RESERVE BANK OF S EVIEW JANUARY EBRUARY OLUME UMBER · Michael R. Pakko Patricia S. Pollard Macroeconomics Richard G. Anderson James B. Bullard Michael J. Dueker Hui Guo Kevin L

FEDERAL RESERVE BANK OF ST. LOUIS

R E V IE WVOLUME 83, NUMBER 1JANUARY/FEBRUARY 2001

The Rise and Fall of a Policy Rule: Monetarism at theSt. Louis Fed, 1968-1986

R. W. Hafer and David C. Wheelock

Comparing Manufacturing Export Growth Across States:What Accounts for the Differences?

Cletus C. Coughlin and Patricia S. Pollard

Membership Structure, Competition, and Occupational CreditUnion Deposit Rates

William R. Emmons and Frank A. Schmid

Retail Sweep Programs and Bank Reserves, 1994-1999

Richard G. Anderson and Robert H. Rasche

Director of Research Robert H. Rasche

Associate Director of ResearchCletus C. Coughlin

Review EditorWilliam T. Gavin

BankingWilliam R. EmmonsR. Alton GilbertFrank A. SchmidDavid C. Wheelock

InternationalChristopher J. NeelyMichael R. PakkoPatricia S. Pollard

Macroeconomics Richard G. AndersonJames B. BullardMichael J. DuekerHui GuoKevin L. KliesenMichael T. OwyangDaniel L. Thornton

RegionalRuben Hernandez-MurilloHoward J. WallAdam M. Zaretsky

Managing EditorGeorge E. Fortier

Assistant EditorLydia H. Johnson

Graphic DesignerJeanette E. Hale

Review is published six times per year by theResearch Division of the Federal Reserve Bank of St. Louis. Single-copy subscriptions are availablefree of charge. Send requests to: Federal ReserveBank of St. Louis, Public Affairs Department, P.O. Box 442, St. Louis, Missouri 63166-0442, or call (314) 444-8808 or 8809.

The views expressed are those of the individualauthors and do not necessarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors.Articles may be reprinted or reproduced if the sourceis credited. Please provide the Public AffairsDepartment with a copy of the reprinted materials.

© 2001, The Federal Reserve Bank of St. Louis. Articlesmay be reprinted, reproduced, published, distributed,displayed, and transmitted in their entirety if this copy-right notice is included. Please provide the ResearchDivision, Federal Reserve Bank of St. Louis, PostOffice Box 442, St. Louis, Missouri, 63166-0442, witha copy of any reprinted, published, or displayedmaterials. Please note: Abstracts, synopses, and otherderivative works may be made only with prior writtenpermission of the Federal Reserve Bank of St. Louis.Please contact the Research Division at the aboveaddress to request permission.

R E V I E W


JANUARY/FEBRUARY 2001 i

ContentsVolume 83, Number 1

1 The Rise and Fall of a Policy Rule:Monetarism at the St. Louis Fed,1968-1986

R. W. Hafer and David C. Wheelock

From the 1960s to the 1980s, the FederalReserve Bank of St. Louis played an importantand highly visible role in the developmentand advocacy of stabilization policy based onthe targeting of monetary aggregates.Research conducted at the St. Louis Bankextended earlier monetarist analysis that hadfocused on the role of money in explainingeconomic activity in the long run. Their suc-cess in finding apparently robust, stable rela-tionships in both long- and short-run data ledmonetarists to apply long-run propositions toshort-run policy questions, effectively com-peting with alternative views of the time.When the short-run correlation betweenmoney and economic activity went astray inthe early 1980s, however, the efficacy of themonetarist rule and appeals for targetingmonetary aggregates to achieve economic sta-bilization quickly lost credibility. This articletraces the evolution of monetary policyresearch at the Federal Reserve Bank of St.Louis as it moved from the identification oflong-run relationships between money andeconomic activity toward short-run policyanalysis. The authors show how monetaristswere lulled into advocating a short-run stabi-lization policy and argue that this experiencecouncils against overconfidence in our abilityto identify infallible rules for conductingshort-run stabilization policy in general.

25 Comparing Manufacturing Export GrowthAcross States: What Accounts for theDifferences?

Cletus C. Coughlin and Patricia S. Pollard

The expansion of United States manufacturingexports has spread unevenly across states.Cletus C. Coughlin and Patricia S. Pollard useshift-share analysis to account for the differ-

ence between a state’s manufacturing exportgrowth and national manufacturing exportgrowth between 1988 and 1998. Three effectsare examined. The industry mix effect indi-cates that a state should have experiencedexport growth above the national average if itsexports were relatively more concentrated inindustries whose exports expanded fasterthan the national average. The destinationeffect indicates that a state should have expe-rienced export growth above the nationalaverage if its exports were concentrated inforeign markets whose purchases from theUnited States expanded faster than the nation-al increase in exports. The competitive effectis what remains after accounting for thesetwo effects. Coughlin and Pollard find that thecompetitive effect, which in previous researchwas related to increases in human capital perworker, is the key determinant of a state’s rel-ative export performance. Furthermore, theindustry mix and destination effects, whichare of similar importance, are generally domi-nated by the competitive effect in accountingfor a state’s relative export performance.

41 Membership Structure, Competition, andOccupational Credit Union Deposit Rates

William R. Emmons and Frank A. Schmid

How do occupational credit unions set depositrates? This article shows that the answer tothis question will depend on (i) who actuallymakes business decisions in credit unions(who is in control), and (ii) whether localdeposit market competition is important. It isnot obvious who controls occupational creditunions. If the sponsor (the employer) is incontrol, then loans and deposits are priced tomaximize the surplus received by all of thecredit union’s current and potential members(those eligible to join). If members are in con-trol, then a group of members with a majoritycan maximize its own surplus. The group incontrol may include members whose primarypurpose for joining the credit union is to bor-

i JANUARY/FEBRUARY 2001

R E V I E W

row money or, alternatively, to lend money(make deposits). If local deposit-market com-petition is the dominant influence, then inter-nal characteristics of the credit union won’tmatter at all. This study tests the sponsor-control, the member-control, and the market-control hypotheses against each other using alarge sample of occupational credit unionsobserved in 1997. Our results suggest thatsponsors exercise effective control over occu-pational credit unions.

51 Retail Sweep Programs and BankReserves, 1994-1999

Richard G. Anderson and Robert H. Rasche

Since January 1994, the Federal Reserve Boardhas permitted depository institutions in theUnited States to implement so-called “retailsweep programs.” The essence of these pro-grams is computer software that dynamically

reclassifies customer deposits from transactionaccounts, which are subject to statutory reserve-requirement ratios as high as 10 percent, tomoney market deposit accounts, which have azero ratio. Through the use of such software,hundreds of banks have sharply reduced theamount of their required reserves. In manycases, this new lower requirement places noconstraint on the bank because it is less than theamount of reserves (vault cash and deposits atthe Federal Reserve) that the bank requires for itsordinary day-to-day business. In the terminologyintroduced by Anderson and Rasche (1996b),such deposit-sweeping activity has allowed thesebanks to become “economically nonbound” andhas reduced to zero the economic burden (“tax”)due to statutory reserve requirements. In thisanalysis, we examine a large panel of U.S. banksand develop quantitative estimates of the impactof sweep software programs on the demand forbank reserves.

Erratum

The article, “The Information Content of Treasury Inflation-Indexed Securities,” published in theNovember/December 2000 issue of Review contained an error regarding the tax treatment of adjustmentsto the principal of the Treasury inflation-indexed securities (TIIS). The article incorrectly stated (on pp. 29and 36 and in Tables 3 and 4, pp. 31-32) that adjustments to principal of TIIS are taxable at capital-gainsrates. Instead, these adjustments are taxable at ordinary-income rates. Thus, all of the income in theexamples used in Tables 3 and 4 is taxed as ordinary income. A corrected version of the article can befound on our Web site, <http://www.stls.frb.org/publications/review/review00.html#NOV>.


JANUARY/FEBRUARY 2001 1

The Rise and Fall of aPolicy Rule:Monetarism at theSt. Louis Fed,1968-1986R. W. Hafer and David C. Wheelock

“Once the quantity theory regained academicrespectability, it was obliged to resume respon-sibility for the short-run forecasting of aggre-gate movements of prices and quantities This it has begun to do, most importantlythrough the research work of the FederalReserve Bank of St. Louis, and with appreciablesuccess; but it has been lured into playing in anew ballpark, and playing according to a differ-ent set of rules than it initially established foritself [I]ts own success is likely to be transi-tory, precisely because it has relied on the samemechanisms of intellectual conquest as the[Keynesian] revolution itself and has alsoespoused a methodology that has put it in con-flict with long-run trends in the development ofthe subject.”—Harry Johnson (1971, pp. 12-13)

Macroeconomists today generally agree thatmonetary policy cannot permanentlyincrease the rate of economic growth

above its potential or decrease the rate of unem-ployment below its market clearing, or “natural,”level (e.g., De Long, 2000; Woodford, forthcoming).In the long run, monetary policy affects only therate of inflation, and many economists argue thatmonetary policy can best promote maximum sus-tainable economic growth by ensuring price level

stability (e.g., Barro, 1996). Monetary policymak-ing, however, both in the United States and else-where, is often concerned with the short-run.Policymakers meet frequently: the FederalReserve’s Open Market Committee (FOMC) meetseight times a year, for example, and a vote is takenat each meeting on whether to maintain or changethe current stance of policy. While price stability iswidely acknowledged as the appropriate long-runobjective of monetary policy, many economistsargue that policymakers should respond to fluctu-ations in real output or employment as part oftheir strategy to achieve price stability and, ulti-mately, to support maximum sustainable econom-ic growth.1

To help guide them in their deliberations,policymakers and their economic advisors rely onboth complex economic and econometric modelsand on simple rules-of-thumb based on empiricalregularities observed in macroeconomic data.Indeed, simple rules, such as the so-called Taylorrule (Taylor, 1993), which describes the response ofthe federal funds rate to past deviations of inflationand output from target values, often appearto explain well how policymakers set policy in theshort-run. Economists generally conclude thatrules-based policies are preferable to those relyingsolely on the discretion of policymakers. The trans-parency of rules reduces uncertainty about theresponsiveness of policy to economic change andcan enhance the accountability of policymakers.

Monetarists have long advocated the use ofrules to guide monetary policy, with Friedman’s(1960) proposal for a constant money stockgrowth rate being the most famous example. Atthe time it was made, Friedman’s proposal wassharply at odds with the prevailing mainstreamview that monetary policy was best conducted bymanipulating interest rates to strike a balancebetween inflation and unemployment. At the time,economists widely believed in the power ofactivist monetary and, especially, fiscal policy tolimit fluctuations in economic activity and toensure sufficient demand to provide full employ-ment economic growth over extended periods.Friedman’s “monetarist” policy rule thus attractedconsiderable attention. Moreover, a growing bodyof empirical work by Friedman and others show-ing the potential for money supply shocks to havelarge short-run impacts on output and employ-

1For a recent argument along these lines, see Mishkin (2000) and thereferences cited therein.

R. W. Hafer is a professor of economics and finance, and Director,Office of Economic Education and Business Research, SouthernIllinois University Edwardsville. David C. Wheelock is an assistantvice president and economist at the Federal Reserve Bank of St.Louis. Heidi L. Beyer provided research assistance. The authorsthank Anatol Balbach, Rachel Balbach, Michael Bordo, KeithCarlson, Brad DeLong, Bill Dewald, Jerry Dwyer, Milton Friedman,Gail Heyne Hafer, Joe Haslag, Bob Hetzel, Ali Kutan, Thomas Mayer,and Anna Schwartz for comments on an earlier draft. The authorsalso thank Gloria Valentine of the Hoover Institution for her assis-tance with the Friedman papers.

2 JANUARY/FEBRUARY 2001

R E V I E W

ment led to the development of an alternativeframework for conducting stabilization policybased on targeting growth of the money stock.

The Federal Reserve Bank of St. Louis playedan important and highly visible role in the devel-opment and advocacy of stabilization policy basedon the targeting of monetary aggregates. This arti-cle examines the development of the monetarist-based stabilization policy framework advocated bythe St. Louis Bank between the late 1960s and the1980s, with an eye toward identifying lessonsfrom that experience for the conduct of stabiliza-tion policy in general.2

This article also illustrates how the FederalReserve System’s decentralized structure fosters aclimate of internal debate. Beginning in the 1960s,the monetary policy actions advocated by the St.Louis Bank in its research publications, in publicforums, and in the participation of the Bank’spresidents in policy meetings often were sharplyat odds with the policies adopted by the FederalReserve System. The Fed’s decentralized structurepermitted the development of alternative policyviews and the exploration of new ideas within theSystem (Wheelock, 2000). Policy debates oftentook place within System publications. Wedescribe the public debate as we trace the evolu-tion of the St. Louis Fed’s monetarist policy, andthe criticism of that policy by other Fed officials,in the Federal Reserve Bank of St. Louis Review,other System publications, and in the publicspeeches of Darryl Francis, President of the Fed-eral Reserve Bank of St. Louis from 1966 to 1976.

To provide a backdrop to our discussion, thenext section summarizes the dominant policypositions taken by economists during the 1960sand early 1970s about the causes of inflation andthe role of monetary policy. We point out keyissues where monetarists disagreed with the con-ventional wisdom. In the subsequent two sectionswe review the research and policy positions takenby St. Louis Fed economists and officials from the1960s through the early 1980s. We limit our dis-cussion mainly to the period 1968-86. In 1968,the Federal Reserve Bank of St. Louis published aneconometric analysis of the relative impacts ofmonetary and fiscal policy on economic activity.That article, by Andersen and Jordan (1968),became one of the most cited papers in econom-ics in the past 40 years. In 1970, the Bank pub-lished the first version of its monetarist model forgauging the impact of monetary policy actions oneconomic activity (Andersen and Carlson, 1970).

The year 1986 marked the appearance of the finalversion of that model (Carlson, 1986). We showhow research conducted at the Federal ReserveBank of St. Louis extended earlier monetaristanalysis that had focused on the role of money inexplaining economic activity in the long run.Their success in finding apparently robust, stablerelationships in both long- and short-run data ledmonetarists to apply long-run propositions toshort-run policy questions, effectively competingwith alternative views of the time. When theshort-run correlation between money and eco-nomic activity went astray in the early 1980s,however, the efficacy of the monetarist rule andappeals for targeting monetary aggregates toachieve economic stabilization quickly lost credi-bility.

THE SETTING: FINE-TUNING THEECONOMY

The 1960s were the glory days of activist,short-run stabilization policy. Policymakers hadconfidence in their ability to achieve full employ-ment using fiscal and monetary policy to “finetune” or manage aggregate demand. For a time,their confidence seemed justified: Before the pres-ent expansion, the 1960s witnessed the longestuninterrupted expansion in U.S. history, with theeconomy operating at full employment (definedthen as a civilian unemployment rate of 4 percentor less) from 1966 to 1969.

