Credit Card Lending and the Performance of U.S. Credit Unions

Is credit card lending by credit unions within the United States primarily a service to members or a profit generating product? This paper examines the impact of credit card lending on the performance of U.S. credit unions from 2000-2017. A panel data approach using fixed effects regression methodology is undertaken to make comparative analyses of credit union performance across several dimensions including the percentage of the firm’s assets in credit card loans and percent of members with a credit card. Credit unions are stratified into deciles by size and significant results are found using this methodology. Controlling for delinquencies and charge-offs among other variables, credit card lending significantly increases ROA for both large and small credit unions, but only after the Financial Crisis of 2008, and the establishment of the CARD Act in 2009. Interestingly, the ROA of small credit unions significantly increases with the percentage of members using the institution’s credit card. This result suggests that small credit unions would benefit by increasing the penetration of credit cards within their membership base. © 2023 by the authors. Licensee SSBFNET, Istanbul, Turkey. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Financial institutions in the United States began issuing credit cards in the 1950s. By the end of 2020, consumer credit car d debt in the U.S. totalled $974.6 billion. 1 Credit unions in the U.S. represent a growing share of outstanding credit card debt. Their share of the credit card lending market has doubled over the last 20 years from 3.2% in 2000 to 6.4% in 2020. 2 The credit unions that offer credit card programs do so partially as a service to their members. Credit unions are member -owned, not-for-profit financial cooperatives that operate for the purpose of providing financial services at competitive interest rates and fees, which are generally lower than those charged by commercial banks. 3 In 2020, the average credit card interest rate was 10.89% at credit unions compared to 12.1% at commercial banks. 4 Banks also experience higher delinquency rates and charge-off rates on credit card loans than credit unions, especially during the 2008 Financial Crisis. Given credit unions' lower interest rates and fees, the purpose of this research is to test if the credit card business adds to the profitability of credit unions. Is credit card lending by credit unions a service to members and/or a profit generating product?
Credit unions are much more likely to issue credit cards than commercial banks. Only about 18% of U.S. banks issue credit cards (measured by whether a bank or credit union is reporting outstanding credit card loans) while 61% of credit unions offer credit cards. Part of the reason for this disparity is because the 10 largest U.S. bank holding companies represent over 90% of the credit card loans outstanding at banks, and thus leaving little incentive for smaller banks to enter that market. In contrast to bank credit c ard issuance, Levitin (2009) finds that 47 of 100 top credit card issuers in the U.S. are actually credit unions. Further, from our data in Table 1, the percentage of credit unions with credit card debt increases from 2000 to 2017 in every asset size decile. The differe nce between credit unions and commercial banks thus provides another motivation for this research. Why are credit unions much more likely to 2 issue credit cards than commercial banks? Is it as a service for members or profit generation or both? While a paper by Goddard, McKillop, and Wilson (2008) does examine the effect of asset portfolio diversification on credit union revenue streams, they do not examine or control for the effects of credit card lending on credit union profitability. Our paper is thu s the first study of the effect of credit card lending on performance (profitability) of credit unions.
This research examines the impact of credit card lending on the performance of credit unions. We collect credit union Call R eport financial data from the National Credit Union Administration (NCUA), the Federal government regulator, for the years 2000 to 2017. We then investigate credit card activities from two standpoints: the total dollar amount of annual credit card lending scale d by total assets and the percentage of credit union members using a credit card issued by the institution. Performance is measured by the annual Return on Assets (ROA), defined as net income divided by total assets. Given our large time -series panel data set, the analysis uses a fixed effects regression model with several control variables, including delinquencies and charge -offs. We find evidence that credit card lending did increase ROA, but mostly after 2008. For smaller credit unions, the percentage of members u sing a credit card issued by the credit union was found to significantly increase ROA over the entire time sample. For larger credit union s, the impact was not as significant.

