Interchange Fees and Credit Card Issuing: An Empirical Investigation

Interchange Fees and Credit Card Issuing: An Empirical Investigation∗ Niyati Ahuja† New York University November, 2008

Abstract This paper develops a framework for the empirical analysis of the role of interchange fees in the issuer–consumer relationship. In the credit card market, the interchange fee is the fraction of each transaction that is retained by the issuing bank. Under a market environment where merchants typically do not surcharge credit card users, these fees have important competitive consequences and their collective setting by card networks raises policy concerns. Although the theory literature in this context is very rich, empirical work on the competitive effects of interchange is relatively behind. I propose a step in addressing this gap. I consider the issuer–consumer side of this market and build a structural model with heterogeneous consumers and competing issuers. The primary goal of this paper is to measure the extent to which issuers pass on changes in interchange fees via interest rates to consumers and the degree of consumer sensitivity to issuers’ policies. Using a combination of micro and macro data, I estimate demand and supply parameters for general purpose credit cards and analyze equilibrium for a given level of interchange fees. A comparison of equilibrium values across different levels of interchange then allows predictions on how consumers’ use of cards and issuer policies respond to changes in these fees. The results indicate that reducing the fee down by about a third increases interest rates on average by less than 2.5% of their original levels and consequently, leads to a 2.9% decline in credit card debt and a 2.7% fall in transactions. The paper also sheds new light on the nature of competition in the market by generating cross– elasticities for competing issuers.

∗

Preliminary and Incomplete. Please do not cite. Department of Economics, New York University, New York 10012. Email: [email protected]. I am indebted to my advisory committee – Ariel Pakes, Michael L Katz and Daniel (Yi) Xu, for their constant guidance and encouragement. I would also like to thank Chaim Fershtman, Fumiko Hayashi, Marc Rysman, Jonathan Zinman and seminar participants at NYU for valuable input. Errors are all mine. †

1

1

Introduction

Each time a credit card transaction takes place through a network such as Visa or MasterCard, the merchant whose location the card was used at pays a proportion of the transaction value as fees to the bank that processes its accounts (the acquiring bank). After retaining a small markup, the acquirer passes on this fee to the consumer’s bank (the issuer) in the form of interchange fees. The levels and structure of the interchange fee are usually determined by the networks (Visa or MasterCard), as opposed to being negotiated between thousands of issuers and acquirers. On the one hand, the interchange fee may offer an instrument for networks to balance card usage among consumers and card acceptance among merchants. On the other hand, this fee may impose externalities on agents not involved in the transaction because merchants typically charge the same price to all their customers and are likely to pass on the fees as higher prices to non-card users as well. It is due to these concerns that the credit card industry in general and the interchange fee in particular, have been a subject of much attention from economists and policy makers. This paper builds an empirical framework to analyze the competitive effects of changes in interchange on the consumer-issuer side of the market under the assumption that the merchant-acquirer side does not respond. The primary goal is to present a model that enables one to estimate demand and supply parameters for general purpose credit cards and use it to measure the extent of issuer pass-through of the fees via interest rates. I develop a structural model that captures two important aspects of this industry, namely, heterogeneity of consumers and differentiation among cards. Consumers choose cards to maximize their utility which depends on card characteristics and their own attributes. Competing issuers choose interest rates to maximize profits given demand and exogenously set interchange fees. The empirical model and estimation procedure fit three quantity variables (credit card accounts, debt and transactions) to their counterparts observed in the data. The model provides a way to tie these three variables together and presents a channel for interchange fees to affect equilibrium interest rates. Counterfactual experiments on the estimated model then allow me to predict the effect of changes in the fees on issuer profits, consumers’ choice of cards and their credit card balances. I find that regulating the interchange fee down by about a third increases interest rates on an average by less than 2.5% of their original levels, which implies that the mean APR would increase from 15.4% to 15.8%. Although this does lead to a drop in credit card usage (accounts, debt and transactions), the change is not dramatic. Even when interchange fee is reduced by nearly 50%, the impact on the key quantity variables is close only to 5%. These findings are in line with the observed impact of regulation in Australia, where among other reforms, the interchange fee was reduced by about 50% in late 2003. Based on market data trends, Chang et. al. (2005)[10] and Gans (2007) [13] report that even after several years, there was no dramatic effect of reforms over a broad range of indicators. The analysis presented here also points to the possible reason. Interchange fee accounts 2

for about one-fourth of issuer profit, however, the rest is mainly interest income. The sensitivity of the estimated demand to interest rates prevents issuers from raising them substantially. This also suggests that the decline in interest rates in recent years may have had a lot to do with demand conditions.

1.1

Industry Background

The credit card industry has seen tremendous growth in the last couple of decades. In 2005, credit cards accounted for 25.2% of the total volume of consumer payments in the U.S., up from 14.5% in 1990.1 According to the 2004 Survey of Consumer Finances, 74.9% of all the families held atleast one credit card and 58.0% of these carried balances on them at the time of the survey, the mean balance being $5,100.2 This industry has a complex, yet fascinating, organizational structure. Payment systems are two-sided markets because for them to work consumers must use them and merchants must accept them. A credit card network needs to bring both sides of end-users on board for it to run successfully. In the United States credit card market, Visa and MasterCard are the two largest card networks collectively accounting for over 70% of all transactions.3 These are commonly categorised as Open-Loop networks or Four-party systems, implying that they do not deal with card holders and merchants directly. They consist of a number of member banks, those that issue cards to consumers (issuing banks or issuers) and those that settle accounts with sellers (acquiring banks or acquirers). The issuer sets all the terms and conditions for the contract with the consumer such as the credit line, APRs, annual fees, reward features, late fees etc., and the consumer settles her monthly statement with the issuer directly. The acquirers deal with the merchant side of the market. They set up the terminal equipment, keep track of transactions and transfer funds to settle accounts on a day to day basis. The network itself is an association of these banks that is responsible for the collective setting of rules and regulations, for maintaining the infrastructure and advertising among other functions. Figure 1 illustrates a simplified example of how a transaction in this system works. It begins with the consumer swiping her credit card at the merchant location. The card information is read through the magnetic strip and relayed along with the transaction amount to the acquirer who then forwards it to the network for authorization. Upon approval, the issuer credits the acquirer the full amount of the transaction minus an interchange fee and the acquirer credits the merchant the transaction amount minus a merchant service fee, usually within a 24-hour period. At the end of the billing cycle, the issuer sends the bill to the consumer who subsequently pays for the goods bought along with any other fees due.4 In this example, the transaction value is $100, the interchange fee is $1.50 1

The Nilson Report, Issues 656 and 869 Bucks, Kennickell and Moore (2006) [5] 3 American Express and Discover form most of the remaining 30%. For 2005, the Nilson Report (Issuer 851) estimates the following market shares for these four networks: Visa-42.7%, MasterCard-29.8%, American Express-22.3% and Discover-5.2%. 4 This system is in contrast to the Closed-Loop network, like Diner’s Club and until recently, American Express and 2

3

CREDIT CARD NETWORK Eg Visa or MasterCard

ISSUING BANK Eg Citibank

ACQUIRING BANK Retains Interchange Fee

CONSUMER

Eg First Data

MERCHANT

Figure 1: Example of a Transaction in an Open–Loop Network and the merchant discount fee is 50 cents. Acquiring is known to be a rather competitive business, so the merchant fee is usually not much higher than the interchange fee. In most cases, the network receives a small percentage as well. The interchange fee structure is a point of interest here. The fees are collectively set by the network and they differ across the two networks and by the type of merchant or the type of card. Supermarkets and convenience store fees are low, hotels and internet sellers incur higher fees. Bigger merchants may be able to bargain for lower fees. Premium cards have higher fees and merchants are required to accept these under their agreements with the network. In the U.S., the Wall Street Journal estimated in 2006 that card issuers earned over $30 billion in interchange fees, up by 117% since 2001 and that the average card fee was about 1.75%. The complete fee structure is mainly hidden from the public because Visa and MasterCard do not disclose details on monthly statements and they keep merchants from disclosing them on receipts. The card networks and issuers argue that the fees are essential for maintaining a balance over the two sides of the market and that collective setting saves negotiation costs for the large numbers of issuers and acquirers involved. Policy makers are concerned that under No-Surcharging (which means that merchants do not Discover, which operate as a single entity and deal with the card holder and the merchant directly. In a closed network, a transaction follows a very similar route except the network independently handles all issuing and acquiring functions and therefore, there is no need for an interchange fee. Visa and MasterCard had previously enforced exclusionary rules that prohibited member banks from issuing rival network credit cards. The U.S. Department of Justice brought an antitrust lawsuit against these networks in 1998 and after a trail and a failed appeal in 2004, banks such as MBNA, Citibank and Bank of America are now able to issue American Express credit cards. (United States of America v. Visa U.S.A. Inc., Visa International Corp., and MasterCard International Incorporated, Complaint for Equitable Relief for Violations of 15 U.S.C. §1, October 7, 1998)

4

charge higher prices to credit card users5 ), the marginal cost to society of using a payment system is not reflected to consumers and this may lead to usage externalities. Higher interchange fees are passed on by acquirers to merchants as higher merchant discount fee. This makes card acceptance costly for merchants6 and they pass on the fee (at least in part) to all their consumers including people who pay by checks or cash. At the issuer’s end, if the higher interchange fee is passed on to consumers as lower fees and higher rewards, it is likely to distort consumer choice of payment system in favor of credit cards. This raises distributional concerns because it implies that non-card users (and partly merchants) end up cross-subsidizing card users. This is important because people who do not possess credit cards on average have lower income than those who do. Also, merchant groups have increasingly expressed dissatisfaction over the fees and there are several ongoing lawsuits filed by retailers in this regard. It is due to these concerns that Australian authorities regulated their interchange fees in 2003 and other countries are considering the same.7 In the U.S., direct regulatory action has not been undertaken yet but the debate is ongoing.8 The latest piece of legislation being considered is the Credit Card Fair Fee Act of 2008, which will allow large and small businesses to negotiate directly with credit card companies in an effort to reduce interchange fees.9

1.2

Research Question & Limitations

The role of interchange fee in the credit card market has been explored extensively in the literature. However, the majority of this research is theoretical. Not surprisingly, there is lack of consensus10 in the theory, emerging primarily due to different modeling assumptions. Many of the theoretical findings pose empirically testable hypotheses. For instance, whether issuers do indeed pass on higher interchange fees to consumers and if this alters consumer choice is largely an empirical question. Nevertheless, empirical work in this area is challenging because of the complexity of the market and the difficulty of obtaining comprehensive data. I circumvent these difficulties here partly by using 5

In the U.S., Visa had imposed this as a contractual condition for a long time before relaxing it recently. MasterCard still prohibits surcharging, but does allow discounts for cash users. However, in practice U.S. merchants rarely offer discounts for cash payments. 6 It is interesting to examine why merchants accept credit cards in the first place if they are costly for them. Rochet and Tirole (2002) [31] use a Hotelling model to show that merchants accept cards for strategic reasons, to attract card-users from rivals who do not accept cards. In equilibrium, merchants do not benefit from accepting cards and end up in a Prisoner’s dilemma kind of situation. 7 See Reserve Bank of Australia (2002) [28] and Katz (2001) [21] for details. See Weiner and Wright (2005) [39] for interchange fees related developments in various countries. 8 See Federal Reserve Bank of New York Conference “Antitrust Activity in Card-Based Payment Systems: Causes and Consequences” (Sept 2005) and Federal Reserve Bank of Kansas City Conference “Interchange Fees in Debit and Credit Card Industries: What Role for Public Authorities?” (May 2005). 9 See Evans and Schmalensee (2005) [12] for the history of regulatory action in the U.S. and for in depth information about this industry. 10 See Section 1.3 for a rather brief look at the literature, Katz (2005) [22] for a summary of key findings and Chakravorti (2003) [8], Rochet (2003) [30], Schmalensee (2003) [34] or Hayashi and Wiener (2006) [19] for detailed surveys.

