The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Two-sided Market, R&D and
Payments System Evolution
WP 19-03R
Bin Grace Li
International Monetary Fund
James McAndrews
TNBUSA
Zhu Wang
Federal Reserve Bank of Richmond
Two-sided Market, R&D, and Payments
System Evolution∗
Bin Grace Li,†James McAndrews‡and Zhu Wang§
June 2019
Working Paper No. 19-03R
Abstract
It takes many years for more efficient electronic payments to be widely used,
and the fees that merchants (consumers) pay for using those services are increasing (decreasing) over time. We address these puzzles by studying payments system
evolution with a dynamic model in a two-sided market setting. We calibrate the
model to U.S. payment card data and conduct welfare and policy analysis. Our
analysis shows that the market power of electronic payment networks plays important roles in explaining the slow adoption and asymmetric price changes, and the
welfare impact of regulations may vary significantly through the endogenous R&D
channel.
Keywords: Two-sided Market, Technology Adoption, R&D, Payments System
JEL Classification: E4, G2, O3
∗Forthcoming,
Journal of Monetary Economics. We thank the editor Ricardo Reis, two anonymous
referees, and Boyan Jovanovic for insightful comments that greatly improved the paper. We are grateful
to Kartik Athreya, Jesus Fernandez-Villaverde, Fumiko Hayashi, Chang-Tai Hsieh, Bob Lucas, Esteban
Rossi-Hansberg, and Nancy Stokey for helpful comments. We also thank Huberto Ennis, Borys Grochulski, Bob Hunt, Todd Keister, Ricardo Lagos, Mark Manuszak, Ned Prescott, Jean-Charles Rochet, Julian
Wright, and participants at numerous seminars and conferences for valuable comments on earlier versions
of the paper. The views expressed herein are solely those of the authors and do not necessarily reflect the
views of the International Monetary Fund, the Federal Reserve Bank of Richmond or the Federal Reserve
System.
†International Monetary Fund. Email: bli2@imf.org.
‡TNB USA Inc. Email: jmcandrews@TNBUSA.com.
§Corresponding author. Research Department, Federal Reserve Bank of Richmond, P.O. Box 27622,
Richmond, VA 23261, USA. Email: zhu.wang@rich.frb.org.
1
1
Introduction
The payments system is a vital component of the overall economy and its scale is enormous. It is estimated that U.S. consumers made 158 billion purchases worth about $8.3
trillion in 2011, not counting payments made by businesses or governments. The past several decades have also witnessed significant changes in payment technology, particularly
the migration from paper to electronic payments. As shown in Fig. 1, the share of card
relative to cash payments increases steadily over time, as does the share of the broadly
measured electronic payments relative to paper payments.
In some ways these developments are puzzling. One key question, for example, is
why it has taken so long for electronic payments to replace paper payments. For several
decades, experts on payments systems have forecast the imminent arrival of a completely
electronic, paperless payments system. While the direction of development seems quite
clear, the pace has been slow. Most electronic payments that we are familiar with today
were introduced decades ago, for example, credit cards in 1950s, ACH in 1970s, and debit
cards in 1980s. However, even until the 2000s, their combined market share was just
around 50-60 percent (see Fig. 1). The slow transition is also happening to some more
recent payment innovations, such as the mobile payments.
Also, there have been controversies regarding the competitive efficiency of electronic
payments systems, particularly for card-based payments. Merchants claim that payment
card networks (e.g., Visa and MasterCard) and their issuing banks have used market power
to drive up the fees that merchants pay for accepting credit and debit card payments (the
so-called interchange fees, shown in Fig. 2).1 This is in sharp contrast to the continuing
technological progress and cost reduction in the payment card industry, and has resulted
in major regulations and a spectacular number of antitrust litigations in recent years.2
1
The fees that merchants pay to accept payment cards are called merchant discount rates. A major
component of the merchant discount rates is interchange fees, which are set by card networks and paid
by merchants to card issuers through merchant acquirers. In the U.S. and many other countries, the
increasing interchange fees have driven up merchants’ costs of accepting cards.
2
For example, Congress passed the Durbin Amendment as a provision of the Dodd-Frank Act in July
2010, which directs the Federal Reserve Board to regulate U.S. debit card interchange fees. Subsequently,
the Federal Reserve Board implements Regulation II which caps debit card interchange fees at about
half of the pre-regulation level. Recently, Visa, MasterCard and their major issuers reached a $6.2
billion settlement agreement in September 2018 with U.S. retailers in a lawsuit over credit and debit
card interchange fees. The settlement, pending court approval, aims to end more than 50 lawsuits that
merchants have filed since 2005, and would be the largest private antitrust settlement in U.S. history.
2
Share (%)
100
80
Card/(Card+Cash)
60
40
Electronic/(Electronic+Paper)
20
0
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Fig. 1. Evolution of Payment Forms.
Data source: Nilson Report (various issues). Card includes credit and debit card payments. Electronic
includes card plus ACH and remote payments. Paper includes cash and check payments.
To address these controversies, we develop a unified framework for understanding the
payments system evolution. In recent years, a growing literature on payments economics
has tried to explore related issues. Some researchers take the money-theoretical approach
and use information economics and mechanism design to characterize alternative payments systems (e.g., Kiyotaki and Wright 1989, Lagos and Wright 2005). These theories
shed light on how payments arrangements overcome frictions of exchange (e.g., limited
information or limited enforcement), but they do not directly explain the slow adoption of
electronic payments systems. Moreover, these theories often assume that payment systems
are operated by a benevolent planner or by a club of participants, so they do not address
the competitive efficiency issues (see Kahn and Roberds 2009, Nosal and Rocheteau 2011
for reviews of the literature).
Some other researchers focus on the industrial organization and pricing issues of payments systems. Particularly, Rochet and Tirole (2002, 2011) argue that the high interchange fees reflect a fundamental failure of the Coase Theorem to hold in the two-sided
3
Cents
100
M C credit card
80
Visa credit card
60
40
PIN Debit Cards
20
0
1997
1998
1999
Star
2000
2001 2002
NYCE
2003
2004 2005
Pulse
2006
2007 2008
Interlink
Fig. 2. Payment Card Interchange Fees.
Data source: American Banker (various issues). Interchange fees are plotted for a $50 non-supermarket
transaction for Visa and MasterCard credit cards plus four largest PIN debit card networks in the U.S.
payment card market (where merchants and consumers are the two sides of card users).
While this literature provides important insights into how payment markets function, it
has some limitations. For instance, those theories focus on usage externalities of payment means but usually ignore payments adoption decisions and endogenous technological progress. Also, the literature typically imposes an ad hoc distribution of “convenience
benefits” from using a particular means of payment and assumes consumers have a fixed
demand for goods invariant to their payment choices. Under those assumptions, the adoption and usage patterns of electronic payments are not fully explained, and the welfare
analysis is restricted (see Rysman and Wright 2014 for a review of the literature).
In this paper, we provide a new analysis of payments system evolution and the accompanying competitive efficiency issues. Our theory lays out a two-sided market environment
where consumers with heterogenous income and merchants of heterogenous size make payment choices under network externalities. This setting is embedded in a fully dynamic
model in which a monopoly electronic payment network sets usage fees and conducts R&D
4
to lower costs. We calibrate the model to match U.S. payment card pricing, adoption and
usage data, and conduct welfare and policy analysis.
Our theory has several novelties. First, we model payment adoption decisions under the natural assumption that adopting a payment innovation requires merchants and
consumers each to pay a fixed cost, which allows them to potentially benefit from lower
marginal cost of making payments. Second, we introduce heterogeneous income (size)
among consumers (merchants) to explain their reliance on alternative payment technologies. Finally, we set aside the typical assumption of inelastic demand for final goods by
consumers.
Our approach yields clear implications for the adoption and usage pattern of payment
innovations. Consider the introduction of payment cards with a high fixed but low marginal cost of use, as compared with cash. More affluent consumers, with higher levels of
consumption and purchases, would be more willing to adopt cards than less affluent consumers. Similarly, larger merchants, or those who sell higher-valued goods, would be more
likely to accept cards than other merchants. Over time, as the card network conducts
R&D to continue lowering card service costs, payment cards penetrate into segments of
lower-income consumers and smaller merchants. These predictions are consistent with
the empirical evidence (see Figs. A1(a)-(c) in Appendix A1), but the literature that assumes an unspecified “convenience benefit” of using payments cards does not yield such
straightforward empirical implications.
It has previously been pointed out (e.g., Wright, 2003) that a merchant serving both
cash and card consumers would be competed out of business in a competitive environment. However, in our model setting, we find that large merchants who serve both cash
and card customers do survive the threat of entry from specialized merchants. Indeed,
our equilibrium is characterized by three categories of merchants based on their size:
Large merchants accept both cash and cards, and set a price that is lower than cash-only
merchants. They attract both cash users and card users. Medium-size merchants are
5
specialized. Some of them accept only cash. The others accept both cash and cards, but
set a price that is higher than the competing cash-only merchants. They attract only
card users, because cards are cost-saving to them overall. Finally, small merchants are
all cash-only merchants. These predictions are consistent with what we observe in reality,
but are not implied by existing theories.
