regulating consumer financial products: evidence...

REGULATING CONSUMER FINANCIAL PRODUCTS:EVIDENCE FROM CREDIT CARDS*

Sumit Agarwal

Souphala Chomsisengphet

Neale Mahoney

Johannes Stroebel

We analyze the effectiveness of consumer financial regulation by consider-ing the 2009 Credit Card Accountability Responsibility and Disclosure (CARD)Act. We use a panel data set covering 160 million credit card accounts and adifference-in-differences research design that compares changes in outcomesover time for consumer credit cards, which were subject to the regulations, tochanges for small business credit cards, which the law did not cover. We esti-mate that regulatory limits on credit card fees reduced overall borrowing costsby an annualized 1.6% of average daily balances, with a decline of more than5.3% for consumers with FICO scores below 660. We find no evidence of anoffsetting increase in interest charges or a reduction in the volume of credit.Taken together, we estimate that the CARD Act saved consumers $11.9 billiona year. We also analyze a nudge that disclosed the interest savings from payingoff balances in 36 months rather than making minimum payments. We detect asmall increase in the share of accounts making the 36-month payment valuebut no evidence of a change in overall payments. JEL Codes: D0, D14, G0, G02,G21, G28, L0, L13, L15.

*We thank the editors, Lawrence Katz and Andrei Shleifer, three anonymousreferees, and our discussants Effie Benmelech, Olivier de Jonghe, BrigitteMadrian, Victor Stango, Jeremy Tobacman, Jialan Wang, and JonathanZinman for thoughtful comments. We are grateful to John Campbell, ChrisCarroll, Raj Chetty, Liran Einav, Alexander Frankel, Matthew Gentzkow,Andra Ghent, Christian Hansen, Benjamin Keys, Theresa Kuchler, AndresLiberman, Monika Piazzesi, Jesse Shapiro, Richard Thaler, Alessandra Voena,and Glen Weyl, and seminar participants at the University of Chicago, New YorkUniversity, Harvard University, Harvard Business School, Duke University,Arizona State University, the University of Michigan, Texas A&M, thePhiladelphia Fed, the NBER meetings in Industrial Organization, CorporateFinance, Law & Economics, and Household Finance, the Boston Fed Conferenceon Payment Systems, the Empirical Macro Workshop in New Orleans, the SloanConference on Benefit-Cost Analysis of Financial Regulation, the Bonn/Bundesbank conference for Regulating Financial Intermediaries, the CFPB, theOCC, the FDIC, Kansas University, University of Virginia, Johns HopkinsUniversity, and the College of William and Mary for helpful comments. Wethank Regina Villasmil for truly outstanding and dedicated research assistance.Mahoney and Stroebel thank the Fama-Miller Center at Chicago Booth for finan-cial support. The views expressed are those of the authors alone and do not nec-essarily reflect those of the Office of the Comptroller of the Currency.

! The Author(s) 2014. Published by Oxford University Press, on behalf of Presidentand Fellows of Harvard College. All rights reserved. For Permissions, please email:[email protected] Quarterly Journal of Economics (2015), 111–164. doi:10.1093/qje/qju037.Advance Access publication on November 25, 2014.

111

I. Introduction

The recent financial crisis triggered a surge of interest inregulating consumer financial products (e.g., Campbell et al.2011; Posner and Weyl 2013). In the United States, the Dodd-Frank Wall Street Reform and Consumer Protection Act of 2010established a Consumer Financial Protection Bureau to monitorand regulate mortgages, student loans, credit cards, and otherconsumer products. In July 2013, the European Commissionfollowed suit and proposed new consumer financial protectionlegislation to simplify disclosures and tighten guidance require-ments related to financial products.

Proponents of this type of regulation argue that consumerfinancial markets have become increasingly unfair. Firms takeadvantage of consumers’ behavioral biases—such as myopia, pre-sent bias, and inattention—to earn large profits, especially fromunsophisticated and poor consumers.1 These proponents suggestthat regulation and additional information can protect less so-phisticated consumers and reduce aggregate borrowing costs.

Critics have expressed skepticism about the effectiveness ofconsumer financial regulations. While limits on hidden fees, forexample, can shift surplus from more sophisticated to less sophis-ticated consumers (Gabaix and Laibson 2006), there is less evi-dence that regulators can bring about an across-the-boardreduction in consumer costs. Regulators, these critics attest, arenaively playing a game of regulatory Whac-A-Mole—efforts tolimit certain fees will simply lead firms to offset reduced revenuewith higher prices on other product dimensions and restrict thesupply of credit (American Bankers Association 2013). Even pro-ponents of regulating late fees, such as Mullainathan, Barr, andShafir (2009), worry that ‘‘the reduced revenue stream to lendersfrom these fees would mean that other rates and fees would beadjusted to compensate.’’

This article aims to advance this debate in the context of the2009 Credit Card Accountability Responsibility and Disclosure(CARD) Act in the United States. We analyze the effectiveness

1. Senator Chris Dodd, lead sponsor of the CARD Act in the Senate, has noted,‘‘My colleague from New York, Senator Schumer, has called this ‘trip-wire pricing,’saying thewhole businessmodel of thecredit card industry isnotdesigned toextendcredit but to induce mistakes and trap consumers into debt. I think he is absolutelyright, unfortunately. This is an industry that has been thriving on misleading itsconsumers and its customers’’ (U.S. Senate 2009).

QUARTERLY JOURNAL OF ECONOMICS112

of two key aspects of the CARD Act: (i) regulatory limits on cer-tain types of credit card fees, which became effective in Februaryand August 2010, and (ii) attempts to affect consumers’ repay-ment behavior by requiring that monthly credit card statementsprovide clear information on the costs of making only the mini-mum payment, which became effective in February 2010.

We conduct a quantitative analysis of the effects of the CARDAct’s provisions using a panel data set on the near universe ofcredit card accounts held by the eight largest U.S. banks. Thesedata, assembled by the Office of the Comptroller of the Currency(OCC), provide us with account-level information on contractterms, utilization, and payments at the monthly level fromJanuary 2008 to December 2012. We observe fees at a disaggre-gated level, allowing us to isolate effects on categories such asover-limit fees and late fees. Our data cover 160 million consumerand small business accounts and a significant share of total in-dustry assets during our period of study.

We estimate the intended and possible unintended conse-quences of the CARD Act by using a difference-in-differences re-search design that compares changes in outcomes over time forconsumer credit cards, which were subject to the new regulations,to changes in outcomes for small business credit cards, which thelaw did not cover. Our strategy expands on the ConsumerFinancial Protection Bureau’s (2013a) analysis that also compa-res average outcomes for consumer and small business creditcards but does not conduct a formal difference-in-differencesanalysis. The identifying assumption is that in the absence ofthe CARD Act, outcomes for consumer and small business ac-counts would have maintained parallel trends.

To argue that our identifying assumption is valid, we showthat outcomes for consumer and small business credit cards movetogether in the pre–CARD Act period, with parallel trends in fees,interest charges, and measures of credit volume such as averagedaily balances (ADBs), credit limits, and the number of newaccounts. This is consistent with a high degree of institutionalsimilarity between consumer and small business cards. Bothtypes of cards are guaranteed by the personal financial assets ofthe account holder, and applicants for both types of cards arescreened on their personal FICO scores. The main difference isthat small business account holders must claim to use their cardsfor business purposes only. Banks in most circumstances do notmonitor this behavior, and survey evidence indicates that a

REGULATING CONSUMER FINANCIAL PRODUCTS 113

significant percentage of charges are used for personal expendi-ture (Federal Reserve Board of Governors 2010).

Using this difference-in-differences strategy, we find thatregulations to reduce fees were highly effective. For borrowerswith a FICO score below 660, total fees declined by 5.3 percentagepoints of ADB, or $57.69 per account per year, led by a largedecline in over-limit fees (3.3 percentage points) and a smallerdecline in late fees (1.5 percentage points). Account holders with aFICO score above 660 had lower pre–CARD Act fee levels, andexperienced a qualitatively similar but smaller decline in fees (0.5percentage point of ADB, or $7.59 per account, per year).Combined across the low and high FICO score accounts, theCARD Act reduced overall fees by an annualized 1.6% of borrow-ing volume, or $22.58 per account each year.2 Extrapolating thisnumber to the total outstanding credit card balances of $744 bil-lion in the first quarter of 2010 (Federal Reserve Bank of NewYork 2013) yields annual cost savings for U.S. credit card users of$11.9 billion a year.

The CARD Act also required monthly credit card statementsto prominently display the cost of repaying the balance when onlymaking minimum payments and compare this to the cost of re-paying the balance when making payments that would pay off thebalance within 36 months. The aim was to nudge consumerstoward paying off a larger fraction of their balances and reducetheir overall interest payments (Thaler and Sunstein 2008). Wefind that these disclosure requirements had a small but signifi-cant effect on borrowers’ repayment behavior. The number of ac-count holders paying at a rate that would repay the balancewithin 36 months increased by 0.4 percentage points on a baseof 5.3%. Cyclical and seasonal variation make it hard to deter-mine whether these account holders would have counterfactuallybeen making higher or lower payments. We estimate as an upperbound that the nudge reduced aggregate interest payments by nomore than 0.01% of ADB, or approximately $57 million a year.

We next examine possible unintended consequences of theCARD Act on credit card pricing and credit volume. We start bymaking the conceptual point that if markets are perfectly com-petitive, so that the aggregate price inclusive of all fees is forced

2. We use the term ‘‘borrow’’ to refer to average daily balances (ADBs). As wediscuss in Section II, ADBs donot include purchase volumethat is paidoff before theend of a consumer’s grace period.


down to marginal cost, any regulation that reduces a certain feewill be offset by a similarly sized increase in another pricingdimension. Similarly, if all fees and prices are perfectly salient,then demand is only responsive to the aggregate price and will beunresponsive to an equally sized reduction in one fee and increasein another. This means that regulators can only be successful inlowering costs if markets are both imperfectly competitive andfees are at least partially nonsalient.

We find little offsetting response in terms of pricing. Usingthe same difference-in-differences approach, we find no evidenceof an anticipatory increase in interest charges prior to the CARDAct, and no evidence of a sharp or gradual increase following theCARD Act implementation periods. Our point estimate for theoffset is approximately zero, and we can rule out an offset ofgreater than 61% with 95% confidence. In addition, we find noevidence of an offsetting increase in interest charge for newaccounts, for which banks are less constrained in their ability toadjust contract terms, and no evidence of an increase in othersources of credit card income (e.g., interchange fees) or a reductionin measures of costs (e.g., marketing and operational expenses).3

Using the same difference-in-differences design, we estimatethat the CARD Act had a precise zero effect on credit limits andADB. We also estimate a zero effect on the number of newaccounts, although our standard errors are too large to preventus from ruling out meaningful effects in either direction. Takentogether, we interpret the results as demonstrating that regula-tion of hidden fees can bring about a substantial reduction inborrowing costs without necessarily leading to an offsettingincrease in interest charges or a reduction in access to credit.We think an alternative interpretation in which banks resistedraising interest rates because of concerns over future regulationis unlikely, since the industry’s main lobby group was advancingthe argument that the CARD Act raised interest charges andlowered access to credit (American Bankers Association 2013).

Our two years of post–CARD Act data do not allow us toinvestigate the longer run effects of the CARD Act on industryexit or entry, or effects on margins with multiyear contracts

3. These findings are consistent with statements by bank executives, such asJ.P. Morgan CEO Jamie Diamond, who argued that the CARD Act would cost hisbank up to $750 million in annual profits, and Bank of America, which said theregulations would cost it $800 million (Tse 2010).


(e.g., promotional agreements) or lumpy long-run investments(e.g., IT infrastructure and credit-scoring models). These ele-ments are important to consider in a complete benefit-cost anal-ysis of the regulation.

Our article contributes to a literature analyzing credit cardusage and pricing (Ausubel 1991; Calem and Mester 1995; Grossand Souleles 2002; Agarwal et al. 2006, 2014a; Kuchler 2013;Stango and Zinman 2013), the debate about regulating consumerfinancial products (Campbell 2006; Bar-Gill and Warren 2008;Mullainathan, Barr, and Shafir 2009; Campbell et al. 2011),and a body of research that analyzes the effectiveness of nudgesand default options in influencing consumer decision making(Madrian and Shea 2001; Thaler and Benartzi 2004; Choi,Laibson, and Madrian 2005; Carroll et al. 2009; Keys and Wang2014).

We also contribute to a small literature on the CARD Act,including Debbaut, Ghent, and Kudlyak (2013), who study therestriction on lending to borrowers under age 21, andJambulapati and Stavins (2013), who, like us, find no evidencethat banks closed credit card accounts or increased interest ratesin anticipation of the CARD Act. Our study complements work byKay, Manuszak, and Vojtech (2014), who find that banks wereonly able to offset about 30% of the decline in revenue from theDurbin Amendment cap on debit card interchange fees.

The rest of the article proceeds as follows. Section II providesbackground on the U.S. credit card industry and describes thekey provisions of the 2009 CARD Act. Section III describesthe data, and establishes important facts about profitability inthe pre–CARD Act period sample. Section IV describes our re-search design and approach to conducting inference. Section Vanalyzes the intended effects of the CARD Act, first examiningthe effect on fees and then turning to the effect of the disclosurenudge. Section VI examines possible unintended effects on inter-est charges and credit volume. Section VII concludes. All appen-dix material is in the Online Appendix.

