berger - 2003 - economic effects of technological progress in the banking industry

7/22/2019 Berger - 2003 - Economic Effects of Technological Progress in the Banking Industry

1/37

M O N E Y , CREDIT, ANDBANKING LECTURE

TheEconomic EffectsofTechnologicalProgress:Evidencefrom the Banking Industry

ALLENN. BERGERThis paper examines technological progressand itseffectsin thebankingindustry. Banksareintensive usersofbothIT andfinancial technologies,and have awealth of data available that may behelpful for thegeneralunderstandingof theeffects of technological change. Theresearchsug-gests improvementsincostsandlending capacitydue toimprovementsin back-office technologies, aswell asconsumer benefits from improved front-office technologies.Theresearch also suggests significant overallproductivity increasesintermsofimproved quality and varietyofbankingservices. In addition, theresearch indicates that technological progresslikely helped facilitate consolidation of the industry.

THIS PAPER EXAMINEStheavailable evidenceontech nologi-cal progressand itseffects in thebanking industry. Innovationsininformationprocessing, telecomm unications,andrelated technologiesknown collectively as information technology or IT areoftencreditedwithhelpingfuelstrong growthin theU.S. economy, although questions remain abouttherelative importance ofIT versus other factors. The extensive researchon thebanking industry may helpinthe generalunderstandingabout the effects oftechnologicalchange.The categoryof DepositoryandNondepository Financial Institutionsof which banking is anintegralpartis the mostIT-intensiveindustryintheU.S.asmeasured by the ratioof computer equipment and softwaretovalue added (Triplett and Bosworth 2002,Table2).

The opinions expresseddo notnecessarily reflect thoseofthe Federal Reserve Boardoritsstaff.Theauthor thanksAnaAizcorbe,BobAvery, Paul Bauer, Mark Carey, Steve Cecchetti, Bill Dewald,BobDeYoung, Paul Evans, Scott Frame, Fred Furlong, Geoff Gerdes, Michael Gordy, Diana Hancock, DaveHumphrey, Erik Kiefel, Leora Klapper, Beth Klee, Myron Kwast, Steve Oliner, Dan Nolle, Dan Sichel,Kevin Stiroh, Phil Strahan, Rick Sullivan, Greg Udell, Haluk Unal, Larry White, and seminar participantsat theJournal ofMoney Creditand BankingAnnual LectureatOhio State University, June 2002,forhelpful suggestions,andNate Millerforoutstanding research assistance.A L L E N N . BERGER is a Senior Economist at the Board of Governors of the Federal

eserveSystem andaSenior Fellowatthe Wharton Financial Institutions Center. E-mail:[email protected] Credit and Banking Vol. 35, No.2(April 2003)


2/37

142 : MO NEY. CRED IT, AND BANKING

Banks are also significant users of financial technologies that employ economicand statistical models to create and value new securities, estimate return distribu tions,and make portfolio decisions based on financial data. Examples include financialengineering used to create new financial derivatives, credit risk and market riskmodels employed to improve portfolio management, and modem credit scoring anddiscriminant analysis used to evaluate credit ap plications. These financial technolog-ies often depend heavily on the use of IT to collect, process, and disseminate thedata, as well as on econom ic and statistical models to evaluate the data. Technolog icalprogress in the banking industry is also important because of the key roles ofbanks in providing financing, deposit, and payments services to other sectors ofthe economy.

We assess the effects of technological progress on productivity growth in thebanking industry and on the structure of this industry. The use of a single industrywith relatively homogenous inputs and outputs may help mitigate problems ofcombining data from heterogeneous industries. Research on banking benefits aswell from detailed data on individual firms to specify cost and profit functions andcontrol for differing business conditions when measuring productivity change, scaleeconomies, and other performance indicators. Some special banking data sets alsoallow for observation of specific technological changes and measurement of someof their effects. In addition, detailed information on the scale, geographic spread,and merger and acquisition (M&A) activity of individual banks aid in evaluatingthe effects of technological progress on the structure of the industry, i.e., the extentto which technological progress facilitates industry consolidation.

Study of the banking industry also demonstrates some of the general problemsin measuring the effects of technological progress and how these problems mightbe addressed. For example, to the extent that markets are competitive, the benefitsfrom technological advances in an industry may be competed away and passedthrough to customers or factors of production and not measured as productivityincreases in that industry. As shown below, banks may have essentially givenaw ay the benefits from the ATM technology in the 1980s as the industry becamemore competitive due to deregulation, and rents from market power shifted toconsumers. It has been shown elsewhere how new products and quality improve-ments from technological progress are often neglected in government statisticsand may lead to overstatements of inflation and understatements of productivitygrowth. In banking, there are many new products and quality improvements thatare not easily captured in standard productivity measures, and we show how somemay be measured in alternative ways.The paper is organized as follows. Section 1 shows some background statisticson changes in the banking industry over time. Section 2 reviews microeconomicresearch on examples of technological changes in the banking industry that provide

some potential general inferences about new technologies. Sections 3 and 5 discussthe research on the two main consequences of technological progress in banking


3/37

ALLEN N. BERGER : 143

1. CHANG ES IN THE BANKING INDUSTRY OVER TIMEWe present statistics that illustrate some of the changes in technology, performance,and structure of the banking industry. Table gives data on changes in the structure ofthe commercial banking industry annually over the period 1984-2001, which illus-trate the consolidation of the industry. The total number of banking organizations(top-tier holding companies plus independent banks) and the number of banks sub-stantially declined over the 17 year period at average annual rates of 3.3% and3.4%, respectively, even while gross total assets (GTA) grew by 3.0% per year. Theconsolidation has primarily occurred through mergers and acquisitions (M&As)that combine institutions in diiferent local marketsthe average local market

Herfindahl index has increased by only 1.1% annually.Table also provides data on changes in the use of selected banking technologies,indicating significant growth in the use of new IT and financial technologies. The

TABLE 1T H E STRUCTURE O F

Year198419851986198719881989199019911992199319941995199619971998199920002001AverageGrowthRate

Number ofBankingOrganizations11,43311,09410,58710,1969 8079 5259 2849 0738 7808 3667 9327 6027 3357,1446 8696 7476 6746 578

- 0 . 0 3 3

THE BANKING INDUSTRY AND T HE

Number ofBanks14,39214,27214,05813,56112,98412,56312,20211,82111,35910,87410,3629 8579 4489 0668,7118,5108 2388,016

- 0 . 0 3 4

Gross TotalAssets (GTA){ trillions)3.443.623.803.783.783.843.773.693.673.783.994.184.354.705.045.205.555.690.030

Local MarketHerflndahlIndex0.17110.16850.16730.16930.17370.17620.18140.18790.19280.19900.20130.20230.20610.20790.20800.20730.20530.20500.011

USE OF SELECTED TECHNOLOGIESNumber ofPbysical BankingOffices(thousands)

5051515354545657585859616264656772

0.021

Number ofATMs(thousands)58616468727680848795109123

13916518722727 33240.101

Ctedlt

, 1984-2001InterestCommitments Rate Swaps( trillion

0.910.970.970.99.01.01.46.47.51.64.952.292.613.063.593.734.114.37

0.092

) ( trillions)

0.250.470.891.121.561.891.872.212.994.415.376.728.4512.6315.4018.8121,46

0.279

N O T S Number of Banking Organizations is the number of top-tier holding companies plus the number of independent commercial hanks.Gross Total Assets (GTA) equals total assets plus loan and lease loss reserves and allocated transfer risk reserve (a reserve for certain foreignloans).Local M arket Herfindahl Index is the average bank w eighted average H erfindahl index of local deposit market concentration acrossthe bank 's markets (Metropolitan Statistical Areas [MSAs] or non-MSA rural counties), where each w eight is the bank's deposit share in themarket. Num ber of Physical B anking Offices generally includes bank head offices, full service b ranches, and limited service branches withhuman tellers (varies slightly by state regulator). ATMs is the number of automated teller mac hines. Credit Commitm ents includes financialand performance standby letters of credit, commercial and similar letters of credit, total unused commitmenLs, and participations inacceptances. Interest Rate Swaps is the notional value of all outstanding interest rate swaps (nol available for 1984). Number of BankingOrganizations, N umber of Ban ks, Gross Total Assets (GTA), Physical Banking O ffices, Credit Com mitments, and Interest Rate Sw aps areobtained from the Decem ber Reports of Condition and Income (Call Reports). Local M arket Herfindahl Index is obtained from the Sum mary of


4/37

144 : MONEY, CREDIT, AND BANKING

num ber of physical banking offices using human tellers has expanded at a 2 ,1 %annual rate, whereas the number of IT-based ATMs has expanded at a 10,1% annualrate, so that ATMs now outnumber physical offices by more than four to one,' Asillustrations of the proliferation of financial technologies, the notional values ofcredit comm itments (standby letters of credit, comm ercial and similar letters of credit,commitments, and participations in acceptances) and interest rate swaps (the onlyderivatives with available data back to the mid-1980s) have also grown at muchfaster rates than bank assets, 9,2% and 27,9%, respectively, annually in real terms,^Table 2 shows selected data on changes in the structure of financial markets. Thedata suggest that public financial markets have grown much faster than the 3,0%annual rate of bank GTA, Money market mutual funds, an altemative to bankdeposits, grew at an average annual rate of 10,8% from 1984 to 2001, Corporateequity and corporate debt (bonds plus commercial paper), which ire alternatives tobank loans, grew at annual rates of 10,0% and 11,3% , respectively, overtakingbank GTA by 20 01 , Finally, mortgage po ols and other asset-backed securitiessome of which are assets that were removed from bank balance sheets and some