Beginning in 1965, however, rising inflationand an increasing balance of payments deficitreflected the cost of expansionary macroeconom-ic policies. Policymakers felt increasing pressureto control inflation but were hesitant to takeactions that might reduce employment and realoutput growth. Because policymaking was viewedas striking a balance between inflation and unem-ployment, disagreements about policy, accordingto one Fed governor, boiled down to one’s prefer-ences between the two outcomes.3 Because un-employment frequently was viewed as a more

2The defining characteristics, technical aspects, and legacy of mone-tarism in general have been explored elsewhere, e.g., DeLong(2000), Melzter (1998), Rasche (1993), and Woodford (forthcoming).

3 Governor Sherman Maisel argued at an FOMC meeting on October20, 1970, “that at least some of the Committee’s differences on pol-icy reflected difference in basic value judgments regarding the rela-tive importance of various conflicting goals—for example, regard-ing the appropriate trade-off between employment and price stabil-ity” (FOMC Minutes, October 20, 1970, p. 41).



serious concern than inflation, for many years theFed opted for maintaining an inflationary bias inmonetary policy to avoid higher rates of unem-ployment. Fed officials, like other governmentofficials, argued that the inflation-unemploymenttradeoff could be improved only through the coop-eration of business and labor in the setting ofprices and wages or, if necessary, by the use ofanti-trust and other polices to make price settingmore competitive. Alternatively, inflation could becontrolled through explicit regulation of wages andprices

Juxtaposed against this mainstream view wasan alternative associated with the work of MiltonFriedman, Anna Schwartz, Karl Brunner, AllanMeltzer, and other so-called monetarists. Monetar-ism was rooted in the Quantity Theory of Money.The core of the Quantity Theory is that, in the longrun, inflation reflects excessive growth of themoney stock relative to real output growth, the lat-ter determined fundamentally by non-monetaryforces such as population growth and productivity.An important component of this position is the sta-bility of the public’s demand for money.Monetarists amassed empirical evidence showingthe demand for money to be more stable thanmoney supply or, equivalently, that velocity is sta-ble.4 Stability of velocity supported monetarists’view that short-run fluctuations in economic activ-ity often are caused by fluctuations in money sup-ply growth—fluctuations brought about by centralbank policy actions.5 Monetarists concluded thatcentral bank attempts to manipulate interest rateshad led to destabilizing fluctuations in money sup-ply growth and, therefore, in economic activity.Hence, monetarists argued that monetary policy-makers should minimize the variation of thegrowth rate of the money stock both in the short-run and over time. Lags in assessing economicconditions and in the effects of policy actions oneconomic activity, they argued, made attempts at“fine tuning” a balance between inflation andunemployment futile. Instead, monetarists arguedfor a policy that maintained growth of the moneystock at a low, fixed rate, irrespective of the busi-ness cycle.

The Federal Reserve Bank of St. Louis was thecenter of monetarism within the Federal ReserveSystem from the 1960s into the 1980s.6 DarrylFrancis, who became president of the St. LouisBank in 1966, was an especially strong advocateof monetarist policy prescriptions. At FOMCmeetings, he argued frequently for a policy of

minimizing variability in money stock growtharound a moderate trend. In public forums,Francis made the case that growth of the moneystock was the most accurate reflection of monetarypolicy and that excessive monetary growth wasthe fundamental cause of inflation. Francis wassupported in his policy views by St. Louis Fedeconomists, whose research findings were largelyin accord with those of other monetarists butsharply at odds with the conventional wisdom ofthe times, including the views about monetarypolicy held in most quarters of the FederalReserve System.

Mainstream Views About MonetaryPolicy in the 1960s

A review of monetarism and all of its differ-ences with mainstream Keynesian macroeconom-ics of the 1960s is beyond the scope of this article.Sharp theoretical differences about the cause orcauses of inflation and economic fluctuations, aswell as methodological differences about theempirical analysis of the effects of policy actionson economic activity, however, are fundamental tounderstanding why St. Louis Fed research and poli-cy positions were controversial. These differencesinclude whether a tradeoff exists between inflationand unemployment, either in the short or the longrun, and whether such a tradeoff is exploitable bypolicymakers. A related question is whether mone-tary policy can, or should, be used to fight inflationwhen the economy is at less-than-full employ-ment. Monetarists and non-monetarists dividedalso over the measures that best reflect the impactof monetary policy actions—monetary aggregatesor interest rates and credit aggregates. Finally,monetarists and non-monetarists debated the toolsand methods for identifying the impact of mone-tary policy on the economy.

4Velocity was stable not as an arithmetic constant, but in the sensethat its behavior was related predictably to changes in the opportu-nity cost of holding money and to changes in income or wealth ofindividuals.

5 Perhaps the most famous statement and evidence of this proposi-tion is Friedman and Schwartz (1963).

6 Consider Friedman’s (1992) appraisal: “The interesting thing to mehas always been that the most important contributions to under-standing of monetary theory and monetary institutions have notcome from Washington during the decades in which I’ve beenactive. The Federal Reserve Bank of St. Louis in the 1950s, ’60s and’70s was by far and away the pre-eminent producer of significantmonetary research within the System.”


R E V I E W

Inflation and the Phillips Curve

The Phillips curve—the inverse relationshipbetween unemployment and wage growth inBritish historical data observed by Phillips(1958)—was a key empirical regularity at the heartof macroeconomic policy discussions in the1960s. The Phillips curve was viewed as a con-straint on stabilization policy; policymakers withboth unemployment and inflation goals had tochoose a feasible combination of the two becausepolicies that, say, reduced inflation would invari-ably slow economic activity and increase unem-ployment. Perry (1966, p. 3) noted that

a fairly general consensus exists among econ-omists [that s]uccessively higher levels ofactivity are associated, roughly at least, withcorrespondingly larger rates of price increase.In this situation, the more traditional problemof adjusting aggregate demand so as to reachfull employment without overshooting intothe area of inflation must be replaced with thedual problems of deciding what combinationof unemployment and inflation to aim at andthen adjusting aggregate demand to reach thispoint.

Fed governor Sherman Maisel (1973, p. 14)observed succinctly: “There is a trade-off betweenidle men and a more stable value for the dollar.A conscious decision must be made as to howmuch unemployment and loss of output is accept-able in order to get smaller price rises.”

Although many influential economistsbelieved that the hyperbolic shape and negativeslope of the Phillips curve were fixed, it was clearthat the position of the Phillips curve could moveover time. The U.S. experience of 1955-58 waswidely discussed: In 1955-57, the unemploymentrate hardly changed when the inflation rateincreased sharply; in 1958, when inflation abated,the increase in the unemployment rate seemeddisproportionately high. Concern arose that theU.S. economy had an “inflation bias,” meaningthat inflation, perhaps accelerating inflation, wasnecessary to achieve and maintain full employ-ment (Wheelock, 1998).

The experience of 1955-58 gave prominenceto the notions of “cost-push” inflation and “wage-price spirals.” The term cost-push inflation wasused to define an ongoing increase in the generallevel of prices caused by firms passing along

increases in production costs, in contrast to“demand-pull” inflation caused by increases inaggregate demand. Cost-push forces were thoughtto explain how inflation could occur at less thanfull employment. The “essence of the problem,”according to Samuelson and Solow (1960, p. 181),stems from the absence of perfect competition infactor and product markets and was, Bronfen-brenner and Holzman (1963) noted, associatedwith the power of “economic pressure groups,”such as labor unions and monopolistic firms.If a powerful union extracts a real wage increase,the quantity of labor demanded falls. If affectedfirms have pricing power, they pass along a por-tion of the increase in wages to consumers in theform of higher prices. If perfect competition existsin other industries and labor markets, prices andwages fall to offset the increases in the first indus-try. If wages are downwardly rigid throughout theeconomy, however, the aggregate price level mightrise alongside an increase in unemployment.

Throughout the 1960s, the Economic Report ofthe President blamed inflation on “excessive” wageand price increases. The clear implication was thatmonetary and fiscal policies were not responsiblefor inflation when the economy was at less thanfull employment. The Economic Report for 1965explained that “in a world where large firms andlarge unions play an essential role, the cost-pricerecord will depend heavily upon the responsibilitywith which they exercise the market power thatsociety entrusts to them” (1966, p. 179). Hence,President Johnson urged, “in the strongest terms Iknow—that unions and business firms exercisethe most rigorous restraint in their wage and pricedeterminations” (Economic Report of the Presidentfor 1967, p. 20). Throughout the 1960s theEconomic Report published detailed wage andprice “guideposts” defining the extent to whichwage and price increases were “justifiable.”

Monetarists dismissed the notion of cost-pushinflation, arguing that inflation is “always andeverywhere a monetary phenomenon.” Non-monetarists generally accepted that “cost-pushmay involve increases in money supply, moneyincome, and money expenditures, particularly ifdecreases in output and employment are to beavoided” (Bronfenbrenner and Holzman, 1963, p. 614). Nonetheless, many economists foundmonetary explanations of inflation wanting,claiming that inflation can arise through wage andprice setting independent of shocks to aggregate



demand. The short-run linkages from money toinflation were considered to be tenuous at best.7

Economists generally did acknowledge thepotential for restrictive monetary policy to elimi-nate cost-induced inflation. The apparent down-ward rigidity of prices and wages, however, con-vinced many that using monetary policy to arresteven a moderate inflation would entail a substan-tial and unacceptable increase in unemployment.The Economic Report of the President for 1961 (p. 47) claimed that “an attempt to restrict aggre-gate demand so severely as to eliminate all risk ofan increase in the general price level might wellinvolve keeping the economy far below full em-ployment.”8 High unemployment was simply anunacceptable cost of reducing inflation. In hisintroduction to the 1967 Report (p. 20), PresidentJohnson argued, “Dealing with inflation by creat-ing a recession or persistent slack is succumbingto the disease—not curing it. The experience of1957 and 1958—when the unemployment ratereached 7 12 percent and consumer prices still rose5 percent—is a clear reminder of the large costs ofsuch a policy and of its limited effectiveness inhalting a spiral in motion. This is a course which Ireject.”

To fight inflation, some mainstream econo-mists advocated incomes policies; others favoredpolicies aimed at enhancing the competitivenessof product and labor markets and policies aimedat raising productivity. Monetary policy, manyargued, should focus on maintaining full employ-ment by keeping interest rates low.

Many Federal Reserve officials expressed simi-lar qualms about using monetary policy to controlinflation. For example, Charles Partee, a Fed staffeconomist who later became a Fed governor,argued in 1970 that “The question is whethermonetary policy could or should do anything tocombat a persistent residual rate of inflation The answer, I think, is negative.” He added that“Product markets generally are substantiallyunderutilized and labor appears to be readilyavailable It seems to me that we should regardcontinuing increases [in the price level] as a struc-tural problem not amenable to macro-economicmeasures” (FOMC Minutes, April 1970, pp. 385-86,379, 396 [quoted in Mayer (1999), p. 99]). FederalReserve Board Chairman Arthur Burns also arguedthat “Monetary policy could do very little to arrestan inflation that rested so heavily on wage-costpressures . . . A much higher rate of unemploy-ment produced by monetary policy would not

moderate such pressures appreciably” (FOMCMinutes, June 8, 1971, p. 51).

The Long-Run Phillips Curve

Wage and price rigidity and the possibility ofcost-induced inflation suggested to manyobservers that using monetary policy to containinflation would invariably cause higher unemploy-ment. Monetarists did not deny this, but arguedthat the tightening of monetary policy would havemerely a temporary adverse impact on unemploy-ment. Andersen (1973), for example, argued that“our economic system is such that disturbingforces, including even changes in money growth,are rather rapidly absorbed and output willnaturally revert to its long-run growth path follow-ing a disturbance” (p. 7).

Some economists, however, argued that usingmonetary or fiscal policy to achieve price stabilitymight cause unemployment to increase perma-nently. Keynesians frequently interpreted theGreat Depression as indicating that privatedemand might be insufficient to generate fullemployment output in the face of downwardlyrigid wages and prices, thereby leaving the econo-my mired permanently at less than full employ-ment. Thus, many economists believed that monetary and fiscal policy should ensure thataggregate demand is sufficient to generate fullemployment, even if that requires some inflation.Samuelson (1960, p. 265), for example, arguedthat “With important cost-push forces assumed tobe operating, there are many models in which itcan be shown that some sacrifice in the require-ment for price stability is needed if short- andlong-term growth are to be maximized, if averagelong-run unemployment is to be minimized, ifoptimal allocation of resources as between differ-ent occupations is to be facilitated.”

The Fed often was accused of paying “exces-sive” attention to price stability. In summarizing asymposium on recent monetary policy, Harris(1960, p. 245) wrote, “In general the disagree-

7Ackley (in Perlman, 1965, p. 47), for example, wrote that “I do notconsider the change in money supply of much short-run impor-tance.” Weintraub (1960, p. 280) contended that “contrary to thewidespread belief that there is a direct tie of money supplies toprice levels, the modus operandi of monetary policy is different: itsimmediate effect is upon consumption and investment demands,thereby upon employment levels, and only indirectly affecting thegeneral level of money wages” (emphasis in the original).

