You are on page 1of 15

Economic Review 2000 Q3

Optimal Use of Scale Economies
in the Federal Reserve’s
Currency Infrastructure
by Paul W. Bauer, Apostolos Burnetas, Viswanath CVSA,
and Gregory Reynolds

Paul W. Bauer is an economic advisor at
the Federal Reserve Bank of Cleveland,
Apostolos Burnetas is an associate
professor of operations research at
Case Western Reserve University’s
Weatherhead School of Management,
Viswanath CVSA is an associate at
McKinsey & Company, and Gregory
Reynolds is an actuarial consultant at
Hewitt Associates.

Introduction interest payments generated by the backing of
these notes is indirectly remitted to the Treasury.2
The Federal Reserve has been providing Ultimately, the interaction of payment-
paper currency since its founding in 1913. In instrument providers, payors, payees, and
fact, the major purpose for creating the Federal regulators will determine various instruments’
Reserve System was to furnish an “elastic market shares, but currency will probably con-
currency” that would make financial crises, such tinue to retain a significant share of transactions.
as the panic of 1907, less severe. While much The Federal Reserve is responsible for oper-
has changed in the financial landscape over the ating its service efficiently. Even if the market
last 87 years, currency remains an integral part were not evolving as a result of new payment
of the U.S. payments system, accounting for instruments, the advent of nationwide branch-
over 80 percent of all transactions.1 (For a brief ing of private depository institutions would
history of the Federal Reserve’s role in currency
processing, see Good and Mitchell [1999].)
For many types of transactions, currency has
withstood the onslaught of checking accounts
and credit cards, and will likely prevail over
debit and smart cards for some types of trans-
actions in the foreseeable future. The reasons
■ 1 Of course, cash accounts for a much smaller share of the value
for this are simple: Currency offers finality, exchanged in trade. Electronic payment services, such as the Clearing House
anonymity, and familiarity at a reasonably low Interbank Payment System (CHIPS), Fedwire, and the Automated Clearing
cost for small-value transactions. This sets a House (ACH), now account for over 99 percent of the value exchanged in
high hurdle for competing payment instruments. trade. See Humphrey, Pulley, and Vesala (1996) for more details.
This is good news for the Treasury because the
■ 2 Anywhere from one-half to two-thirds of U.S. currency is held
20 billion Federal Reserve notes in circulation overseas. Foreign holdings generally involve the higher denominations,
(with a value of $460 billion) are backed by mostly $100 bills (see Porter and Judson [1996]).
Treasury debt. The $20 billion in annual

By adopting this on priced services that have to recover their full economic costs through constraint. the building space previously allocated to it and possibly some overhead expenses must be depository institutions with whatever level of recovered by the remaining services. we explore the most of the social costs that such a closing solutions to two optimization models for the would impose on third parties.14 http://clevelandfed. mathematical models volume to minimize overall processing and analogous to that in the present paper have shipping costs? In the second case. Silver. Beyond this.6 the Federal Reserve has not been given an explicit objective with respect to currency provision. The Federal Reserve generally does not ■ 6 See Adams (1980).7 Whether the Federal Reserve. we ensure that cost savings for the user fees. Consequently. Alternatively. Note that some MTAs encompass more than one city. This accounts for institutions. .3 This paper studies its customers worse off. a higher level of service to depository institu- tions. but processing sites (except Helena) plus 10 others. concave. the last site to be to Economic Review 2000 Q3 have necessitated a reexamination of the Federal Reserve are not obtained by making currency infrastructure. The first a policy matter that we will not consider here. the Uniform Cash Access Policy (see footnote 3) specifies the level of service that depository institutions should receive from existing Federal Reserves sites. removed from any site that is closed. circulating operates only 36 processing sites. but retained paying and lower price or higher quality. 1996. no sites in Hawaii or Alaska are considered in our might lower costs more than enough to offset analysis. Before the Fourth District removed currency processing from its ■ 7 This is not exactly what the Federal Reserve Bank of Cleveland did when the Pittsburgh branch’s high-speed sorting operation was moved Pittsburgh office in 1998. identical in structure and dif. and Peterson (1998) and everything. ■ 3 This process began in April 1996 with the announcement of the Uniform Cash Access Policy (UCAP). if we consider only the these sites. Early studies This is essentially a green-field approach: If include Hanssmann and Hess (1960) and the Federal Reserve were starting from scratch.4 In other words. the universe. linear. Our model assumes that these operations are also in 1938. the Federal Reserve’s perform. address these issues directly. ■ 5 These 46 MTAs include the 37 cities with Federal Reserve fewer sites would lower service levels. more of that receiving operations. Pyke. with reallocating currency volume. Our model accom- whether a redeployment of resources can lower plishes this by having the Federal Reserve the Federal Reserve’s costs while still providing continue to pay for shipping currency to and roughly the same level of service to depository from any site that is closed. how should we allocate processing distribution and logistics. When a product is offered at either a ■ 4 After the data for this study were collected. More sites would mean Federal Register. April 25.8 locations where the Federal Reserve already In the context of production planning. More precisely. For a detailed discussion of the UCAP. the Federal Reserve currently product is demanded. Clearly. Because such issues are far beyond the scope of this paper. Federal Reserve actually picks up these costs is fering only in the data fed into them. consistent level of cash access service across the nation in the question of how many processing sites the distribution of currency. the Pittsburgh site ceased currency processing operations. any such losses to the Treasury. see the Fed should operate. we start been used in operations management to study with the cities that anchor Rand McNally’s 46 problems of volume reallocation and of scale major trading areas (MTAs). Haehling von Lanzenaur (1970) for the joint in which of the 46 largest metropolitan areas problem of production and employment plan- would it locate processing sites to minimize ning.5 economies in processing costs. starts with 37 processing sites and determines The same is true of the cost implications for how the volume should be allocated among other Federal Reserve services remaining at them. In this case. Furthermore. opportunities. our model seeks to provide ■ 8 If a service is removed from a location. except Honolulu. paying and receiving have so far been main- closed was the Twelfth District’s Spokane office tained in Pittsburgh. more currency would indirectly reduce the Treasury’s borrowing needs. Consider the uniform. Thomas and McClain (1993) overall costs? This scenario gives us an estimate provide a survey of mathematical programming of the maximum possible cost savings from models in aggregate production planning. it is expected to run its currency operations in a cost-efficient manner. or convex costs and distribution Like the ultimate question to life. Recently. designed to achieve a ance objective is vague at best. In this case. This could have an adverse effect service they currently receive. While Honolulu is also a designated MTA. has sites.