Literature Review
In the sphere of financial institution research, credit unions are very important participants in our economy but are not wid ely investigated or well understood. Whereas there is a large body of research on commercial banking and thrift institution oper ations and performance, published credit union research in academic journals is minimal. Their nonprofit structure and income tax exemption provide interesting operational differences in the competing structural forms and operating efficiencies, relative to other competing financial institutions. 5 Bauer (2008) states that while credit unions comprise an important part of the depository institutions industry, the lack of existing research is mostly attributable to the nonprofit cooperative nature of the organization. Many credit unions also utilize volunteer labor and some receive benefits of sponsorship from associations or employers. Although there are substantive differences in goals, organizational form, regulatory oversight, and taxation, credit union s do compete with banks and savings institutions in various loan and deposit categories, and they provide an interesting area of research to expand our understanding of the full range of U.S. financial institutions. While there is a wide dispersion in ass et size among credit unions, i.e. the average asset size is less than commercial banks. At the end of 2017, half of all credit unions had total assets of less than 31 million dollars and a quarter had less than 9 million dollars in total assets (NCUA, 202 0). The number of credit unions has also been shrinking because of consolidation in the industry, which is one factor that has generated increases in average credit union size. Between 2000 and 2017, the number of credit unions has fallen from 10,316 to 5,573. Over this same period the mean and median asset size has grown from $42.5 million and $7.2 million in 2000 to $247 million and $31 million in 2017. The increasing siz e has been accompanied by growth in the types of services and products that credit unions offer. This paper researches some factors that are evident in their operations and show how they differ from their larger competitor s in the banking industry. 6 Feinberg (2001) and Feinberg and Rahman (2006) show that credit unions impact consumer loan rates in markets where they are significant participants. Hannan (2003) finds that credit unions present a growing competitive challenge to b anks and thrift institutions within urban lending markets. Wheelock and Wilson (2013) find that technological advances have reduced some advantages of small-scale lenders, such as proximity of location and personal service provided. Their results indicate that costefficiency fell for the average credit union due to changing technology. However, smaller cre dit unions were impacted to a greater extent, and thus provide a rationale for the industry consolidation previously mentioned. Goddard, McKillop, and Wilson (2008) report a similar result regarding the operating results of small versus large credit union s. They found that only the largest credit unions became more scale efficient over time.
The regulation of credit card issuance for all U.S. financial institutions, including credit unions, changed dramatically with the Credit Card Accountability, Responsibility and Disclosure Act of 2009 (CARD Act). The CARD Act introduced reforms to protect consumers, mainly in the areas of fees and transparency. The biggest changes limit what can be charged for late fees and ove r-limit fees. Most of the transparency rules require advance notice of any significant changes, especially in interest rates or f ees. The CARD Act also mandates more uniform reporting of credit card terms so that consumers can more easily understand and compare different credit card offers. Levitin (2009) suggests that with the credit union industry being largely comprised of sm aller lending institutions with less aggressive expansion into credit card lending prior to the 2008 financial crisis, the shock to credit card operations of credit unions resulting from the CARD Act will be less than that for those commercial banks which already possessed large exposure to credit card lending. Levitin (2009) also suggests that the user-friendly business customer/member model of credit unions can attract new customers that may feel frustrated with the credit card lending practices of commercial banks. 5 While legally organized as a non-for-profit cooperative, credit unions are expected to be profitable. As stated previously, credit unions must meet capital adequacy requirements similar to those for commercial banks and must also increase retained earnings in order to grow. We calculate the aggregate Return on Assets (ROA) of NCUA insured credit unions and compare that figure the aggregate ROA of U.S. commercial b anks (obtained from the Federal Reserve FREDS database). We calculate the annual ratio of credit union ROA divided by c ommercial bank ROA annually for the 2000-2017 period and find the mean and median ratios to be 87.70% and 75.54%, respectively. We also find that the credit union RO A exceeds that for commercial banks for the years 2008 and 2009. 6 Li and van Rijn (2022) find results that suggest credit unions are more risk-averse with regard to lending and posit that the customer/member nature of credit union structure contributes to having less aggressive lending practices.
