; — = not applicable
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System. | GAO-23-105269
In general, payments made by revolvers suggest that only a small portion
of their payment decision was associated with the total balance in the
current month (9 cents for every dollar of balance) and a much larger
weight (22 cents for every dollar) was placed on the average payment
over the previous 3 months. The positive coefficient on the change in
revolving balance suggests that cardholders would, on average, increase
their payments slightly in response to a rising revolving balance, though
they only increased it by 5 cents for every dollar the balance increased.
We also found a strong seasonal effect, in line with prior research,
suggesting that, on average, revolvers would pay the most toward their
balances in March or April, consistent with a tax refund effect.
In our baseline model, we estimated that during the pandemic, revolvers
paid on average $43 more per month than prior to the pandemic. This is
consistent with our other findings that revolvers were able to pay off their
balances more quickly during the pandemic.
Page 118 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
Our extended model suggests that variations in payment patterns during
the pandemic were associated with the timing of the disbursements of
federal pandemic cash assistance, though other events concurrent with
the timing of the assistance payments may have been driving some of our
estimates. In the extended model, we found that overall during the period
after March 13, 2020, through the end of December 2021, cardholders
paid an additional $40 each month on average. However, in the initial
shutdown, prior to assistance payments, revolvers paid on average $49
less than prior to March 2020.4 This suggests either that revolvers might
have been keeping cash in reserve due to the extreme uncertainty or that
they had lost income in the days immediately following the shutdown.
When the first economic impact payments were disbursed in April 2020,
cardholders on average increased their payments from the March 2020
decrease by $44, for a net reduction of $5 prior to prepandemic
baselines. The second and third economic impact payments were
associated with increases of $20 and $61 in credit card payments,
respectively, beyond the average increase during the pandemic. This is
consistent with increased proportions of people reporting using the
economic impact payments to pay down debt, as estimated by the
Census Bureau’s Household Pulse Survey, as previously discussed in the
report.5
Finally, in every month in which the advance child tax credit payments
were sent out, revolvers increased their credit card payments by an
average of $37 beyond the average increase during the pandemic.
Because the advance child tax credit payments were recurring generally
4The $49 dollar amount is the net of an increase in $40 for all months after the start of the
pandemic and a decrease during the initial shutdown of $88 (difference was due to
rounding).
5Census conducted the Household Pulse Survey each week from April 23 to July 21,
2020, and every 2 weeks starting in August 2020. We reviewed the surveys conducted
from April 23, 2020, to October 11, 2021. In the surveys conducted from June 11 to July
21, 2020, from January 6 to March 29, 2021, and from April 14 to July 5, 2021,
respondents were asked if they received or expected to receive an economic impact
payment and on what they spent the payment, including paying down credit card
balances, student loans, or other debts. The 95 percent confidence intervals for the
estimates ranged from about 13 to 17 percent based on surveys conducted after the
distribution of the first economic impact payments, from 49 to 54 percent based on
surveys conducted after the distribution of the second economic impact payments, and
from 48 to 59 percent based on surveys conducted after the third economic impact
payments.
Page 119 GAO-23-105269 Credit Cards
Appendix V: Econometric Analysis of Credit
Card Payments
for 6 months, these payments were associated with the largest overall
contribution of additional credit card payments ($219).
Page 120 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Appendix VI: Econometric
Analysis of Credit Card Terms
and Revolving Balances by
Race/Ethnicity in Cardholders’
Zip Codes
This appendix presents the development, estimation, results, and
limitations of our econometric analyses of disparities in credit card interest
rates, credit limits, and revolving balances associated with differences in
racial and ethnic composition in cardholders’ billing zip codes.
Data and Variable Definitions
We used a 0.1 percent nongeneralizable sample of credit card accounts
from the Board of Governors of the Federal Reserve System’s Y-14M
credit card data from June 2013 through December 2021. We
supplemented our analysis with zip-code-level data on race and ethnicity
and median household income from the Census Bureau’s 5-year
American Community Surveys from 2013 through 2020. See appendix II
for details on the datasets and their construction and appendix III for full
details on the use of zip-code-level racial and ethnic composition data.