5

data that is easily accessible to researchers and partly by focusing on a narrower research problem. A comprehensive empirical investigation of interchange would need to explore both the consumer and the merchant sides of the market. It would investigate how both issuers and acquirers adjust their pricing in response to changes in interchange fees and how consumers and merchants, in turn, change their acceptance decisions. The acquirer-side pricing problem in this market is relatively simple, but issuer-pricing is not. A credit card is a product that can be differentiated along several dimensions such as interest rates, rewards and a menu of fees that include annual fees, late fees, minimum finance charges etc. Some of these features may be personalized based on consumer creditworthiness. In addition, a complete analysis would also build in competing networks and allow for substitution to other modes of payment. Due to data limitations, such a detailed study is beyond the scope of this paper. Instead, it confines itself to the following narrower questions: In response to changes in interchange fees in the open-loop network: 1. How do issuers adjust their base interest rates? To what extent do they pass on the change in interchange via interest rates to consumers? 2. Assuming that merchants do not change their acceptance decisions and given the changes in base interest rates by the issuers, how do consumers adjust their demand for cards and credit card debt? This paper focuses the issuer-consumer relationship, assuming that the acquirer-merchant side does not adjust. Essentially, my analysis boils down to considering credit cards as a one–sided market. The underlying assumption is that the cost of not accepting credit cards is relatively high for merchants and hence, small changes in interchange will not alter merchant acceptance. Also, if merchants change overall prices in response to interchange fees, I assume that consumers cannot foresee this and hence, do not take it into account in their decision making process. In addition, this paper confines itself to analyzing the response of interest rates to changes in interchange fee, assuming that other characteristics such as reward features and annual fees are left unchanged. These other features could, by all means, be sensitive to interchange fees and could respond to changes in the fee structure. This would mean that the results presented here over-estimates the actual extent of pass-through. It must also be kept in mind that these results are based on issuer-level data. As with any structural empirical problem, the results would be sharper if micro-level choice data were available. This approach comes with a price. It implies that my analysis is unable to directly address the broader policy question. Nevertheless, it does offer a starting point and a meaningful addition to the existing empirical literature on credit cards. The focus here is on issuer pass-through via interest rates, which is a key component in determining the competitive effects of interchange fees. It is believed that in the United States, acquirers pass through a higher percentage than issuers and 6

issuers’ pricing more involved. I propose a modeling framework that builds together heterogeneous consumers and competing issuers. The model offers a way to estimate demand and cost parameters and cross-price elasticities among competing issuers. The paper generates an initial estimate of the extent of pass-through and presents a new look at the factors that influence demand.

1.3

Related Literature

The theory literature concerning interchange fee focuses mainly on establishing a distinction (or lack thereof) between privately and socially optimal fees and exploring their welfare implications. This work began with Baxter’s (1983) [2] normative analysis who derived the socially optimal fee in a model where agents do not interact strategically (single merchant, single consumer and perfect competition among banks) and a payment transaction is considered a joint-service to consumers and merchants. The paper emphasized that the fees are necessary to balance costs on both sides and that bilateral negotiation of interchange fee would be costly. The work following Baxter considered implications of competition between networks, issuers and acquirers and variations in consumer and merchant demand characteristics. Katz (2001) presented a model with perfectly competitive issuers and acquirers and no surcharging by merchants. He concludes that inefficient use of cards (termed therein as economically excessive use of cards) may result under the absence of surcharging because merchants cost differences represent social cost differences, but the card holder does not bear the costs imposed on merchants. Scwartz and Vincent (2006) [35] similarly conclude that under no surcharging, when rebates to card users are feasible, the networks raise interchange fee which leads to higher total consumer surplus but hurts cash users. Rochet and Tirole (2002) [31] presented the industry as a two-sided market and considered platform competition for a few different governance structures. With competing networks, one of the factors that influences the optimal interchange fee is whether consumers single-home or multi-home, that is, whether they hold cards from a single or multiple networks. For instance, Guthrie and Wright (2007) [17] show that when some consumers multihome, merchants reject the more expensive card, causing competing schemes to lower the interchange fee in an attempt to attract merchants. The literature so far (notably Schamlensee (2002) [33], Gans and King (2003) [14] and Chakravorti and Roson (2004) [9] among several others) highlights that the welfare consequences of interchange fee depend crucially on the underlying market structure - the extent of competition among suppliers and the nature of demand by consumers and merchants. This paper follows the literature in assuming that the extent of pass-through of the fee is crucial and that the degree of pass-through depends on underlying demand conditions and competition among issuers. On the empirical front, a seminal article that examines profitability and pricing in the credit card market is Ausubel (1991) [1]. This article provides evidence for an the apparent puzzle in this market during the 1980s - the credit card interest rates were sticky and profits high, despite

7

the fact that the market structure appears rather competitive. Ausubel offered adverse selection, search costs and switching costs as explanations. Calem and Mester (1995) [6] provided empirical evidence using the 1989 Survey of Consumer Finances that all these three factors may be important in explaining Ausubel’s puzzle. Rysman (2007) [32] comes closest to a comprehensive empirical study of both sides of the market. He exploits a unique data set and shows that consumers concentrate their spending on a single payment network (single-homing), although many maintain unused cards from multiple networks (multi-homing). The paper also establishes a correlation between consumer usage and merchant acceptance and suggests that there exists a positive feedback loop between consumer usage and merchant acceptance. Other empirical work has focused on several different aspects namely, determinants of the demand for credit cards and debt,11 price setting behavior by issuers12 and costs of payment systems.13 This paper draws from all three of the existing empirical literature on the credit card market in so far as it attempts to analyze profit, costs and demand under a unified framework. It also closely follows the the empirical antitrust literature using methodologies set forth in Berry et.al.(1995) [4] and used by several authors in the context of the banking industry.14 The rest of the paper is organized as follows. Section 2 describes data sources and summarizes key trends observed in the data. Sections 3 and 4 outline the empirical model and estimation strategy. Sections 5 and 6 discuss the main findings and results from the policy experiment.

2

Data Description

2.1

Data Sources

The data used in this paper has been compiled from a number of different sources. Product-level data for credit card issuers is pooled together from the Nilson Report and the Terms of Credit Card Plans (TCCP). Consumer-level micro data comes from the Survey of Consumer Finances (SCF). These three sources combined form the core data set used and are supplemented by additional information about issuing banks and network interchange fees. The quantity data comes from the periodical Nilson Report, which is one of the leading trade publications for consumer payment systems. They provide detailed information on a variety of aspects of credit cards and other payment media worldwide. Every year they publish a list of the U.S. credit card industry’s top 150 (or more) issuers providing three key quantity variables used in this paper: number of active accounts, total outstanding debt and total transaction volume per issuer. This information is matched with pricing and credit card characteristics data from the Terms 11

Castronova and Hagstrom (2004) [7], Hayashi and Klee (2003) [18], Gross and Souleles (2002a) [16], Kim and DeVaney (2001) [23] 12 Knittel and Stango (2003) [25], Park (1997) [29], Stango (2002) [36] and (2003) [37] 13 Garcia-Swartz, Hahn and Layne-Farrar (2004) [15] 14 Dick (2008) [11], Ishii (2004) [20] and Zhou (2007) [40]

8

of Credit Card Plans. The TCCP is a semiannual survey of roughly 150 issuers conducted by the Board of Governors of the Federal Reserve. The survey is sent to 25 of the largest credit cards issuers and at least 125 additional institutions including regional issuers. Each respondent provides details about its credit card plan that has the largest outstanding number of cards. They report the annual percentage rate (APR), annual fees, grace period, balance computation method, finance charges, late fees and reward features among other characteristics. For each issuer, I use the second TCCP entry in the year (the July data instead of the January) wherever available. I derive additional issuer-level data from the Federal Reserve’s National Information Center and from the Call Reports filed by the issuing banks. The former is a repository of financial data and institutional characteristics maintained by the Federal Reserve. From here I get information on the type of institution that the issuer is (such as National Bank, Savings Bank etc.) and also the identifying RSSD-ID numbers for the issuers that help match data under TCCP to that in Nilson. The Call Reports are used to obtain some key cost variables that are used as instruments in the estimation.15 The reports also help determine whether an issuer is a monoline credit card bank. If 75% or more of the bank’s reported total assets are held in credit card balances, I regard it as a monoline issuer. Finally, on the issuers’ side, the interchange fees used are the default retail store non-supermarket rates.16 Lastly, the paper uses household-level data from the 1998, 2001 and 2004 Surveys of Consumer Finances. This is a triennial survey of approximately 4,500 respondents17 conducted by NORC at the University of Chicago and sponsored by the Federal Reserve Board. It is the most comprehensive publicly-available source of data on household financial and credit characteristics. It has a few useful questions regarding credit attitudes and the use of payment cards. The SCF has been used in several studies relating to payment cards, especially credit cards.18 Unfortunately, this data does not identify credit card issuers by name. Hence, I cannot match this data with the quantity and price data above. Nevertheless, as outlined in section 4 this data plays a crucial role in the estimation.