II. Credit Cards and the 2009 CARD Act

II.A. A Primer on Credit Cards

Account holders use credit cards to make purchases and toborrow. When an account holder carries a balance from the


http://qje.oxfordjournals.org/lookup/suppl/doi:10.1093/qje/qju037/-/DC1



previous billing cycle, interest charges for the current cycle aregiven by

Interest Charges ¼ ADB�APR

365� Days in Billing Cycle;

where the right-hand side is the product of the ADB, defined asthe arithmetic mean of end-of-day balances over the billing cycle;the daily interest rate, defined as the annual percentage rate(APR) divided by 365; and the number of days in the billingcycle.4 Account holders who do not carry a balance into the cur-rent period have the possibility of repaying current period pur-chase volume without incurring interest charges. If an accountholder pays off her purchase volume completely, interest chargestypically fall within a grace period and are not assessed by thebank. If the account holder does not pay her balance in full, she ischarged interest starting from the date of purchase. Accountholders that fall under the grace period have no ADB in our data.

Credit limits place an upper bound on consumer purchasesand borrowing. Consumers who exceed their credit limit might beassessed an over-limit fee or can have transactions declined.Credit cards have a number of other fees that we discuss later.Credit card borrowing is not secured by collateral, though lendersmay garnish wages or seize assets of account holders who default.Recovery rates are low, in part because credit card debt is juniorto all forms of secured borrowing. To manage and price this de-fault risk, most credit card issuers screen applicants using bothFICO scores and internally generated risk measures. Creditcards are marketed to consumers through a number of channels,including direct mail and TV advertisements. Credit cards oftenprovide consumers with cash back or reward points, which scalewith purchase volume.

For some of our analysis, we compare outcomes for consumercredit cards (referred to in the industry as general purpose creditcards) and small business credit cards. These types of cards are

4. The APR measure does not account for compounding. For instance, a con-sumer with an APR of 15% who carries an ADB of $1,000 for 12 consecutive 30-daymonths would have her balance grow to $1,158 = 1,000� 0:15

365 � 30þ 1� �12

instead of$1,150 = 1,000� (0.15 + 1). In the past, some credit card issuers used a method knownas double-cycle billing to calculate interest payments. This method calculated ADBsover two cycles, rather than just considering the current cycle. Double-cycle billingsometimes added significant interest charges to customers whose average balancevaried greatly from month to month. The CARD Act banned this practice.


institutionally similar. Like consumer credit cards, small busi-ness credit cards are guaranteed by the personal financialassets of the account holder, and applicants for both types ofcards are screened on their personal FICO scores.5 Like consumercards, small business cards have features such as rewards andare marketed by direct mail and TV advertisements. Because ofthe similarities in underwriting and promotion, consumer andsmall business credit cards are issued by the same businessunits of most banks, and regulators conduct joint assessmentsof consumer and small business credit card lending. The maindifference between consumer and small business cards is thatsmall business account holders must claim to use their cards forbusiness and commercial purposes only. Banks in most circum-stances do not monitor this behavior, and survey evidence indi-cates that a significant percentage of charges are used forpersonal expenditure (Federal Reserve Board of Governors 2010).

II.B. The 2009 CARD Act

The CARD Act of 2009, was introduced in the 111th U.S.Congress (H.R. 627).6 On April 30, 2009, it passed the Housewith a majority vote of 357 to 70. The Senate passed an amendedversion of the bill on May 19, 2009, also with an overwhelmingmajority (90 to 5). President Barack Obama signed the bill intolaw on May 22, 2009.

The CARD Act primarily amended the Truth in Lending Actand instituted a number of new consumer protection and disclo-sure requirements for consumer credit cards. The regulation ex-cluded small business credits cards.7 The provisions of the CARDAct were scheduled to take effect in three phases between August20, 2009, and August 22, 2010.

5. Small business cards are also secured by firm assets. However, for manysmall business account holders, personal assets such as home and vehicle equitytend to be much more important than business assets.

6. Congress had previously drafted consumer financial regulation that in-cluded many of the same provisions as the CARD Act. The most recent attemptwas known as the Credit Cardholders’ Bill of Rights Act of 2008 and was introducedin the 110th Congress as H.R. 5244. The bill passed by a 312 to 112 vote in the Housebut was never given a vote in the Senate.

7. Legislation has recently been proposed to extend the CARD Act provisionsto the small business category. The Small Business Credit Card Act of 2013 (H.R.2419) amends the CARD Act to include small business credit cards. It was intro-duced in the House Financial Services Committee on June 18, 2013, but as of July2014 had not advanced.


1. Phase 1: August 20, 2009. The first phase of the CARD Actrequired banks to provide consumers with 45-day advance noticeof rate increases or other significant changes to terms and condi-tions. It also required lenders to (i) inform consumers in the samenotice of their right to cancel the credit card account before theincrease or change goes into effect, and (ii) mail or deliver periodicstatements for credit cards at least 21 days before payment is due.

2. Phase 2: February 22, 2010. The bulk of CARD Act provi-sions came into effect on February 22, 2010. A key requirementwas that no fees could be imposed for making a transaction thatwould put the account over its credit limit unless the cardholderexplicitly opts in for the credit card company to charge such a fee.Credit card companies could choose to either decline such trans-actions or process them without charging a fee. Furthermore, anover-limit fee could be imposed only once during the billing cyclein which the limit is exceeded. The CARD Act also introducedregulation detailing repayment disclosures required in monthlycredit card statements. In particular, it required statements todisplay the following:

(i) The number of months and the total cost to the consumer(including principal and interest) that it would take to paythe outstanding balance, if the consumer pays only therequired minimum payments and no further advancesare made;

(ii) The monthly payment amount that would eliminate theoutstanding balance in 36 months, if no further advancesare made, and the total cost to the consumer, includinginterest and principal payments, of repaying the full bal-ance in this way.

Figure I provides an example of the way credit card statementsdisplay this information.8

The CARD Act also regulated the issuance of credit cards toborrowers below age 21 and included a restriction on interest rateincreases for new transactions within the first year of opening the

8. This new information might be less salient for individuals who do most oftheir banking online. Rather than alerting consumers to the minimum paymentwarnings when they log on, many banks limited the changes to monthly state-ments, which means cardholders have to view a PDF copy of their full statementsto see the minimum payment warnings.


account. Furthermore, it limited the application of increasedrates to existing balances, except if the prior rate was temporary(e.g., an introductory rate), lasting at least six months, or if theminimum payment has not been received for 60 days. For cardswith multiple interest rates (e.g., a balance transfer and a newpurchase rate), issuers were required to apply payments to thehighest-rate balances first. Finally, the CARD Act regulated pay-ment due dates and times.9

3. Phase 3: August 22, 2010. The third phase of the CARD Actfurther regulated the fees banks can charge by requiring them tobe ‘‘reasonable and proportional.’’ Under the new rules, a creditcard company generally cannot charge a late fee of more than $25unless one of the previous six payments was also late (in whichcase the fee may be $35). Second, the late fee cannot be largerthan the minimum payment. Similarly, over-limit fees werecapped at the actual over-limit amount. An additional provisionprevented issuers from charging more than one penalty fee perviolation in a single billing period. The CARD Act also prohibitedthe charging of inactivity fees for not using the credit card for aperiod of time. Finally, it required lenders to reevaluate any newrate increases every six months.10

FIGURE I

Payoff Disclosure

Figure provides an example of the disclosure statement on monthly creditcard reports required by the CARD Act.

9. Credit card issuers are no longer allowed to set arbitrary deadlines for pay-ments, andmust accept payments received before 5 p.m. on the payment due date. Ifpayments are due on a day during which lenders do not receive payments by mail(including weekends and holidays), a payment received on the next business daycannot be treated as late.

10. The regulation requires, ‘‘If a creditor increases the annual percentage rateapplicable to a credit card account . . . based on factors including the credit risk of theobligor, market conditions, or other factors, the creditor shall consider changes in


III. Data and Pre–CARD Act Industry Overview

III.A. Data

Our main source of data is the Credit Card Metrics (CCM)data set assembled by the OCC. The OCC supervises and regu-lates nationally chartered banks and federal savings associa-tions. In 2008, the OCC initiated a request to the nine largestbanks that issue credit cards to submit data on general purpose,private label, and small business credit cards. The purpose of thedata collection was to have more timely information for banksupervision.

The CCM data set has two components. The main data setcontains account-level information on credit card utilization (e.g.,purchase volume, ADB), contract characteristics (e.g., interestrates, credit limits), charges (e.g., interest, assessed fees), andperformance (e.g., charge-offs, days overdue) for the near-uni-verse of credit card accounts at these banks. The second dataset contains portfolio-level information for each bank on itemssuch as operational costs and fraud expenses for the credit cardportfolio managed by the bank. Both data sets are submittedmonthly. Reporting started in January 2008 and continuesthrough the present, although the reporting in the first fewmonths of 2008 is incomplete. Due to mergers and other reportingissues, we observe entry and exit of banks during the time period.

To obtain a balanced panel of banks while maintaining asufficiently wide window around the CARD Act implementationdates, we drop a small bank that enters and exits the sample andrestrict our time period to March 2008 to December 2011. We alsorestrict attention to general purpose and small business creditcard accounts.11 Online Appendix Table A.I presents an overview

such factors in subsequently determining whether to reduce the annual percentagerate for such obligor’’ (12 CFR 226.59). In addition, lenders are required to ‘‘reducethe annual percentage rate previously increased when a reduction is indicated bythe review.’’ In other words, increases in APRs as a result of increases in the FederalFunds rate will have to be reversed if the Federal Funds rate subsequently declines.The regulation also requires that existing customers have interest rates adjusted tothe rate that otherwise similar new customers would be assessed.

11. Our sample does not include private label cards, which can only be used atthe issuing retailer’s stores, but does include affinity and co-branded cards. Weexclude cards from portfolios purchased from third parties, a very small numberof joint credit card accounts backed by more than one individual, and an equallysmall number of secured credit cards. We also drop accounts that do not report aFICO score at origination.



of our sample by reporting quarter. The sample contains datafrom eight banks and covers 150 million consumer accounts and7 million small business accounts in a typically quarter.12

Table I shows annualized summary statistics for keyaccount-level variables for consumer and small business accounts.For the combined sample, the average account carries an ADB of$1,347 and has an annualized purchase volume of $2,138.Multiplying by the number of accounts implies that these dataaccount for about $210 billion in ADB, or 30% of total outstandingU.S. credit card debt over this period.13 The average accountholder pays about $168 in interest charges a year and incurs $58in fees, of which late fees, over-limit fees, and annual fees are thelargest components. Banks charge off an average $175 per accountper year and recover $6 per account per year, or 3.7% of charge-offs.14 We use the term ‘‘net charge-offs’’ to indicate total charge-offs minus recoveries.

Table II shows account-level averages for the variables thatwe construct using the portfolio data and data from other sources.(See Online Appendix A for details on these calculations.)Interchange fees are charged to merchants for processing creditcard transactions and scale with purchase volume. We assessaccount-level interchange income as a constant 2.0% of purchasevolume, or $43 per account annually. Reward and fraud expensescorrespond to about 1.4% of purchase volume on average, or ap-proximately $30 per account per year. We calculate operationalcosts as a percentage of ADB by month in the portfolio data andestimate account-level operational costs assuming they scale pro-portionally with ADB. Operational costs are $50 per account peryear.15 Banks report the total interest expenses for funding their

12. We do not restrict the analysis to a balanced panel of accounts, becausedoing so would require us to drop accounts, for example, that were closed in mid-sample due to delinquency, and thereby create sample selection bias.

13. ADB are somewhat higher for small business accounts than for consumeraccounts, although much of this is driven by a higher average FICO score for smallbusiness accounts. Figure VII, Panel D shows that conditional on the FICO scorebeing below 660, which composes our primary treatment sample, ADB are verysimilar across consumer and small business accounts.

14. ‘‘Charge-offs’’ refer to an expense incurred on the lender’s income statementwhen a debt is considered long enough past due to be deemed uncollectible. For anopen-ended account such as a credit card, regulatory rules usually require a lenderto charge off balances after 180 days of delinquency.

15. About 15% of total operational costs (an annualized 0.5% of ADB) are mar-keting and customer acquisition expenses. Extrapolating this to the industry level



TA

BL

EI

SU

MM

AR

YS

TA

TIS

TIC

SU

SIN

GA

CC

OU

NT-L

EV

EL

DA

TA

(AN

NU

AL

IZE

D$

PE

RA

CC

OU

NT)

Com

bin

edsa

mp

leC

onsu

mer

cred

itca

rds

Sm

all

bu

sin

ess

cred

itca

rds

Mea

nS

td.

dev

.M

ean

Std

.d

ev.

Mea

nS

td.

dev

.

