PRODUCTION AND OPERATIONS MANAGEMENT POMSweb-static.stern.nyu.edu/om/faculty/pinedo/ofs_undergrad/...(Spohrer and Maglio 2008). A few books and a special issue of Management Science

Operations in Financial Services—An Overview

Emmanuel D. (Manos) HatzakisGoldman Sachs Asset Management, Goldman, Sachs & Co., New York, New York 10282-2198, USA, [email protected]

Suresh K. NairDepartment of Operations and Information Management, School of Business, University of Connecticut, Storrs, Connecticut 06269-1041, USA,

[email protected]

Michael L. PinedoDepartment of Information, Operations and Management Science, Leonard N. Stern School of Business, New York University,

New York, New York 10012-1106, USA, [email protected]

We provide an overview of the state of the art in research on operations in financial services. We start by highlighting anumber of specific operational features that differentiate financial services from other service industries, and discusshow these features affect the modeling of financial services. We then consider in more detail the various different researchareas in financial services, namely systems design, performance analysis and productivity, forecasting, inventory and cashmanagement, waiting line analysis for capacity planning, personnel scheduling, operational risk management, and pricingand revenue management. In the last section, we describe the most promising research directions for the near future.

Key words: financial services; banking; asset management; processes; operationsHistory: Received: November 2009; Accepted: July 2010 by Kalyan Singhal, after 2 revisions.

1. IntroductionOver the past two decades, research in service oper-ations has gained a significant amount of attention.Special issues of Production and Operations Managementhave focused on services in general (Apte et al. 2008),and various researchers have presented unified theories(Sampson and Froehle 2006), research agendas (Roth andMenor 2003), literature surveys (Smith et al. 2007),strategy ideas (Voss et al. 2008), and have discussedthe merits of studying service science as a new discipline(Spohrer and Maglio 2008). A few books and a specialissue of Management Science have focused on the oper-ational issues in financial services in particular (seeHarker and Zenios 1999, 2000, Melnick et al. 2000).However, financial services have still been given scantattention in much of the literature relative to otherservice industries such as transportation, health care,entertainment, and hospitality. The dilution of focus, byconcentrating on more general distinguishing featuresdoes not do justice to financial services where some ofthese characteristics are not central. (The more generalfeatures that are typically being considered includeintangibility, heterogeneity, contemporaneous produc-tion and consumption, perishability of capacity, waitinglines (rather than inventories), and customer participa-tion in the service delivery.)

In this overview, we mean by financial servicesprimarily firms in retail banking, commercial lending,

insurance (other than health), credit cards, mortgagebanking, brokerage, investment advisory, and assetmanagement (mutual funds, hedge funds, etc.).

1.1. Importance of Financial ServicesFinancial services firms are an important part of theservice sector in an economy that has been growingrapidly over the past few decades. These firms pri-marily deal with originating or facilitating financialtransactions. The transactions include creation, liqui-dation, transfer of ownership, and servicing ormanagement of financial assets; they could involveraising funds by taking deposits or issuing securities,making loans, keeping assets in custody or trust, ormanaging them to generate return, pooling of risk byunderwriting insurance and annuities, or providingspecialized services to facilitate these transactions.

Services is a large category that encompasses firmsas diverse as retail establishments, transportationfirms, educational institutions, consulting, informa-tion, legal, taxation, and other professional, real estate,and healthcare. Even within financial services, there isa wide variety of firms which are characterized byunique production processes and specialized skills.The processes and skills required for banking arequite distinct from solicitations for credit cards, ac-quisitions of new insurance accounts, or the handlingof equity dividends and proxy voting, for example.

633

PRODUCTION AND OPERATIONS MANAGEMENTVol. 19, No. 6, November–December 2010, pp. 633–664ISSN 1059-1478|EISSN 1937-5956|10|1906|0633

POMSDOI 10.3401/poms.1080.01163

r 2010 Production and Operations Management Society

mailto:[email protected]:[email protected]:[email protected]

Even though services account for about 84% of thetotal employment in the economy, only about 4% ofthis workforce is employed in financial services. Thismight come as a surprise to some because financialservices transactions in one form or another are soubiquitous in our lives. Not surprisingly, however, thenumber of financial services firms is about 7% of thetotal non-farm firms and contributes about 13% of to-tal non-farm sales. Only wholesale trade has a similaremployment and number of firms with a larger con-tribution to sales.

Table 1 provides employment information for thesub-codes within financial services. As can be seen,retail banks, insurance companies, and insurance bro-kers together employ about two-thirds of the financialservices workforce. A cursory look at the table gives asense of the diversity of the services sector. Clearlyoperations management problems and approachesused to solve them have to be customized for partic-ular types of services—we already know that whatworks for manufacturing may not work for services,but by looking at Table 1 we can also realize that whatworks for retail trade or recreational services may notwork for financial services. A quick glance throughMonster.com’s job openings for operations managersin financial service firms shows a wide variety of ti-tles, responsibilities, and ‘‘products’’ related to suchjobs. This is shown in Table 2, and gives a sense of thewide swath of topics that could be covered in aca-demic research on financial services operations. Asfinancial services are such an important segment of

the services economy, we wish to explore whetheroperations in financial services are indeed unique, orshare several characteristics with services in general.That is the motivation for this special issue.

1.2. Distinctive Characteristics of Operations inFinancial ServicesThere are several unique operational characteristicsthat are specific to the financial services industry andthat have not been given sufficient attention in thegeneral treatment of services in the extant literature.We list below a number of these unique operationalcharacteristics and elaborate on them in what follows:

� Fungible products with an extensive use oftechnology

� High volumes and heterogeneity of clients� Repeated service encounters� Long-term contractual relationships between

customers and firms� Customers’ sense of well-being closely inter-

twined with services� Use of intermediaries� Convergence of operations, finance, and marketing.

1.2.1. Fungible Products with an Extensive Useof Technology. One obvious difference betweenoperations in financial services and operations inmanufacturing and in other service industries is thatthe ‘‘widgets’’ in financial services are money, or relatedfinancial instruments. As there is a declining use of thephysical vestiges of money such as coins, currency,

Table 1 Employment within Financial Services

NAICS code Title within financial services

Total employees

Number %

5211 Monetary authorities-central bank

(Federal Reserve banks, etc.)

21,510 0.4

5221 Depository credit intermediation

(banks, credit unions, etc.)

1,816,300 30.7

5222 Non-depository credit intermediation

(credit cards, mortgage lending, etc.)

659,930 11.2

5223 Activities related to credit intermediation

(brokers for lending)

294,910 5.0

5231 Securities and commodity contracts

intermediation and brokerage

516,010 8.7

5232 Securities and commodity exchanges 8,010 0.1

5239 Other financial investment activities

(mutual funds, etc.)

344,950 5.8

5241 Insurance carriers 1,258,050 21.3

5242 Agencies, brokerages, and other

insurance-related activities

907,880 15.3

5251 Insurance and employee benefit funds 47,730 0.8

5259 Other investment pools and funds 41,190 0.7

5,916,470 100.0

Source: Bureau of Labor Statistics, stat.bls.gov/oes/home.htm, May 2008.

Table 2 A sample of Financial Service Operations Job Titles, Responsi-bilities, and Products

Titles Products

Vice President, Operations, Opera-

tions Manager, Financial Operations

Supervisor, Foreclosure and Bank-

ruptcy Operations Manager, Risk

Operations Team Manager, Team

Manager Ops Control-Fixed Income,

National Director Operations, Hedge

Fund Operations Specialist, VP/

Director Of Operations

Premiums, Claims, Refunds, Cash

flow and treasury management,

Customer statements, Loan servi-

cing and support, Trade

confirmations, Reconciliations, Tax

reporting, Security settlements,

Mortgages

Nature of work/responsibility

Brokerage operations, Improve customer service, resolves customer issues,

Review security pricing, Vendor support, Authorize net settlement, Hand-off of

data, ensure data integrity, Verifying transactions, Tracking missing transactions,

Leverage technology, Maintain ops controls, update policies, procedures, Back-

office support, Understand regulations, Ensure compliance, Attain profit and

revenue benchmarks, Reduce risk, Improve quality, Six sigma, Operational

processing efficiency, Problem solving, Ensure best practices, Streamline

activities, Manage key expenses, Work management tools, Monitors work flow,

Production/testing

Source: Monster.com jobs listings during the week of March 8, 2009.

Hatzakis, Nair, and Pinedo: Operations in Financial Services—An Overview634 Production and Operations Management 19(6), pp. 633–664, r 2010 Production and Operations Management Society

bond, or stock certificates, much of the transactions arein the form of bits and bytes. Thus inventory is fungibleand can be transported, broken up, and reconstituted(facilitating securitization, e.g.) in malleable ways thatare simply not possible in manufacturing or in otherservice industries (see, e.g., the recharacterization ofbank reserves in Nair and Anderson 2008).