In these models, we examined three credit card outcomes: interest rates,
credit limits, and revolving balances. Interest rates and credit limits are
credit card terms that are offered by the card issuer and accepted by the
cardholder. Revolving balances are the result of cardholders’ purchase
and payment decisions, which are likely informed by their interest rate
and credit limit, along with other factors. The definitions of these
outcomes are as follows:
· Credit card interest rate. The current interest rate for general
purchases on the account, in percentage points. Note that actual
interest rate assessed may differ if the account has purchases that
qualify for a promotional rate or that are being charged a penalty rate
or different rate for balance transfers or cash advances. Likewise, if
the account is transacting and eligible for the grace period, the
Page 121 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
account holder will not pay the listed rate but will instead pay 0
percent.
· Credit limit. The maximum amount of credit available for a credit card
account.
· Revolving balance (for revolving accounts only). The calculated
balance of the difference between the account’s beginning balance
and payment amount. Revolving balances are treated as positive
when the beginning balance exceeds the payment amount.
Methodology
Credit Card Term Regressions
We used econometric models to estimate relationships between credit
card terms (interest rates and credit limits) and the racial and ethnic
composition of cardholders’ billing zip codes. For each of the two credit
card terms, we estimated the parameters of the following specification:
In these models, y is either the interest rate on the card or the credit limit;
%Asian, %Black, %Hispanic, and %Other Race/Ethnicity are the
percentages of Asian, Black or African American, Hispanic or Latino, and
other people who are not White, respectively, in the cardholder’s billing
zip code; X is a vector of other variables; and is the error term. The
percentage of White residents is the omitted racial category.
In our baseline specification, X included indicators for the month and the
cardholder’s state of residence. In other specifications, X also included
variables measuring the cardholder’s credit score, the status of the
account, and the income distribution in the cardholder’s billing zip code.
We estimated the parameters of these models using all account-month
observations for accounts in transacting, revolving, and inactive status.
We measured account status using indicator variables for whether the
account was revolving or inactive; transacting was the omitted status. We
controlled for account status in our regressions because it may reflect
various factors that affect credit terms, though the sign of the association
and the direction of any underlying causal effects are ambiguous. For
example, inactive accounts may be associated with higher interest rates
Page 122 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
not because the issuer assigned a higher interest rate as a result of
account inactivity, but because the cardholder decided to use the high
interest rate card as little as possible to avoid the high interest charges.
Similarly, revolving accounts may have lower credit limits because they
have had a history of high credit utilization. Alternatively, issuers might
increase their limits seeing an opportunity for a high-return customer who
is signaling, through their use of credit, that they may use more of it if
given the opportunity.
We also controlled for the cardholders’ credit scores because they
incorporate information about an individual’s credit history and credit
usage, and card issuers use credit scores to help make credit
underwriting decisions. Some research suggests that credit scores are
also correlated with individuals’ race and ethnicity.1
We controlled for the income distribution in cardholders’ zip codes. We
measured the distribution of income in cardholders’ zip codes as the
percentage of households with income in each of 16 income groups;
$60,000–$74,999 was the omitted income group.2
In all cases, we assume that our standard errors will have independent
errors between zip codes but we allow for possible correlation across
error terms within zip codes, so we cluster our error terms on zip code,
providing more conservative standard errors and confidence intervals.
Revolving Balance Regressions
We used similar econometric models to estimate relationships between
revolving balance amounts and the racial and ethnic composition of the
cardholders’ billing zip codes. We estimated the parameters of these
models using only account-month observations of accounts in revolving
1See Board of Governors of the Federal Reserve System, Report to the Congress on
Credit Scoring and Its Effects on the Availability and Affordability of Credit (August 2007),
and Robert B. Avery, Kenneth P. Brevoort, and Glenn B. Canner, “Credit Scoring and Its
Effects on the Availability and Affordability of Credit,” The Journal of Consumer Affairs,
vol. 43, no. 3 (2009).