green-field model suggests that the Soon after it is deposited at a processing site. Shredders attached to the high-speed Section I provides an overview of the Federal equipment destroy unfit notes. only a every angle. tutions with currency and accepts deposits of and economic planning. In addition. each note and then pass it through a low-speed The results are presented in section Economic Review 2000 Q3 discuss the advantages and disadvantages of I. in conjunction with the high- together with a discussion on the robustness of speed machine.15 http://clevelandfed. and the entire batch is cataloged on other Federal Reserve services. model is subsequently solved using estimates Rejects are sent through a cancellation proce- for the demand for cash and the processing dure in which operators manually examine and shipping costs as analyzed in section III. The high- that this model would not operate a site in speed equipment then repackages fit currency Helena and would open sites in Phoenix into straps and bundles. the amount of the note is deducted from the sions to the model are discussed in section V. machine which. second. This high-speed equipment are classified as rejects. orders are filled using ■ 9 See Bauer (1998). verified on high-speed sites is very close to the optimum. a decision which would of 10 straps. require examining several issues that we avoid The bundles are counted manually in the here. Currency enters circulation when a bank places an order for it. business. The Federal Reserve supplies depository insti- ming methodologies in industrial. handful of Federal Reserve sites appear to be Deposits are received as bundles made up candidates for closing. vault until needed. and CVSA $460 billion—is maintained at high levels of [forthcoming]). the two components of public enterprises such as joint ventures note quality. A note is Reserve’s currency service. such as transition costs and the impact receiving area. Burnetas. see. holes or if it is too soiled. depositing bank’s reserve-account balance and the note is turned over to the Secret Service. Hillier and role in maintaining the quality of circulating Lieberman (1995). In return. For general applications of mathematical program. for example.9 and franchises. Reserve employees. The rest of the paper is organized as follows. The mathematical deemed unfit for circulation if it is torn or has programming model that we employ to deter. and examined by sensors significant departures from the status quo are that judge its fitness for circulation. When a counterfeit is detected. costs. The model we develop has applicability Processing and distribution expenses to any enterprise that must provide a good exceed $280 million a year. graphically dispersed consumers. our and stored in the vault until it can be processed. The Federal the nation’s stock of 20 billion outstanding Reserve’s own check processing service meets Federal Reserve notes—a total value of these criteria (see Bauer. Reserve Banks acquire new notes from the The key finding from the optimization of Bureau of Engraving and Printing and receive our first model is that while the Federal Reserve used notes from banks depositing their excess may be able to save almost 20 percent of its currency holdings. currency from them. each strap containing 100 notes. it can operations require a highly secure area that obtain most of the $5 million in cost savings by separates armored car personnel from Federal reallocating processing volume without clos. These paying and receiving controllable costs by reallocating volume. but so do many private and fitness and integrity. currency by culling counterfeits and unfit notes. more than $400 million annually. Finally. Notes that are judged mine the optimal allocation of processing to be counterfeit or that cannot be read by the volumes is briefly presented in section II. The most sorting equipment. and Winston (1994). Security cameras cover ing any processing sites. Some conclusions and possible exten. Fed’s current geographic distribution of each note is counted. which are stored in the and Milwaukee. It also plays an important Charnes and Cooper (1961). and the cost to the or service combining some activities that have Federal Reserve for new notes purchased from scale economies with the requirement the Bureau of Engraving and Printing totals of delivering the good or service to geo. reconciles the account of the the solution for a range of estimated shipping depositing bank. . Currency Service using linear programming to model production Overview problems with nonlinear cost structures.

Processing costs would be minimized Book Entry Securities). we briefly present the integer linear programming model we employ to determine the optimal mix of cash processing volumes and shipment schedules among processing sites that will minimize the Federal Reserve’s processing and shipping costs. 16 http://clevelandfed. and shipments. For tractability. Bohn. To cost to banks for depositing or obtaining make our model computationally feasible. and every order must meet the local processed notes (fit plus unfit notes). depend. must meet certain requirements. function is lost in the piecewise approximation. Prior to 1996. at each site we hold the ratio of these three offices must deposit more than 20 bundles in outputs constant at the overall sample mean so that we only have to track aggregate. request new currency from the Federal Reserve. and piecewise linear average cost functions are consistent level of cash access across Federal plotted in figure 1 to demonstrate that very little Reserve districts (see footnote 3). Banks cannot specifically costs on the other. may do so by paying an access fee. and each order must meet the local facility’s minimum threshold for each ■ 11 Bauer. designed to achieve a uniform. To order more often. they must order more than 20 bun. the only if all sites operated at this level of output. model’s utility and limitations. however. To deposit the broad outlines of the model should be described in order to show the more often. costs—were not an important factor. consequently. shipping currency services from the Federal Reserve. fit notes. . are rationed according to the Uniform Cash and Hancock (forthcoming). If offices wish to deposit or order currency more frequently. II. unfit notes.11 ing currency distribution. Fedwire. dles in aggregate. we currency is that of transporting it to and from employ a piecewise linear approximation to the the Federal Reserve. number of bank offices that can obtain free If geography—and. and figure 1). Banks that cannot meet these requirements but still wish to obtain service more Economic Review 2000 Q3 F I G U R E 1 facility’s minimum threshold for each denomi- Comparison of Estimated and Piecewise nation deposited. and Hancock (forthcoming) actually estimate a cost function with three outputs. denomination requested. ACH. Bohn. information about the estimated average cost each district had its own set of policies govern. Generally. then one A bank may designate up to 10 offices to could determine an upper bound on the number receive one free order each week from the local Reserve Bank facility. ing costs on the one hand and transportation ing on availability. Economists define the minimum effi- require Reserve Banks to recover the costs of cient scale (MES) as the lowest level of output providing currency. unlike other Federal Reserve at which average cost reaches its minimum (see payment services (check. they ■ 10 While the general reader can safely skip the mathematical detail. The Fed’s currency services translog cost function estimated by Bauer. Model Description MES In this section. The interaction of scale economies and The Depository Institutions Deregulation shipping costs is the crucial component of our and Monetary Control Act (1980) does not model. A bank may obtain free Linear Approximation of Average Costs access for more than its 10 designated offices under certain conditions: All the offices (including the designated 10) must deposit Average cost (dollars per thousand) and order currency in volumes exceeding the Federal Reserve facility’s high-volume threshold (generally 50 to 100 bundles) and all must meet the facility’s minimum threshold for each denomination deposited or ordered.10 The model is then used to explore the trade-off between scale economies in process- new currency or existing fit currency. Note that MES is achieved at 218 million notes The Uniform Cash Access Policy limits the per quarter. or for more offices than the policy allows. Both the translog Access Policy.