The increased use of online transactions (in broker-ages, credit card payments, retail banking, and retire-ment accounts, e.g.) are forcing fundamental changes inthe way operations managers think about capacity is-sues (for statement mailing and remittance processing,or for transfer of ownership in securities, e.g.). The factthat adoption of online transactions is still growing andhas not yet matured and leveled off makes capacityplanning a big challenge. Yet we are aware of very littleresearch that would help managers deal with this issue.

1.2.2. High Volumes and Heterogeneity of Clients.Financial services are characterized by very highvolumes of customers and transactions. Furthermore,customers are not all alike. In many firms, a smallfraction of the customers generate most of the profits,giving the firms an incentive to view them differentlyand provide differential treatment, given the firms’limited resources. For example, high net worthindividuals may be treated differently by assetmanagement firms; banking clients who keep highbalances in checking accounts and transact heavily maybe handled differently from depositors who keep almostall their funds in savings accounts and certificates ofdeposits (CDs) and transact minimally; revolvers (i.e.,customers who carry over balances from one month tothe next) may be regarded differently from transactors(customers who do not revolve balances) by a creditcard firm. In most non-financial services, because ofa limited number and sporadic interactions withcustomers (e.g., in restaurants and amusement parks),one customer is considered, for the most part, similar tothe next one in terms of margins and attention required.

1.2.3. Repeated Service Encounters. In contrast toother service industries, where research typicallyfocuses on a single encounter (‘‘the moment oftruth,’’ ‘‘when the rubber hits the road’’), financialservices are characterized by repeated serviceencounters or potential encounters between the firmand its customers due to regular monthly statements,year-end statements, buy/sell transactions, insuranceclaims, money transfers, etc. Anecdotal evidence fromthe brokerage and investment advisory industrysuggests that clients with low asset balances andtransaction volumes contribute the least to firm revenueand the most to operational cost through calls forcustomer service. One online bank discouraged calls tocustomer service because it found that just a few calls by

a client could wipe out all the profit from the client’ssavings account. Very little research in service operationsmanagement has focused on this issue. New customersconstitute another major group that is more likely tomake calls with billing questions or inquiries regardingtheir statements. Should billing and statementing to newcustomers be handled differently, perhaps with morecare, than to existing customers? Obviously, if thisobservation is true, differential handling can reduce thetraffic to call centers. Existing call center research usuallyassumes the call volume to be a given, for the most part,and the focus is on ‘‘managing’’ the traffic. This is akin totraditional manufacturing where it was assumed thatlarge setup times were a given, and a good way to‘‘manage’’ would be to use an optimal batch size. Onelearns to live with such a constraint. Not until just intime (JIT) manufacturing came along did managersquestion why setup times were large and what could bedone to reduce them.

At one credit card company, operations managersstruggled for years to cope with volatile demand inbill printing, mailing, remittance processing, and callcenter operations. Daily volumes could fluctuateeasily between half a million and one and a halfmillion pieces of mail in remittance processing, andmanagers were reconciled to high overtimes and idletimes because they felt they had no control overwhen customers mailed in their checks, and reduc-ing float was important. Call volumes at the callcenters were similarly volatile, as were volumes inthe bill printing and mailing operations. This situa-tion continued until someone recognized that allthese problems were interconnected and to a largeextent within the control of the firm. As it happens,the portfolio of current customers is distributed intoabout 25 cycles, one for each working day of themonth. For example, customers in the 17th cycle arebilled on the 17th working day of the month. Care istaken to ensure that customers in the same zip codeare put in the same cycle so volume-mailing dis-counts from the US postal service can be obtained,and the cycles are level loaded. However, this allo-cation to cycles was carried out several years backand over time some customers had closed theiraccounts while new customers had been added toexisting cycles, resulting in large differences inthe numbers of customers in the various cycles,and in a wide variability in the printing and mailingof monthly statements. On the remittance side, ananalysis found that there was less randomness incustomer payment behavior than one would expect.There were broadly four cohorts of customers:one that sent in payments on receipt of the state-ment, a second that mailed checks based on duedate, a third that acted based on salary paymentdate, and a fourth that acted randomly. The first two

Hatzakis, Nair, and Pinedo: Operations in Financial Services—An OverviewProduction and Operations Management 19(6), pp. 633–664, r 2010 Production and Operations Management Society 635

cohorts, the largest ones, were in fact dependent onthe cycles that the firm had set up many years back.Similarly, billing calls were also heaviest soon afterthe statement was received by the customer, againtraffic that was determined by the cycles created bythe firm.

If the cycles could now be level loaded, many ofthese problems would disappear (similar to whathappened in manufacturing when setup times weredramatically reduced thereby enabling lean opera-tions). But there was a problem—the firm needed toinform each customer if their cycles were moved, forgood reason because customers needed to plan theirfinances. However, this notification was not neces-sary if the move was to be within � 3 days fromtheir current cycle. An optimization model foundthat this constrained move was sufficient to take careof the vast majority of moves that were initiallythought to be necessary.

This example illustrates how stepping back andtaking a broader view of the situation and collabo-rating across processes can have a major impact onfinancial service operations, something that is lack-ing in the current literature.

1.2.4. Long-Term Contractual Relationships BetweenCustomers and Firms. Connected to the previouscharacteristic of repeat encounters is the recognitionthat, unlike in other services, in financial services thefirm and the customer have a relatively long-termcontractual arrangement. However, technology and in-formation availability makes comparison shoppingeasy, resulting in easy switching between firms, andtherefore high attrition. This loss of customers makes theacquisition process very important to the continuedgrowth and profitability of the firm. Similarly, loyaltyprograms (such as rewards and balance transferprograms in the credit card business) are important tostanch the bleeding. The design and execution of theseprograms are based on complicated processes that needto consider risks, costs, redemptions, incremental sales,scheduling and sequencing of offers, etc. Researchers infinancial services operations, by not making theirpresence felt in these areas, are missing the boat withregard to issues that are the most important (‘‘must do’’activities) for the firm, and may be paying instead toomuch attention to relatively mundane and low-impactissues (‘‘good to do’’ activities).

Just as the above processes aim at increasing rev-enue, there may be other processes that are put inplace to reduce unnecessary costs. In the insurancebusiness, for example, the claims processes may pri-marily revolve around a call center, which hasattracted sufficient attention in the literature as wewill see later. But unnecessary costs can be reducedby fraud prevention and detection, and subrogation

activities (money the firm pays out but is owedto it by other carriers). Timely intervention canavoid expiry of opportunities to collect dollars owed,and more attention could be paid to even smallopportunities. There is an extensive literature in riskand insurance journals on scoring for fraud pre-vention and detection, but leveraging that informa-tion in the claims process can benefit from anoperations perspective.

Another example from the insurance industryconcerns worker’s compensation claims, where theprocess for handling workplace injury can havelong tails spanning several years before the claimis closed. The process is complicated with inter-actions between the worker, the employer, medicalpractitioners, hospitals, state authorities, and law-yers (both on the staff of the insurance firm andpanel counsel, i.e., lawyers who are hired on anad hoc basis). There are several opportunitieshere (Jewell 1974) to speed up the process (andspeed up the workers’ return to work, which is inthe insurance firm’s interest), reduce costs, detectfraud, ensure that review triggers are not over-looked, increase the utilization of staff counselcompared with panel counsel by better schedulingof appearances for hearings, etc. We are unawareof any recent operations management literature inthis area.

1.2.5. Customers’ Sense of Well-Being CloselyIntertwined with Services. Along with the ease ofmanipulating the putty at the core of the financialservices process comes the responsibility of workingwith something that is so close to the customer’s senseof well-being and worth. Poor operations manage-ment that results in delays, quality issues, or sloppi-ness can and will attract regulatory scrutiny andunfavorable publicity, and will generate immediaterebukes from the customer in the form of calls,complaints, and because the account can be easilymoved around, customer attrition. At least two factorsmake the detection of errors due to operational faultsand their exposure to the clients relatively easy infinancial services:

(i) the amount, frequency, and detail of communi-cation and disclosure as required by regulation,and

(ii) the clients’ heightened propensity and incentiveto check for error in something so closely linkedto their livelihood and sense of security.

Because of the above, tolerance for error is signifi-cantly lower than in other industries. For example,faulty processes resulting in incorrect calculations ofinterest amounts in savings, mortgage loan, or creditcard accounts, or in inappropriate handling of stock


dividend payments, become obvious immediatelyafter monthly statements are sent out.

