The Benefits and Costs of Fee-Income Generation in Small Banks

Jeffrey Stensland and Glenn Pederson

The authors are graduate research assistant and professor, respectively, in the Department of Applied Economics, University of Minnesota, St. Paul. This is Minnesota Experiment Station publication no. 971140011.

Abstract

Banks have been expanding the scope of their operations to include more fee-generating services as part of a strategy to increase overall profitability. This study evaluates whether the added services are improving an average bank’s profit margins. The investigation tests for scale and scope economies in service production and examines the profitability of specific fee-based products using survey data from Minnesota banks. A high fee-income strategy does not appear to improve small banks’ performance and does not provide sufficient benefits to overcome the diseconomies of scale in small Minnesota banks.

Key words: banks, service-fee income, economies of scale and scope.

Article <top>

Bankers may have developed a fee income obsession (Robertson). Even small banks are expanding the scope of their services to generate more noninterest income and provide their customers with a "full service" bank¹. While their motivation is to generate additional fee income, the cost of providing these services is relatively high For example, the Federal Reserve Board’s functional cost analysis (FCA) data for 1991­94 indicate that small banks (those with $0­$50 million in assets) and medium-sized banks (those with $50­$200 million in assets) tend to lose money on their fee-based services.

¹Call report data suggest that the level of noninterest income is increasing. A survey of Minnesota banks (Stensland and Pederson) indicates that, after excluding demand deposit fees, about 50% of the service-fee income represents loan fees and the remaining 50% represents income from other fee-based services. The survey also found that banks are expanding the number of services offered.

Although the FCA data reveal that these services tend to be cost (not profit) centers, there have been recent articles calling for expanded use of these services (see Bird and Harvey; Robertson). Moore and Couch call the shift toward services, and other off-balance-sheet activities, "the definitive trend in the epoch of modern banking" (p. 13). In addition, Lawrence and Shay conclude that these nonbank services exhibit economies of scale and scope. They state that small banks will have to expand this activity if they are going to remain competitive. Similarly, Norwest Bank (a Midwest regional bank) has recently reported that it is attempting to increase the number of products sold from four to eight per household (Berg). At the national policy level, the proposed Financial Services Competitiveness Act of 1997 includes provisions that would allow banks to expand further into the securities and insurance markets.² In this rapidly changing environment, small- and medium-sized banks are being forced to decide if they will increase the scope of their services as a means of competing with the larger banks.

²See H.R. 10, 105th Cong., 1st Sess., 7 January 1997.

For services to increase expected profits, they must exhibit positive profit margins, improve the margins of other product lines, or increase the total volume of business. Even if the FCA is correct in describing these services as cost centers, bankers could still be justified in expanding their use of services if the added services foster improved profitability of traditional deposit-taking and lending activities.

The objective of this article is to evaluate the success of service strategies in small- to medium-sized banks by examining the effect of fee-based services on bank profit margins, economies of scope associated with those services, and the effect of enhanced services on asset growth. This study evaluates a sample of banks that includes all Minnesota banks with less than $200 million in assets. The focus is on the average benefit derived from expanded services, as opposed to the benefit for the most efficient banks located on the production possibility frontier.³ Therefore, the emphasis is not on whether the most efficient small banks can survive. Rather, we explore how profit margins are affected by expanded services in small- to medium-sized banks. One of the key questions is whether a small bank with average management will benefit from expanding its service offerings.

³Prior studies that have looked at the performance of banks on the production possibility frontier include Ferrier et al., and Berger and Humphrey.

Traditional Banking and Service-Fee Income <top>

In order to document the traditional banking and service-fee income activities of banks, annual call report data from 504 Minnesota banks (those with less than $200 million in total assets) are merged with mail survey data collected in 1995 (Stensland and Pederson).⁴ The data suggest that larger banks have healthier profit margins, but it is unclear whether this is due to higher levels of traditional income or service-fee income. Traditional banking income is defined as net interest income and demand deposit fees less the provision for loan losses. Service-fee income includes income from standby letters of credit, trust services, insurance services, brokerage services, safe-deposit services, and credit card fees.