3 What has been the impact of the CARD Act on consumers? Bourke and Hollifield (2010) found evidence that transparency had improved, but also evidence of higher cash advance fees, which were not limited in the CARD Act. Bar-Gill and Bubb (2012) find decreases in late fees and over-limit fees as expected and evidence that annual fees and purchase interest rates, which were not as heavily controlled in the CARD Act, were basically unchanged. Agarwal, Chomsisengphet, Mahony, and Stroebel (2014) found that consumers, especially higher risk borrowers, benefitted from lower borrowing costs. Overall, these researchers find that the CARD Act has helped consumers lower credit card costs.
However, other research shows that financial institutions have reacted in ways to lower access to credit cards. Jambulapati and Stavins (2014) show that banks lowered the credit limits on credit card accounts, although the number of closed credit card a ccounts were about the same as before the CARD Act. Elliehausen and Hannon (2017) found evidence that the CARD Act led to less credit card lending to lower credit borrowers. However, there appeared to be little impact on better credit customers. Debbaut, Ghent, and Kudlyak, (2016) find that borrowers under 21 years of age were more likely to have fewer credit cards after the CARD Act than before its enactment.
What has been the impact of the CARD Act on the performance of financial institutions? To our knowledge, this is the first p aper to investigate the CARD Act and the profitability of credit card lending. This paper seeks to add to the literature by using a panel data fixed effects model to test the influence of credit card lending on profitability (ROA) with a set of control variables.

Data and Descriptive Statistics
For our study we utilize annual financial data collected from the Call Reports of all U.S. state and federally chartered cred it unions collected from 2000 through 2017. The study begins with the year 2000 data because that is when modern U.S. capital adeq uacy requirements for depository institutions were fully applied to the credit union industry (Goddard, McKillop, and Wilson (2016 )). The annual call report data is retrieved from the National Credit Union Administration (NCUA) via this Federal government regulator's website. For our analyses, we require that a credit union have federal deposit insurance via the National Credit Union Share Insurance Fund (NCUSIF). We exclude those credit unions (a very small percentage; this represents approximately 1% o f all credit unions) that while listed in the NCUA call report database, nevertheless, choose to obtain private sector deposit insurance in lieu of the NCUA federal deposit insurance.
For a number of analytical purposes, we break the sample into two sub periods, one from 2000 through 2008 representing the prefinancial crisis and pre CARD Act time frame and the other, 2009-2017 for the post crisis and post CARD Act era. These periods are also compared to the entire time frame from 2000-2017. Prior to the 2008 Financial Crisis, strong evidence exists of lax underwriting standards and even predatory lending in the U.S. market. Part of establishing the CARD Act in 2009 was to reduce predatory lending in credit cards and increase transparency to consumers. It was apparent that the regulatory environment for credit cards shifted significantly in 2009, so it is of interest to investigate the subsequent subperiod 2009-2017.
Tables 1 through 3 report the descriptive statistics. We stratify the data into deciles based on total asset size (assigned each year). Wheelock and Wilson (2013) amongst others find operating and performance differences between large and small credit unions; differences which we also confirm and report in these descriptive statistics.