TABLE 2T H E STRUCTURE O F FINANCIAL MARKETS, 1984 2001

Year198419851986198719881989199019911992199319941995199619971998199920002001AverageGrowthRate

BankGross TotalAssets (OTA)( trillions)3,443.623.803.783.783.843.773.693.673.783.994.184.354.705.045.205.555,690.030

Money MarketMutual Fund Shares,Total Assets( trilliotis)0,320.320.370.390.400.490.540,570.560.570.600.720.840.98 1,241.441.631.970,108

Corporate Equityat Market Value( trillions)2.443.003.463.383.694.413.915.195,656.396.278,209.7812.4714.4817.8615.8113.380.100

Corporate Bondsand Commercial Paper( trillions)0.951.141.381.581.812.002.122.242.422.632.843.193.584.094.815.476.036.530.113

Mortgage Pools ancOther Asset-BackedSecurities. TotalAssets { trillions)0.420.540.780.98.08.25.42.60.75.882.022,212.442.713.163.563.884.350.137

N O T S Gross Total Assets (GTA) equals total assets plus loan and lease loss reserves and allocated transfer risk reserve (a reserve forcertain foreignloans).Money M arket Mutual Fund Shares Includes money m arket mutual fund shares held hy the household sector, nonfinancialcorporate husiness, nonfarm noncorporate business, and the financial sector. Corporate Equity at Market Value includes equity issued bynonfinancial corporate business and financial corporations, and excludes mutual fund shares. Corporate Bonds and Commercial Paperinclude corporate bonds and commercial paper issued by nonfinancial corporate business and financial sectors (excluding holdings offoreign issues by U.S. residents). Mortgage Pools and Other Asset-Backed Securities include federally related mortgage pools, privatemortgage pools (hom e, multifamily residential, and comm ercial), and pools of agency securities, consumer credit, trade receivables, studentloans,and loans to business. GTA is obtained from the D ecember Reports of Condition and Inco me (Call Re ports). Money Market Mutual


5/37

ALLEN N. BERGER 145

of which are alternatives to bank financinggrew at an annual rate of 13.7 overthe interval.These data are consistent with the hypothesis that advances in IT and financialtechnologies have helped these financial markets to grow at faster rates than thebanking industry. Money market mutual funds were helped by IT innovations thatlet them store, keep track of, and move large amounts of information on securitiesand customer accounts m uch more cheaply overtime.Public equity and debt marketswere similarly favored by IT innovations for handling data, and were also propelledby reductions in trading costs. Much of the trading is now done electronically, and thecosts per trade have fallen dramatically. Asset-backed securities markets were aidedby these IT innovations and by financial innovations that allow for more accuratepricing, more securitization instruments, and better risk management models.Table 3 displays some statistics on the performance of the banking industry overtime as measured by accounting ratios (market measures are not available for mostbanks). Return on equity and return on assets are measures of overall performance;total costs/GTA is a measure of total costs per dollar to create assets, which is alsobroken out by noninterest and interest expenses to GTA; revenues/total costs issometimes used as an efficiency measure; and the nonperforming loan ratio, NPL/loans, is an indicator of problem loans that have not yet been charged off. Thebanking industry had sustained good performance after 1991, although the cost

TABLE 3T H E PERFORMANCE OF

Year198419851986198719881989199019911992199319941995199619971998199920002001

ReturnonEquity0 10330 10940 09780 01480 12930 07600 07380 07800 12770 15230 14540 14610 14340 14790 13910 14990 13920 1299

THE BANKING INDUSTRYReturnon GTA

0 00610 00650 00580 00090 00780 00460 00460 00510 00900 01150 01100 01120 01130 01160 01120 01230 01130 0112

TotalCosts/GTA0 09610 08700 07850 07940 08380 09340 09320 08370 07100 06530 06280 06810 06690 06590 06780 06540 06990 0617

1984-2001NoninterestExpenses/GTA0 02920 02990 03030 03190 03190 03200 03360 03570 03680 03720 03550 03420 03460 03340 03520 03520 03440 0336

InterestExpenses/GTA0 06690 05720 04810 04750 05190 06100 05960 04800 03420 02810 02740 03390 03230 03250 03250 03020 03550 0281

Revenues/Total Costs.1399.1652.1739.1822.1909.1755.1702.1941.2694.3038.3079.2922.3067.3230.2974.3463.3183.3627

NPL/Loans0 04960 04680 04580 05170 04570 04860 05920 05700 04730 03270 02450 02420 02360 02230 02180 02030 02330 0270

N O T S Return on Equity is net income divided by the average of this year's and the previous year's total equity capital. Return on GTAis net income divided by total assets plus loan and lease loss reserves and allocated transfer risk reserve (a reserve for certain foreignloans).Total Costs/GTA is total expenses (interest expenses + noninterest expenses) divided by total assets plus loan and lease loss reservesand allocated transfer risk reserve. Noninterest Expenses/GTA is total noninterest expenses divided by total assets plus loan and leaseloss reserves and allocated transfer risk reserve. Interest Expenses/GTA is total interest expenses divided by total assets plus loanand lease loss reserves and allocated transfer risk reserve. R evenues/Total Costs is operating income divided by totalcosts.NPL/Loans is


6/37

146 ; MONEY , CRED IT, AND BANKING

reductions are primarily due to declining interest expenses, rather than noninter-est expenses. Record profits were earned during several years in the 1990s, althoughperformance may be slightly worse after 1999. The extent to which the good per-formance reflects productivity gains from technological progress versus favorablemarket interest rates and other business conditions is investigated in Section 3 below.Table 4 shows m ost of the performance measures as well as the number of banksand share of industry assets by size class. Two findings are apparent. First, banks inthe smallest size class (GTA under $100 million) tended to have worse performancethan other banks in recent years. Second, there is consolidation of the industryinto the largest size class (GTA greater than $10 billion). The number of banks inthe largest size class increased from 38 to 69, and the proportion of industry assetsin this class increased from 39.4% to 69.2% over the 17 year period. The numberof banks in the smallest size class fell by more than half from almost 11,000 to lessthan 5,000, and the share of assets in this class fell by more than two-thirds from12.7% to 4.0%. The middle two size classes each lost some of their banks andalmost half of their industry shares. Evidence presented below will suggest thattechnological progress played a role in this consolidation.2. EXAMPLES OF TECHNOLOG ICAL CHANGESIN THE BANKING INDUSTRY

Rather than reviewing microeconomic research on all banking technologies, wefocus primarily on three exam ples in which the technological changes can be observedand some of their eflFects can be directly measured - Internet banking, electronicpayments technologies, and information exchanges.^ These may not be the mostimportant banking technologies, but they illustrate the multiplicity of potentialdifferent actual and measured effects of technological progress. Our examples alsorepresent both IT and financial technologies and cover both front-oflSce techno log-ies in which the banks deal directly with custom ers and back-office technologiesfor producing services that are generally invisible to customers.2 1 Internet Banking

Internet banking is a relatively new front-office technology. B anks offer a varietyof levels of Internet service and combinations of Internet and physical offices andATM networks. Some banks employ a click-and-mortar implementation strategyin which the banks add a transactional Internet site to their physical offices andATM networks. A transactional site allows customers to make transactions on-linesuch as accessing accounts, transferring funds, applying for a loan, etc. Other bankshave set up informational websites that provide information about the banks andtheir services, but do not allow for on-line transactions. A small number of Internet-only banks offer services through transactional Internet sites and access to ATMnetworks, but with no physical offices open to the public. As of March 2002, there


7/37


TABLE4BANK PERFORMANCE,1984-2001Year Returnon EquityA. Commercial Banks198419851986198719881989199019911992199319941995199619971998199920002001

0.08180.06650.04500.05370.06650.08030.07380.08090.10780.11480.10880.10740.10440.10450.09610.09070.08860.0795

B Commercial Banks198419851986198719881989199019911992199319941995199619971998199920002001

0.11890.10990.09280.08720.09670.11580.09720.09630.12170.13190.13160.13120.12990.13190.13000.13810.13050.1200C Commercial Banks198419851986198719881989199019911992199319941995199619971998

0.13390.13340.11170.07300.11460.09790.05750.07560.13500.16150.15490.14650.14370.14180.1520

NUMBER OFBANKS, ANDSHAREOF

Returnon GTAwith Gross

0.00710.00580.00390.00470.00590.00720.00660.00730.01000.01100.01080.01100.01110.01120.01050.00980.00970.0088

with Gross0.00840.00780.00650.00640.00710.00850.00740.00740.00970.01120.01150.01180.01210.01230.01220.01270.01190.0111with Gross0.00750.00760.00660.00440.00690.00610.00370.00490.00980.01240.01240.01210.01260.01320.0142

