Defined Terms and Documents

Who Gains and Who Loses from Credit Card Payments?  Theory and Calibrations - Federal Reserve Bank of Boston  -  Aug 2010

Scott Schuh, Oz Shy, and Joanna Stavins

Abstract:

Merchant fees and reward programs generate an implicit monetary transfer to credit card

users from non-card (or “cash”) users because merchants generally do not set differential

prices for card users to recoup the costs of fees and rewards. On average, each cash-using

household pays $149 to card-using households and each card-using household receives $1,133

from cash users every year. Because credit card spending and rewards are positively

correlated with household income, the payment instrument transfer also induces a regressive

transfer from low-income to high-income households in general. On average, and after

accounting for rewards paid to households by banks, the lowest-income household ($20,000 or

less annually) pays $21 and the highest-income household ($150,000 or more annually)

receives $750 every year. We build and calibrate a model of consumer payment choice to

compute the effects of merchant fees and card rewards on consumer welfare. Reducing

merchant fees and card rewards would likely increase consumer welfare.

Keywords: credit cards, cash, merchant fees, rewards, regressive transfers, no-surcharge rule

JEL Classifications: E42, D14, G29

Scott Schuh is Director of the Consumer Payments Research Center and a senior economist in the research

department at the Federal Reserve Bank of Boston. Oz Shy is a senior economist and a member of the Consumer

Payments Research Center and Joanna Stavins is a senior economist and policy advisor and a member of the

Consumer Payments Research Center, both in the research department at the Federal Reserve Bank of Boston. Their

email addresses are scott.schuh@bos.frb.org, oz.shy@bos.frb.org, and joanna.stavins@bos.frb.org, respectively.

This paper, which may be revised, is available on the web site of the Federal Reserve Bank of Boston at

http://www.bos.frb.org/economic/wp/index.htm.

We thank Tamás Briglevics for most valuable research assistance, analysis, and advice. We also thank Santiago

Carbó Valverde, Dennis Carlton, Bob Chakravorti, Alan Frankel, Jeff Fuhrer, Fumiko Hayashi, Bob Hunt, Suzanne

Lorant, John Sabelhaus, Irina Telyukova, Bob Triest, Lotta Väänänen, Zhu Wang, Paul Willen, and Michael Zabek,

as well as seminar participants at the Boston Fed and at the Economics of Payments IV conference (New York Fed,

May 2010), the conference on Platform Markets (ZEW Mannheim, June 2010), and the conference on Payment

Markets (University of Granada, June 2010) for valuable comments and suggestions on earlier drafts.

The views and opinions expressed in this paper are those of the authors and do not necessarily represent the views

of the Federal Reserve Bank of Boston or the Federal Reserve System.

This version: August 31, 2010

1. Introduction

The typical consumer is largely unaware of the full rami cations of paying for goods and

services by credit card. Faced with many choices|cash, check, debit or credit card, etc.|

consumers naturally consider the costs and bene ts of each payment instrument and choose

accordingly. For credit cards, consumers likely think most about their bene ts: delayed

payment|\buy now, pay later"|and the rewards earned|cash back, frequent ier miles,

or other enticements. What most consumers do not know is that their decision to pay by

credit card involves merchant fees, retail price increases, a nontrivial transfer of income from

cash to card payers, and consequently a transfer from low-income to high-income consumers.

In contrast, the typical merchant is acutely aware of the rami cations of his customers'

decisions to pay with credit cards. For the privilege of accepting credit cards, U.S. merchants

pay banks a fee that is proportional to the dollar value of the sale. The merchant's bank

then pays a proportional interchange fee to the consumer's credit card bank.1 Naturally,

merchants seek to pass the merchant fee to their customers. Merchants may want to recoup

the merchant fee only from consumers who pay by credit card. In practice, however, credit

card companies impose a \no-surcharge rule" (NSR) that prohibits U.S. merchants from

doing so, and most merchants are reluctant to give cash discounts.2 Instead, merchants

mark up their retail prices for all consumers by enough to recoup the merchant fees from

credit card sales.

This retail price markup for all consumers results in credit-card-paying consumers being

subsidized by consumers who do not pay with credit cards, a result that was rst discussed

in Carlton and Frankel (1995), and later in Frankel (1998), Katz (2001), Gans and King

1Shy and Wang (Forthcoming) show that card networks extract higher surplus from merchants using

proportional merchant fees (rather than xed, per-transaction fees). The amount of surplus that card

networks can extract increases with the degree of merchants' market power.

2See Appendix D for additional discussion on the implications of the NSR. Card associations allow

U.S. merchants to give cash discounts under certain restrictions. However, cash discounts are not widely

observed. Frankel (1998) argues that a prohibition on credit card surcharges can have e
ects di
erent from

those resulting from a prohibition on cash discounts, because card surcharges allow merchants to vary their

charges according to the di
erent merchant fees they pay on di
erent cards, whereas a cash discount is taken

from a single card price.

1

(2003), and Schwartz and Vincent (2006). For simplicity, we refer to consumers who do

not pay by credit card as cash payers, where \cash" represents all payment instruments

other than credit cards: cash, checks, debit and prepaid cards, etc.3 \Subsidize" means that

merchant fees are passed on to all buyers in the form of higher retail prices regardless of the

means of payments buyers use to pay. Thus, cash buyers must pay higher retail prices to

cover merchants' costs associated with the credit cards' merchant fees. Because these fees

are used to pay for rewards given to credit card users, and since cash users do not receive

rewards, cash users also nance part of the rewards given to credit card users.

If the subsidy of card payers by cash payers results from heterogeneity in consumer

preferences and utility between cash and card payments, the subsidy may be innocuous

in terms of consumer and social welfare. However, U.S. data show that credit card use is

very positively correlated with consumer income. Consequently, the subsidy of credit card

payers by cash payers also involves a regressive transfer of income from low-income to high-

income consumers. This regressive transfer is ampli ed by the disproportionate distribution

of rewards, which are proportional to credit card sales, to high-income credit card users.4

Frankel (1998, Footnote 85) was the rst to connect the wealth transfers to average income

of groups of consumers (that is, poorer non-cardholders subsidizing wealthier cardholders).

This idea was later discussed in Carlton and Frankel (2005, pp. 640{641) and Frankel and

Shampine (2006, Footnote 19).5

Our contribution to this line of research is that we are the rst to compute who gains

and loses from credit card payments in the aggregate economy. We compute dollar-value

estimates of the actual transfers from cash payers to card users and from low-income to

3McAndrews and Wang (2008) demonstrates the possibility of a subsidy in the opposite direction (from

card to cash users) in cases where merchants' cost of handling cash exceeds merchants' card fees. McAndrews

and Wang's de nition of cards includes debit cards, which are less costly than credit cards, whereas in our

paper debit cards are considered part of \cash." Humphrey et al. (1996) and Humphrey et al. (2006)

also provide evidence that electronic payment instruments, such as debit cards, are less costly than paper

instruments, such as cash or check. Again, however, we focus only on credit cards, which have high merchant

fees and are more costly than other payment instruments, paper or electronic.

4See Hayashi (2009) and her references for a comprehensive overview of card reward programs.

5Similar points were made recently in New York Times articles by Floyd Norris, \Rich and Poor Should

Pay Same Price," October 1, 2009; and by Ron Lieber, \The Damage of Card Rewards," January 8, 2010.

2

high-income households. A related paper by Berkovich (2009) estimates the total amount

transferred from non-rewards consumers to rewards consumers in the United States resulting

from gasoline and grocery purchases only.6

We propose a simple, model-free accounting methodology to compute the two transfers

by comparing the costs imposed by individual consumer payment choices with actual prices

paid by each buyer. On average, each cash buyer pays $149 to card users and each card

buyer receives $1,133 from cash users every year, a total transfer of $1,282 from the average

cash payer to the average card payer. On average, and after accounting for rewards paid

to households by banks, when all households are divided into two income groups, each

low-income household pays $8 to high-income households and each high-income household

receives $430 from low-income households every year. The magnitude of this transfer is

even greater when household income is divided into seven categories: on average, the lowest-

income household ($20; 000 or less annually) pays a transfer of $21 and the highest-income

household ($150; 000 or more annually) receives a subsidy of $750 every year. The transfers

among income groups are smaller than those between cash and card users because some

low-income households use credit cards and many high-income households use cash. Finally,

about 79 percent of banks' revenue from credit card merchant fees is obtained from cash

payers, and disproportionately from low-income cash payers.

To conduct welfare and policy analysis of these transfers, we construct a structural model

of a simpli ed representation of the U.S. payments market and calibrate it with U.S. micro

data on consumer credit card use and related variables. Parameters derived from the model

are notably reasonable given the simplicity and limitations of the model and data. High-

income households appear to receive an inherent utility bene t from credit card use that

is more than twice as high as that received by low-income households. Eliminating the

merchant fee and credit card rewards (together) would increase consumer welfare by 0.15 to

6This estimated transfer is about $1.4b to $1.9b, and rewards are found to have a disproportionate impact

on low-income minorities and to resemble a regressive tax on consumption. These estimates focus exclusively

on rewards transfers and do not account for the full range of transfers from low- to high-income consumers

resulting from merchant fees.

3

0.26 percent, depending on the degree of concavity of utility, which also can be interpreted

in an aggregate model as the degree of aversion to income inequality in society.

Our analysis is consistent with, but abstracts from, three features of the U.S. payments

market. First, we focus on the convenience use of credit cards (payments only) and do not

incorporate a role for revolving credit, which is an important feature of the total consumer

welfare associated with credit cards.7 U.S. data indicate that household propensity to revolve

credit card spending is surprisingly similar across income groups, so it is unlikely that interest

income plays a major role in the transfers. This fact supports working with a static model

that is more tractable for data analysis. Second, we abstract from the supply-side details

of the payments market for both cash and cards. We take as given the well-established,

seminal result of Rochet and Tirole (2006) concerning the critical role of an interchange fee

between acquiring and issuing banks in the two-sided credit card market, a result that notes

that the optimal level of the interchange fee is an empirical issue.8 By incorporating both

merchant fees and card rewards rates, we can assume that the interchange fee lies between

these rates and is set internally in the banking sector to the optimal level conditional on

fees and rewards. Finally, we do not incorporate a role for the distribution of bank pro ts

from credit card payments to households that own banks, because of a lack of sucient micro

data. Given these three simpli cations, we can assess only the consumer welfare implications

of the payment instrument transfers but not the full social welfare implications.

We want to be clear that we do not allege or imply that banks or credit card compa-

nies have designed or operated the credit card market intentionally to produce a regressive

transfer from low-income to high-income households. We are not aware of any evidence to

7For example, the work of Carroll (1997) provides motivation for credit cards to help consumers smooth

income in the face of income and wealth shocks and achieve optimal consumption plans. However, the

actual impact of credit card borrowing on consumer and social welfare is complicated, as can be seen from

literature, including Brito and Hartley (1995), Gross and Souleles (2002), Chatterjee et al. (2007), and

Cohen-Cole (Forthcoming).

8A complete list of contributions to two-sided markets is too long to be included here. The interested

reader can consult Chakravorti and Shah (2003), Gans and King (2003), Rochet (2003), Wright (2003),

Roson (2005), Evans and Schmalensee (2005), Armstrong (2006), Schwartz and Vincent (2006), Bolt and

Chakravorti (2008), Hayashi (2008), Rysman (2009), and Verdier (Forthcoming). For a comprehensive

empirical study of interchange fees, see Prager et al. (2009).

4

support this allegation or any a priori reason to believe it. However, the existence of a

non-trivial regressive transfer in the credit card market may be a concern that U.S. individ-

uals, businesses, or public policy makers wish to address. If so, our analysis suggests several

principles and approaches worth further study and consideration, which we discuss briey at

the end of the paper. Recent U.S. nancial reform legislation, motivated by concerns about

competition in payment card pricing, gives the Federal Reserve responsibility for regulating

interchange fees associated with debit (but not credit) cards. Our analysis provides a di
er-

ent but complementary motivation|income inequality|for policy intervention in the credit

card market.

Section 2 documents three basic facts about card card use. Section 3 demonstrates a

simple \accounting" of transfers from cash to card users and from low-to high-income buy-

ers. Section 4 presents an analytical model, which is then used in Section 5 to calibrate

the welfare-maximizing merchant fees and rewards to card users, and to compute changes

in welfare associated with a total elimination of card reward programs and merchant fees.

Policy implications are explored in Section 6. Section 7 subjects our computations of income

transfers to a wide variety of tests associated with additional modi cations of the data. Sec-

tion 8 concludes. An appendix provides data details and sensitivity analysis of the calibrated

model.

2. Basic Facts about Credit Cards

This section establishes three basic facts about credit cards: 1) consumer credit card use

has been increasing; 2) consumer credit card use and rewards are positively correlated with

household income; and 3) credit card use varies across consumers due to heterogeneity in

nonpecuniary bene ts from cards, even within income groups. These facts motivate our

analysis and modeling of transfers among consumers, associated with convenience use of

cards.