2We used more granular income categories in our regression analysis as compared with
our descriptive analysis. The income groups used for the regression analysis are as
follows: below $10,000, $10,000–$14,999, $15,000–$19,999, $20,000–$24,999, $25,000–
$29,999, $30,000–$34,999, $35,000–$39,999, $40,000–$44,999, $45,000–$49,999,
$50,000–$59,999, $75,000–$99,999, $100,000–$124,999, $125,000–$149,999,
$150,000–$199,000, and $200,000 and above.
Page 123 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
status and omitted variables measuring account status from our
regression models, but otherwise used the same approach.
Scenario Development
To provide a more concrete scale of effect sizes measured in our credit
term and revolving balance regressions, we created illustrative scenarios
using the representative racial and ethnic clusters of zip codes as
described in appendix III. The cluster centroids allowed us to use realistic
combinations of racial and ethnic groups to estimate the overall marginal
effect of moving from one zip code cluster to another.
To calculate the marginal effects, we
· multiplied the percentage of each group from a zip code cluster
centroid with its respective estimated model coefficient,
· summed the individual products for each centroid, and
· calculated the difference between each centroid and the one
representing predominantly White zip codes.
This allowed us to estimate the overall effect of moving from a
predominantly White zip code to one of our other groupings, while holding
constant all other factors controlled for in the model.
Assumptions and Limitations
Our analysis has limitations and our results should be interpreted with
caution. The sample of credit card accounts we analyzed may not
represent the population of all credit card accounts, so our results reflect
the sample of accounts we analyzed and may not generalize to other
credit card accounts. Similarly, our results may not generalize to other
time periods.
Omitted variable bias poses a validity threat to our models, just as it does
to any ordinary least squares regression. Our analysis was not designed
to determine all the potential reasons for differences in credit card terms
associated with the racial and ethnic composition of cardholders’ zip
codes, which may be due to factors that are either not captured in the
data we analyzed or cannot be measured. Such factors could include the
extent to which cardholders compare credit terms when choosing a card,
competition among lenders, and lender-specific risk evaluation or loan
Page 124 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
approval processes.3 Differences in these factors may be incidentally
associated with race and ethnicity, or these factors may themselves be
influenced by race and ethnicity, and therefore act as channels through
which race and ethnicity influence credit terms.
Results
In all of our models, we found statistically and practically significant
differences in interest rates, credit limits, and revolving balances
associated with different racial and ethnic zip code clusters before and
after controlling for other factors that would be expected to influence
these parameters—that is, cardholders’ credit scores, household incomes
in their zip codes, and revolving status.
Interest rates. Table 7 presents estimated coefficients from different
specifications of our econometric model of credit card interest rates.
3For a discussion of borrowers comparing credit terms, see Victor Stango and Jonathan
Zinman, “Borrowing High versus Borrowing Higher: Price Dispersion and Shopping
Behavior in the U.S. Credit Card Market,” The Review of Financial Studies, vol. 29, no. 4
(2016): 979–1006, and Ping Cheng, Zhenguo Lin, and Yingchun Liu, “Racial Discrepancy
in Mortgage Interest Rates,” Journal of Real Estate Finance and Economics, vol. 51, no. 1
(2015): 101–120. For a discussion of lender competition, see Alexander W. Butler, Erik J.
Mayer, and James P. Weston, “Racial Disparities in the Auto Loan Market,” Review of
Financial Studies, vol. 36, no. 1 (2023): 1–41; Robert Bartlett et al., “Consumer-Lending
Discrimination in the FinTech Era,” Journal of Financial Economics, vol. 143, no. 1 (2022):
30–56; and Neil Bhutta, Aurel Hizmo, and Daniel Ringo, “How Much Does Racial Bias
Affect Mortgage Lending?” Evidence from Human and Algorithmic Credit Decisions”
(working paper, Finance and Economics Discussion Series, Board of Governors of the
Federal Reserve System, Aug. 2, 2022),
https://www.federalreserve.gov/econres/feds/how-much-does-racial-bias-affect-mortgage-
lending.htm. The latter study also discusses how differences in mortgage approval may be
associated with lender-specific loan approval processes, including stricter credit standards
for all borrowers or varying quality of services provided. For a discussion of differences in
how issuers evaluate risk factors, see Stango and Zinman, “Borrowing High versus
Borrowing Higher.” The authors argue that issuers’ internal risk-based pricing models may
place different emphasis on risk factors such as credit scores and late payments, and
hence cardholders may receive different interest rates from different issuers.