TotalCosts/GTATotal Assets

0.09530.09020.08300.07830.07920.08480.08420.08000.06740.06060.06060.06530.06410.06620.06710.06710.06770.0642

Total Assets0.09360.08690.07890.07600.07860.08480.08400.07910.06590.05990.05990.06440.06480.06470.06440.06350.06690.0626Total Assets0.09250.08370.07590.07650.08030.08790.08760.08230.06900.06240.06300.06790.06700.06850.0700

Revenues/Total CostsUnder 100

1.15791.1698.1546.16661.17041.17451.16691.17521.2459.2758.28351.26771.27971.27191.25051.24221.2417.2184

INDUSTRY ASSETS BYBANK

NPL/LoansMillion0.05460.06090.05840.05000.04450.04270.0432

0.04130.03310.02800.02510.02640.02780.02640.02660.02230.02460.0277Between 100Millionand1.1601.17471.17861.19191.19371.2002.19631.20441.27751.3084

Betw

1.32441.3064.31751.32191.31661.33061.30101.2979een 11.14701.16741.18271.20221.20601.19061.20191.22431.32061.37431.36491.34931.37821.41431.3815

0.04570.04820.04700.04230.03790.03910.04390.04340.03440.02810.02310.02410.02390.02180.02120.01830.02000.0223

Numberof Banks

10,92910,72710,40410,0269,5559,1418,8238,4668,0467,6437,2026,6806,2885,9835,5805,3855,0844,784

1 Billion3,0933,1473,2303,1002,9973,0022,9702,9382,9102,8372,7702,7672,7732,7352,7622,7662,8082,876Billionand 10Billion0.04200.03940.03800.04120.03600.04050.05530.05450.04280.03250.02320.02450.02670.02550.0247

310341351357355346333325331327319331312282285

SIZE CLASS,

Proportionof IndustryGTA

0.1270.1190.1120.1080.1030.0970.0970.0960.0940.0880.0790.0710.0640.0570.0510.0470.0420.0400.2130.2060.2030.1950.1900.1900.1920.1950.1930.1810.1680.1600.1530.1440.1350.1320.1260.1270.2660.2740.2730.2840.2930.2790.2740.2870.2870.2860.2640.2440.2210.1840.167


8/37

148 ; MON EY, CRED IT, AND BANKING

T A B L E 1BANK PERFORMANCE1984-2001Year199920002001

Returnon Equity0.15050.12960.1415

D. Commercial Banks198419851986198719881989199019911992199319941995199619971998199920002001

0.07830.11260.1153- 0 . 1 0 5 90.19480.02910.07130.06710.13230.16700.15450.16010.15530.16190.14230.15850.14790.1333

NUMBER OF

Returnon GTA0.01390.01210.0129

BANKS , AND

TolalCosls/GTA0.06600.06970.0643

SHARE O F

Revenues/Total Costs1.39201.34111.3899

with Gross Total Assets Greater than0.00360.00530.0056-0 .00510.00930.00140.00360.00370.00790.01100.01010.01060.01060.01100.01020.01200.01110.0110

0.10010.08850.07880.08330.08980.10270.10270.08780.07530.07030.06410.06980.06770.06530.06790.06550.07060.0609

1.11971.15771.17151.16911.18471.15851.14451.17441.23921.26671.27611.26561.27901.29941.27481.34591.32141.3778

INDUSTRY ASSETS BY BANK

NPULoans0.01970.02110.0233

10 Billion0.05480.04750.04760.06380.05670.05910.07090.06780.05900.03570.02590.02380.02170.02100.02070.02080.02430.0287

Numberof Banks28 326 827 4

384350505361635754556473696066697669

S IZE CLAS S ,

Proportionof Industry GTA0.1590.1370.1410.3940.4010.4120.4130.4140.4340.4380.4210.4260.4450.4890.5260.5610.6150.6480.6620.6950.692

N O T S Return on Equity is net income divided by the average of this year's and the previous year's total equity capital. Return on GTAis net income divided by total assets plus loan and lease loss reserves and allocated transfer risk reserve (a reserve for certain foreignloans).Total Costs/GTA is total expenses (interest expenses+no ninterest expenses) divided by total assets plus loan and lease loss reservesand allocated transfer riskreserve.Revenues/Total Costs is operating income divided by totalcosts.NPL/Loans is nonperfonming loans (eitherpast due at least 30 days or on nonaccrual basis) divided by total loans. Proportion of Bank GTA is the sum of GTA for the size classdividedby total G TA. Ratios are weighted average s, using the denom inators as relative weights. All data are obtaine d from the DecemberReports of Condition and Income (Call Reports). Size classes are defined using real 1994 dollars.

such institutions have failed, been acquired, or voluntarily liquidated.'* A few largebanks set up Internet-only units, and then integrated them into the main bank afterpoor performance. Finally, many banks continue to oifer no Internet services.Internet banking has become widespread in a short time, although there aresubstantial differences by bank size in implem entation strategies. A study of n ationalbanks in the U.S. found that as of the end of 2000, 37.3% of these banks oflferedtransactional Internet sites, and an additional 27.7% offered informational websites(Furst, Lang, and Nolle 2001,200 2).^ The v ast majority of the transactional sites wereset up since the beginning of 1998. The transactional website adoption rate variedgreatly by bank size, with 100% of the banks with over 10 billion in assets havingthese sites and only 20.0% of banks with less than 100 million in assets. Since thelarge banks also have extensive physical branching and ATM networks, these banksare using the click-and-mortar implementation strategy. By the end of 2001, anumber of banks had added transactional sites, but the rest of the relevant factsremained qualitatively unchanged49.7% of national banks had transactional web-sites (100% of the largest banks, 29 .1% of the smallest ban ks), and 21.7% hadinformational sites.^ Although only about half of the national banks offer transac-


9/37


customers because they tend to include the largest banks. A survey of banks in theTenth Federal Reserve D istrict (Kansas City D istrict) yielded co nsistent findings.^Importantly, although there may be scale economies in setting up and maintainingtransactional websites, this technology may still be accessed by small banks. Manysmall banks are able to outsource the provision of their transactional websites tocompanies that specialize in these operations.

Some studies have examined the relative performance of banks offering Internetservices . For national banks with assets over 100 million, those offering transac-tional Internet sites were more profitable than those that did not. This primarilyreflects the choice of profitable banks to adopt the technology, rather than profitabil-ity from Internet services, which currently make up a small portion of output ofmost of these banks (Furst, Lang, and Nolle 2001, 2002). This research also foundthat for small banks with assets below 100 million, there was no statistical differencein profitability between Internet and non-Internet banks for mature institutions inoperation more than three years. However, for small de novo institutions with lessthan three years experience, those with transactional sites performed relativelypoorly. A study of the relatively few Internet-only banks found that these banksperformed more poorly than traditional de novo banks, consistent with the findingabove that some Internet-only banks and units were discontinued due to poor perfor-mance. However, the performance of the Internet-only banks may be improvingfaster than other banks as they ride the learning curve and/or become able to exploitwhat may be substantial scale economies (DeYoung 2002).The fast spread of Internet banking may result in the benefits of this technologygoing primarily to consumers as banks incur the costs of providing these sites tomaintain market shares. That is, competition may currently or in the near futureforce banks to adopt the technology just to keep existing customers and notcharge enough to earn abnormal profits from providing this service. Consistent withthis possibility, banks offering Internet sites and those planning to adopt themgenerally referred to a need to remain competitive and retain customers, rather thanany increase in revenues to cover their costs (Furst, Lang, and Nolle, 2001, 2002,Sullivan, 200 1). This is similar to the experience ofU.S.banks adding ATM networksin the early 1980s without charging the full costs of implementing that technologydue to increased competitive pressures.^2 2 E lectronic P ayments Technologies

Electronic payments technologies are methods of transferring funds electroni-cally with relatively little pape rw ork .' At the front-office level, there has been aswitch from paper payments to electronic payments by theU.S.population. As shownnext, consumers have switched some of their purchases from checks and cash tocredit cards, which are mostly cleared electronically (except for the monthly paperbill and check payment), and to debit cards, which are almost entirely processedelectronically.One study found that the estimated number of checks paid in the U.S. fell by a


10/37

150 : MON EY, CREDIT, AND BANKING

an average annual rate of decline of 3.0%. During the same interval, estimatedcredit card payments grew from 10.4 billion to 15.0 billion, an average annualincrease of 7.3%, and debit card transactions grew from an estimated 1.4 billion to8.3 billion, an average annual growth rate of 35.6% (Gerdes and Walton 2002,Table 2). Thus, by these estimates, the share of checks to total checks plus creditand debit cards used fell from 80.8% to 64.6% in just five years.

Another study found that check use continued to grow, but at a much reducedpace, and that the use of cash in retail payments has declined dramatically. As aresult, both types of paper payments have lost market shares to electronic paym ents.This study estimated that from 1990 to 2000, the share of cash used in personalconsumption spending fell from 25.7% to 16.3%, while the check share droppedfrom 61.8% to 56.0%. During the same interval, the estimated credit card sharerose from 12.2% to 21.2%, and the estimated debit card share grew from 0.4% to6.5% (Humphrey 2002). Thus, according to these estimates, the share of cash andchecks used in personal consumption fell from 87.5% to72.3%in a ten-year interval.