Our analysis shows that the market power of electronic payment networks plays important roles in explaining the slow adoption of electronic payments. In contrast, we find
that a Ramsey social planner who aims to maximize consumer and social welfare would
set lower usage fees and conduct more R&D, thus achieving a higher adoption and usage
of electronic payments. Our analysis also provides a new explanation for high merchant
fees observed in the card payment systems. Imposing a high fee to the merchant side allows a card network to inflate retail prices and extract more rents produced by replacing
costly cash payments. According to our model, the card network tends to raise merchant
fees but reduce consumer fees as card service costs decline over time, a pattern consistent
with the data yet unexplained by prior literature. We find that cash users are disadvantaged by the high merchant fee. However, this is not because they have to subsidize card
users as suggested by previous theories (in our model, at merchants where both cash and
card are used, cash users are subsidized by card users who by their use of a more efficient
payment means, contribute to a lower overall costs of payment at the merchant). Instead,
the card network sets the merchant fee too high so that fewer stores serve both card and
cash users, hence fewer cash users get subsidized by card users.
Our paper also extends the existing literature by offering additional policy insights.
Based on the calibrated model, we compare two regulatory approaches: One requires the
card network to price at marginal cost, and the other caps the merchant fee. We find
that while marginal-cost pricing could maximize consumer welfare in a static setting, it
shuts off the card network’s R&D and may reduce social welfare in a dynamic setting. In
comparison, the merchant fee cap regulation improves consumer welfare without causing
6
much dynamic inefficiency.
The paper is organized as follows. Section 2 lays out the model setup with two competing payment means, “cash” versus “card.” Section 3 characterizes the model equilibrium
and dynamic path. Section 4 calibrates the model to U.S. payment card data. Section 5
conducts welfare and policy analysis. Section 6 concludes.
2
The Model
Our model studies pricing, adoption and usage of payment devices. We first lay out a
market environment in which only a paper device, “cash,” is in use. We then introduce
an electronic device, the “card.” In our analysis, certain market conditions, such as the
variety of goods and the consumer income distribution, are exogenously given.3
2.1
A Cash Economy
The market is composed of a continuous distribution of merchants of measure unity. Each
merchant sells a distinct consumption good characterized by two parameters: consumer
preference and unit cost . The unit cost is measured in terms of a numeraire good
called dollar. For ease of notation, we assume goods with the same value of share the
same value of unit cost denoted as . It will become clear that plays no role in our
analysis except scaling the level of utility.
Cash is the sole payment device in the economy. Merchants incur a transaction cost
per dollar for accepting cash, which includes handling, safekeeping and fraud expenses.
The market is contestable so merchants earn zero profits. As a result, the cash price for
a good is determined as , where
=
1 −
3
(1)
The static environment of the model follows McAndrews and Wang (2012) closely, upon which we
construct a fully dynamic model and conduct quantitative analysis.
7
Each period, a consumer receives her income in terms of the numeraire good (which
could be interpreted as the payoff to her endowment of effective labor). The consumer,
indexed by her income , purchases a variety of goods from merchants, each of which is
associated with a preference parameter , where ∈ (0, ). The consumer has a CobbDouglas utility , and her consumption of each good is denoted as . Let () be
R
the cumulative distribution function of and denote () = 0 (), so
ln =
Z
0
where
R
()
0 ()
ln (),
()
(2)
= 1 The consumer’s budget constraint is
Z
0
(1 + ) () 6
(3)
where is the consumer’s transaction cost of using cash. For simplicity, we assume
there is no intertemporal saving technology, so income is fully consumed each period.
Maximizing the utility (2) subject to the constraint (3) yields consumer ’s demand for
each good :
∗ =
()(1 + )
(4)
As a result, consumer enjoys the utility level such that
ln = ln +
Z
0
ln
()
() ()(1 + )
(5)
This implies that a consumer’s utility is proportional to her income,
=
where = exp
of income.
nR
0 ()
(6)
o
ln ()(1+
()
is the utility associated with one dollar
)
8
Income varies across consumers and is distributed according to a cumulative distriR
bution function () on the support (0, ). We denote () = 0 () as the mean
income and normalize the measure of consumers to be unity. At equilibrium, a merchant’s
sale of good equals its total demand:
=
Z
() =
0
()
()(1 + )
(7)
It follows from Eq (7) that the size of a merchant, measured by the value of sales ,
increases with the preference index of the good that the merchant sells.
2.2
Introducing the Payment Card
We now introduce an electronic payment innovation, referred to as a payment card. The
card service is provided by a monopoly network. The costs of providing the card service
to merchants and consumers are and per-dollar transaction respectively, and we
denote the sum = + . It will become clear that in our two-sided market setting,
only the sum (but not its composition, and ) matters for the analysis. In return,
the card service provider charges merchants and consumers a percentage fee and ,
respectively.4 Figure 3 describes the transaction flow in the card system in which consumers use a payment card to pay merchants. Merchants submit charges to the card
network which then bills consumers.5
4
Assuming payment cards charge percentage fees is consistent with reality. In most countries, credit
cards charge fees proportional to the transaction value. In the U.S., most debit cards also charge percentage fees. Shy and Wang (2011) and Wang and Wright (2017) provide theoretical explanations as to
why many platform industries favor percentage fees.
5
For simplicity, we model a “three-party” system where the payment card network serves consumers
and merchants directly, but our analysis can equally apply to a “four-party” system where the card
network serves consumers and merchants indirectly through card issuers and merchant acquirers. In
the literature, it is typically considered that merchant acquirers are competitive and the card network
maximizes the joint profits of member issuers, so we can simply reinterpret the card network in our model
to be the association of member issuers in a “four-party” system.
9
C a rd N e tw o rk
p a y s p (1 -f m )
p a y s p(1 + f c )
C a rd h o ld e r
M e rc h a n t
sells g o o d a t
p rice p
Fig. 3. Payment Card System.
We adopt the convention that the transaction costs of using payment cards are normalized to 0 for both merchants and consumers. Therefore, the avoided costs of handling
cash, and , are the convenience benefits of card payments to merchants and consumers, respectively. For the card being a more efficient payment means, we impose the
condition:
+
(8)
In order to adopt the payment card, a merchant and a consumer each incur a fixed
cost and every period, respectively. The fixed costs may arise for technological
or institutional reasons. For example, a merchant may incur a fixed cost renting cardprocessing equipment and/or training employees, while a consumer may incur a fixed cost
maintaining her bank balance or credit score. Therefore, in deciding whether to accept or
hold the payment card, merchants and consumers need to weigh the benefit of avoiding
handling cash against their card adoption costs and usage fees. For merchants who decide
to accept payment cards, they still have to accept cash and charge a single price to card
10
and cash customers.6
To study the industry evolution, we proceed in two steps with the following analysis.
(i) Within-Period Analysis. At a point in time, with the card service cost given,
we solve for a subgame perfect Nash equilibrium of the following three-stage game:
Stage I. The payment card network sets the card fees and
Stage II. After observing and merchants make card acceptance decisions and
post retail prices for the period. Meanwhile, consumers decide whether to hold cards.
Stage III. Consumers then decide whether to purchase, which merchants to purchase
from, and what payment device to use.
In making the decisions, consumers and merchants each maximize their utility or
profits, and the card network sets card fees to maximize profit.
(ii) Dynamic Analysis. The industry evolves over time. To study the industry evolution, we assume time is discrete and the payment card is introduced at time 0. We add
back the time subscripts to variables. On the one hand, the mean consumer income ( )
may grow exogenously, together with the changes of card adoption costs and .
On the other hand, the card network can make endogenous R&D investment (in units
of the numeraire good) to reduce the card service cost next period such that
+1 = Γ( )
where Γ 0 and Γ 0. Given the initial value of 0 and the laws of motion
for ( ), and , the card network chooses a sequence of card fees ( ) and
R&D investment to maximize the present value of profits.
6
This is referred to as “price coherence” in the literature, and is commonly seen in reality. Price
coherence may result from network rules or state regulation (e.g., In the U.S., card network rules and
some state laws explicitly prohibit surcharging on payment cards). It can also result from high transaction
costs for merchants to price discriminate based on payment means (e.g., even if surcharging payment cards
were allowed, few merchants would choose to do so).
11
3
Market Equilibrium
In this section, we study the market equilibrium under the profit-maximizing network.
We first study the industry equilibrium within a period, for which the industry has given
values of , (), and . We then embed that in a fully dynamic setting to study
the industry evolution.
3.1
Within-Period Analysis
At a point in time, the industry has given values of , (), and . The market
equilibrium can be solved backward for the three-stage game. In Stages II and III, given
card fees and set by the card network, merchants and consumers each decide their
own card adoption and usage, taking others’ decisions as given. In Stage I, anticipating
consumers’ and merchants’ responses, the card network decides the profit-maximizing
card fees.