5

2.1 Credit cards in the economy

Over the last two decades, payment cards have enjoyed increased popularity in all sectors of

the economy. Our research focuses on credit and charge cards issued by banks, stores, and

gas stations and used by consumers only. Figure 1 shows that the fraction of households who

have a credit card (adopters) has been steady at about 70{75 percent during the past two

decades, reecting the maturity of the market. However, the percentage of total consumption

expenditure paid for by credit card increased from about 9 percent to 15 percent during the

same period.9 As a result, revenue from merchant fees, which are proportional to credit card

spending, also increased. Consumer credit card spending accounts for approximately half of

all credit card spending in 2007.10

0 20 40 60 80 100

Percentage of households

8 10 12 14 16 18

Percentage of consumption expenditure

1990 1995 2000 2005 2010

Consumption spending

volume (left scale)

Credit card adoption

rate (right scale)

Sources: Survey of Consumer Finances 1989−2007

Credit Card Usage

Figure 1: Credit card adoption and spending rates.

9Both series were taken from the Survey of Consumer Finances (SCF), which asked consumers about the

amount of credit card charges they had in the previous month (variable x412 ) since 1989 (\Consumption

spending volume") and about credit card adoption (variable x410 ) since 1989 (\Credit card adoption rate").

10Total credit card spending, which includes business and government expenditures, was about $42 billion

in 2007, according to the Federal Deposit Insurance Corporation's Call Report data (series rcfdc223 and

rcdfc224 ).

6

2.2 Card use and income

Although previous literature found a positive relationship between income and credit card

adoption (Stavins (2001), Mester (2003), Bertaut and Haliassos (2006), Klee (2006), Zinman

(2009a), Schuh and Stavins (2010)), there has been less focus on the relationship between

income and credit card use. Publicly available data sources, such as the 2007 Survey of

Consumer Finances, typically provide only the dollar amounts charged on credit cards, which

we de ne here as use. However, data on the number of transactions consumers make with

credit cards are available from the new 2008 Survey of Consumer Payment Choice (SCPC).

The data reveal a strong positive correlation between consumer credit card use and house-

hold income, as shown in Table 1. (The unequally sized income categories are as reported

in published aggregate data from the Consumer Expenditure Survey.) The proportion of

households who hold (have adopted) at least one credit card increases monotonically with

income ( rst column). Average new monthly charges on all credit cards held by a household

also increases monotonically with income among households who have adopted credit cards

(second column).11 And the share of credit card spending in total household consumption

also increases monotonically with income (third column).12

The data also reveal a strong positive correlation between consumer credit card rewards

and household income, as shown in Table 2. The share of credit card holders earning any

type of rewards increases monotonically with income. A similar pattern is visible for each of

the major types of rewards as well: cash back, frequent yer miles, discounts, and others.

In most of our analysis, we split the consumer population into two income groups: house-

holds earning less than $100; 000 and households earning more than that.13 This decision

11The new charge numbers are based on the following question from the 2007 SCF: \On your last bill,

roughly how much were the new charges made to these [Visa, MasterCard, Discover, or American Express]

accounts?" Because merchant fees are proportional to the amount charged on credit cards, regardless of

whether the cardholder pays his monthly balance or carries it over to the next month, total new credit card

charges for each household is the relevant measure of credit card use.

12The share of credit card spending in household income actually decreases with household income, however,

because the marginal propensity to consume falls with household income.

13Table 7 generalizes our results to multiple income groups.

7

Average monthly cc Share of cc spending

Annual income Have cc charge by adopters in consumption

Under $20; 000 42% $447 8:4%

$20; 000{49; 999 67% $478 9:3%

$50; 000{79; 999 87% $714 12:8%

$80; 000{99; 999 92% $1; 026 15:7%

$100; 000{119; 999 93% $1; 293 17:9%

$120; 000{149; 999 97% $1; 642 20:9%

Over $150; 000 97% $4; 696 27:6%

Under $100; 000 68% $616 11:3%

Over $100; 000 96% $2; 966 24:8%

Whole sample 73% $1; 190 16:9%

Table 1: Households' credit card adoption rates and new monthly charges by annual household

income. Source: 2007 Survey of Consumer Finances.

is motivated by the need for parsimony in modeling, by the signi cant di
erences in credit

card behavior between these two broad income groups shown in Tables 1 and 2, and by

our desire to put the focus more on the transfer to higher-income households (and less on

the transfer from lower-income households). Table 1 shows that credit card spending by

high-income consumers is nearly ve times higher than credit card spending by low-income

consumers, and Table 2 shows that high-income consumers are 20 percentage points more

likely to receive credit card rewards. The difference between the lowest-income (less than

$20,000 per year) and the highest-income ($150,000 per year or more) households' credit

card spending and rewards is markedly greater.

2.3 Non-income factors a
ecting credit card use

Income is not the only factor that is positively correlated with credit card use. Schuh and

Stavins (2010) estimated the use of payment instruments as a function of various characteris-

tics of these instruments, employing a 2006 survey of U.S. consumers. They found that, after

controlling for income, the characteristics of convenience, cost, and timing of payment have

a statistically signi cant e
ect on credit card use. Using the more extensive 2008 SCPC,

we re-estimated the e
ects of payment instrument characteristics on consumer adoption and

8

Income Any Reward Cash Back Airlines Miles Discounts Other Rewards

Under $20,000 48 27 17 13 8

$20,000{49,999 50 28 17 11 10

$50,000{79,999 62 35 26 13 12

$80,000{99,999 68 38 36 15 11

$100,000{119,999 71 37 33 16 15

$120,000{149,999 82 44 39 19 25

Over $150,000 75 33 48 15 19

Under $100; 000 57 32 23 12 10

Over $100; 000 77 37 40 16 19

Whole sample 61 33 27 13 12

Table 2: Percentage (%) of credit card adopters receiving credit card rewards. Source: 2007{2008

Consumer Finance Monthly survey conducted by the Ohio State University.

use of credit cards, using the following speci cation:

CCi

TOTPAYi

= f (CHARi;DEMi; Yi;NUMi) ; (1)

where CCi=TOTPAYi is consumer i's share of the number of credit card payments in total

payments; CHARi is a vector of characteristics of credit cards relative to all other payments

adopted by consumer i, DEMi is a vector of demographic variables for consumer i, including

age, race, gender, education, and marital status; Yi is a set of income and nancial variables;

NUMi is the set of dummy variables indicating the number of other payment instruments

adopted by consumer i.

Table 3 shows the distribution of credit card use, calculated as a share of credit card

payments in all payments for each consumer. The share of credit card transactions is higher

for the over $100K income group than for the under $100K income group across the whole

distribution. However, there is substantial variation within each income group. For example,

among the high-income consumers, the 10th percentile of credit card users pay for 4 percent

of their transactions with credit cards, compared with 70 percent of transactions for the 90th

percentile of users. Therefore, there is variance in credit card use within income groups that

needs to be explained.

Several relative payment-instrument characteristics have a signi cant e
ect on credit card

9

Percentile Under $100K Over $100K Whole Sample

10th 0 4 1

25th 5 13 5

50th 15 30 18

75th 34 55 39

90th 63 70 66

Table 3: Distribution (%) of credit card use within income groups for credit card adopters. Note:

Based on the 2008 Survey of Consumer Payment Choice, and weighted using the population

weights from the 2008 SCPC.

use. Table 4 shows the estimated coecients on payment-instrument characteristics from

estimating equation (1) for three di
erent samples. While the cost of credit cards (which

includes rewards as well as interest rates and fees) is signi cant in all speci cations and for

both income groups, other attributes of credit cards also are important determinants of credit

card use, conditional on cost. Controlling for income categories (column 1 of Table 4), ease

of use and record keeping have a strong and statistically signi cant e
ect on credit card use.

In separate regressions by household income category, record keeping and cost have much

stronger e
ects on higher-income consumers (column 3) than on lower-income consumers

(column 2), while ease of use was not statistically signi cant for the higher-income group.

The preceding results indicate that payment-instrument characteristics are valued dif-

ferently by consumers both within and between income groups. The model in Section 4

captures consumers' nonpecuniary bene ts from using credit cards relative to cash, such as

record keeping, in a utility parameter labeled as bi, speci c to income group i. This param-

eter turns out to be an important factor determining the choice of cash versus credit card

for payments.

3. Transfer Accounting

This section demonstrates a simple, model-free approach to computing two implicit monetary

transfers between U.S. consumers that result when some buyers pay with credit cards and

others do not. One transfer is from cash buyers to credit card buyers; the other is from

10

(1) (2) (3)

Explanatory Variables Whole Sample Under $100K Over $100K

Cost 0:10 *** 0:10 *** 0:13 ***

Speed 0:00 􀀀0:05 0:11

Security 0:01 0:02 􀀀0:02

Control 0:01 0:01 􀀀0:00

Records 0:11 *** 0:08 ** 0:17 **

Acceptance 0:06 0:06 0:08

Ease 0:11 *** 0:12 ** 0:11

Income categories included? Yes No No

Table 4: Three credit card use regressions. Note: Authors' estimation using the 2008 Survey of

Consumer Payment Choice. *** signi cant at the 1% level, ** signi cant at the 5% level.

low-income buyers to high-income buyers. Our methodology decomposes national income

account data on consumption into consumer groups de ned by payment choice and income

level, using micro data on consumption, credit card spending, and related variables (along

with the benchmark estimates of payment costs). Humphrey, Kaloudis, and wre (2004)

use an analogous methodology to estimate cash use in Norway.

3.1 The payments market

Figure 2 illustrates a simpli ed version of the U.S. payments market that frames the computa-

tion of aggregate transfers. There are three types of agents: buyers (consumers), merchants,

and \banks." Buyers can have high or low incomes and pay by credit card or cash (all other

non-credit card payments). A representative merchant sells a representative good to all con-

sumers. This assumption is not strictly true for all markets, so we explore the implications

of relaxing it in Section 7. However, it is a good approximation for most transactions and

it is necessary to compute the transfers, given the lack of micro data on payment choice

at the level of individual transactions.14 Finally, \banks" represents the nancial market

that provides credit card payment services. It includes banks that issue cards to consumers

14It also greatly simpli es the modeling task by avoiding the need to have search and matching of individual

consumers, merchants, and goods|a level of detail for which proper data are not currently available

anyway|in addition to payment choice.

11

(\issuers"), banks that receive card payments from merchants (\acquirers"), and card com-

panies (Visa or MasterCard are examples) that facilitate interactions among banks and be-

tween banks and their customers.15 The literature on two-sided markets analyzes the details

of the \banks" and merchant markets but tends to abstract from consumer heterogeneity,

restricting analysis of transfers among consumers. Our analysis takes the opposite approach.

























Issuer Acquirer



6

?

p = Price ($)

 = Interchange fee (%)

 = Reward (1%)  = Merchant Fee (2%)









Card Companies }

=

\Banks"

Low-income card users -

High-income card users

#

"

!

Merchant Low-income cash users

High-income cash users

#

"

!



p

( <  < )

 = handling cash cost (0:5%)

Figure 2: Fees and payments in a simple market with a card network.

Payments occur as follows. Buyers purchase a good for an endogenously determined price,

p, using cash or credit card according to buyers' preferences for the payment instruments.

The merchant incurs a cost with either payment choice. For cash, the merchant bears a cost,

denoted 0   < 1, associated with handling cash transactions. Thus, the merchant's cost

of accepting a cash transaction is   p.16 For credit cards, the merchant pays a fee, , to

banks (acquirers) that is proportional to card sales. Thus, the merchant's cost of accepting

a credit card transaction is   p. Card buyers receive a partial rebate of the merchant fee

from banks (issuers) in the form of card rewards, , that are proportional to card sales and

15Until recently, Visa and MasterCard were owned by banks. Visa became public in early 2008, and

MasterCard in 2006.

16As drawn, the cash-handling cost  is a marginal cost. However, the actual cost of handling cash may

include a xed cost as well. Footnote 22 presents estimates of the cost of handling cash where  could be

interpreted as average cost that includes possible xed costs because the data do not distinguish well between

xed and marginal costs.

12

are given to encourage use.17 Thus, card buyers receive reward income of   p.

The merchant fee and reward rate are closely related to pricing decisions internal to banks.

Acquirers pay a proportional fee, , to issuers. When the card issuer and card acquirer are

owned by di
erent nancial institutions,
 is called an interchange fee. Because interchange

fees involve the xing of fees by competing card issuers, they have triggered many debates and

court cases against card organizations by antitrust authorities and merchant associations.18

Typically, banks make pro ts by setting  <  < , which we assume holds. Our analysis of

the transfers among consumers requires only the merchant fee and reward rate and not the

inclusion of the interchange fee.

Regardless of whether buyers choose cash or credit card, U.S. merchants tend to charge

the same price, p, despite incurring di
erent costs from the two payment instruments. Under

the no-surcharge rule, merchants cannot charge credit card buyers a higher price than the

price they charge cash buyers to recoup the extra cost (􀀀  1:5 percent in our calculations).

However, under certain conditions card companies do allow the merchant to o
er a discount

to cash buyers, which is conceptually the same as surcharging cards.19 Nevertheless, while

some U.S. merchants have o
ered cash discounts from time to time, they generally do not do

so widely or consistently. One reason may be the cost of o
ering two prices. Another reason

may be concerns about adverse customer reactions to di
erential pricing and especially to

penalizing card buyers, who tend to be higher-income households and to buy more goods.

The simpli ed payments market in Figure 2 covers only convenience use of credit cards

and not the revolving credit feature of cards. In reality, banks also receive revenue from

consumers through interest payments on revolving debt and from credit card fees (annual,

over-the-limit, etc.), so it is possible that card rewards may be funded from sources of

17To fund rewards, banks use revenue from merchant fees and possibly other sources, such as annual fees

or interest from revolving credit card debt. Funding of rewards is discussed more later.