⁴Small- and medium-sized Minnesota banks were mailed surveys in February 1995. The survey requested information on their marketing, staffing, and service strategies. In particular, banks were asked if they currently offered (or planned to offer) conveniences such as ATMs, automated telephone banking, mutual funds, annuities, insurance, travel services, leasing, credit cards, lock box services, trust services, and discount brokerage services. Of the 252 banks randomly selected for survey, 109 with under $200 million in total assets returned a completed survey. Those 109 banks comprise the survey data set.

Tables 1 and 2 show that banks with less than $100,000 in fee income tend to have lower profitability ratios, but the profitability of banks with $100,000­ $300,000 in fees is approximately the same as banks with over $300,000 in fee income. If there are significant economies of scale and scope associated with fee-income generation, we would expect that banks offering more services and earning additional fees would exhibit superior financial performance. Average noninterest expenses increase from 3.3% of assets among the banks with under $300,000 in fee income to 3.8% of assets among the banks with over $300,000 in service-fee income. The extra noninterest expense offsets the extra fee income in high-service banks. Thus, it appears that the additional service-fee income is not highly correlated with improved bank profitability. In this regard, the poor performance of banks with less than $100,000 in fee income may be due to their small loan volume, as opposed to their lack of fee income.

The data also indicate that smaller banks are less likely to offer customer conveniences such as telephone banking and ATM machines, and may offer fewer services overall. However, not all banks offering multiple services generate over $100,000 in fee income. Those banks either may be in the start-up phase of offering services or may be selling the services to a rather limited number of customers. Thus, survey data suggest that banks differ both in the level of services offered and in the level of effort exerted on individual services. Because the offering of a service is only part of the story, we must also examine the total volume of income received from services when evaluating the impact of fee-based services on bank performance.

Scale and Scope Economies in Banking <top>

Recent studies have analyzed the profit margins and the presence of economies of scale and scope in the production of traditional bank products. (Previous work on scale and scope economies has been reviewed by Clark, while Berger, Hunter, and Timme provide an excellent discussion of bank efficiency research.) These studies often used the translog cost function since it is a flexible functional form and the coefficients from the translog function provide direct measures of scale and scope economies.

Economies of scale are generally found to exist up to at least $100 million in assets (Berger, Hunter, and Timme). However, McAllister and McManus provide strong evidence that the translog function is not a good reflection of a global cost function. They suggest that a more flexible form is needed to estimate economies of scale and scope. Using nonparametric methods, they find evidence of economies of scale among banks with up to $500 million in total assets and constant returns to scale above that level. Featherstone and Moss focus on economies of scale in agricultural banks. They conclude that economies of scale may be exhausted at $60 million in total assets. These studies of bank efficiency are similar in their focus on traditional bank product lines and their lack of focus on fee-based services. When fee income was considered, output was measured simply as other income.

In an earlier study, Lawrence and Shay include fee-based services in their analysis of scale and scope economies. They employ FCA data to measure traditional outputs using the levels of deposits, loans, and investments. Service-fee output is assumed to be equal to the estimated cost of generating those services. For example, if the functional cost analysis indicates that on average a bank spends $1,000 on servicing safe-deposit boxes, the Lawrence and Shay study assumes that the bank has $1,000 in output. They divide the sample of banks into quartiles based on asset size, and estimate scale and scope economies using a translog cost function. Lawrence and Shay conclude that substantial economies of scale exist in generating nonbank services. However, service fees are not the main focus of their study, and their methodology with respect to fee-income generation has some significant shortcomings. For example, they assume that a one-to-one relationship exists between service output and the cost of fee-based services.