The above customer service issues are distinctfrom the perceived quality of the performance of in-dividual customer brokerage portfolios, retirementaccounts, annuities, mutual funds, and interest accruedin retail banking. With the plethora of informationavailable comparing a customer’s firm with others inthe same space, moving an account to the competitionis only a few clicks away. Even though performancemay depend on the economy, the stock market, invest-ment research, and fund manager performance,recognizing the costs (Schneider 2010) and capacityissues (the increased transactions during market tur-bulence, e.g.) are important operations managementconcerns that have received little attention.

1.2.6. Use of Intermediaries. This is an importantaspect of the financial services industry. In some casesa direct-to-consumer approach is used (credit cards);in other cases most of the customer facing work isdone by intermediaries (financial advisors, insuranceagents, annuity sales through banks, etc.); and in stillother cases the firm’s employees and its agents haveto collaborate with one another (insurance). Workingthrough an intermediary entails a set of issues notnormally seen in other services that function withoutintermediaries. For example, financial product andservice design and delivery get filtered through theprism of what the agent feels is in his or her own bestinterest. At times, the relationship between the firmand the intermediary is not exclusive, hereby adding alayer of complexity because the customer may choosebetween products from competing firms. Therefore,what gets planned in the corporate offices of thefinancial services firms and what is seen by thecustomer may be quite different. The operationsmanagement literature, to our knowledge, has notpaid attention to product and service design in suchsituations, because the implications on customerlifetime interactions with the firm go much beyondinitial pricing, product features, and the inventory ofbrochures left with the agents.

1.2.7. Convergence of Operations, Finance, andMarketing. There is probably no other industry wherethis convergence is more pronounced. These functionsare supported and enabled by a healthy dose ofstatistics, technology, and optimization. By focusingonly on back-office operations such as call centers,researchers in service operations are leaving a lot on thetable. There is very little research in the serviceoperations literature that leverages this convergence,which requires a choreography, as described by Vosset al. (2008), who put it in a more limited context ofoperations and marketing. For example, the client

acquisition process in full-service retail brokerage andinvestment advisory firms begins at the corporate level,where it draws resources from marketing, strategy,information technology, and operations, and is ulti-mately implemented through the sales force of brokers/financial advisors. Customer acquisition at a credit cardcompany is a competitive differentiator and a complexprocess focused on direct mail campaigns. At manylarge firms, the budgets for direct mail run into severalhundred million dollars annually. By focusing on thebilling mailroom, collections, call centers, and billingcall centers, researchers in service operations are work-ing on a problem akin to quality inspectors at the end ofthe production line—by then it is too late, the volumesof mail and calls are baked in during the mailingcampaign creation, while their skills could have madethe mailings more effective and targeted (given theminuscule response rates), resulting in fewer delin-quent accounts (requiring fewer outbound collectioncalls), and perhaps also fewer billing calls to inboundcall centers.

At a more sophisticated level, very few credit cardfirms use contact history in mailing solicitations,which may result in repeated mailings to chronicnon-responders. The managers developing campaignstrategies may not have the analytical background thata researcher in service operations can bring to bear,and the cycle time for campaign creation is typicallyso long and complicated that much attention getsexpended on scoring for credit and response, filetransfers from credit bureaus and data vendors, scrub-bing of data, etc. These complicated processes leavelittle time to incorporate experience from a previousmailing because a reading of the results of that mailingtakes time (prospective customers may not respondimmediately, even if they do respond), and file struc-tures may not have been designed to carry informationabout previous contacts and the response to them.

Another indication of this convergence is that themajority of the undergraduate and MBA hiring at aninvestment bank in the greater New York City areafrom one of the region’s business schools is in theCOO’s operation, whether the student concentra-tions are in finance, marketing, information systems,or operations.

The foregoing does not imply that no significantwork has been done in the research on financialservice operations, just that areas of work have had anarrow focus. The purpose of this article and thisspecial issue is to begin to expand that focus andencourage research in neglected or emerging areas infinancial service operations. We will survey existingresearch next, not only where the attention is only onfinancial service operations, but also where researchin service industries in general has substantial appli-cation in financial service operations. In Appendix


A, we provide an overview of the various operationsprocesses in financial services and highlight theones that have been addressed in operations man-agement literature.

This survey paper is organized as follows. Sections2 through 9 go over eight research directions that areof interest from an operations management point ofview. The first couple of sections consider the moregeneral research topics, whereas the later sections gointo more specific topics and more narrowly orientedresearch areas. Section 2 focuses on process and sys-tem design in financial services, while section 3considers performance measurements and analysis.Section 4 deals with forecasting, because forecastingplays a major role in virtually every segment of thefinancial services industry. The next section focuseson cash and liquidity management; this section re-lates cash management to classic inventory theory.Sections 6 and 7 deal with waiting line managementand personnel scheduling in retail banking and incall centers. Even though these two topics are stronglydependent on one another, they are treated separately;the reason being that the techniques required for deal-ing with each one of these two topics happen to bequite different from one another. Section 8 focuseson operational risk in financial services. This areahas become very important over the last decade andthis section describes how this area relates to otherresearch areas in operations management, such as to-tal quality management (TQM). Section 9 considersproduct pricing and revenue management issues. Thelast section, section 10, presents our conclusions anddiscusses future research directions.

2. Financial Services System DesignService systems design has attracted quite a bit ofattention in the academic literature. It is clear thatservice design has to be as rigorous an activity asproduct design, because the customer experiences theservice first hand, much like a product, and comesaway with impressions regarding the quality of ser-vice. Although the quality of service delivery dependson a number of factors, such as associate training,technology, traffic, neighborhood customer profile,access to the service (channel access), and quality ofresource inputs, the service experience gets baked intothe process at the time of the service design itself, andtherefore a proper service design is fundamental tothe success of the customer experience.

2.1. Aspects of Service DesignService research has usually focused on capacity man-agement (type of customer contact, scheduling, anddeployment) and the impact of the response to vari-ability on costs and quality. For long the nature ofcustomer contact has influenced service design think-

ing by creating front-office/back-office functions(Sampson and Froehle 2006, Shostack 1984). Shostackalso pioneered the use of service blueprinting foridentifying fail points where the firm may face qualityproblems. She illustrated this methodology for a dis-count brokerage and correctly identified that many ofthe operational processes are not seen by the customer;she then focused on the telephone communication step,the only one with client contact. This focus on clientcontact tasks, whether in the front office or in the backoffice, is widespread in services research in general andin research on financial services operations in particular.One reason may be that service researchers have foundit necessary to motivate their work by differentiatingservices from products (whether it is service marketingvs. product marketing, or service design vs. productdesign), and client contact is an obvious differentiator.

From the outset, it has been clear that serviceprocesses are subject to a significant amount of ran-domness from various sources. Frei (2006) discussesthe various sources of randomness in service processesand how firms react to them in the design of theirservices. She identifies five types of variability—customer arrival variability, request variability, cus-tomer capability variability, customer effort variability,and customer preference variability. She states thatfirms design services to factor in this variability bytrying either to accommodate the variability at a highercost (cross training of employees, increased automa-tion, variable staffing) or to reduce the variability witha view to increasing efficiency rather than cost (offpeak pricing, standard option packages, combo meals).

2.2. Focus on Single EncountersMuch of the services literature, however, focuses onsingle service encounters, which are common in ser-vices such as fast food. Even if a customer repeatedlyvisits the same restaurant, there is not the kind ofstickiness to the relationship as can be found infinancial services. Retail banking seems to haveattracted the most attention among financial serviceswith respect to service design, but here again thefocus is on disparate single visits to the branch orAutomated Teller Machine (ATM), rather than as partof a life cycle of firm–customer interactions. Otherthan meeting the branch manager when opening anaccount, there is usually no other recognition of thestage of the relationship in the delivery of service.Perhaps this will change with time as more firms startexperimenting with their service delivery design asBank of America has been doing (Thomke 2003).

2.3. Descriptive vs. Prescriptive Studies of FinancialServicesSeveral descriptive studies have focused on retailbanking (Menor and Roth 2008, Menor et al. 2001),


substitution of labor with information technology(Fung 2008), the use of customer feedback to improvecustomer satisfaction (Krishnan et al. 1999), the useof distribution channels (Lee et al. 2004, Xue et al.2007), self-service technologies (such as ATMs,pay at the pump, see Campbell and Frei 2010a, b,Meuter et al. 2000), online banking (Hitt andFrei 2002), and e-services in general (see Boyer et al.2002, Ciciretti et al. 2009, Clemons et al. 2002, Furstet al. 2002, Menor et al. 2001). These studies talk aboutthe types of customers who use the various differentchannels and how firms have diversified their deliv-ery of services using these new channels as newertechnologies have become available. However, theyare usually descriptive, rather than prescriptive, inthat they speak about how existing firms and cus-tomers have already adopted these technologies,rather than what they should be doing in the future.For example, there are few quantitative metrics tomeasure a product (e.g., its complexity vis-a-viscustomer knowledge), a process (e.g., face to face vs.automated), and proximity (on-site or off-site) to helpa manager navigate financial service operations strat-egies from a design standpoint based on where herfirm is now. In that sense, financial service systemdesign still has ways to go to catch up with productdesign (product attributes, customer utility, pricing,form and function, configuration, product develop-ment teams, etc.) and manufacturing process design(process selection, batch/line, capacity planning,rigid/flexible automation, scheduling, location analy-sis, etc.). Because batching and lot sizing issues havebeen of considerable interest in the history of thestudy of manufacturing processes, and because onlinetechnologies have made the concept of batching con-siderably less important, it would be interesting to seehow research in service systems design unfolds in thefuture. One paper with prescriptive recommendationsfor service design in the property casualty insuranceindustry is due to Giloni et al. (2003).