have to be constructed). i j. However. If shipping costs are low or if MES is increased significantly. and receiving new cash.... which is not rent application all sites are assumed to shipped anywhere but instead is used to satisfy have the same one.. independent of the type of bills processed.. including the would mean that processing facilities would cost of paying and receiving.. We i. However..j = 1. printing. Decision Variables portation cost between two sites depends on the sites chosen as well as the type of vik = Volume of cash of type k to be processed bill shipped (Insurance costs make $1 notes at site i. approach takes a long-term perspective full-scale economies are achieved at a low and assumes that these capacity constraints level of output). then this approach would are resolved if more volume is shipped to still be roughly correct. • The costs associated with shipping cash (fit or unprocessed) are proportional to the volume of shipped cash. The main assump. a decidedly long-run perspective. System Parameters We allow the model to determine whether some or all the unprocessed cash collected at N = Number of cash processing sites. although in the cur. This parameter denomination because shipping costs and includes all costs due to shipping.. The unit costs of shipping fit and unprocessed currency are vi = Total volume of cash to be processed at the same.. one site should be shipped to other sites for b = Number of generic note types.. miik This function can generally be different for denotes the amount of cash either processed each processing site.... i. k =1. new cash to a site is directly proportional to tijk = Volume of unprocessed cash of type k the volume of new cash shipped.. much cheaper to ship).. ui = Cash processing capacity at site i.... This parameter represents processing (for some of the MTA cities. unit cost dependent on the destination site...17 http://clevelandfed. for i =1. that is fit for circulation after processing (yield).N. when MES such sites. We also allow fit cash at a site to be shipped to other sites for distribution to the cijk = Unit transportation cost for currency local depository institutions...b. bills of low value ($1) and high value ik = Proportion of unprocessed cash at site i (all other denominations). however. The unit trans..j =1. . with the shipped from site i to site j. presented in this section could differenti- ate between all the denominations.. • All the locations can be used for cash j = 1.N... the cost of shipping site i.b. occurs at a relatively high level of output We use the following notation in developing compared to total-system processing volume.. The model dik = Demand for currency of type k at site i. Note that for i=j.N...N. k =1.b.N.b...12 Finally. • There is no restriction on the amount nik = Volume of new cash of type k sent to site of cash that can be stored at any site. i. ing vary by denomination. • The costs associated with currency process.. the proportion found unfit during process...b.. processing. from site i to site j (to satisfy site-j demand) Economic Review 2000 Q3 of sites by dividing system-processing volume space if processing volumes at those sites by MES.. namely. section III we only use two types. mijk = Volume of fit cash of type k shipped ing are a function of the processed volume. paying. k =1. k =1.. our approach: then the number of processing sites will depend on both the degree of scale economies and how expensive it is to ship currency. notes of type k (processed or unprocessed) tions adopted in our analysis are: shipped from site i to site j.. sik = Supply of unprocessed currency of type k in the numerical computations described in at site i. our achieved at a low level of output (that is.. i = 1. This gives our approach the demand at this site. at site i or sent new to site i. should note that some Federal Reserve sites are starting to face vault-capacity constraints and might have to expand vault ■ 12 See Good and Mitchell (1999) for more details.. k =1.. j = 1.. pjk = Unit cost of new cash of type k delivered • Bills are differentiated according to their to site j. this all transportation-related costs.N..

N  fi (vi ) i=1 The Processing Cost Function N N b N N b +   cijk tijk   cijkmijk i=1 j=1 k=1 i=1 j=1 k=1 The model will be computationally tractable j 1 j1 as long as the processing cost functions. wir i .. vi ) is equal to the unprocessed ■ 13 Given our long-run perspective....13 determines the currency volumes vi. 18 http://clevelandfed.nik 0 v for all v>0. the range [0. can be adequately represented by a piece- N b wise linear function as +  pjknjk .b where v is volume.vik .ri –1. Cash demand at for wi j < v wi. wi..N.N. equation (1) is equivalent to j –1 We next explain briefly the motivation for fi (v) = i 1 wi 1+ i l (wi.. and fi (0)=0. According N to this definition. there- fore.ri ... This constraint ensures Model that the amount of incoming fit cash equals the outgoing fit cash at each site. As this capacity was set far above private banks. including cash sent from other sites and cash remaining at the site after processing. ji..j +1. it does not turn out to be a binding constraint. (2) each site must be satisfied. models. k =1. k =1. (b) Unprocessed Cash Balance. mijk..l +1 –wil ) + i j (v –wij ). This constraint ensures that the amount of outgoing unprocessed cash from site i (either shipped to other sites before processing.. all processing sites are cash coming into the site (shipped from assigned the same maximum capacity. k =1. tijk. we transform the original model into a Minimize mixed-integer... j =1. l =2 these constraints. To speed and sorting technology and demand- the computation of the solutions to our various satisfaction constraints... j=1 k=1 fi (v) = fi j + i j (v –wi j ).b j=1 =1.j +1 ]. the largest volume observed at any processing site... Let fi (v) denote the cost of processing cash (d) The total volume of cash processed at a site volume v at site i.tijk . = ui .. such that the processing (d) vi =vik i=1. j =1. j +1.... linear-programming problem.. tij ... or remaining at the site for local processing. .b processing volumes at site i can be divided j=1 into ri –1 consecutive subintervals with end- b points wi1......... j N (a)  mjik=dik i=1.N. Note that the left side of (a) is equal to the total processed cash volume available at site i. (e) 0 vi ui i=1. ui ] of possible (c) ik vik +nik =mijk i=1. comprises the volumes of different types..N. fi (v). i indicates the processing j=1 j=1 j 1 j 1 Economic Review 2000 Q3 The Optimization (c) Fit Cash Balance.. j i ). or from other sites tji .ri –1...N cost is linear with a slope equal to i j within k=1 the subinterval [wi j . subject to for wi j < v wi. (e) This last constraint ensures that processing The mathematical model developed below capacity is not exceeded at any site.... wil = 0. subject to shipping functions fi (v) are generally nonlinear. it follows that fi j = fi (wi j ). In addition. si . (1) N N (b) vik +tijk =sik +tjik i=1. (a) Demand Constraints.. The model above is a nonlinear program- nik that would minimize the total processing ming problem because the processing-cost and transportation cost.... because the cost function is continuous in mijk .