Measuring the Fee-Income Focus <top>

To empirically evaluate whether small banks might benefit from expanding their effort on fee-based services, it is necessary to measure their level of effort. We approach the measurement problem from three different perspectives. The first is to measure effort using cost accounting. The Federal Reserve FCA reports the costs and revenues of specific product offerings. While the cost accounting approach is the most direct, it has some limitations. For example, it requires managers to make subjective estimates about what percentage of employee time is devoted to providing the fee-based services. In addition, the FCA approach does not catch any of the "spillover" effects of offering services (e.g., the existence of a bank-based insurance agency may bring more clients into the bank for home loans).

The second approach is to measure the specific products offered by banks, and then test whether offering specific services shifts profit margins and/or asset growth rates. The existence of specific products can then be regressed on the noninterest expenses of the banks. The problem with this approach is that a list of services does not quantify the amount of effort devoted to each service. For example, some banks may have less than one full-time employee devoted to their insurance business, while other banks may employ several people in their insurance agencies.

The third approach is to use an output measure to estimate the amount of effort exerted on the production of fee-based services. One output measure is the level of other service-fee income, as reported in the bank call reports. The output measure is created by dividing the bank’s level of other service-fee income into three categories (under $100,000, $100,000­ $300,000, and over $300,000). Three categories of service-fee income are used instead of the actual dollar value of service fees generated in order to reduce the likelihood of estimating false economies of scale.⁵ Averaging the data over four years and using categories of output rather than output itself to test for scale economies reduces the effect of the regression fallacy. The potential for misleading results is reduced since a chance fluctuation will rarely cause a bank to jump from one broad fee-income category to another.

⁵See Borts on this "regression fallacy."

There are two caveats that accompany the output approach to measuring effort. First, other service-fee income includes fees from two broad types of services: fees from services such as trust, brokerage, insurance, consulting, and safe-deposit boxes, and fees generated on loan commitments and standby letters of credit.⁶ Loan fees are expected to partially reflect the cost of evaluating a credit and related costs of loan administration, but they also represent payments for giving the customer an option to borrow. Loan fees can be a substantial portion of a bank’s reported other fee income. Among the banks surveyed in this study, loan fees represent about 50% of the service fees unrelated to demand deposit accounts. Thus, service fees represent a wide category of off-balance-sheet activities. Although we can test for other service-fee income effects, this does not reveal which specific off-balance-sheet activities are responsible for the shift in profit margins.

⁶The call report stipulates that other fee income includes: fees from loan commitments when the commitment expires, trust services, safe-keeping charges for safe-deposit boxes, the sale of credit life policies, the collection of utility and other bills, the redemption of U.S. savings bonds, the handling of food stamps, the execution of acceptances of some letters of credit, servicing mortgages held by others, consulting or advisory services, annual or periodic credit card fees, charges to merchants (if the bank does not carry the credit card loan on its books), fees from certain standby and other option contracts, rental fees from furniture rentals, and "any other service charges, commissions, and fees not required to be reported in other items of Schedule RI, Income Statement" [Federal Financial Institutions Examination Council (FFIEC) 1996, p. RI-8]. Banks with over $100 million in assets report fiduciary income as a separate income category. In our analysis, fiduciary income was included as a source of "other fee income." ATM fees were considered to be part of demand deposit fees, as opposed to part of "other fee income."

If we want to look at the effect of specific services on profit margins, we need to turn to indicators of specific product offerings.

A second difficulty with using other fee income as a measure of effort is that a bank holding company or a partnership may operate some services. The fee-based activities of the bank might be a profit center and the owners and/or directors of the bank may operate those services privately to reduce corporate taxes. If profitable services are kept "off the books," the calculations of bank profit margins based on call report data will be understated. Conversely, the relative profitability of banks that report all of their service activities on the bank’s books may be overstated. Hence, it is expected that the profitability of service activities will be slightly overstated when using other service-fee income as a proxy for fee-generating efforts.

None of the above proxy measures for effort are perfect. However, when used together, they allow us to evaluate the profitability of fee-generating services from different perspectives. The accounting measures complement the econometric results, which are based on output data and services offered. This multifaceted approach produces results that are a more adequate test of the effects of fee-based activities on bank performance.

