The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Financial Services WORK ING PAPER 0 1 9 6 Scale Economies, Cost Efficiencies, and Technological Change in Federal Reserve Payments by Paul W. Bauer and Gary D. Ferrier FEDERAL RESERVE FINANCIAL SERVICES • I Financial Services Working Paper Series No. 01-96 SCALE ECONOMIES, COST EFFICIENCIES, AND TECHNOLOGICAL CHANGE IN FEDERAL RESERVE PAYMENTS by Paul W. Bauer and Gary D. Ferrier Paul W. Bauer is a member of the Financial Services Research Group and an economic advisor at the Federal Reserve Bank of Cleveland. Gary D. Ferrier is an associate professor of economics at the University of Arkansas. The authors thank Sigb~rn Atle Berg, Allen Berger, Robert DeYoung, Karl Gambs, Diana Hancock, David Humphrey, Darrell Mak, Jeffrey Marquardt, Louise Roseman, Jeff Stehm, and Florence Young for their useful comments on earlier drafts of this paper. Professor Ferrier gratefully acknowledges the financial support of the Federal Reserve Bank of Cleveland. This paper was presented at the December 1995 Conference on Payment Systems Research and Public Policy, organized by the Board of Governors of the Federal Reserve System and the Journal of Money, Credit, and Banking. Working papers of the Financial Services Research Group are preliminary materials circulated to stimulate discussion and critical comment. The views stated herein are those of the authors and are not necessarily those of the Federal Reserve Bank of Cleveland, the Financial Services Policy Committee, or the Board of Governors of the Federal Reserve System. Working papers are now available electronically through the Cleveland Fed's home page on the World Wide Web: http://www.elev.frb.org. August 1996 • I Abstract This paper uses a stochastic cost frontier to examine the scale economies. cost efficiencies, and technological change of three payment instruments--check, automated clearinghouse (ACH) transfers. and Fedwire processing-provided by the Federal Reserve over the period 1990-94. We find evidence of substantial scale economies and cost inefficiencies in the ACH and Fedwire services. Check processing also exhibits substantial cost inefficiency, but constant returns to scale. Technological progress is found to be sizable for ACH and Fedwire; check processing is found to have experienced technological "regress," probably because of a decrease in processing volume over the sample period . I r Money is not, properly speaking, one of the subjects of commerce; but only the instrument which men have agreed upon to facilitate the exchange of one commodity for another. It is none of the wheels of trade: it is the oil which renders the motions of the wheels more smooth and easy. --David Hume, Essays Moral and Political, 1741 I. Introduction The payments system, the means of conducting transactions in an economy (Hume's "oil"), has undergone tremendous change over the centuries. Commodity money was replaced by fiat (usually paper) money, reducing transportation and storage costs. The invention of checks to supplement fiat money further reduced those costs. lessened the problem of theft, and provided a record of transactions. Most recently, the advent of electronic payment instructions has greatly reduced the time and handling costs associated with checks. Between the great evolutionary leaps forward that created new forms of payment instructions. smaller degrees of gradual evolution occurred within all the payment mechanisms, refining and improving them and reducing the costs associated with their use. The development of the payments system has indeed rendered the functioning of global commerce "more smooth and easy." An advanced economy has many payment instruments, each possessing different characteristics that make it suitable for some transactions but not others. For example, cash (currency and coins) is very convenient for lowvalue consumer purchases; however. few large companies would consider paying their employees in cash. Thus. while cash comprises about 80 percent of the volume of transactions in the U.S., it accounts for less than I percent of their value. Checks, automated clearinghouse (ACH) transfers, and wire transfers combined account for about 99 percent of the value transferred by th~ payments system (see Humphrey and Berger [1990]). Credit cards, point-of-sale, and automated teller machine bill payments are all experiencing rapid growth in volume, but have yet to attract a large share of transaction value. It is unlikely that "e-cash"--digital cash that will permit cybermai:kets to flourish--will account for a significant share for some time to come. 1 In this paper we examine the Federal Reserve's costs of processing three of the most important payment 1 Sec Humphrey. Pulley. and Vesala ( 1996) for infonnation on recent use panems for a variety of payment instruments for developed countries. i services: checks, ACH transfers, and wire transfers of funds (Fedwire). 2 We estimate three frontier cost systems that allow us to derive estimates of marginal cost, scale economies, cost efficiency, and technological change for each service. Each of these properties has important implications for the pricing, delivery, and market structure of these payment services. II. Overview of Check Processing, Automated Clearinghouses, and Fedwire Funds Transfer Services Before the Depository Institutions Deregulation and Monetary Control Act (MCA) of 1980 was passed. the Federal Reserve offered its payments services (check processing, ACH transactions, and Fedwire) at no charge to member banks. The MCA required the Federal Reserve to offer its payments services to all depository institutions. not just member banks, and directed it to begin charging for these services. 3 The Board of Governors has established guidelines for pricing payment services. Prices are set to recover all direct and indirect costs, including a markup over cost (the "Private Sector Adjustment Factor" [PSAF]) that reflects other costs (for example. taxes) incurred by private-sector providers of payments services. The prices of ACH and Fedwire funds transfer are determined nationally; however, because input prices, transportation requirements. and the mix of banks served vary from region to region. fees for check services vary substantially across Federal Reserve offices. While the passage of the MCA increased the number of banks eligible to use Federal Reserve payment services. a large decline in the volume of the Federal Reserve's check processing services occurred as the new pricing requirement made it easier for private providers of payment services to compete for member banks' business. 4 The following is a brief description of each of the Federal Reserve's payments services. It is intended to provide insight into the costs associated with these important services, but it does not reveal all of the complexities faced by a typical Federal Reserve processing site. For example. within each of the three services. transactions processed can be differentiated by the locations of the transmitting and receiving banks involved. the time available 2 1n llus paper. Fedwtre transactions refer lo fund transfers, not to book-entry Treasury security transfers. 3in the remainder of the paper, the term bank will be used lo refer to any depository institution. 4Jbough the Federal Reserve's diedc service volume declined, its ACH and Fedwire services have experienced steady rates of volume growth. -2- I for processing, and the amount of processing required by specific customers. Costs can vary significantly as a result of this myriad of product characteristics. While our analysis attempts to control differences across sites, some possibly important factors are no doubt missing. We hope that future research efforts will address any deficiencies. Check Processing Conceptually, the processing of checks is a straightforward operation. A payor writes a check to a payee, who deposits it at his/her bank. In the case of "on-us" items (when the payor and payee are customers of the same bank), the bank debits the payor's account and credits the payee's account. This situation represents about 25 percent of all transactions involving checks. If the payor and the payee have accounts at different banks, then the payee's bank must somehow present the check to the payor's bank. This type of settlement accounts for the remaining 75 percent of check transactions. In this case, the payee's bank has the option of sending the check directly to the payor's bank for payment, or it can employ the services of a local clearinghouse, a correspondent bank, or a Federal Reserve office to process the "transit" check. In 1994, Federal Reserve Banks processed approximately 35 to 40 percent of the transit checks-- approximately 17 billion of them. If the account on which a check is drawn has insufficient funds to cover it, the check is returned to the bank of first deposit directly or through one or more returning banks. This return process is more labor intensive, and thus more costly. than the forward processing of checks. Automated Clearinghouse Services The ACH system is a value-dated electronic funds transfer system that can be used to make either credit transfers or debit transfers. The five principal participants in ACH transactions are the payor, the payee. the payor's bank, the payee's bank, and the provider of the ACH service. With credit transfers (for example, direct payroll deposits), the payor's bank typically initiates the transfer and funds flow from the payor's bank to the payee's bank. With debit transfers (such as mortgage or utility payments), the payee's bank initiates the transfer and receives funds from the payor's bank. The Federal Reserve handled about 94 percent of the roughly 2.5 billion commercial and government ACH transactions processed in 1994. ACH transactions offer several key advantages over paper instruments. First, in most cases, payors know -3- I exactly when the funds will be removed from their accounts, and payees know exactly when the funds will be deposited to their accounts. Second, ACH transactions may be more convenient, particularly for recurring payments, because the payor need not remember to write and deliver a paper check, and the payee need not cash or deposit it. Third. the total costs to all parties are much lower for ACH transactions than for paper checks. 5 Finally, accounting efficiencies may exist for business payors and payees that have implemented financial electronic data interchange to facilitate communications with trading partners. 6 Fedwire Service The Fedwire funds transfer service is a real-time. gross settlement system in which the sender of funds initiates the transfer. Banks that maintain a reserve or clearing account with a Federal Reserve Bank may use Fedwire to send payments to or receive payments from other account holders directly. In contrast with ACH payments, which take two days to process, Fedwire is an immediate payment mechanism and is therefore used for time-critical payments. Fedwire transfers are used primarily for payments related to interbank overnight loans, interbank settlement transactions. payments between corporations. and settlement of securities transactions. Unlike check and ACH debit transactions, which can be returned unpaid, Fedwire transactions have the advantage that the funds transfer is final when credited to the receiving bank's Federal Reserve account, or when the Federal Reserve Bank sends advice of the payment to the bank, whichever comes first. When funds are transferred via Fedwire, Unifonn Commercial Code (UCC) 4A requires that they be made available to the recipient upon acceptance by the recipient's bank. UCC 4A also applies to corporate ACH credit transactions. While large-value transactions systems. like the Federal Reserve's Fedwire and the private Clearing House Interbank Payments System (Cl-llPS), accounted for only about 120 million transactions in 1994, as compared with 60 billion check transactions and over 2.5 billion ACH transactions (Bank for International Settlements ( 1995]), they accounted for most of the value of noncash transactions. In 1994, Fedwire and CI-IlPS transactions were valued at over $500 trillion, whereas the values of checks processed by the Federal Reserve and total ACH 1nc full social cost of processing an ACH item is only about one-third to one-half as much as for a check (see Humphrey and Berger [ 1990] and Wells [ 1994 ]). 