18Some court cases in the United States and worldwide are discussed in Bradford and Hayashi (2008).

19For example, Section 5.2.D.2 of Visa U.S.A. April 2008 operating regulations states that \A Merchant

may o
er a discount as an inducement for a Cardholder to use a means of payment that the Merchant prefers,

provided that the discount is clearly disclosed as a discount from the standard price and, non-discriminatory

as between a Cardholder who pays with a Visa Card and a cardholder who pays with a `comparable card'."

See also Footnote 2.

13

credit card revenue other than merchant fees.20 However, our data and analysis presented

below suggest that these alternative sources of credit card revenue are unlikely to alter

our qualitative conclusions about transfers. Furthermore, the welfare e
ects of credit card

borrowing and lending are extremely dicult to identify in economic theory and practice|

revolving debt may be welfare improving, even at very high interest rates|whereas the

welfare e
ects of transfers among consumers associated with convenience use of credit cards

are less so.

3.2 Data and assumptions

The payments market discussed in Section 3.1 generates implicit monetary transfers between

consumers, regardless of whether revolving credit is extended for card purchases. Calculation

of these transfers does not require a formal economic model, only data and arithmetic|

hence the terminology \transfer accounting."21 However, the transfer calculations are based

on three key economic assumptions described below.

The quantitative fees and costs portrayed in Figure 2 represent \benchmark" estimates

of recent conditions in the U.S. payments market. The limited available data suggest that

a reasonable, but very rough, estimate of the per-dollar merchant e
ort of handling cash

is  = 0:5 percent.22 Available data suggest that a reasonable estimate of the merchant

fee across all types of cards, weighted by card use, is  = 2 percent.23 And available data

20Section 7.2 discusses the funding of card rewards and the relevant literature.

21See Appendix A for more details about the data.

22Garcia-Swartz, Hahn, and Layne-Farrar (2006) report that the marginal cost of processing a $54.24

transaction (the average check transaction) is $0.43 (or 0.8 percent) if it is a cash transaction and $1.22

(or 2.25 percent) if it is paid by a credit/charge card. The study by Bergman, Guibourg, and Segendorf

(2007) for Sweden found that the total private costs incurred by the retail sector from handling 235 billion

Swedish Crown (SEK) worth of transactions was 3.68 billion SEK in 2002, which would put our measure

of cash-handling costs at  = 1:6 percent. For the Norwegian payment system, Gresvik and Haare (2009)

estimates that private costs of handling 62.1 billion Norwegian Crown (NOK) worth of cash transactions

incurred by the retailers was 0.322 billion NOK in 2007, which would imply  = 0:5 percent.

23Merchant fees in the United States were in the range of $40{$50 billion in 2008; see, for example, \Card

Fees Pit Retailers Against Banks," New York Times, July 15, 2009. This range approximately equals 2

percent of the U.S. credit card sales for that same year in the Call Report data for depository institutions.

Actual merchant fees are complex and heterogeneous, varying over cards and merchants. We estimate

merchant fees across cards as follows: general purpose (Visa, MasterCard, and Discover) 2 percent; American

14

suggest that a reasonable estimate of the reward rate is  = 1 percent.24 However, according

to Table 2, only 55 percent of low-income credit card holders receive rewards, compared

with 75 percent of high-income card holders. For this reason, the average card user in either

income group will not receive the full reward, , but only  multiplied by the fraction of credit

cards with rewards among all credit cards carried by this income group. Thus L = 0:57 and

H = 0:79 denote the e
ective
reward rates received by an average household belonging to

income groups L (low) and H (high), respectively.25

In addition to the benchmark speci cations, the only data needed to calculate the trans-

fers are sales revenues (credit card and total) and the number of buyers. Let t denote the

quantity of transactions and S = t  p denote sales revenue. Sales are measured by consump-

tion from the National Income and Product Accounts (NIPA) and Consumer Expenditure

Survey (CEX), which were S = $9:83 trillion in 2007.26 About 42 percent of this con-

sumption does not involve a payment choice for consumers, for example, imputed rental

of owner-occupied housing, employer-provided health insurance, and fees paid for nancial

services, and thus this portion is excluded from the calculations27. Let N = NL + NH be

the total number of buyers and the sum of buyers with low and high incomes (subscripts L

and H, respectively). Buyers are measured by the number of households, as reported by the

Census Bureau, which was N = 116:0 million in 2007. The proportions of high- and low-

income households and credit card spending data are obtained from the Survey of Consumer

Finances (SCF) and applied to N.28 For reasons described earlier, we set $100; 000 as the

Express 2:2 percent; and speci c purpose (branded) 1 percent, see Hayashi (2009) for some numbers.

24One-percent cash back is widely observed. Most airline mileage and other points systems also have an

approximate cash value of about  = 1 percent.

25Parameters L and H are set to be equal to the credit-card-spending-weighted average of the adoption

numbers in the top half of Table 2, which explains the slight di
erence from 0
:55 and 0:75. In practice, the

actual reward rate could be even lower, because holders of reward credit cards may not claim all of their

rewards or the rewards may expire, but we do not have data on the rate at which consumers actually claim

their rewards.

26For more details about the CEX data source, see Harris and Sabelhaus (2000).

27We would like to thank Tim Chen (Nerdwallet.com), Leon Majors (Phoenix Marketing International),

and Jay Zagorsky (Boston University) for helping us clarify whether credit cards can be used for mortgage

payments.

28Zinman (2009b) compares the SCF with industry data and nds that the two sources match up well on

credit card charges and fairly well on account balance totals.

15

cuto
level of household income (denoted
I).

It is well known that consumption and income are distributed unevenly across households,

and this situation is evident in Table 5. Low-income buyers account for 81 percent of all

households but only 58 percent of transactions. Low-income buyers also tend to favor cash

payments: 70 percent of all households are low-income cash buyers, and 50 percent of all

transactions are conducted by low-income cash buyers. In addition, high-income households

have a disproportionately higher share of credit card transactions (about 13=42  31 percent)

than their population share (19 percent). All this shows that high-income households make

higher use of credit cards.29

Distribution of Households Distribution of Transactions

IL IH Total IL IH Average

Cash buyers 70 13 83 50 29 79

Card buyers 12 6 17 8 13 21

Total 81 19 100 58 42 100

Table 5: Distribution of households and transactions (percentage of total).

Three assumptions are needed to de ne the implicit transfers among households.

A-1 All households pay the same price, p, for the representative product (good or service);

that is, the merchant does not charge di
erent prices to cash buyers and card buyers.

A-2 The merchant passes through the full merchant fee to its customers via the retail price.

A-3 Rewards to card users are not funded by banks' revenue generated by borrowing activ-

ities.

The validity of these assumptions is an empirical matter and the data needed to verify them

are not available. One needs data on individual transactions that identify not only the

payment instrument but also the consumer who uses it and the merchant who receives it.

29The household units in Table 5 are representative agents created across heterogeneous households to

obtain a parsimonious aggregate representation of the data for modeling purposes. Households without

credit cards are literally cash-only households (where cash means non-credit-card). However, there are no

households that strictly use credit cards only, and most households use both cash and credit cards. Our

aggregate transfer calculations cannot account for this within-household heterogeneity, a re nement we leave

for future research.

16

Such matched consumer-merchant data are extremely rare, and may not even be sucient.

If consumers of di
erent income groups buy di
erent products within merchants, and if

merchants price those products not only according to their price elasticities of demand but

also by their probabilities of being paid for by cash versus credit, then consumer-merchant

data are needed at the level of detailed individual products (goods and services) as well.

Future research based on such rich and nely graded data would provide valuable re nements

of our calculations. However, Section 7 considers some alternative calculations that explore

the e
ects of relaxing these assumptions on the transfers.

3.3 Transfer de nitions

Our goal is to measure the actual transfers in the U.S. payments market and their e
ects on

consumer welfare. Thus, we de ne each transfer as the di
erence between the actual money

paid by a household toward merchant payment costs, on one hand, and the reference value

(amount of money) the household would pay if it faced the full cost of its payment choice in

the current payment environment, on the other. The actual money paid is the household's

share of the merchant's total cost of payments (Sd + Sh). The reference value of the

payment depends on the marginal cost of the good for the household. As shown in Section 4,

the marginal cost of producing the good (denoted ) is the same for all households but the

marginal cost of payment varies across households according to the household's payment

choice. Households paying by cash impose a marginal cost of   p for their transactions, and

households paying by credit card impose a marginal cost of   p for their transactions.

With this transfer de nition in mind, consider rst the transfer between cash and credit

card users. Let X denote the transfer made (or subsidy received, if the transfer is negative).

Then the transfer made by cash users (superscript h) is

Xh def =



Sh

S

􀀀

Sd + Sh

􀀀 Sh and xh def =

Xh

Nh

L + NhH

; (2)

where xh denotes the transfer per household, our preferred metric. The term of Xh in braces

is what cash users actually pay toward total merchant payment costs: the cash share of total

17

spending, (Sh=S) = 0:79, times the total merchant cost of transactions, (Sd + Sh) = $47

billion. Cash users indirectly pay a portion of the cost of credit card payments, (Sd) = $24

billion, because cash and credit card buyers pay the same equilibrium price, p, which will

be calibrated later using the model in Section 4. The last term of Xh (outside the braces) is

the total cost of cash transactions: that is, cash-handling costs, (Sh) = $22 billion.

Similar to (2), the transfer (or subsidy received, if the transfer is negative) made by credit

card users (superscript d) is

Xd def =



Sd

S

􀀀

Sd + Sh

􀀀 (LSdL

+ HSdH

)



􀀀 Sd and xd def =

Xd

NdL

+ NdH

: (3)

The term of Xd in braces is what credit card users actually pay toward total merchant

payment costs net of the rewards they receive. The rst term inside the braces is their

contribution to merchants' transaction costs: the card share of total spending, (Sd=S) = :21,

times the total merchant cost of transactions. The second term inside the braces adjusts for

credit card rewards, (LSdL

+HSdH

) = $8:5 billion. The last term of Xd (outside the braces)

is the total merchant cost of credit card transactions, which equals banks' fee revenue from

all credit card transactions.

The credit card transfer, equation (3), contains two components. One is the point-of-sale

(POS) transfer, which occurs at the merchant:

eX

d def =



Sd

S

􀀀

Sd + Sh

􀀀 Sd and exd def =

eX

d

NdL

+ NdH

: (4)

The second component is an adjustment for rewards, 􀀀(LSdL

+HSdH

), which are subtracted

from the POS transfer because rewards are rebated to credit card users by banks and reduce

the contribution of card users to total merchant payment costs. The rewards adjustment

to the POS transfer captures the portion of the overall transfer that occurs because credit

card users do not pay the full value of the rewards they receive. Instead, cash users pay for

part of the rewards, and this rewards-related transfer varies across income groups. Thus,

the POS transfer, which excludes rewards, understates the actual transfer occurring as a

18

result of credit card payments.30 Nevertheless, the POS transfer provides an informative,

lower-bound estimate of the transfer, so we report both estimates. Furthermore, the POS

transfer would be the appropriate measure if credit card users paid the full value of their

own rewards.31

Section 2.2 established a positive correlation between card use and income, which moti-

vates calculation of the transfer between low-income and high-income households. Similar

to the transfer de nitions given by (2) and (3), the transfers paid by each household income

group are

XL

def =



SL

S

􀀀

Sd + Sh

􀀀 LSdL



􀀀 (SdL

+ ShL

); (5)

XH

def =



SH

S

􀀀

Sd + Sh

􀀀 HSdH



􀀀 (SdH

+ ShH

): (6)

The rst terms in braces are what households actually pay toward total merchant payment

costs: the amounts of merchant payment costs borne by income groups L and H, respectively,

((SL=S) = :58 and (SH=S) = :42), less their credit card rewards, (LSdL

) = $2:7 billion and

(HSdH

) = $5:8 billion, respectively. The second terms are the total merchant costs of each

household's own payment choice: (SdL

+ ShL

) = $24 billion and (SdH

+ ShH

) = $23 billion.

Note that the total (aggregate) transfer among households by income level is the same as

between cash-using and card-using households:

X = XL + XH = 􀀀(LSdL

+ HSdH

): (7)

Similar to equation (4), the POS transfers between low-income and high-income house-

30See Appendix B for more details on this point. We especially thank Fumiko Hayashi, Bob Triest, and

Paul Willen for helping us to clarify our thinking about the transfer de nitions, especially the central and

crucial de nition in equation (3).

31A simple way to see this point is think of an alternative payment market in which merchants surcharge

credit card users for their rewards at the POS and then rebate the full rewards instantly to households using

credit cards. In this case, merchants would pay a fee to banks net of rewards, ( 􀀀 ), rather than paying

the full merchant fee and having banks pay rewards to households later.

19

holds are

eX

L

def =



SL

S

􀀀

Sd + Sh

􀀀 (SdL

+ ShL

) (8)

eX

H

def =



SH

S

􀀀

Sd + Sh

􀀀 (SdH

+ ShH

) (9)

and they omit the adjustment for rewards, which varies by income group. At the household

level, the relative magnitudes of the income group transfers are determined primarily by two

facts that favor high-income households: SdH

> SdL

and H > L.