Methodology <top>

This analysis proposes to look at the average return (rather than the optimal return) to different income-generating strategies. Thus, the first task is to estimate the average profit margin generated from fee-based services. The Federal Reserve attempts to measure this through a direct cost accounting approach. Initially, this study compares the cost accounting approach to an econometric approach that uses call report (FFIEC 1991­94) and survey data to measure the overall effect of services on profit margins.

Because the primary concern is with bank services, outputs and inputs are measured as cash flows, as opposed to measuring the "stocks" of loans, deposits, and employees. A benefit of measuring inputs and outputs in terms of flows is that it is not necessary to make strong assumptions about the quality of labor and/or other inputs. Studies that attempt to measure inputs in terms of full-time employees often are required to make strong assumptions about equality of productivity across employees. This is a particularly questionable assumption when dealing with banks that have different product lines. Since the econometric approach and the Federal Reserve’s cost accounting approach look at the total average costs per product line, there is no need to make this strong assumption about bank employees. It is assumed that accounting measures of total costs are reasonable approximations of actual economic costs.

Cost Accounting Approach <top>

The most straightforward approach to measuring the cost of generating fee-based services is to ask the bankers. This is the approach taken in functional cost analysis (FCA). The Federal Reserve asks bankers to estimate their costs of generating specific services. The costs are calculated by summing the salary, fringe benefits, occupancy, and other expenses associated with each specific product line. The number of banks participating in the FCA is small, and banks drop in and out of the sample. Therefore, the estimated cost of generating fee income varies significantly from year to year (see Table 3). For example, the sample size for each FCA category ranges from nine small banks with trust services in 1994, to 100 medium-sized banks offering safe-deposit services in 1992. The small sample size may contribute to the variance in these estimates, but would not account for the striking fact that every service category was estimated to be a cost center in every year from 1991 through 1994.

The FCA data reveal that banks with under $200 million in assets tend to lose money on their services (Table 3). Among small banks, the average cost of generating $1 of fee income ranges from about $1.38 for trust services in 1993, to $2.73 for other services in 1994.

Medium-sized banks experience somewhat lower unit costs, but they, too, tend to exceed $1, and the cost estimates vary greatly (from $1.06 for credit cards to $1.77 for safe-deposit boxes in 1994). Although the FCA process consistently concludes that nonbank services are not directly profitable for most banks, it still may be rational for banks to offer loss-making services if they attract customers who also purchase products with higher, positive margins. In this way, a service might be given credit if it improves the margins on traditional product lines. It is also possible that some services are profitable while others are unprofitable. The overall effect of services on profit margins and the relative value of services can be tested using a combination of call report and survey data.

Econometric Model Approach <top>

An econometric approach is used to estimate the profit margins on specific product areas. Sources of cash flow are divided into three general product areas: net income from traditional bank products⁷ (e.g., net interest income plus demand deposit fees less the provision for loan losses); income from loan fees and service products (e.g., trust, discount brokerage, insurance, credit card, safe-deposit box, and other fees generated from customer services); and other noninterest income products (e.g., data processing fees, gains and losses on certain asset sales, and other noninterest income items). To separate the effects of "high-touch" service fees from the effects of loan fees, dummy variables are added to the regression equation in order to represent the existence of the former category of fee-income generating activities.

⁷Because the relative profitability of different types of loans is not the focus of this study, net interest income from all lending activities is aggregated into a category called "traditional income." If loan fees or other sources of noninterest income are correlated with a particularly unprofitable subset of assets, there is a possibility of biased econometric results. However, the bias would not affect cost accounting estimates, and the potential for bias in the econometric approach is expected to be slight. Nevertheless, future researchers should consider disaggregating the sources of traditional income.