2.2 2.2.1

Data Summary Issuer-level Quantities and Card Characteristics

The market is defined as the market for nationally-issued Visa and MasterCard credit cards. I construct a panel data set spanning ten years from 1996 to 2005 by matching Visa and MasterCard national credit card issuers from the Nilson Report and TCCP. After choosing institutions that are common to both and eliminating missing observations, I am left with a total of 300 issuers. As 15

See section 4 for a description of these. I would like to thank Fumiko Hayashi for granting me access to these data. 17 For the years relevant here, this is not a panel data set. The last panel SCF was conducted during the years 1983-1989. Since then it has only had a different cross-section of households every three years. 18 Calem and Mester (1995) [6], Klee (2006) [24], Stavins (2001) [38] and Zinman (2006) [41] to name a few 16

9

indicated in Table (1), this isn’t a balanced panel. In some years the data has more than forty issuers, while in most there are less than thirty. However, the market concentration in this industry during this period is high and the data includes the biggest issuers, so it captures 70 to 85 percent of the market for most years. Market shares have been calculated using aggregate market quantities which are also reported in Nilson. The mean share of the inside good in the data ranges from 1.39% to 4.37%. The C(1) measure reported in Table (1) shows a general upwards trend over time, somewhat reversed in the last year (2005). Tables (2) and (3) provide definitions and summary statistics for the price and quantity data. The average number of active accounts per issuer is 5.63 million and the mean credit card debt and transactions are $13.85 billion and $25.21 billion respectively. The standard deviation for debt and transactions is very high which again points to the concentrated nature of this market. From the TCCP pricing data, I obtain APR, annual fee, rewards, affiliated network (Visa/MasterCard), type of APR (Fixed or Variable) and length of grace period.19 Note that under TCCP the issuers are required to report the plan that had the largest number of cards outstanding and that was available to new customers as of the report date. I drop other reported characteristics such as late fee, over limit fee, cash advance fees etc., because these are not reported by issuers in a uniform manner (some report dollar values and others report them as percentage of outstanding balance) and in preliminary regressions these did not turn out to be significant determinants of demand. In the profit equation, these fees will be absorbed into the marginal cost of transaction. The mean APR in the data is 15.43%, which is a little more than double the mean prime rate during this ten year period of 6.85%. The average annual fee is $11.48 and 64% of all issuers report their annual fees as zero. The variable rewards is a summary measure of enhancements offered under the credit card plan. In the TCCP data, unfortunately, the issuers do not report details about their reward programs. They simply report a zero or one value for reward features rebates on purchases, purchase protection, travel discounts etc. and these cannot be used to assign a dollar value to the reward features. I construct a summary measure by taking the sum of the enhancements reported. The maximum possible value for this variable is 10 and the average is 1.6. The majority (63%) of the credit cards in the data are Visa cards. Table (3) does not show year-wise statistics, but it is interesting to note that in this dataset the proportion of Visa plans reduces over time. Fixed APR is offered much less frequently than variable APR and among the variable APR plans, most are tied to the prime rate (73.3% of all cards). The average grace period is about 25 days. Turning now to issuer classification, national banks are the biggest group in the data. They make up 61% of all issuers followed by state member banks (12.3%) and savings banks (6.3%). A third 19 The grace period is the interest-free period at the end of the billing cycle (float period). Typically, people that have previous outstanding balances (revolvers) do not enjoy a float period since they start accruing interest charges immediately. This is an important distinction between credit cards and debit cards. Credit cards provide an option to carry debt and a float period if you choose not to carry debt. Zinman (2006) explores how this difference influences use of debit cards.

10

25 Visa $1,000 IF MC $1,000 IF Avg CC Interest Rate (%) Prime Rate (%)

20

15

10

5

0 1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

Figure 2: Interchange Fee and Credit Card Interest Rates of all plans here are offered by institutions that specialize in credit card issuing, known as monoline issuers. Notable examples are MBNA and Capital One. 26% of the plans in the data were offered by issuers that are known to consider risky (subprime) borrowers. Providian and Household bank were major subprime lenders during this period. As mentioned earlier, this is a rather concentrated market and the group of issuers included are very diverse. I include a dummy variable that takes the value one if an issuer was one of the top 25 issuers listed in Nilson (based on outstandings) during the year before. In this data, 55.3% of all issuers fall under this Big Issuer category. This variable is intended to capture the fact that a lot of these issuers are very well know around the country while others may be issuing national credit card plans but have branches in certain regions and are better known only locally. For example, Citibank, Bank of America and Chase are consistently among the biggest issuers. This is similar to the approach taken by previous authors working with demand estimation for the banking industry who classified banks as big, medium or small (Dick (2008) [11] and Zhou (2007) [40]) based on total assets. The advantage of defining the variable this way is that it is predetermined and it captures the reputation effect. Figure (2) depicts the average interest rates observed in the data for each year against the Visa and MasterCard interchange fees for that year. The mean interchange fee during this entire period was 1.46% and Visa’s fee was usually lower than MasterCard’s, though both rose steadily. The

11

graph appears to depict an inverse relationship between mean interest rates and interchange fees. This occurs simultaneously with a positive movement between interest rates and the prime rate. 2.2.2

Household Attributes

From the Survey of Consumer Finances, I obtain information on the subsample of credit card holders. This includes household demographic and financial characteristics, credit worthiness and credit attitude indicators as well as some details regarding credit use. The data used is summarized in tables (4) and (5). The average number of cards per household is 2.4 and the mean interest rate reported is 13.38%.20 The mean credit card debt over all card holders is $2,395 and that amongst revolvers is $4,375. The differences between convenience users and revolvers are interesting to note here. Households that carry debt on their credit cards on average have lower income, financial assets and non-financial assets. Their credit worthiness indicators point towards higher risk as compared to convenience users. In particular, a higher percentage are subprime21 (16.58% compared to 7.8%). Fewer of them own their homes and have checking accounts. 18.24% of households that carry credit card debt report being refused credit in the past or not applying due to the fear of getting rejected. This proportion is only 4.5% among those that pay off their balances regularly. Revolvers also report attitude indicators that support an inclication towards borrowing behavior. For instance, when asked if they think its alright for someone like them to borrow for financing a vacation 19.3% answer affirmatively as compared to 10.9% of convenience users. The demographic characteritics of the two groups differ as well. Among revolving households, the head of the household reports a comparatively lower age, lesser educational attainment and a higher probability of having children.

3

Empirical Model

The empirical model and estimation procedure fit three quantity variables – credit card accounts, debt and transactions – to their counterparts observed in the data. The model presents a way to tie these three variables together. The primary equation is the demand for active accounts. Heterogeneous consumers choose an active account from a menu of differentiated credit cards. Card holders use their accounts for day-to-day transactions and carry some amount of debt on them. The model predicts each issuers total accounts, debt and transactions by aggregating over individual account holders. The issuers’ price setting problem provides a channel for interchange fees to affect equilibrium interest rates. Interchange fees earned on transactions is one source of revenue for the issuers. As the interest rate changes, quantity demanded for credit card transactions adjusts and the 20 For households that have multiple cards, the interest rate reported in the SCF is the rate paid on the card with the largest balance. If balance is zero on all cards, it is the rate on the card obtained most recently. 21 A household is considered as subprime if either their annual income is below the poverty line or they have filed for bankruptcy previously.

12

revenue from interchange fee changes. Hence, an issuer’s profit maximizing interest rate is a function of the exogenously determined interchange fees. This enables me to back out the impact of changes in interchange fee on the equilibrium outcomes.

3.1

Consumer Behavior

Credit cards are differentiated products. They differ along several dimensions - APRs, reward features, annual fees, late fees etc. The financial institutions that issue them also differ substantially in their size, functions, services and scope. Hence, a model of credit card demand must accommodate product differentiation. In addition to this, credit card holders differ as well. A key distinction comes in the use of cards. Some people regularly borrow on cards (termed as revolvers) while others pay off their balances in full at each billing cycle (convenience users). Consumers also differ in their credit worthiness which is an indicator of how likely they are to pay off their credit card debt. All of these factors drive people’s choice of cards, transactions and debt. I follow the discrete choice literature and model demand for credit cards using a random coefficients discrete choice model. Heterogeneous consumers are assumed to pick a single product from an menu of horizontally differentiated credit cards. Their utility from a given card depends on its characteristics and their own attributes such as income and whether they are regular borrowers. The central modeling assumption here is that individuals pick a single active credit card account. It is well known that on an average people carry more than one credit card in their wallets. In the SCF data average person carries 2.4 credit cards. The assumption here is not that people choose one credit card but that they choose one active card, which is the card that they use most of the times. Rysman (2007) [32] finds that although people maintain unused credit cards from multiple networks, they concentrate their spending on a single network. Here I am extending this observation to issuers within the Visa and MasterCard networks. I assume that a card-holder chooses a single active credit card from the available Visa and MasterCard credit cards and accordingly, the data I use is that for number of active accounts per issuer. This brings us to the model. Let j = 0, 1, . . . J be the set of credit cards available in a particular year. Consumer i’s indirect utility from credit card j is a function of both consumer attributes and card characteristics. It is assumed to take the following form:22 Uij

= U (xj , rj , hi , ξj ; θ)

(1)

= λxj + βhi rj + ξj + ij

(2)

where xj and ξj represent observed & unobserved card characteristics, rj is the card’s APR, hi are consumer attributes (including expected credit card debt and an unobservable consumer character22

The data is a panel data set for 10 yrs. Each year represents a separate market. For simplicity, the year (market) subscript is dropped here.

13

istic, among others) and ij captures idiosyncratic taste. The parameters to be estimated are the λs and the βs. This specification is adapted from Berry, Levinsohn and Pakes (1995) (hereafter BLP). Note that it includes interaction terms between the APR and individual characteristics. The hi s represent factors that generate preference differences among heterogeneous consumers. They account for the fact that people’s sensitivity toward credit card interest rates depends on their attributes such as income and how much debt they carry on their cards and the β coefficients determine the impact of these preferences on utility. The paper also presents results from a simplified Logit model that does not include these interaction terms. An implication of the logit model is that substitution patterns are a function of product shares alone. Although the BLP model is less convenient to estimate than the Logit model, it relaxes these restrictions by killing the IIA problem and produces more realistic cross-price effects.23 Now, each individual chooses the card that maximizes her utility. Assuming there are no ties, the subset of h that choose card j is: Aj (θ) = {hi : Uij > Uik , ∀k} Let δj = λxj + ξj be referred to as the mean utility levels. It is assumed that ij follows the Type I extreme value distribution. This implies that we can analytically integrate over ij to get this expression for the probability that individual i chooses card j conditional on h: Pij (δ, r, h; θ) =

exp(δ + βhi rj ) P j 1 + k exp(δk + βhi rk )

(3)

We obtain the predicted share of issuer j in Active Accounts (sCCA) by aggregating this probability over the population, assuming that household characteristics h have a distribution g(h): Z sCCAj

=

Pij (δ, r, h; θ)g(h)dh

(4)

exp(δ + βhi rj ) P j g(h)dh 1 + k exp(δk + βhi rk )

(5)

h

Z = h

Let M be the size of the market. Then, the total number of active accounts for issuer j are: Z CCAj

= M×

Pij (δ, r, h; θ)g(h)dh

(6)

h

Next, we need a model for credit card debt and transactions for the issuers. Let ECCDij denote the amount of credit card debt individual i is expected to carry on card j conditional on obtaining this card. Then, issuer j can expect the amount ECCDij Pij in credit card debt from individual i, where 23 See BLP (1995) [4] for further detail. Ishii (2004) [20], Dick (2008) [11] and Zhou (2007) [40] also use this model for demand estimation in the context of the banking industry.