Another indicator of the switch from paper to electronics is the steep increase inthe use of automated clearinghouse (A CH), which is primarily a substitute for paperchecks for regular payments, such as direct deposit of pay, withdrawal of monthlymortgage paym ents, etc. From 1990 to 2000, ACH volume processed by the FederalReserve (which handles the majority of ACH paym ents) more than quadrupled fromabout 915 million in 1990 to 3.8 billion in 2000, a 14.2% annual rate of increase(Annual Reports of the Board of Governors of the Federal Reserve System).'It is likely that the switch from paper to electronic payments was fueled inlarge part by IT advances that reduced the costs and increased the availability andconvenience of electronic payments. Consumer payments by credit card and debitcard have become available at many more physical retail outlets (e.g., grocery stores)and for more Internet and phone purchases.

Cost studies of electronic payments processing are consistent with the notionthat IT improvements in back-office processing of electronic payments resulted inproductivity gains and scale economies that reduced costs dramatically over time.The studies often focus on the Federal Reserve because it is the largest processorand its data are available. However, the results likely generalize to the privatesector because all processors have access to the same technologies and becausethe Federal Reserve is required to simulate the behavior of aprivate-sector competitorby keeping costs under control and set prices to recover costs.

The two main types of electronic payments processed by the Federal Reservehave had steep declines in unit costs over time. The raw data on ACH showed adecline in nominal unit costs from $0,869 per item processed to $0,176 over the1990-2000 interval. Put into real 1994 dollars using the GDP deflator yields adecline from $0,959 to $0,158, or fall of about 83% in real unit costs. Theraw data on Fedwire (used primarily for large-value wholesale payments) showeda decline in nominal unit costs from $1,029 to $0,518 or from $1,135 to $0,466 inreal unit costs over 1990 -2000, or a decline of about59%in realterms.^Given these


11/37

A L L E N N B E R G E R : 1 51

that substantial unit cost savings have also occurred for private sector electronicpayments processing.Cost function studies of ACH op erations which control fortheinput prices, outputquantities, and other conditionsfound very substantial improvements in cost pro-ductivity over time, although the improvements were smaller in last few years ofthe 1990s (Bauer and Hancock, 1995, Bauer and Ferrier, 1996, Bauer and Higgins,2002). Some of these improvements are clearly due to IT advances. For example,prior to the 1990s, ACH was run primarily as batch jobs with the physical deliveryof computert pesfrom the banks, whereas the data are now sent electronically. Part ofthe gains were also found to be due to the exploitation of scale economies, and thestudies suggest that even more scale economies may remain unexploited. The FederalReserve shifted in 1996 to FedACH, a centralized ACH application software toprocess payments, and consolidated the customer support sites from the 12 Districtsto 2 sites in 2001.

Studies of Fedwire processing also found substantial technological progress andscale economies linked to IT that allowed the Federal Reserve to consolidate from36 processing sites to 3 sites (one main, two back-up; see Bauer and Ferrier, 1996,Hancock, Humphrey, and Wilcox, 1999). A difficult issue in switching to newtechnologies is adjustment costs (having extra machines and personnel during thetransition, training costs, severance pay, etc.) One of the Fedwire studies was able tocapture some of these adjustment costs and found them to be substantial, suggestingthat the productivity gains may be even higher in the long run after the adjustmentcosts have subsided (Hancock, Humphrey, and Wilcox 1999). Adjustment costs willreturn as an issue in productivity studies and in bank M A stud ies.'^

Importantly, although there may be substantial scale economies to processingpayme nts, it is not necessary to have large banks to exploit these econom ies. Smallbanks can share some of these benefits by outsourcing to a large processor, suchas the Federal Reserve, correspondent bank, or other large private-sector processor.Nonetheless, there may be differences by bank size in the use of these technologies,even when most of the processing is performed by a third party.'''2 3 Information Exchanges

Information exchanges, as we use the term here, are intermediaries through w hichbanks and other creditors share data relevant to the creditworthiness of loan appli-cants.These exchanges collect data from financial institutions, trade creditors, publicrecord s, and other sources, aggregate and sum marize the data, and then provide creditreports or credit scores to lending institutions. The exchanges may be private third-party credit bureaus, associations organized by banks, or public credit registersorganized by central banks.'^The technology of information exchanges has been shown to add value. A cross-

country analysis found that bank lending is higher and default rates are lower innations in which lenders use either private or public information exchanges (Jappelli


12/37

152 : MONEY , CREDIT, AND BANKING

explanatory power in predicting firm failure, even after controlling for other informa-tion available to lenders (Kallberg and Udell 2003). Research on consumer creditscoring in the U.S. also generally found that the scores are the best predictors ofrepayment, with only minor improvements from additional data such as householdincome and location (e.g., Avery et al. 2000). The use of information exchangeshasalsobeen showntoenhance the availability of creditinless developed nations (e.g.,Klapper and Kraus 2002).

The information exchange technology is also subject to substantial scale econo-mies. For example, additional observations of payments and other information tendto increase the quality of the credit signals (Pagano and Jappelli, 1993, Kallbergand Udell, 2003). Technological progress has likely increased the productivity andscale economies associated with this technology, as the data are now more oftentransmitted to and disseminated from information exchanges electronically, andimproved financial models are used to analyze the data to predict future creditproblems.

One technology that usually involves information exchanges is small businesscredit scoring (SBC S). Credit scores have long been used in underwriting consume rcredits, but this technology has only recently been adapted to small commercial cred-its. The score is a summary statistic about the firm's expected future loan perfor-mance. A key element of this technology is the use of personal data about theowner of the business. Personal and business information are often obtained fromcredit bureaus and combined with data collected by the bank and entered into aloan-repayment prediction model (Mester 1997). It is a back-oflfice technology, ascustomers typically do not know how their loan application is evaluated.Fair, Isaac and Company, the largest provider of credit scoring models, introducedits first SBCS model in 1995, and many banks adopted SBCS in the succeedingyears. The models are typically designed for credits up to $250,000, although manybanks use the scores to evaluate credits of up to $100,000 only.'*Several research studies employed a 1998 survey of large U.S. banking organiza-tions on whether and how they used SBCS.These studies found that the use of scoring

on small business credits under 100,000 increased lending (Frame, Srinivasan, andWoosley 2001), that the lending was focused more on low- and moderate-incomecensus tracts than on higher-income areas (Frame, Padhi, and Woosley 2001), andthat the loans issued on average had higher interest rates and worse credit ratings thanloans issued by nonscoring institutions (Berger, Frame, and Miller 2002). Thesefindings suggest a net increase in lending to marginal applicants that would nototherwise receive credit and tend to have relatively high loan prices and high riskwhen they are funded. The research also found that SBCS was generally adoptedearlier by larger banks (Akhavein, Frame, and White, 2001, Frame, Srinivasan,and Woosley, 2001). However, banks of any size can access this technology bypurchasing scores from third-party information exchanges.The research also suggests that the way in which the technology is implemented


13/37


and banks that exercise discretion develop their own scoring models and useother inputs in credit decisions (Berger, Frame, and Miller 2002). The expansion ofcredit was concentrated in the rules banks, with discretion banks primarilyimproving accuracy in credit evaluation.2 4 Some Potential Inferences from the Microeconomic

Research on Banking TechnologiesThis research points to some potential general inferences about new technologies.First, there is a multiplicity of difFerent actual and measured eifects of newtechnologies on productivity growth and industry structure. Some new technolog-iessuch as ATMs in the early 1980s and possibly Internet banking currently or in

the near futuremay increase productivity significantly in terms ofthe quality oftheservice to the consumer, but these benefits may not be easily measured. Firms mayprovide the higher quality without charging the full costs due to competitive pres-sures. In contrast, some new technologiessuch as the innovations in processingelectronic paymentsmay have very significant and easily observable efFects interms of productivity gains and increased scale econom ies. Some new technologiessuch as innovations in information exchangesmay alternatively have significantbenefits that can only be measured with nontraditional methods, such as examiningthe composition of the loan portfolio.Second, the research on banking technologies suggests that most of the important

new technologies were generally adopted earlier by large firms than small firms.Large banks have generally been first to (1) adopt transactional Intemet websites, (2)use electronic paym ents technolog ies, and (3) employ the SBC S technology, althoughthere are exceptions, such as some of the new Internet-only banks. The finding thatlarge banks tend to adopt innovations earlier also holds for other banking technologiesnot reviewed in detail here, including the adoption of ATMs (e.g., Hannan andMcDow ell 1984), securitization and off-balance sheet financia l activities (e.g., Bergerand Udell 1993), and the new portfolio risk models for dealing with proposed Basleinternational capital standards.'^Third, the research suggests that even when there are significant scale economiesassociated with back-office operations, small firms are often able to share in thebenefits of technological progress. Small banks can (1) have transactional Intemetsites by outsourcing to companies that specialize in these operations, (2) use elec-tronic payments technologies by outsourcing to the Federal R eserve, correspondentbank, or other large private-sector processors, and (3) use information exchanges bypurchasing data from a large exchange. Similarly, small banks can gain access toother large-scale technolog ies. For example, small banks in theU.S.take advantage ofscale-efficient processing of paper payments provided by large institutions and areable to tap into nationwide and w orldwide ATM networks w ithout each bank settingup its own expansive network. Even some of the benefits of new complex riskmanagement technologies may filter to banks too small to create their own systemsthrough outsourcing, e.g., by purchasing portfolio risk models such as CreditMetrics


14/37

154 ; MON EY, CREDIT, AND BANKING

necessarily mean that small banks can use these technologies at the same unit cost aslarge banksthe parties providing the back-oifice services may charge a relativelyhigh fixed cost or oifer significant volume discounts that put small banks at adisadvantage, which may help explain why small banks may be slower to adoptnew technologies.