3.4 Transfer estimates

Applying the benchmark speci cation and data described in Section 3.2 to the transfer

equations de ned in Section 3.3 yields the central results of this paper. Table 6 displays the

transfer estimates in billions of 2007 dollars and on a per household basis. These two types

of estimates are qualitatively equivalent but we focus on the latter. Recall that positive

(negative) numbers indicate that households using a payment instrument paid a transfer

(received a subsidy).

Total ($ Billions) Per household, total ($)

IL IH Total IL IH Average

Cash buyers 9:0 5:3 14:3 111 352 149

Card buyers 􀀀8:3 􀀀14:5 􀀀22:8 􀀀613 􀀀2; 188 􀀀1; 133

Total/Average 0:8 􀀀9:3 􀀀8:5 8 􀀀430 􀀀73

POS only ($ Billions) Per household, POS ($)

Cash buyers 9:0 5:3 14:3 111 352 149

Card buyers 􀀀5:6 􀀀8:7 􀀀14:3 􀀀414 􀀀1; 311 􀀀710

Total/Average 3:4 􀀀3:4 0 37 􀀀160 0

Table 6: Transfers in the payment market by household income and payment instrument.

To our knowledge, the results in Table 6 are the rst quantitative estimates for the

aggregate economy of theoretical measures of transfers between buyers stemming from the

choice of payment instrument. Two main conclusions can be drawn from the results.

20

Result 1. Cash payers subsidize credit card payers. The average cash-paying household

transfers $149 (xh = 149) annually to card users, and the average credit-card-paying house-

hold receives a subsidy of $1; 133 (xd = 􀀀1; 133) annually from cash users.

The annual transfer gap (di
erence) between the average cash and card users is $1
; 282

(xh 􀀀 xd = $1; 282), which represents 1:8 percent of median income across all households in

2007.

Result 2. Low-income households subsidize high-income households. The average low-income

household transfers $8 (xL = 8) annually to high-income households, and the average high-

income household receives a subsidy of $430 (xH = 􀀀430) annually from cash users.

The annual transfer gap (di
erence) between the average low-income household and the

average high-income household is $438 (xL 􀀀 xH = $438), which represents 0:6 percent

of median income across low-income households in 2007. By far, the bulk of the transfer

gap is enjoyed by high-income credit card buyers, who receive a $2; 188 subsidy every year.

Although low-income credit card buyers also receive a subsidy ($613) and high-income cash

buyers pay a larger transfer ($352) than low-income cash buyers, the greater use of credit

cards and receipt of rewards gives high-income households a non-trivial subsidy each year.

These transfer estimates, based on only two income categories (de ned by a cuto
of

$100; 000), signi cantly understate the magnitude of the transfer between the lowest- and

highest-income households. Dividing households into seven income categories instead, as

in Table 7, reveals that the transfer gap between the lowest-income households (less than

$20; 000) and the highest-income households ( $150; 000) increases to $771 per household

each year. The average lowest-income household pays $21 each year, and the average highest-

income household receives $750 each year, from the convenience use of credit cards. In

between, the transfer gap is nonlinear across groups|relatively at until household income

rises above $100; 000 annually, then sharply increasing in the highest categories. Thus, each of

a large number of lower-income households pays a relatively small dollar amount of transfer,

21

while each household of a small number of higher-income groups receives a relatively large

dollar amount of subsidy.32

Transfers paid

Income range POS Total

Under $20; 000 $32 $21

$20; 000{49; 999 $45 $26

$50; 000{79; 999 $35 􀀀$11

$80; 000{99; 999 $16 􀀀$61

$100; 000{119; 999 􀀀$11 􀀀$113

$120; 000{149; 999 􀀀$50 􀀀$207

Over $150; 000 􀀀$313 􀀀$750

Table 7: Transfers in the payment market by disaggregated income categories.

Section 4 develops a model to quantify the potential loss to consumer welfare result-

ing from these transfers. Before doing so, let us put the payment transfer estimates into

perspective by viewing them in the context of another public policy issue. The literature

on ination nds that the potential welfare gain of reducing steady-state ination from 10

percent to 0 percent ranges between 0:2 and 1:0 percent of the GDP (see Ireland (2009)

and Lucas (2000)). These estimates translate into an annual per household cost of $243 to

$1; 213 (using 2007 GDP data). Thus, the magnitude of the payments transfers would seem

to merit attention from policy makers similar to that devoted to controlling ination.

3.5 Sources of banks' income

This subsection decomposes banks' gross and net income from merchant fees, Sd, into

sources of revenue from each of the four buyer groups. We multiply gross income (revenue)

by the share of total spending of each group of buyers: ShL

=S, SdL

=S, ShH

=S, and SdH

=S. The

results appear in the rst panel of Table 8. We then compute rewards paid to credit card

32Table 7 implies that the transfers computed with only two income groups may be sensitive to the cuto

income level. We chose a cuto
of $100
; 000 because the transfer paid increases nonlinearly with income, so

a higher cuto
level is more representative of the transfer paid by the highest income groups. If the cuto

household income is $50; 000, then the low-income household pays $37 instead of $8, whereas the high-income

household receives $200 instead of $430.

22

users in the second panel of the table. The third panel reports the net income of banks from

merchant fees, that is, gross income ( rst panel) minus rewards (second panel).

Revenue from Merchant Fees

Total ($ billions) Per household ($)

IL IH Total IL IH Total

Cash buyers 12:0 7:0 19:9 149 469 199

Card buyers 2:0 3:1 5:2 149 473 256

Total 14:0 10:1 24:2 149 470 209

Rewards to Consumers (expenditure)

Cash payers 0 0 0 0 0 0

Card payers 2:7 5:8 8:5 199 877 423

Total 2:7 5:8 8:5 28 270 73

Net ($ billions) Net Per household ($)

Cash payers 12:0 7:0 19:0 149 469 199

Card payers 􀀀0:7 􀀀2:7 􀀀3:3 􀀀49 􀀀404 􀀀166

Total 11:4 4:3 15:7 120 200 135

Table 8: Banks' gross income sources and expenditure.

From Table 8 we can derive the following results about sources of banks' income from

merchant fees:

Result 3. Low-income households bear a disproportionately large burden of merchants' cost

of credit cards because they tend to use cash more often than high-income households. Cash

users pay 82 percent ( 19:9=24:2) of banks' gross income from merchant fees, and low-

income cash users pay 50 percent ( 12:0=24:2) of banks' gross income.

Result 4. Cash payers receive no rewards (naturally) and high-income households receive

the lion's share of credit card rewards. The average high-income card payers receive $877

in rewards annually, while the average low-income card payers receive only $199, less than

one-fourth as much.

Result 5. Banks earn negative net income from credit card users, as rewards paid exceed

revenues received from these households (net revenue of 􀀀$3:3 billion), but banks more than

23

o
set this loss with net income from cash-paying households (
$19:0 billion). Almost three-

quarters ( 11:4=15:7) of banks' net income is generated from low-income households, de-

spite the fact that the high-income group uses credit cards more than the low-income group

(13=21  60 percent in Table 5).

Overall, the picture painted by these data and results is one in which low-income cash payers

account for the bulk of the costs (merchant fee revenue) imposed by the payment choices

(credit card purchases) of mostly high-income households.

4. A Model of Cash and Card Users

To investigate the welfare consequences associated with the redistribution of income among

households, we construct an analytical model and then calibrate it. Endogenously deter-

mined variables will be denoted by lower case letters. Exogenous parameters will be denoted

by roman capital and Greek letters.

4.1 Buyers

There are NL low-income buyers and NH high-income buyers. Income levels are denoted by

IL and IH, respectively. Income group i buyers (i = L;H) are uniformly indexed by bi on

the unit interval [ i 􀀀 1; i], (where 0  i  1) according to the bene t they derive from

paying with a card relative to paying with cash, as illustrated in Figure 3 and described

in Section 2.3. Thus, bi measures the nonpecuniary bene t from paying with a card by an

income group i buyer who is indexed by bi. bi = i denotes buyers of income group i who

bene t the most from using a card. bi = i 􀀀 1 are income group i buyers who most prefer

paying with cash over card.

Buyers have an endogenous choice of paying with cash or paying with a card. Banks

(card issuers) reward card users by paying   p as \cash back," where 0 <  < 1 is the

fraction of the price p that is paid back to the buyer. Therefore, the e
ective price paid by

buyers belonging to income group i = H; L is

24

pb =

(

p(1 􀀀 i) paying with a card

p paying cash.

(10)

Thus, assuming that buyers spend their entire budget, low-income buyers perform IL=pb

transactions, whereas high-income buyers perform IH=pb transactions. Therefore, we de ne

the utility function of an income group i buyer who is indexed by bi by

Ubi =

8>><

>>:



(1 + bi)

Ii

p(1 􀀀 i)



paying with a card



Ii

p



paying cash,

for 0 <
 1: (11)

Equation (11) implies that a buyer's utility is increasing with the number of transactions

(income divided by price). In addition, if the buyer pays with a card, the buyer gains an

additional per-transaction bene t bi (loss for buyers indexed by bi < 0).

0

-

bL

0

-

bH

H

z }| {

H 􀀀 1

L

z }| {

L 􀀀 1

NL

Cash Card NH

 Card -

􀀀L 􀀀H

Figure 3: Distribution of buyers according to increased bene ts from paying with cards. Note:

Based on results presented later, the gure assumes NL > NH (most buyers are low

income) and L < H (more high-income buyers prefer paying with a card relative to

low-income buyers).

For each income group i = L;H, buyers who are indi
erent between paying cash and

paying with a card are found by solving



(1 +^b

i)

Ii

p(1 􀀀 i)



=



Ii

p



hence ^b

i = 􀀀i: (12)

Thus, buyers indexed by bi >^b

i pay with cards and buyers bi <^b

i pay cash; see Figure 3. In

the special case where i = 0, buyers indexed by ^b

i = 0 separate those who pay with cards,

bi > 0, from those who pay cash, bi < 0. This means that card rewards induce some buyers

who otherwise prefer to pay cash to use their cards in order to collect rewards.

25

The remainder of this section computes the number of card and cash payers as well as

the number of transactions made with each payment instrument. Recall that superscripts

\h" (for cash) denote cash payers, whereas superscripts \d" (for card) denote card payers.

In view of the \indi
erent" buyers described in (12) and Figure 3, the number of buyers

from group i who pay cash is

nhi

= [􀀀i 􀀀 ( i 􀀀 1)]Ni; hence

nh = nh

L + nh

H = NL[(1 􀀀 L) 􀀀 L] + NH[(1 􀀀 H) 􀀀 H]; (13)

which is the total number of buyers (both income groups combined) who pay cash.

Next, the number of buyers from income group i who pay with cards is

ndi

= ( i + i)Ni; hence nd = nd

L + nd

H = NL( L + L) + NH( H + H); (14)

which is the total number of buyers (both income groups combined) who pay with cards.

The total number of cash and card transactions made by each income group i = L;H,

denoted by , thi

, and tdi

in the model, multiplied by the price p, equals spending. Thus,

Sh

i = pthi

= nhi

Ii and Sd

i = ptdi

= ndi

Ii

1 􀀀 i

: (15)

4.2 Merchants

Merchants supply one \good," which could be either a product or a service. Free entry

results in normal (zero) pro ts. Similar to Wang (2010), we model a \mature" card market

in the sense that we assume that all merchants accept payment cards and cash. Thus, we

assume for simplicity that consumers do not have to search for a merchant who accepts their

preferred payment instrument. Let  denote the unit production (marginal) cost borne by

merchants, and recall that 0   < 1 denotes the e
ort (disutility) of the merchant from a

cash transaction relative to a card transaction. Thus, the merchant's disutility from handling

cash is   p. Under free entry, merchant pro ts are reduced to zero, so

0 = th[p(1 􀀀 ) 􀀀 ] + td[p(1 􀀀 ) 􀀀 ] hence p =

"

1

th

th+td (1 􀀀 ) + td

(th+td)(1 􀀀 )

#

; (16)

26

which is the equilibrium price in a competitive merchant industry. In the above, th[p(1 􀀀

) 􀀀 ] is the pro t from th cash transactions, and td[p(1 􀀀 ) 􀀀 ] is the pro t from td card

transactions, where p(1 􀀀 ) is the net price a merchant receives after paying the fee to the

card acquirer.

4.3 Calibrations

We rst use the model to calibrate the number of cash and card users within each group,

nh

L, nd

L, nh

H, and nd

H. These can be solved from (15) as functions of IL and IH. Because

the numbers of low- and high-income households are known, solving nh

L + nd

L = NL and

nh

H + nd

H = NH yields the calibrated values of IL and IH, which should be interpreted as

consumption expenditures because savings are not modeled.

Next, in view of Figure 3, the key parameters to be calibrated are the maximal bene ts

from using cards relative to cash, L and H. These two parameters are solved directly from

equations (13) and (14), assuming the card reward rates reported in Section 3.1. Transactions

data from the Survey of Consumer Payment Choice (SCPC) show that credit cards accounted

for 21:3 percent of consumer payments in 2008. Table 9 summarizes the model's parameter

values obtained under the above computations.

4.4 Equilibrium price and markup

Substituting the calibrated parameters from Table 9 into (13){(16), the equilibrium price

(16) becomes

pj =2%

=1%

= $27:56;  = 27:34; and L(p; ; ; ) =



p 􀀀 

p



100 = 0:82 percent; (17)

which is the Lerner's index commonly used for measuring markup over marginal cost. Thus,

our calibrations imply the following result:

Result 6. Convenience use of credit cards induces a retail price markup of 0:82 percent over

marginal cost (or 22=c over $27:34).