14

Pij is the probability that i gets card j. Hence, the model’s prediction for issuer j’s total credit card debt is obtained by aggregating each individual’s expected debt over the population: Z CCDj

= M×

ECCDij Pij (δ, r, h; θ)g(h)dh

(7)

h

For credit card transactions, assume that individual i spends a proportion αj of her income Yi as transactions on her active account, i.e. CCTi = αj Yi . As with debt, issuer j can expect the amount αj Yi Pij in credit card transactions from individual i and his total transactions are obtained by aggregating individual transactions over the population of interest: Z CCTj

= M×

αj Yi Pij (δ, r, h; θ)g(h)dh

(8)

h

This completes the description of the consumer’s behavioral model. The underlying hypothesis here is that people choose credit card accounts and then they choose how much debt and transactions to put on each account. The model generates predictions on an issuers expected share in accounts, debt and transactions by aggregating over individual choices.

3.2

Firm Behavior

Issuer profit provides the channel through which interchange fees affects equilibrium in this model. Issuers are assumed to maximize profit conditional on demand and given interchange fee. Interchange fees are determined exogenously and so are card characteristics other than interest rates. There are three sources of profit for issuers. Firstly, issuers earn interest income on credit card debt. Their costs on debt consist of cost of funds and the losses they incur in the form of unrecoverable loans or chargeoffs. Secondly, they earn interchange revenue along with any miscellaneous fees such as late fees, cash advance fees, over-limit penalty etc., on their credit card transactions. There is a marginal cost for each transaction which includes the cost of reward features, data processing and fraud related losses. Thirdly, they collect annual fees (if positive) on their accounts and there is a marginal cost for servicing accounts. The expression for issuer j’s variable profit is: Πj

= (rj − cf )CCDj + (˜ a − mctj )CCTj + (AFj − mcaj )CCAj − Chargeoffsj

(9)

where CCDj , CCTj and CCAj are credit card debt, transactions and accounts respectively, rj denotes the interest rate on balances, a ˜ the interchange fee, AFj is the annual fee, cf is the cost of funds, mctj is the marginal cost of a transaction (net of miscellaneous fees) and mcaj is the cost of servicing an account. The parameters to be estimated here are the two marginal costs and chargeoffs. To impose more structure on chargeoffs, it is assumed that each issuer loses a proportion γj of his total credit card debt holdings as chargeoffs, that is Chargeoffsj = γj CCDj . Consequently, issuer j’s 15

variable profit simplifies to: Πj

= (rj − cf − γj )CCDj + (˜ a − mctj )CCTj + (AFj − mcaj )CCAj

(10)

The issuer chooses APR rj to maximize profit. I assume that there exists an interior Nash equilibrium in interest rates. The first order condition for profit maximization is: ∂Πj ∂rj

= (rj − cf − γj )

∂CCDj ∂CCTj ∂CCAj + CCDj + (˜ a − mctj ) + (AFj − mcaj ) ∂rj ∂rj ∂rj

(11)

Interchange fees do not directly affect demand for credit cards. In response to an exogenous change in interchange fee, issuers recompute profits and choose new interest rates and subsequently, consumers adjust quantities demanded.

4

Estimation

Estimation follows a generalized method of moments approach. For each issuer, I observe three quantity variables in the data - active accounts, credit card debt and transactions. The model’s predicted accounts, debt, transactions and issuer’s first-order condition generate moment conditions that are matched to the data. This section outlines how these moment conditions are set up and estimated. The relevant market is the market for credit cards issued under the Visa or MasterCard logo and the population of interest is that of credit card holders. As mentioned in section 2, I use issuer-level data on quantities and card characteristics (from Nilson and TCCP) and household-level data (from the SCF). The BLP demand model provides a convenient platform for bringing together these two sources of data. Consumer utility is a function of both card characteristics and consumer characteristics: Uij

= U (xj , rj , hi , ξj ; θ) = λxj + βhi rj + ξj + ij = δj + β1 rj ECCDij + β2 rj Yi−1 + β3 rj IiBC + β4 rj ϑi + ij

The consumer characteristics hi used here are the amount of credit card debt that the person expects to hold at interest rate rj (denoted as ECCDij ), the inverse of household income Yi−1 , an indicator for low credit worthiness I BC (where BC stands for Bad Credit) and an unobservable consumer characteristic ϑi . Yi and I BC are observed directly in the SCF data. I BC is defined such that it takes the value one if the household either self reports credit problems or is classified as subprime (poor or bankrupt). The random coefficient ϑi is drawn from a standard normal distribution. This

16

is intended to capture unknown sources of consumer heterogeneity. For expected credit card debt, I estimate credit card debt from as a function of household characteristics and interest rate from the SCF data and use this estimated function to predict the expected credit card debt at each issuer’s rj . The SCF has detailed household data including several demographic, financial and credit worthiness related characteristics. Since a large number of households report zero credit card debt, I use the Type-I tobit censored regression model for the same. In this model the decisions to borrow and the amount to borrow are not estimated separately, which is appropriate here because the objective is to predict the expected amount of debt per household. All households are assumed to borrow, the expected value of debt is a lot more for some than others. One drawback with using the SCF data is that the interest rate reported is the rate on the card which has the largest balance whereas the debt is the total over all cards. To circumvent this issue, we could confine ourselves to the subsample of households that hold exactly one card. However, this approach would introduce a serious sample selection problem as pointed out by Zinman (2006) [41]. Note that I do not explicitly model credit limits here, but the estimated equation does include several variables that control for credit worthiness which largely determines credit limits. Now, the predicted credit card debt, other household attributes and card characteristics are plugged into equation (3) to obtain the probability that individual i chooses credit card j. The structural disturbance term ξj is intended to capture issuer-specific factors that may influence demand but are not observed directly. This term eliminates the over-fitting problem that arises when the only difference between the predicted and actual market shares is the sampling error. However, it leads to an endogeneity problem because if issuers and consumers both observe it, then rj and ξj are determined simultaneously and ξj cannot be assumed to be exogenous. Hence, we must use instrumental variables to fix this problem. Given a set of demand-side instruments zjD , this gives us our first moment condition - at the true parameter value θ0 , the disturbance term ξj is independent of variables zjD : E(ξ(θ0 )zjD ) = 0

(12)

I follow Berry (1994) [3] and BLP and use a contraction mapping to extract out the ξj s for setting up this moment condition.24 Next, the quantities of credit card accounts, debt and transactions per issuer in equations (4), (6), (7) and (8) need to be computed. The integral over households in these equations does not have a tractable solution. Following BLP, these are simulated by taking ns random draws from the SCF dataset and aggregating over these. For each of the 10 years I take a new set of 1000 draws from the closest SCF survey. The SCF has an oversampling of wealthy households but the survey does 24

See Berry et. al. (1995) [4] or Nevo (1998) [27] for detail

17

supply sampling weights to correct this problem.25 The simulated expression for market shares is: \j sCCA

=

1 X Pij (δ, r, h; θ) ns ns

(13)

Credit Card debt and transactions for each issuer are computed in a similar fashion. These are required for the interest rate setting equation. Given a set of demand parameters, total credit card debt for issuer j from equation (7) is obtained simply as a product of the predicted debt per individual and the probability of choosing card j summed across individuals. The estimated debt from SCF is household debt. I use a scaling factor α1 to convert this to expected individual debt. This leads to a second moment condition: \j CCD

=

M×

= CCDObsvd

νj1

=

M

−

1 P 1 \ ns α CCDij Pij (δ, r, h; θ) ns P 1 1 \ × ns ns α CCDij Pij (δ, r, h; θ)

CCDObsvd

E(νj1 (θ0 )zjD ) = 0

(14)

The only new parameter to be estimated from this equation is the scaling factor α1 . However, this equation does involve the demand parameters that enter into Pij (δ, r, h; θ) and adding it to the estimation improves their precision, as does the transactions equation. In the expression for transactions, the proportion of income spent as credit card transactions αj2 is unknown and needs to be estimated. I assume αj to be a linear function of selected card and bank characteristics. In the data, the transactions for each issuer are observed. The corresponding moment condition is set up equating the predicted transactions to those that are observed: \j CCT

=

M×

= CCT Obsvd

νj2

=

M

−

1 P 2 ns αj Yi Pij (δ, r, h; θ) ns 1 P 2 × ns ns αj Yi Pij (δ, r, h; θ)

E(νj2 (θ0 )zjD ) = 0

CCT Obsvd (15)

Further, we need an estimation equation for the cost parameters. Equation (11) represents the issuer’s price setting condition. The proportion of debt lost as chargeoffs γj , marginal cost of a transaction mctj and that of servicing an account mcaj are the parameters to be estimated here. For simplicity, I assume that mcaj is constant across all issuers and that mctj is a function of reward features plus a constant. For cost of funds cf , I use the annual average of the prime rate. It is also assumed that the chargeoffs proportion γj is constant across firms, except for an firm level 25

The SCF uses a multiple imputation technique. Here I have used the first impute from each of the three SCF rounds.

18

disturbance term: mctj γj

= τ1 RW + τ2 = κ + ωj

where κ and τ2 are constants, ωj is an unobservable source of differences in cost of chargeoffs among issuers and RW is summary measure of reward features on the card as observed in the TCCP data. The resulting interest rate setting condition is: ∂Πj ∂rj

∂CCDj + CCDj ∂rj ∂CCTj ∂CCAj +(˜ a − τ1 RW − τ2 ) + (AFj − mca) ∂rj ∂rj

= (rj − cf − κ − ωj )

(16)

This expression is inverted to recover ωj . The identifying assumption here is that ωj is uncorrelated with suitable instruments zjS and the corresponding moment condition is: E(ω(θ0 )zjS ) = 0

(17)

Finally, the four moment conditions in equations (12), (14), (16) and (17) are estimated jointly. They are stacked together as follows: 

ξj (θ)zjD



J  1 D 1 X  νj (θ)zj G(θ) = E[mj (θ)] =  2 D J j=1  νj (θ)zj ωj (θ)zjS

    

It is assumed that at the true parameter value G(θ) = 0. This brings us to a discussion of the instruments. As demand-side instruments, I use card characteristics other than APR and bank characteristics. Following BLP, sums of these variables over competing issuers in the same market are also valid instruments. In addition, cost shifters provide a good source of demand instruments. I obtain two of these from the Call Reports filed by the banks - expenses on premises & equipment and wages, both normalized by assets.26 Since these are exogenous cost shifters, they are also valid cost instruments and so are the exogenous card and bank characteristics. Estimation relies on the validity of these instruments and exogenous variation across issuers and consumers (both within and across markets) for identification. The appendix summarizes the steps involved in estimation and outlines how standard errors are computed. As 26

Other costs reported in the Call Reports, such as provisions for loans and leases, are not included because these may not be exogenous with respect to consumer demand.