8 The Council of Economic Advisors at this time consisted of WalterHeller, Kermit Gordon, and James Tobin.


R E V I E W

ments of [participating] economists with FederalReserve policy have stemmed primarily from afear that the interest in the stability of the currencyhas been at the expense of growth and employ-ment.”9 Not only was the tradeoff between infla-tion and unemployment widely viewed as persist-ent, but some economists argued that policies toachieve price stability in the short-run might makethe tradeoff less favorable over time. Samuelsonand Solow (1960), for example, believed that poli-cies directed at limiting inflation in the short-runmight increase structural unemployment. Thelong-run tradeoff between inflation and unem-ployment would worsen because an increase instructural unemployment would raise the size ofthe increase in inflation that would be needed toachieve a given reduction in the unemploymentrate. Perry (1966, p. 119) even suggested that poli-cies aimed at maintaining low unemployment andminimizing short-run cyclical variation in outputcould moderate wage and price increases.10

The Impact of Monetary Policy

Economists’ conviction that policymakerscould manipulate aggregate demand to stabilizethe growth of real output reached its zenith in the1960s and early 1970s. The apparent failure oflow interest rates to revive the economy duringthe Great Depression, however, was taken as evi-dence that monetary policy is less potent than fis-cal policy. The 1950s witnessed the developmentof new theories about the impact of monetary pol-icy, with the dominant view being that policy iseffective through its influence on both the costand availability of credit.11 Still, even in the 1960s,the mainstream view was that monetary policy,though capable of having some impact, was lesspowerful than fiscal policy.12

Monetarists, by contrast, argued that changesin the quantity of money exert a powerful influ-ence on economic activity. Friedman’s work dur-ing the 1950s helped establish the foundation forlater studies of the link between monetary policyand the economy (e.g., Friedman, 1956). Hisexamination of monetary policy and the businesscycle is reflected in his testimony to the JointEconomic Committee in March 1958.13 At thattime, most Fed officials believed that short-terminterest rates and the quantity of bank credit werethe appropriate instruments of monetary policy.Minutes of FOMC meetings reveal that policymak-ers generally rejected the notion that movements

in the money supply affect economic activity,much less that the Fed can control the moneystock. Although one or two FOMC memberswarned persistently that fluctuations in moneygrowth could cause undue fluctuations in the realeconomy, these concerns largely were ignored.14

Outside the Fed, by contrast, monetarist viewsabout the impact of fluctuations in money stockgrowth were receiving considerable attention.Friedman and Meiselman’s (1963) “The RelativeStability of Monetary Velocity and the InvestmentMultiplier in the United States, 1897-1958,” forexample, stirred much debate. Associated with theCommission on Money and Credit’s inquiry intothe structure of the financial system, this paperwas controversial because it rejected a core ingre-dient in the Keynesian theoretical structure—thevalidity of the expenditure multiplier. Friedman

9 Angell (1960, p. 248) argued similarly, stating that “As to aggregategrowth, a good many students feel—as do I —not only that mone-tary policy has not done much to promote it, but that the intermit-tent restrictions imposed to fight instability and inflation haveprobably retarded it substantially.” Hansen (1960, pp. 255-56) pro-posed that “Monetary policy should seek to achieve a low long-runrate of interest” to raise permanently the ratio of investment to out-put and thus real growth, while fiscal policy could “offset theinflationary pressures caused by the increase in investment inci-dent to rapid technological advance and low interest rates.”

10 Policymakers often agreed with such assessments. The EconomicReport of the President for 1965 concluded “Rising prices thatoriginate from such a [cost-push] process can affect expectations,jeopardize the stability and balance of an expansion, and createinequities and distortions just as readily as demand inflation. Butmeasures to restrain these price increases by reducing over-alldemand will enlarge unemployment and impair the productivityrecord so important to cost-price stability over the longer run” (p.179). The Council of Economic Advisors at this time consisted ofGardner Ackley, Otto Eckstein, and Arthur Okun.

11 The “Availability Doctrine” posits that small changes in interestrates can have large economic effects by affecting banks’ willing-ness to supply loans. The theory was developed by New York Fedeconomist Robert Roosa (1956) and was the dominant view of thetransmission mechanism within the Fed at the time. See Johnson(1962) for a survey of current thinking on monetary theory andpolicy effectiveness as of the early 1960s.

12 The consensus view is perhaps well represented by Ando et al.(1963, p. 2), who concluded “the effect of monetary policy on theflow of expenditures is far from overwhelming, though it exists andis of a magnitude worth exploiting in the interests of economicstability Our findings on fiscal policy are primarily that varia-tions in disposable income, either through transfer payments orpersonal income tax changes, can operate as a powerful short-runstabilizer.”

13 This testimony was based on on-going research with Anna J.Schwartz at the National Bureau of Economic Research. Thisresearch was subsequently published in three volumes: seeFriedman and Schwartz (1963, 1970, 1982).

14 In the late 1950s and early 1960s, Malcolm Bryan, President of theFederal Reserve Bank of Atlanta, argued that the Fed should targetthe growth rate of total reserves, and minimize fluctuations in themoney stock. See Meigs (1976) and Hafer (1999).



and Meiselman’s empirical analysis also foundmoney demand to be relatively more stable thanmoney supply, suggesting that observed move-ments in the money stock and economic activityare dominated by Fed policy actions rather thanby volatility in the public’s demand for money.

The debate that ensued was so important thatthe American Economic Review devoted an entireissue to critical appraisals of the Friedman-Meiselman (1963) study by Ando and Modigliani(1965) and DePrano and Mayer (1965) and aresponse by Friedman and Meiselman (1965).15

Two issues were key to the attacks and rejoinder.One was Friedman and Meiselman’s finding thatvelocity is relatively more stable than theKeynesian expenditure multiplier. In other words,output appeared to be related more to movementsin money than to other measures of autonomousexpenditure.

The other issue, one that is more important forwhat would later be the attack on the St. Louisapproach, was the procedure Friedman andMeiselman used to produce that finding. Thedebate displayed a fundamental difference inviews about how to estimate economic relation-ships for policy purposes. Friedman andMeiselman, whose work was grounded in the tra-dition of the Quantity Theory of Money, basedtheir conclusions on simple reduced-form rela-tions observed in the data. Ando and Modigliani(1965) argued that such simple regressions, notmuch more than correlations, were inferior to theoutput of the large-scale, structural models thencoming into vogue to evaluate policy. In otherwords, though different approaches may producedifferent results, only the most sophisticated isuseful for policy analysis.16

Many Federal Reserve officials embraced theuse of large-scale econometric models then beingdeveloped. Board staff participated with econo-mists from the Massachusetts Institute ofTechnology in constructing the FRB-MIT model,which was used for policy analysis and evaluationat the Board. Such models were widely judgedsuperior to single-equation or small-model sys-tems for studying the effects of policy actions oneconomic activity. An official of the FederalReserve Bank of New York argued that large-scalemodels produced “quantitative estimates of thetiming and magnitude of the effects of centralbank actions on the money supply and otherfinancial magnitudes and the subsequent effects,in turn, of these variables on each of the variousmajor components of aggregate demand” (Davis,

1968, p. 73). Thus, large structural modelsappeared to give policymakers the informationthey needed to make short-term, fine-tuning poli-cy adjustments to stabilize economic activity.

Monetary Policy Target

Alongside large-scale macroeconometric mod-els, the 1960s witnessed the development ofincreasingly complex analyses of the effects ofmonetary policy actions, interacting with financialregulatory policies, on financial flows and interestrates. With the exception of monetarists, manymacroeconomists believed that monetary policywas not represented accurately by the behavior ofany single variable, such as the rate of interest, thequantity of bank credit, or the stock of money.This view was reflected in the Economic Report ofthe President for 1968:

In the formulation of monetary policy,careful attention should be paid to inter-est rates and credit availability as influ-enced by and associated with the flows ofdeposits and credit to different types offinancial institutions and spending units.Among the financial flows generally con-sidered to be relevant are: the total offunds raised by nonfinancial sectors ofthe economy, the credit supplied by com-mercial banks, the net amount of newmortgage credit, the net changes in thepublic’s holdings of liquid assets, changesin time deposits at banks and other thriftinstitutions, and changes in the moneysupply. Some consideration should begiven to all of these financial flows aswell as to related interest rates in formu-lating any comprehensive policy programor analysis of financial conditions. (p. 89)

The Report also dismissed monetarist appealsfor focusing on money stock growth. In responseto calls for setting monetary policy according to a

15 This is the so-called AM-FM or “radio” debate. Hester (1964) wasalso critical of the Friedman-Meiselman approach.

16 Brunner (1986) provides an excellent and wide-ranging overview ofthis debate. He defends the Friedman-Meiselman (1963) approach,arguing that “the use of a single equation with a single independentvariable should now be clear. It was the appropriate choice for anassessment of the core class [of hypotheses]. It did not represent asingle equation model or a [direct] disposition to favor simple, asagainst sophistical models” (p. 41, emphasis in original). Rather,Brunner suggests (p. 40) that “the strong assertions conveyed bythe basic core of the income-expenditure approach, which fre-quently spilled over into categorical policy statements, were thusshown to have little substantive foundation.”


R E V I E W

fixed-growth rule for the money stock, the Reportargued (p. 92) that “given the complex role ofinterest rates in affecting various demand cate-gories and the likely variations in so many otherfactors, any such simple policy guide could proveto be quite unreliable.” Similar reasoning wasreflected in the analysis of Federal Reserve Boardeconomists. In a paper published in the FederalReserve Bulletin, Gramley and Chase (1965, pp.1403-04) wrote that “the money stock [is] anuntrustworthy indicator of the effects of policyactions on financial asset prices and yields Financial market behavior is too complex for sim-ple monetary rules to work.”17

Monetarists, of course, disagreed that the com-plexity of financial markets made targeting amonetary aggregate infeasible. In an article in theJournal of Political Economy, Federal Reserve Bankof St. Louis economist Leonall Andersen (1968)presented evidence that the Fed could use openmarket operations to smooth fluctuations in totaland free (i.e., excess less borrowed) reserves and,by implication, in the monetary base.18 AlbertBurger (1971), another St. Louis Fed economist,presented a detailed analysis of how the monetarybase can be manipulated so as to control themoney stock.

The Fed’s ability to control monetary aggre-gates and the efficacy of monetary control for eco-nomic stabilization remained hotly debated issues.They also were central research themes at theFederal Reserve Bank of St. Louis throughout the1960s and 1970s.

ST. LOUIS—THE LONG-RUN VIEW

Under the leadership of its director ofresearch, Homer Jones, the Federal Reserve Bankof St. Louis emerged as the center of monetaristeconomics within the Federal Reserve System inthe early 1960s.19 The Bank’s Review, which thenappeared monthly, tracked the behavior of theeconomy and the money stock in nearly everyissue. Review articles often described the recentbehavior of the money stock and related it tomonetary growth during previous expansions. Forexample, the June 1962 article, “MonetaryDevelopments,” compared graphically the recentbehavior of bank reserves and the money stock(M1) with their patterns during the 1953-57 and1957-60 cycles. This type of chart, used byFriedman in his 1958 Senate testimony and inFriedman and Schwartz (1963), provided a visual

analysis of expansionary and contractionarymovements of the money stock.

A companion article in the June 1962 Reviewprovided one of the first monetarist explanationsfrom the St. Louis Fed of how monetary policyactions are transmitted to changes in nominalincome and prices. The article, “Changes in theVelocity of Money: 1951-1962,” addressed one ofthe more difficult questions monetarists faced inattempting to use the Quantity Theory to explainhow monetary policy affects the economy in theshort-run. If velocity is highly variable, then theconnection between changes in the money supplyand nominal income is uncertain at best. The arti-cle provided a “tentative and exploratory analysis”of the behavior of velocity. The analysis showedthat a “rapid change in money [is] to be matchedtemporarily by an opposite change in velocity”(p. 13). Over time, however, “as the public recog-nized the change in its [money] balances therewas an increase in spending, and velocity movedupward” (p. 13). Using this pattern to explain theeffects of past policies, the article noted that“Within a few months after money began expand-ing at a rapid rate in 1954, 1958, and 1961,spending and the velocity of money began rising”(p. 13).

The article is an early example of the researchcoming out of St. Louis in support of the mone-tarist position that changes in nominal incomelargely reflect prior movements in the money sup-ply. Even though velocity might vary, its variabilityappeared to be less than that of money supply.Hence, monetarists argued, observed cycles in

17 It was a long-standing view among Board officials that monetarypolicy should not focus on any one variable. For example, as earlyas 1932, Governor Eugene Meyer stated: “Our credit machinery isentirely too delicate and responsive to too many influences to makeit desirable to have any one indicator, whether it be the price levelor the level of member bank reserves, be the sole guide in determin-ing credit policy” (quoted by A. James Meigs in a letter to MiltonFriedman: Friedman Papers, Hoover Library, Stanford University,Box 30, Folder 17).

18 See also Meigs (1966), who reviews the debate about whether finan-cial innovation had made control of the money stock infeasible orinefficacious.

19 Jones was hired by the St. Louis Fed in 1958. At that time, the Bank’sresearch staff consisted of one Ph.D. economist, two graduate stu-dents, an agricultural economist, a geographer and several juniorstaff members. During 1958, Jones corresponded regularly withFriedman concerning potential hires for the department (FriedmanPapers, Hoover Library, Stanford University, Box 28, Folder 36). Inrecognition of Jones’ accomplishments, in 1976 a special issue ofthe Journal of Monetary Economics was devoted to papers in hishonor.



nominal income growth are caused mainly bychanges in the money supply—over which theFed has some control—rather than by changes invelocity.

As inflation worsened over the 1960s, Reviewarticles reflected an increasingly aggressive appli-cation of monetarist arguments by St. Louis Fedeconomists, as critical reviews of policy replacedtentative and exploratory studies. For example, aJuly 1966 article, titled “Total Demand andInflation,” stated that “Excessively stimulativeGovernment policies lead to marked increases inthe price level. Rapid monetary expansion isregarded by many as a means of stimulating totaldemand” (p. 1). In the same issue, Keran (1966)studied the relationship between nominal and realoutput in eight countries. He concluded that infla-tion results when total demand—nominalincome—rises faster than the economy’s potentialrate of real output growth, suggesting that “If therecent acceleration in total demand is continuedat a time of high-level resource utilization, priceswill probably begin to rise even faster” (p. 12). Inaddition, Keran hinted at the possibility of usingmonetary policy for short-run stabilization, notingthat “To the extent that policy tools control thegrowth in total demand they are useful in achiev-ing cyclical stability in the economy because year-to-year movements in real output can be influencedby changes in total demand” (p. 12, emphasisadded). These and other Review articles, many ofwhich were written anonymously, reflect clearlythe monetarist-oriented research and policy analy-sis carried out by St. Louis Fed economists in theearly-to-mid 1960s.

The St. Louis Bank’s visibility increased signifi-cantly, however, with the publication of articles byAndersen and Jordan (1968) and by Andersen andCarlson (1970). These articles provided two ofmonetarism’s most challenging attacks on policyorthodoxy and mainstream Keynesian macroeco-nomics. The former presented an econometricevaluation of the relative impacts of monetary andfiscal policy on economic activity. The latteroffered a small monetarist econometric modelthat the authors proposed as an alternative tool tosimulate alternative policy scenarios. Both papersconcluded that changes in the growth of nominalincome and inflation are linked closely to changesin the growth rates of monetary aggregates.Although such relationships had been demonstrat-ed over relatively long time-horizons, these two articles suggested that they hold over even the

short time horizons of concern to policymakers.The articles were instrumental in developing themonetarist policy rule as an alternative to the con-ventional interest rate and fiscal-policy orientedideas of the time.