3. Financial Services PerformanceMeasurement and Analysis

3.1. Best Practices and Process ImprovementMany service firms are measuring success by factorsother than profitability, using such factors as customerand employee loyalty, as measured by retention,depth of relationship, and lifetime value (Heskettet al. 1994). Chen and Hitt (2002), in an empiricalstudy on retention in the online brokerage industry,found that ease of use, breadth of offerings, and qual-ity reduce customer attrition. Balasubramanian et al.(2003) find that trust is important for online transac-tions, because physical appearance of branches, etc.no longer matter in such situations. Instead, perceived

environmental security, operational competence, andquality of service help create trust.

In general, service quality is difficult to manage andmeasure because of the variability in customer expec-tations, their involvement in the delivery of theservice, etc. In general, there may be two differentmeasures of service quality that are commonly used:the first refers to and measures the actual service pro-vided (e.g., customer satisfaction, resolution, etc.), thesecond may refer to the availability of service capac-ity/personnel (e.g., service level, availability, waitingtime, etc.). The first type of quality measure is not asnebulous in financial services where the output isgenerally related to monetary outcomes. If there is anerror in the posting of a transaction, or if quarterlyreturns from a mutual fund are below industry per-formance, there is an immediate customer reactionand the points in the service design that caused suchfailures to occur is apparent, whether it is in remit-tance processing or in the hiring of a fund manager.Quality in financial services is not influenced by suchmatters as the mood of the customer, as may be thecase in other services. This makes ensuring quality infinancial services more doable and one of the foci ofthe research in operational risk management whichwe will discuss later.

Roth and Jackson (1995) found that market intelli-gence and imitation of best practices can be aneffective way of improving service quality, and thatservice quality is more influenced by service processchoices and the cumulative impact of investmentsthan by people’s capabilities. Productivity measure-ment in services is also a challenge (Sampson andFroehle 2006). Bank performance as a result of processvariation has been studied by Frei et al. (1999).

This current special issue of Production and Opera-tions Management provides some interesting newcases of process improvement in financial services.The paper by Apte et al. (2010), ‘‘Analysis andimprovement of information-intensive services: Evi-dence from insurance claims handling operations,’’presents a classification of information-intensiveservices based on their operational characteristics;this paper proposes an empirically grounded concep-tual analysis and prescriptive frameworks that can beused to improve the performance of information- andcustomer contact-intensive services. The paper by DeAlmeida Filho et al. (2010) focuses on collection pro-cesses in consumer credit. They develop a dynamicprogramming model to optimize the collections pro-cess in consumer credit. Collection processes havebeen the Cinderella of consumer lending research,because psychologically lenders do not enjoy analyz-ing their mistakes, and also once an accounting loss isascribed to a defaulted loan, there had been littleincentive for senior managers to keep track of how


much will be subsequently collected. The paper byBuell et al. (2010) investigates why self-servicecustomers are more reluctant to change their serviceprovider. This paper’s primary contribution is toinvestigate how satisfaction and switching costs con-tribute to retention among self-service customers. Thisis a particularly important issue in the financial ser-vices industry where considerable investments havebeen made in developing self-service distributionchannels, and migrating customers to them.

3.2. An Example of Best Practices: AssetManagementAsset management provides an interesting example ofan area within the financial services sector that hasbeen receiving an increasing amount of researchattention with regard to best practices from variousoperations management perspectives. The body ofresearch on operations management in asset manage-ment is growing, however, not always produced byoperations management researchers, but often bythose in the finance world (Black 2007, Brown et al.2009a, b, Kundro and Feffer 2003, 2004, Stulz 2007),who examine operational risk issues in hedge funds.A collection of operations management researchpapers in asset management can be found in a recentbook by Pinedo (2010). Alptuna et al. (2010) present abest practices framework for the operational infra-structure and controls in asset management and arguethat it is possible to effectively implement such aframework in organizations that enjoy a strong, prin-ciple-based governance. They examine conditionsunder which the cost-effective strategy of outsourc-ing asset management operations can be successfulfor asset managers and their clients. Figure 1, whichhas been adapted from Alptuna et al. (2010), shows

the multiple constituent parts that must work togetherin order for a typical asset management organizationto function effectively. Figure 2, also adapted fromAlptuna et al. (2010), lists the functions in the invest-ment management process according to their distancefrom the end client. Typically, the operations-intensivefunctions reside in the middle and back offices;accordingly, the untapped research potential of oper-ations in asset management must be sought there. Onecan create a similar framework, as shown in Figures 1and 2, for a typical retail bank, credit card issuer,mortgage lender, brokerage, trust bank, asset custo-dian, life or property/casualty insurer, among others,none of which is less complex than an asset manager.Outsourcing operations adds to the complexity by in-troducing elements of quality control for outsourcedpieces and coordination between the main organiza-tion and the third-party provider (State Street 2009).To develop their framework, Alptuna et al. (2010)draw heavily on asset management industry resourceson best practices, namely the Managed Funds Associ-ation’s Sound Practices for Hedge Fund Managers(2009), the Report of the Asset Managers’ Committee tothe President’s Working Group on Financial Markets(2009), the Alternative Investment Management Asso-ciation’s Guide to Sound Practices for European HedgeFund Managers (2007), and the CFA Institute’s AssetManager Code of Professional Conduct (2009).

Schneider (2010) provides a framework for assetmanagement firms to analyze their costs. Arfelt (2010)proposes an adaptation of the Lean Six Sigma frame-work used in automobile manufacturing for assetmanagement. Biggs (2010) advocates a decentraliza-tion of risk management, accountability as well astechnology and expense control in asset managementfirms. Cruz (2010) argues that the focus of cost man-

Asset managemento Investment research, management,

and execution o

o

o

Sales and client relationship managementProduct development

Marketing

Independent internal oversight functions

o Compliance, legal and regulatory o Controllers o Credit and market risk

management o Internal audit o Valuation oversight

Internal support teamso Billing o Human resources o Operations o Operational risk o Performance o Tax o Technology o Treasury

External service providerso Brokerage, clearing and execution o Custody and trust services o Fund administrator o Prime brokerage and financing o Reputable auditor o Valuation (reputable third-party

valuation firm)

Figure 1 Typical Structure of an Asset Management Organization


agement programs at asset management firms shouldbe strategic and tactical (see also Cruz and Pinedo2009). Nordgard and Falkenberg (2010) give an ITperspective on costs in asset management. Campbelland Frei (2010a) examine cost structure patternsin the asset management industry. Amihud andMendelson (2010) examine the effect of transactioncosts on asset management, and study their implica-tions for portfolio construction, fund design, tradeimplementation, cash and liquidity management, andcustomer acquisition and development strategies.

3.3. Performance Analysis Through DataEnvelopment Analysis (DEA)There are numerous studies on performance and pro-ductivity analyses of retail banking that are based onDEA. DEA is a technique for evaluating productivitymeasures that can be applied to service industries ingeneral. It compares productivity measures of differ-ent entities (e.g., bank branches) within the sameservice organization (e.g., a large retail bank) to oneanother. Such a comparative analysis then boils downto the formulation of a fractional linear program. DEAhas been used in many retail banks to compareproductivity measures of the various branches withone another. Sherman and Gold (1985), Sherman andLadino (1995), and Seiford and Zhu (1999) performedsuch studies for US banks; Oral and Yolalan (1990)performed such a study for a bank in Turkey; Vassi-loglou and Giokas (1990), Soteriou and Zenios (1999a),

Zenios et al. (1999), Soteriou and Zenios (1999b), andAthanassopoulos and Giokas (2000) for Greek banks;Kantor and Maital (1999) for a large Mideast bank;and Berger and Humphrey (1997) for various inter-national financial services firms. These papers discussoperational efficiency, profitability, quality, stock mar-ket performance, and the development of better costestimates for banking products via DEA. Cumminset al. (1999) use DEA to explore the impact oforganizational form on firm performance. They com-pare mutual and stock property liability companiesand find that in using managerial discretion and cost-efficiency stock companies perform better, and in linesof insurance with long payouts mutual companiesperform better.