Aver

age

dail

ybala

nce

1,3

46.5

52,8

87.5

21,3

19.5

52,8

40.5

61,9

22.7

63,7

05.3

9P

urc

hase

vol

um

e2,1

38.0

710,6

75.0

31,9

46.9

09,4

04.0

26,2

17.8

125,3

20.9

9In

tere

stch

arg

es168.0

5448.2

9165.3

1438.1

1226.5

1624.3

1T

otal

fees

57.6

4230.7

757.7

2229.5

655.8

6255.3

5A

nn

ual

fee

6.4

763.9

56.5

864.3

24.2

055.5

7B

ala

nce

tran

sfer

3.9

6124.9

33.9

8125.6

73.5

8107.9

4C

ash

ad

van

ce4.1

985.6

74.3

286.7

11.4

359.3

3D

ebt

susp

ensi

on8.1

061.9

48.3

762.8

22.1

537.9

7L

ate

fee

24.3

9104.0

624.1

8101.2

628.8

3151.9

2N

otsu

ffici

ent

fun

ds

0.5

418.2

00.5

317.3

00.7

031.7

5O

ther

fees

1.4

852.1

21.4

650.1

31.8

483.9

2O

ver

lim

it8.5

161.6

88.2

960.0

113.1

390.1

3T

otal

charg

e-of

fs174.8

24,6

19.8

3167.6

54,5

35.3

8327.8

86,1

50.9

8P

rin

cip

al

144.1

44,2

43.9

9138.0

64,0

99.3

1273.8

46,6

14.6

9In

tere

stan

dfe

es30.6

8710.9

029.5

9669.3

454.0

41,3

14.9

7R

ecov

ery

6.4

9469.3

96.3

4453.0

49.7

1736.4

1R

epaym

enta

Min

imu

mor

less

(%)

26.1

26.2

23.9

Fu

llor

mor

e(%

)30.2

29.4

43.9

Not

es.

Valu

esare

calc

ula

ted

usi

ng

all

acc

oun

t-m

onth

sin

the

sam

ple

per

iod

.S

eeO

nli

ne

Ap

pen

dix

Afo

rm

ore

det

ail

s.E

xce

pt

for

AD

Ban

dre

paym

ent

stati

stic

s,valu

esare

an

nu

ali

zed

.T

he

sam

ple

per

iod

isA

pri

l2008–D

ecem

ber

2011.

All

vari

able

sare

infl

ati

on-a

dju

sted

to2012

usi

ng

the

CP

I-U

.aT

hes

est

ati

stic

sare

calc

ula

ted

onth

esa

mp

leof

acc

oun

t-m

onth

sw

ith

pos

itiv

ecy

cle-

end

ing

bala

nce

s.



credit card liabilities at the portfolio level by month. These ex-penses scale with ADB and vary over time. Over the sampleperiod, funding the average account’s credit card receivables forone year costs banks $22.

III.B. Pre–CARD Act Industry Overview

In this section we analyze data on average credit card issuerincome, costs, and profits across the FICO score distribution forthe pre–CARD Act period (April 2008 to January 2010). Table IIIshows key summary statistics on account-level credit card utili-zation and profitability, grouped by FICO score at account origi-nation.16 About 29.9% of accounts have FICO scores below 660,and 26.8% of accounts have FICO scores of 760 or higher.

Table III, Panel A describes credit card capacity and utiliza-tion. Credit limits increase from $2,025 for account holders withFICO scores below 620 to $12,400 for borrowers in the 760–799range, and then tail off moderately. ADB are hump-shaped in

TABLE II

SUMMARY STATISTICS USING PORTFOLIO-LEVEL DATA (ANNUALIZED $ PER ACCOUNT)

Mean Note on construction

Interchange income 42.76 2% of purchase volumeRewards + fraud expense 29.93 1.4% of purchase volumeCost of funds 22.49 Share of ADB (time varying)Operational costs 49.70 Share of ADB (time varying)

Collection 5.81 Share of ADB (time varying)Marketing + acquisition 7.31 Share of ADB (time varying)Other operational cost 36.58 Share of ADB (time varying)

Notes. Variables are constructed by combining account-level measures of ADB and purchase volumewith information from the portfolio-level data. Values are calculated using all account-months in thesample period. Operational expenses include expenses for marketing and acquisition, collections, servic-ing, card-holder billing, processing payments, card issuing and administration. See note on constructionand Online Appendix A for more details. The sample period is April 2008–December 2011. All variablesare inflation-adjusted to 2012 using the CPI-U.

suggests total industry advertising spending of about $3.75 billion. This level ofspending is consistent with numbers reported in Consumer Financial ProtectionBureau (2013b).

16. We use FICO score at account origination to avoid the reverse causality thatcould arise if an account is assigned a low FICO score precisely because it missed apayment and now has to pay a late fee. Using FICO score at origination introducessome measurement error if the object of interest is profitability by contemporane-ous FICO score.



TA

BL

EII

I

PR

E-C

AR

DU

TIL

IZA

TIO

NA

ND

RE

AL

IZE

DP

RO

FIT

SB

YF

ICO

SC

OR

E

FIC

Osc

ore

ran

ge

Tot

al

<620

620–659

660–719

720–759

760–799

800+

Per

cen

tof

acc

oun

ts100.0

17.3

12.6

24.6

18.6

19.2

7.6

Pan

elA

:ca

paci

tyan

du

tili

zati

on(a

nn

uali

zed

$p

eracc

oun

t)

Cre

dit

lim

it8,0

42

2,0

25

3,5

46

7,7

81

11,1

56

12,4

00

11,3

90

AD

B1,4

10

804

1,4

69

2,0

29

1,7

97

1,1

10

486

Pu

rch

ase

vol

um

e1,8

20

730

1,0

19

1,6

51

2,3

06

2,8

92

2,2

82

Pan

elB

:re

ali

zed

pro

fits

(%of

AD

B)

Tot

al

inco

me

25.0

45.7

31.5

21.0

16.9

17.1

19.9

Inte

rest

charg

es14.3

20.6

19.2

15.2

11.8

9.3

7.6

Tot

al

fees

7.6

23.3

10.9

4.1

2.5

2.4

2.9

Inte

rch

an

ge

inco

me

3.2

1.8

1.5

1.7

2.6

5.4

9.5

Tot

al

cost

s23.4

37.8

30.2

22.5

17.2

15.6

16.8

Net

charg

e-of

fs15.6

30.8

23.4

15.8

9.7

6.3

4.7

Cos

tof

fun

ds

2.3

2.4

2.3

2.2

2.2

2.2

2.2

Rew

ard

san

dfr

au

d2.2

1.3

1.0

1.2

1.8

3.7

6.5

Op

erati

onal

cost

s3.4

3.4

3.4

3.4

3.4

3.4

3.4

Col

lect

ion

0.4

0.4

0.4

0.4

0.4

0.4

0.4

Mark

etin

g+

acq

uis

itio

n0.5

0.5

0.5

0.5

0.5

0.5

0.5

Oth

erop

erati

onal

cost

2.5

2.5

2.5

2.5

2.5

2.5

2.5

Rea

lize

dp

rofi

t1.6

7.9

1.3

�1.6

�0.2

1.5

3.1

Not

es.

Table

show

sin

com

ean

dco

stco

mp

onen

tsfo

rco

nsu

mer

acc

oun

tsd

uri

ng

the

pre

–C

AR

DA

ctp

erio

d,

defi

ned

as

Ap

ril

2008–Jan

uary

2010.

Th

efi

rst

colu

mn

show

saver

ages

over

all

acc

oun

ts;

the

oth

erco

lum

ns

show

aver

ages

by

FIC

Osc

ore

at

acc

oun

tor

igin

ati

on.

Net

charg

e-of

fsare

the

sum

ofp

rin

cip

al

an

din

tere

st/f

ees

charg

e-of

fsm

inu

sre

cover

ies.

Pan

elB

show

svalu

esas

an

nu

ali

zed

per

cen

tof

aver

age

dail

ybala

nce

s.A

lld

olla

rvari

able

sare

infl

ati

on-a

dju

sted

to2012

usi

ng

the

CP

I-U

.


FICO score, rising from $804 for borrowers with FICO scoresbelow 620 to $2,029 for borrowers in the 660–719 range, beforefalling to $1,110 or less for account holders with FICO scoresabove 760. Purchase volume rises over much of the FICO scoredistribution, increasing from an annualized $730 for accountholders with a FICO score below 620 to $2,892 for account holdersin the 760–799 range. Overall, the share of people using creditcards to borrow rather than facilitate transactions is declining inFICO score.

We next examine components of profitability by FICO score.To compare across different components of profits, we report allvariables as an annualized percentage of ADB. For example,given monthly data on total fees and ADB, we calculate

Total fees as an annualized percentage of ADB¼Total fees

ADBþ1

� �12

�1:

For an account holder with a constant interest rate, interestcharges as an annualized percentage of ADB is simply the inter-est rate. Our measure can thus be interpreted as an interest rateequivalent for different components of income and costs.17

We define profits for a credit card account as the differencebetween total income and total costs. Total income for an accountis the sum of interest payments, fee payments, and interchangefees. The most basic measure of total costs includes realized netcharge-offs, the cost of funds, rewards and fraud expenses, andoperational costs. We call this measure realized costs.

Table III, Panel B examines the components of profits as apercentage of ADB. Borrowers with a FICO below 620 payan annualized 20.6% of ADB in interest charges and 23.3% ofADB in total fees. Interest charges decline modestly by FICOscore; total fees decline precipitously. Interchange income isnot quantitatively important, except for the highest FICOscore borrowers, who generate interchange income of more

17. We use ADB as the common denominator to normalize outcomes acrossaccounts with different levels of activity. An alternative approach would be to nor-malize outcomes as an annualized percent of purchase volume. Because inter-change income scales with purchase volume, the resulting measure could beinterpreted as converting our outcomes into interchange income equivalents.This approach seems less natural, because interchange income makes up only asmall fraction of total revenue. In addition, our results that low FICO score accountholders pay large fees would be even more extreme if we normalized by purchasevolume instead of ADB.


than 9.5% of ADB.18 Figure II, Panels A and B show plots of in-terest charges and fees by FICO score.

The main component of realized costs is net charge-offs.During our time period, account holders with FICO scoresbelow 620 incurred annualized net charge-offs of 30.8% of ADB,and account holders with FICO scores of 760 and above incurrednet charge-offs of less than 6.3%. Similar to interchange income,rewards and fraud costs as a share of ADB are larger for higherFICO account holders, who generate more purchase volume perunit of borrowing. The cost of funds as a share of ADB is relativelylow, at about 2.3%. Figure II, Panels C and D show plots of netcharge-offs and interchange income net of rewards expenses byFICO score.19

The income and costs data combine to produce a U-shapeddistribution of realized profits by FICO score. Account holderswith FICO scores below 620 generated realized profits of 7.9%of ADB. Realized profits bottom out at �1.6% of ADB for accountswith FICO scores in the 660–719 range. They rise to above 1.5%for accounts with the highest FICO scores. Figure II, Panels Eand F plot realized profits and the number of accounts by FICOscore.

Although realized profits do not account for ex ante risk, andwe have insufficient time-series data to estimate risk premiaacross the FICO distribution, the data suggest that credit cardswere a very profitable segment of the banking industry. Table IIIshows that across all FICO scores, realized profits averaged 1.6%of ADB in the pre–CARD Act period. Adjusting for taxes and bankleverage, this translates into a return on equity in excess of 10%,more than five times larger than the average financial sectorreturn on equity during this time period. (See Online AppendixA.1.4 for details.) Indeed, at the same time that bank divisionsmaking subprime home loans were losing large amounts of

18. This is not surprising given the ratio of purchase volume to average dailybalances for different FICO score groups. The highest FICO score account holdersprimarily use credit cards to facilitate transactions, not to borrow. Hence, inter-change income relative to overall receivables managed by the bank increases sig-nificantly as FICO rises.

19. We do not show the cost of funds and operational expenses because they aredefined as a constant fraction of ADB.




51

01

52

02

5In

tere

st C

harg

es (

% o

f A

DB

)

480 520 560 600 640 680 720 760 800 840

FICO Score

01

02

03

04

0T

ota

l F

ees (

% o

f A

DB

)

480 520 560 600 640 680 720 760 800 840

FICO Score

Interest Charge Income

Fee Income

A

B

FIGURE II

Pre–CARD Act Profit Components by FICO Score

Figure shows revenue and cost components, realized profits, and number ofaccounts for consumer credit cards by FICO score at origination binned ingroups of five. Revenue and cost components and realized profits are measuredas an annualized percentage of ADB. Number of accounts are per reportingmonth. Realized profits are the difference between revenues (interest charges,fees, and interchange income) and costs (net charge-offs, cost of funds, opera-tional expenses, and fraud and rewards expenses). The sample is restricted tothe pre–CARD Act period, defined as April 2008–January 2010.

(continued)


01

02

03

04

05

0

Net C

harg

eoffs (

% o

f A

DB

)

480 520 560 600 640 680 720 760 800 840

FICO Score

Interchange Income - Rewards Expenses

Net Chargeoffs

01

23

4

Inte

rch

an

ge

- R

ew

ard

s (

% o

f A

DB

)

480 520 560 600 640 680 720 760 800 840

FICO Score

C

D

FIGURE II

(continued)


-50

51

01

5

Re

aliz

ed

Pro

fits

(%

of A

DB

)

480 520 560 600 640 680 720 760 800 840

FICO Score

Number of Accounts

Realized ProfitsE

F

01

23

4

Nu

mb

er

of A

cco

un

ts (

mill

ion

s)

480 520 560 600 640 680 720 760 800 840

FICO Score

FIGURE II

(continued)


money, credit card issuers were earning their largest profits fromthe subprime segment of the market.20

IV. Research Design

Having established basic facts about the importance of feerevenue in the pre–CARD Act period, we turn to evaluatingthe intended and unintended consequences of the consumer pro-tections that were implemented by the CARD Act. The empiricalchallenge is that the CARD Act was introduced shortly after thefinancial crisis, when there was much instability in the macro-economy, which complicates the interpretation of a simple event-study analysis.