Page 125 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Table 7: Estimated Differences in Credit Card Interest Rates by Race and Ethnicity in Cardholders’ Billing Zip Codes, June
2013–December 2021
Variables Model 1 Model 2 Model 3 Model 4
coefficient coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic) (t-statistic)
Percentage of residents in billing zip code who are: Black or 2.17*** 1.42*** 0.83*** 0.72***
African American (38.91) (26.91) (16.29) (13.02)
Percentage of residents in billing zip code who are: Asian -0.81*** 0.26*** 0.35*** 0.52***
(-8.88) (3.19) (4.34) (5.96)
Percentage of residents in billing zip code who are: Hispanic or 2.22*** 1.56*** 1.06*** 0.94***
Latino (41.59) (31.85) (22.63) (18.84)
Percentage of residents in billing zip code who are: Other 2.68*** 1.84*** 1.06*** 0.96***
(10.84) (7.9) (4.7) (4.25)
Account status: Revolving — 3.72*** 2.60*** 2.60***
(241.31) (154.79) (154.29)
Account status: No activity — 2.38*** 2.45*** 2.45***
(167.26) (170.72) (170.27)
Credit score — — -0.01*** -0.01***
(-156.98) (-155.71)
Percentage of households in billing zip code with household — — — 0.08
incomes: Under $10,000 (0.26)
Percentage of households in billing zip code with household — — — 0.42
incomes: $10,000–$14,999 (1.05)
Percentage of households in billing zip code with household — — — 0.58
incomes: $15,000–$19,999 (1.39)
Percentage of households in billing zip code with household — — — -0.77*
incomes: $20,000–$24,999 (-1.82)
Percentage of households in billing zip code with household — — — -0.30
incomes: $25,000–$25,999 (-0.69)
Percentage of households in billing zip code with household — — — -0.38
incomes: $30,000–$34,999 (-0.89)
Percentage of households in billing zip code with household — — — 0.09
incomes: $35,000–$39,999 (0.2)
Percentage of households in billing zip code with household — — — 0.55
incomes: $40,000–$44,999 (1.22)
Percentage of households in billing zip code with household — — — -0.06
incomes: $45,000–$49,999 (-0.13)
Percentage of households in billing zip code with household — — — 0.27
incomes: $50,000–$59,999 (0.71)
Percentage of households in billing zip code with household — — — 0.39
incomes: $75,000–$99,999 (1.19)
Page 126 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Variables Model 1 Model 2 Model 3 Model 4
coefficient coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic) (t-statistic)
Percentage of households in billing zip code with household — — — -0.11
incomes: $100,000–$124,999 (-0.3)
Percentage of households in billing zip code with household — — — -0.15
incomes: $125,000–$149,999 (-0.37)
Percentage of households in billing zip code with household — — — -0.50
incomes: $150,000–$199,999 (-1.42)
Percentage of households in billing zip code with household — — — -0.42*
incomes: $200,000 and above (-1.67)
Constant 13.83*** 11.80*** 22.69*** 22.67***
(84.74) (84.19) (150.53) (82.96)
F-test of all income variables — — — F(15, 30175) =
5.01***
State fixed effects Yes Yes Yes Yes
Month fixed effects Yes Yes Yes Yes
N 30.3 million 30.3 million 30.0 million 30.0 million
R2 0.04 0.11 0.13 0.13
Legend: * = significantly different from zero at 90 percent confidence level; *** = significantly different from zero at the 99 percent confidence level; — =
not applicable
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and Census Bureau. | GAO-23-105269
As shown in table 7, in all of our models, our racial composition variables
were statistically significant, as were all control variables, with the
exception of the income groups. However, the income groups were jointly
statistically significant.