The Andersen-Jordan Equation

Andersen-Jordan (1968) (hereafter, AJ) is anintellectual and analytical descendant ofFriedman-Meiselman (1963).20 Andersen andJordan, like Friedman and Meiselman, were inter-ested in isolating statistically the impact of moneyon nominal income. AJ went further, however.They provided a straightforward empirical test ofa related and critical policy issue—the relativeimpacts of monetary and fiscal impulses on nomi-nal income. Rather than building a complexeconometric model like those in vogue at thetime, AJ took a relatively simple approach toassessing alternative policies. They estimated asingle empirical relation—a “reduced-form”model—between income and different measuresof monetary and fiscal policy actions. Their equa-tion can be written as:

(1)

where Y represents nominal gross national prod-uct, M is the money stock (M1 or the monetarybase), and E is a measure of fiscal policy actions.The variables were measured as changes in theirlevels. In their estimation, AJ accounted for lags inthe effect of policy actions on economic activityusing a new econometric technique that con-strained the estimated parameters to lie along apredetermined polynomial. This was thought toprovide more precise estimation of the effects ofchanges in the policy variables.21

∆ ∆ ∆Y M Et ii

t i jj

t j t= ∑ + ∑ +=

−=

− β δ ε0

3

0

3

20 Jordan (1986) refers to it as a “sequel” to Friedman-Meiselman(1963), though it clearly is linked to earlier work by Karl Brunner.This is evident in the statement by AJ that their purpose is not to“test rival economic theories [i.e., Keynesian vs. Monetarist] of themechanism by which monetary and fiscal actions influence eco-nomic activity” (AJ, 1986, p. 29). A decade earlier Brunner wasexamining the logical structure of empirically testing betweenKeynesian models, in which money played a very minor role, andmonetary models, in which “money matters.” For example,Brunner and Balbach (1959) present a structure of models in whichthey test empirically the relative roles of money and fiscal policyactions.

21 See Batten and Thornton (1986) and the articles cited therein for adiscussion of this and other technical issues regarding the estima-tion of equation (1).


R E V I E W

Using quarterly data covering the period 1952to mid-1968, AJ estimated equation (1) to testthree hypotheses. By comparing the sizes of theestimated impacts of fiscal and monetary policyon GNP, AJ rejected the hypothesis that outputresponds more to fiscal policy actions than tochanges in the money stock. Comparison of thestatistical significance of the coefficient estimatesfor monetary and fiscal policy actions led AJ toreject the hypothesis that fiscal actions have amore “reliable” impact on GNP than monetaryactions. Finally, comparison of coefficient esti-mates on lagged monetary and fiscal policyactions, led AJ to reject the hypothesis that fiscalactions affect GNP faster than do monetary policyactions. They succinctly summed up their evi-dence: “The response of economic activity tomonetary actions compared with that of fiscalactions is (I) larger, (II) more predictable, and (III)faster” (p. 22).

Early Criticism of Andersen-Jordan(1968)

Andersen-Jordan (1968) was subject to imme-diate and critical analysis by economists insideand outside of the Federal Reserve System.Technical criticisms have been dealt with at lengthelsewhere.22 Of interest here is the fact that muchof the early debate over the usefulness and theconclusions of the article took place among Fedeconomists within the pages of System publica-tions. For example, the first published criticism ofthe AJ approach and findings was by DeLeeuwand Kalchbrenner (1969), both of whom were orhad been with the Board of Governors.23 Theircomment was published in the Federal ReserveBank of St. Louis Review, along with a response byAJ. DeLeeuw and Kalchbrenner (hereafter, DK)raised several technical issues, but focused on AJ’suse of the monetary base as the appropriatemeasure of monetary policy. DK argued that theFed controls neither the borrowed reserves ofmember banks nor the currency stock. Hence,they argued, the base is not statistically independ-ent of the model’s dependent variable—changesin GNP. DK reestimated the AJ equation, using thebase less borrowed reserves and currency as themonetary policy variable, and found that,although monetary policy appeared “to exert apowerful influence” on GNP, money was not asdominant as AJ’s results had suggested. DK notedalso that their results were more consistent with

the output from large-scale econometric models,suggesting that their results were more plausiblethan those of AJ.

Another early criticism of AJ appeared in theMonthly Review of the Federal Reserve Bank ofNew York (Davis, 1969). That study defended theview that monetary policy affects income throughinterest rates, not the money stock or monetarybase, with its author noting that the St. Louisequation “portrays a world in several respects[that is] sharply at variance with the expectationsof most of us” (p. 121). Like DK, Davis reestimatedthe AJ equation using different measures of mone-tary policy, as well as different polynomial lagspecifications and different sample periods.Davis’s analysis led him to conclude that “we can’taccept the St. Louis equations at face valuebecause neither money nor the total reserve basemay be sufficiently exogenous” (p. 126). The onlyrecourse, he suggested, is to build a structuralmodel (like the FRB-MIT model) and reject thereduced-form approach used by AJ. The onus formonetarists, he implied, was to put their ideasinto a structural model that details the transmis-sion mechanism of monetary policy.

The controversy generated by the appearanceof AJ marked an abrupt change in the “monetaryversus fiscal policy” debate. Despite criticism ofthe AJ study, the earlier view that business cycleevidence relating money and income was “theprovince of an obscure sect with headquarters inChicago” (Davis, 1969, p. 119) was changed bytheir results. Monetary aggregates now were con-sidered plausible alternatives to interest rates andfiscal policy as tools for short-run economic stabi-lization.

Darryl Francis and the Andersen-Jordan Results

While the technical analysis, criticism, andresponses of the AJ equation took place in bothSystem and academic publications, its policy im-plications were being disseminated in publicforums. Darryl Francis, the president of the St.Louis Bank, used the AJ results to promote the role

22 Reviews include Meyer and Rasche (1980), Batten and Thornton(1986), McCallum (1986), and Brunner (1986).

23 DeLeeuw, then a Senior Staff Member at the Urban Institute, hadbeen the Chief of the Special Studies Section, Division of Researchand Statistics at the Board of Governors and a principal in thedesign and development of the FRB-MIT model. Kalchbrenner wasan economist in the Special Studies Section.



of monetary aggregates in setting stabilization pol-icy. Francis (1968, p. 8) rejected the use of fiscalactions as a tool for stabilization, arguing that“monetary actions are a major determinant ofshort-run movements in total spending.” He alsorejected the common view that interest rates andbank credit reflect accurately the stance of mone-tary policy. Francis (1968) argued that “move-ments in interest rates should be viewed no differ-ently than movements in commodity prices” (p. 8). Instead, Francis pushed for the “primaryand consistent use of monetary aggregates” in set-ting policy, noting that “all of these aggregates canbe rather precisely controlled by monetaryauthorities” (p. 8). This approach would serve thedual purpose of holding the authorities account-able for their actions and instituting “scientificmethodology and modern quantitative analysis”to monetary policy (p. 7).

At an FOMC meeting on February 4, 1969,Francis reviewed how the Committee had beenmisled by the behavior of interest rates and bankcredit:

For about four years the Committeehad been led into unintended inflationarymonetary expansion while followinginterest rate, net [free] reserves, and bankcredit objectives If the Committeemeant business now, it should try someother guides. Not only could the oldguides lead to further inflation as long asdemands for credit continued to rise, butwhen and if contrary trends set in theycould lead to an undue contraction oftotal spending.

Francis also made clear his preferred policyguides: “He urged the Committee to give some evi-dence that it was exercising restraint by limitingthe growth of bank reserves, the monetary base,and the narrow measure of money supply” (FOMCMinutes, February 4, 1969, p. 47).

Francis believed strongly that inflation wasthe consequence of excessive monetary growth,and that the Fed had erred in pursuing policiesthat resulted in accelerating growth of the mone-tary aggregates. In essence, Francis attacked thedominant view that policy should be aimed at sta-bilizing short-run variation in economic activity,as reflected in the unemployment rate, at theexpense of higher inflation. At an FOMC meeting

on May 11, 1971, Francis reviewed the course ofmonetary policy and inflation over the previous20 years:

During the ten-year period ending in late1962, money grew at an average annualrate of 1.5 per cent With the econom-ic sluggishness of the early 1960’s monetary stimulation was increased, andmoney rose at a 3.5 per cent averageannual rate from late 1962 to the end of1966 [T]hat rate of monetary expan-sion resulted in a gradual increase ininflation to a 3 per cent rate. Followingthe credit crunch of 1966 money growthwas again accelerated, producing a 6.3per cent average annual rate from early1967 to the present [A] 6 per centtrend rate of monetary expansion implieda sustained 4 per cent rate of inflation. Ineach case the rate of growth in moneywas accelerated in order to overcomeweakness in the economy. Despite thoseprogressively more stimulative monetaryactions, the rate of unemployment hadaveraged about the same whether thetrend growth of money was 6 per cent,3.5 per cent, or 1.5 per cent. The trendgrowth had had its chief impact on prices,whereas fluctuations around the trendhad had the greatest impact on produc-tion and employment. (FOMC Minutes,May 11, 1971, pp. 57-58)

Francis’s perspective reflected his monetaristoutlook: He argued that inflation is primarilydetermined by the rate of growth of the moneystock and that, in the long run, real output growthand the unemployment rate are unaffected bymonetary policy. In other words, the long runPhillips curve is vertical.24 Francis also arguedthat while monetary policy has no effect on realgrowth or employment in the long run, fluctua-tions in monetary growth could have substantialeffects on these variables in the short-run. Heused the Andersen and Jordan (1968) results, andthose of other Bank economists, to support hisclaim.

24 The distinction between short- and long-run Phillips curves was for-malized by Friedman (1968) and Phelps (1967).


R E V I E W

Although some members of the Committeeshared Francis’s views, the chairman and a majori-ty of others did not. In one of the most frank dec-larations of the opposing view, Chairman WilliamMcChesney Martin stated at the October 7, 1969,FOMC meeting that he “did not accept the mone-tarist’s position regarding the critical importanceof the specific rate of change in the money supply.In particular, he did not agree that the conse-quences of deviating significantly from some pre-ferred rate for a period of time would be as disas-trous as the monetarists believed” (1969, p. 1100).Board economists Gramley and Chase (1965, p.1403) went even farther, arguing, “there is littledoubt that such a simple rule [based on changesin the money stock] for appraisal of central bankoperations is no longer appropriate.”25

Francis cited Federal Reserve Bank of St. Louisresearch often and at least once entered St. Louisstaff forecasts of real output and inflation underalternative money stock growth rates into the for-mal record of FOMC deliberations. His first record-ed reference came at the December 17, 1968,meeting when he discussed “a recent study done atthe St. Louis Reserve Bank [which] indicated thatwith the existing stance of fiscal policy, if moneycontinued to grow at a 6 per cent annual ratethroughout the coming year, gross national prod-uct would rise at an excessive 8 per cent annualrate” (FOMC Minutes, December 17, 1968, p. 54).

Francis’s discussion illustrates that St. LouisFed officials (and monetarists in general) hadbegun to actively engage the prevailing wisdom onits own short-run grounds. The AJ results, ineffect, provided a platform by which monetaristpolicy prescriptions, oriented to the behavior ofthe monetary aggregates, could be discussed interms of short-run stabilization issues. With oneeye cocked to the longer-term inflationary effectsof policy, something that the conventional viewdid not provide, monetarists could also discuss theshort-run effects of monetary policy.

Francis’s use in FOMC meetings of researchconducted by St. Louis Fed economists illustrateshow the Fed’s decentralized organizational struc-ture can affect policy deliberations. The participa-tion of Federal Reserve Bank presidents in mone-tary policymaking provides an outlet for alterna-tive perspectives to be heard, including direct criti-cism of System policy. In addition to bringing theresearch findings of his staff to FOMC delibera-

tions, Francis also promoted St. Louis Fed researchin his numerous public appearances. Citing “forth-coming articles,” Francis often talked about lags inthe impact of monetary policy and how theymade attempts to “fine tune” a balance betweeninflation and unemployment difficult, if notimpossible.26 Even so, the behavior of the econo-my in the late 1960s gave credence to the claimthat nominal spending responded, albeit with alag, more to changes in money supply growththan to fiscal policy. The “mini-recession” of 1966and the failure of tax increases in 1968 to halt theupward march of inflation seemed to support theefficacy of monetary over fiscal policy. Whilesome of his FOMC colleagues had urged tighterfiscal policy to stem inflation in 1968, and sup-ported the temporary tax increase that had beenenacted, Francis contended that the tax increasewas unlikely to have a significant effect on eco-nomic activity.27 As Francis predicted, the tempo-rary fiscal measures adopted in 1968 had minimalimpact on economic activity.

THE ST. LOUIS MODEL: MONETARISMFOR THE SHORT-RUN

The predictive success of the St. Louis (AJ)equation and the apparent failure of fiscal policyto stem the inflation of the late 1960s gave mone-tarists credibility in policy discussions. The behav-ior of the money stock began to get more consid-

25 Gramley recalls the policy debates this way: “ if the FederalReserve had appreciated how serious the inflationary problem wasgoing to become, they would have paid more attention to thegrowth of the monetary aggregates and relied less on money mar-ket conditions if you weren’t worried too much about long-runinflation you were inclined therefore not to pay sufficient attentionto what was happening to those aggregates” (quoted in Mayer,1995, pp. 7-8).

26 For example, at an FOMC meeting on March 9, 1971, he noted that“In the past the System had, on occasion, persisted in a policycourse too long. Knowledge of current developments in the econo-my was available only with a delay, and the effects of monetaryactions on spending, production, prices and employment contin-ued for months” (FOMC Minutes, March 9, 1971, p. 66).

27 As Francis summarized his position before the FOMC, “Over thepast year the System had aggressively advocated fiscal restraint as anecessity to rational stabilization policy. Yet now that such restraintappeared likely, there seemed to be growing fear of its destabilizingimpact. [He] did not share those views Those fiscal measureswere generally expected to be temporary, and thus much of the taxburden on consumers would probably come from reduced savingand much of the burden on corporations would probably comefrom increased borrowing The tax measure, because of its tem-porary nature, might actually cause some acceleration of invest-ment spending” (FOMC Minutes, June 18, 1968, pp. 87-88).



eration in the setting of stabilization policy.28 Butsome observers, even Homer Jones, the St. LouisBank’s research director when the St. Louis modelwas developed, sounded a note of caution:

Our own econometric studies at St. Louishave long indicated strong, roughly pre-dictable, relations between monetary action,intentional or unintentional, and the courseof the economy does this mean we canexpect to engage usefully in active monetarymanagement in the future? I concludethat we cannot in the near future engage intel-ligently in short-run manipulative monetarymanagement. (1970, p.15, emphasis added)

Darryl Francis also warned against using mone-tary policy to fine-tune economic activity. Thesuccess of the AJ equation and the building empir-ical evidence in support of monetarist views, how-ever, led to a greater focus on the short-run. In hisretrospective of the equation’s development anduse, Jordan (1986, p. 8) notes that “The [AJ] arti-cle’s impact on economic policymaking wouldhave been more favorable had it not led to anincreased reliance on monetary over fiscal policy,but had it instead contributed to a general de-emphasis of fine-tuning attempts by policymak-ers.”

The increasingly short-run emphasis undoubt-edly reflected, in part, the natural focus of policy-making at the central bank. Dewey Daane, aFederal Reserve governor, noted that the FOMCwas “always concentrating on what’s the immedi-ate problem over the next four to six weeks andnot really thinking in terms of long-run forecastsand inflation” (quoted in Mayer, 1995, p. 16).Governor Andrew Brimmer’s recollection corrobo-rates this view, noting that there was “clearly ashort-term horizon. [Chairman] Martin put a lot ofemphasis on the long run, but that was unusual”(quoted in Mayer, 1995, p. 4).