IV.A. Identification Strategy

We estimate the effects of the CARD Act using a difference-in-differences research design, where we compare outcomes forconsumer credit cards (treatment group) and small businesscredit cards (control group) during the different phases of theCARD Act implementation. The role of the control group is toestablish a counterfactual of what would have happened to con-sumer credit cards if the law had not been implemented. Theidentifying assumption is that, in the absence of the CARD Act,outcomes for consumer credit cards and small business cardswould have maintained parallel trends.

We argue that the parallel trends assumption is likely to bevalid for two reasons. First, we show that outcomes for consumerand small business credit cards move together in the pre–CARDAct period, with parallel trends in fees, interest charges, andmeasures of credit volume, such as ADB, credit limits, and thenumber of new accounts. We also show that conditional on FICOscores, consumer and small business accounts look similar onobservable characteristics such as credit limits and interestcharges.

The second reason is that consumer and small business cardsare institutionally similar. As we discuss in Section II, both types

20. The data show that earnings increased substantially as the economy recov-ered from the Great Recession. By the end of our sample in December 2011, annu-alized realized profits had increased from 1.6% of ABD to 5.2% of ADB, largely dueto a decline in annualized net charge-offs from a peak of 18.8% of ADB in May 2009to 7.8% of ADB in December 2011.


of cards are guaranteed by the personal financial assets of theaccount holder, and applicants for both cards are screened ontheir personal FICO scores. Both types of cards are issued bythe same business unit of most banks, and regulators conductjoint assessments of consumer and small business credit cardlending. The main difference between these types of cards isthat small business account holders must claim to use theircards for business purposes only, although banks in most circum-stances do not monitor this behavior (Federal Reserve Board ofGovernors 2010).

IV.B. Econometric Model

We specify the econometric model at the account level. Sinceour panel data set of 160 million accounts over 45 months hasmore than 7 billion observations, we estimate the model on datacollapsed to means for each bank�product type�FICO scoregroup�month.21 A product type is defined as the interaction ofa consumer card indicator and whether the card is co-branded, oiland gas, affinity, student, or ‘‘other.’’ FICO score groups are< 620, 620–659, 660–719, 720–759, 760–799, and � 800. Weshow in Online Appendix B that regressions using these collapseddata recover the parameters of interest from the account-levelspecification.

Our baseline econometric model is a difference-in-differencesspecification. Let yit be an outcome for account i in month t. Theregression specification is given by

yit ¼ �t þ �C1i2Consumer þ �1 1i2Consumer � 1t2Phase 2

þ�2 1i2Consumer � 1t2Phase 3 þ X 0it�X þ �it;ð1Þ

where �t are month fixed effects, 1i2Consumer is a treated indicatorthat takes a value of 1 if the account is a consumer credit cardaccount, Xit is a vector of possibly time-varying covariants, and �itis the error term that we assume is uncorrelated with unobserveddeterminants of the outcome. The indicator 1t2Phase 2 takes a valueof 1 for the months between the implementation of phase 2 and

21. In the regressions, we weight the group-level observations to allow us tointerpret the resulting estimates as aggregate effects. In particular, when the de-pendent variable is denominated as an annualized percentage of ADB (e.g., over-limit fees as an annualized percentage of ADB), we weight by total ADB in eachgroup. For other dependent variables (e.g., credit limits), we weight by the numberof accounts in each group.



the implementation of phase 3 (March 2010–August 2010), andthe indicator 1t2Phase 3 takes a value of 1 for the months after theimplementation of phase 3 (after August 2010).22 The time periodprior to phase 2 is the omitted group, so the coefficients can beinterpreted as the differential effect relative to the preimplemen-tation mean. For some outcome variables, we also include a con-sumer account� anticipation period interaction term to captureanticipatory responses that take place between the month the billwas passed and the month phase 2 came into effect (May 2009–February 2010). In these specifications, the time period prior tothe passage of the law is the omitted group and the coefficientsshould be interpreted relative to this period.23

In the Online Appendix we also present results from an al-ternative specification, where we allow for the coefficient on thetreatment group to evolve nonparametrically by month. Althoughthis specification has less statistical power than the baseline spe-cification, plotting the coefficients of interest over time allows usto establish whether there are spurious pretrends in the out-comes as well as examine the timing of the response to the law.The regression specification with treated�month-specific coeffi-cients is given by

yit ¼ �t þ �C1i2Consumer þX

t 6¼May 2009

�t1i2Consumer þ X 0it�X þ �it:ð2Þ

The coefficients of interest are the �t’s and the omitted group isMay 2009, the month the CARD Act was signed. In the plots, wenormalize the coefficient on May 2009 to the pre–CARD Act con-sumer account mean so that the effects can be interpreted rela-tive to this baseline level.

IV.C. Inference

We conduct statistical inference using two complementarystrategies. In our first approach, we construct standard errorsto account for (i) serial correlation in outcomes within accountsover time and (ii) correlation in outcomes across accounts thathave the same type of credit card and therefore have interest

22. We drop February 2010 and August 2010 from the sample because phase 2and phase 3 came into effect partway through these months.

23. We do not include a dummy variable for phase 1 of the CARD Act imple-mentation, since we do not study any of the provisions that came into effect onAugust 20, 2009.



rates and other contract characteristics jointly determined. Wespecify cluster-robust standard errors at the bank�product typelevel. We view this approach as conservative because many bankshave multiple types of co-branded cards, for example, with con-tract characteristics that are adjusted individually. The numberof product types varies across banks and yields no fewer than 46clusters in the regression specifications.

Our second approach is to construct p-values using a permu-tation test where we compare our estimate of the actual CARDAct to estimates of placebo reforms specified at other periods oftime (Conley and Taber 2011). To conduct inference on the over-limit fee restriction, which applied to consumer credit cardsduring the 22 months between February 2010 and the endof our sample, we assign placebo over-limit fee restrictions toconsumer credit cards in 22 randomly selected months drawnwithout replacement and estimate a placebo effect on thissample. We then compare the true effect of the fee restriction tothe distribution of placebo estimates derived from 1,000 ran-domly constructed samples. We use an analogous approach forthe other dependent variables.

V. Intended Effects

In this section, we examine the intended effects of the CARDAct. We first analyze the effects of the fee regulations and thenturn to considering the effects of the disclosure nudge.

V.A. Fees

The CARD Act had two primary elements that aimed to sig-nificantly reduce over-limit fees and late fees (see Section II.B).Figure III examines the effects of these regulations by plottingaverage fee revenue for consumer and small business accountsover time. In each plot, the vertical axis shows mean fee revenueas an annualized percentage of ADB. The horizontal axes showmonths, with the vertical bars in May 2009, February 2010, andAugust 2010 indicating the dates when the CARD Act was signed,and when phase 2 and phase 3 of the provisions came into effect.Because fee payments vary substantially by FICO score (seeSection III.B), we separately examine effects on accounts withFICO scores below and above 660 at origination, approximatelythe 30th percentile of the FICO score distribution in our data.


02

46

An

nu

aliz

ed

Ove

r L

imit F

ee

s (

% o

f A

DB

)

2008m1 2009m1 2010m1 2011m1 2012m1

Consumer Accounts Small Business Accounts

0.5

1

An

nu

aliz

ed

Ove

r L

imit F

ee

s (

% o

f A

DB

)

2008m1 2009m1 2010m1 2011m1 2012m1


A

B

Over Limit Fees: FICO < 660

Over Limit Fees: FICO ≥ 660

FIGURE III

Over-Limit, Late, and Total Fees

Figures show over-limit fees, late fees, and total fees as an annualizedpercentage of ADB for account holders with a FICO score less than 660 atorigination and a FICO score of at least 660 at origination. The sampleperiod is April 2008–December 2011. Vertical lines are plotted in May 2009,February 2010, and August 2010, the date when the bill was signed and thetwo key implementation dates of the CARD Act, respectively.

(continued)


02

46

81

0

Annualiz

ed L

ate

Fees (

% o

f A

DB

)

2008m1 2009m1 2010m1 2011m1 2012m1


0.5

11.5

2

Annualiz

ed L

ate

Fees (

% o

f A

DB

)

2008m1 2009m1 2010m1 2011m1 2012m1


C

D

Late Fees: FICO < 660

Late Fees: FICO ≥ 660

FIGURE III

(continued)


05

10

15

20

An

nu

aliz

ed

To

tal F

ee

s (

% o

f A

DB

)

2008m1 2009m1 2010m1 2011m1 2012m1


12

34

5

An

nu

aliz

ed

To

tal F

ee

s (

% o

f A

DB

)

2008m1 2009m1 2010m1 2011m1 2012m1


E

F

Total Fees: FICO < 660

Total Fees: FICO ≥ 660

FIGURE III

(continued)


Figure III, Panels A and B examine the effects on over-limitfees. In the pre–CARD Act period, over-limit fees for consumerand small business accounts move together, confirming the par-allel trends identifying assumption. In February 2010, when thelaw required consumer accounts to opt in to the processing ofover-limit transactions, over-limit fees for consumer accountsdrop to virtually zero. Over-limit fees for small business accounts,which were not affected by the CARD Act, trend smoothlythrough this implementation date.

Table IV shows the corresponding difference-in-differencesregressions for the effect on over-limit fees. Column (1) showsthe baseline specification that has consumer card by phase 2(March 2010–August 2010) and phase 3 (after August 2010) in-teraction terms and consumer card and month fixed effects. Thepre–February 2010 period is the omitted category, so that theeffects can be interpreted relative to the outcomes prior to theimplementation of the CARD Act. Column (2) addsbank�FICO score group fixed effects to this specification.Panel A shows outcomes for accounts with FICO scores below660, and Panel B shows outcomes with FICO scores above thislevel. The point estimates indicate that over-limit fees fell by 3.3percentage points and 0.3 percentage points for low and highFICO score accounts, respectively. The estimates are highlystable across specifications and statistically distinguishablefrom zero at conventional levels.

Figure III, Panels C and D examine the effects of the CARDAct on late fees. As with over-limit fees, there is no evidence ofdifferential pretrends for consumer and small business accounts.Late fees for consumer accounts decline in February 2010, whenrestrictions on which payments could be considered late becamemore stringent, and drop more sharply in August 2010, when the$25 maximum for late fees came into effect. The regression esti-mates for late fees, shown in columns (3) and (4) of Table IV,confirm these results. For accounts with FICO scores below660, late fees decline by 1.5 percentage points over both imple-mentation phases, from a pre–CARD Act mean of 5.9%. The dropfor high FICO score account holders is 0.3 percentage point off apre–CARD Act mean of 1.3%.

Some industry observers conjectured that credit card issuerswould respond to the loss in over-limit and late fee revenue byincreasing other fees, in particular annual fees. We examine thispossibility in columns (5) and (6) of Table IV by estimating


TA

BL

EIV

FE

ES

RE

ST

RIC

TIO

NS:

DIF

FE

RE

NC

E-I

N-D

IFF

ER

EN

CE

SR

EG

RE

SS

ION

S

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:%

ofA

DB

Over

-lim

itfe

esL

ate

fees

Oth

erfe

esT

otal

fees

Pan

elA

:F

ICO<

660

Con

sum

er�

ph

ase

2�

3.3

0**

�3.2

7**

�0.6

1�

0.5

7�

0.7

0�

0.6

2�

4.6

1**

�4.4

6**

(1.5

0)

(1.5

2)

(0.4

2)

(0.3

8)

(0.4

6)

(0.4

1)

(2.2

5)

(2.2

2)

[.03]

[.04]

[.15]

[.14]

[.13]

[.14]

[.05]

[.05]

Con

sum

er�

ph

ase

3�

3.3

7**

�3.3

3**

�1.4

1*

�1.4

7**

�0.5

7�

0.5

2�

5.3

5**

�5.3

2**

(1.5

2)

(1.5

3)

(0.7

6)

(0.6

9)

(0.4

5)

(0.3

7)

(2.6

7)

(2.5

5)

[.03]

[.03]

[.07]

[.04]

[.21]

[.17]

[.05]

[.04]

Pre

–C

AR

DA

ct,

con

sum

erm

ean

3.7

73.7

75.8

55.8

55.0

55.0

514.6

814.6

8R

-squ

are

d0.3

20.7

30.1

00.8

40.0

40.8

70.1

10.8

4N

um

ber

ofob

serv

ati

ons

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

Pan

elB

:F

ICO�

660

Con

sum

er�

ph

ase

2�

0.3

0**

*�

0.3

0**

*�

0.0

7�

0.0

8�

0.0

8�

0.0

7�

0.4

5**

*�

0.4

5**

*(0

.11)

(0.1

1)

(0.0

7)

(0.0

7)

(0.1

0)

(0.1

0)

(0.1

4)

(0.1

5)

[.01]

[.01]

[.32]

[.26]

[.43]

[.49]

[.00]

[.00]

Con

sum

er�

ph

ase

3�

0.2

5**

�0.2

6**

�0.2

5**

*�

0.3

0**

*0.0

60.0

7�

0.4

5**

�0.4

9**

(0.1

2)

(0.1

2)

(0.0

8)

(0.1

0)

(0.0

8)

(0.0

7)

(0.2

0)

(0.2

0)

[.04]

[.04]

[.00]

[.00]

[.46]

[.32]

[.03]

[.02]

Pre

–C

AR

DA

ct,

con

sum

erm

ean

0.4

00.4

01.2

71.2

71.5

01.5

03.1

73.1

7R

-squ

are

d0.2

30.6

90.1

10.7

30.1

90.5

20.1

60.7

2N

um

ber

ofob

serv

ati

ons

6,9

86

6,9

86

6,9

86

6,9

86

6,9

86

6,9

86

6,9

86

6,9

86


TA

BL

EIV

(CO

NT

INU

ED)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:%

ofA

DB

Over

-lim

itfe

esL

ate

fees

Oth

erfe

esT

otal

fees

Con

trol

s:all

pan

els

Main

effe

cts

Con

sum

erca

rdF

EX

XX

XX

XX

XM

onth

FE

XX

XX

XX

XX

Ad

dit

ion

al

covari

ate

sB

an

kF

E�

FIC

Osc

ore

gro

up

FE

XX

XX

Not

e:T

able

show

sco

effi

cien

tsfr

omd

iffe

ren

ce-i

n-d

iffe

ren

ces

regre

ssio

ns

that

com

pare

fees

for

con

sum

ercr

edit

card

s(t

reatm

ent

gro

up

)an

dsm

all

bu

sin

ess

card

s(c

ontr

olgro

up

)d

uri

ng

the

dif

fere

nt

ph

ase

sof

the

CA

RD

Act

imp

lem

enta

tion

.T

he

dep

end

ent

vari

able

sare

show

nas

an

an

nu

ali

zed

per

cen

tof

AD

Ban

dare

calc

ula

ted

inth

esa

me

man

ner

as

the

valu

esin

Pan

elB

ofT

able

III.