Note that the interactions between the two sides of the market can easily yield multiple equilibria. For example, an equilibrium may exist at which neither merchants nor
consumers adopt cards. In the following analysis, we will focus on an equilibrium with
positive card adoption and usage. We show that there exist threshold values of merchant
size and consumer income, above which merchants and consumers accept and hold cards.
We will first construct such an equilibrium and then characterize its properties.
3.1.1
Merchants’ Choices
Merchants take card fees as given and expect that consumers ≥ 0 will hold cards
when making their card acceptance decision. Recall that the index indicates merchant
size. It can be shown that there exist two threshold values 1 and 0 , where 1 0 ,
so that merchants fall into three categories: (1) Large merchants ( ≥ 1 ) accept both
cash and cards, and charge price ≤ so they are patronized by both cash and
12
card customers. (2) Intermediate merchants (0 ≤ 1 ) specialize. Some accept both
cash and cards, and charge where
1+
1+
≥ , so they are patronized only
by card customers. The others do not accept cards and charge , so they only serve
cash customers. (3) Small merchants ( 0 ) only accept cash, so all customers make
purchases there using cash regardless of whether they have adopted a card.
As we will show next, the thresholds 1 and 0 are endogenously determined under
card fees and . Note that because merchants who accept payment cards still have
to accept cash and charge a single price to card and cash users, the consumer card fee
has to be below the consumer cost of handling cash, i.e., ≤ . This condition can be
thought of as a card pricing constraint to provide consumers “incentive at the counter”
to use payment cards. As we show in the following analysis, card fees also need to satisfy
additional pricing constraints that make card-accepting stores attractive to consumers, in
other words, providing consumers “incentive at the door.”
Large Merchants: ≥ 1 Merchants in this category charge ≤ and receive
revenues from both card customers ( ≥ 0 ) and cash customers ( 0 ):
=
where ≥0 () ≡
R
0
[≥0 ( − )]
[0 ()]
,
,
=
()(1 + )
()(1 + )
(9)
().
Because merchants need to post their prices and cannot make changes within the
same period, they each set a lowest possible price to prevent being undercut by potential
entrants. This contestable market assumption leads to zero profit so that a merchant’s
total revenue equals total cost,
(1 − )
+ (1 − ) = + +
13
(10)
Equations (9) and (10) pin down the price :
=
[≥0 (− )]
[0 ()]
+ (1+
(1+ )
)
[≥0 (− )]
[0 ()]
)
+ (1 − ) 1+
1+
(1 −
(11)
− ()
Recall from Eq. (1) that = (1 − ). Hence, ≤ implies that there
exists a threshold
1 =
()
[≥0 ( − )]( 1−
−
1+
1−
)
1+
if ≤ ,
(12)
so that merchants whose size ≥ 1 fall into this category. Note Eq. (12) suggests that
no merchant would belong to this category if because the prices they offer would
not attract cash users.
Medium-Size Merchants: 0 ≤ 1 Merchants in this category specialize. For
each good, there are two merchants. One accepts both cash and card, and charges
where
1+
1+
≥ , so it is patronized only by card customers (i.e., ≥ 0 ).7
The other does not accept cards and charges , so it only serves cash customers (i.e.,
0 ).
A card-accepting merchant in this category receives revenues only from card customers
and earns zero profit, which implies
=
Therefore,
1+
1+
[≥0 (− )]
(1+ )
[≥0 (− )]
)
−
()
1+
(1 −
(13)
≥ implies that merchants 0 ≤ 1 are in this
group, where 1 is given in Eq. (12) and 0 is determined by
0 =
()
[≥0 ( − )]( 1−
−
1+
7
1−
)
1+
.
(14)
Because , the merchant would not attract cash customers. But given (1 + ) ≤
(1 + ) , the merchant would attract card users because cards are cost-saving to them overall.
14
Equations (12) and (14) suggest that
1−
1+
≥
1−
1+
has to hold. In fact, if
1−
1+
1−
,
1+
there would be no merchant in either category (1) or (2).8
Small Merchants: 0 Given
1−
1+
≥
1−
,
1+
small merchants 0 are in the
third category. Due to small transaction values, these merchants cannot afford paying
the fixed adoption cost to accept cards. Otherwise, it would result in
1+
.
1+
Therefore, they only accept cash and all consumers, regardless of holding a card or not,
make purchases there using cash.
3.1.2
Consumers’ Choices
An individual consumer takes market prices, card fees and merchants’ card acceptance as
given and decides whether to hold a payment card or not. Recall that merchants in the
market fall into three categories according to their sizes. A consumer , if not adopting
from shopping at card-acceptance stores in
the card, may use cash and derive utility
category (1) and at cash-only stores in categories (2) and (3). Accordingly,
ln
=
Z
min(1 )
0
Z
ln
()+
ln
()
() (1 + ) ()
(1 + ) ()
min(1 ) ()
In contrast, if she holds a payment card, she can use the card to shop at card-accepting
stores in categories (1) and (2), though she still needs to use cash at stores in category
so that
(3). As a result, she derives utility
ln
=
Z
0
8
0
( − )
ln
() +
() (1 + ) ()
Z
0
( − )
ln
()
() (1 + ) ()
Because ≥ has to hold for any consumers to use cards,
15
1−
1+
1−
1+
implies
Therefore, consumer card adoption requires ln
≥ ln
, which implies
1 +
) ≥ () ln(
)+
≥0 () ln(
1 +
−
where ≥0 () ≡
R
0
Z
min(1 )
ln(
0
)()
(15)
().
Equation (15) suggests that an adopter’s income has to satisfy ≥ 0 :
≥0 ()()
)
( 1+
1+
0 =
R
min(
)
1
≥0 ()()
( 1+
)
−
exp(
ln(
)())
1+
()
0
(16)
where
) ≤min(1 ) =
(
0
1−
[≥0 (
1+
1−
[≥0 (
1+
− )]
− )] −
()
(17)
follows Eqs. (1) and (13).
3.1.3
Two-sided Market Interactions
As shown above, the interactions between consumer and merchant card adoption are
described by the threshold equations (12), (14), (16), and (17).
To simplify the notations, we denote:
1 = (
1 − 1 −
−
)
1 +
1 +
0 = (
1 − 1 −
−
)
1 +
1 +
(18)
And we can summarize our findings in the following proposition.
Proposition 1 Given card fees ( and ) that satisfy
≥
and
1 −
1 −
≥
1 +
1 +
(19)
there exist threshold values of merchant size (0 and 1 ) and consumer income(0 ), above
16
which merchants and consumers accept and hold cards, where
0 =
1 =
()
[≥0 ( − )]0
0
0
1
if
≤
≥0 ()()
)
( 1+
1+
0 =
R
min(
)
(1− )
1
≥0 ()()
( 1+
)
−
exp(
ln
())
1+
()
(1− )−(1+ )0 0
0
3.1.4
(20)
(21)
(22)
Within-Period Equilibrium
The card network, anticipating merchants’ and consumers’ card adoption and usage decisions in Stages II and III, would set card fees ( ) in Stage I to maximize its profit.
Accordingly, the card network solves the following problem:
(; () ) = max
≥0 ()≥0 ( − )
( + − )
()(1 + )
(23)
(19) (20) (21) (22)
As shown in (23), the network’s profit equals consumers’ total spending on cards times
its markup. In order to maximize profit, the network internalizes the two-sided market
∗
externalities by setting the optimal card fees (∗
) which determine the card adoption
∗
), the market
thresholds for merchants and consumers. Given the optimal card fees (∗
achieves a within-period equilibrium, at which the card network maximizes profit, consumers maximize utility, merchants break even, and goods and payments markets clear.
3.2
Dynamic Analysis
For the dynamic analysis, we add back time subscripts to variables. Over time, the market
may evolve due to exogenous changes (e.g., in mean consumer income ( ) and card
17
adoption costs , ) and the network’s endogenous R&D investment . Embedding
the within-period equilibrium into the dynamic setting, the value function of the card
network is defined as follows:
( ; ( ) ) = max ( ; ( ) ) −
(24)
+ (+1 ; (+1 ) +1 +1 )
+1 = Γ( )
(25)
( ; ( ) ) ≥
(26)
where ( ; ( ) ) is the card network’s within-period profit, is the discount
rate, and Γ is the R&D function
To solve the dynamic optimization problem, we can rewrite the constraint (25) as
= Ψ( +1 )
(27)
where Ψ is the inverse function of Γ As a result, if the budget constraint (26) never binds
(which is verified in our calibrated model), the problem (24) is equivalent to
( ; ( ) ) = max ( ; ( ) ) − Ψ( +1 )
+1
(28)
+ (+1 ; (+1 ) +1 +1 )
As long as the functions and Ψ and the laws of motion for ( ) and satisfy
standard regularity conditions, the optimal solution to (28) can be pinned down by the
following first-order and envelope conditions:
Ψ2 ( +1 ) = 0 (+1 ; (+1 ) +1 +1 )
18
(29)
0 ( ; ( ) ) = 0 ( ; ( ) ) − Ψ1 ( +1 )
(30)
Combining (29) and (30) yields a second-order difference equation
Ψ2 ( +1 ) = [0 (+1 ; (+1 ) +1 +1 ) − Ψ1 (+1 +2 )]
which, together with boundary conditions of (i.e., 0 is given and ∞ = 0) and the laws
of motion for ( ) and , pins down the dynamic equilibrium path.