6 Sec Knudson. Walton. and Young (1994) for a disrussion of the potential benefits of financial electronic data interchange (a combination of electroruc rerruuance data and electroruc funds transfers) for business payments. -4- i transactions were only $13 trillion and $10 trillion, respectively. The Fedwire service accounts for more than half of all large-value transactions, though it handles less than half of the dollar volume. III. Estimation Technique Frontier cost functions for each of the three payment services are estimated to measure their marginal costs, economies of scale, cost efficiencies, and rates of technological change. A variety of methodologies exist for calculating cost frontiers; each of them measures efficiency relative to a "best-practice" cost curve. In this paper 8 cost frontiers are estimated using a stochastic, parametric model.7· A stochastic form is chosen so that noise is less likely to be commingled with inefficiency; a parametric form so that estimates of the underlying technology's various properties, such as marginal costs and scale economies, can be derived. The estimation of cost frontiers requires data on cost, output quantities, input prices, and any environmental factors that might influence the level of costs. Let C, 1 be the level of observed cost incurred by the I (I= 1,. .. .N) processor in period t (t = l, .... n, y 11 be the vector of output quantities produced by the I th processor in period t, w, 1 be the vector of input prices facing processor I in period t, and E, 1 be a vector of characteristics describing the environment faced by the I th processor in period t. A stochastic, parametric frontier (log) cost function can be written as: lnC,, = lnC(y 11 ,w11 ,E 11 ,1;p) + v, 1 + u;,· (1) That is, the observed (log) cost, lnC,,, is the sum of frontier (log) cost, lnC 11 (y, 1 , w, 1 , £ 11 , t; P), random deviations from minimal cost, v11 , and deviations from minimal cost due to inefficiencies, u, 1 • The random disturbance may be positive, zero. or negative (v,, ~ 0), while the disturbance due to inefficiency is non-negative (u, 1 L 0), since inefficiency cannot cause cost to be less than the frontier level. Before estimating the cost frontier given by equation (l), two further modeling decisions must be made. First, a functional form must be chosen to represent the parametric relationship between cost, output, input prices, 7 Sec Chames et al. ( 1994) and Greene ( 1993), respectively. for overviews of the programming and econometric approaches to frontier estimation. Tur examples of frontiers applied to financial services, sec Aly et al. ( 1990), Bauer, Berger, and Humphrey (1993), Bauer and Hancoclc (1993), Berger (1993). Berger, Hancoclc, and Humphrey ( 1993). Berger and Humphrey (1991 ). Elyasiani and Mehdian (1990), Ferrier and Lovell (1990), Ferner et al. (1993). Fried. Lovell. and Vanden Eeckaut (1993), Mester (1993), and Rangan et al. (1988). -5- i and environmental variables. Because of its flexibility, we have chosen a hybrid cost function that combines the terms of the standard translog model and the first-, second-. and third-order trigonometric terms of the Fourier functional form. This hybrid cost function offers a close. global approximation to any underlying functional form (Gallant [1981] and Berger, Leusner, and Mingo (1995]). Second. the inefficiency term must be modeled. The availability of repeated observations over time (that is, panel data) alleviates the need to make an assumption about the particular distribution followed by the inefficiency term. Instead, we use the distribution-free "within" frontier model of Schmidt and Sickles (1984), modified for use with a cost function. This model identifies site-specific, time-invariant measures of inefficiency (that is, u,, = u,) based on observation-specific constants. The within model provides consistent estimates of the individual intercepts as T be consistently separated from the overall residual as N - 00 • 00 • and allows the individual inefficiency effect to Furthermore, it allows for correlation between the inefficiency terms and the regressors. Unfortunately. any variables that do not change over time must be excluded from the estimation, and the effects of these variables are included in the efficiency estimate. Given our assumptions, the cost function to be estimated can be written as: l L InC11 "' el + I: al lnyl,t L l K L Pk L K 1n w.b, + - k•l + I: J•l K P1k 1n W111 In w.b, M 6/k lnYt,, lnw.b, L + k■ I y m In£,,.,, m • I 3 + K L kL■ I 21.1 LL I• I a,k lnYi,, lny.b, /•l k•l /•! + L L L 2 + - N 4>1 QTR1 + I: e, PS, + }..YEAR (2) r•2 L + L [tlJ cos : 1 111 + w 1 sin: 111 ] I• I L L I• I k• I L L +LL [t,kcos(:111 +=.ti,)+ w 1*sin(: 111 + =.ti,>] L +LL L [w,m,cos(:,11 + z.b, + =m11> + w,m,sin(zl,r + z.b, + z,,.,,)] /•l k•l m•l + v,,, where C, y, w, and E are as defined above. QTR is a set of dummy variables to indicate the quarter in which an -6- i observation operates (one for each quarter except the first), PS is a set of dummy variables indicating which processing site is being observed (one for each site except the first), YEAR, which indicates the year of observation, is included as a proxy for technological change, and z is logged output mapped into the interval [0.1 ·21t, 0.9·21t] using a linear transformation (see Berger, Leusner, and Mingo [1995]). Note that the inefficiency portion of the "error" term (u) in equation (l) has been replaced by the site-specific dummy variables in equation (2). To improve the statistical efficiency of the estimates, equation (2) was estimated together with its corresponding input share equations (derived via Shephard's lemma). The usual linear homogeneity and symmetry restrictions were imposed prior to estimation. Note also that the translog functional form, a second-order local approximation, is nested within equation (2). If all of the 'ljJ and w coefficients are restricted to zero in equation (2), then the hybrid translog-Fourier model reduces to the standard translog model. These restrictions were tested in the empirical analysis reported below. Given that u, ? 0, the observation-specific constants can be normalized so that the processing site with the lowest intercept is deemed 100 percent efficient (that is, u, = 0) and serves as the benchmark against which other sites' efficiencies are assessed. This can be accomplished as follows: Lei 0 then u, min;(0,), 0; - 0, (3) i = 1. ... ,N. Since the estimated cost functions are in log form, the measures of cost efficiency, which range from 0 to I, are given by exp!-u,) = exp!-(0, - 0)) . IV. Data We estimated three cost function/cost share systems of equations. one for the operations of each payment system considered: check. ACH. and Fedwire. Each of the data sets used in our analysis consists of 20 quarterly -7- i observations over the years 1990-94. 9 Data on total costs, output volumes, input prices, and environmental variables are included for the 47 Federal Reserve check processing sites, 12 ACH sites, and 12 Fedwire sites. 10 The total number of observations is 931 for check processing, 232 for the ACH service, and 240 for Fedwire. 11 Though we used data from the individual check processing sites, the check results are aggregated to the District level in the discussion of our findings. The primary data source used was the annual functional cost accounting reports collected by the Federal Reserve's Planning and Control System (PACS). Since the purpose of PACS is to monitor costs and to improve resource allocation within the System, the reported data should be fairly accurate; however. some data errors are likely to be present. The use of a stochastic frontier should mitigate the effects of measurement error. The PACS data were supplemented with data from various Federal Reserve surveys, Bureau of Economic Analysis and Bureau of Labor Statistics price indexes. and pricing data from industry sources. 12 N.a Total Cost and Input Prices Total costs for all three payments services were proxied by their activity production costs. which include direct and support costs. but exclude imputed costs and certain overhead expenses. such as special District projects. The processing cost of each payment mechanism is composed of payments for four inputs--labor (L), materials (M), communications equipment and transit (7), and buildings (B). Table I reports the average share of total cost attributable to each input over the 20 quarters for each of the three payments services. 9 We chose this penod for several reasons. First. the data series are all complete for this period. Second. the Monetary Control Act of 1980 required !hat full-cost pncing be introduced for each of the Federal Reserve's payments services. For ACH, full-cost pricing was only gradually introduced and was not completed until 1985. By 1990 markets should have adjusted fully to MCA's full-cost pricmg requirement. Third, processing me consolidanon could cloud !he effects of scale econonues: relanve to !he 1980s. little consolidation of Federal Reserve processing sites occurred dunng our sample period. Fourth. Expedited Funds Availability (Title 6 of the Competitive Banking Equality Act [CEBAJ of 1987) may have changed !he technology of check processing. By 1990, the Federal Reserve could talce return items for which they hadn't handled !he forward processmg. Furthermore. all return items were to be in a new format that would allow increased automation of thetr processing. Finally. such dramanc technological changes have taken place that a smgle cost function may be unable 10 fit a longer sample period adequately. Consequently, we concentrate on !he most recent data available. 10 0ne processing S1le each for ACH and Fedwire services was excluded from the sample for reasons detailed below. 11 Not all processmg s11es were in operation for !he full 20-quarter sample period. One check-processing site ceased operations at the end of 1992 (eight quarters pnor to the end of the sample penod): another site operated for just one month of the fourth quarter of 1994 and was therefore dropped for !hat penod. Thus, nine "site-quarters" are missing from the check service data. One of !he ACH processing sites also ceased operations at the end of 1992. reducmg the number of ACH service observations by eight site-quarters. 