We

defi

ne

ph

ase

2as

Marc

h2010–A

ugu

st2010

an

dp

hase

3as

the

mon

ths

aft

erA

ugu

st2010.

Th

ep

erio

dp

rior

toF

ebru

ary

2010

isth

eom

itte

dgro

up

,so

the

coef

fici

ents

can

be

inte

rpre

ted

as

the

dif

fere

nti

al

effe

ctre

lati

ve

toth

ep

reim

ple

men

tati

onp

erio

d.

Th

esa

mp

lep

erio

dis

Ap

ril

2008–D

ecem

ber

2011.

Th

ere

gre

ssio

ns

are

esti

mate

don

data

aggre

gate

dto

the

ban

k�

pro

du

ctty

pe�

FIC

Osc

ore

gro

up�

mon

thle

vel

,an

dw

eigh

ted

by

tota

lA

DB

inea

chgro

up

.A

pro

du

ctty

pe

isd

efin

edas

the

inte

ract

ion

ofth

eco

nsu

mer

card

ind

icato

ran

dw

het

her

the

card

isco

-bra

nd

ed,

oil

an

dgas,

affi

nit

y,

stu

den

t,or

oth

er.

FIC

Osc

ore

gro

up

sare

:<

620,

620–659,

660–719,

720–759,

760–799,

an

d�

800.

Sta

nd

ard

erro

rscl

ust

ered

by

ban

k�

pro

du

ctty

pe

are

show

nin

pare

nth

eses

an

dth

eass

ocia

ted

p-v

alu

esare

show

nin

bra

cket

s.T

her

eare

46

such

clu

ster

sin

Pan

elA

an

d47

clu

ster

sin

Pan

elB

.S

ign

ifica

nce

level

s:*

p<

.10,

**p<

.05,

***

p<

.01.


difference-in-differences specifications with other fees, defined asall fees except late fees and over-limit fees, as the dependent var-iable. We find no evidence for an offsetting response in other fees.However, while we observe no offsetting medium-term increasein fee revenue, we cannot assess whether firms will respond byintroducing novel fees in the long run, as theorized in Heidhues,Ko00szegi, and Murooka (2012).

Figure IV, Panels E and F combine the analysis of the sepa-rate fee categories by showing the effects of the CARD Act on totalfees. The corresponding coefficient estimates, shown in columns(7) and (8) of Table IV, indicate that over the implementationphases, total fees dropped by 5.3 percentage points for lowFICO score accounts and by 0.5 percentage points for highFICO score accounts. Both estimates are statistically distinguish-able from zero with p-values of .04 and .02, respectively.

In the Online Appendix we show two additional pieces ofsupporting analysis. Appendix Figure A.IV plots the coefficientson consumer account�month interactions from difference-in-differences regression specifications where we allow for sepa-rate coefficients of interest by month (equation (2)). These figuresconfirm the lack of preexisting trends and the sharp response tothe implementation of the law. In Appendix Figure A.V we showresults from permutation tests where we compare our estimate ofthe actual effect of the CARD Act to the distribution of placeboestimates derived from 1,000 samples where ‘‘treatment’’ is ran-domly assigned (see discussion in Section IV). This alternativeapproach to conducting inference confirms that the drop in feerevenue was statistically significant.

The last part of this section examines the mechanisms un-derlying the observed drops in fee revenue. Did banks respond tothe virtual elimination of over-limit fees by more frequently de-clining over limit transactions? Figure IV, Panel A examines thisquestion by plotting the share of consumer and small businessaccounts with cycle-ending balances larger than the credit limitover time. This share of accounts declines for both types of ac-counts, most likely driven by the economic recovery. There is asmall additional decline for consumer credit cards around theFebruary 2010 CARD Act implementation date, suggesting thata small number of issuers chose to decline over limit transactionsfollowing the regulation. However, we continue to observe a sig-nificant number of accounts being allowed to process transactionsthat would take them over their credit limit.





04

81

2

Acco

un

ts O

ve

r-L

imit (

%)

2008m1 2009m1 2010m1 2011m1 2012m1


0.0

5.1

.15

.2

Sh

are

of

Acco

un

t-M

on

ths

0 50 100 150Account Utilization (% of Credit Limit)

Pre-CARD Act Post-CARD Act

A

B

Account Utilization: Percent Over Limit

Account Utilization for Consumer Accounts

FIGURE IV

Account Utilization and Late Payments

Figures show information on account utilization and late payments. PanelA shows the percentage of consumer and small business accounts with cycle-ending balances exceeding the credit limit in each month. Panel B shows thedistribution of cycle-ending balances as a share of the credit limit for consumercredit cards in the year before (dashed line) and after (solid line) the February2010 CARD Act implementation date.

(continued)


02

46

La

te F

ee

s In

cid

en

ce

(%

)

2008m1 2009m1 2010m1 2011m1 2012m1

$40 $25 $35 Other

05

10

15

La

te F

ee

s In

cid

en

ce

(%

)

2008m1 2009m1 2010m1 2011m1 2012m1


C

D

Late Payments: Percent Late

Late Fee Levels for Consumer Accounts

FIGURE IV

(continued)

Panel C shows the percentage of consumer and small business accountswith positive late fees in each month. Panel D shows the percent of consumeraccounts with late fees of $25, $35, $45, and an ‘‘other’’ category that combinesall other positive levels. The $25 category is defined as accounts with late feesin the $21–$25 range, the $35 category as accounts with late fees in the $31–$35 range, and the $40 category as accounts with late fees in the $36–$40range. The sample period is April 2008–December 2011. Vertical lines are plot-ted in May 2009, February 2010, and August 2010, the date when the bill wassigned and the two key implementation dates of the CARD Act, respectively.


Figure IV, Panel B further investigates this result by show-ing the distribution of account utilization, defined as the ratioof cycle-ending balances to credit limits, in the year before andafter the February 2010 implementation date. Even in thepost–CARD Act period, many credit card borrowers were al-lowed to exceed their credit limits, sometimes by substantialamounts. This evidence confirms that the vast majority of thedecline in over-limit fee revenue did not come from banks nolonger processing over-limit transactions, but was insteaddriven by banks no longer charging for processing thetransaction.

Figure III shows that the decline in late fee revenue appearsto diminish over time. To investigate this partial reversal, FigureIV, Panel C shows the overall incidence of late fees for consumerand small business accounts, defined as the share of accounts thatincurred a late fee. In the pre–CARD Act period, about 8% ofaccounts experienced a late fee each month. These values driftdownward over the sample period, with similar declines for con-sumer and small business accounts. In particular, the plot showsno effect of the August 2010 implementation date on the fre-quency of late payments. This suggests consumers did not re-spond to the reduction in the late fee ‘‘price’’ by increasing thefrequency of late payments.

Because the data show no decline in the incidence of latepayments, the drop in late fee revenue is attributable to achange in the fee amount. Figure IV, Panel D examines thisprice effect by plotting the frequency of late fees of differentdollar amounts. Recall that the CARD Act reduced late fees to$25 unless one of the previous six payments was also late, inwhich case a $35 fee was allowed. The plot provides evidence ofa one-for-one substitution from $40 to $25 late fees in August2010. The plot also shows a gradual substitution from $25 to$35 late fee amounts in the months after implementation,with the frequency of $35 late fees exceeding the frequency of$25 late fees by November 2010. Banks seem to have providedaccount holders with a ‘‘clean slate’’ in August 2010 with no latepayments in their six-month look-back period. Over time, as thefraction of account holders with a late payment in the previous sixmonths rose, banks were able to increase their fees to $35 forthose accounts, contributing to the observed partial reversal inlate fees.


This analysis shows that the CARD Act brought about asharp drop in fees, with drops of 5.3 percentage points of ADBfor accounts with a FICO score below 660 and 0.5 percentagepoints of ADB for accounts with a FICO score above this level.Low and high FICO score cards make up 23.0% and 77.0% ofborrowing in our data, so if we take a weighted average, we cal-culate a reduction of 1.6% as an annualized percentage of ADB.Given an outstanding U.S. credit card borrowing volume of $744billion in the first quarter of 2010 (Federal Reserve Bank of NewYork 2013), extrapolating to the entire market suggests theCARD Act’s fee regulation reduced borrowing costs by $11.9 bil-lion a year.

V.B. Payoff Nudge

In addition to regulating the fees that banks may charge,the CARD Act introduced rules requiring repayment disclosuresin monthly credit card statements. The aim of these disclosurerequirements was to provide information on the cost of onlymaking the minimum payment, as well as information on thereduction in interest payments that could be achieved bymaking payments that would eliminate the balance within 36months. Indeed, information such as the 36-month paymentamount might be understood by consumers as a payment recom-mendation or nudge, anchoring repayment at this level (Navarro-Martinez et al. 2011). However, it is not obvious whether such apayoff nudge would actually lead to a change in repaymentbehavior, both because it is unclear if the nudge would be suffi-ciently powerful and because prevailing repayment levels mightalready have been optimal.

1. Pre–CARD Act Payoff Behavior. We begin by documentingcredit card payoff behavior in the pre–CARD Act period. For thisanalysis, we restrict the sample to consumer credit cards with anonzero cycle-ending balance. Figure V, Panel A shows the shareof account holders making full payments by FICO score at ac-count origination. About 10% of borrowers with a FICO scorebelow 620 fully repay their balance at the end of the cycle.This share rises monotonically in FICO score, with about 25%of borrowers with a FICO score of 720 and about 75% of borrowerswith a FICO score above 800 making the full payment.


0.2

.4.6

.8

Sh

are

of A

cco

un

t-M

on

ths

480 520 560 600 640 680 720 760 800 840

FICO Score

.1.2

.3.4

.5

Sha

re o

f Acc

ount

-Mon

ths

480 520 560 600 640 680 720 760 800 840

FICO Score

A

B

Full Payoff

Minimum Payment or Less

FIGURE V

Pre–CARD Act Payoff Distribution

Figure shows payoff behavior by FICO score at origination in the pre–CARD Act period, defined as April 2008–January 2010. Panel A shows theshare of account-months making the full payment. Panel B shows the shareof account-months making the minimum payment or less. The sample excludesaccounts that have a zero cycle-ending balance.


On average, 30.1% of account holders pay their cycle-ending bal-ance in full.24

Figure V, Panel B shows the share of account holders makingminimum payments or less by FICO score. About 40% of bor-rowers with a FICO score below 620 pay the minimum or less.The number of account holders making the minimum payment orless declines monotonically in FICO score, with about 27% ofborrowers with a FICO score of 720 and 10% of borrowerswith a FICO above 800 making payments of the minimum orless. We calculate that on average 13.0% of borrowers onlymake the minimum payment, and 14.7% make payments of lessthan the minimum amount.

2. Payoff Nudge. The CARD Act mandates the disclosure ofthe monthly payment that would eliminate an account holder’scycle-ending balance if the account holder makes 36 equal-sizedpayments and avoids new purchases (see Figure I). Let T be thenumber of months it would take to pay off a cycle-ending balancefor a constant payment amount:25

T ¼ 1�ln 1� APR

12Cycle-Ending Balance�Payment

Payment

� �ln ð1þ APR

12 Þ:ð3Þ

Full repayment is indicated by T = 1. At the pre–CARD Actaverage APR of 16.5%, T = 5 implies a payment of 20.5% of thecycle-ending balance; T = 10 implies a payment of 10.6% of thebalance; and T = 83 implies a payment of 2% of the balance,which is a typical minimum payment amount in our data. TheCARD Act requires a disclosure of the payment that would resultin T = 36. At the average interest rate, this implies a payment of3.7% of the cycle-ending balance. Consumer credit accounts havea median T of 15.