4
Model Calibration
In this section, we calibrate the model to U.S. payment card data. We first lay out
functional form assumptions and characterize the model properties, and then choose parameter values to match the model with U.S. credit and debit card pricing, adoption and
usage data.
4.1
Functional Forms
To take our model to data, we make the following functional form assumptions F1-F4.
F1. Merchant size and consumer income follow explicit distributions: ∈ (0 1) is
uniformly distributed where () = and () = 12; and ∈ (0 ∞) is exponentially
distributed where ( ) = 1 − (− ) and ( ) = 1 .
F2. The mean consumer income has a constant growth rate such that +1 =
(1 + ).
F3. The card adoption costs are proportional to the mean consumer income: =
( ) = and = ( ) = , where and are fixed parameters.
F4. The R&D function Γ takes the following form:
1
+1
−
1
−−1
=(
with 1 0
)
19
(31)
where R&D investment (scaled by income), ( ) = , enters the function and
is a fixed parameter measuring the efficiency of R&D.
The functional form assumptions are made mainly for tractability. Note that one
assumption in F1, that merchant size follows a uniform distribution, is inessential, and
our model findings continue to hold even if follows a positively skewed distribution
(such as an exponential distribution). Another assumption, that consumer income
follows an exponential distribution, matches U.S. data well, as shown by Dragulescu and
Yakovenko (2001). Moreover, the exponential distribution has a constant Gini coefficient
at 0.5, which also agrees with U.S. data (according to Dragulescu and Yakovenko 2001,
the Gini coefficient for U.S. personal income was between 0.47 and 0.56 from 1979-1997).
Therefore, this assumption allows our analysis to focus on the change in the mean income
while holding the income inequality fixed.
F2-F3 are simplifying assumptions that reduce free parameters of the model so that
we can draw sharp results for the dynamic analysis. Particularly, by tying card adoption
costs and to the mean consumer income 1 , the exogenous growth of income
would affect card transaction values but not the adoption thresholds, with the latter being
fully driven by the endogenous decline in card service costs . F4 models the decline in
card service costs as a result of R&D, and because R&D may entail opportunity costs
of labor hours, we scale it by the mean consumer income (i.e., ( ) = ). As will
become clear, the R&D functional form (31) allows us to derive a balanced growth path
in the dynamic analysis.
4.2
Model Characterization
Before we take the model to data, we characterize some important properties of the model
equilibrium given the functional form assumptions we made.
20
4.2.1
Within-Period Equilibrium
We start with the within-period equilibrium. First, given adoption costs, two-sided market
interactions may yield multiple equilibria. Assume 1 ̄ = 1. Under our functional
form assumptions, Eqs. (20) and (22) can be rewritten into
0 = (−0 )
2
(1 + 0 − )0
2
(L1),
0 =
1−0
)
( 1+
1+
2
1−0
( 1+
)
− exp(20 )
1+
(L2),
where is a scalar determined by the model parameters (see the proof in Appendix A2).
For plausible parameter values, we plot Eqs. (L1) and (L2) in Fig. 4, which illustrates
the interactions between merchant and consumer card adoption. For a given pair of card
fees ( , ), there exist two equilibria with positive levels of card adoption (Note that
no adoption is an equilibrium as well): a high-adoption equilibrium (0∗ ,∗0 ) and a low0
0
adoption equilibrium (0 ,0 ). The high equilibrium is stable while the low equilibrium is
not, so we always select the stable high-equilibrium in our following analysis.9
Second, under Assumption F3, the optimal card adoption thresholds (0∗ , ∗0 ) depend
on but not . As a result, changing the mean consumer income 1 just shifts the
profit function (; ) proportionally. We can also verify that for a given , the network’s
optimal profit (; ) is approximately linear in . Therefore, the profit function can be
simplified as
(; ) =
1
(0 − 1 )
(32)
where 0 and 1 are scalars determined by the model parameters (see Appendix A3 for
the detailed derivation).
9
The idea of using stability to select an equilibrium goes back to Cournot (1838), which provided a
dynamic adjustment argument by imagining sequential play by each agent myopically best-responding
to all rivals. This is referred to as best-reply dynamics, and when the process converges, the solution
is termed stable (Fudenberg and Tirole, 1991). In the context of our model, the existence of multiple equilibria implies that when the payment card is initially introduced, the card network may need
to push the card adoption to overcome the unstable (low-adoption) equilibrium to achieve the stable
(high-adoption) equilibrium. This could be interpreted as the adoption hurdles due to merchant and/or
consumer learning.
21
0
1
L1
u
0
L2
0*
0
I0*
I0
H igh Eqm
Low E qm
I0
Fig. 4. Interaction of Merchants and Consumers in Card Adoption.
4.2.2
Dynamic Equilibrium
The within-period equilibrium can then be embedded in the dynamic analysis. Note that
Assumption F4 implies a R&D investment function such that
∙
¸ 1
−1
with 1 0
=
+1
The function, in the spirit of adjustment costs studied in the macro/investment models
(e.g., Hall, 2001, 2004), is strictly increasing and convex in technological progress ( +1 )
and constant returns to scale in ( +1 ).10
We can then formulate the dynamic problem of the card network as follows:
∙
¸ 1
− 1 + (+1 ; +1 )
( ; ) = max ( ; ) −
+1
+1
10
(33)
See Khan and Thomas (2008) for a comprehensive review of the adjustment cost literature. Our
modeling of R&D allows us to solve for a balanced growth path at the industry level. This is also related
to the endogenous growth theories (e.g., see Aghion and Howitt 1998), though the focus there is to study
how R&D could drive balanced growth at the aggregate economy level.
22
Given the profit function ( ; ) represented by Eq. (32), the dynamic optimization
problem (33) can be explicitly solved for a balanced-growth path (see Appendix A3 for
the detailed proof).
4.3
Calibration Results
4.3.1
Industry Background
Given the functional forms of the model, we choose parameter values to match U.S.
payment card data from 1997-2008.11 During this period, credit and debit cards became
the most popular electronic payment means for retail transactions in the U.S. economy.
The early history of general purpose credit cards in the U.S. can be traced back to 1950s
when the Diners Club card was first launched, but it was not until the 1970s that credit
cards eventually picked up, with the market eventually becoming highly concentrated
(Visa and MasterCard became the two largest credit card networks, followed by Amex
and Discover). The introduction of debit cards was more recent. ATM and ATM cards
were invented in 1970s. By adding the debit function, some of the ATM cards became
debit cards in early 1980s, which were used to pay at the point of sale. The adoption and
usage of debit cards started to gain steam in the mid-1990s, and Visa, MasterCard and
several ATM networks became dominant players.12
Since the late 1990s, with the wide adoption of credit cards and rapid expansion of
debit cards, the fees associated with these cards have raised great controversies. The key
issue is the asymmetric fees that card networks charge to merchants and consumers. In
a card transaction, merchants are typically charged a much higher fee than consumers,
with the major component being the interchange fee. Despite rapid technological progress
in the card industry, the fees charged to merchants continue to increase while the fees
11
Our sample starts from 1997 when the card fee data became available. The sample ends in 2008
because the adoption of credit and debit cards had almost saturated by then and we also want to avoid
the disruption of the Great Recession.
12
For the historical development of credit and debit cards in the U.S. market, see Evans and Schmalensee
(2005), Hayashi, Sullivan and Weiner (2006), and Hayashi, Li and Wang (2017).
23
charged to consumers decline (and in some cases consumers pay a negative fee by receiving
rewards). The controversy of payment card fees led to numerous lawsuits and Congress
hearings, and eventually Congress mandated the Federal Reserve to regulate debit card
interchange fees in 2011.
While our model is about electronic payments in general, credit and debit cards are
particularly good applications. Note that while our model does not consider credit functionality, it is still very relevant for credit cards. By the late 1990s, about three-fourths
of American families had adopted credit cards, but of those families nearly half only used
the payment function of the cards (i.e., by paying off all purchases made with the card at
the end of each month, not utilizing any revolving credit), so-called “convenience users.”
However, a consumer’s spending on a credit card is limited by the card’s credit line, so
credit cards may not cover the consumer’s entire payment needs. This made way for
the fast adoption of debit cards since the mid-1990s, especially among lower-income consumers who are constrained by their credit limits (Zinman, 2009). Because debit cards
do not provide the credit function, they serve purely as a payment device. In 2008, credit
card were used in 26.5 billion transactions worth about $2.1 trillion in the U.S. economy,
while debit cards were used in 34 billion transactions worth about $1.3 trillion.