zil  wi.vik . linear-programming The functional form of the processing costs problem is given below: proposed in equations (1) and (2) is consistent N N N ri –1 with empirical findings about economies of  fi l δi +i 1zi1+  i 1(zil –zi. ■ 14 See Bauer. open for cash processing (δi =1) or not (δi =0).ri }. While they start with our The above expression describing the pro.l –1) scale in cash processing.ri –1 volume is shipped to another site to more fully 0  z 1 z 2 .. our model keeps track of low.. cost minimization.. j =1. ... for j =1. N Minimize Our piecewise linear approximation can be  mjik = dik j=1 made arbitrarily accurate by increasing the number of subintervals ri . Although our model is similar in its intent to ming problem: that developed by Good and Mitchell (1999).. fi (v) = min i 1 z1 + il (zl –zl –1) their model is geared to calculating the costs l=2 when various sites are closed. note that the piecewise linear 0 zi1zi2 . average cost begins to increase slowly.j –1. While cost. ij . given the trade. and Hancock (forthcoming).. Because fi (v) is not continu. Therefore. they reduce protection costs from the fixed processing costs—to model the fixed sites that are closed.l +1 l =1.14 Figure 1 illustrates i=1 i=1 i=1 l=2 how close the piecewise linear approximation N N b N N b is to the estimated translog. We employ a piecewise linear approxima- N tion to f (v) based on nine subintervals. therefore. Stemming from this first difference. δi  {0.and high-denomination notes because Economic Review 2000 Q3 The slopes. once they +  pjk njk j=1 k=1 are exhausted at about 218 million notes per quarter..ri –1= vi cost can be equivalently defined by the mijk .1}i . vik + tijk = sik + t jik j=1 j=1 off between the number of segments and the j i j i speed of convergence. and zj =zj +l = . is. then the optimal mainly because of lower insurance costs. fi j ).nik  0. Bohn. diseconomies.l +1 To see this. they assume that cessing cost function at each site i must now variable unit costs remain constant once be incorporated into the original problem of minimum efficient scale is achieved. = zr = v... they consider some indirect costs that we ignore.. of the cost function in con.zi.. but we deem nine + N N segments to be close enough. it can although both have the same processing costs.l +1. the cost is convex for v >0. be shown that if v is in the j th subinterval.   cijk tijk +  cijk mijk i=1 j=1 k=1 i=1 j=1 k=1 The presence of scale economies is also ji ji clearly evident. Initially. however. First. site i.j +1 for some j.19 http://clevelandfed. that low-denomination bills are cheaper to ship.. with the processing-cost function employed by Good objective function value precisely equal to the and Mitchell does not allow for scale processing cost as defined in equation (2). Another significant difference is that the l=1.ui ].. with subject to ik vik +nik = mijk endpoints wj (in thousands of notes) and j=1 costs fj (in dollars). we define a binary variable δi for each these are all potentially significant differences. such that δi =1 if v > 0 and δi =0 if v =0. The variable δi is associated with keeping site i secutive segments are increasing in j for anyi. exploit scale economies. because i j is increasing in j. for anyv (0.. solution to the problem above is zl = wi.. processing-cost function.zri –1=v. fi (v) can be expressed as the solution to the following linear program. ous for v =0—the discontinuity representing For example. sequence of points {(wi j .10 as specified b in table 1...tijk . Finally. ri –1 there are several important differences. whereas our model can examine the case when only some zl wi.. average cost falls as a N b result of scale economies. wij < v wi . The resulting mixed-integer.. whereas we do not.. The piecewise linear approximation vi = vik k=1 model for fi (v) allows us to reformulate the vi  ui δi original nonlinear programming problem using mixed-integer linear programming.

lected from the new sorters (which will require the further passage of time). Limited vault capacity could 8 318. centers and distribution depots and in essence portation costs are the best available. Except as noted Our results can only suggest situations that below.000 406.646 processing sites.726. Including these factors would Approximation result in cost-function variations across pro- cessing sites. a more complete cost-benefit expense reports.717 factors. volume.16 Analysis of this set explores sample that included sites with both old and where currency-processing sites would be new (high-speed) currency-sorting machines. (MTA) data sets. while most of the cost savings can 9 418.238. one that specifies either higher (forthcoming). costs. Alternatively. and Hancock currency was derived for the 46 MTA sites.166 adopt best-practice techniques). the contemplated shipments.000 3. tions of sites.687 is a long-run model at heart.088 capacity.000 1.000 1. except Helena. Solving our model with this alternative performance objective could provide additional insight into such important the next section will show that. Bohn. consequently. be reoptimized. On a related issue. the two models come up with very similar sets of results. the 46-site MTA set function estimated by Bauer. the existing Federal Reserve sites can lower A third potential weakness is that the cost overall costs. they are represent alternative ways of allocating the still just estimates.000 612. adopting a green-field approach. we intend to model long-run operational Two sets of data are used for our computations. . given the loca. First. Finally. so an argument 3 28. and shipping costs actually observed.15 located if the Federal Reserve were to start When a large enough sample has been col. all the data comes from the Federal require additional study.913 125. Hancock (forthcoming) used data from a plus 10 others. processing sites of the Federal Reserve System In light of this.000 245. Without actually performing same national data. the cost functions ■ 16 Good and Mitchell (1999) provide details on how demand for estimated by Bauer. a different objective function SOURCES: Authors’ calculations. respectively. although our estimates of trans. By opti- results by solving the model with transportation mizing the model with these data. and Hancock (for example. we do not consider any hereafter referred to as the Current Processing of the transition costs that would surely be Sites (CPS) and the Major Trading Areas incurred if a site were to be opened or closed.688. The two sets differ in the analysis should be performed. the cost functions could be re-estimated and our models could ■ 15 All of the older sorting machines are now retired.538 be obtained by reallocating volumes without 10 788.207 can be made for omitting potentially transient 4 48. Before actually closing Reserve’s Planning and Control System quarterly a processing Economic Review 2000 Q3 T A B L E 1 (forthcoming) incorporate site-specific environ- mental variables and allow for varying levels of Endpoints of Cost-Function cost efficiency. which varies significantly across 7 218. Analysis Some caveats should be kept firmly in mind. However.865 preclude some reallocations. it is unlikely that The CPS set comprises the existing 37 cash- our estimates of shipping costs can be improved. scale economies. we will costs 10 percent higher and 10 percent lower explore whether reallocating volume among than our best estimates.716 closing any sites. number of sites considered cash processing Second. Bohn. or lower service levels) might suggest more fundamental changes. one short-run factor that we 5 98. and includes the entire 37 CPS set. 20 http://clevelandfed.000 833. from scratch. we tested the sensitivity of our (see table 2 for a list of these sites). III. policy questions. which could affect our results. and our model 2 8. the sources of these vary- number point wj (dollars) ing productivity levels can be mitigated over j (thousands of notes) time (by encouraging underperforming sites to 1 0 100.000 183. and Bauer. Subinterval Subinterval starting Cost fj On the other hand.406 may want to include in future work is vault 6 158. Bohn.