In April 1970, the Federal Reserve Bank of St.Louis published “A Monetarist Model forEconomic Stabilization,” by Andersen and Carlson(hereafter, AC). The AC, or “St. Louis” model as ithas become known, reflected the latest stage inthe development of an empirical model of mone-tarist propositions and expanded the on-goingdebate over the role of money in determiningaggregate spending and inflation in an importantway. Unlike the large-scale macroeconometric

models being developed elsewhere, the St. Louismodel built upon previous research at the Bank inwhich the money stock is the central focus of sta-bilization policy.

The original St. Louis model consisted of eightequations, only four of which were estimated: thetotal spending equation—the AJ equation, a priceequation, an equation for the long-term interestrate, and an unemployment equation. Theremaining equations are definitions.29 The interestrate equation, based on earlier work by Yohe andKarnosky (1969), reflected the view that interestrates are determined by past inflation and pastchanges in money growth. The unemploymentequation was essentially that developed by Okun(1962). The price equation rejected the typicalwage-price markup approach popular at the time.Instead, AC specified the change in the price levelas a function of demand pressures and anticipatedprice changes.

The St. Louis model is “monetarist” in that themoney stock is treated as exogenous and its effecton total spending is central to the workings of themodel.30 As AC state:

The change in total spending is combinedwith potential (full employment) output toprovide a measure of demand pressure.Anticipated price change, which dependson past price changes, is combined withdemand pressure to determine the changein the price level. The total spending iden-tity enables the change in output to bedetermined, given the change in totalspending and the change in prices. (p.10)

28 In reviewing the discussion of monetary policy at a recent confer-ence, Friedman wrote to Homer Jones in July 1969 that “I, too,have been very much impressed with the evidence that a new dayhas dawned I almost fell over when he [Board Governor DeweyDaane] started talking about the importance of paying attention tomonetary aggregates” (Friedman Papers, Box 28, Folder 36).

29 The original model is summarized in the Appendix. Carlson (1986)provides a comparison of the original version and the then “cur-rent” version which reflects modifications over the interveningyears.

30 AC determine real output as a residual; that is, output is deter-mined as the difference between total spending and the price level.As they note, “This method of determining the change in totalspending and its division between output change and price changediffers from most econometric models. A standard practice ineconometric model building is to determine output and prices sep-arately, then combine them to determine total spending” (p.10).


R E V I E W

The structure of the model meant that for agiven change in the money supply or governmentexpenditures, one could solve for changes in totalspending, prices, real output, the unemploymentrate, and interest rates. The model was simple incomparison with the complex structural modelsused by the Board staff and elsewhere. For exam-ple, whereas the Wharton model, a representativeKeynesian structural model, had 43 exogenousvariables, the St. Louis model had just three. Themodel also omitted details about specific sectorsfor the simplicity of determining the impact of achange in money growth on the economy broadly.This development fit nicely with the view of manymonetarists that “the Federal Reserve should beconcerned with the aggregate effects of policy,and should leave the allocative details to the oper-ation of the market” (Francis 1973, p. 9).31 AsCarlson (1986, p. 18) recollects the developmentof the model,

we wanted a model that was small enoughthat the interrelationships among the varablescould be understood easily We were notconcerned about respecifying behavioralequations [and] we wanted to captureempirical relationships between a relativelyfew key macroeconomic variables that wereimplicitly grounded in economic theory.

AC state explicitly that their statistical analysisis used “to estimate the response of output andprices to monetary and fiscal actions, not to test ahypothesized structure” (pp. 10-11).32 The originalestimates appeared to support the monetarist viewof the world: an increase in the money supplyleads first to an immediate increase in nominalspending and real output, and only after pricesadjust to the higher demand pressure does theprice level rise to stifle the increase in real output.

Using the Model for Short-RunAnalysis

Andersen and Carlson used the St. Louismodel to simulate nominal spending, real output,inflation, and the unemployment rate for differenthypothesized growth paths for money. They alsocompared their monetarist model’s forecastingability with that of the Wharton model during1963-64, a period that included a major fiscalaction—the tax cut of 1964. How would the St.

Louis model, in which fiscal policy plays a minorrole, fare in comparison with the Wharton modelin which fiscal policy has a much larger role thanmoney? The St. Louis model’s simulations werebetter (i.e., produced lower root-mean-squarederrors) than the Wharton model for nominal GNPand the unemployment rate, about the same forreal GNP, and worse for the price level. The upshotwas that this small, monetarist-oriented modelcould prove as valuable to policymakers as thelarge-scale Keynesian models then in use.Importantly, it seemed to demonstrate the useful-ness of a small monetarist model for currentanalysis. As AC state, “The purpose of the follow-ing statistical section is to estimate the response ofoutput and prices to monetary and fiscal actions,not to test a hypothesized structure. The focus ison the response in the short run—periods of two orthree years—but the long-run properties also areexamined” (1970, pp. 10-11, emphasis added).

The success of the St. Louis model was impor-tant to monetarism’s growing impact on policydiscussions. It also appeared soon after Friedman(1968) and Phelps (1967) provided theoreticalmodels in which the popular Phillips curve trade-off between inflation and unemployment (and,hence, real output growth) was shown to be transi-tory. The Phillips curve, a version of whichappears in the St. Louis model, was a critical com-ponent of most Keynesian macro-models of thetime. The results of AC provided an empiricaldemonstration that although expansionary mone-tary policy might produce a short-run increase inreal output growth and a dip in the unemploy-ment rate, these effects would vanish over time asinflation increased and unemployment and outputgrowth returned to their “natural” or trend rates.

The St. Louis model enabled monetarists toproduce short-run forecasts of alternative policyscenarios, thereby putting them on similar footing

31 A recurrent theme in discussions about the role of policy was therecognition that policy actions sometimes affected certain indus-tries—most notably housing—more than others. Such attentionwas disruptive to the working of market forces, Francis believed. Henoted that “Regulation of interest rates paid by commercial banksand thrift institutions unduly disrupts the allocation function ofmarkets. Furthermore, excessive concern for the well-being ofthese institutions and the housing industry has caused monetaryauthorities to expand the money stock at a rapid rate during muchof the current inflationary period” (1968, p. 9).

32 Francis (1973, p. 8) makes the point that “The bewildering strug-gles that occur between model builders over specification errors,structural versus reduced-form models, recursive versus non-recur-sive systems, etc., are meaningless to most policymakers.”



with other mainstream economists, both in andoutside of the Federal Reserve System. Dewald(1988, p. 6) contends that with the development ofthe St. Louis model, “monetarism was widely inter-preted as providing an alternative to short-runKeynesian model forecasts.” The St. Louis model,though grounded in the long-run conditions of theQuantity Theory, increasingly was used to counterthe short-run policy prescriptions coming fromthe larger structural models in use at the Board.The St. Louis model, estimated using quarterlyobservations and with the money stock—notinterest rates—as the policy instrument, led manyobservers to conclude that the money stock couldbe an effective tool for economic stabilization.

The shifting emphasis at the St. Louis Banktoward short-run policy analysis can be found incomments of participants in the model’s develop-ment. Carlson (1972, p. 25) warned against usingthe model for anything but interpreting the “gen-eral time path” of important macroeconomic vari-ables. Even so, analysis of the short-run impact ofalternative policies is precisely what the modelcame to be used for. His own admonition aside,Carlson (1972, p. 20) noted that the model “pres-ents a set of simulations using alternative steadygrowth rates which can aid in assessing the eco-nomic impact over several quarters of differenttrend growth rates of money” (emphasis added).In keeping with this view, his analysis of themodel’s performance was based on a six-quarterhorizon, hardly the long run used by early mone-tarist studies. But such a use for the modelappeared justified by the empirical results: “themodel succeeded in roughing out the average timepaths of total spending, real product, prices,unemployment, and interest rates during the peri-od from late 1969 to mid-1971” (Carlson, 1972, p.26, emphasis added).

The model’s success as a forecast tool gavesupport to monetarist calls for a policy aimed atstable money growth. In reviewing the debate overstabilization policy, Andersen (1973, p. 3) summa-rized the model’s success at forecasting real outputand inflation, stating that “The key proposition isthat changes in money dominate other short-runinfluences on output and other long-run influenceson the price level and nominal aggregate demand”(emphasis added). The St. Louis model suggestedthat stable money growth would lessen anymonetary-induced instability in the real economywhile promoting price stability in the long run.

Research from St. Louis continued to provide along-term, inflation-oriented perspective on mon-etary policy actions, reflecting rising inflation ofthe early 1970s. At the same time, the Review con-tained numerous studies, often authored byAndersen or Keran, of the short-run response ofthe economy to changes in the growth of themonetary aggregates. The allure of short-runanalysis perhaps is best illustrated by Carlson’s(1975) estimation of the St. Louis equation usingmonthly data. Replacing nominal GNP with per-sonal income, Carlson found that the lag fromchanges in the growth of money to nominalincome was completed in about one year, similarto that found by AJ, though slower than reportedin other studies.33 The implication of this findingwas clear: Carlson (1975, p. 17) suggests that the“Use of monthly data thus appears to carry thepotential for evaluating the thrust of monetaryand fiscal actions before quarterly data on GNPbecome available.”

In the late 1960s, public interest in mone-tarism rose as inflation continued to increase.Monetarists were called upon by the incomingNixon administration for advice. Milton Friedmanwrote a regular column for Newsweek alongsideone by Paul Samuelson, who had been a leadingadvisor to the previous two administrations and aleading architect of the so-called New Economics.Increased attention, however, brought more stri-dent criticism. By 1972 there already were claimsthat monetarism had “failed.” For example, in hisNewsweek column of August 2, 1971, Samuelsonobjected to the monetarist claim that rapid moneygrowth in 1971 would subsequently lead to fasternominal GNP growth. He suggested that “the fore-casting ability of monetarism is selling at a hugediscount on the markets of informed opinion”(Samuelson 1971, p. 70) and that the “pseudoposi-tivism which prevails among monetarists. . . [is]still another reason why the peculiar tenets ofmonetarism have to be rejected” (quoted inFrancis, 1972, p. 32).

Members of the Board of Governors also criti-cized the policy advice of monetarists. For exam-ple, Andrew Brimmer, echoing arguments madeduring the previous two decades, rejected any pol-icy based on control of the monetary aggregates:“I am convinced that it would be a disastrous

33 Another example of attempts to model the short-run effects ofmoney on the economy is Laffer and Ransom (1971). Unlike the St.Louis results, Laffer and Ransom report that monetary actions leadto an immediate and permanent effect on the level of GNP.


R E V I E W

error for the Federal Reserve to try to conductmonetary policy on the basis of a few simple rulesgoverning the rate of expansion of the moneysupply” (1972, p. 351).

Francis (1972, p. 32) considered such attacks“strident,” “doctrinaire,” and “no more precisethan in the past.” He answered these criticisms bypointing to the St. Louis model’s ability to forecasteconomic activity over the short-run. He com-pared income, inflation, and unemployment pre-dictions for 1969, 1970, and 1971 derived fromthe St. Louis model to those of the consensusLivingston forecasts. The overall forecasting abilityof the St. Louis model compared favorably withthe consensus forecasts. Although Francis main-tained that policy should take a longer-term viewto be effective, he focused on a few years of fore-casting results to justify applying the model toshorter time horizons.

Refining the Model

As the 1970s progressed, neither Keynesianmodels based on the Phillips curve and interestrates, nor simple monetarist models based ongrowth of the money stock, successfully forecastthe rapid inflation and higher unemployment thatactually occurred. Monetarists faced the task ofexplaining why inflation had increased so dramat-ically without a similar-sized increase in moneystock growth. Monetarists, including St. Louis Fedofficials, responded that their critics had confusedchanges in the aggregate price level, which arecaused by monetary policy, with changes in rela-tive prices brought on by special factors. Francispresented the Bank’s position in a series ofspeeches in 1974. He pointed out that the increasein inflation was due largely to an increase inmoney growth over the preceding few years. Anyinflation over and above the underlying monetarygrowth rate was caused by the removal of wageand price controls and the increase in oil prices bythe OPEC nations.34

St. Louis Fed economists soon integrated suchspecial factors in their studies. Karnosky (1976)demonstrated that money growth continued toexplain longer-term movements of inflation oncethe oil price shock effects were accounted for.Rasche and Tatom (1977) extended this idea intheir examination of the effects of supply shockson the economy and how they could distort thestatistical relationship between money andincome in the short-run. Although aspects ofthese works were criticized, they suggested that

the money supply remained useful for stabiliza-tion purposes.

The Breakdown of the Monetarist Rule

Kane (1990) observes that as the rate of inflationcontinued to increase over the 1970s, the growingweight of evidence supporting the monetarist posi-tion pushed the FOMC to incorporate money stockgrowth into their policy deliberations and evalua-tions. This was quite a change from their position ofa decade earlier when “they treated monetarism asan eccentric and quasi-religious belief system thatno responsible macroeconomist or public officialcould possibly take seriously”(p. 292).

Monetarism as a policy approach, however,had a relatively short stay in the limelight. InOctober 1979, the Federal Reserve adopted newoperating procedures that it claimed wouldenhance its control of the money stock. Highlyrestrictive policies also were enacted to reduceinflation, which had reached double-digit levels.Although inflation eventually declined significant-ly, the more immediate effect was to send theeconomy into the deepest recession of the post-war period. Critics associated the policy withmonetarism, referring to the policy as the Fed’s“monetarist experiment,” and this perception con-tributed to the widespread discounting of mone-tarism as a viable policy option. Monetaristsprotested that the Fed had not, in fact, adoptedtheir preferred policy of slow and steady moneygrowth. Rather, they noted, the variability ofmoney growth actually increased after 1979 andgave rise to increased fluctuations in real econom-ic activity without any appreciable short-run effecton inflation.35

34 Francis argued similarly in FOMC deliberations. At a meeting onJanuary 22, 1974, for example, Francis contended that “the actualand prospective slowdown in economic activity resulted whollyfrom capacity, supply, and price-distorting constraints and not froma weakening in demand. Therefore, to ease [monetary] policy andallow a faster rate of monetary growth would be to increase infla-tionary pressures without expanding real output or reducing unem-ployment” (FOMC Minutes, January 22, 1974, p. 102).

35 Batten and Stone (1983) provide an overview of the issues and evi-dence in support of the monetarist position. For contrasting assess-ments of this episode, see B. Friedman (1984) and M. Friedman(1984). Charles Schultz, Chairman of the Council of EconomicAdvisors from 1977-81, considered the monetarist experiment inthis light: “What monetarism really is for the Fed (and I’m morallycertain this is what Volcker thinks, too) a political cover. They’renot monetarists, but it allowed them to do what they could neverhave done They could never have done what needed to bedone if it looked as if they were the ones raising interest rates,when they were targeting interest rates, per se. But with fixedmonetary targets they could just say, ‘Who, us?’” (quoted inHargrove and Morely, 1984, p. 486).