4.3.2
Model Fit
To calibrate the model, we first collect data on the pricing, adoption and usage for credit
and debit cards. The pricing data we collect include the interchange fees charged by Visa,
MasterCard and major debit card networks between 1997-2008 and we convert them into
merchant card fees by adding a reasonable merchant acquiring fee. We also collect data
on consumers’ adoption of credit and debit cards in the same period, as well as annual
transaction values for cash, credit card and debit card.13 We then calibrate our model to
13
Data sources: Interchange fee data are from the American Banker (various issues). We then calculate
the average credit card interchange fee based on credit card data for Visa and MasterCard. We also calculate the value weighted average debit card interchange fee based on debit card data for Visa, MasterCard,
and top PIN debit networks. We then add a merchant acquiring fee on top of the interchange fee so that
24
match these patterns. Our calibrated model also delivers additional patterns, including
the decline in consumer card fees and card service costs, which are consistent with industry
observations.
Table 1. Parameter Values for Model Calibration
Parameter Definition
Value
Parameter Definition
Value
Merchant cost of handling cash
4.0%
Initial value of card service costs
0
2.25%
Consumer cost of handling cash
2.5%
R&D function curvature
0.5
Merchant cost of adopting card
2.5%
R&D efficiency parameter
10
Consumer cost of adopting card
0.3%
Initial value of mean income
10
21,215
Merchant cost of goods
1
Growth rate of mean income
2%
Parameter values used for the calibration are reported in Table 1. We set merchant
cost of handling cash = 4%, which gives the upper limit of the merchant card fee
implied by the model. We set consumer cost of handling cash = 25%. We calibrate
merchants’ card adoption cost = 25% of the consumer mean income, which implies an
adoption cost about $500 in 1997. Consumer card adoption cost is set as = 03%, which
implies an adoption cost about $60 in 1997. These numbers allow the model to match
the card adoption and usage pattern in the data.14 We set the mean consumer income
to be $21,215 in 1997 with an average growth rate 2%. This is derived from U.S. Census
data on deflated annual Personal Consumption Expenditures.15 Because merchant cost of
the total merchant discount rate for accepting credit cards is around 3% in the late 2000s, which is about
the industry standard (e.g., Square Inc. charges merchants 2.75% for each swiped credit card transaction
and 3.5% per manually-entered credit card transaction). The data on consumers’ adoption of credit and
debit cards are from the Survey of Consumer Finance (various years). The data on annual transaction
values for cash, credit card and debit card are from the Nilson Report (various issues).
14
Note that if we set symmetric costs for merchants and consumers = and = , the model
would deliver the same qualitative pattern of card pricing that the merchant card fee continues to increase
over time while consumer card fee continues to decrease.
15
In our model, a consumer’s income equals her total purchases. Therefore, we use U.S. Census data
on deflated annual Personal Consumption Expenditure (PCE) from 1997-2008 to calibrate the consumer
income/purchase in our model. Using the estimates from the Nilson Report, we count purchases of goods
and services as 78% of the PCE (the rest are nonpurchases — transactions for which no payment was
made, such as food and lodging received by employees).
25
goods plays no role in our analysis except for scaling utility, we normalize that = 1.
We set initial value of card service cost 0 = 225%, below the costs of handling cash. We
set R&D function parameters = 05 and = 10 to match the dynamic pattern of data.
The model calibration yields patterns that are consistent with data, as shown in Fig.
5A. The model generates a rising merchant fee over time and the level falls into the range
between the average merchant fees charged by credit and debit cards in the data. The
model generates a declining consumer fee over time, which is consistent with an increase of
consumer card rewards during the period.16 The model also generates increasing adoption
of cards by consumers similar to the data, and the share of card transaction rises in parallel
with the data (for the sum of credit and debit cards) though at a slightly higher level.
Beyond comparison with the data, Fig. 5B reports additional patterns of interest
generated by the model. First, the overall card adoption by merchants increases over
time, with the shares of large merchants (i.e., those who accept both cash and card but
charge prices lower than cash-only stores) and small merchants (who accept cash only)
declining. Second, the card network invests in R&D to reduce card service costs , with
the R&D expenditure to mean income ratio declining over time. Third, with the decline
in card costs, the card network charges an increasing markup, which together with a rising
card spending share indicates that the card network earns an increasing profit. Finally,
consumer welfare in each period, scaled by the mean income, continues to rise, while in
a cash economy it would be constant (which we normalize to 100 in the figure). The
detailed derivation of consumer welfare is explained next.
16
It is hard to estimate the average consumer card fees. Credit card users, if not rolling over any
balances on their cards, may not need to pay a fee (or even receive rewards) for each card transaction.
However, there is a chance that some of those users may end up borrowing from their cards, in which case
they then need to pay a very high interest rate for every transaction made. For debit cards, consumers
often need to pay a PIN fee for using online debit cards in our sample period, but some of them may
also receive rewards. On the other hand, industry studies show that card rewards have gained increasing
popularity over time. In 2001, less than a quarter of credit card offers included the promise of a rewards
program. But by 2005, the share was 58 percent, according to Mail Monitor, a unit of consumer research
company Synovate. A similar trend also happened to debit cards, as shown in 2005/2006 Study of
Consumer Payment Preferences conducted by the American Bankers Association and Dove Consulting.
26
A.
4
Merchant Card Fee: f m (%)
2
3
1.5
2
1
Model: Monopoly
Data: Credit
Data: Debit
1
0.5
Model: Monopoly
0
0
1998
100
Consumer Card Fee: f c (%)
2000
2002
2004
2006
2008
1998
Card Adoption by Consumers (%)
100
80
80
60
60
40
2000
2002
2004
2006
2008
Card Transaction Share (%)
40
Model: Monopoly
Data: Credit
Data: Debit
20
20
0
Model: Monopoly
Data: (Card)/(Card+Cash)
0
1998
2000
2002
2004
2006
2008
1998
2000
2002
2004
2006
2008
B.
Share of Small Merchants:
55
0
(%)
Share of Large Merchants: 11.5
1
(%)
1
50
0.5
45
0
1998 2000 2002 2004 2006 2008
1
R&D to Income Ratio:
1998 2000 2002 2004 2006 2008
R (%)
3
Card Cost: d (%)
2
0.5
1
0
0
1998 2000 2002 2004 2006 2008
1998 2000 2002 2004 2006 2008
Card Markup: f +f -d (%)
4
m
c
102
3
101
Consumer Welfare
Card Economy
Cash Economy
2
100
1
1998 2000 2002 2004 2006 2008
1998 2000 2002 2004 2006 2008
Fig. 5. Model Calibration.
27
5
Welfare and Policy Analysis
This section provides a normative analysis of the card payment system. First, we show
introducing payment cards increases welfare for both card adopters and nonadopters.
Second, we show a Ramsey social planner who maximizes consumer and social welfare
using the same set of pricing and R&D instruments would behave differently than a profitmaximizing platform. Finally, we discuss welfare implications of policy interventions.
5.1
Payment Card and Welfare Improvement
In our model framework, introducing the payment card improves every consumer’s welfare,
even for those who do not adopt the card. This can be shown in the following welfare
comparison between a cash economy and a card economy.
Recall in a cash economy, an individual consumer enjoys the utility level such
that
ln =
Z
0
()
ln
() ()(1 + )
Consider the same economy but with the introduction of the payment card. A consumer
decides whether to adopt the card based on her income. For a consumer who adopts the
card (i.e., ≥ 0 ), her utility
implies that
ln
=
Z
0
0
( − )
ln
() +
() ()(1 + )
Z
0
( − )
ln
()
() ()(1 + )
such that
while a nonadopter (i.e. 0 ) derives the utility level
ln
=
Z
0
min(1 )
Z
ln
()+
ln
()
() (1 + ) ()
(1 + ) ()
min(1 ) ()
Therefore, a consumer receives different welfare gain depending on her income. For a card
28
adopter, her welfare gain is
ln
− ln
Z
Z
ln(
ln(
=
)() +
)()
()
0
min(1 ) ()
1 +
−
>0 ()
ln(
)
(34)
) + ln(
+
()
1 +
min(1 )
while for a nonadopter, her welfare gain is
ln
− ln =
Z
min(1 )
)()
ln(
()
(35)
Equations (34) and (35) are intuitive: A card consumer enjoys utility gains from
shopping at card-accepting stores in both categories (1) and (2), subject to the card
adoption and usage costs, while a cash consumer only benefits from lower retail prices
charged by merchants in category (1). In fact, we can derive the following proposition.
Proposition 2 No consumer is worse off given any positive measure of card adoption.
Proof. Equation (35) shows that cash consumers’ welfare cannot be lower in a card economy than in a cash economy, given that ≤ in merchant category (1). Equation
(34) shows this also holds true for the marginal card consumers = 0 , while all other
card consumers 0 are strictly better off in the card economy.
The findings of Proposition 2 stand in contrast to the previous studies (e.g., Rochet
and Tirole 2002, 2011), which argue that cash consumers subsidize card consumers. In
our setting, because cards are a more efficient payment means, cash consumers are indeed
subsidized by card consumers. However, this does not mean that cash consumers’ welfare
cannot be further improved in a card economy, as we will study next.