19

a reference point, I also estimate demand for active accounts using a restricted form of the utility function wherein consumer heterogeneity enters preferences only through the idiosyncratic term ij . This leads to the logit demand model of McFadden (1974) [26].

5

Results

The estimation procedure involves several equations and a few sets of parameters. In this section, I discuss the results and their implications in order, starting with the demand side. Credit Card Debt Equation (SCF) The first step in estimation is computing the demand for credit card debt from the SCF data as a function of interest rate and household characteristics. This is used to predict the amount of debt an individual i is expected to carry on card j at the interest rate rate offered by the issuer, rj . The predicted debt along with a few more household attributes is then plugged into the utility function for further estimation. I fit a censored regression (Type I Tobit) model with credit card balances as the dependent variable and interest rate and household observables as explanatory variables. From the SCF data, households that reported negative incomes or annual incomes over one million were dropped. A small percentage of households report credit card debt exceeding their credit limits were also omitted. The resulting sample has 8,397 observations. Table (6) shows the results from this step. Credit attitude, credit worthiness, demographic and financial factors all turn out to be important in explaining household balances. Firstly, the interest rate coefficient is significant. It implies a mean elasticity of -0.39 when computed at the interest rates offered by the issuers. The other coefficients reflect the trends observed in the average statistics and it is interesting to note their relative importance. Debt holding increases with age, marriage and children and is lesser at both the lowest and the highest levels of education. Households that self report credit problems and an inclination toward borrowing are significantly more likely to carry higher credit card debt. Households that save and those with higher income and assets carry less debt on their cards. Demand for Accounts, Debt and Transactions The empirical model and estimation procedure fits three quantity variables to their counterparts observed in the data. The primary equation is the demand for active accounts. Credit Card holders are expected to use their active accounts by charging some amount of debt on their cards and for conducting day to day transactions. I match the model’s predictions regarding shares in active accounts, total outstanding debt and total transactions with those observed in the data. The demand side estimation follows the BLP framework. For comparison, the first equation (demand for active accounts) was also fit using the logit model which is a simplified version of the full BLP model. The dependent variable for the logit model is ln(sj ) − ln(s0 ) and the explanatory 20

variables are card and issuer characteristics. Table (7) presents the results from two different specifications of this equation. The first was estimated without instrumenting for interest rate (OLS version). The second includes exogenous card and issuer characteristics as instruments (IV Logit version). After instrumenting for APR, its coefficient increases about ten-fold from -0.017 to -0.147. This provides strong evidence for the endogeneity of the price variable. Correlation between the unobservable and APR would tend to bias the coefficient toward zero. The instruments correct this bias. The logit model also indicates that demand for active accounts is elastic. For instance in 2005, the predicted mean elasticity with respect to APR for the logit model was -1.92. Among the other explanatory variables, the dummy variables that indicate the type of bank seem to be more important (both in terms of their absolute levels and their significance) than the other card characteristics such as type of APR and grace period etc. This trend is preserved in the full BLP model. The R-squared for the IV logit specification is 0.68 which indicates that the majority of the variance in mean utility is explained by these independent variables. The full model is an extension of the simple logit model in that it allows for consumer heterogeneity to be built into the demand analysis. I include interactions between consumer attributes and APR to allow for people’s sensitivity to APR to depend on factors such as their income, their credit worthiness and the amount of debt they are expected to carry on their cards. Table (8) shows the results from this specification. The linear APR coefficient is -0.16, higher than that under the logit \ij and APR is significant model. Among the interaction variables, that between predicted debt CCD and positive. This indicates that people with higher than average credit card debt care less about interest rates than others. This may seem surprising at first but it has an explanation. Credit card interest rates are usually higher than rates on consumer loans and installment loans available from other sources. People who carry large amounts of credit card debt possibly do so for the convenience rather than being motivated by interest rates. The other interaction terms in the utility function were not estimated with sufficient precision. Among the linear variables, just as in the logit model, the coefficients of the issuer classification variables are large in magnitude and significant when compared with card characteristics other than APR. The coefficient for savings bank is negative implying that people prefer credit cards from national banks or state member banks over these. (Non-member banks is the omitted category here.) Visa credit cards generate negative marginal utility. Although Visa is the bigger of the two networks, the share of MasterCard during this period was rising and some of the large issuers such as Citibank were aggressively promoting MasterCard credit cards over Visa. Annual fees, rewards and grace period all have the expected signs though their magnitudes are low. The second and third equations match the predicted credit card debt and transactions for the issuers to those observed in the data. The new parameters estimated here are first, the scaling factor that is intended to transform household-level credit card debt to individual-level debt (α1 in the

21

model) and second, the proportion of income spent as credit card transactions (αj2 ). The scaling factor is estimated at 61.4%, implying that on average there are less than two active account holders per household. The mean proportion of income spent using credit cards as the method of payment was estimated at 5.4%. This proportion is estimated as a function of card and issuer charateristics. It reduces with APR and is lower for monoline banks and subprime lenders. It is higher for national banks and big issuers and rises with reward features. Several of the estimated coefficients in this equation are significant at the 1% level. Tables (9), (10) and (11) present a sample from year 2004 of the own and cross semi-elasticities for all three of these quantity variables with respect to interest rates. Table (12) shows the mean and weighted mean of the own price elasticities across all issuers, with initial shares as the weights. It is interesting to note that demand for all three variables is on average elastic. The mean own price elasticities indicate that a 1% increase in an issuer’s APR would lead to a 2.05% decline in accounts, 2.06% decline in transactions and 1.87% fall in debt on average. The weighted means are somewhat lower. As can be seen from tables (9)-(11), the larger issuers tend to have lower elasticities. For instance, in table (9) First National, Cross-Country bank, Direct Merchant’s and Merrick Bank all have elasticities exceeding 2% in absolute value, whereas Chase and Bank of America, which are among the top ten issuers, have inelastic demand. Costs, Revenue and Profits Turning now to the supply side, I estimate issuer chargeoffs, the marginal cost of transactions and the cost of servicing an account from the issuer’s price setting condition. These estimates are presented in table (13). The estimated chargeoffs are 5.13% of credit card debt outstandings. This figure is well within the range of chargeoffs reported by the banks in their call reports. The Federal Reserve’s report on chargeoff and delinquency rates27 for the 100 largest banks shows that the chargeoff rate on consumer credit card loans during this time period varied between 3.97% and 8.08% with a quarterly average of 4.93%. The mean marginal cost of a transaction (net of miscellaneous fees) is estimated to be about 0.57 cents. This cost increases with reward features, but the coefficient on rewards is rather small. The estimated cost of servicing an account is $4.24, but this parameter has a high standard error of $15.8. These parameter estimates allow predictions on the composition of issuer costs, revenues and profits. Figure (3) depicts this composition. Debt heavily outweighs transactions and accounts as the source of both costs and revenue. Cost of funds and chargeoffs collectively account for 90.1% of the issuer costs, whereas transaction costs and account servicing are just 8.3% and 1.6% respectively. Evans and Schmalensee (2005) [12] also report that the largest chunk of costs and revenue are attributable to credit card debt. Annual fees form 2.6% of profits and account servicing is 1.6% of 27

Source:

http://www.federalreserve.gov/releases/chargeoff/chgtop100sa.htm

22

72% Interest on Debt

2.6% Membership 25.4%

Cost of Funds

Interchange Fees & Misc. Fees

50.7% 1.6% Account Servicing 8.3%

Chargeoffs

Interest Income

39.4%

Transaction Costs Net of Misc. Fees

83.6% 2% Annual Fees 14.4% Interchange Revenue

Figure 3: Composition of Estimated Profits, Costs and Revenue costs. Income earned as interest on debt is 83.6% of all revenue. Interchange fee contributes 14.4% of the total revenue. A quarter of the profits come from these fees, but the majority (72%) are net earnings on debt. The average credit card profitability on debt was estimated to be 3.09%. This is in line with the return on assets reported by the Federal Reserve for a sample of large credit card banks, which was ranging between 2.14% and 3.66% for the years 1996-2005.28 Although this rate of return has fallen as compared to the late 1980s and the early 1990s, credit cards still remain more profitable than most other commercial bank activities.

28

Source: http://www.federalreserve.gov/boarddocs/rptcongress/creditcard/2006/default.htm

23

6

Counterfactual Experiment

Given the parameters estimated from the structural model, a policy experiment to analyse the effect of changes in interchange fees is relatively straight-forward. Interchange fees are exogenously set to a new level and a new equilibrium in interest rates is computed numerically, starting from the observed interest rates. The issuer behavior follows a Nash pricing assumption, so that the new vector of interest rates is such that no firm can gain profits by deviating.29 I consider three new levels of interchange - 1%, 0.75% and 0.5%. The average interchange fee in the data used in this paper was 1.5% and its mean contribution to issuer revenue was about 14.4%. In Australia, the fee was regulated in 2003 and brought down to 0.5% which is about half of its pre-regulation level. In most countries, possible regulation of the fee is unlikely to reduce it to levels lower than 0.5%. For each new level of interchange fee considered, Table (14) presents percentage changes in the key quantity variables and profit under the new equilibrium. The results indicate that regulating the interchange fee down by a third increases interest rates on an average by less than 2.5% of their original levels, which implies that the mean APR would increase from 15.4% to 15.8%. This causes credit card debt to fall by 2.9% and transactions to decline by 2.7%. Even when interchange fee is reduced by 50%, the impact on the quantity variables is close only to 5%. It would take a decline in the fee of the order of a dollar for every hundred for the impact on the equilibrium quantities to be over 7%. This impact is not significant especially when considering that annual fees, late fees etc. are exogenous to this model. Demand is expected to be relatively inelastic with respect to a few of these fees and issuers might change them in response to changes in interchange fee to partially recover their losses. This would imply that the figures presented here over-estimate the increase interest rates. Thus, it is possible that the actual impact might be even less. Demand elasticity also explains why issuers are not able to increase interest rates substantially. The elasticities with respect to APR reported in Tables (9)-(11) are high for most banks and particularly so for smaller issuers. Some of the large issuers face inelastic demand at the original APRs, but their ability to raise interest rates is also curtailed by the possibility of losing market share to the smaller banks. The increase in interest rates is not completely able to compensate for the decline in interchange fee revenue and issuer profits suffer in the process. The average decline in profits is 6.3% when interchange fee is reduced to one percent. The loss in profits is higher for larger issuers30 and so is the change in their debt, transactions and accounts. In all three cases considered, large issuers raise interest rates more than smaller ones do. Due to this difference in response of large and small issuers, market concentration reduces following a drop in the interchange fee. The C(4) ratio drops in almost all years. The share of the largest issuer falls on average by 8.5% of its original 29 30

I restrict the search to APRs between 4% and 35% but neither of these limits turn out to be binding. Those with market share greater than 5%

24

level. These changes indicate that even though aggregate quanitites may not change drastically in response to regulation of interchange fee, the market does undergo some reorganization.