12 Data construcnon parallels Bauer and Hancoclc. ( 1992). who provide details. -8- • i The price of labor (PJ was constructed as the sum of expenditures on labor (including salaries, retirement, and other benefits) divided by the number of employee processing hours. Based on cost shares, check processing is the most labor-intensive payments service, due largely to the paper-based nature of the service and. to a lesser extent, return items. ACH is much less labor-intensive than check processing. though some return items continue to be initiated manually. Because most banks initiate transfers electronically, Fedwire is the least laborintensive of the payments services. Due to the massive amounts of clearing data to track. all three payments services make heavy use of materials, which consist of computers and data processing, office equipment and supplies, printing and duplicating. and. in the case of check processing, check reader-sorters. The price of materials (PM) is given by a Tornquist approximation to a Divisia price index. It was constructed from the service prices of supplies, machines, and check reader-sorters. The service price of supplies (office equipment and supplies, and printing and duplicating) was represented by the implicit price deflator for gross domestic product (GDP). The service price of machines (computers and data processing) was constructed from cost-accounting expenditure data supplemented with the implicit price deflator for office. computing. and accounting machinery. To construct a price for data system support services (primarily used for in-house. product-specific software development), we utilized expenditures for labor and hours worked in that area of each Reserve Bank. Unlike prices for other computer hardware. those of check reader-sorters did not decline over the sample period. Therefore, a separate price index for check reader-sorters was constructed using historical data from industry sources. For computer hardware and check reader-sorters, an estimate of the service value. or price. of machines was constructed using a perpetual mvcntory model derived by Hall and Jorgenson (1967). 13 Communications and transit expenditures consist of the costs associated with data and other communications. shipping. and travel. Communications costs fonn the bulk of this cost category for ACH and Fedwire. Better than half of Fedwire's costs are associated with this input category. due to the communications equipment that is needed because most transfers are sent and received electronically. Check processing is the least 1 -'The Tomqmst mdex was constructed usmg the rates of growth m the pnces for each mput category. These rates of growth were weighted by the average propomonate shares of matenals expenses anribuiable 10 each category over adjomrng periods. -9- communications- and transit-intensive service. Most of its expenditures in this category are due to the costs of flying paper checks around the country. The implicit price deflator for communications equipment purchases by nonresidential producers was used for data and other communications expenses. The fixed-weight aircraft price index for private purchases of producers' durable equipment was employed for shipping and travel expenditures. The Tornquist approximation of a Divisia price index was calculated for transit (P 1 ), based on the expenditure shares of communications and shipping and their individual price indexes. Buildings have the smallest cost share among the four inputs, because the Federal Reserve does not finance buildings; thus, interest expenses associated with the acquisition of fixed assets are not present in the PACS's cost-accounting framework. Instead, interest costs are included in the PSAF used to set prices for Federal Reserve payments services. The share of buildings is greatest for check processing. owing to the bulkiness of check-sorting machines. The price of buildings (P8 ), measured as square-foot replacement costs adjusted by sitespecific depreciation rates, was constructed using cost accounting information from the PACS data and annual replacement-cost indexes available from Means (1995). !Vh Outputs and Environmental Variables Check Processing Check processing is a multi product operation consisting of forward items and return items. 14 Forward items <YFoR) are much more numerous than are return items (vRrr), 15 hut have a much lower per-item processing cost. A collection of other factors that may affect the cost of processing checks were included as elements of the environmental vector.£. The number of endpoints (EP). locations to which checks are delivered. was included in the cost equation together with its squared value. The item-pass ratio (IPR). the average number of times a check must pass through a reader-soner. is a proxy of the check-sort pattern. IPR is a function of the number of "pockets" on a reader-sorter. the number of banks for which a site processes checks. and the distribution of checks 1 '1'me-son ,terns (1.e .. checks that are fully prcsoned by hanks) were not included m the check processing output. nor were thetr production costs mclude<l in the cost m"-<lsure. 1 7ne ratio of forward Items to return Items handled by the Federal Reserve was approximately 70: I over the sample period. -IO- I across endpoints, as well as other factors. A dummy variable indicating whether IBM (IBM= 1) or Unisys (IBM= 0) machines were used at a site during each quarter was included to control for potential differences in maintenance expenses. failure rates, and down times. Some sites process forward and return items separately; other sites process them simultaneously. The latter method is likely more costly. A dummy variable was used to indicate whether a site intermingled the two items (NTRMNGLE =1) or processed them separately (NTRMNGLE = 0). Checks are processed at three different types of offices--District Banks, branches of District Banks. and regional check processing centers (RCPCs). Since costs are likely to vary across these settings, two dummy variables were included to control for office type. District Banks served as the reference group; RCPCs were indicated by RCPC = I (0 otherwise); branches of District Banks were indicated by BRANCH= I (0 otherwise). RCPCs have three potential cost advantages over District Banks: They were set up specifically to process checks. they are typically located outside of the central business district. and. because they do not handle securities or currency. their physical security costs are relatively low. Branches of District Banks are also likely to have cost advantages relative to the District Banks: Branches do not offer ACH or Fedwire services nor do they house monetary policy functions. Government checks are processed at just one site per district; a dummy variable. GO\'CK. was included to indicate those sites that process government checks. Site-specific figures that focus on transactions processed. rather than the number of payments, served as our measure of output for the ACH service (V,1rn), 1° The number of ACH processing sites fell over the sample period; hy 1993. only the 12 District Banks processed ACH items. During the period under study. the largest volumes were handled hy the 12 District Banks and the Los Angeles branch of the San Francisco Federal Reserve Bank. Thus. with the exception of the Los Angeles ACH site. we aggregate ACH data to the District level. Los Angeles 1s treated as a separate site because of its large volume of transactions. One of the 12 Districts was omitted from our empirical analysis because the hulk of its transactions were processed by a private provider of ACH services. 16 ACH pavment, m111a1ed and received at the same processing site are counted as only one transaction. Payments panially processed at one site. then 1ransm1ued to a sc:cond slle, are counted as transamons at both the transmining and receiving sites. Thus. the processing volume of ACH sites exceeds the actual number of ACH payments made hy the system as a whole. Note that with a single processing site, volume processed would equal the actual number of pavments transacted. -11- i The empirical analysis included three control variables to account for differences in ACH sites' processing environments. These sites have some discretion over processing schedules for government items, which may therefore be less costly to process than items processed for commercial customers. However, because government items are concentrated within certain relatively short periods each month, they may cause peak-load problems that would make them more costly than commercial items. In view of these considerations, the proportion of government items (PG[) processed was included. As was the case for check processing, the number of endpoints (EP) and endpoints squared (EP 2 ) were included. In this case, the number of endpoints refers to the number of banks or processors receiving ACH payments information. Finally, the proportion of banks receiving electronic payment information (PEER) was included. since nonelectronic delivery of information via computer tapes, diskettes, paper, and so on, increases transportation costs. 17 By contrast, increased use of electronic networks for infonnation delivery might give rise to greater scale efficiencies. Fedwire Output (_vFw) was measured as the total number of Fedwire funds transfers processed at each site. PACS contains infonnation on the number of transfers that originated within each district, the number of interdistrict originations. and the sum of total originations and interdistrict receipts. The number of intradistrict receipts could be derived from these three numhers. Total transfers processed for each site were defined as the sum of originations (interand intradistrict) and receipts (inter- and intradistricl), which reflects the number of reserve account entries ,L'isociated with Fedwire funds tranfers. This measure of output is consistent with the current pricing strategy employed for the Fedwire service. under which hoth sending and receiving banks are charged a fee for Fedwire transfers. One processing site employed a slightly different technology than the others (for example. it had a mainframe computer dedicated to Fedwire). Furthennore. the site's output was more than twice as large as any other processing site's. Because it wa-, the only site operating in the upper range of output, disentangling the effects of scale economies and cost efficiency was problematic. Therefore. the cost function for Fedwire was ,-'Over tune. ACH transacnons have migrated to vanous electroruc fonns (tapes. diskettes. and on-hne connections). As of July I. 1993, all cornmernal I non-federal government I ACH transactions were delivered electrorucally: as of July I. 1994. all federal government ACH transactions were also delivered electromcall~. -12- i estimated both with and without this site in the data set to determine its impact on our findings. Three environmental variables for each Fedwire site were included to control for differences across sites that might affect costs. Extension time (EX]). the number of extra minutes a site had to remain open to clear all of its daily transactions. may affect costs. 