Fourth, the research suggests that the effects of a new technology may varysignificantly with the way in which it is imp lemen ted. The research on front-officetechnologies suggests that combining new technologies with existing technologies tooffer m ore consum er choice may often b e the most effective implem entation strategy.In the 1980s, banks generally combined the new ATM networks with traditionalphysical offices and today, the large banks that serve most customers are addingtransactional Internet sites to their physical offices and ATM networks. For back-office technologies, the implementation strategy also appears to matter. The FederalReserve appeared to achieve dramatically lower unit costs in processing electronicpayments in part by consolidating operations to take advantage of scale economies.The effects of small business credit scoring on lending also appear to vary consider-ably with whether the bank follows rules versus exercises more discretion andother factors.

3 . TECHNOLOGICAL PROGRESS AND PRODUCTIVITY GROWTHWe next examine the evidence linking technological progress to productivitygrowth. We include evidence on the aggregate economy as well as on the bankingindustry to provide background information and contrast the research methodsand results.

3 1 Technological Progress and Government Indexes of Aggregateand Banking Productivity GrowthMuch has been made of the recent increased productivity growth for the U.S.economy as measured in government productivity index es. Table 5, column 1, showsthe Bureau of Labor Statistics (BL S) labor productivity measure for the U.S . privatenonfarm bus iness sector for 1967 through 200 0 (1996 value = 100). Colum n 2shows the one-year (continuous) growth rates or natural log differences betweenyears t and t - 1 . An aggregate productivity slowdown is often identified as havingoccurred from about 1973-1995, and the recent upswing in productivity growthis often identified as having occurred approximately over 1995-2000. As shown,the average annual growth rate was 2.57% over 1967-1973, which fell to 1.37%over 1973-1995, and then rose to 2.54% over 1995-2000, although there is signifi-

cant variation within all of these intervals.It is often argued that IT advances played a substantial role in the recent speedup,


15/37


TABLE 5PRODUCTIVITY1967-2000

Year19671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920001967-19731973-19951995-20001967-19731973-19821982-2000

IN THE P R I V A T E 1

1)IndexofOutputPerHourin thePrivateNonfanrBusinessSector

61.863.763.864.967.669.972.171.073.075.876.977.877.577.378.378.081.483.284.487.187.588.689.290.391.494.895.396.597.5100.0102.0104.7107.1110.7

NONFARM B U S I N E S S S E C T O R

2)

ProductivityGrowthinthe Private1 NonfarmBusinessSector

0.03030.00160.01710.04080.03350.0310-0.01540.02780.03760.01440.0116-0.0039-0.00260.0129-0.00380.04270.02190.01430.03150.00460.01250.00670.01230.01210.03650.00530.01250.01030.02530.01980.02610.02270.0331

verage Productivity Growth Rates over0.02570.01370.02540.02570.00870.0195

AND COMMERCIAL BANKING INDUSTRY

3)

IndexofOutputPer Hour Iinthe CommercialBankingItidustry70.972.371.172.575.176.581.276.176.380.584.785.884.178.577.880.187.189.694.396.2100.0102.8104.8107.7110.1111.0118.5121.7126.4129.7133.0132.6135.9143.2

Selected Intervals0.02260.02010.02500.0226-0.00150.0323

4)

ProductivityGrowthinthe CommercialBankitigIndustry

0.0196-0.01670.01950.03520.01850.0596-0.06490.00260.05360.05090.0129-0.0200-0.0689-0.00900.02910.08380.02830.05110.01990.03870.02760.01930.02730.02200.00810.06540.02660.03790.02580.0251-0.00300.02460.0523

N O T S Column 1) is an index of the output per hour of all persons for the private nonfarm business sector 1996 = 1), and column 2)shows the natural log differences by year for this sector. Column 3) is the output per hour of all persons for the commercial bankingindustry SIC-602, 1987 = I), and column 4) shows the natural log differences by year for this industry. Data are from the Bureau ofLabor Statistics, www.bls.gov.


16/37

156 : MON EY. CREDIT, AND BANKING

across industries and found that the highest productivity gains have generally oc-curred in industries that tend to use IT intensively and those that manufacture ITequipment (e.g., McKinsey Global Institute, 2001, Stiroh 2001).

Some analysis has used modified forms of the Solow (1957) neoclassical growthmodel (e.g., Jorgenson and Stiroh, 2000, Oliner and Sichel, 2000). Essentially,aggregate output Y) is modeled as a simple function of IT capital services iK\j),other capital services KQJH), labor (L), and a multifactor produ ctivity term (MFP).Technological change is embodied in the MFP variable. A number of neoclassicalassumptions are imposed, including perfect competition, constant returns to scale,no adjustment costs, equal retums to all types of capital. Hicks-neutral technologicalchange, etc. The growth in labor productivity is given by:

A{Y/L) = a , A ( / : , T / L ) -I- ttzA{KOTH/L) +AMFP , (1)where denotes a growth rate, and the a a are income shares. Technological progressis measured by the Solow residual or AMFP.

These studies generally found that IT contributed significantly to the recentupswing in aggregate productivity in two ways. First, the very large investments inIT equipment over time resulted in capital deepening or increases in A(A'IT/L),growth in IT capital per unit of labor. Second, IT contributed to AMFP primarilyas a result of productivity gains in the production of this equipment.Table 5, columns 3 and 4, show U.S. BLS labor productivity index for thecommercial banking industry (SIC code 602, 1987 value = 100) and its one-yeargrowth rates, respectively. For the 1967-1973,1973-1995, and 1995-2000 intervals,the average annual growth rates were 2.26%, 2.01%, and 2.50%, respectively.Com paring these figures to those of the entire U .S. nonfarm business sector shownabove(2.57%,1.37 ,and 2.54%, respectively), suggests similar productivity growthrates for first and third intervals, but much less of a drop in productivity growth over1973-1995.A more important difference between the series (noted by Furlong 2001) lies inthe timing of the start of the increase in measured productivity. Breaking up the

banking series into the intervals 1967-1973, 1973-1982, and 1982-2000, givesaverage annual growth rates of 2.26%, 0.15%,and 3.23 % , respectively. Thu s, theBLS statistics suggest that the banking industry suffered a more significant dropin productivity than the nonfarm business sector starting after 1973, and had aneven stronger recovery that occurred much earlierin the early 1980s, ratherthan in the mid-1990s. As well, the strongest part of the growth was earlier than the199 5-20 00 interva l. It was sugges ted that this earlier measured upturn for the bankingindustry may have occurred because banks generally invested heavily in IT beforeother industries (Furlong 2001). ^

3.2 Difficulties in Linking Technological Progress to Productivity GrowthThere are a numb er of difficulties in linking technologica l progre ss to produ ctiv-


17/37


firms do not necessarily always employ the best available technology, observedchanges in productivity or firm performance also reflect factors other than techn ologi-cal progress. Econometric analysis can at most estimate the best-practice frontier,which reflects the behavior of the best existing firms, but not the true efficientfrontier that embodies the best available technology.

Second, it is difficult to account for the efFects of technological progress inimproving the quality and variety of goods and services. For example, prior researchsuggested that the introduction of high-technology consumer products such as cellphones, voicemail, and the Internet involved large increases in consumer welfarethat went unmeasured in the U.S. Consumer Price Index (CPI) for many years (e.g.,Hausman 1998). Research also suggested that improvements in quality associatedwith technological progress may have resulted in price indices that significantlyoverstate inflation (e.g., Pakes 2002). The Boskin Commission estimated that theCPI bias from new products and quality change was on the order of a 1.1% annualupward bias, with a plausible range of 0.8% to 1.6% (Boskin et al. 1996).

Third, the benefits from technological progress do not necessarily accrue to thefirm or industry where they occur, making them difficult to measure. To the extentthat markets are competitive, the associated rents may be competed away andpassed through to customers or factors of production, provided there are no barriersto the adoption of the technology or in the product or factor markets. For example, ifbanking were perfectly competitive, and all technological improvements weresuccessfully copied by other banks or nonbank competitors, any abnormal returnswould be competed away through more favorable prices to customers (e.g., lowerrates on loans, higher rates on deposits), improved quality and variety of servicesprovided (e.g., ATMs, Internet banking), rents to the providers of the improvedhardware/software (e.g., Intel, Microsoft), key employees that know how to use thetechnology, etc.