5.2
Ramsey Social Planner
Given the utility measures described above, the Ramsey social planner would choose a
sequence of card fees ( ) and R&D investment to maximize the present value
29
of consumer welfare subject to the adoption incentive constraints of merchants and consumers as well as the network balanced-budget constraint each period. Because the Ramsey social planner makes zero profit for running the card network, it also maximizes the
social welfare under those constraints. The Ramsey social planner’s problem is analyzed
as follows.
Within-Period Decision Within each period, for given values of card costs and
R&D investment , the Ramsey social planner who runs the card network would choose
( ) to maximize consumer welfare with the revenue covering operation and R&D
expenditures. This means that
( ; () ) =
Z
∞
()
0
+
Z
0
()
(36)
0
(11) (17) (19) (20) (21) (22) (34) (35) and
(; () ) =
≥0 ()≥0 ( − )
( + − ) ≥
()(1 + )
(37)
To minimize distortion, the Ramsey social planner would always keep the constraint (37)
binding.
Dynamic Decision To describe the dynamic problem, we add back the time subscript
to each variable. The value function of the Ramsey social planner is as follows.
( ; ( ) ) = max ( ; ( ) )
(38)
+ (+1 ; (+1 ) +1 +1 )
+1 = Γ( )
30
(39)
where Γ is the R&D function.
Based on the same set of functional form assumptions and parameter values as above,
we can solve the Ramsey social planner’s problem and compare the result with that of
the profit-maximizing network (see Appendix A4 for the detailed proof).
For illustrative purposes, we consider a hypothetical scenario where the Ramsey social
planner takes over and runs the card network at the beginning of our sample period.
Figure 6 compares the Ramsey outcome with the monopoly network case and shows
several important findings. First, both merchants and consumers enjoy lower fees under
the Ramsey social planner, and unlike the monopoly case, both the merchant fee and the
consumer fee decrease over time. These lead to higher card adoption and usage, which
suggests that the market power of card networks does contribute to a slow diffusion
of electronic payments. Second, under the Ramsey social planner, the share of large
merchants whose sizes are large enough to attract both card and cash consumers is much
higher and increasing over time. This suggests that cash users would benefit much more
from electronic payments than under the monopoly network case. Third, the Ramsey
social planner would conduct more R&D, which results in a faster decline in card service
cost. Finally, all things together, consumer welfare in each period increases faster under
the Ramsey social planner, which is due to the fact that the Ramsey social planner charges
a decreasing markup over time, while the monopoly charges an increasing one.
A few reasons explain why the monopoly network and the Ramsey social planner
would behave differently. For the monopoly network, imposing a high merchant fee leads
to high retail prices of goods and allows the network to extract more rents produced by
replacing costly cash payments. Moreover, because the network does not earn profits
from cash users, charging a high merchant fee reduces cross subsidies from card users to
cash users through large merchants who serve both card and cash users. In contrast, the
Ramsey social planner cares about consumers’ real purchases rather than their nominal
card spending, and cares about the welfare of both card and cash users. In terms of R&D
31
decisions, the monopoly only sees the benefit of increased profit, which is a subset of
the social welfare that the social planner would value, so the monopoly makes less R&D
investment than the Ramsey social planner.
A.
Merchant Card Fee: f
4
m
(%)
Consumer Card Fee: f (%)
c
4
3
2
2
0
1
0
-2
1998 2000 2002 2004 2006 2008
1998 2000 2002 2004 2006 2008
Card Adoption by Consumers (%)
100
100
80
80
60
60
40
40
20
20
0
Card Transaction Share (%)
Monopoly
Social Planner
0
1998 2000 2002 2004 2006 2008
1998 2000 2002 2004 2006 2008
B.
Share of Small Merchants:
60
0
(%)
Share of Large Merchants: 160
1
(%)
40
40
20
20
0
1998 2000 2002 2004 2006 2008
1.5
1998 2000 2002 2004 2006 2008
R&D to Income Ratio: R (%)
3
1
2
0.5
1
0
0
1998 2000 2002 2004 2006 2008
4
Card Cost: d (%)
1998 2000 2002 2004 2006 2008
Card Markup: fm+f c-d (%)
106
104
2
Consumer Welfare
Monopoly
Social Planner
102
100
0
1998 2000 2002 2004 2006 2008
1998 2000 2002 2004 2006 2008
Fig. 6. Monopoly Network vs. Ramsey Social Planner
32
Note that in the Ramsey social planner case, because the network earns zero profit,
consumer welfare equals social welfare. However, in the monopoly network case, evaluating social welfare needs to take into account both consumer welfare and network profits by
assigning appropriate weights to each. Here, we consider a natural approach by assuming
that the monopoly network rebates its profits to consumers proportionally to consumers’
income (e.g., consumers own shares of the network in proportion to their income). In
so doing, we can assign consumption values to the network profits to calculate the total
social welfare. Of course, there are other ways of assigning welfare weights to profits, and
we keep the distinction between “consumer welfare” (without profit rebates) and “social
welfare” (with profit rebates) in our following analysis.
104
Monopoly: Consumer Welfare
Monopoly: Social Welfare
Social Planner: Consumer/Social Welfare
103.5
103
102.5
102
101.5
101
100.5
100
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Fig. 7. Social Welfare Comparison.
Based on our calibrated model, consumers’ income with profit rebates continues to
follows an exponential distribution but just has a higher mean. Because the card adoption costs are proportional to the mean income, card adoption thresholds would remain
unchanged for merchants and consumers. Figure 7 compares social welfare under the
monopoly network versus the Ramsey social planner. The results show that in the beginning years, because the social planner invests more in R&D, social welfare in a given
period is actually lower than that under the monopoly, even though consumer welfare
(without profit rebates) is higher. However, social welfare in each period grows faster
33
under the Ramsey social planner and surpasses that under the monopoly in a few years
and the gap becomes wider over time.
5.3
Policy Experiments
In this section, we use our calibrated model to study two regulatory approaches on payment cards. One is the marginal-cost pricing regulation, and the other is the merchant
fee cap regulation. These two approaches are the most popular ones in policy debates,
and can each be justified by some existing theories.
The marginal-cost pricing regulation can find its root in traditional one-sided markets,
and a naive argument is to require the card network to set fees to merchants and consumers
equal to the marginal cost of serving each side, which implies = and = .
However, as Baxter (1983) pointed out, because payment card markets are two-sided, it
would be decidedly inefficient to block side payments between merchants and consumers.
Instead, the socially optimal card pricing should be + = + . While this approach
considers the two-sided nature of card markets, it is based on a static analysis and ignores
the endogenous R&D of card networks.
On the other hand, the merchant fee cap regulation has been adopted in many countries. Compared with the marginal-cost pricing regulation, it is easier to implement
because it regulates only the fee on the merchant side. In practice, the merchant (interchange) fee cap is rationalized either by an issuer-cost argument (e.g., in Australia and
the U.S.) or by a merchant-benefit argument (e.g., in the EU).
We simulate our calibrated model for each policy experiment by assuming the regulation is implemented at the beginning year of our sample period. For the marginal-cost
pricing regulation, we require the card network to set card fees ( ) to maximize
consumer welfare subject to the zero markup constraint + = . Figure 8 shows
the results. Compared with the unregulated monopoly case, the regulated network lowers
card fees to both merchants and consumers. This boosts card adoption by both sides and
34
the fraction of large merchants who serve both card and cash users increases substantially.
All these result in a higher level of consumer welfare. However, this regulation deprives
the card network of R&D incentive and resources, so the card service cost stays at the
initial level 0 . Figure 9 shows that compared with the unregulated monopoly and the
Ramsey social planner, this regulation yields a lower level of social welfare in each period
except for the initial few years.
We also simulate alternative regulations that cap the merchant fee at various levels.
Figure 8 shows the results for the cap set at 1%. Given the binding cap on the merchant
fee, the card network turns to a higher consumer fee to make up the lost revenue. As a
result, consumer card adoption is lower but the adoption by merchants (including large
merchants who serve both card and cash users) becomes higher compared with the unregulated monopoly case. The card transaction share does not change much but the markup
is lower, so the network profit is reduced. As a result, the card network’s R&D spending
is constrained in the first few years, which leads to a slower decline in card service costs
until the late stage of the sample period.17 Consumer welfare in each period is higher
under the merchant fee cap regulation compared with the unregulated monopoly case, but
as shown in Fig. 9, social welfare gets lower in the early years before it becomes higher
in the longer run.
Figure 10 provides welfare comparison between different scenarios by computing the
present values of consumer and social welfare at the beginning year of our sample period,
with the present value of the cash economy being normalized to 100. The marginal-cost
pricing regulation maximizes consumer welfare in a static setting, but it leaves no profit
for the card network to conduct R&D. As a result, it yields a higher present value of
consumer welfare than the unregulated monopoly, but the present value of social welfare
17
Our calibrated model shows that the merchant fee cap regulation reduces the card network’s profit
level but not the marginal return of doing R&D. Recall the monopoly network’s profit function (32):
(; ) = 1 (0 − 1 ). The merchant fee cap reduces the value of 0 but not the value of 1 (and even
increases it slightly), and 1 determines the rate of cost decline at the balanced growth path. Therefore,
the decline in card service cost does not slow down except for the first few years when the network’s
optimal R&D spending is constrained by its period profit.