In Table 1 we report aggregate descriptive statistics, including credit card lending data, for all NCUA insured credit unions for the years ending 2000, 2008, and 2017. 7 The firms are sorted into size deciles based on total assets. As seen in Table 1, the top 3 decile groups (8, 9, and 10) account for about 95% of the industry's assets. In fact, the total assets represented by only the cred it unions in decile 10 grew from 78.3% in 2000 to 83.43% in 2008 to 85.62% in 2017. We categorize deciles 8, 9, a nd 10 as "Large" credit unions and deciles 1 through 7 as "Small" credit unions. Table 1 also reports that most of the credit union industry credit card lending is contained within size decile 10: 78.3% of the total in 2000, 83.43% in 2008, and 85.62% in 2017, so not only is most credit card lending contained within the largest asset size decile, but also that percentage has been increasing over time. 1: Descriptive statistics; aggregate assets and credit card loans All credit unions contained in the National Credit Union Association (NCUA) annual Call Report records are sorted into 10 deciles based on total assets for each year. Panels A, B, and C below report results for calendar years ending 2000, 2008, and 2017, respectively. Results below report total assets and credit card (CC) loans for each decile, as well as cumulative totals for all credit unions. Also shown are each asset decile's percentage of total credit union assets and total credit union credit card loan s. The penultimate column reports the total credit card loans of each asset size decile as a percentage of that decile's total loans . Amounts shown below are listed in U.S. dollars.   Table 2 reports the mean annual return of ROA by asset size decile for the entire 2000-2017 sample time period, with Panels A and B reporting results for credit card issuing and non-credit card issuing credit unions, respectively. Panel A also lists the mean level of credit card lending (credit card loans as percentage of assets), as well as the percentage of members with a credit card b alance. 8 We observe that credit card issuing credit unions in the highest two size deciles report higher ROA than non -credit card issuing credit unions, while the opposite is observed for size deciles 1 through 8. We further examine this ROA difference in Table 3.
In Table 3, we report the results of a difference in means test between the ROA of credit unions that issue credit cards vers us those that do not issue. The ROA for credit card issuers is definitely higher and statistically significant (at far better than the 1% level) for decile 10 credit unions versus the decile 10 non credit card issuers. The opposite was observed for the smaller deciles wher e non credit card issuing credit unions had higher ROA. The decile 10 credit card issuing firms, where most of the credit union credit card lending occurs, are the most profitable credit unions as measured by ROA; however, we note that around 90% of decile 10 firms do issue credit cards, so our observed result may be a self -selection bias of less profitable large credit unions choosing to forego the issuance of credit cards. 8 The two main independent variables used for the fixed effects regression analysis in Tables 4 and 5 are (1) credit card balance divided by total assets, and (2) percentage of CU members with a credit card. We also repeat the analysis for the 2000 -2008 and 2009-2017 subperiods (unreported for brevity), and find that the results are qualitatively similar to the full 20 00-2017 period. We also generated (also unreported for sake of brevity) results for the control variables used for the fixed effects regression results in Tables 4 and 5. We find that all of these variables are certainly related to asset size: operating expenses, loan delinquencies, loan write-offs, and liquidity variables are all inversely related to asset size, and thus indicate the increasing economies of scale of large credit unions. We also find that the level of uninsured deposits is increasing with asset size. Our primary results are obtained by using a panel data approach using two -way fixed effects regressions, controlling for both the firm and year for analyzing the effects of independent variables on our choices of dependent variables. We perform a Hausman test to determine whether the preferred regression model is random effects versus fixed effectsthe results of the test strongly suggest that the preferred model for this analysis is fixed effects.
We selected return on assets (ROA) as our measure of financial performance. This dependent variable was regressed separately on the level of credit card loans and percentage of credit card holding members, and on control variables consisting of the LN o f assets, LN of squared assets, delinquent loans, net loan write-offs, operating expenses, liquidity, and uninsured deposits. Except for the LN assets and the LN of squared assets, all other independent variables and the dependent variables are all defined as ratios. All items are tabulated for each year for each credit union in our 2000 to 2017 sample, with the specific call report variable codes listed in the Appendix. Credit card loans are defined as the dollar amount of credit card loans outstanding divided by the firm's total assets. ROA is defined as Net Income (less "dividends"; the interest paid on member deposits) divided by total assets. LN Assets is found by t aking the natural log of the assets (total assets are deflated to end-of-year or December 2000 using the U.S nominal GDP index retrieved from the Federal Reserve's FRED database). Delinquent loans are defined as the amount of delinquent loans divided by total loans. Net loan write-offs are the total net loan write-offs (write-offs minus recoveries) for the year divided by the total loans. Operating expenses are the total non-interest operating expenses divided by total assets. Liquidity is cash-on-hand plus cash held in deposits at other institutions divided by total assets. Uninsured deposits are the dollar amount of deposits not insured by the NCUA (de posit accounts in excess of $100,000 prior to the fall of 2008 and in excess of $250,000 following fall 2008) divided by total deposits.