Fourth, the simplifying assumptions of the productivity models may be signifi-cantly violated by the data. For example, the AMFP residual in Equation (1) abovemay em body factors o ther than the effects of technological prog ress if there is marketpower, increasing returnst capital, adjustment co sts, unequal returnst different typesof capital, nonneutral technological change, etc. One study found that adjustmentcosts to new technology during the late 1990s were substantial, suggesting that thecontribution of technological progress to productivity growth may be higher thanis usually measured and may be revealed more at a later date (Basu, Fernald, andShapiro 2001).^'

Som e difficulties relate specifically to the govem me nt m easures of bank produ ctiv-ity. The BLS statistics for commercial banking shown in Table 5 above, SIC code602, measure physical banking output using a number-of-transactions approachbased on demand deposits (number of checks written and cleared, and number ofelectronic funds transfers), time deposits (weighted index of number of depositsand withdrawals on regular savings accounts, club accounts, CDs, money marketaccounts, and IRAs), ATM transactions, loans (indexes of new and existing real


18/37


transactions), and number of trust accounts, weighted by the proportions of e mployeehours used in each activity. Employee labor hours are used as the denominator ofthe index.^^

This measure does not include the financial market activities of banks (eitheron or off the balance sheet) and does not incorporate improvements in financialtechnologies except to the extent that they afFect the amount of labor in doingtraditional banking activities. By focusing on labor-weighted outputs, this mea-sure tends to understate the outputs that are capital-intensive, where presumablymuch of any effects of IT advances would be concentrated. The BLS index excludesinputs that account for most of the costs, especially interest expenses on the inputs offunds. During the 1984-2001 interval, total wages, salaries, and fringe benefitsexpenses accounted for only between 15 and23 of bank costs, with the fluctuationover time largely driven by changes in market interest rates, which affect interestexpenses.

As well, the focus on transactions, as opposed to intermediation, may not ac-curately describe the business of banking. Banks are financial intermediaries thattransform funds from depositors and other investors into loans and other financialinvestments. Presumably, intermediating m ore dollars contributes m ore to the outputof a financial intermediary, even if it is accomplished with the same number oftransactions. The exclusion of interest expenses also makes it difficult to evaluateban ks' productivity gains in intermediation. For example, banks may have substitutedhigher rates on deposits for real resources used to provide services on these depositswhen deposit rates were deregulated in the early 1980s. Such a substitution wouldnot by itself necessarily constitute a productivity decrease.In addition, the BLS measure does not include all of the labor input. Much ofthe labor that goes into back-office processing of payments, computer operations,etc. have been moved over time into service bureaus, which are often affiliates ofthe same bank holding company. As a result, banks might appear to become moreproductive over time by shifting some of their workers into service bureaus, payingfor these labor hours through transfer paym ents to service bureaus rather than throug h

labor hours paid directly by the banks, but this shift does not necessarily constitutea real increase in labor productivity.'^^ As w ell, to the extent that in novations in dataprocessing, payments processing, and other back-office technologies occur in theservice bureaus, they may not be incorporated as measured im provements in commer-cial bank productivity.'^'*3.3 Econom etric Studies Linking Technological Progress to Productivity

Growth in BankingThe econometric studies of bank productivity address some of these difficulties.The most common methodology uses multiproduct cost functions with detailed

variables collected at the individual bank leve l. A typical (log) banking cost functionmay be written as:


19/37


where In indicates n atural log; C is variable co sts; / is the (log) cost functionembodying the best-practice frontier; w, q, z, and v are exogenous business condi-tions that afFect costsvariable input prices, variable output quantities, fixed inputand output quantities, and env ironmen tal variables, respec tively; lnM is an ineffi-ciency factor that is zero for best-practice firms and positive for other firms, andIn e is mean-ze ro random error. As show n in Berger and M ester (20 03), thepredicted gross change in cost at the mean values of all the exogenous factorsbetween period tand t \ may be represented as:

C, \/C, = {exp[/,+| .j:,+ ,)] exp[ln, i]}/{exp[/; x,)] exp[ln ,]} , (3)where x denotes the vector of business conditions (w, q., z, v), and the subscript tdenotes the cost function or the mean value for a set of variables at time t. Therandom error In e drops out because it is assumed to be mean zero each period.The predicted gross change in costs is decomposed into three multiplicative termsrepresenting (1) the movement of the best-practice frontier, (2) the change in industryinefficiency or average dispersion from the frontier, and (3) the effect of changesin business conditions:

C, \/C,= {exp[/,+,(x,)]/exp|/,(-^,)]} (Movement of best-practice frontier){exp[lnu, 1 ]/exp[lnu,] (Change in inefficiency){exp[/,+i x,+i)]/exp[/,+|(x,)] (Changes in business con ditio ns) .

(4)The movement of the best-practice frontier incorporates technological change andtheextenttowhichthebest-practicefirmsadoptit.C hanges in inefficiency measu re thechanges in dispersion from best practices, and the change in business conditions isthe effect of the changes between periods in input prices, output quantities, fixednetputs, and environmental conditions.Thechange in cost productivity is measured bythe product of the movement in the best-practice frontier and the change ininefficiency, or the first two terms above:

PRO D C,,,+, = {exp [/,+ |(x,)]/exp[/,(x ,)]l {exp[lnM, ,]/exp[lnu,]] . (5)Thus, cost productivity is the change in predicted costs due to movement of thebest-practice frontier and changes in inefficiency, abstracting from random errorand changes in exogenous business conditions. This is a gross measure of theincrease in costs due to productivity change, so a finding of PROD C,,,+ | = 0.95would indicate a 5% improvement in cost productivity.

The cost function approach addresses a number of problems in the governmentmeasures . It accounts for financial market activities of banks, including off-balancesheet activities in the outputs. It also incorporates both financial and nonfinancialinputs, inclusive of interest expenses for financial inputs and payments to servicebureaus. The cost function approach also statistically removes the effects ofexogenous business conditions, so that changes in costs due to changes in market


20/37


the individual inputs and outputs to contribute to the measurement of productivityin a nonarbitrary way, letting the data dec ide on their relative contribution s tovalue. PRODC, , | is also a multifactor productivity measure, rather than a laborproductivity measure. Banks are financial intermediaries and their major cost itemis the cost of funds, rather than labor compensation.

A number of studies measured productivity growth for U.S. banks during the1980s and in some cases through 1993, splitting it into movements of the best-practice frontier and changes in inefficiency using either cost productivity (e.g.,Berger and Humphrey, 1992a, Bauer, Berger, and Humphrey, 1993, Berger andMester, forthcoming) or linear programming methods (e.g., Wheelock and Wilson,1999, Devaney and Weber, 2000, Alam, 2001).^^ Although these studies are notfully consistent, a general finding was that bank productivity growth was poor ornegative during the early 1980s and may have improved during the late 1980s andearly 1990s.^*The unfavorable changes in the early 1980s primarily occurred throughshifts in the best-practice frontier, although in some cases, changes in inefficiencywere larger (e.g., Wheelock and Wilson, 1999, Berger and Mester, 2003). Thesestudies also often found that bank business conditions deteriorated substantiallyduring the early 1980s.

The primary driving force behind poor results during the early 1980s appears tobe adjustment to industry deregulation, rather than technological change. There wassubstantial deregulation in the U.S. during this period, including deregulation ofdepositrates,liberalization of charter policy, and reductions in geographic restrictionson banking within states and across state lines.'^^ The deregulation brought aboutsubstantial increases in competition, which benefited cus tomers but hurt the m easuredproductivity of the industry. The deregulation of deposit rates and other increasesin competition caused bank interest expenses to rise. In part, this was substitutionof interest expenses for labor expenses by freeing banks to choose the optimal mixbetween interest payments and service, and in part this was higher rates to competefor customers. Banks also expanded their ATM networks with little or no charge tocustomers. To the extent that the industry performed more poorly because of anincrease in competitiveness that raised deposit rates and increased bank spendingon ATMs, this may be a social good, because the benefits to depositors may outwe ighthe higher costs to banks. Thus, the adoption of a highly valued technologicaladvance, the ATM network, may appear in the data as a reduction in measuredproductivity, but in fact it likely raised actual productivity.The finding of poor productivity growth in the early 1980s by the research studiesstands in stark contrast to the increases in the BLS labor productivity measure forthis period. As noted, the research measures likely understate the productivity gainsduring this unusual period because they do not account for the transfers toconsumers from increased competition. The BLS measure may understate or over-

state the productivity growth. Like the research studies, the BLS measure does notaccount for the transfers to consumers. The BLS measure also excludes gains from


21/37


by counting any substitution of interest expenses for labor expenses that likelyoccurred as gains. As well, the BLS measure counts as a productivity increase whenlabor is moved into service bureaus, which occurred during the early 1980s, Thedata suggest that the overstatements may dominate the understatements.