Increases in the percentage of residents who were Black, Hispanic, and
Other race in a cardholder’s zip code were associated with a significantly
higher credit card interest rate. While the estimated effects are reduced
somewhat when control variables are added to the model, the differences
remain substantial.
In the model without controls, an increase in the percentage of Asian
residents in the cardholder’s zip code was associated with a modest
reduction in interest rates. However, when we controlled for additional
factors that might affect interest rates, the sign of the association flipped,
and the percentage of Asian residents in a cardholder’s billing zip code
became associated with an increase in interest rates. This happens when
a variable is added to the model that is both correlated with the Asian
percent coefficient and with lower interest rates. Cardholders living in zip
codes with more Asian residents were more likely to be transactors and
Page 127 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
have higher incomes and credit scores than other minority groups. Once
we controlled for those correlations, our model revealed a disparity in
interest rates for zip codes higher in Asian residents relative to similarly
situated zip codes with lower concentrations of Asian residents and
higher concentrations of White residents.
Both revolvers and inactive cardholders faced higher interest rates than
transactors—over 2 percentage points after controlling for zip-code
incomes and credit scores. As expected, cardholders with higher credit
scores generally had lower interest rates—every 10-point increase in
credit score was associated with a reduction of 0.1 percentage point in
interest rate. While controlling for zip-code-level income composition
overall was statistically significant, the coefficients on individual income
groups were generally not statistically significant and their sign followed
no particular pattern. This suggests that any association between zip-
code incomes and interest rates was sufficiently weak and was not
captured by the zip-code-level income measures.
Credit limits. Table 8 presents estimated coefficients from alternative
specifications of our econometric model of credit limits on credit cards.
Table 8: Estimated Differences in Credit Limits by Race and Ethnicity in Cardholders’ Billing Zip Codes, June 2013–December
2021
Variable Model 1 Model 2 Model 3 Model 4
coefficient coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic) (t-statistic)
Percentage of residents in billing zip code who are: Black or -5,440*** -4,822*** -2,638*** -1,036***
African American (-48.71) (-45.75) (-29.78) (-13.15)
Percentage of residents in billing zip code who are: Asian 1,221*** 310 125 -3,112***
(3.51) (0.97) (0.47) (-19.46)
Percentage of residents in billing zip code who are: Hispanic or -7,100*** -6,511*** -4,582*** -2,347***
Latino (-57.26) (-56.17) (-46.06) (-30.72)
Percentage of residents in billing zip code who are: Other -6,498*** -5,842*** -3,020*** -1,576***
(-13.64) (-13.16) (-7.92) (-4.78)
Account status: Revolving — -3,507*** 160*** 269***
(-135.76) (7.02) (12.18)
Account status: No activity — -3,678*** -3,655*** -3,563***
(-139.7) (-147.34) (-148.13)
Credit score — — 47*** 46***
(341.99) (344)
Percentage of households in billing zip code with household incomes:
Page 128 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Variable Model 1 Model 2 Model 3 Model 4
coefficient coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic) (t-statistic)
Percentage of households in billing zip code with household — — — 2,374***
incomes: Under $10,000 (4.73)
Percentage of households in billing zip code with household — — — -864
incomes: $10,000–$14,999 (-1.41)
Percentage of households in billing zip code with household — — — -2,594***
incomes: $15,000–$19,999 (-4.03)
Percentage of households in billing zip code with household — — — -296
incomes: $20,000–$24,999 (-0.47)
Percentage of households in billing zip code with household — — — -666
incomes:$25,000–$25,999 (-1.03)
Percentage of households in billing zip code with household — — — -782
incomes: $30,000–$34,999 (-1.24)
Percentage of households in billing zip code with household — — — -1,660**
incomes: $35,000–$39,999 (-2.52)
Percentage of households in billing zip code with household — — — -906
incomes: $40,000–$44,999 (-1.37)
Percentage of households in billing zip code with household — — — -156
incomes: $45,000–$49,999 (-0.21)
Percentage of households in billing zip code with household — — — 556
incomes: $50,000–$59,999 (0.99)
Percentage of households in billing zip code with household — — — -568
incomes:$75,000–$99,999 (-1.12)
Percentage of households in billing zip code with household — — — 617
incomes: $100,000–$124,999 (1.13)
Percentage of households in billing zip code with household — — — -483
incomes: $125,000–$149,999 (-0.76)