24. This number is similar to estimates from other sources. For example, usingdata from the 1995 and 1998 Survey of Consumer Finances, Laibson, Repetto, andTobacman (2007) calculate that 32.2% of households pay their credit card bill in fulleach month. Note that our data are at the account level and the survey is a house-hold-level survey, so the numbers are not directly comparable.

25. This equation obtains from rearranging the standard monthly paymentformula for a series of payments starting in the current month:

Payment ¼APR12

1�ð1þAPR12 Þ�ðT�1Þ

� �ðCycle-Ending Balance� PaymentÞ.


Figure VI shows the distribution of months-to-payoff T in theyear before and after the CARD Act for consumer and small busi-ness cards, focusing on values of T in the vicinity of 36.26

Following the implementation of the CARD Act, there was asmall but significant increase in the share of consumer creditcard holders paying the 36-month payment amount. No suchchange can be detected for small business card holders, whowere not shown the payoff disclosure.

We use a difference-in-differences approach to quantify theimpact of the nudge, comparing the change in repayment behav-ior before and after the February 2010 implementation date forconsumer and small business credit cards. Table V showsthe results of these regressions. Columns (1) and (2) show theeffect on the share of account holders making payments close tothe 36-month value (30�T� 38).27 The results show that thenudge increased this share by a precisely estimated 0.4 percent-age point on a base of 5.3%.28

26. There is seasonality in repayment behavior, so analyzing the 12 monthsbefore and after the CARD Act is more representative than analyzing the entirepre–CARD Act and post–CARD Act periods. Online Appendix Figure A.VI showsthe full distribution of T in the year preceding the implementation of the CARD Actdisclosure requirement, for both consumer credit cards and small business creditcards.

27. The choice of this range, which is asymmetric around the CARD Act man-dated disclosure amount of T = 36, is motivated by an inspection of Figure VI, whichshows the shift of account holders to values of T within that range. We would expectincreases in a small range around 36 months for a number of reasons. The mostimportant is that we use the current interest rate for our calculation of T, whereasbanks are required to account for contractually determined changes in interestrates over the 36-month period. In particular, ‘‘if the interest rate in effect on thedate on which the disclosure is made is a temporary rate (such as an introductoryrate) that will change under a contractual provision applying an index or formulafor subsequent interest rate adjustment, the creditor is required to apply the inter-est rate in effect on the date on which the disclosure is made for as long as thatinterest rate will apply under that contractual provision, and then apply an interestrate based on the index or formula in effect on the applicable billing date’’ (15 U.S.Code §1637).

28. This estimate is relatively small compared to the experimental estimates inStewart (2009) and Navarro-Martinez et al. (2011). However, one caveat to inter-preting our results is that we are unsure which credit card users actually saw theCARD Act disclosures. For example, payment disclosures are only required onmonthly credit card statements, which are rarely viewed by individuals doingonline banking. This means that while we pick up the full effect of the CARD Actregulation, we might understate the possible effect of a regulatory design thatachieved wider awareness of the disclosures.



.00

5.0

06

.00

7.0

08

.00

9.0

1S

hare

of A

ccount-

Month

s

10 20 30 40 50 60Months to Payoff


.00

2.0

04

.00

6.0

08

.01

Sh

are

of A

cco

un

t-M

on

ths

10 20 30 40 50 60Months to Payoff


A

B

Months-to-Payoff: Consumer Accounts

Months-to-Payoff: Small Business Accounts

FIGURE VI

Months-to-Payoff (T)

Figure plots distributions of months-to-payoff (T) in the year before (dashedline) and after (solid line) the February 2010 CARD Act implementation date.Months-to-payoff (T) is the number of months it would take to pay off the cycle-ending balance if the account holder makes constant payments and makes no newpurchases, and is calculated using equation (3). We present the distribution for Tbetween 10 and 60. The share of account-months is top-coded at 1% to focus on thedistribution around the CARD Act target payoff amount (T = 36). Panel A showsthe distribution for consumer credit cards, Panel B for small business credit cards.


TA

BL

EV

PA

YO

FF

NU

DG

E:

DIF

FE

RE

NC

E-I

N-D

IFF

ER

EN

CE

SR

EG

RE

SS

ION

S

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

:p

erce

nt

ofp

aym

ents

Nu

dge

ran

ge

(30�

T�

38)

Les

sth

an

nu

dge

ran

ge

(T>

38)

Min

imu

mp

aym

ent

Con

sum

er�

ph

ase

20.3

9**

0.3

8**

0.7

5**

0.0

3�

0.1

50.4

6�

0.2

2�

0.2

5�

0.3

5(0

.19)

(0.1

7)

(0.3

4)

(1.2

3)

(1.0

7)

(1.2

9)

(0.3

6)

(0.4

0)

(0.4

7)

[.05]

[.03]

[.03]

[.98]

[.89]

[.72]

[.54]

[.54]

[.46]

Con

trol

sM

ain

effe

cts

Con

sum

erca

rdF

EX

XX

XX

XX

XX

Mon

thF

EX

XX

XX

XX

XX

Ad

dit

ion

al

covari

ate

sB

an

kF

E�

FIC

Osc

ore

gro

up

FE

XX

XX

XX

Sam

ple

rest

rict

ion

sE

xcl

ud

ing

acc

oun

tsw

ith

bala

nce<

$1,0

00

XX

X

Pre

–C

AR

DA

ct,

con

sum

erm

ean

5.3

15.3

17.1

935.3

535.3

547.7

812.9

512.9

514.9

7R

-squ

are

d0.0

80.8

70.7

80.0

60.9

00.8

80.0

60.8

70.8

8N

um

ber

ofob

serv

ati

ons

10,4

33

10,4

33

10,4

33

10,4

33

10,4

33

10,4

33

10,4

33

10,4

33

10,4

33

Not

es.

Table

show

ses

tim

ate

sfr

omd

iffe

ren

ce-i

n-d

iffe

ren

ces

regre

ssio

ns

ofp

ayof

fbeh

avio

rfo

rco

nsu

mer

cred

itca

rds

(tre

atm

ent

gro

up

)an

dsm

all

bu

sin

ess

card

s(c

ontr

olgro

up

)aro

un

dth

eim

ple

men

tati

onof

the

payof

fn

ud

ge.

Inco

lum

ns

(1)–

(3),

the

dep

end

ent

vari

able

isth

ep

erce

nt

ofacc

oun

th

old

ers

that

mak

ea

paym

ent

corr

esp

ond

ing

toth

eta

rget

CA

RD

Act

paym

ent

nu

dge

am

oun

t(3

0�

T�

38).

Inco

lum

ns

(4)–

(6),

the

dep

end

ent

vari

able

isth

ep

erce

nt

ofacc

oun

th

old

ers

that

mak

ep

aym

ents

small

erth

an

the

CA

RD

Act

paym

ent

nu

dge

am

oun

t(T>

38).

Inco

lum

ns

(7)–

(9),

the

dep

end

ent

vari

able

isth

ep

erce

nt

ofacc

oun

th

old

ers

that

mak

eex

act

lyth

em

inim

um

paym

ent.

Th

esa

mp

lep

erio

dis

Ap

ril

2008–

Dec

ember

2011.

Th

ere

gre

ssio

ns

are

esti

mate

don

data

aggre

gate

dto

the

ban

k�

pro

du

ctty

pe�

FIC

Osc

ore

gro

up�

mon

thle

vel

,an

dw

eigh

ted

by

the

nu

mber

ofacc

oun

tsin

each

gro

up

.A

pro

du

ctty

pe

isd

efin

edas

the

inte

ract

ion

ofth

eco

nsu

mer

card

ind

icato

ran

dw

het

her

the

card

isco

-bra

nd

ed,

oil

an

dgas,

affi

nit

y,

stu

den

t,or

oth

er.

FIC

Osc

ore

gro

up

sare

<620,

620–659,

660–719,

720–759,

760–799,

an

d�

800.

Inco

lum

ns

(3),

(6),

an

d(9

),th

esa

mp

leis

rest

rict

edto

acc

oun

tsw

ith

acy

cle-

end

ing

bala

nce

larg

erth

an

$1,0

00.

Sta

nd

ard

erro

rscl

ust

ered

by

ban

k�

pro

du

ctty

pe

are

show

nin

pare

nth

eses

an

dth

eass

ocia

ted

p-v

alu

esare

show

nin

bra

cket

s.T

her

eare

47

such

clu

ster

sin

the

sam

ple

.S

ign

ifica

nce

level

s:*

p<

.10,

**p<

.05,

***

p<

.01.


Account holders were not shown the 36-month paymentamount if this amount was lower than their minimum payment.This primarily occurred for accounts with low balances for whichthe minimum payment was set to a nominal lower bound. Incolumn (3), we restrict the sample to account holders with morethan $1,000 in cycle-ending balances. We find a larger response inthis subsample, with about 0.8% of these accounts shifting to the36-month payment amount.

Determining whether the nudge affected overall repaymentbehavior requires estimating whether the account holderswho shifted to the 36-month payment amount would have coun-terfactually been making higher or lower payments. Estimatingthis effect is difficult because the 0.4% shift is small relative tocyclical and seasonal variation in repayments. Columns (4)–(6)show the effect of the CARD Act on the share of account holdersmaking payments less than the nudge amount (T >38). A precise0.4 percentage point decline in this share would indicate that thenudge shifted account holders from lower payments to the 36-month amount. The estimates, however, are too imprecise toallow us to draw any conclusions. This is confirmed by the per-mutation tests presented in Online Appendix Figure A.VII, whichshow that the effect on the number of account holders makinglower payments is small relative to normal month-on-month var-iation in that number.

In principle, the CARD Act could have decreased overall pay-ments if the disclosure of the minimum payment amount hadan anchoring effect and increased payments at the minimumpayment level. This would be consistent with experimental evi-dence from Stewart (2009), who shows that presenting a mini-mum payment reduces overall willingness to repay. In columns(7)–(9) we test for the effect of the repayment nudge on the shareof people making exactly the minimum payment. Contrary to theanchoring hypothesis, the point estimates suggest a decline of 0.2percentage point in the share of accounts making the minimumpayment on a base of 13%, but the estimates are too imprecise toallow us to draw definitive conclusions.

Although we are unable to precisely estimate the exact effectof the nudge on overall repayment behavior, we can use our esti-mates to construct an upper bound of the effect of the nudge onannualized interest payments. Assume that the nudge (i) shiftedaccount holders from making no payment to paying at the36-month value and (ii) did not affect the cycle-ending balance



of these account holders. The one-month change in interest pay-ments for account holders that shift their behavior is given by theproduct of the change in the percentage of balance paid, the cycle-ending balance, and the monthly interest rate:

� Interest Payments ¼ � % of Balance Paid

�Cycle-Ending Balance�APR

12:ð4Þ

Account holders making no payment had a pre–CARD Actaverage cycle-ending balance of $2,957 and an average APR of21.7%. Plugging this average APR into equation (3) implies achange in the percent of balance paid from 0% to 3.7%. Takingthe product of these numbers and multiplying by 12 to annualizeyields an estimated $24.00 reduction in annualized interest pay-ments for account holders that shifted their repayment behaviorin response to the nudge.

Although a reduction in interest payments of this amountwould be nonnegligible for the account holders who shift theirbehavior, the fact that few account holders respond to thenudge leads us to estimate a small upper bound for the aggregateeffect. The estimate of $24.00 annualized savings for the 0.4% ofaccounts that switch translates to aggregate savings of 0.0076%

(=0:4%�$24:00$1;251

) of total ADB. If we extrapolate these results to the

$744 billion national credit card market, we estimate an upperbound for the nudge of $57 million (= $744 billion� 0.0076%) inannualized savings.

VI. Unintended Consequences

In this section, we assess possible unintended consequencesof the CARD Act, focusing on whether lenders responded tothe decline in fee revenue by increasing interest charges or byrestricting access to credit.

From a theoretical perspective, fees will be offset if (i) mar-kets are perfectly competitive or (ii) fees are perfectly salient.If markets are perfectly competitive, so that aggregate price in-clusive of all fees is equal to marginal cost, any regulation thatreduces a certain fee will be offset by a similarly sized increase inanother pricing dimension. If all fees and prices are perfectly sa-lient, then demand is only responsive to the aggregate price, and


firms will adjust other prices to keep the aggregate priceunchanged. This means that regulators can only be successfulin lowering aggregate borrowing costs when markets are imper-fectly competitive and fees are at least partially nonsalient. Wepresent a model that formalizes this argument in OnlineAppendix C, and we expand on this mechanism in Agarwalet al. (2014b).

In the empirical analysis that follows, we restrict our atten-tion to accounts with a FICO score below 660 at origination,which experienced a decline in fees of 5.3% of ADB. Since ac-counts with a FICO score of at least 660 only experienced a0.5% decline in fees, we would not expect an economically signif-icant offset for these account holders. In the Online Appendix, wereplicate our analysis on the sample of accounts with a FICOscore of 660 or higher.

VI.A. Interest Charges

Figure VII, Panel A shows mean interest charges as anannualized percent of ADB for consumer and small business ac-counts with a FICO score less than 660 at origination.29 Interestcharges for both types of cards move together in the pre–CARDAct period and continue to move together over the rest of themonths in the data. We find no evidence of anticipatory increasesfor consumer accounts after the May 2009 passage date and noevidence of any increase during the implementation periods.