18• 19 The number ofFedwire customers is also likely to affect costs. Unfortunately. quarterly data on the number ofFedwire customers were not available for the full period of study. Instead, the number of accounts at each processing site was used as a proxy for the number of customers (CUSJ).']J.) Note that this variable is similar to the number of endpoints used in the analyses of check and ACH processing. The number of financial-institution accounts should be a good proxy for the number of Fedwire customers. since there was a 0.98 correlation between the two variables during periods for which data on both were available. Finally, the percentage of Fedwire transfers processed on-line (ONUNE) was included, since these transactions were probably less costly than those processed off-line. (Note the similarity with the control variable PEER used in the analysis of ACH transactions.) IV.c Other Variables Three final sets of variables used in the cost analysis were dummy variables to control for quarterly effects (QTR,, i =2.3.4 ), dummy variables for all but the first processing site (PS,, r =2, ... N), and a time trend (YEAR = 1,. .. ,5). The site-specific dummy variables were included to allow for the measurement of cost efficiency using the within frontier cost model. The time trend was included to capture the effects of technological change that may have occurred over time. Given the relatively short time period, a more elaborate model of technological change was not practical. i s1 ;n1ike the other environmental vanable. the actual value of EXT was used rather than its natural log. This was due to the fact that some sites used no ex1ens10n time. 1 'Extension lime is hkely to be needed to clear transactions on peak activity days. However, extension time also may be an endogenous measure of processmg-sne effie1ency: A less efficient site may need extension lune 10 settle the same number of 1ransacttons that an efficient site could settle dunng 11s normal hours of operauon. ~ile the number of customers setved by each processmg site will probably affect costs, the composition of each site's customer base is likely to he more 1llummatmg. For example. the number of customers by connection type (dial-up, leased-line, multi-drop, etc.). the proponion of low volume customers. or the geographic dispersion of customers are all factors that future research might consider. -13- i V. Results The parameter estimates and their t-statistics for the hybrid translog-Fourier frontier cost functions for the check and ACH services appear in tables 2 and 3, respectively. Though the nested translog fonns of the hybrid translogFourier cost frontiers were rejected for all three payments services,21 our Fedwire results (tables 4a and 4b) are based on the nested translog frontier cost function. For the check and ACH services, the hybrid translog-Fourier models fit the data well. most signs being as expected, and most parameter estimates being statistically significant. However, the hybrid translog-Fourier results for Fedwire violated monotonicity: Predicted costs fell as output increased in some ranges. In effect, the hybrid model was too flexible a functional form. as it allowed a violation of a fundamental property of the cost function given by economic theory. Since we choose economic theory over statistical considerations. the (nested) translog is our preferred model for Fedwire. Furthermore. due to the strong influence of one particular processing site (PS2) on the Fedwire models, two sets of results are reported. The first set (table 4a) includes all 12 processing sites; the second set (table 4b) excludes site PS2. V.a Unit Cost, Marginal Cost, and Scale Economies We are interested in the unit cost (based on activity production costs), mar~inal cost. and scale economies associated with each of the three payment services offered by the Federal Reserve. The scale economies are represented hy the cost elasticity. which is defined as the percentage increase in cost for a 1 percent increase in (one of the) output(s): cllnC(w.v.£.1:~) oln Y, (4) and. in the case of check processing. the ray cost elasticity. which is given hy the following expression: 11 For all three payments services. the null hypothesis that all of the sine and cosine coefficients were equal to zero was firmly rejected. For check processmg. F(3662. I 8) = 14.57 (Prob> F: 0.000 I): for ACH transactions. F(899 ,6) = 4.24 (Prob> F: 0.0003 ); for Fedwire (with PS2). F(884.6) = 8.69 (Prob> F: ().()()OJ). Full translog results are available from the authors upon request. -14- i 2 11 E 11;- (5) ,• I The unit costs, marginal costs, and cost elasticities and their 95 percent confidence intervals for check, ACH, and Fedwire services are given in tables 5, 6, and 7. respectively; table 5 also shows the ray cost elasticity for check processing. For comparison purposes. tables 5 and 6 contain the results for both the hybrid translog-Fourier cost systems and the nested translog models for the check and ACH services. Table 7 contains the translog results for Fedwire. both with and without PS2. The weighted average cost of processing a check across the 47 sites is just under 2 cents per check. 22 The marginal cost of forward items. about 1.4 cents on average. is substantially less than that of return items, 38.6 cents on average. corroborating the view that return items are much more costly to process than forward items. 23 The returns to scale for the joint provision of forward and return items (ray cost elasticity) averages .97 for the 47 processing sites. implying that a one percent increase in both outputs would increase costs by just .97 percent. The confidence intervals for the cost elasticity suggest that 12 processing sites have insufficient volume to achieve scale efficiency and that three operate with scale diseconomies: the remaining 32 sites are characterized by constant returns to scale. indicating that they operated very near the scale efficient rate of output. Figure 1 plots the average cost curves based on the unit costs at the sample means of the check processing sites for both the hybrid translogFourier and translog models. Note that the two plots are quite similar. The figure reveals that unit cost drops dramatically a,; output initially rises. but quickly flattens out. 14 Our findings on scale economies for Federal Reserve check processing services differ from those of ~ecall that check processing 1s a multi product activny--both forward and return items are processe<i. However, the share of forward items is more than 98 percent: therefore. umt cost is calculate<i on the basis of forward items only. This simplification should have a negligible effect on the measure of umt cost. ¾e hybnd translog-Founer cost function results exhibit a few "local" violations of monotonicity; that is, some point estimates of marginal cost are negauve. How,-ver, these violations only occur for very small processing sites. Overall. the hybrid translog-Fourier results are very similar to those of the neste<i translog model. These problems are not as severe as for Fedwtre, where there were "nonlocal" (occurring over a broad range of output) v10la11ons. ;,,,~ote the inh,-rent wav,- forms embedded in the hybrid translog-Founer functional form as a result of the sine and cosine terms. This is the source of the local v10lauons of mono1omc11y for the smallest processing sites. -15- i Humphrey (1980, 1981a, 1984, 1985). which were based on a single-output model for earlier time periods, and from those of Bauer and Hancock (1993), who estimated a multiproduct cost function over the period 1979-90. It appears that constant returns to scale, rather than increasing returns. characterize the operation of all but the smallest. and perhaps the very largest, processing sites. The difference between our results and those of previous studies may be partly due to our choice of functional form. First, because the hybrid translog-Fourier model allows for a "flatter" function. the range of constant returns to scale may be broader than for the functional forms used in earlier studies. Second, the standard-error terms for some of the hybrid translog-Fourier's coefficient are large. so statistically the range of constant returns to scale is broader. The hybrid translog-Fourier and the (nested) translog point estimates of cost elasticity were roughly similar. However, the standard errors of the Fourier estimates were much larger than those of the translog estimates. making it less likely that a null hypothesis of constant returns to scale would be rejected. The weighted average of the unit cost of an ACH is 1.7 cents for the 12 Federal Reserve processing sites in our sample: marginal cost averages less than a penny (0.9 cents) per transaction. The mean level of the cost elasticity for ACH is .48. indicating that the Federal Reserve's ACH service is characterized by increasing returns to scale. Plots of the hybrid translog-Fourier and translog average cost curves for ACH appear in figure 2. which also reveals that increa<;ing returns exist for most of the observed output levels for ACH services provided by the Federal Reserve. While the average cost curve for the hybrid translog-Fourier model appears to rise at the highest level of output. this finding is ba<;ed on just one processing site whose output is significantly higher than that of any other site. Interestingly. Humphrey (1981h. 1982. 1984. 1985) found roughly the same magnitude of economies of scale for Federal Reserve ACH services a, we did. even though the volume of output in our sample period is much greater than in his. The ACH results are also broadly consistent with those of Bauer and Hancock ( 1995a). The weighted average unit cost of a Fedwire transaction is 24.2 cents for the 12 processing sites in the data set. As mentioned above. site PS2 exerts a strong influence over the estimation results. The average marginal cost is 26.5 cents when PS2 is included. but just 20.4 cents when it is excluded. For the full sample. the cost elasticity averages l. 14: the two smallest Fedwire sites are found to experience scale economies. the largest site -16- I (PS2) experienced scale diseconomies (with a cost elasticity of 1.70); all other sites appear to operate under constant returns to scale. With PS2 excluded from the analysis, the cost elasticity averages just 0.79. This cost elasticity indicates that the 11 remaining sites experienced scale economies, though none of the cost elasticities are statistically significantly different from 1 (constant returns to scale). The biggest differences in the cost elasticities across the two sets of estimates are for the smallest and largest sites; for the medium-sized sites, all the results are roughly similar. The average cost curves associated with the two translog models estimated for Fedwire. one with and one without PS2. appear in figure 3. which also contains a scatter plot of_the raw data. The figure conveys the influence of PS2. which is represented by the cluster of data points at the highest output levels. The right half of the average cost curve is based solely on PS2. making it unlikely that scale economies and cost efficiency have been correctly disentangled for PS2. If PS2 is cost efficient. then diseconomies of scale exist for the highest output levels: however. if PS2 is cost inefficient. then it is likely that constant returns to scale--or possibly even scale economies--exist throughout the observed range of output levels. Intuitively. the latter possibility is more appealing. Like ACH. Fedwire would seem to be the type of service that would offer scale economies. since both are highly dependent on communications and computers. resources for which unit costs are likely to decline as volume increases. Our Fed wire results are similar to those of Humphrey (1982. 1984 ). which seemed to suggest relatively slight scale economies in the Fed wire services. Humphrey ( 1982. 