Cook and Seiford (2009) present an excellent over-view of the DEA developments over the past 30 yearsand Cooper et al. (2007) provide a comprehensivetextbook on the subject. For a good survey andcautionary notes on the pitfalls of improper interpre-tation and use of DEA results (e.g., loosely using theresults for evaluative purposes when uncontrollablevariables exist), see Metters et al. (1999). Zhu (2003)discusses methods to solve imprecise DEA (IDEA),where data on inputs and outputs are either bounded,ordinal, or ratio bounded, where the original linearprogramming DEA formulation can no longer be used.

Koetter (2006) discusses the stochastic frontieranalysis (SFA) as another bank efficiency analysisframework, which contrasts to the deterministic DEA.

Asset management - Investment research - Portfolio and risk

management -

-

Sales and client relationshipmanagementProduct development

Trade execution - Financial

InformationeXchange (FIX) connectivity

- Trade order management and execution

Middleoffice

Investment operations - Billing - Cash administration - Client data warehouse - Client reporting

- Corporate actions processing

- Data management - OTC derivatives

processing

- Performance and analytics

- Portfolio recordkeepingand accounting

- Reconciliation processing

- Transaction management

Back office Fund accounting - Daily, monthly, and ad-

hoc reporting - General ledger - NAV calculation - Reconciliation - Security pricing

Global custody - Assets safekeeping - Cash availability - Failed trade

reporting- Income/tax

reclaims- Reconciliation - Trade settlement

Transfer agency - Shareholder

servicing

Frontoffice

Figure 2 Investment Management Process Functions


4. ForecastingForecasting is very important in many areas of thefinancial services industry. In its most familiar form inwhich it presents itself to customers and the generalpublic, it consists of economic and market forecastsdeveloped by research and strategy groups in broker-age and investment management firms. However, thetypes of forecasting we discuss tend to be more inter-nal to the firms and not visible from the outside.

4.1. Forecasting in the Management of CashDeposits and Credit LinesDeposit-taking institutions (e.g., commercial banks,savings and loan associations, and credit unions) areinterested in forecasting the future growth of theirdeposits. They use this information in the process ofdetermining the value and pricing of their depositproducts (e.g., checking, savings, and money marketaccounts, and also CDs), for asset–liability manage-ment, and for capacity considerations. Of specialinterest to these institutions are demand deposits,more broadly defined as non-maturity deposits.Demand deposits have no stated maturity and thedepositor can add to the balance without restriction,or withdraw from ‘‘on demand,’’ i.e., without warningor penalty. In contrast, time deposits, also known asCDs, have a set maturity and an amount establishedat inception, with penalties for early withdrawals.Forecasting techniques have been applied to demanddeposits because of their relative non-stickiness due tothe absence of contractual penalties. A product withsimilar non-stickiness is credit card loans. Jarrow andVan Deventer’s (1998) model for valuing demanddeposits and credit card loans using an arbitrage-freemethodology assumes that demand deposit balancesdepend only on the future evolution of interest rates;however, it does allow for more complexity, such asmacroeconomic variables (income or unemployment),and local market or firm-specific idiosyncratic factors.Janosi et al. (1999) use a commercial bank’s demanddeposit data and aggregate data for negotiable orderof withdrawal (NOW) accounts from the FederalReserve to empirically investigate Jarrow and Van Dev-enter’s model. They find demand deposit balances to bestrongly autoregressive, i.e., future balances are highlycorrelated with past balances. They develop regressionmodels, linear in the logarithm of balances, in whichpast balances, interest rates, and a time trend are pre-dictive variables. O’Brien (2000) adds income to the setof predictive variables in the regression models. Shee-han (2004) adds month-of-the-year dummy variables inthe regressions to account for calendar-specific inflows(e.g., bonuses or tax refunds), or outflows (e.g., taxpayments). He focuses on core deposits, i.e., checkingaccounts and savings accounts; distinguishes betweenthe behavior of total and retained deposits; and devel-

ops models for different deposit types, i.e., business andpersonal checking, NOW, savings, and money marketaccount deposits.

Labe and Papadakis (2010) discuss a propensityscore matching model that can be used to forecast thelikelihood of Bank of America’s retail clients bringing innew funds to the firm by subscribing to promotionalofferings of CDs. Such promotional CDs carry anabove-market premium rate for a limited period oftime. Humphrey et al. (2000) forecast the adoption ofelectronic payments in the United States; they find thatone of the reasons for the slow pace of moving fromchecks to electronic payments in the United States is thecustomers’ perceived loss of float. Many electronic pay-ment systems now address this, by allowing forpayment at the due date rather than immediately.

Revolving credit lines, or facilities, give borrowersaccess to cash on demand for short-term funding needsup to credit limits established at facility inception. Bankstypically offer these facilities to corporations with in-vestment grade credit ratings, which have access tocheaper sources of short-term funding, for example,commercial paper, and do not draw significant amountsfrom them except:

(i) for very brief periods of time under normalconditions,

(ii) when severe deterioration of their financialcondition causes them to lose access to thecredit markets, and

(iii) during system-wide credit market dysfunction,such as during the crisis of 2007–2009.

Banks that offer these credit facilities must set asideadequate, but not excessive, funds to satisfy the de-mand for cash by facility borrowers. Duffy et al. (2005)describe a Monte Carlo simulation model that MerrillLynch Bank used to forecast these demands for cashby borrowers of their revolver portfolio. The modeluses industry data for revolver usage by borrowercredit rating, and assumes Markovian credit ratingmigrations, correlated within and across industries.Migration probabilities were provided by a majorrating agency, and correlation estimates were calcu-lated by Merrill Lynch’s risk group. The model wasused by Merrill Lynch Bank to help manage liqui-dity risk in its multi-billion portfolio of revolvingcredit lines.

Forecasting the future behavior and profitability ofretail borrowers (e.g., for credit card loans, mortgages,and home equity lines of credit) has become a keycomponent of the credit management process. Fore-casting involved in a decision to grant credit to a newborrower is known as ‘‘credit scoring,’’ and its originsin the modern era can be found in the 1950s. A dis-cussion of credit scoring models, including relatedpublic policy issues, is offered by Capon (1982). Fore-


casting involved in the decisions to adjust credit ac-cess and marketing effort to existing borrowers isknown as ‘‘behavioral scoring.’’ The book by Thomaset al. (2002) contains a comprehensive review of theobjectives, methods, and practical implementation ofcredit and behavioral scoring. The formal statisticalmethods used for classifying credit applicants into‘‘good’’ and ‘‘bad’’ risk classes is known as ‘‘classifi-cation scoring.’’ Hand and Henley (1997) reviewa significant part of the large body of literature inclassification scoring. Baesens et al. (2003) examinethe performance of standard classification algorithms,including logistic regression, discriminant analysis,k-nearest neighbor, neural networks, and decision trees;they also review more recently proposed ones, such assupport vector machines and least-squares supportvector machines (LS-SVM). They find LS-SVM, andneural network classifiers, and simpler methods such aslogistic regression and linear discriminant analysis tohave good predictive power. In addition to classifica-tion scoring, other methods include:

(i) ‘‘response scoring,’’ which aims to forecast aprospect’s likelihood to respond to an offer forcredit, and

(ii) ‘‘balance scoring,’’ which forecasts the pros-pect’s likelihood of carrying a balance if theyrespond.

To improve the chances of acquiring and maintainingprofitable customers, offers for credit should be mailedonly to prospects with high credit, response, and bal-ance scores. Response and balance scoring models aretypically proprietary. Trench et al. (2003) discuss amodel for optimally managing the size and pricing ofcard lines of credit at Bank One. The model usesaccount-level historical transaction information to selectfor each cardholder, through Markov decision processes,annual percentage rates, and credit lines that optimizethe net present value of the bank’s credit portfolio.