Two recent studies applied the cost productivity methodology to more recent dataonU,S,banks from 1991 to 1997, One found sm all cost productivity im provem ents ofless than 1% annually (Stiroh 2000), and the other found cost productivity declinesof 12,5% annually for this period, primarily reflecting a significant u nfavorable shiftin the best-practice frontier (Berger and Mester 2003), The studies both foundthat total costs rose over the 1991-1997 period, but decomposed these differentlybetween productivity growth and changes in business conditions, or thexvariables(w,q z v) shown in the cost function above . The studies em ployed similar specifica-tions of outputs and inputs, but only the latter study included env ironment variables (v),which likely explains most of the difference in results. Some of these variables,such as local market nonperforming loans and state income growth, were generallyimproving over the period and likely improved bank performance in a number ofways, including fewer costs expended dealing with problem loans.

It maybe matter of taste sto w hether market conditions that relargely exogeno usto any one bank should be included or excluded from measured productivity growth,but it is clear that the extent to which researchers control for these conditions canmake a substantial difference to measured productivity growth. It might also beargued that to examine the effects of technological progress in particular, it maybebettertocontrol formoreexogenous bu siness conditions that are not directly relatedto technological change.

It remains to be explained why costs increased somewhat and at least by onemeasure, cost productivity declined dramatically during this period of the 1990swhen it seems likely that technological progress would have increased productivity.The answ er may lie in unmeasured im provem ents in service quality and variety. Usingthe same specification of business condition variables, Berger and Mester (2003)found that profit productivity improved by between 13,7% and 16,5% annuallyduring the 1991-1997 period even w hile cost productivity declined by 12,5% annu-ally. Profit productivity includes revenues as well as costs and is estimated using aprofit function in place of a cost function. Profit productivity may be superior tocost productivity because profit maximization more completely describes theeconomic goals of managers and owners, who take revenues into account as wellas costs. An additional benefit is that changes in profits include to some degreeunmeasured changes in service quality and variety to the extent that customers payfor these improvements above and beyond the costs of providing the improvedquality and variety.

The cost and profit productivity findings are consistent with the hypothesis thatover time, banks have provided an improved array of services (e,g,, mutual funds,derivatives, on-line services, etc) that increased bank costs, but were able to raiserevenues to more than cover these costs. The small business credit scoring (SBCS)


22/37

162 : MONEY , CRED IT, AND BANKING

extended loans to more costly marginal applicants, but charged them higher loanrates. As argued above, to the extent that markets are competitive, the benefits fromthese improvements may be competed away, and so this hypothesis requires thepresence of market power. The finding of continued increases in profit productivityover time suggests that banks may have retained some market power over theseinnovations over time until other banks were able to fully adopt them. Forexample, adoption of SBCS took a number of years even for large banks, eventhough the product was franchised and seemingly relatively easy to adopt.

Some additional banking research has looked directly at IT investments and usebased on a survey of retail banking practices conducted in 1993 and 1994 bythe Wharton Financial Institutions Center. Respondents included most of the largeU.S. banks and holding companies, covering over 75% of industry assets. Severalstudies were conducted matching the survey data to the bank Call Report. Thestudies were performed on a cross-section basis comparing efficiency of individualinstitutions because of the one-time nature of the data. Nonetheless, the findingsmay be useful for assessing the effects of IT investments on productivity growth.Presumably, if IT investment increases productivity over time, the firms that investthe most in IT will have sup erior efficiency at any point in time i.e., be close tothe best-practice frontier.

The findings from these studies, some of which are summarized in Frei, Harker,and Hu nter (2000 ), do not suggest that IT investment by itself increases efficiencybanks that invested more in IT equipment were not significantly more efficient.How ever, an important efficiency source may be in the proper u se of the equip-ment. The authors found that investment in IT laboremployees that specialize inIT hardware or softwareis efficiency increasing. It is also possible that most ofthe productivity gains from IT investment m ay have occurred after the sam ple periodin the mid- and late 1990s, when productivity gains became apparent in the rest ofthe economy.

4. TECHNOLOGICAL PROGRESS AND THE STRUCTUREOF THE BANKING INDUSTRYTechnological progress may also affect industry structure, facilitating consolida-tion by making it more efficient or less inefficient at the margin for banks tobe larger, more geographically dispersed, and/or to engage in M&A activity. Thesearguments do not imply that it is efficient to have a highly consolidated industryjust that at the margin, there may be more economies or fewer diseconomies toconsolidation due to technological advances. Of course, it is also theoreticallypossible that technological changes may deter consolidation, but in the interest of

brevity, we focus only on the more likely case in which consolidation is facilitated.We also focus only on comm ercial banking, and do not discuss potential econom ies


23/37


4 1 Technological Progress and Ba nking Organization SizeTechnological progress may facilitate increases in bank size in at least fourdifferent ways. First, it may create new services that are subject to more scaleeconomies or fewer diseconomies than traditional services. For example, IT-driveninnovations for delivering depo sitor services , such as call centers, ATMs, and Internetbanking, may exhibit greater economies or less diseconomies ofsc lethan traditionalbranching networks (Radecki, Wenninger, and Orlow 1997), Similarly, some whole-

sale products that are financial technology-driven, such as securitization, derivatives,and other off-balance activities may be more efficiently provided at the margin bylarge banks, consistent with the dominance of large banks in these products.Second, technological progress may create new technologies for producing ex-

isting banking services that are subject to greater scale economies or fewer disecono-mies than the technologies they replace. As discussed above, newer electronicpayments processing technologies may have increased scale economies due to ITinnovations. Similarly, credit scoring may be characterized by greater scale econo-mies or fewer diseconomies than the lending technologies it replaces. This maybe most likely when credit scoring supplants relationship lending, which may bemost efficiently provided by small institutions.

Third, technological progress may allow large banks to push out their risk-expected return frontiers more than small banks,'^^ Any scale economies in new riskmanagement systems may help large banks control risks more than small banks.This may expand or create advantages for large banks to make high risk-highexpected return investments, improve access to uninsured funding, and/or economizeon costly equity capital.

Fourth, technological progress may reduce managerial diseconomies of scale, ITadvances may improve m onitoring and control w ithin large banks more than withinsmall banks. These technologies may make it easier for managers of large banksto monitor the behavior of their staff reducing agency problems and better aligningincentives within the bank. New technologies may also help spot operational prob-lems and keep track of the profits and risks associated with different operations. Inaddition, many of the advances in financial technologies are based more on hardquantifiable and verifiable information that may be easier for the management oflarge organizations to track. To the extent that new products of financial engineering,such as derivative contracts or loans based on credit scores, are more the provincesof large banks, they may improve their monitoring of performance more thansmall banks.

Turning to the empirical ev idence, research on bank co st scale efficiency usingdata on U,S, banks from the 1980s generally found that the average cost curve hada relatively fiat U-shape, Even for small banks , the measu red inefficiencies wereusually relatively small, on the order of 5% of costs or less (e,g,. Hunter andTimme, 1986, Berger, Hanweck, and Humphrey, 1987, Noulas, Miller, and Ray,1990, Clark, 1996), Som e research u sing data from the early and mid-199 0s suggests


24/37


1990s than in the 1980s and that these economies may have continued increasingduring the 1990s (e.g., Berger and Mester, 1997, Stiroh, 2000).Technological progress may also be reflected in improved revenue scale efficienc-ies.This may occur if technological change creates new services or improved qualitythat increase revenues or if the technological change improves the risk-expectedreturn frontier and banks take some of the benefits in higher expected returns. Onestudy ofU.S.bank revenue scale efficiency found slight scale efficiencies on the orderof 1% to 4% of revenues in 1984, but these efficiencies were dissipated by 1990(Berger, Hum phrey, and Pulley 1996). Some research has exam ined profit efficiency,which includes hoth costs and revenues. The effects of scale on bank profitefficiency are ambiguous, with profit efficiency sometimes being highest for largebanks (Berger, Hancock, and Humphrey 1993), sometimes being highest for smallbanks (Berger and Mes ter 1997), and sometimes about equal for large and small banks(Clark and Siems 1997).

Some studies also found a superior risk-expected return frontier for large banksrelative to small banks . One early study found scale efficiency from diversificationof loan risk as bank loan portfolio sizes increased up to about 1 billion (M cAllisterand McManus 1993). Others found that higher ratios of equity capital are associatedwith greater resources devoted to managing risks, and that these resource costs arelower for the largest U.S. banking organizations, consistent with scale efficiency(e.g., Hughes et al., 1996, Hughes and Mester, 1998). Other research found thatlarge institutions take the benefits of an improved risk-expected return frontierprimarily in higher expected returns by increasing risky loans and reducing equityratios (e.g., Demsetz and Strahan 1997).One study examined changes over time in the ability of the senior managers ofa multibank holding company (MBHC) to control their affiliate banks by measuringthe extent to which the efficiency rank of a nonlead bank affiliate varies with theefficiency rank of the lead bank (the largest bank in the MBHC). The increases incontrol based on profit efficiency ranks over time w ere statistically and econom icallysignificant, and increased on the order of 50% to 100% over the period 1984-1998,consistent with the hypothesis that technological progress has reduced managerialdiseconomies of scale (Berger and D e Young 2 002).