1984) found that 98 percent of all Fedwire transfers took place in offices with constant returns to scale. However. he employed only cross-sectional data and included the largest processing site in his model. Given the large jump in volume between the largest and nextlargest site. his model. like ours. would have had trouble differentiating scale economies and cost efficiency. Without PS2. we find ahout the same level of scale economies for Fedwire as did Bauer and Hancock (1995b). who employed data from 1988 to 1992. \'.h Em·ironmemal Variables Each of the estimated frontier cost systems contains a number of environmental variables to control for differences -17- i in the operating characteristics across processing sites. All of the environmental variables have statistically significant effects on the cost of check processing. The number of endpoints (EP and EP 2) is found to have a quadratic relationship with cost. Following an inverted-U shape. cost initially rises as the number of endpoints increases. but eventually falls as the number of endpoints increases. Predictably, cost increases with the number of times a check must pass through a reader-sorter (given by the item-pass ratio [IPR]). The processing sites that use IBM equipment (IBM= I) had higher costs than those using Unisys equipment. However, this finding may result from differences in geography rather than differences in machinery. Most of_ the population of the U.S. is located in Districts that use Unisys equipment: most of the area of the U.S. is located in Districts that use IBM equipment. Thus. IBM may be serving as an indirect proxy for transportation costs. The type of site where checks are processed also has a statistically significant effect on cost: as a group. both RCPCs (RCPC = I) and branches of District Banks (BRANCH= 1) have lower costs than did the District Banks. The intermingling of forward and return items is found to be more costly than the separate processing of the two. Finally, sites that processed government checks appear to have higher costs than those that did not. This finding warrants further study. It may suggest that there are diseconomies of scope between the processing of government checks and private checks: however. it could also result from an accounting anomaly. For ACH transactions. the coefficient on the proportion of government items (PG/) processed is positive. hut is not statistically significant. This suggests that either the benefit of having some discretion regarding when to pnx:ess these items halances with any peak-load problems that they may create. or else that neither of these 2 considerations has any noticeable effect on cost. The nurnhcr of endpoints (EP and EP ) has a negative relationship with cost--the larger the nurnhcr of endpoints. the lower the cost of carrying out ACH transactions. The final environmental variable. the proportion of hanks receiving their payment information electronically (PEER). docs not have a statistically significant effect on cost. possibly hecause all banks were required to receive data electronically by the end of the sample period. leaving relatively little variation in this variable. For Fedwire services. there is no statistical evidence that any of the three environmental variables (CUST. EXT. ONUNE) affects processing costs. The lack of significance for ONUNE probably results from lack of variability in the proportion of transfers processed on line; during the sample period. nearly all transfers were -18- i processed on line. V.c Cost Efficiency Site-specific and summary statistics for the three payments services' cost efficiency scores are reported in table 8 for all models. The average levels of cost efficiency for the check and ACH services are 68.1 percent and 59.4 percent, respectively, for the hybrid translog-Fourier frontier. As discussed above. the largest Fedwire site has a strong influence on the estimated scale economies and cost efficiencies associated with this service. The average cost efficiencies of the Fedwire service are 58.9 percent when PS2 is included, and 66.0 percent when it is excluded from the analysis. The measures of cost efficiency indicate the proportion of observed cost that would have been expended had all the sites operated on the best-practice cost frontier defined by the site with an efficiency score of l. At constant input prices. and given the linear homogeneity of the cost function. the efficiency measures also may be interpreted as the proportion of observed input quantities needed to produce the observed level of output relative to the best-practice performance. Comparing the consequences of unrealized scale economies and the presence of cost inefficiencies, our findings for check processing and Fedwire are similar to that of Ferrier and Lovell ( 1990) and Berger and Humphrey ( 1991) with regard to banks: The effects of cost inefficiencies dominate those of scale economies. This is especially true for check processing where scale economies have been exhausted. Fed wire. on the other hand. is characterized by substantial scale economies as well as substantial cost inefficiency. For ACH. we find that the effect of scale economies is slightly greater than that of cost inefficiencies, though both are substantial. The relatively low levels of average efficiency indicate a great deal of dispersion in the ability of processing sites to convert their inputs into services. The inefficiency may result from principal-agent problems. Since there is no market for corporate control to "discipline" managers at Federal Reserve processing sites. monitoring costs will be higher at the Federal Reserve than at publicly traded corporations. Furthermore, the Federal Reserve lacks two important monitoring devices available to publicly traded firms (Puttennan [ 1993]). First. for publicly traded firms. the value of tradeable shares serves as an indicator of incumbent managers' perfonnance. Second. interest rates that private firms pay to finance expenditures indicate a project's perfonnance -19- I or prospects. Since the Federal Reserve does not borrow funds to finance the purchase of buildings. this signal is not available to assess managerial performance. Alternatively, the high degree of market concentration (less for check processing than for the ACH and Fedwire services) and attendant market power within the markets for these payment services may reduce the competitive pressure on processing sites to perform as efficiently as possible. Instead of realizing higher "profits," the processing sites may indulge in the "quiet life" that market power affords (Hicks [1935]). Berger and Hannan (1994) found evidence of the "quiet life" in commercial banking. concluding that the efficiency cost of market concentration was several times greater than the social cost as measured by the welfare triangle. However. mergers and changes in technology give the Reserve Banks increasing competition from private providers of payment services. Finally. despite the inclusion of several environmental variables. heterogeneity across processing sites may not have been adequately modeled. The measured "inefficiency" in this case would reflect operating/environmental differences across sites. not true inefficiency. As Stigler (1976) noted. measured inefficiency may result from not incorporating the right variables or the right constraints in the analysis, or from failing to consider the correct economic objective of the organization under analysis. Table 9 reports the rank correlation coefficients for the efficiency scores of the check, ACH, and Fedwire operations. The efficiency rankings based on the two different models estimated for each of the three payments services are all highly correlated. However. the efficiency rankings across the three payments services are all negatively correlated. though not with statistical significance. Thus, while the operating performances of processing sites are widely scattered. there is no evidence that the efficient operation of one payment service is related to the efficiency with which the other services are provided. Interestingly. the use of IBM. as opposed to Unisys. check processing equipment is associated with higher costs based on the estimated parameters of the hybrid translog-Fourier frontier cost system. Indeed the mean level of cost efficiency is 9.2 percent higher for IBM-based processing sites. This suggests that the cost disadvantage of using IBM equipment is partially offset by other factors. \'.d Techrwlogical Change The rate of technological change (more properly. the rate of cost diminution) is modeled by including a time trend -20- i in the frontier cost functions estimated for each payment service. Thus, disembodied technological change, which manifests itself by shifting the cost function, is implicitly assumed. The technological change coefficients for all three payments services are shown in table 10. Note that the results across the two models estimated for each payment service are very similar. The cost of ACH transactions diminished at an average rate of about 11 percent per year over the 1990-94 period. Fedwire also enjoyed technological gains during this time, as its cost diminished at an annual rate of about 6 percent. Check processing, however, appears to have suffered the fate of technological "regress." as costs rose about 1.7 percent per year over the period. A number of factors offer likely explanations. First. and probably most important, processing volume declined at many sites over the sample period; cost reductions will almost certainly lag volume declines, since sites require time to shed their fixed costs. Second. the "quality" of checks processed declined over the period, because a high proportion of "high quality" check items (such as payroll checks and Social Security items) migrated to ACH. The remaining, "lower-quality" items are likely to involve greater processing costs. Third. the quality of service provided by Federal Reserve processing sites increased over the sample period. For example. magnetic ink character recognition (MICR) information services were added. Because the output measures are not adjusted for quality. the observed regress may reflect costs associated with capital-intensive quality improvements. Fourth. Federal Reserve Districts were changing to the new Funds 5.0 application during 1994 and 1995. ~ About half of the Districts made the conversion in 1994, the last year of our study. Costs were probably increased by conversion efforts and experimentation with the new technology to determine how to use the available advances most effectively (in other words. sites had not moved very far along the learning curve). 16 The lack of a me,L'>ured change in output to accompany the cost increase associated with the new technology would result in the appeara11ce of regress. \ .e Decompositio11 The unit cost for each payment service varies substantially across processing sites (see the second column of tables 15 In addt110n. the \1mneapohs Federal Reserve 1mplemen1ed imaging software in 1994. ""An altemauve model thal employed dummy vanables to indicate the year of operation rather than the variable YEAR= 1.2.3.4,5, produced the following results 1990. O; 199 I. -4.5: 1992. 0: 1993.2.9; I 994. 6.1. Thus. technolog1cal "advance" is observed in 1991. but "decline" is observed in 1993 and 1994. These results lend suppon to our explanauons for the technological "regress" found for the full sample penod. -21- i 11, 12, 13a, and 13b). Unit cost is a useful summary measure of operating performance and is thus interesting in its own right. However. the sources of variation in unit cost may be of even greater interest. An advantage of the functional forms used is that differences in unit cost can be traced back to their "source" by examining logarithmic differences between a site's unit cost and the unit cost at the mean of the sample data (see Bauer [1993]). 