569 El Paso 11.551 Memphis–Jackson 21.641 Knoxville 9.271 Charlotte–Greensboro–Raleigh 87.520 82.399 Little Rock 16.014 42. In thousands of notes.293 Oklahoma City 11.028 36.476 28.702 21.529 Philadelphia 83.244 73.263 Seattle 35.949 52.194 Spokane–Billings 9. Louis 26.646 Salt Lake City 12.952 Richmond 50.025 65.175 Helena 3.859 Miami–Fort Lauderdale 40.260 Birmingham 18.976 Richmond–Norfolk 34.094 Baltimore–Washington 69.605 Pittsburgh 32.821 25.072 22.776 Memphis 17.232 Detroit 71.002 Minneapolis 39.865 19. Louis 25.631 191. .054 32.591 San Antonio 25.285 142.483 80.111 Columbus 13.674 Omaha 9.139 24.501 77.891 39.464 Minneapolis–St.627 43.118 82.379 33.480 El Paso–Albuquerque 16. SOURCES: Good and Mitchell (1999).929 81.137 70.690 33.949 Portland 16.021 Pittsburgh 28.083 168.089 18.200 18. and Planning and Control System Expense Report.652 129.628 123.588 33.566 Houston 35.777 69.887 Los Angeles–San Diego 165.300 St. Petersburg–Orlando 34.776 Houston 33.720 102.255 Miami 40.837 Nashville 9.664 13.299 118.122 52.163 Buffalo–Rochester 41.977 139.139 79.255 Portland 14.942 25.227 New York 365.337 70.056 46.467 398.025 71.930 77.730 Salt Lake City 12.722 100.129 38.348 174.360 73.295 24.032 26.075 Dallas 43.031 108.775 182.685 San Antonio 19.886 Kansas City 12.475 56.586 Cincinnati–Dayton 35.21 http://clevelandfed.760 52.328 Wichita 5.006 87.817 203.735 11.361 209.830 7.028 Jacksonville 36.112 Cleveland 23.876 Boston Economic Review 2000 Q3 T A B L E 2 Demand for Unprocessed Casha 37-site case (Current Processing Sites) 46-site case (Major Trading Areas) Low-value notes High-value notes Low-value notes High-value notes Atlanta 49.480 Seattle 31.207 71.729 63.969 4.404 Kansas City 16.510 35.299 70.453 Omaha 7.061 Denver 33.213 132.322 68.554 St.664 331.458 New York 293.859 Milwaukee 38.026.594 Little Rock 12.865 53. 1996.714 a.771 68.683 Philadelphia 66.328 Louisville 18.495 Tampa–St.884 Buffalo 39. Paul 35.427 Phoenix 28.183 67.303 56.300 Boston–Providence 110.469 Totals 1.236 Charlotte 89.674 Cleveland 35.938 37.823 Jacksonville 10.643 Tulsa 6.854 San Francisco–Oakland–San Jose 95.861 38.028 2.603 Louisville–Lexington–Evansville 22.248 18.388 44.597 Baltimore 60.274 Nashville 19.511 Oklahoma City 18.322 41.633 Cincinnati 36.695 Chicago 176.372 Denver 26.730 New Orleans 37.065 Detroit 63.757 Chicago 101.337 22.009 New Orleans–Baton Rouge 31.172 595.265 14.955 Birmingham 20.942 221.755 731.208 357.244 Des Moines–Quad Cities 13.316 Dallas–Fort Worth 54.476 291.952 Atlanta 59.982.884 Los Angeles 196.782 San Francisco 103.097 62.599 Indianapolis 21.537 27.