Percentage of households in billing zip code with household — — — 557
incomes: $150,000–$199,999 (0.91)
Percentage of households in billing zip code with household — — — 13,544***
incomes: $200,000 and above (30.3)
Constant 11,866*** 14,239*** -22,757*** -23,066***
(25.92) (34.18) (-61.39) (-50.89)
F-test of all income variables — — — F(15, 30185) =
315.24***
State fixed effects Yes Yes Yes Yes
Month fixed effects Yes Yes Yes Yes
N 30.4 million 30.4 million 30.0 million 30.0 million
Page 129 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Variable Model 1 Model 2 Model 3 Model 4
coefficient coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic) (t-statistic)
R2 0.03 0.07 0.21 0.23
Legend: ** = significantly different from zero at the 95 percent confidence level; *** = significantly different from zero at the 99 percent confidence level;
— = not applicable
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and Census Bureau. | GAO-23-105269
Our results for our credit limit analysis paralleled those of the interest rate
analysis. Higher concentrations of Black, Hispanic, and Other residents in
a zip code were associated with substantially lower credit limits, though
the scale of the penalty was reduced as controls were added to the
model. In the baseline model without other controls, higher concentrations
of Asian residents in a zip code were associated with higher credit limits,
but this association switched sign once additional controls were added.
Once all of our controls were added to the model, Asian concentrations
were associated with the largest credit limit penalties.
In addition, both revolving and inactive accounts were associated with
credit limits of more than $3,000 less than transacting accounts in the
model without controlling for characteristics that may influence credit
limits. After we controlled for credit scores, revolving accounts had
somewhat higher credit limits than transacting accounts with similar credit
scores. Our analysis estimated that higher credit scores were associated
with higher credit limits—an increase of 10 points in credit score was
associated with an almost $500 increase in credit limits. Once again, our
analysis found a more ambiguous association between zip-code incomes
and credit limits.
Revolving balances. Table 9 presents estimated coefficients from
alternative specifications of our econometric model of balances on credit
cards in revolving status.
Table 9: Estimated Differences in Revolving Balances by Race and Ethnicity in Cardholders’ Billing Zip Codes, June 2013–
December 2021
Variables Model 1 Model 2 Model 3
coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic)
Percentage of residents in billing zip code who are: Black or African -1,517*** -1,426*** -775***
American (-27.03) (-25.38) (-13.17)
Percentage of residents in billing zip code who are: Asian 314* 312* -1,009***
(1.95) (1.95) (-7.08)
Page 130 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Variables Model 1 Model 2 Model 3
coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic)
Percentage of residents in billing zip code who are: Hispanic or Latino -2,008*** -1,939*** -1,129***
(-35.02) (-33.84) (-20.41)
Percentage of residents in billing zip code who are: Other -1,663*** -1,518*** -811***
(-6.74) (-6.23) (-3.47)
Credit score — 2*** 2***
(23.86) (20.13)
Percentage of households in billing zip code with household incomes: — — 260
Under $10,000 (0.67)
Percentage of households in billing zip code with household incomes: — — -718
$10,000–$14,999 (-1.42)
Percentage of households in billing zip code with household incomes: — — 495
$15,000–$19,999 (0.71)
Percentage of households in billing zip code with household incomes: — — -391
$20,000–$24,999 (-0.78)
Percentage of households in billing zip code with household incomes: — — 45
$25,000–$25,999 (0.08)
Percentage of households in billing zip code with household incomes: — — -282
$30,000–$34,999 (-0.52)
Percentage of households in billing zip code with household incomes: — — -1,567***
$35,000–$39,999 (-2.84)
Percentage of households in billing zip code with household incomes: — — -589
$40,000–$44,999 (-1.04)
Percentage of households in billing zip code with household incomes: — — -97
$45,000–$49,999 (-0.16)
Percentage of households in billing zip code with household incomes: — — 809
$50,000–$59,999 (1.58)
Percentage of households in billing zip code with household incomes: — — 165
$75,000–$99,999 (0.39)
Percentage of households in billing zip code with household incomes: — — 824*
$100,000–$124,999 (1.76)
Percentage of households in billing zip code with household incomes: — — 1139**
$125,000–$149,999 (2.14)
Percentage of households in billing zip code with household incomes: — — 1,816***
$150,000–$199,999 (3.35)
Percentage of households in billing zip code with household incomes: — — 4,204***
$200,000 and above (11.45)