Regardless of whether the Fed had in factadopted a monetarist policy in 1979, large, volatilemovements in velocity began to erode profession-al support for monetarist policies. Inflation andattendant increases in market interest ratesencouraged financial innovations that gave con-sumers more options for holding liquid balances.This, in turn, spurred regulatory changes that pro-duced sharp changes in the relative demands forliquid financial assets. At the same time, the Fed’sabrupt tightening brought a sharp decline in therate of inflation, which probably contributed to areversal of the upward trend in velocity that hadcharacterized the previous three decades.36 Assetdemand shifts and uncertainty about the Fed’scommitment to a new path for inflation probablyexplain why the velocity of traditional measuresof the money stock, especially the narrow meas-ures, such as M1, favored by officials of theFederal Reserve Bank of St. Louis, began to moveerratically. Stable velocity was crucial for thereduced-form models used at St. Louis to hold. Ifthe path of velocity changed unpredictably, thenthe predictions of the St. Louis equation andmodel could be unreliable.37 As velocity began todeviate seriously and continuously from its previ-ously stable path, monetarist policy prescriptionsbecame increasingly suspect.

As shown in Figure 1, the velocity of M1—themonetary aggregate used in the St. Louis model—maintained a fairly steady upward climb duringthe 1960s and 1970s. During the 1980s, however,M1 velocity deviated considerably from its previ-ous path. M1 velocity appeared to become “unsta-ble,” thus justifying critics’ opposition to monetarytargeting.38

As M1 velocity deviated further from its his-toric trend, St. Louis Fed researchers devotedincreasing effort to understanding velocity andmodifying their forecasting model. Meyer andVarvares (1981), for example, made two modifica-tions: one modification was to model the rate ofinflation as a direct outcome of money growthand oil price shocks; the other modification incor-porated a new Phillips curve relation. Other St.Louis Fed studies researched the lag betweenmoney and prices (Carlson, 1980), the effects offiscal policy (Hafer, 1982), and the longer-runconsequences of policy (Carlson and Hein, 1983).These studies all concluded that the reduced-formapproach continued to be a reasonable way ofmodeling the impact of monetary policy on the

economy, and early success at improving modelforecasts encouraged further research along thesame lines. But, because the shift in velocity hadoccurred only recently, and therefore affectedonly a limited number of observations used in theestimation of the model, these papers largelyignored the potential effects of the shift. Indeed,even Carlson’s (1986) version of the model, onethat made minor revisions that recognized the

1960 1970 1980 1990 20003

4

5

6

7

8

9

10

Pre-1980 Trend

M1 Velocity

Velocity of Money (M1)Quarterly Data, 1960 Through 1999



36 Gavin and Dewald (1989) show that disinflation leads to a reduc-tion in velocity once the public comes to expect that a new trendgrowth rate of money and the price level have been established.

37 This potential problem was already known. For example, Rasche(1972) noted that the successful forecasting ability of the St. Louisequation was, in part, based on the small interest elasticity of themoney demand function. As he states (p. 31): “if the short-runinterest elasticity of the money demand function is very small,then an estimated equation omitting this term [the interest rate]would most likely produce a credible forecasting record” (emphasisin original). In some sense, the constant term in equation (1) is theempirical representation of velocity. Financial innovations alteredthe underlying short-run interest elasticity of money demand andadversely affected the equation’s forecasting ability. Because nochanges along these lines were made to the model, its forecastsbegan to stray. More recent investigations of the behavior of veloci-ty include Stone and Thornton (1987), Rasche (1993), Hoffman andRasche (1996), and Laurent (1999).

38 Monetarists had long divided over whether a narrow aggregate,such as the monetary base or M1, or a broader aggregate, such asM2 or M3, was a preferable target for monetary policy. St. LouisFed officials advocated M1, while Milton Friedman favored M2.When M1 velocity began to deviate from its trend in the early1980s, M2 velocity remained relatively stable. By the early 1990s,however, M2 velocity also had deviated substantially from its long-run trend.

Figure 1


R E V I E W

impact of energy prices, wage and price controls,and several other technical changes, continued tofit the data reasonably well. Thus, for a time, mon-etarists continued to argue that nominal incomegrowth reflected changes in money growth overshort periods and that inflation reflected moneygrowth over longer periods, even during thisturbulent time. As the 1980s progressed, however,continued instability of velocity caused all but themost diehard supporters to abandon short-runmonetary aggregate targets.

LESSONS FROM THE ST. LOUISEXPERIENCE

The development and decline of the St. Louismonetarist model as a guide to short-run stabiliza-tion policy is not unlike the evolution of stabiliza-tion policies designed to exploit a tradeoffbetween inflation and unemployment. Policiesbased on the Phillips curve arose from an appar-ently robust empirical relationship between infla-tion and unemployment observed in macroeco-nomic data. As discussed in King and Watson(1994), the negative correlation between the vari-ables in the short-run suggested the presence of along-run structural relationship that could beexploited for policy purposes. Attempts to manip-ulate interest rates to increase the growth of realoutput and employment above potential, however,gave rise to what has been called “the great infla-tion” in the United States, a period encompassingthe 1960s and 1970s.39

Over the 1970s, as inflation and unemploy-ment rose simultaneously, monetarists gained astronger voice in monetary policy debates as thelong-run relations among money, prices, andnominal income seemed to hold even in theshort-run. Due in part to work at the FederalReserve Bank of St. Louis, Woodford (forthcoming,p. 18) suggests that “the monetarist viewpoint hadbecome the new orthodoxy by the mid-1970s.”

The St. Louis Bank and its officials had specialprominence because they provided an avenue formonetarist research and views to potentiallyinfluence policy deliberations. To be influential,however, monetarists had to offer a viable policyfor the short-run—a policy that could be discussedand voted on at meetings some six weeks apart.The success of the St. Louis model at forecastingoutput, nominal income, and prices over suchshort horizons during the 1970s convinced St.

Louis Fed officials that they could credibly advocatea monetarist stabilization policy for the short-run.

The Federal Reserve took a step toward mone-tary aggregate targeting in October 1979, whennew procedures were implemented to better con-trol the money stock with the goal of reducinginflation. The targeting of monetary aggregates waslargely abandoned in 1982, however, when velocity,particularly of M1—the narrow aggregate favoredby St. Louis Fed officials—proved too erratic.

Deregulation, other institutional changes, anduncertainty about the Fed’s commitment to disin-flation probably explain much of the unstablebehavior of velocity in the 1980s. The DepositoryInstitution Deregulation and Monetary Control Actof 1980 (DIDMCA) instituted a six-year processending the prohibition of interest payments ontransaction accounts at commercial banks andderegulating rates on other accounts. Thesechanges, and various financial innovations, werefollowed by volatile flows between classes offinancial assets that altered the empirical relation-ships between national income and monetaryaggregates. Monetary aggregates quickly lost favoras short-run policy targets when movements invelocity became difficult to explain or predict. Inessence, changes in the structure of the economyaltered the short-run relationships between tradi-tional monetary aggregates and policy objectives.Monetarist models, including the St. Louis model,were not equipped to handle such changes, andtheir forecasting performance suffered as a result.

Typical macroeconomic models of the 1970s,including the St. Louis model, also were notequipped to deal with the so-called “LucasCritique.” Lucas (1976) demonstrated that coeffi-cient estimates of typical forecasting models areunlikely to be stable across policy regimes. Asindividuals learn about and modify their behaviorin response to a change in regime, empirical rela-tionships among macroeconomic variables maychange. Consequently, economic projectionsbased on estimation of a model over one regimemay not be valid for a different regime. For exam-ple, the close short-run correlations amongmoney, output, and prices observed during the1960s and 1970s under a regime characterized byinterest rate targeting would not necessarily havebeen so close in a regime of monetary aggregate

39 For alternative views of this period, see DeLong (1997), Mayer(1999), Sargent (1999), Taylor (1999), or Wheelock (1998).



targeting.40

The search for robust empirical relationshipsacross different regimes has shown that money,nominal income, and inflation remain closelylinked in the long run (Dewald, 1998; Dwyer andHafer, 1999; Lucas, 1996; Rolnick and Weber,1997). There also is some evidence that the link-ages between money and economic activity arerobust even at relatively short-run frequencies.41

Monetary aggregates again may prove useful foreconomic forecasting or as guides for conductingmonetary policy. Experience has shown, however,that empirical relationships between policy vari-ables and goals can change, sometimes unpre-dictably. The experience with the monetarist ruleas developed at the St. Louis Fed councils againstoverconfidence in our ability to identify infalliblerules for conducting short-run stabilization policy.

REFERENCES

Andersen, Leonall C. “Federal Reserve DefensiveOperations and Short-Run Control of the Money Stock.”Journal of Political Economy, March/April 1968, p. 275-88.

__________. “The State of the Monetarist Debate.”Federal Reserve Bank of St. Louis Review, September1973, 55(9), pp. 2-8.

__________ and Jordan, Jerry L. “Monetary andFiscal Actions: A Test of Their Relative Importance inEconomic Stabilization [reprint].” Federal Reserve Bankof St. Louis Review, October 1986, 68(8), pp. 29-44.

__________ and __________. “Monetary andFiscal Actions: A Test of Their Relative Importance inEconomic Stabilization.” Federal Reserve Bank of St.Louis Review, November 1968, 50(11), pp. 11-24.

__________ and Carlson, Keith M. “A MonetaristModel for Economic Stabilization.” Federal Reserve Bankof St. Louis Review, April 1970, 52(4), pp. 7-25.

Ando, Albert; Brown, E. Cary; Solow, Robert M. andKareken, John. “Lags in Fiscal and Monetary Policy,” inStabilization Policies: Commission on Money and Credit.Englewood Cliffs, NJ: Prentice Hall, 1963, pp. 1-164.

Ando, Albert and Modigliani, Franco. “The RelativeStability of Monetary Velocity and the InvestmentMultiplier.” American Economic Review, September 1965,pp. 693-728.

Angell, James W. “Appropriate Monetary Policies andOperations in the United States Today.” Review ofEconomics and Statistics, August 1960, pp. 247-52.

Barro, Robert J. “Inflation and Growth.” Federal ReserveBank of St. Louis Review, May/June 1996, 78(3), pp. 153-69.

Batten, Dallas S. and Stone, Courtenay C. “Are Monetaristsan Endangered Species?” Federal Reserve Bank of St.Louis Review, May 1983, 65(5), pp. 5-16.

__________ and Thornton, Daniel L. “The Monetary-FiscalPolicy Debate and the Andersen-Jordan Equation.”Federal Reserve Bank of St. Louis Review, October 1986,68(8), pp. 9-17.

Brimmer, Andrew F. “The Political Economy of Money:Evolution and Impact of Monetarism in the FederalReserve System.” American Economic Review, May 1972,pp. 344-52.

Bronfenbrenner, Martin and Holzman, Franklin D. “Surveyof Inflation Theory.” American Economic Review,September 1963, 52, pp. 593-661.

Brunner, Karl. “Fiscal Policy in Macro Theory: A Surveyand Evaluation,” in R.W. Hafer, ed. The Monetary VersusFiscal Policy Debate: Lessons from Two Decades. Totowa,NJ: Rowman and Allanheld, 1986, pp. 33-116.

__________ and Balbach, Anatol B. “An Evaluation of Two Types of Monetary Theories.” Proceedings of theThirty-Fourth Annual Conference of the WesternEconomic Association, 1959, pp. 78-84.

Burger, Albert E. The Money Supply Process. Belmont, CA:Wadsworth Publishing Co., 1971.

40 The monetarist experience, in which proponents sought to replaceone regime with another, contrasts with Taylor’s (1993) proposal forthe Fed to continue the policy rule that he found described recentpolicy well. Researchers have, however, questioned how well theTaylor rule actually represents Fed policy. Orphanides (1997), forexample, shows that a Taylor rule derived from real-time data pro-duces a federal funds rate path over the past decade that is quitedifferent from that found using revised data. Kozicki (1999) showsthat Taylor rule predictions are not robust to differences in meas-ures of inflation, the output gap, the real interest rate, and to differ-ent parameterizations. Finally, Hetzel (2000) argues that use of theTaylor rule within an activist model of monetary policy will misleadpolicymakers as they attempt to control inflation.

41 Dwyer and Hafer (1988, 1999) find that for a large cross-section ofcountries the correlation between money growth and nominalincome is close to unity even over short time horizons. Similarly,based on estimates of a small VAR model with quarterly U.S. data,Dwyer (1998) concludes that the link between money and nominalincome may be much closer than some have claimed. Finally, usinga recursive estimate of equilibrium M2 velocity, Orphanides andPorter (2000) find that monetary aggregate-based forecasts of infla-tion for the 1990s are quite accurate.


R E V I E W

Carlson, Keith M. “Projecting with the St. Louis Model: AProgress Report.” Federal Reserve Bank of St. LouisReview, February 1972, 54(2), pp. 20-26.

__________. “The St. Louis Equation and Monthly Data.”Federal Reserve Bank of St. Louis Review, January 1975,57(1), pp. 14-17.

__________. “Money, Inflation, and EconomicGrowth: Some Updated Reduced Form Results and Their Implications.” Federal Reserve Bank of St. LouisReview, April 1980, 62(4), pp. 13-19.

__________. “A Monetarist Model for EconomicStabilization: Review and Update.” Federal Reserve Bankof St. Louis Review, October 1986, 68(8), pp. 18-28.

__________ and Hein, Scott E. “Four Econometric Modelsand Monetary Policy: The Longer-Run View.” FederalReserve Bank of St. Louis Review, January 1983, 65(1),pp. 13-24.

Davis, Richard G. “The Role of the Money Supply inBusiness Cycles.” Federal Reserve Bank of New YorkMonthly Review, April 1968, pp. 63-73.

__________. “How Much Does Money Matter? A Look at Some Recent Evidence.” Federal Reserve Bankof New York Monthly Review, June 1969, pp. 119-31.

DeLeeuw, Frank and Kalchbrenner, John. “Monetary andFiscal Actions: A Test of Their Relative Importance inEconomic Stabilization—Comment.” Federal ReserveBank of St. Louis Review, April 1969, 51(4), pp. 6-11.

DeLong, J. Bradford. “America’s Only Peacetime Inflation:The 1970s,” in Christina Romer and David Romer, eds.Reducing Inflation. Chicago: University of Chicago Press,1997, pp. 247-76.

__________. “The Triumph of Monetarism?” Journal ofEconomic Perspectives, Winter 2000, pp. 83-94.

DePrano, Michael and Mayer, Thomas. “Tests of theRelative Importance of Autonomous Expenditures andMoney.” American Economic Review, September 1965,pp. 729-52.