27

Parameter Notation Value Procedure

Cash e
ort
 0:5% Assumed

Merchant fee  2:0% Assumed

Card reward  1:0% Assumed

Rewards to low-income (cc-spend. weighted avg.) L 0:57% OSU 2007

Rewards to high-income (cc-spend. weighted avg.) H 0:79% OSU 2007

Number of credit card transactions td 43:9bn SCPC 2008

Total Spending Low-income NL  p  tL $3:33tr NIPA 2007

Total Spending High-income NH  p  tH $2:35tr NIPA 2007

Total Credit Card Spending Low-income NL  p  td

L $0:47tr SCF 2007

Total Credit Card Spending High-income NH  p  td

H $0:74tr SCF 2007

Low income level (excluding saving) IL $34; 879 Calibration

High income level (excluding saving) IH $110; 153 Calibration

Maximum card bene t (low income) L 0:137 Calibration

Maximum card bene t (high income) H 0:300 Calibration

Price p $27:56 Calibration

Marginal cost  $27:34 Calibration

Table 9: Computed values of model parameters and variables.

To assess the sensitivity of this result, Figure 4 plots the retail price markup as a function

of  and . The graph excludes all points in which banks make negative pro t, which is

depicted by the shaded triangle on the oor of the three-dimensional graph. Each relationship

between the markup and the two parameters is each approximately linear, but the markup

is more sensitive (steeper slope) to the merchant fee than to the reward rate. The reason

for this result follows from equation (16), which shows that the merchant fee a
ects price

directly because it is a cost for the merchant, whereas the reward rate has only an indirect

e
ect by making credit cards more attractive, thereby increasing the number of card users,

see equation (14).

The elasticity of the markup with respect to the merchant fee (evaluated at  = 2 percent,

 = 1 percent, and  = 0:5 percent) is 0:52. In other words, eliminating the merchant fee

(a change of 􀀀100 percent) would about halve the markup (from 0:82 percent to around

0:40 percent). These numbers are illustrated in Figure 4 by the point corresponding to no

28

Figure 4: Consumer price markup as a function of the merchant fee and the reward rate.

Note: The color gradations facilitate distinguishing among levels (dark red, the highest,

through dark blue, the lowest).

merchant fee and no rewards,33 in which case the markup would be 0:40 percent to cover

the costs of cash-handling ( = 0:5 percent) imposed by the 79 percent of the population

who pay cash. On the other hand, rewards have a much smaller e
ect on the markup; the

corresponding elasticity of the markup (measured at the same point) is only 0:014, meaning

that abolishing rewards (􀀀100 percent change) would yield only a 1:4 percent reduction in

the markup to 0:79 percent.

4.5 Banks' income from consumer credit cards

Banks' net income from income group i buyers is given by p  tdi

( 􀀀 i), i = L;H. Like the

transfers analyzed in previous sections, banks' net income is nonlinear with respect to the

merchant fee and reward rate. Banks' income from consumer credit card payments, net of

rewards, was $15:7 billion in 2007 (see Table 8). Thus, banks keep 65 percent of the revenues

from merchant fees, while consumers receive 35 percent in rewards.

33Since the markup responds very little to a change in the reward rate, the vast majority of the reduction

in the markup comes directly from the change in the merchant fee.

29

Figure 5 displays banks' net income from credit card spending as a function of the

merchant fee, , and the reward, . One interesting feature of the net income function

Figure 5: Banks' net income as a function of the merchant fee and the reward rate.

evident in the graph is that the iso-pro t lines are nearly linear with respect to  and .

Thus, banks can keep the same net income using di
erent combinations of merchant fee

and reward rates, while keeping ( 􀀀 ) approximately constant. This result is shown in

Figure 6. The dashed line shows the combinations of parameters for which bank pro ts are

0 1 2 3 4 5

0

1

2

3

4

5

Merchant fee (%)

Reward rate (%)

Iso−profit line for P=15.7bn

Iso−profit line for P=0

μ = 2%

r = 1%

μ = 1.36%

r = 0%

Figure 6: Banks' iso-pro t lines as functions of the merchant fee and the reward rate

zero|combinations of reward rates and merchant fees to the left of this line would result

30

in losses to the banks. Since the rates at which households actually receive rewards (i)

are both less than one, the slope of the iso-pro t curves is greater than one, meaning that

banks could o
er a higher reward rate than the merchant fee, since they earn merchant

fees on every credit card payment while they have to give rewards for only a fraction of

these transactions. The solid line, which runs through the benchmark point, shows the

combinations of parameters for which bank pro ts are constant at $15:7 billion. Reducing

the merchant fee and reward rate to the point ( = 1:36 percent,  = 0 percent) would

not alter bank pro ts, but would result in a lower retail price markup, as explained in the

previous subsection.

5. Consumer Welfare Calibrations

The analytical framework developed in this paper enables us to calibrate the consequences of

merchant fees and card rewards on consumer welfare stemming from the implicit monetary

transfers between the two income groups.34 In view of the buyers' utility function (11) and

Figure 3, aggregate consumer welfare of income group i buyers is given by

cwi(i; ) = Ni

8<

:



Ii

p



[􀀀i 􀀀 ( i 􀀀 1)] +



Ii

p(1 􀀀 i)



Z i

􀀀i

(1 + bi)dbi

9=

;; i = L;H; (18)

where the equilibrium price p is given in (16). The above expression consists of the sum

of utilities gained by cash users and card users (whose utilities must be integrated over bi

because buyers derive di
erent bene ts from card use). Therefore, total buyer welfare as a

function of the reward rate, , and merchant fee, , is given by cw(L; H; ) = cwL(L; )+

cwH(H; ), and is plotted in Figure 7.35

34This partial equilibrium model does not take into consideration how changes in banks' pro ts a
ect

consumption demand, because we do not have micro data on bank ownership (stocks). For this reason, we

do not extend this analysis to include social welfare. However, if household ownership of banks is increasing in

income too, then taking bank pro ts into consideration would likely magnify our central results. Section 7.3

and Appendix B discuss the implications of income changes due to redistribution of banks' pro ts.

35A more general formulation of aggregate consumer welfare could take the form of cw(cwL; cwH) =

(cwL)(cwH)1􀀀. For our limited calibration purposes, the additive function is sucient.

31

Figure 7: Consumer welfare as a function of the merchant fee and the reward rate (assuming


= 0:5)

Consumer welfare increases monotonically with the reward rate, keeping  constant.

The reason for this result is that rewards are pure windfalls received by the households

from the banks in this partial equilibrium setup. On the other hand, consumer welfare

falls very fast with an increase in the merchant fee. More precisely, the elasticity of the

welfare function with respect to the merchant fee evaluated at the benchmark (point C

on the graph, where  = 2 percent,  = 1 percent) is 􀀀0:0021, meaning that eliminating

the merchant fee (while leaving rewards unchanged) would increase aggregate consumer

welfare by 􀀀0:0021(􀀀100 percent) = 0:21 percent. However, this change is infeasible without

reducing  as well. The elasticity with respect to the reward rate at point C is 0:0006. Hence,

eliminating rewards, while leaving the merchant fee unchanged would lead to a 0:06 percent

decline in aggregate consumer welfare.

Using these elasticities, we can infer the welfare implications of certain changes in the pay-

ment fee structure. If, for example, the merchant fee is cut in half to 1 percent, the economy

would move to point B ( = 1 percent,  = 1 percent). Based on the aforementioned elastic-

ities, this move would entail a 0:105 percent (= 􀀀0:0021(􀀀50 percent)) increase in consumer

welfare. However, Figure 7 reveals that this is not the maximum attainable level of welfare.

32

A move from point B to point A ( = 0 percent,  = 0 percent) would further increase con-

sumer welfare, although this move would raise welfare by a smaller amount than the move

from point C to B. The elasticities calculated above con rm this. The welfare improvement

would amount to only a further 0:045 percent, which is the di
erence between the welfare

gain from another 1-percent reduction in the merchant fee and the welfare loss from the

elimination of rewards (0:0006(􀀀100 percent) = 􀀀0:06 percent).36 So, eliminating the mer-

chant fee, and hence rewards, would result about in a 0:105 percent + 0:045 percent = 0:15

percent increase in consumer welfare compared with the benchmark starting point.

The parameter
a
ects the shape of the utility function and hence the optimal transfer

levels. As
declines, the transfer between household income groups becomes less desirable

because the marginal utility loss from the low-income transfer becomes larger, while the

marginal utility gain from the high-income subsidy gets smaller. When applied to aggregate

data, as we do here, the parameter
can be interpreted equivalently as a measure of the

economy's aversion to income inequality (lower
means greater inequality aversion).

0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

−0.5

0

0.5

1

1.5

2

2.5

a

Merchant fee, reward rate (%)

Optimal μ

Optimal r

Figure 8: Consumer welfare-maximizing merchant fee and reward rate as functions of
(assuming

zero bank pro ts)

Figure 8 plots the welfare-maximizing values of the merchant fee and reward rate for

di
erent values of

and portrays the following result:

36This computation is slightly imprecise because we assume that the elasticity at point C is the same as

at point B. The exact calculation is given in Table 11 below.

33

Result 7. The merchant fee and card reward that maximize total consumer welfare decline

with an increase in the degree of concavity of buyers' utility function (11) with respect to the

number of transactions (a decrease in
).

Result 7 highlights the distortion in the income distributions caused by the merchant fee and

card use programs. When buyers' utility becomes more concave (
decreases), any transfer

from low- to high-income buyers has a greater impact on low-income buyers. For low values

of
, eliminating merchant fees and card rewards is optimal. In the opposite-extreme case of

linear utility, the loss to low-income buyers is smaller than the gain to high-income buyers,

so positive merchant fees and rewards become optimal.

However, even for high levels of
, such as linear utility (
= 1), the move from point C

to point A in Figure 7 would still be welfare improving. In fact, with a linear utility function,

welfare would increase by 0:26 percent (relative to the case in which
= 0:5). Whereas the

consumer optimum in this case would be at  = 2:66 percent and  = 3:79 percent, a move

to  = 0 percent and  = 0 percent would still raise welfare, because such a move eliminates

banks' net income, so all households would be paying lower prices.37

Finally, Figure 9 illustrates the combinations of merchant fee and card rewards such that

it is possible to reduce the merchant fee from  = 2 percent to  = 1:36 percent, and card

reward from  = 1 percent to  = 0, while keeping banks' net income constant and also

improving total consumer welfare. The consumer welfare maximum is at  = 1:36 percent

and  = 0 percent, the same point as depicted in the banks iso-pro t function in Figure 6.

6. Policy Implications

Our model and analysis suggest that aggregate consumer welfare likely can be increased

by reducing transfers between consumers, especially between low-income and high-income

37The reason why this improvement is bigger than the one in our benchmark model follows from the

di
erent shapes of the utility functions. In particular, a higher

results in higher marginal utilities, so the

welfare e
ects of zero banks' net income are magni ed.

34

1 2 3 4 5

109.6784

109.6785

109.6786

Merchant fee (%)

Consumer Welfare (log scale)

μ = 2%

r = 1%

μ = 1.36%

r = 0%

Figure 9: Welfare-improving fee and reward reductions along banks' iso-pro t line

consumers. While it is natural to consider public policy initiatives in this endeavor, our

research and discussions suggest preemptive actions that private sector agents (households,

merchants, and banks) could take that would reduce the transfers. However, if private

agents are not willing or able to take these actions to reduce the transfers, then public policy

makers may wish to enact policies that would do so. Given the limitations of our model and

analysis, we cannot provide precise policy recommendations that would necessarily optimize

social welfare. Nevertheless, our research suggests some general principles and implications

pertaining to consumer welfare that may be useful for policy deliberations:

 Cost-based pricing|One condition supporting the transfers is uniform pricing across

payment instruments. Policies that would allow and encourage merchants to charge

di
erential prices according to the costs imposed by payment instruments could help

to reduce the transfers by reducing payment cross subsidies. Eliminating the NSR

would seem to be an obvious option, but it may not be a sucient incentive to induce

di
erential pricing (for example, see Bolt and van Renselaar (2009)).

 Full information|Another condition supporting the transfers is the lack of full infor-

mation about about merchant fees and other aspects of payment costs that have an

impact on retail prices and consumer welfare. Policies that would require merchants,

banks, or credit card companies to fully disclose fees, costs, and price markups to

35

consumers could help to reduce transfers by giving consumers the incentive to make

optimal payment choices.

 Redistribution|The transfers can be reduced by compensating low-income households,

using tax policies to redistribute money from high-income households according to

credit card use and receipt of rewards. Direct methods may be complicated and costly,

but tax deductions for reward contributions may be feasible.

 Competition|If there is inadequate competition in the credit card market, then gov-

ernment e
orts to promote alternative payment instruments could help to reduce the

transfers. Expanding access to low-cost existing networks, such as the Automatic

Clearing House (ACH), is one possibility.

 Regulation of fees and rewards|The transfers likely can be reduced by regulating the

merchant fee, but two important caveats apply. First, economists would caution as

usual that regulators may have diculty determining the optimal fee, so regulation

of the merchant fee could actually reduce consumer welfare if the wrong level of the

fee were selected. Second, and unique to our analysis, regulators should consider the

merchant fee and reward rate simultaneously.