Page 131 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Variables Model 1 Model 2 Model 3
coefficient coefficient coefficient
(t-statistic) (t-statistic) (t-statistic)
Constant 4,693*** 3,347*** 2,888***
(17.62) (12.31) (7.32)
F-test of all income variables — — F(15, 27644) =
72.05***
State fixed effects Yes Yes Yes
Month fixed effects Yes Yes Yes
N 10.6 million 10.6 million 10.6 million
R2 0.01 0.01 0.02
Legend: * = significantly different from zero at 90 percent confidence level; ** = significantly different from zero at the 95 percent confidence level; *** =
significantly different from zero at the 99 percent confidence level; — = not applicable
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and Census Bureau. | GAO-23-105269
As shown in table 9, our analysis suggests cardholders in zip codes with
more residents of Black, Hispanic, and Other race and ethnicity carried
statistically significant and substantially lower balances than cardholders
in predominantly White zip codes. After we controlled for credit scores
and zip-code incomes, the size of the association reduced by about half
but remained substantively and statistically significant.
Our analysis also suggests cardholders in zip codes with high shares of
Asian residents had slightly higher revolving balances (ones that were
only statistically different from zero at the less strict p < 0.1 confidence
level). Once we controlled for credit scores and zip-code incomes,
cardholders in zip codes with high shares of Asian residents had
substantially lower revolving balances relative to otherwise similar
cardholders in zip codes with higher concentrations of White residents.
Cardholders who carried a balance and lived in zip codes with greater
concentrations of higher-income households and households with higher
credit scores had higher revolving balances.
Results of scenarios. Table 10 shows the estimated differences in
interest rates, credit limits, and revolving balances of credit card accounts
for cardholders in four hypothetical zip codes compared with those of
accounts for cardholders in a zip code that is predominantly White. We
derived these estimates using estimated coefficients from specifications
with both the baseline model without controls and the full set of control
variables.
Page 132 GAO-23-105269 Credit Cards
Appendix VI: Econometric Analysis of Credit
Card Terms and Revolving Balances by
Race/Ethnicity in Cardholders’ Zip Codes
Table 10: Estimated Differences in Credit Terms and Revolving Balances by Race and Ethnicity in Cardholders’ Billing Zip
Codes, June 2013–December 2021
Difference compared with cardholders in predominantly White zip codes
Percentage point difference in Dollar amount difference in credit Dollar amount difference in
interest rates limits revolving balances
With controls With controls
Without controls for revolving Without controls for revolving With controls
for revolving status, credit for revolving status, credit Without controls for credit
Racial and status, credit scores, and status, credit scores, and for credit scores scores and
ethnic zip scores, and zip- zip-code scores, and zip- zip-code and zip-code zip-code
code clusters code incomes incomes code incomes incomes incomes incomes
Majority 1.4 0.6 -$4,285 -$1,477 -$1,212 -$711
Hispanic or
Latino
residents
Majority Black 1.3 0.5 -3,412 -710 -953 -492
or African
American
residents
Mixed race and 0.4 0.2 -1,330 -540 -374 -258
ethnicity with
majority White
residents
Mixed race and 0.2 0.4 -1,079 -1,489 -308 -564
ethnicity with
no majority
residents
Source: GAO analysis of data from the Board of Governors of the Federal Reserve System and Census Bureau. | GAO-23-105269
As shown in table 10, our analysis suggests that cardholders in zip codes
with higher shares of Black, Hispanic, or Asian residents had higher
interest rates, lower credit limits, and lower revolving balances compared
with zip codes with higher shares of White residents. For example,
cardholders in majority-Black zip codes had 1.3 percentage points higher
interest rates than cardholders in predominantly White zip codes, and
cardholders in majority-Hispanic zip codes had interest rates 1.4 points
higher. Consistent with lower credit limits, cardholders in majority-Black
and majority-Hispanic zip codes carried lower revolving balances. Also,
zip codes with mixed race and ethnicity but no majority had the highest
share of Asian residents among the five zip code clusters. Cardholders in
these zip codes had interest rates that were 0.2 percentage points higher,
credit limits that were $1,079 lower, and revolving balances that were
$308 lower. The differences in credit terms and revolving balances for
cardholders in zip codes of different racial and ethnic mixes continued to
exist after we controlled for factors that can contribute to differences in
credit terms or revolving balances.