Dewald, William G. “Monetarism is Dead; Long Live theQuantity Theory.” Federal Reserve Bank of St. LouisReview, July/August 1988, 70(3), pp. 3-18.

__________ “Historical U.S. Money Growth, Inflation andInflation Credibility.” Federal Reserve Bank of St. LouisReview, November/December 1998, 70(6), pp. 13-23.

Dwyer, Gerald P. Jr. “Is Money Growth a Leading Indicatorof Inflation?” Conference volume from the internationalconference on The Conduct of Monetary Policy, Taipei,Taiwan, June 1998 (forthcoming).

__________ and Hafer, R.W. “Is Money Irrelevant?” FederalReserve Bank of St. Louis Review, May/June 1988, 70(3),pp. 3-17.

__________ and __________. “Are Money and Inflation StillRelated?” Federal Reserve Bank of Atlanta EconomicReview, Second Quarter 1999, pp. 32-43.

Economic Report of the President (various issues).

FOMC Minutes, 1968, 1969, 1970.

Francis, Darryl. “An Approach to Monetary and FiscalManagement.” Federal Reserve Bank of St. Louis Review,November 1968, 50(11), pp. 6-10.

__________. “Has Monetarism Failed? The RecordExamined.” Federal Reserve Bank of St. Louis Review,March 1972, 54(3), pp. 32-38.

__________. “The Usefulness of Applied Econometrics tothe Policymaker: An Address.” Federal Reserve Bank ofSt. Louis Review, May 1973, 55(5), pp. 7-10.

Friedman, Benjamin M. “Lessons from the 1979-82Monetary Policy Experiment.” American EconomicReview, Papers and Proceedings, May 1984, pp. 382-87.

Friedman, Milton, ed. Studies in the Quantity Theory ofMoney. Chicago: University of Chicago Press, 1956.

__________. A Program for Monetary Stability. New York:Fordham University Press, 1960.

__________. “The Role of Monetary Policy.” AmericanEconomic Review, March 1968, pp. 1-17.

__________. “Lessons from the 1979-82 Monetary PolicyExperiment.” American Economic Review, Papers andProceedings, May 1984, pp. 397-400.

__________. “An Interview with Milton Friedman.” FederalReserve Bank of Minneapolis The Region, June 1992.



__________ and Meiselman, David. “The Relative Stabilityof Monetary Velocity and the Investment Multiplier inthe United States, 1897-1958,” in Stabilization Policies.Englewood Cliffs, NJ: Prentice Hall, 1963, pp. 165-268.

__________ and __________. “Reply to Ando andModigliani and to DePrano and Mayer.” AmericanEconomic Review, September 1965, pp. 753-85.

__________ and Schwartz, Anna J. A Monetary History ofthe United States, 1867-1960. Princeton: PrincetonUniversity Press, 1963.

__________ and __________. Monetary Statistics of theUnited States: Estimates, Sources, and Methods. New York:National Bureau of Economic Research, 1970.

__________ and __________. Monetary Trends in the UnitedStates and the United Kingdom: Their Relation to Income,Prices, and Interest Rates, 1867-1975. Chicago: Universityof Chicago Press, 1982.

Gavin, William T. and Dewald, William G. “The Effect ofDisinflationary Policies on Monetary Velocity.” The CatoJournal, Spring/Summer 1989, pp. 149-64.

Gramley, Lyle E. and Chase, Samuel B. Jr. “Time Deposits inMonetary Analysis.” Federal Reserve Bulletin, October1965, pp. 1380-404.

Hafer, R.W. “The Role of Fiscal Policy in the St. LouisEquation.” Federal Reserve Bank of St. Louis Review,January 1982, 64(1), pp. 17-22.

__________. “Against the Tide: Malcolm Bryan and theIntroduction of Monetary Aggregate Targets.” FederalReserve Bank of Atlanta Economic Review, First Quarter1999, pp. 20-37.

Hargrove, Erwin C. and Morely, Samuel A., eds. ThePresident and the Council of Economic Advisors:Interviews with CEA Chairmen. Boulder: Westview Press,1984.

Harris, Seymour E. “Controversial Issues in RecentMonetary Policy: A Symposium, Introduction andSummary.” Review of Economics and Statistics, August1960, pp. 245-47.

Hansen, Alvin H. “Appropriate Monetary Policy, 1957-1960.” Review of Economics and Statistics, August 1960,pp. 255-56.

Hester, Donald D. “Keynes and the Quantity Theory: AComment on the Friedman-Meiselman CMC Paper.”Review of Economics and Statistics, November 1964, pp.364-68.

Hetzel, Robert L. “The Taylor Rule: Is it a Useful Guide toUnderstanding Monetary Policy?” Federal Reserve Bankof Richmond Economic Quarterly, Spring 2000, pp. 1-33.

Hoffman, Dennis and Rasche, Robert H. Aggregate MoneyDemand Functions: Empirical Applications in CointegratedSystems. Boston: Kluwer Academic Publishers, 1996.

Johnson, Harry G. “Monetary Theory and Policy.” AmericanEconomic Review, June 1962, pp. 335-84.

__________. “The Keynesian Revolution and the MonetaristCounter-Revolution.” American Economic Review, Papersand Proceedings, May 1971, pp. 1-14.

Jones, Homer. “Observations on Stabilization Management.”Federal Reserve Bank of St. Louis Review, December1970, 52(12), pp. 14-19.

Jordan, Jerry L. “The Andersen-Jordan Approach AfterNearly 20 Years.” Federal Reserve Bank of St. LouisReview, October 1986, 68(8), pp. 5-8.

Kane, Edward J. “Bureaucratic Self-Interest as an Obstacleto Monetary Reform,” in Thomas Mayer, ed. The PoliticalEconomy of American Monetary Policy. Cambridge:Cambridge University Press, 1990, pp. 283-98.

Karnosky, Denis S. “The Link Between Money and Prices—1971-76.” Federal Reserve Bank of St. Louis Review, June1976, 58(6), pp. 17-23.

Keran, Michael W. “The Effect of Total Demand on RealOutput.” Federal Reserve Bank of St. Louis Review, July 1966, 48(7), pp. 7-12.

King, Robert G. and Watson, Mark W. “The Post-War U.S.Phillips Curve: A Revisionist Econometric History.”Carnegie-Rochester Conference Series on Public Policy,December 1994, pp. 157-219.

Kozicki, Sharon. “How Useful are Taylor Rules forMonetary Policy?” Federal Reserve Bank of Kansas CityEconomic Review, Second Quarter 1999, pp. 5-33.

Laffer, Arthur B. and Ranson, R. David. “A Formal Model ofthe Economy.” The Journal of Business, July 1971, pp.247-70.


R E V I E W

Laurent, Robert D. “Is the Demise of M2 GreatlyExaggerated?” Contemporary Economic Policy, October1999, pp. 492-505.

Lucas, Robert E. Jr. “Econometric Policy Evaluation: ACritique.” Carnegie-Rochester Conference Series on PublicPolicy Vol. 1, The Phillips Curve and Labor Markets, 1976,pp. 19-46.

__________. “Nobel Lecture: Monetary Neutrality.” Journalof Political Economy, August 1996, pp. 661-82.

Maisel, Sherman J. Managing the Dollar. New York: WWNorton & Company, 1973.

Mayer, Thomas. Interviews, Special Collections, GeneralLibrary, University of California–Davis, 1995.

__________. Monetary Policy and the Great Inflation in theUnited States: The Federal Reserve and the Failure ofMacroeconomic Policy, 1965-1979. Cheltenham, UK:Edward Elgar, 1999.

McCallum, Bennett T. “Monetary Versus Fiscal PolicyEffects: A Review of the Debate,” in R.W. Hafer, ed. TheMonetary Versus Fiscal Policy Debate: Lessons from TwoDecades. Totowa, NJ: Rowman & Allanheld, 1986, pp. 9-32.

Meigs, A. James. “Recent Innovations in the Functions ofBanks.” American Economic Review, May 1966, 56,pp. 167-77.

__________. “Campaigning for Monetary Reform: TheFederal Reserve Bank of St. Louis in 1959 and 1960.”Journal of Monetary Economics, November 1976,pp. 439-53.

Meltzer, Allan H. “Monetarism: The Issues and theOutcome.” Atlantic Economic Journal, March 1998, pp. 8-31.

Meyer, Laurence H. and Rasche, Robert H. “EmpiricalEvidence on the Effects of Stabilization,” in StabilizationPolicies: Lessons from the ‘70s and Implications for the‘80s, Proceedings. St. Louis: Center for the Study ofAmerican Business and Federal Reserve Bank of St.Louis, 1980, pp. 41-102.

__________ and Varvares, Chris. “A Comparison of the St.Louis Model and Two Variations: Predictive Performanceand Policy Implications.” Federal Reserve Bank of St.Louis Review, December 1981, 63(10), pp. 13-25.

Mishkin, Frederic S. “What Should Central Banks Do?”Federal Reserve Bank of St. Louis Review,November/December 2000, 82(6), pp. 1-14.

Okun, Arthur M. “Potential GNP: Its Measurement andSignificance.” Proceedings of the Business and EconomicStatistics Section of the American Statistical Association,1962, pp. 98-104.

Orphanides, Athanasios. “Monetary Policy Rules Based onReal-Time Data.” FEDS Working Paper 1998:03, FederalReserve Board of Governors, December 1997.

__________ and Porter, Richard D. “P* Revisited: Money-Based Inflation Forecasts with a Changing EquilibriumVelocity.” Journal of Economics and Business,January/April 2000, 52(1-2), pp. 87-100.

Perlman, Richard, ed. Inflation: Demand-Pull or Cost-Push?Boston: D.C. Heath, 1965.

Perry, George L. Unemployment, Money Wage Rates, andInflation. Cambridge, MA: MIT Press, 1966.

Phelps, Edmund S. “Phillips Curves, Expectations ofInflation and Optimal Unemployment Over Time.”Economica, August 1967, pp. 254-81.

Phillips, A.W. “The Relation Between Unemployment andthe Rate of Change of Money Wage Rates in the UnitedKingdom, 1861-1957.” Economica, November 1958, 25,pp. 283-99.

Rasche, Robert H. “Comments on a Monetarist Approachto Demand Management.” Federal Reserve Bank of St.Louis Review, January 1972, 54(1), pp. 26-32.

__________. “Monetary Aggregates, Monetary Policy andEconomic Activity.” Federal Reserve Bank of St. LouisReview, March/April 1993, 75(2), pp. 1-35.

__________ and Tatom, John A. “The Effects of the NewEnergy Regime on Economic Capacity, Production andPrices.” Federal Reserve Bank of St. Louis Review, May1977, 59(5), pp. 2-12.

Rolnick, Arthur J. and Weber, Warren E. “Money, Inflation,and Output under Fiat and Commodity Standards.”Journal of Political Economy, December 1997, pp. 1308-21



Roosa, Robert V. Federal Reserve Operations in the Moneyand Government Securities Market. Federal Reserve Bankof New York, 1956.

Samuelson, Paul A. “Reflections on Monetary Policy.”Review of Economics and Statistics, August 1960, pp.263-69.

__________. Newsweek, 2 August 1971, p. 70.

__________ and Solow, Robert M. “Problem of Achievingand Maintaining a Stable Price Level: Analytical Aspectsof Anti-Inflation Policy.” American Economic Review,Papers and Proceedings, May 1960, pp. 177-94.

Sargent, Thomas J. The Conquest of American Inflation.Princeton: Princeton University Press, 1999.

Stone, Courtenay C. and Thornton, Daniel L. “Solving the1980s’ Velocity Puzzle: A Progress Report.” FederalReserve Bank of St. Louis Review, August/September1987, 69(7), pp. 5-23.

Taylor, John B. “Discretion Versus Policy Rules in Practice.”Carnegie-Rochester Conference Series on Public Policy,December 1993, pp. 195-214.

__________. “The Robustness and Efficiency of MonetaryPolicy Rules as Guidelines for Interest Rate Setting bythe European Central Bank.” Journal of MonetaryEconomics, June 1999, pp. 655-79.

Weintraub, Sidney. “Monetary Policy, 1957-59: Too Tight,Too Often.” Review of Economics and Statistics, August1960, pp. 279-82.

Wheelock, David C. “Monetary Policy in the GreatDepression and Beyond: The Sources of the Fed’sInflation Bias,” in Mark Wheeler, ed. The Economics ofthe Great Depression. Kalamazoo: The Upjohn Institute,1998, pp. 127-69.

__________. “National Monetary Policy by Regional Design:The Evolving Role of the Federal Reserve Banks inFederal Reserve System Policy,” in Jürgen von Hagen andChristopher J. Waller, eds. Regional Aspects of MonetaryPolicy in Europe. Boston: Kluwer Academic Publishers,2000, pp. 241-74.

Woodford, Michael. “Revolution and Evolution inTwentieth-Century Macroeconomics,” in P. Gifford, ed.Frontiers of the Mind in the Twenty-First Century.Cambridge, MA: Harvard University Press (forthcoming),<www.princeton.edu/~woodford/macro20C.pdf>.

Yohe, William P. and Karnosky, Denis S. “Interest Rates andPrice Level Changes, 1952-69.” Federal Reserve Bank ofSt. Louis Review, December 1969, 51(12), pp. 18-38.

R E V I E W


AThe first version of the St. Louis monetarist

model appeared in Andersen and Carlson (1970),Exhibit 1. It is summarized below. For a compari-son of the original and last versions of themodel, see Carlson (1986).

Exogenous Variables:∆Mt = change in money stock∆Et =change in high-employment Federal

expendituresXt

F =potential output

Endogenous Variables:∆Yt =change in total spending (nominal GNP)∆Pt = change in price level (GNP deflator)Dt= demand pressure∆Xt = change in output (real GNP)Rt = market interest rate∆Pt

A = anticipated change in price levelUt= unemployment rateGt = GNP gap

Appendix

Total Spending Equation

1)

Price Equation

2)

Demand Pressure Identity

3)

Total Spending Identity

4)

Interest Rate Equation

5)

∆ ∆ ∆ ∆ ∆

∆ ∆

∆

∆ ∆ ∆

∆ ∆ ∆ ∆ ∆

Y f M M E E

P f D D P

D Y X X

Y P X

R f M X X P P

t t t n t t n

t t t nA

t t tF

t

t t t

t t t t n t

=

=

= − −

= +

=

− −

−

−

−

1

2

1

3

( ... , ... )

( ... , )

( )

( , ... , , ttA

tA

t t n

t t t

t tF

t tF

P f P P

U f G G

G X X X

)

( ... )

( , )

( ) /

Anticipated Price Equation

6)

Unemployment Rate Equation

7)

GNP Gap Identity

8)

∆ ∆ ∆=

=

= −

− −

−

4 1

5 1



ComparingManufacturing ExportGrowth Across States:What Accounts for theDifferences?Cletus C. Coughlin andPatricia S. Pollard

Until the reduction in manufacturing exportscaused by the Asian crisis, U.S. manufactur-ing exports as a share of gross domestic

product had trended upward since the mid-1980s.As shown in Figure 1, this share increased from4.1 percent in 1986 to 7.3 percent in 1997, beforedecreasing to 7.0 percent in 1998. This feature ofthe internationalization of the U.S. economy hasspread unevenly across regions and states. Asshown in Table 1, from 1988 through 1998 theannual rate of change of manufacturing exportsranged from a low of –10.9 percent in Alaska to ahigh of 28.2 percent in New Mexico.1 In this paperwe examine the differences in the growth of man-ufacturing exports across states. Using a techniquecalled shift-share analysis, we isolate the effectsthat account for the difference between a state’smanufacturing export growth and U.S. manufac-turing export growth between 1988 and 1998.