4.2. Forecasting in Securities Brokerage, Clearing,and ExecutionIn the last few decades, the securities brokerageindustry has seen dramatic change. Traditional wire-houses charging fixed commissions evolved or werereplaced by diverse organizations offering full service,discount, and online trading channels, as well as re-search and investment advisory services. Thisevolution has introduced a variety of channel choicesfor retail and institutional investors. Pricing, servicemix and quality, and human relationships are keydeterminants in the channel choice decision. Firms areinterested in forecasting channel choice decisions byclients, because they greatly impact capacity planning,revenue, and profitability. Altschuler et al. (2002) dis-cuss simulation models developed for Merrill Lynch’s

retail brokerage to forecast client choice decisions onintroduction of lower-cost offerings to complementthe firm’s traditional full-service channel. Clientchoice decision forecasts were used as inputs in theprocess of determining the proper pricing for thesenew offerings and for evaluating their potentialimpact on firm revenue. The results of a rational eco-nomic behavior (REB) model were used as a baseline.The REB model assumes that investors optimize theirvalue received by always choosing the lowest-costoption (determined by an embedded optimizationmodel that was solved for each of millions of clientsand their actual portfolio holdings). The REB model’sresults were compared with those of a Monte Carlosimulation model. The Monte Carlo simulation allowsfor more realistic assumptions. For example, clients’decisions are impacted not only by price differentialsacross channels, but also by the strength and qualityof the relationship with their financial advisor, whorepresented the higher-cost options.

Labe (1994) describes an application of forecastingthe likelihood of affluent prospects becoming MerrillLynch’s priority brokerage and investment advisoryclients (defined as clients with more than US$250,000in assets). Merrill Lynch used discriminant analysis, amethod akin to classification scoring, to select highquality households to target in its prospecting efforts.

The trading of securities in capital markets involveskey operational functions that include:

(i) clearing, i.e., establishing mutual obligations ofcounterparties in securities and/or cash trades, aswell as guarantees of payments and deliveries,and

(ii) settlement, i.e., transfer of titles and/or cash tothe accounts of counterparties in order to final-ize transactions.

Most major markets have centralized clearingfacilities so that counterparties do not have to settlebilaterally and assume credit risk to each other. Thecentral clearing organization must have robust pro-cedures to satisfy obligations to counterparties, i.e.,minimize the number of trades for which delivery ofsecurities is missed. It must also hold adequate, butnot excessive, amounts of cash to meet payments.Forecasting the number and value of trades during aclearing and settlement cycle can help the organiza-tion meet the above objectives; it can achieve this bymodeling the clearing and settlement operation usingstochastic simulation. A different approach is used byde Lascurain et al. (2011): they develop a linear pro-gramming method to model the clearing and settlementoperation of the Central Securities Depository ofMexico and evaluate the system’s performance throughdeterministic simulation. The model’s formulation in deLascurain et al. (2011) is a relaxation of a mixed integer


programming (MIP) formulation proposed by Güntzeret al. (1998), who show that the bank clearing problemis NP-complete. Eisenberg and Noe (2001) includeclearing and settlement in a systemic financial riskframework.

4.3. Forecasting of Call Arrivals at Call CentersForecasting techniques are also used in various otherareas within the financial services sector; for example,the forecasting of arrivals in call centers, which is acrucial input in the personnel scheduling processin call centers (to be discussed in a later section). Toset this in a broader context, refer to the framework ofThompson (1998) for forecasting demand for services.Recent papers that focus on forecasting call centerworkload include a tutorial by Gans et al. (2003), asurvey by Aksin et al. (2007), and a research paper byAldor-Noiman et al. (2009).

The quality of historical data improves with thepassage of time because call centers become increas-ingly more sophisticated in capturing data with everynuance that a modeler may find useful or interesting.Andrews and Cunningham (1995) describe the auto-regressive integrated moving average (ARIMA) fore-casting models used at L.L. Bean’s call centers. Timeseries data used to fit the models exhibit seasonalitypatterns and are also influenced by variables such asholiday and advertising interventions. Advertisingand special calendar effects are addressed by Antipovand Meade (2002). More recently, Soyer andTarimcilar (2008) incorporate advertising effects bymodeling call arrivals as a modulated Poisson processwith arrival rates being driven by customer calls thatare stimulated by advertising campaigns. They use aBayesian modeling framework and a data set from acall center that enables tracing calls back to specificadvertisements. In a study of Fedex’s call centers,Xu (2000) presents forecasting methodologies usedat multiple levels of the business decision-makinghierarchy, i.e., strategic, business plan, tactical, andoperational, and discusses the issues that each meth-odology addresses. Methods used include exponentialsmoothing, ARIMA models, linear regression, andtime series decomposition.

At low granularity, call arrival data may have toomuch noise. Mandelbaum et al. (2001) demonstratehow to remove relatively unpredictable short-termvariability from data and keep only predictable vari-ability. They achieve this by aggregating data athigher levels of granularity, i.e., progressively mov-ing up from minute of the hour to hour of the dayto day of the month and to month of the year. Theelegant textbook assumption that call arrivals follow aPoisson process with a fixed rate that is known or canbe estimated does not hold in practice. Steckley et al.(2009) show that forecast errors can be large in com-

parison to the variability expected in a Poissonprocess and can have significant impact on the pre-dictions of long-run performance; ignoring forecasterrors typically leads to overestimation of perfor-mance. Jongbloed and Koole (2001) found that the callarrival data they had been analyzing had a variancemuch greater than the mean, and therefore did notappear to be samples of Poisson distributed randomvariables. They addressed this ‘‘overdispersion’’ byproposing a Poisson mixture model, i.e., a Poissonmodel with an arrival rate that is not fixed but ran-dom following a certain stochastic process. Brownet al. (2005) found data from a different call center thatalso followed a Poisson distribution with a variablearrival rate; the arrival rates were also serially corre-lated from day to day. The prediction model proposedincludes the previous day’s call volume as an auto-regressive term. High intra-day correlations werefound by Avramidis et al. (2004), who developedmodels in which the call arrival rate is a randomvariable correlated across time intervals of the sameday. Steckley et al. (2004) and Mehrotra et al. (2010)examine the correlation of call volumes at later peri-ods of a day to call volumes experienced earlier in theday for the purpose of updating workload schedules.

Methods to approximate non-homogeneous Pois-son processes often attempt to estimate the arrival rateby breaking up the data set into smaller intervals.Henderson (2003) demonstrates how a heuristic thatassumes a piecewise constant arrival rate over timeintervals with a length that shrinks as the volumeof data grows produces good arrival rate functionestimates. Massey et al. (1996) fit piecewise linear ratefunctions to approximate a general time-inhomoge-neous Poisson process. Weinberg et al. (2007) forecastan inhomogeneous Poisson process using a Bayesianframework, whereby from a set of prior distributionsthey estimate the parameters of the posterior dis-tribution through a Monte Carlo Markov Chainmodel. They forecast arrival rates in short intervalsof 15–60 minutes of a day of the week as the productof a day’s forecast volume times the proportion ofcalls arriving during an interval; they also allow for arandom error term.

Shen and Huang (2005, 2008a, b) developed modelsfor inter-day forecasting and intra-day updating ofcall center arrivals using singular value decomposi-tion. Their approach resulted in a significant dimen-sionality reduction. In a recent empirical study, Taylor(2008) compared the performance of several univari-ate time series methods for forecasting intra-day callarrivals. Methods tested included seasonal autore-gressive and exponential smoothing models, and thedynamic harmonic regression of Tych et al. (2002).Results indicate that different methods perform bestunder different lead times and call volumes levels.


Forecasting other aspects of a call center with asignificant potential for future research include, forexample, waiting times of calls in queues, see Whitt(1999a, b) and Armony and Maglaras (2004).

5. Inventory and Cash Management

5.1. Cash Inventory Management UnderDeterministic and Stochastic DemandOrganizations, households, and individuals need cashto meet their liquidity needs. In the era of checks andelectronic transactions, an amount of cash does nothave to be in physical currency, but may correspondonly to a value in an account that has been set up forthis purpose. To meet short-term liquidity needs, cashmust be held in a riskless form, where its value doesnot fluctuate and is available on demand, but earnslittle or no interest. Treasury bills and checkingaccounts are considered riskless. Cash not needed tomeet short-term liquidity needs can be invested inrisky assets whereby it may earn higher returns, butits value may be subject to significant fluctuations anduncertainty, and could become wholly or partiallyunrecoverable. Depending on the type of risky asset,its value may or may not be quickly recoverable andrealizable at a modest cost (as with a public equitythat is listed in a major stock exchange). Determiningthe value of certain types of risky assets (e.g., privateequity, real estate, some hedge funds, and asset-backed fixed income securities) may require special-ized valuation services, which could involvesignificant time and cost. Risky assets can also besubject to default, in which case all or part of the valuebecomes permanently unrecoverable.

Researchers have produced over the last few decadesa significant body of work by applying the principles ofinventory theory to cash management. We review thecash management literature from its beginnings sowe can put later work in context, and we have notidentified an earlier comprehensive review that accom-plishes this purpose. Whistler (1967) discussed astochastic inventory model for rented equipment thatwas formulated as a dynamic program; this workserved as a model for the cash management problem.One of the early works produced an elegant result thatbecame known as the Baumol–Tobin economic modelof the transactions demand for money, independentlydeveloped by Baumol (1952) and Tobin (1956).The model assumes a deterministic, constant rate ofdemand for cash. It calculates the optimal ‘‘lot sizes’’ ofthe risky asset to be converted to cash, or the optimalnumbers of such conversions, in the presence of trans-action and interest costs. Tobin’s version requires aninteger number of transactions and therefore approxi-mates reality more closely than Baumol’s, which allowsthat variable to be continuous.