27 There are six potential sources for cost differences in our analysis. First, differences in cost efficiency occur across sites. Some sites operate on the best-practice cost frontier, while others do not. Ceteris paribus, sites that lie above the cost frontier will have higher unit costs. Second, scale economies may account for cost differences. Other things being equal, a processing site that is too small to fully exploit scale economies will suffer a cost disadvantage. Third. sites may face disparate input prices. Those with higher input prices will have higher unit costs, other factors held constant. Fourth, a site's environment may be more or less advantageous compared to another site's. Holding other influences constant. sites with a more hospitable processing environment will experience lower unit costs. Fifth. there is a residual category that comprises all of the interactive terms of the hybrid translog-Fourier cost function. Fortunately. these "interaction effects" account for less than I percent of any of the unit cost differentials. Finally. there are random effects. To discern the sources of unit cost variation. we form the (log) ratio of a site's mean cost over the sample peri{xl to the cost at the mean of the sample data: l (C, / 5',) In - - <t I 5' l I = lI IC(5',. "·,. E, )·exp(E,) In - - - - - - - - - In 5', C(v.· "'· E)·exp(E) l . (6) _v where the terms with the subscript i represent site-specific means and the other terms represent overall sample means. Using the cost function defined in equation (2). the percentage difference between the mean unit cost of a processing site and the unit cost at the mean of the data can be written as: 28 7 Loganthms have the property that, close to zero. they can be interpreted roughly as percentages. For example, a logarithmic difference of 0.1 converts to roughly a I fl percent difference. For the exact percentage. you would need to calrulate ( I - exp! 0.1 ) ). : 1o simplify notation, the tngonometnc Founer terms do not appear in (7). They were. however, included in the empmcal decompositions reported helm• -22- I l i + + + [aY(lny, - lny) :E K [ + ½a:YY{(lny, Pk(lnwki - lnw1 ) + k•I [f [t - lny 2 ) - (lny 1 -lny)}] .!.:E :E P1/lnwkilnwj, 2 K K - lnwklnw) l k•l J•I t\(lny;lnwb - lnylnw1 k•I + 2 (7) )l , Ym(ln£m, - ln£m)l m•l + [ E, - E] where the hracketed terms on the right-hand side of equation (7) are the efficiency effects. scale effects. input price effects. interaction effects (between processing volumes and input prices), environmental effects, and random dfects. respectively. The decomposition of cost differences into the first five sources is reported in tables 11. 12, and 13 (a and h). for the check. ACH. and Fedwire operations. respectively. For check processing. cos! efficiency appears to be the largest single factor in explaining unit cost differences. hut scale economies. input prices. and the operating environment also account for some large cost difkrcnccs (sec tahle 11 ). In general. these results are very similar to those reported by Bauer ( 1993), who used data for 1983-90. Our higgest departure from Bauer (1993) is that environmental effects play a larger role for more pnx:essing sites in our findings. For example. the coefficient on the IBM indicator variable rose from Bauer's es11mate of 0.0925 to 0.205: thus. Unisys sites appear to have a significant cost advantage. Cost efficiency and output effects drive the results for ACH services (see tahle 12). The existence of largescale economies places smaller processing sites at a significant cost disadvantage. However, it is worth noting that the logarithmic differences for the cost efficiency and output effects usually take opposite signs. implying that the less scale-efficient sites make up for this disadvantage hy being more cost efficient. and vice versa. Given that data processing inputs. which are priced nationally. account for ahout 75 percent of the costs of this service, the processing sites show relatively little difference on this score. -23- i Two sets of results are reported for Fedwire, one which includes PS2 in the analysis (table 13a) and one which excludes it (table 13b). The first set of results indicates that PS2's superior cost efficiency appears to offset its large disadvantages of scale. Recall, however, that the relative effects of scale economies and cost efficiency on PS2's observed costs is suspect, given that PS2 is the sole source of data for the upper third of the observed range of output levels. In general, the results for Fedwire are similar to those for ACH services, though environmental factors appear to account for a bit more of the observed unit cost differences with Fedwire. As with ACH services, the penalty for failing to fully exploit scale economies can be large. The input price effect accounts for a very small share: again, this is not surprising, given that data processing inputs account for more than 85 percent of Fedwire costs. VI. Conclusions We estimate cost functions for the check. ACH, and Fedwire services provided by the Federal Reserve to derive estimates of marginal costs. scale economies. cost efficiency. and technological change over the pericxi 1990-94. There are wide differences in performance across processing sites. even after controlling for volume. input prices. and various environmental variables. Scale economies appear to have been exhausted for all but the 12 smallest check processing sites, indicating that some additional consolidation may be in order. particularly if volume declines are projected to continue. However. one cannot decide whether it would be wise to reallocate volume among processing sites. or even to close some sites. hy looking at costs alone: one must also consider the demand for check processing services. For example. hy locating closer to their customers. processing sites can offer higher quality service. receiving checks later and delivering them earlier. Closing a processing site could lead to lower quality service and loss of volume. Customers may be more than willing to pay the higher processing costs at these suboptimally sized sites. Of course. the move towards electronic check presentment and imaging make the processing site's location much less important. All of the ACH processing sites were found to have statistically significant scale economies. which appears to justify the Reserve Banks' plan to consolidate to one processing site with one backup -24- • I site. The decision to consolidate the processing of all but the largest Fedwire site receives some support from our findings, as there may be significant scale economies throughout the relevant range of output. There appears to be a great deal of dispersion in the operating performances of the various processing sites for all three payments services considered in this paper. This suggests that the costs of providing these services could be reduced considerably if all sites were to move to the best-practice frontier. No single site's performance dominated across services. In fact, there appears to be no tendency for a site's cost efficiency in one service to spill over to other services. The electronic services (ACH and Fedwire) have both experienced rapid technological change over the last five years. while check processing costs have risen over that period. The first finding is consistent with the rapid decline in the price of computer and communication equipment. key inputs into electronic services. Check processing, on the other hand. is largely dependent on labor and the speed at which paper checks can be read through check reader-sorters. neither of which has benefited much from the productivity improvements caused by recent technological change. Clearly. more empirical research is needed on how new technologies affect the efficiency of the payments system. For example. scope economies among the various services. particularly between ACH and Fedwire transfers. could also be important in determining the scale efficiency and optimal product mix for payment-service providers. Such scope economies could enable many more suppliers to operate efficiently and to reduce the real resource costs associated with processing payments. Finally. in order to construct a pricing mechanism that enwurages efficiency in the payments system and still recovers costs. the demand side of the payment-service market--including cross-elasticities between payment instruments--needs to be more fully understood. The emergence of new payment instruments and nationwide branching will greatly increase the competitive pressure on the Federal Reserve Banks. and they will need to compete both on quality and price in providing payment services. -25- I References Aly, Hassan Y .. Richard L. Grabowski, Carl A. Pasurka, Jr., and Nanda Rangan. "Technical, Scale, and Allocative Efficiencies in U.S. Banking: An Empirical Investigation." Review of Economics and Statistics 72 (May 1990), 211-218. Bank for International Settlements. Payment Systems and the Group of Ten Countries. Basel, Switzerland: BIS, December 1995. Bauer. Paul W. "Efficiency and Technical Progress in Check Processing." &onomic Review, Federal Reserve Bank of Cleveland 29 (Quarter 3. 1993). 24-38. _ _ _ _ . Allen N. Berger. and David B. Humphrey. "Efficiency and Productivity Growth in U.S. Banking, in The Measurement of Productive Efficiency: Techniques and Applications, edited by Harold 0. Fried, C.A. Knox Lovell. and Shelton S. Schmidt. pp. 386-413. Oxford: Oxford University Press. 1993. _ _ _ _ . and Diana Hancock. "The Efficiency of Federal Reserve Check Processing Facilities." Finance and Economics Discussion Series (FEDS) no. 214, Board of Governors of the Federal Reserve System (1992). _ _ _ _ .and _ _ _ _ . "The Efficiency of Federal Reserve Check Processing Facilities." Journal of Ba11ki11R and Fi11ance 17 (April 1993). 287-330. _ _ _ _ . and _ _ _ _ . "Scale Economies and Technological Change in Federal Reserve ACH Payment Prtx:essing." Eco11omic Review. Federal Reserve Bank of Cleveland 3 I (Quarter 3. 1995). 14-29 (a). _ _ _ _ . and _ _ _ _ . "Estimation of the Fed wire Funds Transfer Cost Function." Mimeo (I 995) (b). Berger. Allen N. "'Distrihution-Free' Estimates of Efficiency in the U.S. Banking Industry and Tests of the Standard Distrihutional Assumptions." Journal of Productirity Analysis 4 (September I 993). 261-292. _ _ _ _ . Diana Hancock. and David B. Humphrey. "Bank Efficiency Derived from the Profit Function." Journal of BankinR and Fina11ce 17 (April 1993). 317-348. _ _ _ _ . and Timothy M. Hannan. "The Efficiency Cost of Market Power in the Banking Industry: A Test of the 'Quiet Life' and Related Hypotheses." Board of Governors of the Federal Reserve System, Finance and Economics Discussion Series: 94-36 (November 1994 ). _ _ _ _ . and David B. Humphrey. "The Dominance of Inefficiencies Over Scale and Product Mix Economies -26- • i in Banking." Journal of Monetary Economics 28 (August 1991), 117-148. _ _ _ _ , John H. Leusner, and John J. Mingo. "The Efficiency of Branch Banks and Implications for Mergers and Interstate Banking." 1n Proceedings: A Conference on Bank Structure and Competition, pp. 332348. Chicago: Federal Reserve Bank of Chicago, 1995. Charnes, Abraham, William Cooper, Arie Y. Lewin, and Lawrence M. Seiford. "Basic DEA Models." 