Of course, these policy implications and ideas would require more research and formulation

before they could be considered and adopted.

Finally, these policies to reduce transfers are closely related to recent policies enacted to

regulate payment card interchange fees worldwide. Policy makers in Australia and Spain, as

well as the European Commission, have already taken actions to limit the interchange fees

associated with credit cards. Actions taken by various countries are discussed in Bradford and

Hayashi (2008). The recent U.S. nancial reform bill (ocially, the \Dodd-Frank Wall Street

Reform and Consumer Protection Act" of 2010), signed into law on July 21, 2010, includes

the Durbin Amendment, giving the Federal Reserve responsibility for regulating interchange

fees associated with debit cards. In each of these cases, regulation of interchange fees was

36

motivated in part by concerns over an alleged lack of competition in payment card markets.

Our analysis provides a di
erent but complementary motivation|income inequality|for

policy intervention.

Given that policy makers have been and will be focusing on regulating interchange fees, we

can provide some potentially helpful information about the properties of merchant fees and

rewards for policy makers who wish to take these parameters into consideration. Table 10

summarizes the key elasticities with respect to the merchant fee and the reward rate in

the model. Recall from Section 4.4 that regulating the merchant fee without changing the

Variable Merchant Fee Reward rate

Markup 0:52 0:014

Transfer paid by low income (XL) 5:99 􀀀3:560

Transfer received by high income (􀀀XH) 0:50 0:658

Consumer Welfare 􀀀0:0021 0:0006

Table 10: Key elasticities (at  = 2%;  = 1%) with respect to  and  in the model

reward rate would have a much larger e
ect on the price markup and consumer welfare

than regulating the reward rate without changing the merchant fee ( rst and last lines

of Table 10). However, it is important to remember that optimal policy would require

simultaneous regulation of the merchant fee and the reward rate. It would also require an

analysis and treatment of household claims to banks' pro ts, which we have not considered

here.

Table 11 provides a guide to the e
ects of policy changes by showing the percentage

changes in consumer welfare associated with reductions in merchant fee and reward rates

below their benchmark values ( = 2 percent and  = 1 percent). A positive number

indicates an increase in consumer welfare. The maximum possible increases in consumer

welfare are found at the top of each column where banks' net income is the smallest for the

column.

37

Reward rate ()

 0 0:25 0:50 0:75 1:0

0:00 0:147    

0:25 0:121 0:137   

0:50 0:095 0:111 0:128  

0:75 0:069 0:085 0:101 0:118 0:134

1:00 0:043 0:059 0:075 0:091 0:107

1:25 0:018 0:033 0:049 0:065 0:081

1:50 􀀀0:008 0:007 0:022 0:038 0:054

1:75 􀀀0:034 􀀀0:019 􀀀0:004 0:011 0:027

2:00 􀀀0:060 􀀀0:045 􀀀0:030 􀀀0:015 0:000

Table 11: Percentage changes in consumer welfare associated with reductions in merchant fee and

reward rates below their benchmark values ( = 2% and  = 1%).

7. Quali cations and Extensions

Our analysis relies on several assumptions and simpli cations imposed due to lack of data

or for tractability. Relaxing these restrictions could alter the magnitudes of the transfer

estimates. This section explores the potential impact of these restrictions, and provides

some quali cations and extensions to the central results.

7.1 Transfer accounting assumptions

Section 3.2 lists three key assumptions underlying the estimates of the transfers between

cash and card payers and between low-income and high-income households. In reality, each

assumption may not hold exactly. So we designed some alternative transfer calculations to

approximate more realistic conditions in the payments market that would occur if we relaxed

the assumptions. Table 12 reports the results of our alternative transfer calculations and

their deviations from the benchmark estimates based on two household income categories.

To simplify the analysis, columns three and four report only the transfer gap, which we

de ned earlier as the di
erence between the average transfer per low-income household and

the average transfer per high-income household. The remaining two columns report the

percentage change for the alternative transfer estimate relative to the benchmark estimate.

38

Transfer Gap ($) Change (%)

Assumption Alternative Card Income Card Income

{ Benchmark (two income categories) 1; 282 438 { {

A-1 Partial price di
erentiation 1
; 234 365 􀀀3:7 􀀀16:7

A-2 Imperfect competition (merchants) 1; 004 421 􀀀21:7 􀀀3:9

A-2a Price markup (10%) 1; 292 494 0:8 12:8

A-2b Bargaining power over  995 372 􀀀22:4 􀀀15:1

A-3 Interest funding of rewards 1; 148 314 􀀀10:4 􀀀28:4

Table 12: Changes in the transfer gap estimates due to relaxing the underlying assumptions.

First, we relaxed assumption A-1, one price for all buyers, and instead allowed for partial

price di
erentiation between cash and credit card buyers. Price di
erentiation could arise for

many reasons, including the following: the representative merchant could surcharge credit

cards or discount cash purchases; there may exist heterogeneous merchants and/or products

for which only cash or only credit cards are accepted; or low-income and high-income house-

holds may shop at di
erent merchants so that cash and credit card purchases are segregated.

Each of these reasons can be simulated in observationally equivalent fashion by excluding a

portion of cash or card spending (or both) from the transfer calculations. We excluded 4:2

percent of consumption from broad NIPA categories that are likely paid for by cards only or

cash only.38 With partial price di
erentiation in the economy, the card transfer gap falls by

3.7 percent and the income transfer gap falls by 16.7 percent.

Next, we relaxed assumption A-2, complete (100 percent) pass-through of the merchant

fee to consumers, and instead allowed for the pass-through to be more or less than complete

by introducing two forms of imperfect competition. One form is classic market power for

the merchant, which results in a traditional price markup over marginal costs and the cost

of the payment instrument. The transfer formula for this price markup is:

Xi

def =



 + 1

Si

S

(Sd + Sh) 􀀀 (Sd

i + Sh

i ) 􀀀 iSd

i i = L;H: (19)

We simulate the e
ects of a 10-percent markup based on an elasticity of
 = 10. The

38We subtracted from aggregate consumption the spending on \household furnishings and equipment,"

\air transportation," and \accommodations," which are likely paid mostly with credit cards.

39

other form is market power held by a very large merchant (for example, Walmart) over

banks, giving the merchant leverage in bargaining over the merchant fee. We simulated

this possibility by reducing the aggregate merchant fee 0.5 percentage points to 1.5 percent.

The price markup of 10 percent increases the income transfer gap by 12:8 percent because

the pass-through of payment costs in the retail price is more than 100 percent; the card

transfer gap is only slightly higher. In contrast, bargaining power over the merchant fee

reduces the card transfer gap by 22:4 percent and the income transfer gap by 15:1 percent.

Combining these two di
erent e
ects of market power, we see that imperfect competition

tends to a
ect primarily the card transfer gap (21
:7 percent lower) but leaves the income

transfer gap largely the same (3:9 percent lower).

Finally, we relaxed assumption A-3, no funding of credit card rewards from revolving

debt activity, and instead assumed that interest revenue from revolving debt held by high-

income households is used to fund rewards paid to low-income households. As we show in

more detail below, this alternative transfer calculation is not supported well by the data,

even though it is often alleged in the literature. In any case, this alternative does not a
ect

the card transfer gap, but it reduces the income transfer gap by 28:4 percent because of the

direct transfer of interest payments from high-income to low-income household rewards.

One clear overall conclusion emerges from these alternative transfer calculations: both

transfers remain economically signi cant even after adjusting for alternative conditions in

the payments market. Although relaxing some assumptions leads to reductions in some of

the estimates, the adjusted transfers are still about three-quarters (or more) as large as their

benchmark values. Furthermore, we have omitted from the benchmark transfer calculations

two very important features of credit card markets|redistribution of bank pro ts (discussed

in Appendix B) and business credit card use (discussed below)|that likely would increase

the transfer estimates. We believe that these increases to the transfer estimates are most

likely greater (in absolute value) than the reductions reported in Table 12.

40

7.2 Revolving credit

It is important to emphasize once more that our model and analysis focus on the convenience

use of credit cards and do not incorporate a role for revolving credit. Revolving credit is an

important part of the value of credit cards to the economy, and we support future research

that expands our analysis in this direction. We also recognize that debt activity could be

another source of revenue for banks and credit card companies. This subsection explores the

evidence on this issue further to reassure the reader that we have not grossly mischaracterized

the transfers.

High interest and penalties paid by credit card borrowers on revolving debt may directly

or indirectly fund some of the bank issuers' expenses on card rewards. In fact, Chakravorti

and Emmons (2003) demonstrate an equilibrium in the market for credit cards (as opposed

to debit and charge cards) in which the \convenience use" of credit cards by nonborrowing

consumers is subsidized by liquidity-constrained consumers who borrow on their credit cards

and pay high interest. Chakravorti and Emmons's results explain that borrowers pay high

interest rates on credit card debt because this interest is used to reward all credit card users,

including those who avoid interest charges by paying their full balances on time. However,

the evidence suggests that rewards are funded at least partly by merchant fees. Levitin

(2007) reports that 44 percent of interchange fees goes to fund reward programs. Hayashi

(2009) also investigates the degree to which card reward programs are nanced by merchant

fees, but does not draw de nite conclusions. In our calculations, rewards make up about 35

percent ( 8:5=24:2) of merchant fees. If we look at interchange fees instead of merchant

fees, subtracting 0:5 percent (acquiring banks' pro t) from 2 percent we compute 35 percent

times 4=3  47 percent, which is fairly close to the result in Levitin (2007).

The SCF provides data on credit card revolving debt, reported in Table 13, that help one

to evaluate the idea of Chakravorti and Emmons (2003). The survey poses two questions

related to revolving credit, and both show surprisingly little di
erence between low-income

and high-income households. First, the SCF asks whether respondents usually pay o
their

41

balances. For high-income households, 30:7 percent answer \sometimes" or \hardly ever,"

while for the low-income group, 32:9 percent provide the same answer. The second ques-

tion is about the outstanding balance after the last payment, showing that 43:2 percent

of low-income and 47:5 percent of high-income households carried debt. The similarity in

revolving credit between income groups belies the conventional notion that credit card debt

is predominantly a problem for low-income households.

Low-income High-income

Revolving debt (reported incidence) 32:9% 30:7%

Revolving debt (actual incidence) 43:2% 47:5%

Revolving debt (revolvers) $6; 243 $11; 709

Interest rate (card holders/revolvers) 12:90%=12:20% 12:85%=11:15%

Annual interest payment (debt rate) $788 $1316

Aggregate interest payment (payment households) $30:9 billion $13:4 billion

Aggregate annual rewards (from Table 8) $2:7 billion $5:8 billion

Table 13: Revolving credit activity by household income group

The remainder of Table 13 shows the implications of revolving credit for interest revenues

to banks. Among revolvers, high-income households carry about twice as much revolving

debt as low-income households, but their credit cards have interest rates about 1 percentage

point lower.39 The last two rows of Table 13 reveal that both income groups pay more than

enough interest to cover the credit card rewards earned by the group. Thus, it seems unlikely

that interest from either group cross-subsidizes the rewards of the other, so we conclude that

the transfer calculations based only on convenience use of credit cards are likely accurate.

7.3 Other extensions

We close this section with a brief discussion of some extensions to our model and some

analysis that we leave for future research.

39The interest rates in Table 13 are for all credit card holders (the rst rate shown) and the debt-weighted

average for all revolvers (the second rate shown). The other gures in the table, except for those shown in

the last two rows, are averages over the entire income group.

42

Bank pro ts: We have not incorporated household ownership of banks (including card

companies). In our analysis, banks make $15:7 billion of undistributed pro ts on consumer

credit card services, which would be distributed to households in reality. Because the wealth-

iest 20 percent of the U.S. population holds the majority of all stocks, bank pro ts from

merchant fees likely would be distributed disproportionately to high-income households.

Thus, incorporating household ownership of banks is likely to increase the transfers from

low-income to high-income households.

Business credit cards: We use data on credit card use by consumers only. The Call

Report data on total U.S. credit card transactions indicate that total credit card spending

by business (and including government) is about equal to consumer credit card spending. If

businesses used credit cards at the same establishments as consumers, they would impose

further costs on the merchants and raise retail prices even more. If businesses (and their

pro ts) are more likely to be owned by high-income households, then incorporating business

use of credit cards into the analysis is likely increase the transfers from low-income to high-

income households.

Congestion (externality) e
ects:
Murphy and Ott (1977) suggests that cash buyers

impose more costs on merchants' sales sta
s than on card users. If cash transactions take

signi cantly longer to handle than credit card transactions, cash users may impose an ex-

ternality on card users by slowing them down at the point of payment. This externality

would o
set, at least partly, the transfer from cash users to card users. However, the avail-

able data on the time it takes to handle a transaction by payment method do not provide

strong support for this view.40 It is possible that cash congestion e
ects may be relevant for

highway toll booths, as discussed in Amromin, Jankowski, and Porter (2007). But electronic

40According to a 2000 study by the Food Marketing Institute, titled \It All Adds Up: An Activity Based

Cost Study of Retail Payments," a credit card transaction takes longer to handle than a cash transaction:

49 seconds compared to 29 seconds. However, a 2006 study by MasterCard International titled \MasterCard

PayPass: The Simpler Way to Pay," nds that the average cash transaction is slower than the average credit

card transaction if no signature is required: 34 seconds compared to 27 seconds.