The concept of transactions demand for money,addressed by the Baumol–Tobin model, is related to,but subtly different from, precautionary demand forcash that applies to unforeseen expenditures, oppor-tunities for advantageous purchases, and uncertaintyin receipts. Whalen (1966) developed a model with astructure strikingly similar to the Baumol–Tobinmodel, capturing the stochastic nature of precaution-ary demand for cash. Sprenkle (1969) and Akerlof andMilbourne (1978) observed that the Baumol–Tobinmodel tends to under-predict demand for money,partly because it fails to capture the stochastic natureof precautionary demand for cash. Sprenkle’s paperelicited a response by Orr (1974), which in turnprompted a counter-response by Sprenkle (1977).

Robichek et al. (1965) propose a deterministic short-term financing model that incorporates a great degreeof realistic detail involved in the financial officer’sdecision-making process, which they formulate andsolve as a linear program. They include a discussionon model extensions for solving the financing prob-lem under uncertainty. Sethi and Thompson (1970)proposed models based on mathematical control the-ory, in which demand for cash is deterministic butdoes vary with time. In an extension of the Sethi–Thompson model, Bensoussan et al. (2009) allow thedemand for cash to be satisfied by dividends anduncertain capital gains of the risky asset, stock.

In what became known as the Miller–Orr modelfor cash management, Miller and Orr (1966) extendedthe Baumol–Tobin model by assuming the demand forcash to be stochastic. The cash balance can fluctuaterandomly between a lower and an upper boundaccording to a Bernoulli process, and transactions takeplace when it starts moving out of this range; units ofthe risky asset are converted into cash at the lowerbound, and bought with the excess cash at the upperbound. Transaction costs were assumed fixed,i.e., independent of transaction size. In a critique ofthe Miller–Orr model, Weitzman (1968) finds it to be‘‘robust,’’ i.e., general results do not change muchwhen the underlying assumptions are modified.

Eppen and Fama (1968, 1969, 1971) proposed cashbalance models that are embedded in a Markovianframework. In one of their papers, Eppen and Fama(1969) presented a stochastic model, formulated as adynamic program, with transaction costs proportionalto transaction sizes. Changes in the cash balance canfollow any discrete and bounded probability distribu-tion. In another one of their papers, Eppen and Fama(1968) developed a general stochastic model thatallowed costs to have a fixed as well as a variable com-ponent. They showed how to find optimal policies forthe infinite-horizon problem using linear programming.In their third paper, Eppen and Fama (1971) proposed astochastic model with two risky assets, namely ‘‘bonds’’


and ‘‘stock’’; the stock is more risky but has a higherexpected return. They also discussed using ‘‘bonds’’(the intermediate-risk asset) as a ‘‘buffer’’ between cashand the more risky asset. Taking a similar approach,Daellenbach (1971) proposed a stochastic cash balancemodel using two sources of short-term funds. Girgis(1968) and Neave (1970) presented models with bothfixed and proportional costs and examined conditionsfor policies to be optimal under different assumptions.Hausman and Sanchez-Bell (1975) and Vickson (1985)developed models for firms facing a compensating-balance requirement specified as an average balanceover a number of days.

Continuous-time formulations of the cash manage-ment problem were based on the works of Antelmanand Savage (1965), and Bather (1966), who used aWiener process to generate a stochastic demand intheir inventory problem formulations. Their approachwas extended to cash management by Vial (1972),whose continuous-time formulation had fixed andproportional transaction costs, and linear holding andpenalty costs, and determined the form of the optimalpolicy (assuming one exists). Constantinides (1976)extended the model by allowing positive and negativecash balances, determined the parameters of theoptimal policy, and discussed properties of theoptimal solution. Constantinides and Richard (1978)formulated a continuous-time, infinite-horizon,discounted-cost cash management model with fixedand proportional transaction costs, linear holding andpenalty costs, and the Wiener process as thedemand-generating mechanism. They proved that therealways exists an optimal policy for the cash manage-ment problem, and that this policy is of a simple form.Smith (1989) developed a continuous-time model witha stochastic, time-varying interest rate.

5.2. Supply Chain Management of PhysicalCurrencyPhysical cash, i.e., paper currency and coins, remainsan important component of the transactions volumeeven in economies that have experienced a significantgrowth in checks, credit, debit and smart cards, andelectronic transactions. Advantages of cash includeease of use, anonymity, and finality; it does notrequire a bank account; it protects privacy by leavingno transaction records; and it eliminates the need toreceive statements and pay bills. Disadvantagesof cash include ease of tax evasion, support of an‘‘underground’’ economy, risk of loss through theft ordamage, ability to counterfeit, and unsuitability foronline transactions.

Central banks provide cash to depository institu-tions, which in turn circulate it in the economy. Thereare studies on paper currency circulation in variouscountries. For example, Fase (1981) and Boeschoten

and Fase (1992) present studies by the Dutch centralbank on the demand for banknotes in the Netherlandsbefore the introduction of the Euro. Ladany (1997)developed a discrete dynamic programming model todetermine optimal (minimum cost) ordering policiesfor banknotes by Israel’s central bank. Massoud (2005)presents a dynamic cost minimizing note inventorymodel to determine optimal banknote order size andfrequency for a typical central bank.

The production and distribution of banknotes, andthe required infrastructure and processes, have alsobeen studied. Fase et al. (1979) discuss a numericalplanning model for the banknotes operations at acentral bank, with examples from pre-Euro Nether-lands. Bauer et al. (2000) develop optimization modelsfor determining the least-cost configuration of the USFederal Reserve’s currency processing sites given thetrade-off between economies of scale in processingand transportation costs. In a study of costs and econ-omies of scale of the US Federal Reserve’s currencyoperations, Bohn et al. (2001) find that the FederalReserve is not a natural monopoly. Opening currencyoperations to market competition and charging feesand penalties for some services provided for free bythe Federal Reserve at that time could lead to moreefficient allocation of resources.

The movement of physical cash among centralbanks, depository institutions, and the public must bestudied as a closed-loop supply chain (see, e.g.,Dekker et al. 2004). It involves the recirculation ofused notes back into the system (reverse logistics),together with a flow of new notes from the centralbank to the public through depository institutions(forward logistics). The two movements are so inter-twined that they cannot be decoupled. Rajamani et al.(2006) study the cash supply chain structure in theUnited States, analyze it as a closed-loop supplychain, and describe the cash flow management systemused by US banks. They also discuss the new cashrecirculation policies adopted by the Federal Reserveto discourage banks’ overuse of its cash processingservices, and encourage increased recirculation at thedepository institution level. Among the practices to bediscouraged was ‘‘cross shipping,’’ i.e., shipping usedcurrency to the Federal Reserve and ordering it in thesame denominations in the same week. To compareand contrast new and old Federal Reserve policies forcurrency recirculation, Geismar et al. (2007) introducemodels that explain the flow of currency between theFederal Reserve and banks under both sets of guide-lines. They present a detailed analysis that providesoptimal policies for managing the flow of currencybetween banks and the Federal Reserve, and analyzebanks’ responses to the new guidelines to help theFederal Reserve understand their implications.Dawande et al. (2010) examine the conditions that


can induce depository institutions to respond in so-cially optimal ways according to the new FederalReserve guidelines. Mehrotra et al. (2010), which is apaper in this special issue of Production and OperationsManagement, address the problem of obtaining effi-cient cash management operating policies fordepository institutions under the new Federal Re-serve guidelines. The mixed-integer programmingmodel developed for this purpose seeks to find‘‘good’’ operating policies, if such exist, to quantifythe monetary impact on a depository institution op-erating according to the new guidelines. Anotherobjective was to analyze to what extent the newguidelines can discourage cross shipping and stimu-late currency recirculation at the depositoryinstitution level. Mehrotra et al. (2010a) study pricingand logistics schemes for services such as fit-sortingand transportation that can be offered by third-partyproviders as a result of the Federal Reserve’s newpolicies.