7

Conclusion

This paper outlines an empirical model and estimation strategy for the evaluation of the role of interchange fee in credit card issuing. The motivation for this research stems from the recent attention that the credit card industry has received from economic theorists and regulators. One of the central themes explored in the literature is interchange fees – the determination of optimal interchange fee, the competitive effect of its collective setting and the justification, if any, for intervention by regulatory authorities. The theoretical underpinnings of interchange fees have been investigated pretty thoroughly, but there is a lack of consensus in the theory and it raises several empirically testable questions. The question explored in this paper is: given the nature of competition in the market, to what extent are issuers likely to pass on decreases in interchange fees to consumers via higher interest rates? I propose a model that builds together consumers and credit cards issuers in one framework and estimates optimal policies by combining data from several sources. Counterfactual experiments allow me to predict the impact of changes in interchange fees on consumer usage and issuer profit, assuming that merchant behavior is invariant. I find that competition in the market is intense and demand is elastic with respect to interest rates. This prevents issuers from raising interest rates substantially in response to a fall in interchange fees and since interchange is an important part of their revenue, issuer profits suffer. Although aggregate quantities of transactions and debt do not seem to fall significantly, there is some amount of restructuring involved as market shares readjust and larger issuers suffer higher losses. A comprehensive evaluation of this subject requires accounting for changes in other card characteristics (such as reward features) and allowing for acquirers and merchants to respond as well. That, however, remains open for further research.

25

References [1] L. M. Ausubel. The Failure of Competition in the Credit Card Market. The American Economic Review, 81(1):50–81, 1991. [2] W. F. Baxter. Bank Interchange of Transactional Paper: Legal and Economic Perspectives. Journal of Law and Economics, 26(3):541–588, 1983. [3] S. Berry. Estimating Discrete-Choice Models of Product Differentiation. The RAND Journal of Economics, 25(2):242–262, 1994. [4] S. Berry, J. Levinsohn, and A. Pakes. Automobile Prices in Market Equilibrium. Econometrica, 63(4):841–890, 1995. [5] B. K. Bucks, A. B. Kennickell, and K. B. Moore. Recent Changes in US Family Finances: Evidence from the 2001 and 2004 Survey of Consumer Finances. Federal Reserve Bulletin, 92:A1–A38, Feb 2006. [6] P. S. Calem and L. J. Mester. Consumer Behavior and the Stickiness of Credit-Card Interest Rates. The American Economic Review, 85(5):1327–1336, 1995. [7] E. Castronova and P. Hagstrom. The Demand for Credit Cards: Evidence from the Survey of Consumer Finances. Economic Inquiry, 42(2):304–318, April 2004. [8] S. Chakravorti. Theory of Credit Card Networks: A Survey of the Literature. Review of Network Economics, 2(2):50–68, 2003. [9] S. Chakravorti and R. Roson. Platform Competition in Two-sided Markets the Case of Payment Networks. Technical report, 2004. [10] H. Chang, D. Evans, and D. Garcia-Swartz. The Effect of Regulatory Intervention in TwoSided Markets: An Assessment of Interchange-Fee Capping in Australia. Review of Network Economics, 4:328–358, 2005. [11] A. Dick. Demand Estimation and Consumer Welfare in the Banking Industry. Journal of Banking and Finance, 32(8):1661–1676, 2008. [12] D. S. Evans and R. Schmalensee. Paying With Plastic: The Digital Revolution In Buying And Borrowing. MIT Press, 2005. [13] J. Gans. Evaluating the Impact of the Payment System Reform. Technical report, Submission to the RBA Review of Payment System Reforms., 2007.

26

[14] J. S. Gans and S. P. King. A Theoretical Analysis of Credit Card Regulation. Economic Record, 79(247):462–472, 2003. [15] D. Garcia-Swartz, R. Hahn, and A. Layne-Farrar. The Economics of a Cashless Society: An Analysis of the Costs and Benefits of Payment Instruments. AEI-Brookings Joint Center Working Paper (04-20), 2004. [16] D. B. Gross and N. S. Souleles. An Empirical Analysis of Personal Bankruptcy and Delinquency. Review of Financial Studies, 15(1):319–347, 2002. [17] G. Guthrie and J. Wright. Competing Payment Schemes. Journal of Industrial Economics, 55(1):37–67, 2007. [18] F. Hayashi and E. Klee. Technology Adoption and Consumer Payments: Evidence from Survey Data. Review of Network Economics, 2(2):175–190, 2003. [19] F. Hayashi and S. E. Wiener. Interchange fees in australia, the uk, and the united states: Matching theory and practice. Economic Review, (Q III):75–112, 2006. [20] J. Ishii. Interconnection Pricing and Compatibility in Network Industries: ATM Networks in the Banking Industry. Harvard University Working Paper, 2004. [21] M. L. Katz. Reform of Credit Card Schemes in Australia II. Sydney, Australia: Reserve Bank of Australia, 2001. [22] M. L. Katz. What Do We Know About Interchange Fees and What Does it Mean for Public Policy? Proceedings–Payments System Research Conferences, Federal Reserve Bank of Kansas City, 2005. [23] H. Kim and S. A. DeVaney. The Determinants of Outstanding Balances among Credit Card Revolvers. Financial Counseling and Planning, 12(1):67–77, 2001. [24] E. Klee. Families’ Use of Payment Instruments During a Decade of Change in the U.S. Payment System. Finance and Economics Discussion Series, Board of Governors of the Federal Reserve System, (1), 2006. [25] C. R. Knittel and V. Stango. Price Ceilings as Focal Points for Tacit Collusion: Evidence from Credit Cards. American Economic Review, 93(5):1703–1729, 2003. [26] D. McFadden. Conditional Logit Analysis of Qualitative Choice Behavior. P. Zarembka, Ed. Frontiers in Econometrics, 1974.

27

[27] A. Nevo. A Research Assistant’s Guide to Random Coefficients Discrete Choice Models of Demand. February 1998. [28] Reserve Bank of Australia. Reform of Credit Card Schemes in Australia IV: Final Reforms and Regulation Impact Statement. Sydney, Australia: Reserve Bank of Australia, 2002. [29] S. Park. Effects of Price Competition in the Credit Card Industry. Economics Letters, 57(1):79– 85, 1997. [30] J. C. Rochet. The Theory of Interchange Fees: A Synthesis of Recent Contributions. Review of Network Economics, 2(2):97–124, 2003. [31] J. C. Rochet and J. Tirole. Cooperation among Competitors: Some Economics of Payment Card Associations. The RAND Journal of Economics, 33(4):549–570, 2002. [32] M. Rysman. An Empirical Analysis of Payment Card Usage. Journal of Industrial Economics, 55(1):1–36, 2007. [33] R. Schmalensee. Payment Systems and Interchange Fees. Journal of Industrial Economics, 50(2):103–122, 2002. [34] R. Schmalensee. Interchange Fees: A Review of the Literature. The Payment Card Economics Review, 1:25–44, Winter 2003. [35] M. Schwartz and D. Vincent. The No Surcharge Rule and Card User Rebates: Vertical Control by a Payment Network. Review of Network Economics, 5(1):72–102, 2006. [36] V. Stango. Pricing with Consumer Switching Costs: Evidence from the Credit Card Market. Journal of Industrial Economics, 50(4):475–492, 2002. [37] V. Stango. Strategic Responses to Regulatory Threat in the Credit Card Market. The Journal of Law and Economics, 46(2):427–452, 2003. [38] J. Stavins. Effect of Consumer Characteristics on the Use of Payment Instruments. New England Economic Review, 3(4-5):19–31, 2001. [39] S. E. Weiner and J. Wright. Interchange fees in various countries: Developments and determinants. Review of Network Economics, 4(4):290–323, Dec 2005. [40] X. Zhou. Estimation of the Impact of Mergers in the Banking Industry. Yale University Working Paper, 2007. [41] J. Zinman. Debit or Credit? Forthcoming, Journal of Banking and Finance, 2006.

28

Appendix A. Summary of Estimation Procedure The empirical model involves four equations that are estimated together. The objective function for estimation is set up as follows: • Step 1: Estimate Credit Card Debt as a function of interest rate r from the SCF Data using a censored regression model. The household’s desired stock of debt is: CCD∗ (r) = max {0, ρHi + ρ˜r˜ + η} where Hi are household demographic, financial and credit characteristics. Predict CCD∗ (r) at \ij . interest rate rj for each issuer to obtain CCD • Step 2: Take simulation draws from the SCF dataset. Plug in the estimated debt along with other individual characteristics into Uij and predict the probability that individual i chooses card j as: Pij (δ, r, h; θ) = =

exp(δ + βhi rj ) P j k exp(δk + βhi rk ) \ij + β˜h˜i rj ) exp(δj + β1 rj CCD P \ik + β˜h˜i rk ) exp(δk + β1 rk CCD 1+ 1+

k

• Step 3: Sum over the simulation draws to obtain the predicted active accounts, credit card debt and transactions for issuer j: CCAj CCDj CCTj

1 X Pij (δ, r, h; θ) ns ns 1 X 1 \ = M× α CCDij Pij (δ, r, h; θ) ns ns 1 X 2 α Yi Pij (δ, r, h; θ) = M× ns ns j = M×

and compute expressions for their first derivatives:

∂CCAj ∂CCDj ∂rj , ∂rj

and

∂CCTj ∂rj .

• Step 4: Recover the two disturbance terms - ξj using the contraction mapping and ωj from the first order condition. Compute the difference between the predicted and the observed debt and transactions, νj1 and νj2 respectively.