43

toll transponders that serve as a faster alternative to cash are not credit cards, and the

proportion of toll payments is relatively small.

Credit card annual fees: Card fees are another potential source of revenue to fund card

rewards that could a
ect the transfer estimates. If credit card holders pay for their rewards

with high annual fees, then our transfer calculations would overstate the transfers. However,

this possibility is unlikely to be a major factor. According to the 2003 Synergistics Credit

Card Market survey, low-income households paid an average annual fee of $5:7, while high-

income households paid $7:7. These data imply trivial changes in the transfer estimates.41

The preceding list of extensions suggests that the magnitudes of our estimates and results

for transfers from low-income to high-income consumers may be altered quantitatively by

future research. However, if anything, the qualitative nature of the regressive transfer is

almost surely robust and the quantitative estimates are likely to increase relative to our

benchmark.

8. Conclusion

We proposed an accounting methodology to calculate two types of implicit monetary transfers

occurring in a simpli ed representation of the U.S. payments market: 1) the transfer between

cash buyers and credit card buyers; and 2) the transfer between low-income and high-income

households. Both of these transfers are estimated to be economically signi cant and robust

to potential changes in the assumptions underlying the accounting methodology.

We also built an empirically tractable theoretical model of payment for consumption that

includes all of the salient and economically important features of U.S. credit card payments.

We calibrated this model with the best, most detailed data available to us and derived

estimates of the average payment, retail price markup over marginal cost, and nonpecuniary

41Including credit card annual fees would reduce the card transfer gap by 0:6 percent to $1274, and reduce

the income transfer gap by 0:5 percent to $436.

44

utility bene t of card use over cash use. The results are remarkably plausible given the

relative simplicity of our data and model.

Extending our model and analysis with better data and more realistic features of the

credit card market surely would provide more re ned quantitative estimates of the two

transfers. However, we are con dent that the qualitative existence of these two transfers is

robust to changes in the model and data. On balance, our estimates of the transfers likely

understate the true values of the transfers, especially between income classes. Taking into

account the quantitative impact of all potential improvements and extensions to the data

and model, it is most likely that including in future research the factors we omitted from

this analysis will yield higher estimates of the transfers.

Appendix A Data

To get total consumption expenditure, we looked at the National Income and Product Ac-

counts (NIPA) for 2007. From the Personal Consumption Expenditure gure, we subtracted

a number of subcategories, where we believe that the transfers analyzed in the paper did

not take place, because assumption A-1 was not satis ed. Table 14 below details these

calculations for our benchmark model and for the alternative simulation with partial price

di
erentiation (the second row in Table 12): This adjustment resulted in dropping $4
:13

trillion of personal consumption expenditures from the headline gure of $9:83 trillion. The

drawback of using NIPA data is that we cannot break down consumption expenditure by

income categories. To do that, we used the 2008 edition of the Consumer Expenditure

Survey (CEX). Tables 2 and 2301 of the 2008 CEX contain the most detailed breakdown

available of consumption by income. To make our calculations consistent with our NIPA

consumption spending gure we had to take the same spending categories out of the CEX

consumption gure as we took out of the NIPA. Unfortunately, the subcategories in the CEX

and the NIPA do not map into each other one-for-one. So, from the CEX \Average annual

expenditure" gure, we took out the entire \Healthcare" category as well as \Mortgage and

45

Line Category name Amount ($ Billions)

1 Personal consumption expenditures $9; 826:4

Benchmark model

29 Food produced and consumed on farms $0:4

46 Net expenditures abroad by U.S. residents $6:1

50 Housing $1; 472:9

60 Health care $1; 465:4

84 Food furnished to employees (including military) $14:1

87 Financial services $507:9

93 Net health insurance $158:3

106 Social services and religious activities $134:3

109 Foreign travel by U.S. residents $113:9

111 NPISHs $254:2

Total adjustments made $4; 127:5

Partial price di
erentiation

8 Furnishings and durable household equipment $277:7

74 Air transportation $51:6

85 Accommodations $80:8

Additional adjustments made $410:1

Table 14: Adjustments to PCE gure using NIPA Table 2.4.5, revised, August 5, 2010.

interest charges," \Property taxes," \Rented dwellings," \Cash contributions," and \Pen-

sions and Social Security." (Expenditures on nancial services are not measured in the CEX

at all.) Once we had the relevant consumption and income gures from the CEX (readily

available in Tables 2 and 2301 of the CEX publication), we could construct the average

propensity to consume by each income category (except for the bottom income group). For

the lowest income group, where consumers' average income was negative, the average total

consumption expenditure per weighted respondent was matched so that it was equal to that

of the second-lowest group. These average propensities could then be multiplied by the in-

come gure in the Survey of Consumer Finances (SCF) 2007 to yield an estimate of total

consumption expenditure by income group. We measured household income as the sum of

variables x5702, x5704, x5706, x5708, x5710, x5712, x5714, x5716, x5718, x5720, x5722, and

x5724. To make the resulting consumption number consistent with the NIPA data, we also

multiplied the resulting number by a scalar so that it matched our adjusted total Personal

46

Consumption Expenditures gure.

Total annual credit card spending was computed as the sum of the values data gathered

in response to questions in the SCF asking about consumers' total use of credit cards in the

past month, x412, x426, x420, and x423, all multiplied by 12. For the partial price discrimi-

nation scenario we subtracted \Other lodging" and \Household furnishings and equipment"

spending by income group from their respective total credit card spending. The gures for

total annual credit card transactions were taken from Table 19 (monthly credit card use

multiplied by 12) in SCPC 2008 (Foster et al. (2009)).

Appendix B Transfer Accounting Details

To understand better why it is appropriate to adjust the POS transfer for rewards, consider

the aggregate accounting of the complete ow of funds among households, merchants, and

banks. The revenue from merchant fees is paid to banks, which then distribute rewards to

households that use credit cards. Thus, banks' pro ts () are:42

= Sd 􀀀 (LSdL

+ HSdH

): (20)

Viewed this way, credit card rewards act as a claim on banks' pro ts that is paid to credit card

users only, rather than to all owners of banks, and the distribution of rewards precedes the

distribution of pro ts to owners of banks.43 Because households own banks (either publicly

or privately), banks' pro ts ultimately are income for households. Thus, rewards represent

a transfer of pro ts and dividend income from owners of banks to credit card users who may

or may not be owners of banks. Let Di denote the dividends received by household type i.

After rewards are paid to credit card holders, the distribution of pro ts to the owners of

banks is

= Dh + Dd: (21)

42In this equation, we omit the costs of providing credit card services for simplicity and clarity. If banks

are perfectly competitive, then these pro ts would be zero.

43In a sense, owners of banks have subordinated claims to pro ts and credit card users have primary

claims. However, this would be irrelevant if credit card users were the sole owners of banks.

47

Because our computations omit the distribution of banks' pro ts given in (21), the sum of

transfers equals the negative amount of rewards,

X = Xh + Xd = 􀀀(LSdL

+ HSdH

) < 0; (22)

rather than zero. Therefore, estimates of the full aggregate transfer between cash-paying

and card-paying households depends crucially on the structure of ownership of banks by

households. In contrast, the sum of transfers at the point of sale,

eX

= Xh + eX

d = 0; (23)

does equal zero and does not depend on the ownership of banks.

Data on household ownership of banks by household payment choice and household in-

come are not available, so we cannot estimate the full aggregate transfer. However, unless

cash-paying households own a large portion of the banks, the full aggregate transfer likely is

greater than the POS transfer for two reasons: (1) the dividend income of cash-paying house-

holds is reduced by the payment of rewards to credit card users; and (2) the post-rewards

distribution of dividend income to households may not be proportional to the payment costs

imposed on merchants by household payment choices. The credit card transfer, equation (3),

includes rewards as an estimate of (1), but it does not include an estimate of (2). However, if

the ownership of banks is positively correlated with income, the net e
ects of bank ownership

on total (pre-reward) pro ts is likely to make the full aggregate transfer at least as large as

the rewards-adjusted transfer. The rewards-adjusted transfer also allows evaluation of the

independent e
ects of changes in the merchant fees versus changes in rewards. Clearly, more

data and additional research in this area would produce more complete and re ned estimates

of the full aggregate transfer.

Transfer equations (2) and (3) can be rewritten using the de nitions above to clarify the

role of rewards in the transfers. Let wh = (Sh=S) and wd = (Sd=S) denote the spending

shares of cash and card users, respectively, so that wh+wd = 1, and recall that Sd = SdL

+SdH

.

Merchant fee revenue is divided between credit card users (in the form of rewards) and owners

48

of banks (in the form of pro ts), so that  = ( + ), where  is pro t expressed as a rate.

Substituting this identity into the transfer equations, and then collecting and rearranging

terms, yields

Xh def =



wh

Sd + Sh

+ wh(SdL

+ SdH

)

􀀀 Sh (24)

Xd def =



wd

Sd + Sh

+ (wd 􀀀 L)SdL

+ (wd 􀀀 H)SdH

􀀀

Sd + Sd

: (25)

The structure of the rewritten transfer equations mirrors the original equations. In both

equations, the rst term in braces represents what payment users actually pay toward total

merchant payment costs, and the second term (outside braces) represents the merchant cost

of the household's payment choice. With regard to rewards, it is now clear from equation

(25) that the credit card transfer represents the amount of imbalance between the rewards

portion of credit card costs borne by the merchant (􀀀Sd), on the one hand, and the portion

of that cost paid by credit card users ((wd 􀀀 L)SdL

+ (wd 􀀀 H)SdH), on the other. The

portion paid by credit card users clearly shows that card users do not pay the full value of

their rewards: (wd􀀀L) = (0:21􀀀0:55) = 􀀀0:34 and (wd􀀀H) = (0:21􀀀0:75) = 􀀀0:54.

Appendix C Sensitivity Analysis

The following sections present the sensitivity analysis to changes in H and . Since we

are not aware of any study that has directly estimated H, we would like to see how our

assumption that richer people derive higher utility from using credit cards a
ects our results.

Also, as noted above, some empirical studies nd values that di
er from our estimates for the

costs of handling the payment instruments that we labeled as \cash," and these di
erences

could have important implications for our results.

When thinking about the welfare implications of di
erent parameter values, one has to

look carefully at the utility of all four groups in the model: (i) low-income cash users, (ii) low-

income card users, (iii) high-income cash users and (iv) high-income card users. The di
erent

parameter values considered below lead to di
erent estimates of the transfers between these

49

groups. In general, since our social welfare function is utilitarian, a redistribution to groups

with higher marginal utility will be desirable. With our concave individual utility functions,

low-income households will have higher marginal utilities, but the (1+bi) (with bi > 0) term

in card users' utility will raise their marginal utility above cash users' within their respective

income groups.

C.1 Sensitivity analysis with respect to H

We will now analyze what would happen if H decreased all the way to the level of L.

Having H > L means two things in the model: (i) a higher share of card users in the

high-income group (see equation (14)) and (ii) a higher average marginal utility of card

users in that income category. The former means that for H > L, the cash-payer-to-

card-payer transfer will amplify the redistribution of income between the income groups as

well. Intuitively, there will be more card payers who underpay in the high-income group,

so the cash payers (in both income categories) will have to overpay by more, but with the

number of card payers in the low-income category xed (for a given L), this overpaying will

result in a cross-subsidy from low-income households to their high-income counterparts. For

concave utility functions, this redistribution will lower total consumer welfare. At the same

time, a higher H also results in a higher utility gain from redistributing money from cash

users to card users within the high-income group. Remember that in both income groups

card payers derive higher marginal utilities from an additional transaction (for a given t),

so a redistribution from cash to card payers within each income group is welfare increasing

until the marginal utilities of cash and card users within the income groups are equalized.

As H increases, this utility gain is traded o
against the utility loss from a simultaneous

redistribution of income from low- to high-income groups.

The top panel of Figure 10 helps to gauge the e
ect of a change in
H on the aggregate

consumer welfare function. The mean change in the consumer welfare function has the exact

same shape as the maximum change (not shown) or the change at the point of ( = 2

percent,  = 1 percent). This nding indicates, that changes in H will not a
ect the shape

50

of the consumer welfare function drastically, so we expect our results to remain robust to

changes in H. The bottom panel of the same gure shows that the shape of the transfers

paid by the low-income group changes with the value of H, as we would expect based on

the discussion above, but the magnitude of the transfer at  = 2 percent and  = 1 percent

stays fairly constant.

0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

11.7

11.8

11.9

12

12.1

12.2

12.3

bH

Percentage (%)

Change in consumer welfare compared to benchmark

Mean change

Change at μ=2%, r=1%

0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

−20

−10

0

10

20

bH

Dollar billions

Change of XL

Min XL Max XL At μ=2%, r=1%

Figure 10: Welfare and transfers as a function of H

Figure 11 plots the welfare-maximizing level of  as a function of H and , illustrating the

story about the within- and across-income-group redistribution outlined above. A higher H

leads to a relatively higher number of card payers among the rich, and thus more of the cash-

to-card-payer redistribution becomes also low-income-to-high-income redistribution. Since

this latter is detrimental to aggregate welfare, the optimal level of  decreases with H to

curtail the amount of cash-to-card-payer redistribution.