888 7.975 36.890 Salt Lake City 11.456 Charlotte–Greensboro–Raleigh 82.799 Omaha 6.453 Atlanta 55.222 24.247 42.411 Little Rock 15.495 Miami 58.076 San Francisco 105.727 Miami–Fort Lauderdale 59.675 41.321 Spokane–Billings 9.131 Salt Lake City 11. Louis 24.119 97.621 84.538 53.220 12.522 94.284 1.021 Boston 132.445 65.696 76.781 15.731 31.407 El Paso–Albuquerque 21.855 30.583 Buffalo 38.411 22.512 Pittsburgh 24.342 55.682 25.964 67.697 Richmond–Norfolk 32.618 Los Angeles–San Diego 192.503 Jacksonville 12.22 http://clevelandfed.109 68.882 Birmingham 16.802 Louisville–Lexington–Evansville 20.774 82.805 21.636 494.927 Louisville 16.989 a.541 51.911 Oklahoma City 17.714 Wichita 5.501 Kansas City 14.621 70.975 Nashville 10.156 22.655 New York 303.842 195.312 Phoenix 37.519 81.777 Tampa–St.938 Little Rock 12.705 19.537 El Paso 14.161 58.389 18.685 Seattle 33.791 205.268 120.411 70.458 Tulsa 6.439 Philadelphia 88.649 33. Louis 25.425 29.729 91.561 49.956 607.059 Pittsburgh 27.496 Oklahoma City 10.120 Baltimore–Washington 67.749 25.718 135.839 San Antonio 34.621 Portland 14.024 Chicago 93.816 New Orleans 40.136 Milwaukee 35.698 43.822 Knoxville 9. In thousands of notes. Petersburg–Orlando 45.962 Baltimore 58.935 269.276 56.419 22.313 75.565 51.081 Des Moines–Quad Cities 11.287 Portland 12.261 Detroit 60.823 Minneapolis 37.279 Cincinnati 29.477 118.330 Chicago 162.894 Houston 30. Planning and Control System Expense Report.994 65.319 10.312 Detroit 53.443 Dallas 37.068 118.165 164.657 180.551 Cleveland 37.349 Memphis 17.497 Richmond 48.494 3.162 64.274 22.217 New York 243.833 76.929 Nashville 20.123 Columbus 12.637 Totals 1.656 141.514 49.908 78.436 Birmingham 17.995 Buffalo–Rochester 40.290 San Antonio 26.015 67.610 Omaha 7.770 62.990 Jacksonville 41.770 St.717 Denver 32.481 110. and authors’ calculations. SOURCES: Good and Mitchell (1999).899 Boston–Providence 102.757 Cleveland 24.408 20.963 385.534 Dallas–Fort Worth 47.671 119.181 34.363 Los Angeles 228.203 34.397 Helena 3.898.927 Memphis–Jackson 21.359 94.039 Charlotte 84.408 3.813 Minneapolis–St.324 Indianapolis 17.750 214.057 329.784 New Orleans–Baton Rouge 34.881 Cincinnati–Dayton 28.473 Houston 32. 1996.448 171.160 Kansas City 11. .250 24.001 Philadelphia Economic Review 2000 Q3 T A B L E 3 Supply of Unprocessed Casha 37-site case (Current Processing Sites) 46-site case (Major Trading Areas) Low-value notes High-value notes Low-value notes High-value notes Atlanta 46.030 Seattle 29.310 48.929.873 Denver 25.706 San Francisco–Oakland–San Jose 97.815 109.011 32.001 58.917.182 40.842 464.988 St.357 12.639 187.484 29.949. Paul 33.995 35.030 74.

org/research/review/ Economic Review 2000 Q3 In both data sets. Results value shipments is much stricter. processing in all cases. processing. Low-value notes represent closed under at least one scenario.18 Alternatively. severe abuse than other denominations. the cost of shipping them to a particular site is ■ 19 The total volumes between the two data sets differ because of not. The no-shipment case of computer time required to obtain a solution. Consequently. IV. The of transportation costs. volumes observed across the Federal Reserve..and high-value notes.. notes differ in unit transportation cost as well as in the yield of the cash-processing opera- tion. and the but this addition would add little to our case where shipments are allowed but no sites analysis while greatly lengthening the amount are permitted to close. On the basis of inter. justify further differentiation.22 The information in cities we rely on estimates from Good and tables 4 and 5 includes only sites that are Mitchell (1999). Transportation costs increase 1 billion unprocessed notes (mostly $1 bills).N. for i =1. to site j is pij = $41 per 1. separate assumptions about transportation views with cash personnel.C. We determined that refining this More formally. we assume that a costs to determine the sensitivity of our results. closer to Bureau’s sites in Washington. D. an advantage over those located farther away. The level of security required for high. the currency notes come with both volume and distance shipped. and new-cash components in the The demand and supply volumes are pre- sented in tables 2 and 3. notes to be the sum of the average cost of printing new notes plus the average cost per ■ 20 Allowing for differential shipping costs from the Bureau of note of shipping currency (Bureau-to-Federal Printing and Engraving to the various processing sites would give sites Reserve shipping costs/number of new notes). Sites not one-third of the total of unprocessed cash present in these tables remain open for cash and high-value notes represent two-thirds. and 2j =70 for all sites. uniformly lower (90 percent) and higher most likely because they are employed in (110 percent).000 notes. are allowed between processing sites. The results are summarized in table 4 for We determined that the differences in the 37-site (CPS) and in table 5 for the 46-site transportation costs and yields for notes of $5 (MTA) data sets. and Fort Worth.17 The two types of between low. ■ 22 By “no-shipment case” we mean that the model has not added The unit shipping cost between any two any shipments. for the MTA volume reallocation.. fluctuations in the demand for currency across the various locations.. after all. also take into account the cost difference value notes (all others). corresponds to the state of affairs in cash The data on the demand for fit cash and processing that was current at the time the data the supply of unprocessed cash in each of the were collected and serves as the basis of com- 37 CPSs are derived from average quarterly parison for estimating cost savings through values for 1997. j = 1. Printing’s average printing cost per thousand notes and the total cost of shipping new notes ■ 18 We do not study the possible complications of seasonal to Federal Reserve processing sites are known. Recall that the model takes sites is based on estimates from Good and the existing configuration as given and so starts with shipments of about Mitchell (1999).19 ■ 17 The proportion of demand for these two types of notes is Although the Bureau of Engraving and assumed to be the same for all sites and is set to their nationwide averages. The tables also show This ratio is based on average processing how costs break down into transportation. The model could as well as a control case. These scenarios are referred to more transactions and receive much more as the low-cost and high-cost cases. They in two varieties. . not that no shipments are made. The tables include information and higher are not significant enough to on the overall cost for the three cost scenarios.2.21 bundle of $1 bills can be shipped at one-tenth After solving the model with our best estimates the cost of a bundle of high-value notes.23 http://clevelandfed. low-value ($1) notes and high. Texas. where no shipments be modified to distinguish each denomination.20 The percentage yield of the cash. ■ 21 The optimal solutions for the two data sets and the various processing operations is equal to 1j = 60 scenarios are obtained by employing CPLEX optimization software. we take the cost of new assumptions made by Good and Mitchell (1999). at least five times the value is being shipped for a We solve the optimization model using three given number of notes.. for the CPS and MTA data sets. the cost of new cash delivered aspect of the model was not a high priority at this time. we also estimated other difference between the two note types is the model with the unit transportation costs that the processing yield is lower for $1 bills.