29

• Step 5: Set up the sample analogue of the moment condition: 

ξj (θ)zjD



J J  1 D 1X 1 X  νj (θ)zj GJ (θ) = mj (θ) =  2 D J J j=1 j=1  νj (θ)zj ωj (θ)zjS

    

• Step 6: Minimize kGN (θ)kA , where A is a weight matrix, to obtain the parameter estimates. √ The estimated parameters are N −consistent and asymptotically normal (under certain conditions) with the following variance-covariance matrix is given by the GMM formula: (Γ0 AΓ)−1 Γ0 AV AΓ(Γ0 AΓ)−1 where Γ = ∂G(θ)/∂θ is the gradient of the objective function and V = E[m(θ0 )m(θ0 )0 ] plus two other terms that adjust variance in the moment condition due to simulation error and prediction error. See BLP (1995) for detail. These standard errors allow for arbitrary heteroskedasticity.

30

Table 1: Market Dataset Summary Year

N

Share Mean

Share S.Dev.

Share Sum

C(1)

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

49 45 35 28 27 25 27 25 22 17

1.39 1.60 2.37 2.89 3.07 3.39 3.15 3.35 3.71 4.37

2.29 2.66 4.08 4.29 4.45 4.25 4.54 5.16 5.98 6.88

68.09 72.13 82.98 81.02 82.85 84.67 85.16 83.78 81.64 74.25

11.69 10.82 16.59 16.58 18.28 14.68 16.78 20.86 20.35 19.17

31

Table 2: Data Definitions for Quantities and Characteristics Dataset Variable

Source

Definition

APR Annual Fee Rewards

TCCP TCCP TCCP

Visa

TCCP/Nilson

Fixed APR Variable APR (Prime)

TCCP TCCP

Grace Period

TCCP

National Bank State Member Bank Savings Bank Monoline CC Bank

Fed NIC Fed NIC Fed NIC Call Reports

Subprime Lender Big Issuer

Nilson Nilson

Annual Percentage Rate Membership fee paid annually The issuer reports 1 or 0 for the following enhancements offered: Rebates on purchases, extension of warranty, purchase protection, travel accident insurance, travel discounts, automobile rental insurance, discounts on purchases other than travel discounts, card registration and other. This summary measure is the sum of all reported enhancements. Max value = 10 Equals 1 if Network name mentioned under TCCP credit card plan name is Visa or if no name was mentioned, if Number of Visa cards > Number of MasterCard Cards in Nilson If the credit card plan APR is fixed If the credit card plan APR is variable, tied to the prime rate. Variable APRs tied to some other rate such as Tbill rates are the omitted category. The period of time from the end of the billing cycle in which credit extended for purchases during that billing cycle may be repaid without incurring a finance charge Federal Reserve’s Bank Classification Federal Reserve’s Bank Classification Federal Reserve’s Bank Classification If 75% or more of the bank’s reported total assets are held in credit card balances If the issuer offers credit cards to risky individuals If the issuer was one of the top 25 issuers reported in Nilson during the previous year

Note:

Under TCCP, issuers report the plan that had the largest number of cards outstanding AND that was available to new customers as of the report date

32

Table 3: Summary Statistics: Quantity and Characteristics Variable

Mean

Std.Dev.

Quantities Outstandings ($ Bil.) Transaction Volume ($ Bil.) Active Accounts (Mil.)

13.8504 25.2087 5.6304

25.3862 49.8536 9.5368

Card Features APR (%) Annual Fee ($) Zero Annual Fee Rewards Visa Fixed APR Variable APR (Prime) Grace Period

15.4270 11.4839 0.6400 1.6967 0.6300 0.1900 0.7333 25.7833

3.9316 18.6335 0.4808 1.9709 0.4836 0.3930 0.4430 5.0006

Bank Type National Bank State Member Bank Savings Bank Monoline CC Bank Subprime Lender Big Issuer

0.6100 0.1233 0.0633 0.3367 0.2600 0.5533

0.4886 0.3294 0.2440 0.4734 0.4394 0.4980

Interchange Fee (%)

1.4591

0.1757

33

Table 4: Data Definitions for SCF Dataset Variable

Definition

Married, Age, etc. Home: Regular OK to borrow for vacation

Refer to the Head of the Household.a Home type is other than farm, ranch, mobile home or RV The respondent feels that it is all right for someone like herself to borrow money for a vacation The respondent feels that it is all right for someone like herself to borrow money for a fur coat or jewelry Indicator of whether the household saved over the past 12 months Household seeks advice from an Accountant, Banker, Broker, Lawyer or Financial Planner when making decisions about credit or borrowing Household income for previous calendar year Total value of financial assets held by household Total value of nonfinancial assets held by household Total value of debt held by the household excluding installment loans and credit card balances Last yr income higher than what they would expect in a normal yr Last yr spending higher than what they would expect in a normal year The number of financial institutions where the household has accounts or loans, or does regular personal financial business with Whether the household owns their home Household has no checking account, ownership includes checking accounts with a zero balance. In the last 5 yrs, have the respondent or partner been turned down for credit or not given as much credit (even after reapplying) or not applied for credit due to the fear of being rejected Of all loan and mortgage payments, the household has been behind in any by over two months Income lower than the Federal poverty line The household has filed for bankruptcy previously Poor or Bankrupt, as above Number of general purpose credit cards: Visa, MasterCard, Discover or Optima Total credit limit on all general purpose credit cards The balance still owed on these accounts after the last payments Amount of Debt/Credit Limit The interest rate paid on the card with the largest balance. If balance is zero on all cards, the rate on the card obtained most recently

OK to borrow for luxury Savings Dummy Have Professional Advice

Income Financial Assets Non Financial Assets Other Debt Last yr Income high Last yr Spending high Number of Institutions Home Ownership Dummy No Checking Account Credit Problems

Late Payments Poor Bankrupt Subprime Number of Cards Credit Limit Amount of Debt Utilization Rate Interest Rate

a

The respondent may not be the head of the household. This variable has been suitably adjusted using the ’switch’ variable x8000.

34

Table 5: Descriptive Statistics for Credit Card Holders from the SCF Dataset

Variable

All Card Holders Meana Std.Dev.

Convenience Users Mean Std.Dev.

Revolvers Mean Std.Dev.

Demographic Married Single Female Age (yrs.) White Any Children Dummy Education: No High School Education: College Home: Regular

0.6564 0.2171 48.74 0.8239 0.4317 0.0784 0.3903 0.9375

0.4749 0.4123 15.98 0.3809 0.4954 0.2688 0.4878 0.2420

0.6711 0.1927 53.67 0.8933 0.3471 0.0761 0.4726 0.9422

0.4699 0.3945 16.75 0.3088 0.4761 0.2651 0.4993 0.2333

0.6443 0.2372 44.66 0.7666 0.5016 0.0803 0.3222 0.9337

0.4788 0.4254 14.06 0.4231 0.5001 0.2719 0.4674 0.2489

Credit Attitude OK to borrow for vacation OK to borrow for luxury Savings Dummy Have Professional Advice

0.1548 0.0673 0.6422 0.4630

0.3617 0.2506 0.4794 0.4987

0.1085 0.0536 0.7480 0.5016

0.3110 0.2253 0.4342 0.5001

0.1931 0.0787 0.5547 0.4311

0.3948 0.2693 0.4971 0.4953

Financial Indicators Incomeb Financial Assets Non Financial Assets Other Debt Last yr Income high Last yr Spending high

76.72 227.88 340.83 74.14 0.1061 0.1667

165.07 1258.33 1784.04 140.35 0.3079 0.3727

96.82 398.35 516.82 76.33 0.1087 0.0991

235.65 1807.05 2551.95 178.27 0.3114 0.2989

60.11 86.87 195.37 72.33 0.1038 0.2225

57.18 387.00 620.52 98.56 0.3051 0.4160

Credit Worthiness Number of Institutions Home Ownership Dummy No Checking Account Credit Problems Late Payments Subprime: Poor or Bankrupt

2.7317 0.7752 0.0290 0.1202 0.0915 0.1261

1.6875 0.4175 0.1679 0.3252 0.2883 0.3320

2.9023 0.8530 0.0130 0.0449 0.0363 0.0780

1.8084 0.3541 0.1133 0.2071 0.1870 0.2683

2.5907 0.7109 0.0423 0.1824 0.1371 0.1658

1.5670 0.4534 0.2013 0.3862 0.3440 0.3720

Credit Use Number of Cards Credit Card Limit Amount of Debt Utilization Rate Interest Rate

2.3837 20.16 2.40 0.2101 13.38

1.6615 47.77 6.00 0.6037 5.69

2.1844 22.78 0 0 13.35

1.5315 63.77 0 0 5.37

2.5485 18.00 4.38 0.3838 13.40

1.7448 28.20 7.56 0.7740 5.95

N a b

10198

5987

These are weighted means. All monetary values in are in ’000s and deflated to 2001 dollar.

35

4211

Table 6: Demand for Credit Card Debt: Censored Regression Model Variable

Coeff.

Std.Err.

Demographic Married Single Female Age Age Squared White Any Kids Dummy Education: No High School Education: College Home: Regular

1.2998** 0.7307 0.4673** -0.0056** -0.7842* 0.9056** -1.1403** -1.5153** 0.0735

0.4170 0.4606 0.0584 0.0006 0.3163 0.3140 0.4256 0.3443 0.4801

Credit Attitude OK to borrow for vacation OK to borrow for luxury Savings Dummy Have Professional Advice

2.7792** 1.7121** -2.4464** -0.4511

0.4288 0.6016 0.3582 0.2909

Credit Worthiness Number of Institutions Home Ownership Dummy No Checking Account Credit Problems Subprime: Poor or Bankrupt

0.0834 -1.2163** -0.6814 3.9645** 1.3114**

0.0878 0.3615 0.5583 0.5018 0.3357

Financial Last yr Income high Last yr Spending high Income Financial Assets Non Financial Assets Other Debt

-0.1605 2.6111** -0.0132** -0.0005 -0.0001 0.0048**

0.4556 0.4254 0.0023 0.0003 0.0001 0.0012

Interest Rate 1998 SCF 2004 SCF Constant Sigma

-0.1371** 0.0190 0.6754* -7.5806** 10.3182

0.0304 0.3167 0.3176 1.5819 0.5366

N

8397

The dependent variable is household credit card balances in ’000s. Significance at 1% and 5% levels is indicated by ** and * respectively

36

Table 7: Demand for Credit Card Accounts: Logit Model OLS Coeff. Std.Err.

Variable

IV LOGIT Coeff. Std.Err.