1n Data Envelopment Analysis: Theory, Methodology and Applications, edited by Abraham Chames, William Cooper, Arie Y. Lewin, and Lawrence M. Seiford, pp. 23-47. Boston: Kluwer Academic Publishers. 1994. Elyasiani. Elyas, and Seyed M. Mehdian. "A Non-Parametric Approach to Measurement of Efficiency and Technological Change: The Case of Large U.S. Commercial Banks." Journal of Financial Services Research 4 (June 1990). 157-168. Ferrier. Gary D.. Shawna Grosskopf, Kathy J. Hayes. and Suthathip Yaisawarng. "Economies of Diversification in the Banking Industry: A Frontier Approach." Journal of Monetary Economics 31 (April 1993). 229-249. _ _ _ _ . and C.A. Knox Lovell. "Measuring Cost Efficiency in Banking: Econometric and Linear Programming Evidence." Journal of Econometrics 46 (October/November 1990), 229-245. Fried. Harold 0 .. C.A. Knox Lovell. and Philippe Vanden Eeckaut. "Evaluating the Performance of U.S. Credit Unions." Journal of Rankin[? and Finance 17 (April 1993). 251-266. Gallant. A. Ronald. "On the Bias in Flexihle Functional Forms and an Essentially Unbiased Form: The Fourier Flexihle Fonn." Journal of Econometrics 15 (June 1981 ). pp. 211-245. Greene. William H. "The Econometric Approach to Efficiency Analysis." ln The Measurement of Productive Efficiencv: Techniques and Applications. edited by Harold 0. Fried. C.A. Knox Lovell and Shelton S. Schmidt. pp.68-119. Oxford: Oxford University Press. 1993. Hall. Robert E.. and Dale W. Jorgenson. "Tax Policy and lnvestmenl Behavior." American Economic Review 57 (June 1967). 391-414. Hicks. John R. "The Theory of Monopoly: A Survey." Econometrica 3 (January 1935). 1-20. Humphrey. David B. "Are There Economies of Scale in Check Processing at the Federal Reserve?" Journal of -27- I Bank Research 11 (Spring 1980), 8-19. _ _ _ _ . "Economies to Scale in Federal Reserve Check Processing Operations." Journal of Econometrics 15 (January 1981), 155-173 (a). _ _ _ _ . "Scale Economies at Automated Clearinghouses." Journal of Bank Research 12 (Summer 1981), 7181 (b). _ _ _ _ . "Costs. Scale Economies. Competition. and Product Mix in the U.S. Payments Mechanism." Staff Studies, Board of Governors of the Federal Reserve System (April 1982). _ _ _ _ . "The U.S. Payments System: Costs. Pricing, Competition and Risk." Monograph Series in Finance and Economics, Board of Governors of the Federal Reserve System (1984). _____ . "Resource Use in Federal Reserve Check and ACH Operations after Pricing." Journal of Bank Research 12 (Spring 1985), 45-53. _ _ _ _ , and Allen N. Berger. "Market Failure and Resource Use: Economic Incentives to Use Different Payment Instruments." In The U.S. Pavment System: Efficiency, Risk and the Role of the Federal Resen·e. edited by D. B. Humphrey, pp. 45-86. Boston: Kluwer Academic Publishers, 1990. _ _ _ _ . Lawrence B. Pulley, and Jukka M. Vesala. "Cash. Paper, and Electronic Payments: A CrossCountry Comparison." Journal of Money. Credit. and Banking 28 (November 1996), forthcoming. Knudson, S.E .. J.K. Walton. and F.M. Young. "Business-to-Business Payments and the Role of Financial Data Electronic Interchange." Federal Resen1e Bulletin. Board of Governors of the Federal Reserve System (1994). 269-278. Means. R.S. Means Square Foot Costs, 15th edition. Kingston. MA: R.S. Means Company. 1995. Mester. Loretta J. "Efficiency in the Savings and Loan Industry." Journal of Banking and Finance 17 (April 1993). 267-286. Puuerman. Louis G. "Ownership and the Nature of the Finn." Journal of Comparative Economics 11 (June 1993 ). pp. 243-263. Rangan, Nanda. Richard L. Grabowski. Hassan Y. Aly, and Carl A. Pasurka. Jr. "The Technical Efficiency of U.S. Banks." Economics Lellers 28 (1988). 169-175. -28- i Schmidt, Peter J., and Robin C. Sickles. "Production Frontiers and Panel Data." Journal of Business and Economic Statistics 2 (October 1984), 367-374. Stigler, George J. "The Xistence of X-Efficiency." American Economic Review 66 (March 1976), pp. 213-216. Wells. Kirsten E. The Social Costs of Paper and Electronic Payments. Master of Arts thesis, Virginia Polytechnic Institute, 1994. -29- Table 1: Cost Shares for Payment Services Check ACH Fedwire Labor 0.493 0.213 0.139 Materials 0.275 0.408 0.320 Communication and Transit 0.170 0.354 0.523 Building 0.062 0.025 0.019 Source: Authors' calcularions. I Table 2: Hybrid Parameter Estimates for Check Processing Intercept lnJFOR (hlYFoR) 2 myFoR lnwL IJtyFOR lnWM lnYFoR lnws lnYFOR lnw,IJtJFOR lnyRET lnyRET (hlyRET)2 lnyRET lnWL IJtyRET lnWM lnyRET lnws ln y RET ln W,COS(ZFoR) sin(ZFoR) cos(2zFoR) sin(2zFoR) cos(3zFoR) sin(3zFoR) cos(zRET) sin(zRET) cos(2zRET) sin(2zRET) cos(3zRET) sin(3zRET) COS(ZFOR+ZRET) sin( ZFoR+zRET) cos(2zFOR+zRET) sin(2ZFoR+ZRET) COS(ZFoR+ 2zRET) sin(ZFOR+2zRET) Year Qi Q1 Q4 IJtwL lnwM lnw8 hlWr ht/PR IBM lnEP Parameter Estimate 14.878 -1.262 -3.934 -0.042 0.004 -0.011 0.049 -0.120 0.432 1.201 0.071 -0.022 0.010 ,-statistic 19.88 -2.63 -2.71 -6.63 0.65 -6.89 12.35 -2.13 LSO -0.060 1.83 14.27 -4.18 7.56 -18.78 2.166 -0.501 0.480 -0.068 0.183 0.124 -0.935 -0.314 -0.152 -0.072 -0.118 -0.072 -0.098 -0.l 15 -0.143 -0.237 0.146 0.112 0.017 -0.017 -0.029 -0.054 0.505 0.271 0.063 0.161 0.157 0.205 0.091 -3.77 3.61 -1.34 5.14 5.00 -1.49 -3.17 -1.31 -1.82 -3.65 -2.91 -2.18 -2.53 -3.40 -4.85 3.74 2.28 6.56 -2.15 -3.51 -6.60 248.74 126.47 119.82 127.60 3.70 6.10 7.66 3.04 i Table 2: Hybrid Parameter Estimates for Check Processing (continued) (ln£P)2 (lnwL) 2 lnWtlnWM lnWtlnW9 lnwtlnWr (lnwM) 2 lnw.,lnw9 lnwMlnWr (lnws>2 lnw9lnWr (lnWr)2 PS2 PS3 PS4 PSs PS6 PS1 PSs PS9 PSw PS11 PS12 BRANCH RCPC t-.'TRMNGLE GOVCK Parameter Estimate -0.061 0.085 0.024 0.013 -0.122 -0.003 0.004 -0.025 0.001 -0.018 0.165 -0.086 -0.229 -0.223 -0.077 -0.457 -0.252 -0.475 -0.136 -0.246 -0.673 -0.394 -0.055 -0.088 0.122 0.034 Source: Authors' calculations. r-statistic -9.96 7.53 3.37 3.12 -12.46 -0.38 .1.65 -4.86 0.19 -3.42 13.66 -2.43 -4.51 -10.74 -4.25 -19.36 -9.94 -10.62 -2.85 -5.48 -12.50 -10.10 -4.11 -5.50 5.26 2.77 i Table 3: Hybrid Parameter Estimates for ACH Processing lnw.wlnwa Parameter Estimate 12.670 0.450 4.677 -0.130 -0.070 0.010 0.010 0.210 0.400 0.030 0.360 -0.010 -0.030 0.000 0.030 -1.170 -0.070 -0.180 -0.010 0.000 0.000 -0.010 0.020 -0.010 -0.010 0.020 0.()00 In wMln »'r (lnwB)" -0.()40 0.()1() lnwBlnwr 0.010 0.()40 Intercept JnJACH (lnYACH) 2 ln.EP (ln.EP)2 PEER PG/ lnwl IDWM IDWB lnw-r IDYACH IDWL lnyAcH lnwM lnyACH lnwB ffiYACH illl-l'f cos(:Arn) sin(:Arn) cos(2:Arn) sin(2:AcH) cos(3:Arn) sin(3:Arn) <mwd lnwLlnW.w lnwLlnwe IDWLID»'r (lnwM)" (lntt'r}" Q2 Q_, Q4 YEAR PS 3 PS4 PSs PS6 PS? PS8 PSQ PSw 0.010 0.000 0.000 -0.l 00 -0.410 0.030 0.260 0.418 0.474 0.000 0.033 0.149 t-statistic 11.65 1.39 0.96 -2.26 -1.05 0.23 0.06 65.37 84.79 49.07 59.21 -0.78 -2.08 0.72 2.06 -0.92 -0.36 -0.79 -0.20 0.08 -0.04 -0.68 2.56 -8.51 -0.71 1.00 -1.47 -1.62 1.89 1.83 1.32 0.39 -0.06 -0.08 -4.10 -7.71 0.35 2.52 3.81 3.73 0.01 0.57 1.33 Table 3: Hybrid Parameter Estimates for ACH Processing (continued) Parameter Estimate PS11 PS12 PS13 Source: Authors' calculations. t-statistic 0.223 3.27 0.418 -0.124 4.16 -2.27 i Table 4a: Translog Parameter Estimates for Fedwire (with PS2) Intercept lnyFED (lnYFED) 2 lnwL lnwM JnwB inWr lnYFED InwL lnyFED lnwM lnyFED lnwB lnyFED lnw.,<mwd JnwLJnWM JnwLJnWB JnwLinWr (JnwM)2 JnwMlnWB JnwMlnWr (JnwB) 2 lnw8 lnw7 <mwd YEAR lnONUNE lnCUST EXT PS2 PS 3 PS4 PSs PS 0 PS7 PSx PSo 1.219 0.837 0.549 0.137 0.334 0.021 0.509 -0.029 0.029 0.006 -0.006 0.095 0.127 -0.002 -0.220 0.306 -0.013 -0.421 0.007 0.008 0.633 -0.056 2.494 0.144 0.000 -0.798 -0.202 -0.571 -0.211 -0. l 05 -0.171 -0.173 -0.041 0.12 3.33 2.76 58.28 47.03 39.85 64.99 -6.25 2.45 5.85 -0.46 5.58 4.84 -0.53 -6.08 3.31 -1.81 -4.07 1.77 0.84 5.03 -4.09 1.13 0.66 1.52 -1.88 -1.66 -6.04 -1.75 -0.70 -0.68 -0.63 -0.13 -1.00 0.76 PS12 -0.226 0.136 -0.461 Q2 Q, Q4 -0.043 -0.063 -0.072 -1.83 -2.49 -2.75 PS1t, PS 11 Source: Authors· calculations. -1.91 i Table 4b: Translog Parameter Estimates for Fedwire (without PS2) Intercept lnyFED 2 (ITIYFED) lnwl lnwM lnw8 lnw-,lnYFED lnwL lnYFED lnwM lnyFED lnwn lnyFED lnw,(tnwd lnwLlnWM lnwllnwn lnwLlnw,(lnwM) 2 lnwMlnwn lnwMlnwr (lnw8 ) 2 lnwelnw,(lnw,-)2 YEAR lnONUNE lnCUST EXT PS3 PS4 I'Ss PS0 PS7 PSR PS4 PS1n PS11 PS12 Q2 Q, Q. Source: Authors· calculations. 1.145 0.758 0.151 0.139 0.320 0.019 0.523 -0.029 -0.025 0.001 0.052 0.094 0.107 -0.004 -0.197 0.421 -0.009 -0.518 0.003 0.011 0.704 -0.064 2.501 0.134 0.0002 -0.209 -0.574 -0.224 -0.095 -0.()45 -0.083 0.(149 -0.230 0.121 -0.262 -0.()45 -0.065 -o.mm 0.11 2.79 0.53 53.95 44.62 39.50 66.49 -4.66 -1.55 1.13 2.84 5.22 3.78 -1.08 -5.18 4.40 -1.43 -4.90 0.81 1.33 5.50 -4.46 l.10 0.59 l.68 -1.66 -5.88 -1.79 -0.61 -0.17 -0.29 0.15 -0.98 0.65 -0.99 -l.76 -2.40 -2.87 Table 5: Estimates of Unit C'osts. Marginal C'osts, :md C'ost Elasticities for Check Translog Hyhrid Processing Site SI S2 S3 S4 S5 S6 S7 S8 S9 SIO S 11 Sl2 Sl3 Sl4 S15 Sl6 S17 Sl8 Sl9 S20 S21 S22 S23 S24 S25 S26 S27 S28 S29 S30 S31 S32 Unit Cost 0.015 0.020 0.013 0.026 0.021 0.020 0.019 0.030 0.019 0.019 0.022 0.014 0.022 0.021 0.021 0.016 0.019 0.024 0.022 0.018 0.017 0.013 0.028 0.022 0.014 0.016 0.028 0.ol8 0.021 0.014 0.016 0.021 Marginal Cost Return Forward 0.0 IO 0.013 0.009 0.014 0.()()9 0.019 0.013 0.014 0.012 0.016 0.018 0.{J09 0.012 0.013 0.019 0.0 I 5 0.011 -0.0 I 3 0.024 0.010 0.010 0.009 0.026 0.014 0.014 0.013 0.020 0.007 0.005 0.006 0.008 0.019 0.565 0.515 0.257 0.618 -0.056 -0.597 0.153 0.615 0.402 0.261 0.183 0.491 0.719 0.596 0.192 0.542 0.264 0.550 -0.239 0.352 0.224 0.486 0.207 0.290 0.149 -0.085 0.243 -0.044 -0.047 0.258 0.010 0.681 Cost Elasticity Upper Lower Overall l.lfd 1.060 0.903 0.891 0.399 0.696 0.811 0.877 0.968 0.979 0.937 1.()4) 1.065 1.178 1.054 1.237 0.777 -0.276 1.()49 1.014 0.826 1.159 1.030 0.915 1.065 0.779 0.970 0.316 0.188 0.662 0.510 1.165 1.529 1.192 1.235 1.038 0.556 0.958 1.020 1.130 1.067 1.236 1.211 1.173 1.213 1.350 1.147 1.510 0.936 0.554 1.576 1.386 1.007 1.557 1.146 1.172 1.370 0.973 1.544 0.911 0.815 0.890 0.814 1.772 II. 797 0.927 0.571 0.744 0.242 0.435 0.601 0.623 0.869 0.722 0.662 0.909 0.916 1.007 0.961 0.964 0.618 -1.107 0.522 0.641 0.646 0.760 0.914 0.658 0.760 0.586 0.396 -0.278 -0.440 0.435 0.206 0.557 Marginal Cost Forward Return 0.013 0.016 0.010 0.021 0.013 0.012 0.014 0.025 0.013 0.014 0.014 0.010 0.018 0.017 0.016 0.011 0.015 0.012 0.013 0.013 0.012 0.01 I 0.021 0.015 0.007 0.009 0.024 0.010 0.012 0.011 0.01 I O.ot8 0.393 0.394 0.232 0.562 0.062 -0.152 0.369 0.672 0.216 0.347 0.164 0.139 0.477 0.418 0.363 0.182 0.396 -0.280 -0.645 0.218 0.201 0.359 0.516 0.273 -0.306 0.007 0.463 0.082 0.092 0.315 0.081 0.679 Overall 1.171 1.123 0.947 l.106 0.654 0.540 l.010 1.298 0.856 0.958 0.754 0.800 1.152 1.222 1.021 0.797 1.062 0.352 0.441 0.999 0.896 1.169 0.993 0.946 0.395 0.608 1.322 0.689 0.685 1.078 0.709 1.132 Cost Elasticity Upper l.251 1.167 0.990 1.155 0.714 0.610 l.050 1.376 0.891 0.996 0.830 0.831 1.208 l.282 1.051 0.823 1.108 0.501 0.509 · 1.040 0.939 1.251 1.025 0.985 0.492 0.691 1.383 0.801 0.799 1.138 0.741 1.229 Lower 1.092 1.078 0.904 1.056 0.594 0.469 0.969 1.220 0.822 0.921 0.679 0.769 1.096 1.163 0.990 0.771 1.016 0.203 0.372 0.959 0.853 1.088 0.962 0.907 0.299 0.525 1.262 0.577 0.571 1.017 0.676 1.036 Table 5: Estimates of Unit Costs. Marginal Costs, and Cost Elastil:ities for Check (continued) Hyhrid Marginal Cost Return Forward Translog Cost Elasticity Lower Upper Overall Marginal Cost Return Forward Overall Cost Elasticity Upper Lower Processing Site S.B S34 S35 S36 S37 S38 S39 S40 S41 S42 S43 S44 S45 S46 S47 Unit Cosl 0.013 0.014 0.022 0.022 0.024 0.017 0.017 0.021 0.019 0.017 0.027 0.015 0.023 0.016 0.020 0.014 0.0 IO 0.015 0.016 0.013 0.012 0.008 0.02 I 0.015 0.0IO 0.021 0.0 IO 0.014 0.013 0.020 0.434 0.234 0.203 0.