618. various cases.154 19.972 19. cost of new cash can be considered a fixed Comparing the last two columns of table 4 cost of the currency operation.236.706.2 million per to the sum of transportation and processing quarter.786 Total cost 98.719.786 75. by spending an addi- of new cash is assumed to be the same for tional $717.563.706. When the total total cost can be affected by reallocating cash cost is considered by including the fixed cost volumes among sites.901.826. a every site.077 28. costs cash between processing sites results in total increase for both transportation and processing.744 282.980 3.822 0 Total transportation cost 647. for optimization purposes. and yield umes can be achieved by exploiting scale figures and is independent of possible realloca.236.549.851 23.477. as ized merely by allowing for cash shipments well as cost savings resulting from the trans.863 104.972 435.105.295 0 High-value note 386. are very similar between the the no-closure constraint yields additional sav- 37.24 http://clevelandfed.236.316 163.776.000.692 SOURCE: Authors’ calculations. supply. Relaxing portation option.070 23. between sites without closing any.236.786 75.117 0 Variable cost Economic Review 2000 Q3 T A B L E 4 Optimal Solution Summary for Current Processing Sites Data Set (dollars) Low cost Reference cost High cost All sites open No cash (90%) (100%) (110%) shipments El Paso OPEN CLOSED OPEN Helena CLOSED CLOSED CLOSED Kansas City OPEN CLOSED OPEN Little Rock CLOSED CLOSED CLOSED Louisville CLOSED CLOSED CLOSED Oklahoma City CLOSED OPEN OPEN Omaha CLOSED CLOSED CLOSED Portland CLOSED CLOSED CLOSED Salt Lake City CLOSED OPEN CLOSED Low-value note 260.825.714.906 Controllable costs 23. the avoiding scale diseconomies at others.134 22. to be the critical factor in improving the system’s A comparison of the controllable costs (total efficiency.830. whereas closing sites has a much cost less the cost of acquiring new currency) in smaller effect.044 449.142 3.269. the savings are approximately new cash is not controllable. The unit cost 5 percent. Thus. Although the sub.179 23.980 3.637 98. .026 23. Therefore.146 3. This corresponds to savings in control- costs.786 75.955.728 285. the cost of of new cash.906 New cash cost 75.818 25.227 717. In contrast.070. ings of only about $200.764 Fixed cost 3.795.and the 46-site data sets.717 717.142 Total processing cost 22.236. The controllable-cost figures refer cost savings of approximately $5. because only these two components of lable costs of nearly 18 percent.075 410.856 98.854. economies more fully at some sites while tions.960 28.831.965 98. Specifically. it applies to both.004.879 75.785. reallocation sequent discussion concentrates on the 37-site of cash volume through cash shipments seems scenario.150 306. and the total amount required is more efficient allocation of processing vol- determined by the demand.952 22. the reference-cost and no-shipments columns As the unit shipping costs rise from 90 per- in table 4 reveals that allowing for shipment of cent to 110 percent of the reference case.000 in transportation per quarter. also demonstrates that the major part of these A first observation from tables 4 and 5 is savings (approximately $5 million) can be real- that transportation and processing costs.

However.374.883 New cash cost 76.088.883 Controllable costs 23.607.323 28.133 19.247 Fixed cost 3. which indicates that the optimal at a combination of other sites under this cost cost is fairly robust with respect to shipping.25 http://clevelandfed.275 23. Several observations can be made from Most of the cash that gets shipped consists table 4 regarding the behavior of site closings of low-value notes.759. moving from normal to high shipping levels of transportation costs.222 76.976 23.463. relatively few high-value as a function of shipping costs.903.714 Economic Review 2000 Q3 T A B L E 5 Optimal Solution Summary for Metropolitan Trading Areas Data Set (dollars) Low cost Reference cost High cost All sites open No cash (90%) (100%) (110%) shipments Birmingham CLOSED CLOSED OPEN Columbus CLOSED CLOSED OPEN Des Moines–Quad Cities CLOSED CLOSED CLOSED Indianapolis CLOSED OPEN OPEN Jacksonville CLOSED CLOSED OPEN Knoxville CLOSED CLOSED OPEN Little Rock CLOSED CLOSED CLOSED Nashville CLOSED CLOSED OPEN Oklahoma City CLOSED CLOSED CLOSED Omaha CLOSED CLOSED CLOSED Portland CLOSED CLOSED OPEN Salt Lake City CLOSED CLOSED OPEN Spokane–Billings CLOSED CLOSED CLOSED Tulsa CLOSED CLOSED CLOSED Wichita CLOSED CLOSED CLOSED Low-value note 245.105.222 Total cost 98. This makes sense intuitively.374.343.088.633 24.839 276.088. the model’s objective is economies in currency processing.098. configuration (for instance. for Salt Lake Only five sites are closed under all three City.431.636 Total processing cost 22. makes the transportation option more viable.607 23.105 SOURCE: Authors’ calculations. to minimize the Federal Reserve Banks’ costs.430 816. fewer notes are shipped. for notes are shipped. volume goes to Kansas City).692. higher shipping costs tend to lower the volume The lower shipping costs for low-value notes of cash shipped. The more expensive ship- example. ing open. However.222 76.198 99.321 24. leading to more sites remain.088.919.811.553.963 19.088. On the other hand.222 76. Four others .519 586.442 104.831.202.153.722.636 4. Consider.587.647 19.222 76.146 3. much of the cost variations.543 100. This is expected.687 23.474 4.897 0 Variable cost 19.183 1.793 22.706.358 862.299.030. ping costs for high-value notes appear to It is optimal for this site to be closed in the prohibit reallocating the processing of these low-shipping-cost case and open in the other notes given the relatively small cost savings two cases. as from further exploiting scale economies.767. This obser- costs rise. because as transportation costs results in closing the site. resulting vation is counterintuitive when considered in in smaller cost savings from exploiting scale isolation. the increase in controllable costs is about and the volume can be more cheaply handled 1 percent.855 100.065.037 0 Total transportation cost 1.220 28.122 334.592 302.607.753 255.072.860 0 High-value note 784.205. the processing site in Oklahoma City.312 3.