35
is lower. In comparison, the merchant fee cap regulation redistributes between network
profit and consumer welfare but without hurting the network’s R&D much. We do see that
the lower the cap, the higher the consumer welfare, but the social welfare (which slightly
increases in the cap value) changes little compared with the unregulated monopoly.
A.
Merchant Card Fee: f
4
m
(%)
Consumer Card Fee: f (%)
c
4
3
3
2
2
1
0
1
-1
0
-2
1998 2000 2002 2004 2006 2008
100
Card Adoption by Consumers (%)
1998 2000 2002 2004 2006 2008
100
80
80
60
60
40
40
20
20
0
Card Transaction Share (%)
Monopoly
Monopoly with Cap 1%
Marginal Cost Pricing
0
1998 2000 2002 2004 2006 2008
1998 2000 2002 2004 2006 2008
B.
Share of Small Merchants:
60
0
(%)
Share of Large Merchants: 160
50
40
40
20
30
(%)
0
1998
1
1
2000
2002
2004
2006
2008
R&D to Income Ratio: R (%)
1998
2000
2002
2004
2006
2008
2006
2008
Card Cost: d (%)
3
2
0.5
1
0
0
1998
4
2000
2002
2004
2006
2008
Card Markup: fm+f c-d (%)
1998
2000
2002
2004
Consumer Welfare
104
Monopoly
Monopoly with Cap 1%
Marginal Cost Pricing
2
102
0
100
1998
2000
2002
2004
2006
2008
1998
2000
2002
Fig. 8. Policy Experiments
36
2004
2006
2008
104
103.5
103
102.5
102
101.5
101
Monopoly
Social planner
Monopoly with cap 1%
Marginal cost pricing
100.5
100
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Fig. 9. Social Welfare Comparison of Policy Experiments.
104
Consumer Welfare
Network Profits
103.5
103
102.5
102
101.5
101
100.5
100
Monopoly
Cap 0.5%
Cap 1%
Cap 2%
Marginal Cost Pricing
Social Planner
Fig. 10. Present Value Comparison of Consumer and Social Welfare.
6
Conclusion
In this paper, we provide a new analysis of payments system evolution and the accompanying competitive efficiency issues. Our theory lays out a two-sided market environment
where consumers and merchants make adoption and usage decisions for electronic payments. The economics of these payment choices, specifically the fixed costs of adoption
37
and marginal benefits of usage intertwined with the heterogeneity of consumer income
and merchant size, together with price coherence, generate nontrivial network externalities. This setting is embedded in a fully dynamic model, in which a monopoly electronic
payment network sets usage fees and conducts R&D to lower costs. We calibrate the
model to U.S. payment card pricing, adoption and usage data, and conduct welfare and
policy analysis.
Our analysis suggests that the market power of electronic payment networks plays
important roles in explaining the slow adoption of electronic payments. In contrast, a
Ramsey social planner who aims to maximize consumer and social welfare would set lower
usage fees and conduct more R&D, thus achieving higher levels of adoption and usage
of electronic payments. We also consider several regulatory interventions. Our findings
suggest that while marginal-cost pricing regulation could maximize consumer welfare in
a static setting, it shuts off the card network’s R&D and may reduce social welfare in a
dynamic setting. In comparison, the merchant fee cap regulation may improve consumer
welfare without causing much dynamic inefficiency.
Our analysis can be extended in several directions. First, our model assumes that the
fixed costs of adopting electronic payments move together with the consumer mean income. This is a simplifying assumption that allows our analysis to focus on cost-reducing
technological progress. In case data becomes available, it would be useful to study how
the adoption costs actually change over time and how that affects diffusion of electronic
payments. Second, by assuming consumer income follows an exponential distribution,
our analysis focuses on changes in the average income while holding the income inequality
fixed. For future research, it might also be interesting to study the effect of income inequality, perhaps by choosing more flexible distributions. Third, the competitive merchants
that we consider provides a first approximation to modeling retail market competition. It
would be interesting to study alternative merchant market structures and their effects on
payment system evolution. Finally, one may extend the framework to consider multiple
38
payments networks. It would be interesting to study how cooperation and competition
among networks affect the market outcome.
References
[1] Aghion, P., Howitt, P., 1998. Endogenous Growth Theory. MIT Press, Cambridge,
MA.
[2] Baxter, W., 1983. Bank interchange of transactional paper: legal perspectives. Journal of Law and Economics 26, 541-588.
[3] Dragulescu, A., Yakovenko, V.M., 2001. Evidence for the exponential distribution of
income in the USA. European Physical Journal B 20, 585-589.
[4] Evans, D., Schmalensee, R., 2005. Paying with Plastic: The Digital Revolution in
Buying and Borrowing, 2nd Edition. MIT Press, Cambridge, MA.
[5] Fudenberg, D., Tirole, J., 1991. Game Theory. MIT Press, Cambridge, MA.
[6] Hall, R., 2001. The stock market and capital accumulation. American Economic
Review 91(5), 1185-1202.
[7] Hall, R., 2004. Measuring factor adjustment costs. Quarterly Journal of Economics
119(3), 899-927.
[8] Hayashi, F., Sullivan, R., Weiner, S., 2006. A Guide to the ATM and Debit Card
Industry: 2006 Update. Federal Reserve Bank of Kansas City.
[9] Hayashi, F., Li, B., Wang, Z., 2017. Innovation, deregulation, and the life cycle of a
financial service industry. Review of Economic Dynamics 26, 180—203.
[10] Kahn, C., Roberds, W., 2009. Why pay? an introduction to payments economics.
Journal of Financial Intermediation 18, 1-23.
[11] Khan, A., Thomas, J., 2008. Adjustment Costs. In: The New Palgrave Dictionary of
Economics. Palgrave Macmillan, London.
39
[12] Kiyotaki, N., Wright, R., 1989. On money as a medium of exchange. Journal of
Political Economy 97, 927-954.
[13] Lagos, R., Wright, R., 2005. A unified framework for monetary theory and policy
analysis. Journal of Political Economy 113, 463-484.
[14] McAndrews, J., Wang, Z., 2012. The Economics of Two-Sided Payment Card Markets: Pricing, Adoption and Usage. Federal Reserve Bank of Richmond Working
Paper No. 12-06.
[15] Mester, L., 2012. Changes in the use of electronic means of payment: 1995-2010.
Federal Reserve Bank of Philadelphia Business Review, Third Quarter, 25-36.
[16] Nosal, E., Rocheteau, G., 2011. Money, Payments, and Liquidity, MIT Press, Cambridge, MA.
[17] Rochet, J., Tirole, J., 2002. Cooperation among competitors: some economics of
payment card associations. RAND Journal of Economics 33, 549-570.
[18] Rochet, J., Tirole, J., 2011. Must-take cards: merchant discounts and avoided costs.
Journal of the European Economic Association 9(3), 462—495.
[19] Rysman, M., Wright, J., 2014. The economics of payment cards. Review of Network
Economics 13, 303-353.
[20] Shy, O., Wang, Z., 2011. Why do payment card networks charge proportional fees?
American Economic Review 101(4), 1575-1590.
[21] Wang, Z., Wright, J., 2017. Ad valorem platform fees, indirect taxes, and efficient
price discrimination. RAND Journal of Economics 48, 467—84.
[22] Wright, J., 2003. Optimal card payment systems. European Economic Review 47,
587-612.
[23] Zinman, J., 2009. Debit or credit? Journal of Banking and Finance 33(2), 358-366.
40
Appendix
A1. Card Adoption by Consumer Income and Merchant Size18
Share %
100
80
60
40
20
0
1970
1977
1983
Income Quintile 1
1989
IQ2
1992
IQ3
1995
1998
IQ4
2001
IQ5
Fig. A1(a). Percentage of U.S. Households Holding Credit Cards
Share %
100
80
60
40
20
0
1995
Low Income
1998
2001
Moderate Income
2004
2007
Middle Income
2010
Upper Income
Fig. A1(b). Percentage of U.S. Households Using Debit Cards
18
Data sources: Mester (2012) and Evans and Schmalensee (2005). They also show similar adoption
and usage patterns for other electronic payment means, such as smart cards and automatic bill paying.
41
Share %
80
1996
2001
60
40
20
0
movie
tickets
fast food mid-price high-price
(Entertainment)
(Restaurant)
grocery
dept.
(Retail Store)
Fig. A1(c). Percent of Transactions Using Payment Cards
A2. Derivation of Eqs. (L1) and (L2).