Methodology: Two-Way Fixed Effects Regression Model
The two-way fixed effects panel data regression model that we use controls for both firm and year fixed effects. 9 Our results reported in Table 4 are estimated using the following regression model (the Appendix describes the construction of the dependent, credit card independent variable of interest, and the control variables): Where ROA is regressed on a credit card independent variable CC_IV and set of control variables. The credit card independent variable CC_IV takes the form of either: (1) credit card loans outstanding divided by total assets or (2) percent of members with an outstanding credit card loan balance. Also, K = number of control variables (not including firm or time dummy variables). The firm dummy variables FD control for the firm-specific time-invariant factors (where N is number of firms, resulting in N-1 firm dummy variables), while TD are the time dummy variables (where T is number of years, resulting in T-1 year dummy variables) which control for time-variant changes that have the same effect on all firms. Table 5 uses a modification of Equation (1) above, and regresses annual changes in ROA on annual changes in both the credit card independent variable and the K control variables (except for the asset size controls). Annual changes in affected variables are defined as the first difference between time period t and t-1. Table 5 reports the results of the following model (2): The panel data regression analysis is run using Stata, and utilizes the XTREG function, controlling for both firm level and y early fixed effects. We report t-statistics using robust standard errors. The Call Report database contains two firm identifiers: (1) cu_number and (2) join_number. The cu_number is reassigned upon a merger or change in charter (Federal versus state), whereas the assigned join_number is pe rmanent. We thus use the join_number as our unique firm identifier. Acquired firms cease to be reported the quarter or year that they are acquired. 8

Findings and Discussion
Our analysis focuses on the impact of credit card issuance on credit union ROA performance. To our knowledge, this is the fi rst study of this performance relationship, and our results should prove useful to scholars as well as practitioners. We measure credit card issuance in two different ways. First, we define it as credit card loan balances outstanding as a percentage of total a ssets. Second, we use the percentage of members with a credit card balance with the credit union. We calculated this using d ata from the call reports taking the number of credit card loans divided by the number of members, i.e., loans per member. Our results find interesting similarities and differences based upon the credit card variable, size of credit union, and time period.
Given the doubling in the relative market share of outstanding credit card debt for U.S. Credit Unions, and the growth in the number of cards issued, it is important to study the impact that these changes have on ROA, and this is reported in Tables 4 and 5. The signs (and t-statistics) for the control variables in the regressions are what we would expect, and thus demonstrate that our controls are properly specified as constructed. As shown in Panel A of Table 4, regression results for all credit unio ns indicate that higher loan delinquencies and higher operating expenses are associated with lower ROA. Obviously, the same is observed for liquidity and net write-offs which confirms the known tradeoff between liquidity and ROA. As expected, higher leve ls of uninsured deposits are associated with lower ROA (a higher rate must be paid on deposits to make the uninsured deposit investments attractive). The coefficient for this is negative and significant at the 1% level for the entire period of study and in the first sub-period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008). However, in the second sub-period it is negative, but the level of significance is low.