a particular site Another important result is that only a few is too costly because transition costs are a processing sites appear to be candidates for friction that make change less likely. the Tampa is preferred to Jacksonville in the MTA current Federal Reserve sites generally remain solution. cost function may be some opportunities for additional cost estimates are based on an evolving technology. However. data set do not. the sole performance objective of the Federal ably good job of selecting processing sites.26 http://clevelandfed. although there shipping costs are uncertain. Phoenix and Milwaukee are the alternative configuration. this case does have a number appears to be another manifestation of differ- of interesting points. it may overstate the diseconomies of several robust results. This some boost to their economic vitality because depository institutions located there would incur lower costs in shipping currency between difference is probably because Good and their branches and the currency depot. Also. This unexpected result cannot and vault space is constructed if necessary. This set. . of processing volume among the 46 MTAs. Given the small cost advantage of open. Summary ■ 23 In July 1999. because their model was not set up to examine value analysis of the site-closing decision may this question. Because reallocated. First. at Economic Review 2000 Q3 might also be candidates for closing. a bit more than the 10 percent ■ 25 While cities with currency-processing sites have received predicted by Good and Mitchell (1999). in a Finally. Transition costs are neglected. Intriguingly. a nearby city is preferred to approach and search for the optimal allocation an existing Federal Reserve site. processing sites ent assumptions about the existence of scale in Phoenix and Milwaukee (not currently diseconomies for larger sites leading to slightly Fed sites) remain open under all three different results. Good and Mitchell’s MTA optimum conclusions can be made from the MTA data has 34 processing sites versus our 31. tively.24 Consequently. open under all three transportation-cost As discussed earlier. We have tried to allow for these which continue to satisfy currency processing needs nearly a hundred years later. as mentioned above. Good and Mitchell’s model suggests that Whereas Phoenix and Milwaukee remain a site in Spokane would be viable. Among current processing sites. savings. do not appear to be good choices for process- Lastly. only it would be costly to reopen processing sites. We have ■ 24 Because our cost function is based on a translog approximation to the underlying true functional form. analogous ing sites. Milwaukee’s volume would most likely may be able to save up to about 20 percent be sent to Chicago for processing. it appears that the founders of the and finally. For instance.25 V. transportation-cost scenarios. Of Mitchell do not get any cost savings from course. be confirmed by Good and Mitchell (1999) Factoring these additional costs into a present. In the green-field simulation. any sites that were closed the impact on the only nine warrant further study to determine cost of providing other Federal Reserve whether their processing operations should be services would have to be considered. which is scheduled to begin operating in Federal Reserve currency processing sites September 2001. If so. First. the study takes a achieved without closing any processing sites long-run approach and assumes that all inputs by shipping mostly low-denomination bills adjust fully to the new processing volumes. this endogenous effect is likely to be small. factors that must be considered before any Second. even when we adopt a green-field handful of cases. the Federal Reserve scale once MES is achieved. Reserve. eight other sites added by the MTA our results. of controllable costs by reallocating processing volumes. Alterna- moves impractical. from sites with scale diseconomies to sites with This means that currency sorters are reallocated scale economies. cost minimization may not be Federal Reserve System in 1913 did a remark. 15 MTA sites (8 of them Federal Reserve sites) there is an option value for retaining them. For example. this study does not examine all the avoiding scale diseconomies. reveal that closing.23 Second. the Federal Reserve Bank of San Francisco We set out to construct a model that we could announced that it had signed a contract to purchase a Phoenix site for a use to determine the least-cost configuration of future cash operations center. most of these cost savings can be sites are closed. the transition cost the only added sites that remain open under of relocating processing sites makes any such all three shipping-cost assumptions. or opening. some caveats apply to scenarios. given the trade-off between processing scale economies and transportation costs.

.27 http://clevelandfed. Joseph. “A Linear Programming Approach to Production and References Employment Scheduling. to the Galaxy. Christoph. 17.” Management Technology. “Federal Reserve Banks and Branches Winston. 28. eds. 1994. and Economic Commentary.S. Credit. Good. Haehling von Lanzenaur. McClain. Research... Banking. Frederick S. Currency: How Much Economic Review. 1998. Paul W. 46–51. 1. Vesala. “Scale Economy.” Federal Reserve Bank Applications and Algorithms. Adams. Paul W. & Sons.” Porter. Inventory Management and of Cleveland. forthcoming. vols. 1999. (October 1996). 2 (1970). Research Logistics. Industrial Applications of Linear Program. “Production and Employment Scheduling in . Economies in Federal Reserve Check Pro. 1961. Paper. Economic Review 2000 Q3 shortcomings in various ways and feel that Multistage Production Systems. F. Federal Reserve Bank of Cleveland. and Jukka M. Tech. Electronic Payments: A Cross-Country Bauer. I and II. Working Paper. Wayne L. pp. 914–39. forthcoming. of Dallas. David B. Judson. vol. and P. no.. Rinnooy Kan. Pulley. Hanssmann. Federal Reserve Currency Processing. Paul W. and Gerald J. and William New York: John Wiley & Sons. and Rein cessing Operations. Logistics of Production and Inventory. Charnes. Federal Reserve Bank of Cleveland. 193–8. W. Hess. 883–903. and Analysis. Douglas. David F..: Duxbury Press. Operations Research: Location Study. Abraham. Introduction to Operations Books. 1998.. Richard D. no. Apostolos Burnetas. 4. pp. “Cash. James Bohn. vol. and John O. pp. and Ruth A. vol.” Journal of Money. Cooper. Federal Reserve Viswanath CVSA. New York: McGraw-Hill. New York: Harmony Lieberman. “Employing Scale Bulletin. Graves. Management Models and Thomas. Barbara. 1993.” in ming. “An Overview of Production Planning. 1995. and W. 4. Bauer. The Hitchhiker’s Guide Hillier. A. October 1. Production Planning and Scheduling. Pyke. Lawrence B. no. Belmont. and Cost Efficiency of 1996). unpublished manuscript. and Diana Hancock. and Harvey Mitchell. vol. Amsterdam: Elsevier. “The Location of U.. and Federal Reserve System.” Naval our qualitative results are robust. chapter 7. Is Abroad?” Board of Governors of the Bauer. Calif. New York: John Wiley S. 1 (1960).” Federal Reserve Bank Peterson.. “Currency: Time for Change?” Humphrey. 1980. part 2 (November nological Change. Zipkin. Edward A. Silver.