(147 0.932 0.215 -0.030 0.200 0.161 0.188 0.572 0.417 0.254 0.593 0.330 1.361 ().867 0.847 0.767 1.075 0.871 0.474 1.134 0.923 0.775 1.218 1.007 0.802 1.05 3 1.182 1.609 1.174 1.090 0.929 1.423 1.090 0.629 1.278 1.162 0.985 1.382 1.108 1.076 1.502 1.386 1.111 0.561 0.604 0.604 0.727 0.652 0.319 0.990 0.684 0.566 1.054 0.905 0.528 0.604 0.978 0.010 0.011 0.014 0.013 0.021 0.011 O.OIO 0.016 0.012 0.011 0.021 0.011 0.016 0.012 0.015 0.198 0.280 0.186 0.065 0.656 0.111 0.055 0.351 0.096 0.161 0.442 0.194 0.239 0.237 0.333 0.879 0.966 0.805 0.647 1.242 0.728 0.653 0.955 0.682 0.853 1.095 0.848 0.878 0.864 0.929 0.911 1.009 0.870 0.723 1.330 0.800 0.716 0.984 0.767 0.906 1.130 0.881 0.932 0.910 0.961 0.846 0.924 0.740 0.570 1.154 0.655 0.590 0.926 0.597 0.799 1.061 0.815 0.824 0.818 0.897 Weighted Average 0.020 0.014 0.385 0.966 1.242 0.691 0.015 0.339 1.003 1.060 0.946 Source: Authors' calculations. Table 6: Estimates of Unit Costs. Marginal Costs, and Cost Elasticities for ACH Hyhrid Processing Site PSI PS3 PS4 PSS PS6 PS7 PS8 PS9 PSIO PSII PSl2 Unit Cost 0.020 0.015 o.ot5 0.016 0.017 0.019 0.023 0.019 0.016 0.022 0.020 Marginal Cost 0.0117 11.0088 0.0066 0.0109 II.Oil I O.(KJl9 0.0184 0.0053 0.0113 0.1Kl50 0.0142 Upper 0.024 0.019 0.014 0.017 0.019 0.010 O.o35 0.012 0.019 0.018 0.024 Lower -0.001 -0.001 -0.001 0.005 0.003 -0.006 0.002 -0.002 0.004 -0.008 0.004 Cost Elasticity 0.571 0.584 0.442 0.684 0.650 0.101 0.785 0.281 0.720 0.225 0.724 Marginal Cost 0.0094 0.0068 0.0072 0.0077 0.0084 0.0094 0.0106 0.0086 · 0.0076 0.0104 0.0095 Weighted Average 0.017 0.009 0.017 0.000 0.478 0.0082 Source: Authors' calculations. Translog Upper Lower O.Dl5 0.004 0.003 O.Oll 0.004 0.010 0,004 0.011 0.005 0.012 0.014 0.005 0.004 0.017 0.004 0.013 0.011 0.004 0.005 0.016 0.005 0.014 O.Qll 0.004 I Cost Elasticity 0.460 0.454 0.480 0.485 0.493 0.502 0.452 0.458 0.485 0.465 0.483 0.403 Table 7: Estimates of Unit Costs, Marginal Costs, and Cost Elasticities for Fedwire Without PS2 With PS2 Processing Site Unit Cost Marginal Cost Upper Lower Cost Elasticity Marginal Cost Upper Lower Cost Elasticity 0.773 na 0.720 0.730 0.748 0.809 0.863 0.653 0.643 0.745 0.766 0.883 0.793 PSI PS2 PS3 PS4 PSS PS6 PS7 PS8 PS9 PSI0 PSI I PS12 0.228 0.206 0.229 0.206 0.278 0.268 (1.280 0.321 0.357 0.252 0.3()4 0.220 0.189 CU51 0.143 0.135 0.201 0.251 0.320 0.116 0.119 0.176 0.240 0.272 0.283 0.488 0.249 0.229 0.321 0.358 0.438 0.303 0.333 0.288 0.367 0.370 0.096 0.214 0.037 0.(l42 0.080 0.144 0.202 -0.071 -0.095 0.065 0.112 0.174 0.830 1.703 0.626 0.657 0.723 0.939 1.141 0.361 0.334 0.699 0.788 1.240 0.176 na 0.165 0.150 0.208 0.217 0.242 0.210 0.230 0.188 0.233 0.194 0.274 na 0.276 0.250 0.339 0.337 0.366 0.382 0.465 0.361 0.372 0.290 0.078 na 0.054 0.050 0.076 0.096 0.118 0.038 -0.006 0.015 0.094 0.098 Weighted Average 0.242 0.265 0.387 0.143 1.144 0.204 0.329 0.079 Source: Authors' calculations. Table 8: Cost Efficiency by Service A'CH Check Translog 0.699 Processing Site PSI PS2 PS3 PS4 PSS PS6 PS7 PS8 PS9 PSIO PSI I PS12 Hybrid 0.510 0.556 0.641 0.637 0.551 0.806 0.656 0.820 0.584 0.652 1.000 0.756 Translog 0.598 0.628 0.762 0.745 0.622 0.936 0.767 0.974 0.662 0.748 1.000 0.855 Hybrid 0.665 na na 1.000 0.648 0.513 0.438 0.414 0.665 0.643 0.573 0.532 0.438 1.000 0.639 0.505 0.445 0.416 0.632 0.644 0.555 0.514 0.442 Mean Median Standard Dev. 0.681 0.647 0.141 0.775 0.755 0.139 0.594 0.573 0.165 0.590 0.555 0.166 Source: Authors' calculations. Fedwire With PS2 Without PS2 0.563 0.450 1.000 na 0.551 0.694 0.797 1.000 0.556 0.704 0.500 0.619 0.589 0.535 0.536 0.612 0.469 0.536 0.709 0.565 0.499 0.393 0.714 0.731 0.589 0.543 0.170 0.660 0.619 0.136 Table 9: Rank Correlation of the Cost Efficiency Estimates for the Three Services (significance levels in parentheses) ACH Check Payment Service Hybrid Translog Check, Hybrid 1.000 (0.000) 0.960 (0.000) -0.260 (0.440) -0.313 (0.348) -0.202 (0.552) -0.263 (0.435) Check, Translog 0.960 (0.000) 1.000 (0.000) -0.211 (0.534) -0.266 (0.429) -0.095 (0.782) -0.179 (0.599) ACH, Hybrid -0.260 (0.440) -0.211 (0.534) 1.000 (0.000) 0.995 (0.000) -0.024 (0.945) 0.130 (0.703) ACH. Translog -0.313 (0.348) -0.266 (0.429) 0.995 (0.000) 1.000 (0.000) -0.036 (0.916) 0.115 (0.737) Fedwire. with PS2 -0.202 (0.552) -0.095 (0.782) -0.024 (0.945) -0.036 (0.916) 1.000 (0.000) 0.928 (0.000) Fedwire, without PS2 -0.263 (0.435) -0.179 (0.599) 0.130 (0.703) 0.115 (0.737) 0.928 (0.000) 1.000 (0.000) Source: Authors· calculations. Translog Fedwire WithPS2 Without PS2 Hybrid Table 10: Technological Change in Payment Service Provision Translog Hybrid Payment Service Time ,-statistic Time ,-statistic Check 0.017 6.56 0.018 6.67 ACH -0.112 -6.35 -0.110 -4.66 WithoutPS2 WithPS2 Payment Service Time I-statistic Time I-statistic Fedwire -0.056 -4.09 -0.064 -4.46 Source: Authors' calculations. ■ Table 11: Check Processing Unit Cost Decomposition, Hybrid Cost Function (Processing Site Means Relative to Overall Sample Means, 1990-1994) Processin Site SI S2 S3 S4 S5 S6 S7 S8 S9 S10 S 11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21 S22 S21 S24 S25 S26 S27 S28 S29 S30 S31 S32 S33 S34 S35 S36 S37 S38 S39 S40 S41 S42 S43 Unit Cost ($) 0.015 0.020 0.013 0.026 0.021 0.020 0.019 0.030 0.019 0.019 0.022 0.014 0.022 0.021 0.021 0.016 0.019 0.024 0.022 0.oI8 0.017 0.013 0.028 0.022 0.014 0.016 0.028 0.018 0.021 0.014 0.016 0J)21 0.013 0.014 0.022 0.022 0.024 0.017 0.017 0.021 0.019 0.017 0.027 Unit Cost -0.247 0.066 -0.382 0.318 0.094 0.063 -0.023 0.441 -0.022 -0.008 0.165 -0.316 0.166 0.096 0.112 -0.204 0.009 0.227 0.142 -0.061 -0.115 -0.383 0.394 0.133 -0.333 -0.198 0.398 -0.065 0.093 -0.341 -0.147 0.114 -0.349 -0.282 0.134 0.128 0.250 -0.102 -0.136 ().()90 0.oI5 -0.130 U.346 Logarithmic Differences from Sample Mean Effect of Effect of Direct Total Cost In ut Price Interactions Environment Efficienc Ou ut 0.048 -0.002 0.049 -0.124 -0.166 -0.159 0.001 0.029 0.013 0.214 -0.085 0.005 -0.038 -0.088 -0.166 -0.099 0.000 0.222 -0.048 0.291 -0.047 -0.003 0.019 -0.027 0.205 -0.164 -0.004 -0.104 0.113 0.214 -0.158 -0.043 -0.060 0.000 0.214 0.075 0.005 0.100 0.113 0.039 -0.108 -0.047 -0.068 0.000 0.068 -0.055 0.001 -0.036 -0.082 0.068 -0.182 -0.003 -0.045 0.059 0.214 -0.257 0.001 -0.087 -0.155 0.068 -0.096 -0.001 0.147 -0.004 0.205 0.380 0.000 0.023 0.111 -0.383 0.178 0.000 -0.010 -0.022 0.045 -0.100 -0.008 -0.134 -0.136 0.039 -0.001 -0.056 0.024 -0.061 0.039 0.084 -0.006 -0.103 0.619 -0.383 -0.015 0.010 -0.191 0.146 0.155 0.259 -0.001 -0.026 0.083 -0.383 -0.114 -0.001 -0.136 -0.001 0.039 -0.121 0.000 0.033 -0.148 -0.166 0.138 -0.001 0.234 -0.053 0.205 0.261 0.()45 0.001 0.059 0.006 -0.796 0.006 0.145 0.138 0.291 0.()49 0.139 -0.099 0.004 -0.184 0.151 0.269 0.003 0.080 -0.103 0.168 -0.001 -0.042 0.129 -0.184 0.105 -0.002 -0.089 0.167 -0.184 -0.092 -0.016 -0.131 0.000 -0.166 0.027 -0.002 -0.087 -0.200 0.039 0.193 -0.241 0.000 0.072 0.155 -0.074 -0.012 -0.123 0.005 -0.166 -0.071 0.014 0.000 -0.077 -0.166 0.187 -0.001 -0.073 0.049 0.045 0.179 -0.003 -0.105 0.030 0.045 0.219 -0.002 -0.027 -0.049 0.062 -0.155 -0.001 0.024 -0.030 0.068 -0.050 -0.005 0.023 -0.009 -0.103 -0.032 -0.022 -0.051 0.214 0.000 0.012 -0.002 -0.005 0.053 -0.103 0.290 -0.001 -0.072 0.026 -0.383 0.209 0.002 0.128 0.061 -0.103 i Table 11: Check Processing Unit Cost Decomposition, Hybrid Cost Function (continued) ProcessinJ?: Site S44 S45 S46 S47 Unit Cost ($) 0.D15 0.023 0.016 0.020 Unit Cost -0.213 0.199 -0.170 0.033 Standard Dev. 0.004 0.219 Source: Authors' calculations. Logarithmic Differences from Sample Mean Effect of Direct Effect of Total Cost Price Input Environment Output Interactions Efficiency -0.003 -0.103 0.061 -0.035 -0.092 0.000 0.017 0.270 0.041 -0.184 0.009 0.057 -0.210 -0.174 0.205 0.000 0.291 0.140 -0.239 -0.075 0.191 0.140 0.091 0.003 0.199 i Table 12: ACH Unit Cost Decomposition, Hybrid Cost Function (Processing Site Means Relative to Overall Sample Means, 1990-1994) Processing Site PSI PS3 PS4 PS5 Unit Cost ($) 0.0204 0.0150 Unit Cost 0.112 PS12 0.0160 0.0171 0.0188 0.0234 0.0188 0.0157 0.0224 0.0196 -0.195 -0.203 -0.135 -0.066 0.028 0.248 0.030 -0.149 0.203 0.072 Standard Dev. 0.0029 0.157 PS6 PS7 PSS PS9 PSl0 PSll 0.0149 Source: Authors· calculations. Logarithmic Differences from Sample Mean Environmental Effect of Direct Total Cost Effect Efficiency Input Price Interactions Output -0.001 0.058 0.191 -0.122 0.003 0.076 0.001 0.192 0.084 -0.530 0.023 -0.037 -0.079 0.000 -0.096 -0.001 -0.123 -0.087 0.000 0.137 -0.049 0.001 -0.019 -0.212 0.296 -0.059 -0.366 0.005 0.000 0.352 -0.001 -0.006 -0.007 0.295 -0.122 -0.007 0.002 0.101 -0.089 0.072 -0.054 -0.002 0.000 -0.140 0.026 -0.004 0.071 0.006 0.000 0.100 -0.027 -0.001 -0.185 0.296 0.072 0.255 0.205 0.056 0.044 0.001 SO.OS W..()4 Tr ■ mlog SO.OJ 11Hl2. Hybrid 10.0] so l In mill Ioli, rt1 Ll~n So,1r~: A1,1rhl.}fS' cak11latiu115. 100 lSU 200 Forward I t~:m~ h1,no.af'd pmpnrUon:111lh• '111'1 ti!! torun.l i!!ffl.:!I j ~!~ms inl"r~a~ 1:'5(i ■ Table 13a: Fedwire Unit Cost Decomposition, with PS2 (Processing Site Mean Relative to Overall Sample Means, 1990-1994) Office PSI PS2 PS3 PS4 PSS PS6 PS7 PSS PS9 PSIO PSl 1 PS12 Standard Dev. Logarithmic Differences from Sample Mean Environmental Effect of Direct Total Cost Interactions Effect Unit Cost ($) Unit Cost Efficiency Output Input Price 0.001 -0.070 -0.122 0.036 0.235 -0.126 0.228 -0.009 0.040 0.042 0.311 -0.562 -0.227 0.206 0.001 -0.129 0.004 -0.020 -0.123 0.033 0.229 0.001 -0.061 -0.005 -0.039 -0.335 -0.228 0.206 0.001 -0.013 -0.006 -0.077 0.025 0.072 0.278 0.001 0.031 -0.027 -0.144 0.131 0.035 0.268 -0.001 0.080 -0.007 -0.128 0.064 0.081 0.280 0.002 -0.035 -0.048 0.223 0.062 0.217 0.321 -0.003 0.001 -0.009 0.263 0.194 0.324 0.357 0.001 0.056 -0.036 -0.065 0.009 -0.024 0.252 0.001 0.034 -0.01 l -0.106 0.371 0.163 0.304 0.004 0.066 0.066 -0.096 -0.226 -0.163 0.220 0.048 0.179 0.259 0.165 0.033 0.063 0.003 Source: Authors· calculations. Table 13b: Fedwirc Unit Cost Decomposition. without PS2 (Processing Site Mean Relative to Overall Sample Means. 1990-1994) Office PSI PS3 PS4 PS.S PS6 PS7 PS8 PS9 PSI0 PSI I PSI2 Standard Dev. Logarithmic Differences from Sample Mean Effect of Environmental Direct Total Cost Interactions Effect Unit Cost (S) Unit Cost Efficiency Output Input Price -0.()64 0.000 0.040 -0.050 0.141 -0.147 0.228 0.002 0.(l44 -0.118 0.008 -0.068 -0.144 0.229 0.001 -0.052 -0.001 o.cno -0.433 -0.249 0.206 0.001 -0.009 -0.002 -0.002 -0.083 0.052 0.278 0.002 0.039 -0.023 -0.090 0.046 0.014 0.268 0.000 0.078 -0.002 -0.152 0.096 0.060 0.280 -0.002 -0.032 -0.043 0.196 0.058 0.196 0.321 -0.001 0.005 -0.006 0.217 0.191 0.303 0.357 0.001 0.053 0.0<)9 -0.()45 -0.032 -0.089 0.252 0.001 0.032 0.3()4 -0.031 -0.007 0.262 0.143 -0.004 0.067 0.069 -0.171 -0.121 -0.183 0.220 0.()47 Source: Authors· calculations. 0.172 0.190 0.123 0.031 0.061 0.002 Figure 2: ACH Average Cost 0.0200 0.0180 0.0160 0.0140 0.0120 0.0100 0.0080 0.0060 Translog 0.0040 0.0020 0.0000 0 20,000 40,000 60,000 80,000 100,000 Items Processed (millions) Source: Authors' calculations. 120,000 140,000 160,000 Figure 3: Fedwire Average Cost 0.9 0.8 • • 0.7 Unit Cost - - - - Without PS2 -- 0.6 0.5 With PS2 0.4 ♦. 0.3 _)~- - 0.2 0.1 0.0 5,000 Items Processed (millions) Source: Authors' calculations. 10,000 15,000