In Table 4, when we use credit card balances as a percentage of total assets in the regression, the level of credit card loan s is found to significantly increase ROA over the entire sample period for both small credit unions and large credit unions as shown in panels B and C. This is a likely result of higher rates charged on credit card balances and economies of scale and scope resu lting from greater balances and more diversity in the firms' product/credit mix. However, all the increase occurs in the post CARD Act subperiod. For the pre-CARD Act subperiod, the quantity of credit card loans has no significant impact on ROA. When the percentage of credit union members with a credit card is used in the regression (level of issuance penetration among members) credit card issuance significantly increases ROA for small firms across all time periods. However, the impact is greater and mo re significant in the post-CARD Act period. Our results imply that for small firms, the economies of scale and thus profitability improve as more members have and use a credit card which has obvious implications for profitable management of operations.

Table 4: Panel data fixed effects regressions of Return on Assets (ROA)
We report coefficients of two-way fixed-effects regressions of Return on Assets (ROA) on either (1) credit card loans (level of lending or credit card loans divided by total assets) or (2)  We report the t-statistics in brackets (using robust standard errors) for differences in means; *, **, and *** indicate statistical significance (two -tailed tests) at the 10%, 5%, and 1% levels, respectively. Panel A: All credit unions (all asset size deciles) 2000-2017 2000-2008 2009-2017 2000-2017 2000-2008 2009-2017   For large credit unions, ROA significantly increases with the percentage of members only for the post CARD Act period of 2009-2017, although with a much smaller scale impact than for the small credit unions. For large credit unions the percentage of members with a credit card account actually has a negative but statistically insignificant coefficient on ROA for the pre CARD Act pe riod of 2000-2008. This adds support to our findings in Table 5 where we show evidence that large firms were apparently overextended into credit card lending. In Table 5, the results are shown for the model run by regressing annual changes in ROA for the credit unions against changes in the independent variables (except for the asset size controls). For the full credit union sam ple (Panel A), increasing credit card balances results in increasing ROA with a 5% significance level for the post CARD Act period and lesser degrees of significance for the pre-CARD Act period, as well as the entire period.   For large credit unions (Panel C), they do not benefit from increases in credit card lending balances during the pre-CARD Act period. Moreover, increasing card penetration among members significantly negatively impacted ROA. These results imply that lax underwriting standards in the pre-CARD Act period really hurt large credit unions. For the large credit unions, po st-CARD Act increases in the level of lending and increased credit card penetration among members both have a positive and statistically significant effect on ROA at the 5% level. Overall, the large credit unions gained by growing credit card balances and credit card accounts in the post-card/crisis period. For the smaller credit union sample in Panel B, we find statistically significant gains in ROA from increases in credit card lending over the entire period, but the gains are substantially in the post-CARD Act era. Gains are also achieved through increases in percent of members, over the entire period, but most of the result is due to post -crisis gains in members.
These results indicate the importance of credit card lending for credit unions as we move f orward.

Conclusion
Our research finds interesting and useful results for U.S. credit union operations in the wake of the implementation of the C ARD Act of 2009. This paper analyzes how the Card Act differentially impacted the operations and profitability of credit unions over time that varied in size and scope of operations. Analysis used credit card variables for the firms that varied based upon differ ences in credit card lending, member usage, size of the credit union, and the time period evaluated. Surprisingly, the results show that the proportion of credit card loans on a credit union's balance sheet only has a significant impact on ROA for large and small cr edit unions in the post CARD Act time period of 2009 to 2017. This seems to confirm the literature regarding the service aspect of this product for credit unions as well as evidence of lax lending standards in the pre -CARD Act time period. Interestingly, when measured as the percent of members using a credit card issued by the credit union, credit card issuance only significantly increases ROA for small firms across both the 2000 to 2008 and 2009 to 2017 time periods. This result implies that for small firms, economies of scale and scope exist and thus profitability improves as a higher percentage of members use a credit card. Finally, these results provide some evidence that the CARD Act has helped to increase the ROA from credit card activities. Given the paucity of extant cred it union research, other researchers may find it fruitful to examine similar member-owned financial cooperatives in countries outside of the U.S.