Card Features APR Annual Fee Rewards Visa Fixed APR Variable APR (Prime) Grace Period

-0.0167 -0.0085* 0.0075 -0.2094 0.3999 0.1988 -0.0181

0.0185 0.0037 0.0321 0.1305 0.2511 0.2212 0.0139

-0.1466** -0.0099* 0.0068 -0.2256 0.4159 0.1985 0.0095

0.0204 0.0040 0.0353 0.1436 0.2763 0.2434 0.0153

Bank Type National Bank State Member Bank Savings Bank Monoline CC Bank Subprime Lender Big Issuer

0.7762** 0.7920** -0.3459 0.3270* 0.4072* 2.0403**

0.1615 0.2106 0.2729 0.1446 0.1643 0.1300

0.6367** 0.6323** -0.7170* 0.5135** 0.5280** 1.9331**

0.1777 0.2317 0.3002 0.1592 0.1808 0.1430

Constant

-4.2927**

0.5263

-2.8611**

0.5791

R-squared N Time Trend Included

0.6766 300 Yes

The dependent variable is ln(sj ) − ln(s0 ). The IV Logit specification uses instruments for the price variable APR Significance at 1% and 5% levels is indicated by ** and * respectively

37

0.6811 300 Yes

Table 8: Demand Parameter Estimates : Full Model Variable

Coeff.

Demand for Active Accounts: Linear Coefficients APR Annual Fee Rewards Visa Fixed APR Variable APR (Prime) Grace Period National Bank State Member Bank Savings Bank Monoline CC Bank Subprime Lender Big Issuer Constant Time Trend Included Interactions with APR N (0, 1) ECCD Bad Credit Income Inverse CCD Scaling Factor

Std.Err.

-0.16396** -0.00001 0.00022** -0.29837* 0.47155 0.40551 0.01211 0.70866** 0.64716** -0.61458* 0.45920** 0.46055** 2.04231** -3.43897** Yes

0.01807 0.00019 0.00007 0.13352 0.26488 0.23692 0.01410 0.17209 0.22525 0.29165 0.15153 0.16787 0.13347 0.52130

0.00392 0.01428** -0.05002 -0.00004

0.03355 0.00484 0.04185 0.00006

0.61483**

0.01328

Proportion of Income spent as CCT Coefficients APR -0.00016** Rewards 0.00089 National Bank 0.00970* State Member Bank 0.01931** Savings Bank 0.00224 Monoline CC Bank -0.01892** Subprime Lender -0.01706** Big Issuer 0.01484** Constant 0.04956** Mean(%) 5.443 N

0.00004 0.00081 0.00399 0.00526 0.00692 0.00353 0.00383 0.00320 0.00512

300

Significance at 1% and 5% levels is indicated by ** and * respectively

38

39 0.1280 0.1283 0.1103 0.0889 0.1150

MBNA

Capital One

Direct Merchant’s Bank

Cross Country Bank

Merrick Bank

0.0508

0.0393

0.0487

0.0566

0.0565

0.0635

0.0458

0.0646

0.0621

-1.2444

0.0608

Nat’l City

0.1007

0.0779

0.0967

0.1124

0.1121

0.1260

0.0910

0.1282

-1.0524

0.1207

0.1207

Citi

0.0603

0.0466

0.0578

0.0672

0.0671

0.0754

0.0544

-0.8227

0.0738

0.0722

0.0722

JPM Chase

0.0046

0.0035

0.0044

0.0051

0.0051

0.0057

-2.7779

0.0058

0.0056

0.0055

0.0055

First Nat’l

0.0450

0.0348

0.0432

0.0502

0.0501

-0.9617

0.0406

0.0572

0.0551

0.0539

0.0539

BofA

0.0835

0.0646

0.0801

0.0931

-1.6374

0.1044

0.0753

0.1062

0.1021

0.0999

0.0999

MBNA

0.1251

0.0968

0.1201

-1.5776

0.1393

0.1565

0.1130

0.1592

0.1531

0.1499

0.1499

Capital One

0.0858

0.0664

-2.3833

0.0958

0.0955

0.1074

0.0775

0.1092

0.1051

0.1028

0.1028

Direct Merch.

0.0132

-3.4573

0.0127

0.0147

0.0147

0.0165

0.0119

0.0168

0.0162

0.0158

0.0158

CrossC’try

-2.2871

0.0036

0.0044

0.0052

0.0051

0.0058

0.0042

0.0059

0.0057

0.0055

0.0055

Merrick Bank

Cell entries (i, j), where i indexes row and j indexes column, give the percentage change in the accounts of issuer i with a 1% change in issuer j’s APR

0.1438

0.1463

JP Morgan Chase

Bank of America

0.1407

Citigroup

0.1038

0.1377

National City Bank

First National Bank

-1.1759

Providian

Prov.

Table 9: Own and Cross Semi-Elasticities of Credit Card Accounts wrt APR Sample from year 2004

40 0.0515 0.0596 0.0810 0.0688 0.0836

MBNA

Capital One

Direct Merchant’s Bank

Cross Country Bank

Merrick Bank

0.0629

0.0517

0.0609

0.0448

0.0387

0.0271

0.0582

0.0285

0.0336

-1.2746

0.0473

Nat’l City

0.1050

0.0864

0.1018

0.0748

0.0646

0.0452

0.0971

0.0475

-1.0788

0.0464

0.0790

Citi

0.0767

0.0631

0.0743

0.0546

0.0472

0.0330

0.0709

-0.8442

0.0410

0.0339

0.0577

JPM Chase

0.0020

0.0016

0.0019

0.0014

0.0012

0.0009

-2.5474

0.0009

0.0011

0.0009

0.0015

First Nat’l

0.0593

0.0488

0.0575

0.0422

0.0365

-0.9862

0.0548

0.0268

0.0317

0.0262

0.0446

BofA

0.0695

0.0572

0.0674

0.0496

-1.6744

0.0300

0.0643

0.0315

0.0372

0.0307

0.0523

MBNA

0.0903

0.0742

0.0875

-1.6141

0.0556

0.0389

0.0835

0.0408

0.0482

0.0399

0.0679

Capital One

0.0385

0.0316

-2.1635

0.0274

0.0237

0.0166

0.0356

0.0174

0.0206

0.0170

0.0289

Direct Merch.

0.0058

-3.2013

0.0057

0.0042

0.0036

0.0025

0.0054

0.0026

0.0031

0.0026

0.0044

CrossC’try

-2.0807

0.0004

0.0005

0.0004

0.0003

0.0002

0.0005

0.0002

0.0003

0.0002

0.0004

Merrick Bank

Cell entries (i, j), where i indexes row and j indexes column, give the percentage change in the transactions of issuer i with a 1% change in issuer j’s APR

0.0360

0.0378

JP Morgan Chase

Bank of America

0.0447

Citigroup

0.0773

0.0369

National City Bank

First National Bank

-1.2049

Providian

Prov.

Table 10: Own and Cross Semi-Elasticities of Credit Card Transactions wrt APR Sample from year 2004

41 0.2011 0.2014 0.1795 0.1523 0.1853

MBNA

Capital One

Direct Merchant’s Bank

Cross Country Bank

Merrick Bank

0.0818

0.0673

0.0793

0.0889

0.0888

0.0969

0.0757

0.0981

0.0954

-1.1120

0.0939

Nat’l City

0.1649

0.1356

0.1598

0.1793

0.1790

0.1954

0.1525

0.1978

-0.9376

0.1892

0.1892

Citi

0.1017

0.0837

0.0986

0.1106

0.1104

0.1206

0.0941

-0.7307

0.1186

0.1167

0.1167

JPM Chase

0.0063

0.0052

0.0061

0.0069

0.0068

0.0075

-2.5474

0.0076

0.0073

0.0072

0.0072

First Nat’l

0.0749

0.0616

0.0726

0.0814

0.0813

-0.8557

0.0693

0.0899

0.0873

0.0859

0.0859

BofA

0.1283

0.1055

0.1243

0.1395

-1.4709

0.1520

0.1186

0.1539

0.1495

0.1471

0.1471

MBNA

0.1926

0.1584

0.1866

-1.4143

0.2090

0.2282

0.1781

0.2310

0.2246

0.2210

0.2210

Capital One

0.1216

0.1000

-2.1635

0.1323

0.1320

0.1442

0.1125

0.1459

0.1418

0.1396

0.1396

Direct Merch.

0.0169

-3.2013

0.0164

0.0184

0.0184

0.0201

0.0157

0.0203

0.0197

0.0194

0.0194

CrossC’try

-2.0807

0.0055

0.0065

0.0073

0.0073

0.0079

0.0062

0.0080

0.0078

0.0077

0.0077

Merrick Bank

Cell entries (i, j), where i indexes row and j indexes column, give the percentage change in the debt of issuer i with a 1% change in issuer j’s APR

0.2195

0.2223

JP Morgan Chase

Bank of America

0.2160

Citigroup

0.1713

0.2125

National City Bank

First National Bank

-1.0491

Providian

Prov.

Table 11: Own and Cross Semi-Elasticities of Credit Card Debt wrt APR Sample from year 2004

Table 12: Mean Own-Price Semi-Elasticities Quantity Variable

Mean

Wtd. Mean

Accounts

-2.0502

-1.9027

Transactions

-2.0647

-1.9340

Debt

-1.8657

-1.7259

Initial shares are used as weights to compute the weighted mean.

Table 13: Cost Parameter Estimates : Full Model Variable

Coeff.

Std.Err.

Chargeoffs Proportion: (%)

5.13**

0.1300

Marginal Transaction Cost: (Net of Misc. Fees) Coefficients Rewards Constant Mean (cents)

0.00019** 0.00563** 0.569

0.00004 0.00035

Account Servicing Cost ($)

4.24776

15.7704

Significance at 1% and 5% levels is indicated by ** and * respectively

42

Table 14: Counterfactual Results: Average Percentage Changes in Key Variables

Variable

IF = 0.5%

IF = 0.75%

IF = 1.0%

All Issuers Profits Debt Transactions Accounts Interest Rates

-12.518 -7.926 -7.412 -7.025 +6.652

-9.437 -5.593 -5.229 -4.957 +4.656

-6.277 -2.888 -2.682 -2.538 +2.434

Large Issuers Profits Debt Transactions Accounts Interest Rates

-13.839 -7.990 -7.550 -7.166 +6.667

-10.810 -5.957 -5.516 -5.231 +4.875

-7.615 -3.051 -2.802 -2.651 +2.564

Small Issuers Profits Debt Transactions Accounts Interest Rates

-11.802 -7.892 -7.337 -6.948 +6.644

-8.694 -5.397 -5.074 -4.808 +4.537

-5.552 -2.800 -2.617 -2.477 +2.364

Large issuers are those with initial market shares more than 5%.

43