Page 133 GAO-23-105269 Credit Cards
Appendix VII: GAO Contact and Staff
Acknowledgments
Appendix VII: GAO Contact and
Staff Acknowledgments
GAO Contact
Alicia Puente Cackley, 202-512-8678 or cackleya@gao.gov
Staff Acknowledgments
In addition to the contact named above, Karen Tremba (Assistant
Director), Anna Chung (Analyst in Charge), Ben Bolitzer, James
Boohaker, Abigail Brown, Chelsea Carter, Giselle Cubillos-Moraga, Jill
Lacey, Courtney LaFountain, Matthew Levie, Nicholas Pigeon Rossy,
Jessica Sandler, Eric Schwab, Jennifer Schwartz, Daniel Speer, Jena
Sinkfield, and Jeff Tessin made key contributions to this report.
Page 134 GAO-23-105269 Credit Cards
GAO’s Mission
The Government Accountability Office, the audit, evaluation, and investigative
arm of Congress, exists to support Congress in meeting its constitutional
responsibilities and to help improve the performance and accountability of the
federal government for the American people. GAO examines the use of public
funds; evaluates federal programs and policies; and provides analyses,
recommendations, and other assistance to help Congress make informed
oversight, policy, and funding decisions. GAO’s commitment to good government
is reflected in its core values of accountability, integrity, and reliability.
Obtaining Copies of GAO Reports and Testimony
The fastest and easiest way to obtain copies of GAO documents at no cost is
through our website. Each weekday afternoon, GAO posts on its website newly
released reports, testimony, and correspondence. You can also subscribe to
GAO’s email updates to receive notification of newly posted products.
Order by Phone
The price of each GAO publication reflects GAO’s actual cost of production and
distribution and depends on the number of pages in the publication and whether
the publication is printed in color or black and white. Pricing and ordering
information is posted on GAO’s website, https://www.gao.gov/ordering.htm.
Place orders by calling (202) 512-6000, toll free (866) 801-7077, or
TDD (202) 512-2537.
Orders may be paid for using American Express, Discover Card, MasterCard,
Visa, check, or money order. Call for additional information.
Connect with GAO
Connect with GAO on Facebook, Flickr, Twitter, and YouTube.
Subscribe to our RSS Feeds or Email Updates. Listen to our Podcasts.
Visit GAO on the web at https://www.gao.gov.
To Report Fraud, Waste, and Abuse in Federal
Programs
Contact FraudNet:
Website: https://www.gao.gov/about/what-gao-does/fraudnet
Automated answering system: (800) 424-5454 or (202) 512-7700
Congressional Relations
A. Nicole Clowers, Managing Director, ClowersA@gao.gov, (202) 512-4400, U.S.
Government Accountability Office, 441 G Street NW, Room 7125, Washington,
DC 20548
Public Affairs
Chuck Young, Managing Director, youngc1@gao.gov, (202) 512-4800
U.S. Government Accountability Office, 441 G Street NW, Room 7149
Washington, DC 20548
Strategic Planning and External Liaison
Stephen J. Sanford, Managing Director, spel@gao.gov, (202) 512-4707
U.S. Government Accountability Office, 441 G Street NW, Room 7814,
Washington, DC 20548