5.3. Other Cash Management Applications inBanking and Securities BrokerageUS banks are required to keep on reserve a minimumpercentage (currently 10%) of deposits in client trans-action accounts (demand deposits and othercheckable deposits) at the Federal Reserve. Until veryrecently, banks had a strong incentive to keep fundson reserve at a minimum, because these funds wereearning no interest. Even after the 2006 Financial Ser-vices Regulatory Relief Act became law, authorizingpayment of interest on reserves held at the FederalReserve, banks prefer to have funds available for theirown use rather than have them locked up on reserve.Money market deposit accounts (MMDA) with check-ing allow banks to reduce the amounts on reserve atthe Federal Reserve by keeping deposits in MMDAaccounts and transferring to a companion checkingaccount only the amounts needed for transactions.Only up to six transfers (‘‘sweeps’’) per month areallowed from an MMDA to a checking account, and aclient’s number and amount of transactions in thedays remaining in a month is unknown. Therefore, thesize and timing of the first five sweeps must be care-fully calculated to avoid a sixth sweep, which willmove the remaining MMDA balance into the checkingaccount. Banks have been using heuristic algorithmsto plan the first five sweeps. This specialized inven-tory problem has been examined by Nair andAnderson (2008), who propose a stochastic dynamicprogramming model to optimize retail accountsweeps. The stochastic dynamic programming modeldeveloped by Nair and Hatzakis (2010) introducescushions added to the minimum sweep amounts. Itdetermines the optimal cushion sizes to ensurethat sufficient funds are available in the transaction

account in order to cover potential future transactionsand avoid the need for a sixth sweep.

The impact of the sequence of transaction postingson account balances and resultant fees for insufficientfunds, similar to the cost of stock-outs in inventorymanagement, has been studied by Apte et al. (2004).They investigate how overdraft fees and non-sufficient funds (NSF) fees interact in such situations.

Brokerage houses make loans to investors who wantto use leverage, i.e., to invest funds in excess of theirown capital in risky assets, and can pledge securitiesthat they own as collateral. In a simple application ofthis practice, known as margin lending, the brokerageextends a margin loan to a client of up to the valueof equity securities held in the client’s portfolio. Theclient can use the loaned funds to buy more equitysecurities. Calculating the minimum value required ina client’s account for a margin loan can become com-plex in accounts holding different types of securitiesincluding equities, bonds, and derivatives, all withdifferent margin requirements. The complexity in-creases even more with the presence of long and shortpositions and various derivative strategies practicedby clients. Rudd and Schroeder (1982) presented asimple transportation model formulation for calculat-ing the minimum margin, which represented animprovement over the heuristics used in practice. Asignificant body of subsequent work has been pub-lished on this problem, especially by Timkovsky andcollaborators, which is more related to portfolio strat-egies and hedging. We believe that the approach in thepaper by Fiterman and Timkovsky (2001), which isbased on 0–1 knapsack formulations, is methodolog-ically the most relevant to mention in this overview.

6. Waiting Line Management in RetailBanks and in Call Centers

6.1. Queueing Environments and ModelingAssumptionsIn financial services, in particular in retail banking,retail brokerage, and retail asset management (pen-sion funds, etc.), queueing is a common phenomenonthat has been analyzed thoroughly. Queueing occursin the branches of retail banks with the tellers beingthe servers, at banks of ATM machines with themachines being the servers, and in call centers, wherethe operators and/or the automated voice responseunits are the servers. These diverse queueing envi-ronments turn out to be fairly different from oneanother, in particular with regard to the followingcharacteristics:

(i) the information that is available to the cus-tomer and the information that is available tothe service system,


(ii) the flexibility of the service system with regardto adjustments in the number of serversdependent on the demand,

(iii) the order of magnitude of the number of servers.

Even though in the academic literature the arrivalprocesses in queueing systems are usually assumed tobe stationary (time-homogeneous) Poisson processes,arrival processes in practice are more appropriatelymodeled as non-homogeneous Poisson processes. Overthe last couple of decades, some research has been doneon queues that are subject to non-homogeneous Poissoninputs (see, e.g., Massey et al. 1996). The more theoret-ical research in queueing has also focused on variousaspects of customer behavior in queue, in particularabandonment, balking, and reneging. For example,Zohar et al. (2002) have modeled the adaptive behav-ior of impatient customers and Whitt (2006) developedfluid models for many-server queues with abandon-ment. In all three queueing environments describedabove, the psychology of the customers in the queuealso plays a major role. A significant amount of researchhas been done on this topic, see Larson (1987), Katz et al.(1991), and Bitran et al. (2008). As it turns out, reducingwait times may not always be the best approach in allservice encounters. For example, in restaurants andsalons, longer service time may be perceived as betterservice. In many cases, customers do not like waiting,but when it comes their turn to be served, would likethe service to take longer. In still others, for example, ingrocery checkout lines, customers want a businesslikepace for both waiting and service. The latter category,which we may call dispassionate services, are more com-mon in financial service situations, though the former,which we may call hedonic services, are also present—forexample, when a customer visits their mortgage brokeror insurance agent, they would not like to be rushed. Inthe following subsections, we consider the variousdifferent queueing environments in more detail.

6.2. Waiting Lines in Retail Bank Branches and atATMsThe more traditional queues in financial servicesare those in bank branches feeding the tellers. Sucha queue is typically a single line with a number ofservers in parallel. There are clearly no priorities insuch a queue and the discipline is just first come firstserved. Such a queueing system is typically modeledas an M/M/s system and is discussed in many stan-dard queueing texts. One important aspect of thistype of queueing in a branch is that managementusually can adjust the number of available tellersfairly easily as a function of customer demand andtime of day. (This gives rise to many personnel sched-uling issues that will be discussed in the next section.)

In the early 1980s retail banks began to make ex-tensive use of ATMs. The ATMs at a branch of a bankbehave quite differently from the human tellers. Incontrast to a teller environment, the number of ATMsat a branch is fixed and cannot be adjusted as afunction of customer demand. However, the tellerenvironment and the ATM environment do havesome similarities. In both environments, a customercan observe the length of the queue and can, therefore,estimate the amount of time (s)he has to wait. Inneither the teller environment, nor the ATM environ-ment, can the bank adopt a priority system that wouldensure that more valuable customers have a shorterwait. Kolesar (1984) did an early analysis of a branchwith two ATM machines and collected service timedata as well as arrival time data. However, it becameclear very quickly that a bank of ATMs is capable ofcollecting some very specific data automatically (e.g.,customer service times and machine idle times), butcannot keep track of certain other data (e.g., queuelengths, customer waiting times). Larson (1990), there-fore, developed the so-called queue inference engine,which basically provides a procedure for estimatingthe expected waiting times of customers, given theservice times recorded at the ATMs as well as themachine idle times.

6.3. Waiting Lines in Call CentersSince the late 1980s, banks have started to investheavily in call center technologies. All major retailbanks now operate large call centers on a 24/7 basis.Call centers have therefore been the subject of exten-sive research studies, see the survey papers by Pinedoet al. (2000), Gans et al. (2003), and Aksin et al. (2007).The queueing system in a call center is actually quitedifferent from the queueing systems in a teller envi-ronment or in an ATM environment; there are anumber of major differences. First, a customer nowhas no direct information with regard to the queuelength and cannot estimate his waiting time; he mustrely entirely on the information the service systemprovides him. On the other hand, the service organi-zation in a call center has detailed knowledge con-cerning the customers who are waiting in queue. Theinstitution knows of each customer his or her identityand how valuable (s)he is to the bank. The bank cannow put the customers in separate virtual queueswith different priority levels. This new capability hasmade the application of priority queueing systemssuddenly very important; well-known results inqueueing theory have now suddenly become moreapplicable, see Kleinrock (1976). Second, the callcenters are in another aspect quite different from theteller and the ATM environments. The numberof servers in either a teller environment or an ATMenvironment may typically be, say, at most 20,


whereas the number of operators in a call center maytypically be in the hundreds or even in the thousands.In the analysis of call center queues, it is now possibleto apply limit theorems with respect to the number ofoperators, see Halfin and Whitt (1981) and Reed(2009). Third, the banks have detailed informationwith regard to the skills of each one of its operators ina call center (language skills, product knowledge,etc.). This enables the bank to apply skills-based-rout-ing to the calls that are coming in, see Gans and Zhou(2003) and Mehrotra et al. (2009).

In call centers, typically, an enormous amount ofstatistical data is available that is collected automat-ically, see Mandelbaum et al. (2001) and Brown et al.(2005). The data that are collected automatically aremuch more extensive than the data that are collectedin an ATM environment. It includes the waiting timesof the customers, the queue length at each point intime, the proportion of customers that experience nowait at all, and so on.

Lately, many other aspects of queueing in call cen-ters have become the subject of research. This specialissue of Production and Operations Management as wellas another recent special issue contain several suchpapers. For example, Örmeci and Aksin (2010) foc

Documents

PRODUCTION AND OPERATIONS MANAGEMENT POMSweb-static.stern.nyu.edu/om/faculty/pinedo/ofs_undergrad/...(Spohrer and Maglio 2008). A few books and a special issue of Management Science