C.2 Sensitivity analysis with respect to 

According to Figure 12, changes in  lead to changes in the consumer welfare function that

are of similar magnitude to the changes produced by di
erent values of
H. Again the upper

51

Figure 11: Optimal merchant fee as a function of H and 

panel of Figure 12 suggests that the shape of the consumer welfare function does not change

by much as  takes on di
erent values. Surprisingly, the redistribution also stays fairly

constant as  changes. From Equation 5 one can see that

@XL

@

=

SL

S

Sh 􀀀 ShL

=

SL

S

Sh 􀀀

ShL

ShSh = 􀀀0:05  Sh;

where the last line makes use of the gures in Table 5. In words, a change in  changes low-

income households' contribution to the costs imposed and to the costs paid by roughly the

same amount. A rise in the cost of handling cash leads to a redistribution from card to cash

payers, just as the increase in the merchant fee leads to a transfer from cash payers to card

payers. Again, the no-surcharge rule forces merchants to recover the higher costs imposed by

cash payers by charging higher prices to all customers, so as  increases, the price paid by card

users will increase, even though their purchases do not impose any additional costs to the

merchants. Since this transfer means a redistribution from high- to low-income households

(with H > L), it can increase social welfare as long as it helps to equalize marginal utilities

between the income groups. As can be seen from Figure 11, however, this redistribution can

become ineciently high for high values of , which would then validate a nonzero merchant

fee to redirect some of the transfer to low-income households back to high-income households.

52

0.5 1 1.5

11.8

11.9

12

12.1

12.2

Cash−handling cost (e) in %

Percentage (%)

Change in consumer welfare compared to benchmark

Mean change

Change at μ=2%, r=1%

0.5 1 1.5

0

15

Cash−handling cost (e) in %

Dollar billions

Change in XL

Min XL Max XL At μ=2%, r=1%

Figure 12: Welfare and transfers as a function of 

However, in our benchmark model with a high H, a 1.6 percent cash-handling cost would

still not warrant a positive merchant fee to maximize consumer welfare. Also, as noted above,

for high cash-handling costs the optimal merchant fee changes markedly with different values of H, as the di
erence between
L and H (di
erence between the fraction of card users in

the two income groups) increases the between-income group redistribution. If there were no

redistribution between income groups, the transfer resulting from cash-handling costs would

decrease welfare, since it would channel income from (high marginal utility) card payers to

(lower marginal utility) cash payers. This is why, in the case of equal s and high , a

high merchant fee (0:9 percent) would be optimal to o
set the transfer from card payers

to cash payers. As H increases, however, the redistribution towards cash payers becomes

more desirable, as it becomes a subsidy from high-income to low-income households, while

the redistribution caused by the merchant fee becomes less desirable, since it works in the

opposite direction. Note that in Figure 11, a high merchant fee is optimal only for low H

and high .

Cash-handling costs play an important role in determining the markup. Because of the

high fraction of cash payers (approximately 86 percent in the low- and 69 percent in the high-

53

income group), the markup moves almost one-for-one with . Figure 13 plots the markup

as a function of cash-handling costs and the merchant fee. Note that while the merchant

fee goes from 0 to 5 percent, cash-handling costs vary only between 0.5 and 1.6 percent.

Keeping this in mind, Figure 13 shows that the markup is almost ve times more responsive

to changes in  than to changes in .

Figure 13: Markup as a function of  and 

Appendix D Discussions of the NSR

Our analysis is conducted under the assumption that merchants obey the no-surcharge rule

(NSR). Under the NSR, merchants sign an agreement under which they cannot charge con-

sumers an additional fee for using a card. Over the years, formal NSR agreements have been

declared illegal by several antitrust authorities but not in the United States. Most merchants

in the United States still do not impose a surcharge on card payments and many do not give

discounts for cash payments. Bolt and van Renselaar (2009) provide an empirical analysis

of the effect of surcharging card payments on actual payment behavior in the Netherlands,

where surcharging is currently allowed.

54

There are a number of explanations for why merchants do not surcharge buyers for card

payments, despite having to pay a high fee for each card transaction.

Buyers' perceptions: Most buyers are not aware of the high fees imposed on merchants.

Buyers may suspect that the sole purpose of a card surcharge is to enhance merchants'

pro t with no cost justi cation. Clearly, educating consumers may solve this problem.

Proper marking: Most states require shops to mark prices on all items they sell. Imposing

a surcharge on cards may require placing two labels. By itself, this should not be a

big problem; however, when a sale is declared, merchants will have diculties with

marking down di
erent prices associated with the di
erent means of payment.

Competition: Card acceptance under high merchant fees may reect a \bad" equilibrium

for merchants, in which no merchant can pro tably deviate by refusing to accept card

payments. See Hayashi (2006) for a theoretical study.44

References

Amromin, G., C. Jankowski, and R.D. Porter. 2007. \Transforming payment choices by

doubling fees on the Illinois Tollway." Economic Perspectives 31 (2).

Armstrong, M. 2006. \Competition in two-sided markets." RAND Journal of Economics

37 (3):668{691.

Ausubel, L. 1991. \The failure of competition in the credit card market." The American

Economic Review 81 (1):50{81.

Bergman, M., G. Guibourg, and B. Segendorf. 2007. \The costs of paying|Private and

social costs of cash and card payments." Bank of Sweden Working Paper No. 212 .

Berkovich, E. 2009. \Trickle-Up Wealth Transfer: Cross-subsidization in the payment card

market." The Hispanic Institute Working Paper .

44Borzekowski and Kiser (2008) present evidence showing that merchants can substantially reduce their

cost by not accepting credit cards. In fact, Ausubel (1991) has already suggested that the use of plastic cards

by buyers cannot always be explained in a rational matter. Merchants may also manifest similar behavior.

55

Bertaut, C. and M. Haliassos. 2006. \Credit cards: Facts and theories." In The economics

of consumer credit, edited by G. Bertola, R. Disney, and C. Grant. Cambridge, MA: MIT

Press.

Bolt, N. Jonker, W. and C. van Renselaar. 2009. \Incentives at the counter: An empirical

analysis of surcharging card payments and payment behavior in the Netherlands." Journal

of Banking and Finance 34 (8):1738{1744.

Bolt, W. and S. Chakravorti. 2008. \Economics of payment cards: A status report." Eco-

nomic Perspectives 32 (4):15{27.

Borzekowski, R. and E. Kiser. 2008. \The choice at the checkout: Quantifying demand across

payment instruments." International Journal of Industrial Organization 26 (4):889{902.

Bradford, T. and F. Hayashi. 2008. \Developments in interchange fees in the U.S. and

Abroad." Federal Reserve Bank of Kansas City Payment System Research Brie ng :37{

63.

Brito, D. and P. Hartley. 1995. \Consumer rationality and credit cards." Journal of Political

Economy 103 (2):400{433.

Carlton, D. and A. Frankel. 1995. \The antitrust economics of payment card networks."

Antitrust Law Journal 63 (2):643{668.

|||. 2005. \Transaction costs, externalities, and two-sided payment markets." Columbia

Business Law Review :617{642.

Carroll, C.D. 1997. \Bu
er-stock saving and the life cycle/permanent income hypothesis."

Quarterly Journal of Economics 112 (1):1{55.

Chakravorti, S. and W. Emmons. 2003. \Who pays for credit cards?" Journal of Consumer

a
airs
37 (2):208{231.

56

Chakravorti, S. and A. Shah. 2003. \Underlying incentives in credit card networks." Antitrust

Bulletin 48 (1):53{76.

Chatterjee, S., D. Corbae, M. Nakajima, and J.V. Rios-Rull. 2007. \A quantitative theory

of unsecured consumer credit with risk of default." Econometrica 75 (6):1525{1589.

Cohen-Cole, E. Forthcoming. \Credit card redlining." Review of Economics & Statistics .

Evans, D. and R. Schmalensee. 2005. \The economics of interchange fees and their regulation:

An overview." In Interchange Fees in Credit and Debit Card Industries: What Role for

Public Authorities? Kansas City: Federal Reserve Bank of Kansas City.

Foster, K., E. Meijer, S. Schuh, and M. Zabek. 2009. \The 2008 survey of consumer payment

choice." Federal Reserve Bank of Boston, Public Policy Discussion Paper 09{10.

Frankel, A. 1998. \Monopoly and competition in the supply and exchange of money."

Antitrust Law Journal 66:313{361.

Frankel, A. and A. Shampine. 2006. \The economic e
ects of interchange fees."
Antitrust

Law Journal 73:627{673.

Gans, J. and S. King. 2003. \The neutrality of interchange fees in payment systems." Topics

in Economic Analysis and Policy 3 (1):1069{1069.

Garcia-Swartz, D., R. Hahn, and A. Layne-Farrar. 2006. \The move toward a cashless

society: Calculating the costs and bene ts." Review of Network Economics 5 (2):199{228.

Gresvik, O. and H. Haare. 2009. \Costs in the Norwegian payment system." Norges Bank

Sta
memo .

Gross, D. and N. Souleles. 2002. \Do liquidity constraints and interest rates matter for

consumer behavior? Evidence from credit card data." Quarterly Journal of Economics

117 (1):149{185.

57

Harris, E. and J. Sabelhaus. 2000. \Consumer expenditure survey family level extracts."

www.nber.org/data/ces_cbo.html, National Bureau of Economic Research.

Hayashi, F. 2006. \A puzzle of card payment pricing: Why are merchants still accepting

card payments?" Review of Network Economics 5 (1):144{174.

|||. 2008. \The economics of interchange fees and payment card fee structure: What

is the optimal balance between merchant fee and payment card rewards?" Economic

Research, Federal Reserve Bank of Kansas City Working Paper :08{06.

|||. 2009. \Do U.S. consumers really bene t from payment card rewards?" Federal

Reserve Bank of Kansas City Economic Review :37{63.

Humphrey, D., A. Kaloudis, and G. wre. 2004. \The future of cash: Falling legal use

and implications for government policy." Journal of International Financial Markets,

Institutions and Money 14 (3):221{233.

Humphrey, D., L. Pulley, J. Vesala, and R. Avery. 1996. \Cash, paper, and electronic

payments: a cross-country analysis." Journal of Money, Credit & Banking 28 (4):914{

939.

Humphrey, D., M. Willesson, G. Bergendahl, and T. Lindblom. 2006. \Bene ts from a

changing payment technology in European banking." Journal of Banking & Finance

30 (6):1631{1652.

Ireland, P. 2009. \On the welfare cost of ination and the recent behavior of money demand."

American Economic Review 99 (3):1040{1052.

Katz, M. 2001. \Reform of credit card schemes in Australia II: Commissioned report."

Reserve Bank of Australia Public Document. August .

Klee, E. 2006. \Families' use of payment instruments during a decade of change in the

U.S. payment system." Federal Reserve Board, Finance and Economics Discussion Paper

2006-01 .

58

Levitin, A. 2007. \Priceless? The costs of credit cards." UCLA Law Review 55 (2):1321{

1405.

Lucas, R. 2000. \Ination and welfare." Econometrica 68 (2):247{274.

McAndrews, J. and Z. Wang. 2008. \The economics of two-sided payment card markets:

Pricing, adoption and usage." Federal Reserve Bank of Kansas City, Research Working

Paper :08{12.

Mester, L. 2003. \Changes in the use of electronic means of payment: 1995-2001." Business

Review (Q3):18{28.

Murphy, M. and M. Ott. 1977. \Retail credit, credit cards and price discrimination." South-

ern Economic Journal 43 (3):1303{1312.

Prager, R., M. Manuszak, E. Kiser, and R. Borzekowski. 2009. \Interchange fees and pay-

ment card networks: Economics, industry developments, and policy issues." Finance and

Economics Discussion Series .

Rochet, J. 2003. \the Theory of Interchange Fees: A synthesis of recent contributions."

Review of Network Economics 2 (2):97{124.

Rochet, J. and J. Tirole. 2006. \Two-sided markets: A progress report." RAND Journal of

Economics 37 (3):645{667.

Roson, R. 2005. \Two-sided markets: A tentative survey." Review of Network Economics

4 (2):142{160.

Rysman, M. 2009. \The economics of two-sided markets." Journal of Economic Perspectives

23 (3):125{143.

Schuh, S. and J. Stavins. 2010. \Why are (some) consumers ( nally) writing fewer checks?

The role of payment characteristics." Journal of Banking and Finance 34 (8):1745{1758.

59

Schwartz, M. and D. Vincent. 2006. \The no surcharge rule and card user rebates: Vertical

control by a payment network." Review of Network Economics 5 (1):72{102.

Shy, O. and Z. Wang. Forthcoming. \Why do card networks charge proportional fees?"

American Economic Review .

Stavins, J. 2001. \E
ect of consumer characteristics on the use of payment instruments."

New England Economic Review 3 (2001):19{30.

Verdier, M. Forthcoming. \Interchange fees in payment card systems: A survey of the

literature." Journal of Economic Surveys .

Wang, Z. 2010. \Market structure and payment card pricing: What drives the interchange?"

International Journal of Industrial Organization 28 (1):86{98.

Wright, J. 2003. \Optimal card payment systems." European Economic Review 47 (4):587{

612.

Zinman, J. 2009a. \Debit or credit?" Journal of Banking and Finance 33 (2):358{366.

|||. 2009b. \Where is the missing credit card debt? Clues and implications." Review of

Income and Wealth 55 (2):249{265.

60