Empirical Model <top>

Noninterest expense is modeled as a linear function of output along three general product lines: traditional banking products, service products, and other products. In addition, noninterest expense is specified as a linear function of bank equity capital:

(1)

where NIE is total noninterest expense; TRBI is traditional banking income (net interest income plus demand deposit fees less the provision for loan losses); SFINC is service-fee income (e.g., loan fees, trust, discount brokerage, travel services, credit card fees, and safe-deposit fees); OTHNII is the level of other noninterest income including data processing fees and gains on certain asset sales; and EQTY is the average level of bank equity capital. We expect that banks with high levels of equity financing will have lower interest expense and lower costs of administering deposits in relation to their competitors at the same level of gross income. Therefore, the equity capital variable is added to account for the cost savings associated with equity as opposed to debt financing. The constant term (CONST) represents the fixed overhead costs of operating the bank at any positive level of output. The error term is denoted by ε.

The coefficients on the traditional income and fee income variables represent the average noninterest expenses associated with generating $1 of income from those products. These coefficients are equivalent to efficiency ratios for specific product lines.⁸ The coefficient on equity capital is interpreted as the unit cost reduction associated with using equity financing instead of debt financing. To mitigate the problems associated with product cycles and nonmeasurable changes in the level of competition, the data are averaged over the period 1991­94. Also, since heteroskedasticity creates a problem when the dependent and independent variables covary with the bank size, the variables on both sides of (1) are divided by average total assets.

⁸ The efficiency ratio equals noninterest expense divided by net interest income and fee income. The product line efficiency ratio for fee-based services refers to the noninterest expense associated with the fee income less any cost savings associated with economies of scope stemming from fee-based services, all divided by service-fee income. Since an econometric approach looks at the net impact of service fees on noninterest expenses, the benefits from economies of scope are included in the measure of fee income efficiency.

Since (1) is an estimation of average profit margins, the estimated coefficients reveal little about how margins may shift as the scale of fee income activity or traditional banking changes. To remedy this problem, dummy variables are added in (2) to test for shifts in profit margins attributable to economies of scale and/or scope in fee income generation. Three levels of fee income generation are identified: under $100,000, $100,000­$300,000, and over $300,000. The first level of fee income consists of banks with less than $100,000. The $100,000 cutoff was chosen to separate banks with a minimal fee income focus from the other banks in the sample. Banks with $100,000­$300,000 in fee income may have two to three people working in service-fee areas, but those banks still have fairly limited fee-generating operations. Banks with over $300,000 in service-fee income are likely to have several people providing services, and are considered high-service banks. The use of three dummy variables representing the level of service-fee income is one way of evaluating how bank services can affect the profit margins on traditional product lines.

Three dummy variables are added to test for economies of scale in the traditional product lines. The identified levels of traditional income are: under $1 million, $1­$2 million, $2­$4 million, and over $4 million. Banks with less than $1 million in traditional income tend to have less than $20 million in assets.⁹ Moreover, bankers operating small rural banks tend to indicate that this is the minimum size for an efficient bank (Stensland and Pederson). Therefore, banks with less than $1 million in traditional income are expected to have lower than average profit margins. All Minnesota banks with under $4 million in traditional income had less than $100 million in assets, while most banks with over $4 million of traditional income had over $100 million in assets.¹⁰ Therefore, the $4 million level approximates the point where previous studies indicate that banks exhibit full economies of scale. The $1­$2 million and $2­$4 million categories serve to separate the banks that may lack economies of scale from those that are assumed to exhibit close to full-scale economies.

⁹Of the 197 banks with less than $1 million in traditional income, 163 have less than $20 million in assets and all have less than $32 million in assets.

¹⁰Past banking studies have tended to find economies of scale in banks with up to $100 million in assets.

Adding the traditional income dummy variables to (1) results in model (2) for noninterest expenses:

(2)

where SF100 is a dummy variable indicating the level of fee income is less than $100,000, and SF300 is a dummy variable indicating that the level of fee income is in the $100,000­$300,000 range. The TRAD1, TRAD2, and TRAD4 dummy variables denote that the level of traditional banking income is between $0­$1 million, $1­$2 million, and $2­$4 million, respectively.