4 2 Technological Progress and the Geographic Expan sionof Banking Organizations

Technological progress may also facilitate the geographic expansion of bankingorganizations beyond the effects of the increases in bank scale associated withthe expansion. First, some new services created by technological progress may bedelivered with fewer distance-related diseconomies than traditional services. Forexample, customers do not need to be geographically proximate to receive servicesover the Internet or to purchase financial derivatives, and the ban k's cost of prov iding


25/37


Second, new technologies may deliver some traditional banking services withfewer distance-related diseconomies. As noted above, ACH data was moved fromthe physical delivery of computer tapes to electronic delivery, reducing costs associ-ated with distance. As well, credit scoring does not require geographic proximityor local knowledge to screen a potential loan applicant the way that the relationshiplending do es, and may be provided at greater distances with little or oadditional co st.

Third, changes in lending technologies and innovations in risk management tech-nologies may allow banks to monitor and control risk exposures at long distancesat less cost. As examples, banks' abilities to observe deteriorations in the qualityof loans issued in other nations and to model country risks may have improved overtime, allowing them to issue credits at greater distances with less additional costand risk.Fourth, technological progress may reduce managerial diseconomies of distance.Similar to the arguments above regarding managerial diseconomies of scale, im-provements in IT and financial technologies may make it easier for bank managersto improve monitoring and control over more distant staffA number of recent empirical studies found that U,S, banks have been increasingthe distances at which they make small business loans (e,g,, Cymak and Hannan,2000, Petersen and Rajan, 2002, Wolken and Rohde, 2002), although one studyfound very little change in distance between small firms and their lender in Belgium(Degryse and Ongena 2002), One study also examined the elfects of distance onthe efficiency of hank ho lding com pany affiliates over time . It found that the negativeeffect on efficiency of the distance between a nonlead bank and the lead bank in aU,S, MBHC was slightly lessening o ver time from 1985 to 1998 (Berger andDe Young 2002 ), The greater ability to lend over distances and the slightly lesseningnegative effect of distance on bank efficiency in the U ,S, are consistent w ith at leastslightly reduced distance-related diseconomies due to technological progress,^'4 3 Technological Progress and the C onsolidation Process

Technological progress may also facilitate the consolidation process itself byhelping banks engaged in M As improv e X-efficiencyi,e,, m ove them closer tothe best-practice frontieror reduce the X-efficiency losses associated with anM A, First, new banking products created by technological prog ress may createopportunities for efficiency improvements through the faster spread of new productsthrough consolidation. For example, a bank that operates a transactional Intemetwebsite may bring this technology to a bank it acquires and raise the X-efficiencyof the institution.

Second , banks m ay achieve a faster spread of new efficiency-increasing technolo g-ies for producing traditional banking services through M As, For example, a bankthat employs SBCS may apply this technology to some of the loan applications ofan acquired bank , or a bank using advanced risk managem ent techniques m ay applythese to the acquired bank's portfolio.Third, improvements in IT and financial models may make it easier for acquiring


26/37

166 ; MON EY. CRED IT, AND BANKING

improvements. That is, acquirers may better identify banks that do not offerrecently developed services, that do not employ up-to-date technologies in producingtraditional services, that are poorly managed, or that have risk profiles that meshwell with the acquiring institution.

Fourth, technological progress may reduce the transition costs and risks associatedwith M& As. For example, improved IT and risk managem ent m ethods may speed theprocesses of integrating the computer and risk control systems, reducing the costsof integration and the amount of time during which managers may be unaware ofdeveloping problems.The em pirical evidence on U.S . bank M& As using data from the 1980s generallyfound little or no cost X-efficiency improv emen t after co nsolida tion. Ty pically

5% or less of costs were saved on average, although some M&As yielded substan-tial gains and others yielded substantial losses (e.g., Berger and Humphrey, 1992b,Rho ades, 1993, De Young, 1997, Peristiani, 1997). The studies using U .S. data fromthe early 1990s were mixed. As examp les, one study of M&A s of largeU.S.institutionsfound modest cost X-efficiency gains (Rhoades 1998), while another study foundvery little improvement in cost X-efficiency for M&As of either large or smallbanks (Berger 1998). Studies of European M&As also give mixed cost efficiencyfindings (e.g., Vander Vennet, 1996, Resti, 1998, Haynes and Thompson, 1999).Profit efficiency studies of U.S. bank M&As from the 1980s and early 1990sfound that M&As improved profit efficiency and that this improvement could be

linked to improved diversification of risks (Akhavein, Berger, and Humphrey, 1997,Berger, 1998). After consolidation, banks shifted their portfolios from securities toloans, had lower equity ratios, and more uninsured purchased funds raised atreduced rates. Studies using similar measures found consistent results (e.g., Fixlerand Zieschang, 1993, Hughes et al., 1999). These findings are consistent with thehypothesis that an improvement in portfolio diversification allowed institutions tomake additional high risk-high expected return investments without additionalequity. The study above that found that profit productivity increased while costproductivity declined during the 1990s also found that banks that had recentlymerged appeared to be responsible for much of the findings (Berger and Mester2003). This is consistent with the hypothesis that banks involved in M&Asspread the new or improved services afforded by technological advances to theacquired banks.^

5. CONCLUSIONSResearch on the banking industry provides a wealth of information about techno-logical progress. Banks intensively use modem technologies and the detailed dataon this industry allow for investigations of the effects of advances in both IT andfinancial technolog ies and in both front-office and back-office techn ologie s.Banking industry data give opportunities to investigate examples in which individual


27/37


The detailed data also allow researchers to link technolog ical progress to prod uctivityand other indicators of performance using multiproduct cost and profit functions andother methods. These methods help account for improvements in service qualityand variety and address other difficulties inherent in the use of labor productivity in-dexes. The banking data also allow for analysis of the effects of technologicalprogress on banking industry structureor the extenttowhich technological progressfacilitates consolidationusing statistics on bank scale, distances, and mergers andacquisitions (M&As).

The raw data on the banking industry show the changes over time in the use oftechnology, including the shift to IT-based delivery systems like ATMs and Internetbanking and the proliferation of financial technologies such as financial derivativesand off-balance sheet credit commitments. The data also show improvements inbank performance and consolidation of the industry during the deployment of newtechnologies, although establishing the links between technological progress andboth banking industry productivity growth and industry structure require multivari-ate analyses.

The microeconomic research on banking technologies also yields a number offindings. IT advances appear to have increased productivity and scale economiesin processing electronic payments that have reduced costs dramaticallyin somecases by more than 50% during the 1990s and may help explain some of the recentshift of customers from paper to electronic payments. The use of the small businesscredit scoring technology also appears to have yielded benefits by increasing lendingto marginal applicants that might not otherwise receive bank credit. Notably, thisfinding required use of special banking data sets with information on loan portfoliocompositioni.e., banks using the technology issued more small loans with rela-tively high interest rates and risky credit ratings than otherwise comparable banks.Such a finding would not be easily captured in conventional output or productivitymeasures. The evidence on Internet banking is less clear because of the lack ofexperience with this technology, but it appears the largest U.S. banks that serve thevast majority of customers have adopted this technology, adding transactional In-ternet sites to existing phy sical offices and ATM networks. Sim ilar to the experien cewith ATMs in the early 1980s, competitive pressures m ay result in customers gainingmost or all of the benefits of Internet banking, making it difficult to measure theproductivity gains. These examples illustrate the multiplicity of different actualand measured effects of technological progress. The microeconomic research alsosuggests that (1) most of the important new technologies were generally adoptedearlier ylarge banks than small banks, (2) even small banks can take some advantageof scale economies associated with back-office operations through outsourcingand other methods, and (3) the effects of a new technology may vary significantlywith how it is implemented.

There are a numb er of difficulties in linking technological progress to productivitygrowth, and the econometric studies of bank productivity address some of thesedifficulties. The BLS labor productivity index for the commercial banking industry


28/37


in the early 1980s versus the micl-1990s. The econometric studies, in contrast,generally found that bank productivity growth was poor or negative during the early1980s and improved in later periods. Both the research measures and the BLSmeasure may understate the productivity gains during the early 1980s because theydo not account for the transfer paymen ts to consum ers from increased co mpe tition dueto deregulation. The BLS measure may also exclude gains from the increasedoutput of financial market products. However, the BLS measure may also overstateproductivity gains by counting any substitution of interest expenses for laborexpenses and any movement of labor into service bureaus as productivity increases.The data suggest that these overstatements may dominate the understatements.

The research findings for the 1990s suggest that after controlling for exogenousmarket conditions, bank cost productivity declined but profit productivity improved.These findings are consistent with the hypothesis that technological progressresulted in improved quality and variety of banking services that increased costs, andthat customers were willing to pay for these improvements so banks were able toraise revenues sufficiently to more than cover the higher costs.

Technological progress may also have important effects on the structure of anindustry, facilitating consolidation by making it more efficient or less inefficient atthe margin for firms to be larger, more geographically dispersed, and/or to engagein M&A activity. The banking data are at least somewhat consistent with all three ofthese eifects. The research suggests that the

berger - 2003 - economic effects of technological progress in the banking industry

Documents