Table VI shows the corresponding difference-in-differencesregressions. We show coefficients on consumer cards interactedwith indicators for the anticipation (June 2009–January 2010),phase 2 (March 2010–August 2010), and phase 3 (after August2010) time periods. The pre–May 2009 period is the omitted cat-egory, so that the effects can be interpreted relative to interestcharges prior to the passage of the CARD Act. Column (1) showsthe baseline specification with the interaction terms and con-sumer card and month fixed effects. Column (2) adds fixed effectsfor bank�FICO score groups. The regression estimates confirm

29. Interest charges are the total interest payments made by the borrower. Thisis a better measure for the cost of credit than APRs, because most accounts have anumber of different APRs applying to different types of transactions (e.g., balancetransfers, cash advances) that are in place simultaneously. Interest charges aggre-gate across these different APRs to provide a measure of the weighted average costof credit.





12

14

16

18

20

22

24

Annualiz

ed Inte

rest C

harg

es (

% o

f A

DB

)

2008m1 2009m1 2010m1 2011m1 2012m1


02

00

04

00

06

00

0

Cre

dit L

imit (

$)

2008m1 2009m1 2010m1 2011m1 2012m1


A

B

Interest Charges

Credit Limits

FIGURE VII

Interest Charges and Credit Volume: FICO< 660

Panel A shows interest charges as an annualized percentage of ADB. Panel Bshows credit limits. Panel C shows new accounts measured as a percentage of theaverage total pre–CARD Act number of accounts. Panel D shows average dailybalances. All panels focus on account holders with a FICO score below 660 atorigination, and display monthly averages for consumer and small businesscredit cards. The sample period is April 2008–December 2011. Vertical lines areplotted in May 2009, February 2010, and August 2010, the date when the bill wassigned and the two key implementation dates of the CARD Act, respectively.

(continued)


0.5

11

.52

Ne

w A

cco

un

ts (

% o

f P

re-C

AR

D A

ct L

eve

l)

2008m1 2009m1 2010m1 2011m1 2012m1


01

00

02

00

03

00

0

Ave

rag

e D

aily

Ba

lan

ce

($

)

2008m1 2009m1 2010m1 2011m1 2012m1


C

D

New Accounts

Average Daily Balances

FIGURE VII

(continued)


TA

BL

EV

I

INT

ER

ES

TC

HA

RG

EO

FF

SE

T:

DIF

FE

RE

NC

E-I

N-D

IFF

ER

EN

CE

SR

EG

RE

SS

ION

S(F

ICO<

660)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:%

ofA

DB

Inte

rest

charg

esO

ffse

tT

otal

inco

me

Cos

tsex

clu

din

gch

arg

e-of

fs

Con

sum

er�

an

tici

pati

on0.2

50.2

90.0

50.0

6�

0.3

70.0

1�

0.0

6�

0.0

5(1

.40)

(1.4

0)

(0.2

6)

(0.2

6)

(1.5

1)

(1.4

7)

(0.1

7)

(0.1

5)

[.86]

[.84]

[.86]

[.84]

[.81]

[1.0

0]

[.71]

[.77]

Con

sum

er�

ph

ase

20.2

20.1

70.0

40.0

3�

4.6

8**

�4.4

1**

�0.4

8*

�0.5

0**

(2.3

1)

(2.2

6)

(0.4

3)

(0.4

2)

(2.2

7)

(2.2

0)

(0.2

5)

(0.2

3)

[.92]

[.94]

[.92]

[.94]

[.05]

[.05]

[.06]

[.04]

Con

sum

er�

ph

ase

3�

0.3

4�

0.4

8�

0.0

6�

0.0

9�

6.5

6**

�6.4

7**

*�

0.6

1**

�0.6

6**

*(2

.03)

(1.8

9)

(0.3

8)

(0.3

6)

(2.7

2)

(2.3

6)

(0.2

9)

(0.2

4)

[.87]

[.80]

[.87]

[.80]

[.02]

[.01]

[.04]

[.01]

Con

trol

sM

ain

effe

cts

Con

sum

erca

rdF

EX

XX

XX

XX

XM

onth

FE

XX

XX

XX

XX

Ad

dit

ion

al

covari

ate

sB

an

kF

E�

FIC

Osc

ore

gro

up

FE

XX

XX

Pre

–C

AR

DA

ct,

con

sum

erm

ean

19.1

419.1

4N

/AN

/A35.4

035.4

03.6

53.6

5R

-squ

are

d0.0

60.7

8N

/AN

/A0.0

50.5

00.2

00.3

0N

um

ber

ofob

serv

ati

ons

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

3,4

47

Not

e:T

able

show

sco

effi

cien

tsfr

omd

iffe

ren

ce-i

n-d

iffe

ren

ces

regre

ssio

ns

that

com

pare

outc

omes

for

con

sum

ercr

edit

card

s(t

reatm

ent

gro

up

)an

dsm

all

bu

sin

ess

card

s(c

ontr

olgro

up

)d

uri

ng

the

dif

fere

nt

ph

ase

sof

the

CA

RD

Act

imp

lem

enta

tion

.W

ed

efin

eth

eA

nti

cip

ati

onp

erio

das

the

mon

ths

bet

wee

nth

ep

ass

age

ofth

ebil

lin

May

2009

an

dth

eim

ple

men

tati

onof

ph

ase

2in

Feb

ruary

2010.

We

defi

ne

ph

ase

2as

Marc

h2010

toA

ugu

st2010

an

dp

hase

3as

the

mon

ths

aft

erA

ugu

st2010.

Th

ep

erio

dp

rior

toA

pri

l2009

isth

eom

itte

dgro

up

,so

the

coef

fici

ents

can

be

inte

rpre

ted

as

the

dif

fere

nti

al

effe

ctre

lati

ve

top

re–C

AR

DA

ctp

erio

d.

Th

esa

mp

leis

rest

rict

edto

acc

oun

tsw

ith

FIC

Osc

ores

bel

ow660

at

orig

inati

on.

Th

esa

mp

lep

erio

dis

Ap

ril

2008–D

ecem

ber

2011.

Th

ere

gre

ssio

ns

are

esti

mate

don

data

aggre

gate

dto

the

ban

k�

pro

du

ctty

pe�

FIC

Osc

ore

gro

up�

mon

thle

vel

,an

dw

eigh

ted

by

tota

lA

DB

inea

chgro

up

.A

pro

du

ctty

pe

isd

efin

edas

the

inte

ract

ion

ofth

eco

nsu

mer

card

ind

icato

ran

dw

het

her

the

card

isco

-bra

nd

ed,

oil

an

dgas,

affi

nit

y,

stu

den

t,or

oth

er.

FIC

Osc

ore

gro

up

sare

<620,

620–659,

660–719,

720–759,

760–799,

an

d�

800.

Sta

nd

ard

erro

rscl

ust

ered

by

ban

k�

pro

du

ctty

pe

are

show

nin

pare

nth

eses

an

dth

eass

ocia

ted

p-v

alu

esare

show

nin

bra

cket

s.T

her

eare

46

such

clu

ster

sin

the

sam

ple

.S

ign

ifica

nce

level

s:*

p<

.10,

**p<

.05,

***

p<

.01.


the finding that the CARD Act had at most a limited effect oninterest charges, with point estimates of approximately zeroacross the different phases.30

The absence of an aggregate response does not mean thatinterest rates were completely unresponsive to the CARD Act.Online Appendix Figure A.VIII shows the frequency of ‘‘upwardrepricing’’ over time, where upward repricing is defined, followingConsumer Financial Protection Bureau (2013a), as an APR in-crease of at least 1 percentage point on a base of at least 10% inthe previous month, to exclude increases that occur at the end ofintroductory rate periods. The figure shows a small spike in thenumber of accounts with APR increases in the two months priorto the CARD Act. This suggests that banks were reacting to theCARD Act, but that the competitive environment and the nonsa-lience of fees limited the aggregate significance of the response.

Columns (3) and (4) of Table VI convert the estimates of thechange in interest changes into an implied offset, which we cal-culate as the interest charge estimates from columns (1) and (2)divided by the phase 3 estimate of the drop in fee revenue fromTable IV with the same controls.31 The standard errors and cor-responding p-values for the offset are calculated using the deltamethod assuming no covariance in the error terms. The phase 3estimates provide the longest adjustment period and representour preferred estimates of the medium-run effects of the law. Thephase 3 estimate of �0.09 from column (4) allows us to rule outoffset effects of greater than 0.61 with 95% confidence.32

Credit card issuers in principle have wide latitude to increaseinterest rates on account holders. The CARD Act somewhat re-duced this flexibility with a set of provisions that came into effectin August 2009, which (i) required lenders to notify consumers 45days in advance of rate changes and (ii) limited lenders’ ability to

30. Online Appendix Figure A.IX, which plots the coefficients on consumeraccount�month interactions from a difference-in-differences specification(equation (2)), further confirms that interest charges do not rise to offset the 5.3percentage point decline in fee revenue, shown by a horizontal line in the figure.

31. We use the phase 3 fee estimate from column (7) of Table IV to construct theoffset estimates in column (3), and the phase 3 fee estimate from column (8) ofTable IV to construct the estimates in column (4).

32. In Online Appendix D we provide ancillary evidence on this estimate using arestriction on the relationship between the offset and the ‘‘pass-through rate’’ of acost shock that we derive from our model. This approach to estimating offsets isdiscussed in greater detail in Agarwal et al. (2014b).





change interest rates on existing accounts, in particular in thefirst year after origination. Lenders, of course, could announcerate changes 45 days in advance, so the main practical effect ofthis provision was to make rate increases more salient and slowdown the implementation of rate increases. Nevertheless, we ex-amine interest charges for new accounts to provide additionalevidence from a setting where banks face fewer constraints intheir pricing.

Online Appendix Table A.II shows difference-in-differencesestimates of the effect on interest charges as a percentage of ADBfor new accounts. Columns (1) and (2) repeat the estimates on theset of all accounts with FICO scores below 660 from Table VI forreference. Columns (3) and (4) show estimates for the sample oflow FICO score new accounts, defined as accounts in their firstfull month since origination. Interest charges for new accountsare noisy because promotional rates and marketing campaigns bya single bank can have a meaningful effect on monthly outcomes.However, the estimates clearly indicate that interest charges fornew accounts did not rise by the 5.3 percentage points needed tooffset the reduction in fee revenue. The preferred phase 3 esti-mate with all controls, shown in column (4), takes on a value of0.17, and we can reject an increase of the full 5.3 percentage pointamount with a p-value below .01.

Online Appendix Figure A.X, Panel A shows interest chargesas an annualized percentage of ADB for account holders withFICO scores of 660 or higher. This figure provides a placebotest for the interest charge response. Since fee revenue barelydeclines for high FICO score account holders, a large change ininterest charges would suggest that there are other contempora-neous effects that are not being captured by the difference-in-differences specification. The plot shows no evidence of adifferential effect of the CARD Act on high FICO score consumeraccounts. Columns (5)–(6) of Online Appendix Table A.II showdifference-in-differences regression specifications that confirmthis finding.

VI.B. Total Income and Costs

The reduction in fee revenue and the lack of an offsettinginterest charge response suggest that the CARD Act reducedbanks’ total income. Recall from Section III.B that we definetotal income as the sum of fee payments, interest payments,





and interchange fees. Although the evidence shows a drop in feesand no change in interest charges, the reduction in fee revenuecould lead to a ‘‘waterbed effect,’’ where credit card issuers offsetthe reduction in fees with higher interchange fee revenue frommerchants.

Columns (5) and (6) of Table VI examine this directly byshowing difference-in-differences specifications for the sampleof low FICO score accounts with total income as a percentage ofADB as the dependent variable. The point estimates show aphase 3 drop in total income of 6.5 percentage points, similar tothe 5.3 percentage point decline in fees. Online Appendix FigureA.II shows that interchange income as a share of purchasevolume was a stable 2% over the entire time period. Thus thecombination of the decline in fee revenue and flat interest chargesand interchange income translates directly into a decline in totalincome from low FICO score account holders.

Another way banks could mitigate their exposure to theCARD Act is by reducing their spending on awards, marketing,or other credit card costs. We examine this potential response byestimating difference-in-differences specifications where the de-pendent variable is costs excluding charge-offs, defined using in-formation on the cost of funds, rewards and fraud expenses, andoperational costs.33 The estimates, shown in columns (7) and (8)of Table VI, show no evidence of an economically significant de-cline in costs.34 However, this does not rule out that the decline inprofitability will lead to a medium-term decline in banks’ invest-ment on IT infrastructure and credit-scoring models.

VI.C. Credit Volume

In the final part of this section, we examine the effects of theCARD Act on the equilibrium volume of credit, as measured bycredit limits, new accounts, and ADB. Figure VII, Panel B showsmean credit limits for consumer and small business cards with aFICO score below 660 at origination. Columns (1) and (2) of TableVII present the corresponding regressions specifications, whichhave credit limit as the dependent variable, and are otherwise

33. We exclude contemporaneous charge-offs because they are not controlled inthe short term by actions taken by credit card issuers.