Positive coefficients on any of the dummy variables in (2) indicate that banks in this range of production have average profit margins lower than larger banks (those with over $300,000 in service-fee income and over $4 million in traditional banking income). Lower profit margins among banks with lower volumes of traditional banking income suggest that economies of scale exist in traditional banking products.

Since the dummy variables represent discrete shifts in noninterest expenses, (2) implies a discontinuous cost function. One method of smoothing the cost function is to use a spline function approach (Green, pp. 235­38). A spline function creates a continuous cost function where the slope is allowed to shift at nodes that separate the levels of production. This method is in contrast to the use of intercept dummy variables where each variable reflects a discrete shift in the whole function. The spline function is based on the assumption that the unit cost of generating income changes gradually as the bank moves to increasingly higher levels of income. The corresponding unit cost of production shifts at four different rates depending on the bank’s level of production.

The specification for the spline function model is

(3)

where SSF100 is the spline variable for the level of service-fee income if it is under $100,000; SSF300 is the spline variable for the level of service-fee income if it is under $300,000; STRAD1, STRAD2, and STRAD4 are the spline variables, respectively, for the level of traditional banking income if it is less than $1 million, less than $2 million, and less than $4 million.

The coefficients on SSF100, SSF300, STRAD1, STRAD2, and STRAD4 indicate how a bank’s cost structure shifts with every dollar of additional service-fee income and traditional income. For example, the model may estimate that the noninterest expense ratio for banks with $2 million in traditional income is 0.24% higher than for the average bank with over $4 million in traditional income. Since the model is linear and continuous, the excess noninterest expense gradually moves from 0% to 0.24% of assets as the level of traditional income falls to $2 million. The rate of change is allowed to change at the three nodes: $1 million, $2 million, and $3 million. The SF100 and SF300 variables have similar interpretations.

A common limitation of models (1)­(3) is that above-average performance of a particular product will not be identified in this type of aggregate analysis. In particular, we cannot separate out the effect of loan service fees from "high-touch" service fees. For this reason, (4) is specified to see how individual products affect bank profit margins. A set of binary variables is used to represent whether the following specific services are offered by the bank: automatic-teller machines, tele-banking, credit cards, annuities, mutual funds, insurance agency, brokerage services, and trust services. Positive coefficients on these variables indicate that they are decreasing the bank’s profit margin. The revised model in (4) tests for the individual effects of insurance, credit cards, ATMs, and discount brokerage services on bank profit margins. The other products that are identified are omitted, since they add little to the model and are closely related with the list of products which are included in (4). The full model is represented as

(4)

where ATM is a dummy variable indicating the availability of an ATM service, CARD is a dummy variable denoting use of credit cards, MFUND is a dummy variable for selling mutual funds, BROK is a dummy variable indicating the bank provides a discount brokerage service, and INSUR is a dummy variable signifying the bank has a full insurance agency.

Test Results <top>

The results from estimating models (1)­(3) are presented in Table 4, where the error terms are assumed to be independently and identically distributed after correcting for heteroskedasticity by dividing both sides of the equation by average asset size. Model (1) provides baseline estimates of the average pre-tax profit margins on the three product lines. Model (2) is a test for economies of scale and scope. Model (3) imposes continuity on the estimates of scale and scope economies so that any shift in the cost structure is assumed to occur gradually as the volume of business changes.

The estimated coefficient on the traditional banking income variable (TRBI) varies from 0.707 to 0.732 in the three models.¹¹ These results suggest that banks are spending 70­74¢ on average to generate $1 of traditional operating income. This cost figure includes both human resource and physical asset costs, but does not include the cost of acquiring equity financing. The positive margin on the traditional banking business (from .26­.30) can be interpreted as a gross pre-tax margin.

¹¹The estimated range for the coefficient is higher than the average efficiency ratios of banks. This is the result of subtracting loan losses from operating profits and including equity capital as a control variable in the regression model for the effect of higher levels of equity capitalization on the overall profit margins of the banks.