Under Assumptions F1 and F3, we can rewrite Eqs. (20) and (22) in the main paper
into
0 =
2(−0 ) (1
+ 0 − )0
(40)
2
(1−0 )
)
( 1+
1+
0 =
R
min(1 1)
(1− )
(1−0 2 )
( 1+
)
−
exp(2
ln(
))
1+
(1− )−(1+ )0 0
0
(41)
Equation (40) is exactly Eq. (L1). We can further rewrite Eq. (41) by noting that
Z
(1 − )
)
(1 − ) − (1 + )0 0
2
1
0 0 (1 + )
=
ln(1 − ) − 2 ln((1 − ) −
)
2
2
0 0 (1 + )
2 2 (1 + )2
0 0 (1 + )
ln( −
+ 0 0
)
+
2
2(1 − )
2(1 − )
(1 − )
ln(
42
Hence, we derive
2
Z
0
1 0
= 20
ln(
(0
(1 − )
)
(1 − ) − (1 + )0 0
)
0
0 (1+ )
− (1−
1 + 0 (1 + ) 0
02 (1 + )2
1
)
ln
( − 1) +
+
ln(
)
0 (1+ )
1 +
(1 − ) 1
(1 − )2
1 − (1− )
Equation (41) can then be rewritten into Eq. (L2) where
0
0 (1+ )
− (1−
1 + 0 (1 + ) 0
02 (1 + )2
1
)
= ln
( − 1) +
+
ln(
)
(1+
2
)
0
1 +
(1 − ) 1
(1 − )
1 − (1−
)
A3. Model Solution: Monopoly Network
Given functional form assumptions F1-F4, we can solve the model equilibrium under
the monopoly network.
Within-Period Decision Within each period, the monopoly card network maximizes its profit as follows.
(; ) = max
(−0 )
1 − 0 2
(1 + 0 − )(
)( + − )
1 +
0 =
1 =
2(−0 ) (1
0
0
1
+ 0 − )0
if ≤
(42)
(43)
(44)
2
(1−0 )
)
( 1+
1+
0 =
R
min(1 1)
(1− )
(1−0 2 )
( 1+
)
−
exp(2
ln(
))
1+
(1− )−(1+ )0 0
0
≥
1 −
1 −
≥
1 +
1 +
43
(45)
(46)
∗
For a given marginal cost and mean income 1, we can derive ∗
0∗ ∗0 . The
corresponding maximum profit is
∗
(; ) =
1 − ∗0 2 ∗
(−0 )
∗
)( +
− )
(1 + 0∗ − )(
1 + ∗
(47)
∗
Conditioning on , Eqs. (43-45) imply that the optimal values (
∗ 0∗ ∗0 ) are
invariant to . Therefore, according to Eq. (47), the profit function is proportional to the
mean income 1 . Moreover, we can verify that is approximately linear in (as shown
in Fig. A2 based on the model calibration), so that
(; ) =
1
(0 − 1 )
where 0 and 1 are scalars determined by the model parameter values.
Fig. A2. Network Profit Function and Linear Fitting
Dynamic Decision We now add back the time subscript to each variable. The
value function of the monopoly network is as follows.
( ; ) = max ( ; ) − + (+1 ; +1 )
44
(48)
1
+1
−
1
−−1
)
=(
( ; ) ≥
(49)
(50)
Assuming the budget constraint (50) never binds, the dynamic problem can be rewritten as
∙
¸ 1
1
( ; ) = max (0 − 1 ) −
− 1 + (+1 ; +1 )
+1
+1
The first-order condition implies that
∙
¸2
¸ 1 −1 ∙
−1
+ 0 (+1 ; +1 ) = 0
+1
+1
(51)
The envelope condition implies that
∙
∙
¸ 1
¸ 1 −1
1
−1 −
−1
( ; ) = − −
+1
+1
+1
0
which suggests that
0
(+1 ; +1 )
+1
∙
∙
¸ 1
¸ 1 −1
1
+1
+1
+1
=−
−
−1 −
−1
+1 +1 +2
+1 +2
+2
(52)
We can combine Eqs. (51) and (52) to get
∙
¸2
∙
∙
¸ 1 −1 ∙
¸ 1
¸ 1 −1
+1
1
+1
+1
−1
=
+
−1 +
−1
+1
+1
+1 +1 +2
+1 +2
+2
Calibrating = 12, we can rewrite the above equation as
∙
¸2
∙
∙
¸∙
¸2
¸
2
1
+1
2 +1
+1
−1
=
+
−1 +
−1
+1
+1
+1 +1 +2
+1 +2
+2
Given the constant income growth rate
+1
45
(53)
= , we can solve for the balance growth
path where
+1
=
+1
+2
= . Note that we can rewrite Eq. (53) as
2
[ − 1] 2 = 1 + [ − 1]2 + 2 [ − 1]
and solve for a constant cost declining rate . Based on , we can then solve dynamic
paths of all the endogenous variables in the model. We can also verify that the budget
constraint (50) never binds.
A4. Model Solution: Ramsey Social Planner
Given functional form assumptions F1-F4, we can solve the model equilibrium under
the Ramsey social planner.
Within-Period Decision
Within each period, for a given income-scaled R&D expenditure , the Ramsey social
planner maximizes consumer welfare as follows.
( ; ) =
Z
∞
−
+
0
Z
0
−
0
= exp(ln
− ln )
⎧
R
R
⎪
⎨ min(1 1) 2 ln( ) + 1
2 ln(
)
0
min(1 1)
= exp
⎪
⎩
+(1 − 2 ) ln( 1+ ) + ln( − )
0
=
exp(ln
(
)1≥≥min(1 1) =
− ln ) = exp
1+
½Z
⎫
⎪
⎬
⎪
⎭
(54)
¾
2 ln(
)
min(1 1)
1
1− −0
(1 + 0 − ) + 1−
[1 − −0 (1 + 0 )] − 2
1+
1+
1− −0
1−
(1 + 0 − ) + (1+
[1 − −0 (1 + 0 )]
1+
)
46
(55)
(
( 1−
)−0 (1 + 0 − ) −
1+
)min(1 1)>0 =
1− −0
(1 + 0 − )
1+
0 =
0
0
1
+ 0 − )0
2(−0 ) (1
1 =
2
if ≤
2
0 =
1−0
)
( 1+
1+
2
1−0
( 1+
)
− exp(2
1+
≥
R min(1 1)
0
(1− )
ln( (1− )−(1+
))
)0 0
1 −
1 −
≥
1 +
1 +
= (−0 ) (1 + 0 − )(
1 − 20
)( + − ) ≥
1 +
∗
0∗ ∗0 .
For a given marginal cost and the profit target , we can derive ∗
The corresponding maximum consumer welfare is
( ; ) =
Z
∞
0
=
−
½Z
Z
+
0
−
0
∞
−
exp{·}
0
+ exp{∗}
(56)
Z
0
−
0
¾
where exp{·} and exp{∗} are given in Eqs. (54) and (55), respectively.
∗
0∗
We can again verify that conditioning on and , the optimal values (∗
∗0 ) are invariant to , so according to Eq. (56) the welfare function is proportional to
the mean income 1 . Moreover, we can verify that is approximately linear in and
(as shown in Fig. A3 based on the model calibration). Therefore, the welfare function
47
can be simplified as
( ; ) =
1
(0 − 1 − 2 )
where 0 , 1 and 2 are scalars determined by the model parameter values.
Fig. A3. Consumer Welfare Function and Linear Fitting
Dynamic Decision We now add back the time subscript to each variable. The
value function of the Ramsey social planner is as follows.
( ; ) = max ( ; ) + (+1 ; +1 )
1
+1
−
1
−−1
)
=(
The dynamic problem can be rewritten as
∙
¸ 1
1
( ; ) = max (0 − 1 − 2 ) − 2
− 1 + (+1 ; +1 )
+1
+1
The first-order condition implies that
∙
¸ 1 −1
2
0
−1
+ (+1 ; +1 ) = 0
2
2
+1 +1
48
(57)
The envelope condition implies that
∙
∙
¸ 1
¸ 1 −1
1
2
− 1 − 2
−1
( ; ) = (−1 ) −
+1
+1 +1
0
which suggests that
∙
∙
¸ 1
¸ 1 −1
+1
2 +1
+1
(−1 ) −
− 1 − 2
−1
(+1 ; +1 ) =
+1
+1 +2
+2 +1 +2
1
0
(58)
Combining Eqs. (57) and (58), we get
∙
∙
∙
¸ 1 −1
¸ 1
¸ 1 −1
+1
2
2 +1
+1
2
−1
=
1 +
− 1 + 2
−1
2+1 +1
+1
+1 +2
+2 +1 +2
Calibrating = 12, we can rewrite the above equation as
∙
∙
∙
¸
¸2
¸
+1
2
2 +1
+1
−1 =
1 +
− 1 + 22
− 1 (59)
22 2
+1 +1
+1
+1 +2
+2 +1 +2
Given the constant income growth rate
path
+1
=
+1
+2
+1
= , we can solve for the balance growth
= . Note that Eq. (59) can be rewritten as
22
[ − 1] 2 = 1 + 2 [ − 1]2 + 22 [ − 1]
and we can solve for the constant cost declining rate . Based on , we can then solve
dynamic paths of all the endogenous variables in the model.
49
@FedFRASER