34. As an additional piece of evidence, Online Appendix Figure A.I shows thatthe ratio of rewards and fraud expenses to interchange income was constant overthe sample and therefore that these costs were approximately constant.





identical to the interest charge regressions (columns (1) and (2) ofTable VI). The plot and regressions provide clear evidence thatthe CARD Act did not bring about a differential reduction incredit limits for consumer accounts. The preferred phase 3point estimate from column (2) of Table VII indicates an increasein consumer account credit limits of $114 or approximately 5% ofthe $2,808 pre–CARD Act mean. We can rule out a drop in creditlimits of greater than than $722, or 26% of the pre–CARD Actmean, with 95% confidence.

Figure VII, Panel C shows plots for the number of new con-sumer and small business accounts with a FICO score below 660.The number of new accounts is measured as a percentage of theaverage pre–CARD Act number of accounts in the data.35

Originations of consumer and small business accounts follow astrong U-shaped pattern over the time period. Both types of ac-counts drop in parallel between the start of our sample (April2008) and the depth of the financial crisis (early 2009) before re-covering over the final two years of our sample period. Themonth-to-month numbers exhibit considerable noise, reflectingthe fact that the percentage of new accounts can be shifted bylarge promotional or marketing campaigns at a single bank.However, the preferred phase 3 point estimate of 0.09 fromcolumn (4) of Table VII indicates that there is virtually zero dif-ferential change in the percentage of new consumer accounts,although the standard error of 0.23 prevents us from ruling outmeaningful effects in either direction.

Figure VII, Panel D shows mean ADB for consumer andsmall business accounts with FICO scores below 660 at origina-tion. There is no evidence of a differential change in consumercredit card ADB. The regression estimates in columns (5) and (6)of Table VI confirm this result.

In the Online Appendix, we present three additional pieces ofevidence on the volume of credit. Appendix Figure A.X plots meancredit limits, new accounts, and ADB for consumer and smallbusiness accounts with a FICO score of 660 or above. AppendixFigure A.IX plots the coefficients on the consumer ac-count�month interactions from a difference-in-differences

35. We use the average pre–CARD Act level of accounts as the denominator,instead of the contemporaneous number of accounts, to minimize noise that arisesfrom bank decisions to write off accounts in blocks, which makes the denominatordrop in discrete increments over time.






regression specification (equation (2)) for the sample of accountswith a FICO score below 660. Appendix Figure A.XI shows anal-ogous plots for the sample of accounts with FICO scores of 660or higher. The plots provide a more granular view of the month-to-month variation and confirm our finding of no volume response.

The nonresponse of credit volume is fully consistent with themodel in Online Appendix C. The lack of price offset suggests that

TABLE VII

CREDIT VOLUME: DIFFERENCE-IN-DIFFERENCES REGRESSIONS (FICO< 660)

(1) (2) (3) (4) (5) (6)Dependent variable:

Credit limits ($)

New accounts(% of pre–CARD

Act level)Average dailybalances ($)

Consumer�anticipation 124.31 40.62 �0.19 �0.17 �105.66* �128.09**(153.63) (163.42) (0.11) (0.11) (59.32) (58.28)

[.42] [.80] [.09] [.13] [.08] [.03]Consumer�phase 2 229.37 101.72 0.02 0.02 �98.13 �128.14

(285.74) (295.80) (0.17) (0.12) (142.37) (136.72)[.43] [.73] [.90] [.87] [.49] [.35]

Consumer�phase 3 291.23 113.95 0.08 0.09 9.30 �39.28(326.38) (309.64) (0.21) (0.23) (190.25) (176.21)

[.38] [.71] [.71] [.70] [.96] [.82]Controls

Main effectsConsumer card FE X X X X X XMonth FE X X X X X X

Additional covariatesBank FE�FICO

score group FEX X X

Pre–CARD Act,consumer mean

2,807.88 2,807.88 0.76 0.76 1,159.29 1,159.29

R-squared 0.00 0.74 0.04 0.24 0.03 0.79Number of observations 3,447 3,447 3,447 3,447 3,447 3,447

Notes. Table shows coefficients from difference-in-differences regressions that compare measures ofcredit volume for consumer credit cards (treatment group) and small business cards (control group) duringthe different phases of the CARD Act implementation. Columns (1) and (2) show regressions with creditlimits as the dependent variable. Columns (3) and (4) show regressions with new accounts as a percentageof the pre–CARD Act average number of accounts as the dependent variable. Columns (5) and (6) showsregressions with average daily balances as the dependent variable. We define the anticipation period asthe months between the passage of the bill in May 2009 and the implementation of phase 2 in February2010. We define phase 2 as March 2010–August 2010 and phase 3 as the months after August 2010. Theperiod prior to April 2009 is the omitted group, so the coefficients can be interpreted as the differentialeffect relative to pre–CARD Act period. The sample is restricted to accounts with FICO scores below 660at origination. The sample period is April 2008–December 2011. The regressions are estimated on dataaggregated to the bank�product type�FICO score group�month level, and weighted by the number ofaccounts in each group. A product type is defined as the interaction of the consumer card indicator andwhether the card is co-branded, oil and gas, affinity, student, or other. FICO score groups are <620, 620–659, 660–719, 720–759, 760–799, and �800. Standard errors clustered by bank�product type are shownin parentheses and the associated p-values are shown in brackets. There are 46 such clusters in thesample. Significance levels: * p< .10, ** p< .05, *** p< .01.




fees are nonsalient to consumers and that consumers are onlyresponsive to the interest rate. Since interest charges areunchanged, from the perspective of the borrower, the cost of bor-rowing does not change and there is no change in the demand forcredit. Since firms continue to make money by lending to con-sumers, equilibrium credit volume is also unchanged.

VII. Conclusion

The recent financial crisis has focused considerable attentionon regulating consumer financial products, with the newly cre-ated CFPB and other federal agencies given an explicit mission to‘‘promote fairness and transparency for mortgages, credit cards,and other consumer financial products and services.’’ We agreewith Campbell et al. (2011) that an important priority for eco-nomic research is to ‘‘evaluate both potential and existing regu-lations to determine whether interventions actually deliver thedesired improvements in the metrics for success.’’

This article takes a step in this direction by providing a quan-titative analysis of the impact of the CARD Act, which introducedsignificant changes to the regulation of credit cards. We find thatthe CARD Act successfully reduced borrowing costs, in particularfor borrowers with the lowest FICO scores. We find no evidencefor offsetting increases in other costs or a decline in credit volume.In addition, we find that the disclosure requirements of the CARDAct had a small but significant impact on borrowers’ repaymentbehavior. Our two years of post–CARD Act data do not allow us toinvestigate the longer run effects of the CARD Act on industryexit or entry, or effects on margins with multiyear contracts (e.g.,promotional agreements) or lumpy long-run investments (e.g., ITinfrastructure and credit-scoring models).

National University of Singapore

Office of the Comptroller of the Currency

University of Chicago

New York University

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qje.oxfordjournals.org).



References

Agarwal, Sumit, Souphala Chomsisengphet, Chunlin Liu, and Nicholas Souleles,‘‘Do Consumers Choose the Right Credit Contracts?,’’ Federal Reserve Bankof Chicago Working Paper WP 2006-11. Working Paper, 2006.

Agarwal, Sumit, Souphala Chomsisengphet, Neale Mahoney, andJohannes Stroebel, ‘‘Selection and Market Power in Consumer CreditMarkets,’’ Working Paper, 2014a.

———, ‘‘A Simple Framework for Estimating the Consumer Benefits fromRegulating Hidden Fees,’’ Journal of Legal Studies, 43 (2014b), S239–S252.

American Bankers Association, ‘‘Response to Request for Information Regardingthe Credit Card Market,’’ 2013.

Ausubel, Lawrence M., ‘‘The Failure of Competition in the Credit Card Market,’’American Economic Review, 81 (1991), 50–81.

Bar-Gill, Oren, and Elizabeth Warren, ‘‘Making Credit Safer,’’ University ofPennsylvania Law Review, 157 (2008), 1–101.

Calem, Paul S., and Loretta J. Mester, ‘‘Consumer Behavior and the Stickiness ofCredit-Card Interest Rates,’’ American Economic Review, 85 (1995),1327–1336.

Campbell, John Y., ‘‘Household Finance,’’ Journal of Finance, 61 (2006),1553–1604.

Campbell, John Y., Howell E. Jackson, Brigitte C. Madrian, and Peter Tufano,‘‘Consumer Financial Protection,’’ Journal of Economic Perspectives, 25(2011), 91–114.

Carroll, Gabriel D., James J. Choi, David Laibson, Brigitte C. Madrian, andAndrew Metrick, ‘‘Optimal Defaults and Active Decisions,’’ QuarterlyJournal of Economics, 124 (2009), 1639–1674.

Choi, James J., David I. Laibson, and Brigitte C. Madrian, ‘‘Are Empowermentand Education Enough? Underdiversification in 401 (k) Plans,’’ BrookingsPapers on Economic Activity, 2 (2005), 151–213.

Conley, Timothy G., and Christopher R. Taber, ‘‘Inference with ‘Difference inDifferences’ with a Small Number of Policy Changes,’’ Review of Economicsand Statistics, 93 (2011), 113–125.

Consumer Financial Protection Bureau, ‘‘CARD Act Report,’’ 2013a.———, ‘‘Navigating the Market: A Comparison of Spending on Financial

Education and Financial Marketing,’’ 2013b.Debbaut, Peter, Andra Ghent, and Marianna Kudlyak, ‘‘Are Young Borrowers

Bad Borrowers? Evidence from the Credit CARD Act of 2009,’’ FederalReserve Bank of Richmond WP 13-09R. Working Paper, 2013.

Federal Reserve Board of Governors, ‘‘Report to the Congress on the Use of CreditCards by Small Businesses and the Credit Card Market for SmallBusinesses,’’ 2010.

Federal Reserve Bank of New York, ‘‘Quarterly Report on Household Debt andCredit, May 2013,’’ 2013.

Gabaix, Xavier, and David Laibson, ‘‘Shrounded Attributes, Consumer Myopia,and Information Suppression in Competitive Markets,’’ Quarterly Journal ofEconomics, 121 (2006), 505–540.

Gross, David B., and Nicholas S. Souleles, ‘‘Do Liquidity Constraints and InterestRates Matter for Consumer Behavior? Evidence from Credit Card Data,’’Quarterly Journal of Economics, 117 (2002), 149–185.

Heidhues, Paul, Botond Ko00szegi, and Takeshi Murooka, ‘‘Exploitive Innovation,’’UC Berkeley Working Paper, 2012.

Jambulapati, Vikram, and Joanna Stavins, ‘‘The Credit CARD Act of 2009:What Did Banks Do?,’’ Federal Reserve Bank of Boston Working Paper 13-7, 2013.

Kay, Benjamin, Mark D. Manuszak, and Cindy M. Vojtech, ‘‘Bank Profitabilityand Debit Card Interchange Regulation: Bank Responses to the DurbinAmendment,’’ Federal Reserve Board Working Paper 2014-77, 2014.

Keys, Benjamin, and Jialan Wang, ‘‘Perverse Nudges: Minimum Payments andDebt Paydown in Consumer Credit Cards,’’ University of Chicago WorkingPaper, 2014.


Kuchler, Theresa, ‘‘Sticking to Your Plan: Hyperbolic Discounting and CreditCard Debt Paydown,’’ New York University Working Paper, 2013.

Laibson, David, Andrea Repetto, and Jeremy Tobacman, ‘‘Estimating DiscountFunctions with Consumption Choices over the Lifecycle,’’ NBER WorkingPaper, 2007.

Madrian, Brigitte C., and Dennis F. Shea, ‘‘The Power of Suggestion: Inertia in401 (k) Participation and Savings Behavior,’’ Quarterly Journal ofEconomics, 116 (2001), 1149–1187.

Mullainathan, Sendhil, Michael Barr, and Eldar Shafir, ‘‘The Case forBehaviorally Informed Regulation,’’ in New Perspectives on Regulation,David Moss, and John Cisternino, eds. (Cambridge, MA: Tobin Project,2009), 25–62.

Navarro-Martinez, Daniel, Linda Court Salisbury, Katherine N. Lemon,Neil Stewart, William J. Matthews, and Adam J. L. Harris, ‘‘MinimumRequired Payment and Supplemental Information Disclosure Effects onConsumer Debt Repayment Decisions,’’ Journal of Marketing Research, 48(2011), S60–S77.

Posner, Eric, and E. Glen Weyl, ‘‘Benefit-Cost Analysis for Financial Regulation,’’American Economic Review, 103 (2013), 393–397.

Stango, Victor, and Jonathan Zinman, ‘‘Borrowing High vs. Borrowing Higher:Sources and Consequences of Dispersion in Individual Borrowing Costs,’’NBER Working Paper, 2013.

Stewart, Neil, ‘‘The Cost of Anchoring on Credit-Card Minimum Repayments,’’Psychological Science, 20 (2009), 39–41.

Thaler, Richard H., and Shlomo Benartzi, ‘‘Save More TomorrowTM: UsingBehavioral Economics to Increase Employee Saving,’’ Journal of PoliticalEconomy, 112 (2004), S164–S187.

Thaler, Richard H., and Cass R. Sunstein Nudge: Improving Decisions aboutHealth, Wealth, and Happiness (New Haven, CT: Yale University Press,2008).

Tse, Tomoeh Murakami, ‘‘J.P. Morgan’s Dimon Says New Laws Have Hurt HisCompany.’’ The Washington Post, April 2, 2010.

U.S. Senate, ‘‘S5314,’’ Congressional Record, 155 (2009).


regulating consumer financial products: evidence...

Documents