Applying the shift-share method generates ameasure of each state’s net relative change overthe period. States in which manufacturing exportsgrew more (less) rapidly than the national averagebetween 1988 and 1998 have a positive (negative)net relative change. In classic shift-share models astate’s net relative change is separated into anindustry mix effect and a competitive effect. Theindustry mix effect is the change due to differ-ences in the initial industry makeup of the staterelative to the nation. A positive (negative) indus-try mix effect indicates that a state’s exports wererelatively more concentrated in industries whoseexports expanded faster (slower) than the overallnational average. Meanwhile, the competitiveeffect in these models is the change in exports

due to differences between the export growth of astate’s industries and export growth at the nation-al level, assuming the state’s industry mix was thesame as the nation’s.

Recent work by Gazel and Schwer (1998)extended the classic shift-share model to incorpo-rate the destination of a state’s exports. This ispotentially important because the geographic dis-tribution of exports differs a great deal acrossstates, a fact stressed by Erickson and Hayward(1991) and Cronovich and Gazel (1998) in generalstudies and by Coughlin and Pollard (2000) in arecent study of the impact of the Asian crisis onindividual states. These studies highlight theimportance of developments in foreign markets asa source of differential export performance acrossstates. In the present context, a positive (negative)destination effect indicates that a state's manufac-turing exports were initially relatively more con-centrated in export markets that subsequentlyexpanded faster (slower) than the overall nationalaverage.

In the following section we provide details onthe data used in our study and the differences inexport behavior across states. We also highlightthe differences in the overall growth of manufac-turing exports across states as well as the differ-ences in the industrial compositions and geo-graphic destinations of these exports. In the sub-

1986 87 88 89 90 91 92 93 94 95 96 97 19984

5

6

7

8

Manufacturing Exports as a Share of GDP1986-98

Perc

ent

Source: U.S. Department of Commerce, Bureau of EconomicAnalysis and Bureau of the Census; U. S. Department of Labor, Bureau of Labor Statistics

Figure 1

1Alaska was the only state with a decline in manufacturing exports.Between 1988 and 1998 these exports declined $1.4 billion, prima-rily as a result of a decline in exports of food products to Japan.

Cletus C. Coughlin is a vice president and associate director of re-search and Patricia S. Pollard is an economist and research officer atthe Federal Reserve Bank of St. Louis. Heidi L. Beyer providedresearch assistance.


R E V I E W

sequent section we discuss shift-share analysisand the two models we calculate—a classic shift-share model and Gazel and Schwer’s (1998)model. Next, we examine our results to assess theimportance of the industry mix, competitive, anddestination effects. A summary of the key findingscompletes the paper.

EXPORT DATA DETAILSThe data on state manufacturing exports used

in this study were prepared by the MassachusettsInstitute for Social and Economic Research(MISER) at the University of Massachusetts. Thesedata are export shipments by state of origin ofmovement at the two-digit Standard IndustrialClassification (SIC) industry level. The MISERexport data are regarded as the best available datasource for state exports; however, these data havesome well-known weaknesses that have been dis-cussed in Cronovich and Gazel (1999) andCoughlin and Mandelbaum (1991). One potentiallyimportant problem is that the identified exportstate may not be the state of manufacture, butrather the state of a broker (or wholesaler) or thestate where a number of shipments were consoli-dated. This problem is more pronounced forexports of agricultural commodities than the focusof our study, manufactured goods.

State exports exhibit much variety in boththeir absolute size and relative importance for eco-nomic activity in their respective states. Exportsduring 1998 ranged from $98.9 billion inCalifornia to $0.2 billion in Hawaii. As shown inTable 1, California and Texas led the nation during1998 with 15.8 percent and 13.1 percent of thenation’s manufacturing exports, respectively.Meanwhile, primarily because of their small eco-nomic size, seven states—Alaska, the District ofColumbia, Hawaii, Montana, North Dakota, SouthDakota, and Wyoming—had shares of the nation’smanufacturing exports that were 0.1 percent orlower.2

Adjusting for the size of a state’s economy,using 1998 gross state product data, produces adifferent picture of the importance of a state’sexports. As shown in Table 2, the relative impor-tance of exports as a share of gross state productvaried substantially across states. Vermont (one ofthe smaller states in terms of total exports) andWashington had export shares exceeding 20 per-cent.3 Michigan and Texas were also leading states,with export shares exceeding 10 percent. At theother extreme, the states showing the lowest rela-

tive manufacturing export involvement—theDistrict of Columbia and Hawaii—had exportshares of less than 1 percent.

A final point illustrated in Table 2 is theincreasing importance of manufacturing exportsfor state economies. Between 1988 and 1998, only6 of the 51 states experienced a decline in theirratios of manufacturing exports to gross stateproduct—Alaska, Delaware, the District ofColumbia, Louisiana, Michigan, and Montana. Inseven states the share of manufacturing exportsincreased by more than 3 percentage points. Theincrease was largest in Vermont—more than 13percentage points.

In the present paper, we focus on the growthof manufacturing exports and connect this growthto differences among states in the competitive,industry mix, and destination effects. For com-pleteness we examined whether expressing thechanges in exports in real terms, as opposed tonominal terms, altered our results in any meaning-ful way. The short answer is no.4 One reason isthat state manufacturing export growth rates inreal and nominal terms are virtually identical.Between 1988 and 1998 the real and nominalcompound annual growth rates are within 1 per-centage point of each other for 47 of the 51states. Not surprisingly, the simple correlationbetween these two growth rate measures is quitehigh (0.99). Thus, all our calculations use nominalvalues.

An Industry View of ManufacturingExport Growth

U.S. manufacturing exports grew at differentrates across industries. Table 3 shows these differ-ent rates using two-digit SIC codes. Lumber and

2 We treat the District of Columbia as the 51st state.3 A note of caution is in order. Because the measure of manufacturing

exports is based on shipments, the value of exports for a state is notequivalent to value added. Thus, we are not suggesting that morethan 23 percent of Vermont’s gross state product was due to manu-facturing exports. We are using this measure only as suggestive evi-dence that the importance of manufacturing exports varies acrossstates.

4 To calculate the real percentage change in exports, exports in 1998were deflated by their change in price between 1988 and 1998;however, export price data are available using the StandardInternational Trade Classification (SITC) system rather than by SICcode. Thus, we started with an export price index that groups thedata based on the SITC system and matched these industries withthe appropriate SIC codes. When multiple SITC codes fit one SICcategory, a weighted average of the price indices for those cate-gories was constructed to produce the price index on an SIC basis.For additional details, see Pollard and Coughlin (1999).



State Manufacturing Exports

Share of national Annual growth rate manufacturing exports

1988-98 1998State (percent) (percent)

Alabama 10.6 1.04Alaska –10.9 0.10Arizona 13.4 1.86Arkansas 13.8 0.38California 9.1 15.83Colorado 10.7 0.88Connecticut 7.9 1.22Delaware 6.9 0.37District of Columbia 3.2 0.05Florida 8.3 4.35Georgia 12.7 2.24Hawaii 5.5 0.04Idaho 9.8 0.24Illinois 10.6 4.92Indiana 10.9 2.11Iowa 9.3 0.80Kansas 9.2 0.65Kentucky 12.7 1.33Louisiana 4.3 1.61Maine 8.7 0.29Maryland 7.9 0.80Massachusetts 5.9 2.65Michigan 4.1 4.90Minnesota 6.9 1.45Mississippi 6.6 0.38Missouri 9.2 0.98Montana 3.4 0.05Nebraska 9.6 0.32Nevada 12.0 0.11New Hampshire 6.4 0.29New Jersey 7.7 2.58New Mexico 28.2 0.31New York 4.7 6.05North Carolina 10.5 2.50North Dakota 12.9 0.10Ohio 8.8 4.19Oklahoma 7.6 0.47Oregon 10.4 1.33Pennsylvania 8.6 2.69Rhode Island 9.1 0.17South Carolina 11.4 1.34South Dakota 18.1 0.07Tennessee 13.3 1.58Texas 10.4 13.14Utah 14.0 0.52Vermont 14.1 0.62Virginia 7.9 1.92Washington 9.7 6.25West Virginia 4.4 0.23Wisconsin 9.5 1.58Wyoming 8.6 0.08

Table 1

Manufacturing Exports as a Share of Gross State Product

1998 1988 Difference betweenState (percent)* (percent) 1998 and 1988

(percentage points)†

Alabama 5.93 3.64 2.30Alaska 2.60 9.36 –6.76Arizona 8.69 5.24 3.44Arkansas 3.88 1.91 1.97California 8.84 6.11 2.73Colorado 3.88 3.00 0.87Connecticut 5.38 4.04 1.34Delaware 6.88 6.97 –0.09District of Columbia 0.59 0.66 –0.07Florida 6.49 5.46 1.03Georgia 5.52 3.37 2.15Hawaii 0.56 0.51 0.05Idaho 4.90 3.99 0.91Illinois 7.22 4.53 2.69Indiana 7.56 4.76 2.80Iowa 5.92 4.26 1.65Kansas 5.30 3.66 1.64Kentucky 7.77 4.14 3.64Louisiana 7.79 8.06 –0.28Maine 5.52 3.64 1.88Maryland 3.05 2.31 0.74Massachusetts 6.92 6.21 0.71Michigan 10.39 11.68 –1.28Minnesota 5.62 5.23 0.39Mississippi 3.79 3.50 0.29Missouri 3.75 2.62 1.13Montana 1.58 1.89 –0.32Nebraska 3.84 2.74 1.10Nevada 1.11 0.90 0.21New Hampshire 4.35 4.21 0.15New Jersey 5.06 3.93 1.12New Mexico 4.06 0.68 3.38New York 5.35 5.27 0.09North Carolina 6.62 4.53 2.09North Dakota 3.72 1.96 1.76Ohio 7.67 5.47 2.20Oklahoma 3.60 2.71 0.90Oregon 7.95 6.32 1.63Pennsylvania 4.63 3.34 1.28Rhode Island 3.54 2.31 1.23South Carolina 8.36 4.93 3.43South Dakota 1.95 0.70 1.25Tennessee 6.19 3.26 2.93Texas 12.72 9.16 3.56Utah 5.46 3.26 2.20Vermont 23.80 10.07 13.72Virginia 5.21 4.29 0.92Washington 20.24 16.36 3.89West Virginia 3.66 3.61 0.05Wisconsin 6.27 4.51 1.76Wyoming 2.90 1.90 1.00

*Export shares of 10 percent or above are shown in bold; values of 1 percent or below are shown in italic.†Increases in export share exceeding 3 percentage points are shown in bold.


R E V I E W

Table 2



wood products (SIC 24) exports grew slowest (1.1percent), while furniture and fixtures (SIC 25)exports grew fastest (18.6 percent). If the industrymix of exports was identical across states, thenthese differences in growth rates of industryexports at the national level would not explain dif-ferences in export growth at the state level. Anobvious question is: Was the industry mix of stateexports identical?

Using an index proposed by Finger andKreinin (1979) for a slightly different purpose, wecalculate a measure of the similarity between thesectoral concentration of a state’s exports and thatof the United States overall.5 The range of thisindex is from zero, indicating complete dissimilar-ity, to 100, indicating the state’s sectoral distribu-tion of exports is identical to the national distribu-tion. Table 4 reveals a wide range of export simil-

arity with values for 1988 ranging from 19.4 forAlaska, which indicates very little similarity withthe national distribution, to levels exceeding 80.0for Florida, Indiana, Kentucky, Maryland, andPennsylvania.

Table 4 also provides information on howexport similarity in each state changed during1988-98. For 36 of the 51 states, this export-similarity index increased, indicating that theindustry distribution of exports from these statesbecame more similar to the national distribution.Twenty-three states had increases of more than 5

U.S. Export Growth by Industry 1988-98

Compound annualSIC Description rate (percent)

20 Food and kindred products 5.3

21 Tobacco products 5.4

22 Textile mill products 11.4

23 Apparel and related products 16.8

24 Lumber and wood products (except furniture) 1.1

25 Furniture and fixtures 18.6

26 Paper and allied products 6.8

27 Printing and publishing 9.9

28 Chemicals and allied products 6.4

29 Refined petroleum and coal products 4.5

30 Rubber and miscellaneous plastics products 11.2

31 Leather and leather products 7.3

32 Stone, clay, glass, and concrete products 8.3

33 Primary metal products 6.6

34 Fabricated metal products (except machinery and transportation equipment) 10.3

35 Industrial and commercial machinery and computer equipment 7.9

36 Electrical and electronic machinery, equipment, and supplies 13.1

37 Transportation equipment 8.7

38 Scientific and professional instruments; photographic and optical goods, etc. 8.8

39 Miscellaneous manufactured goods 10.0

20-39 All manufacturing industries 8.7

Table 3

5 This export-similarity index is calculated quite easily: For a specificstate, calculate the state’s share of its total exports by a specificindustry and corresponding national share for each of the 20 SICcategories. For each industry, compare the state share with thenational share, take the minimum, and then sum these 20 values;next, multiply by 100.


R E V I E W

Export-Similarity Index on an Industry BasisDifference between

State 1988 1998 1998 and 1988*

Alabama 72.4 71.4 –1.0Alaska 19.4 34.1 14.7Arizona 69.6 72.5 3.0Arkansas 66.0 66.8 0.8California 79.5 77.0 –2.5Colorado 55.9 64.0 8.1Connecticut 78.3 73.9 –4.3Delaware 46.1 55.6 9.5District of Columbia 56.7 59.1 2.4Florida 81.8 89.5 7.7Georgia 69.7 78.5 8.8Hawaii 37.9 52.3 14.4Idaho 55.0 58.7 3.7Illinois 74.2 88.2 14.1Indiana 82.5 78.5 –4.0Iowa 66.9 69.6 2.7Kansas 59.1 59.0 –0.2Kentucky 80.1 74.2 –5.9Louisiana 36.6 36.7 0.1Maine 39.4 45.6 6.2Maryland 82.7 81.8 –0.9Massachusetts 61.2 71.9 10.8Michigan 47.3 63.1 15.8Minnesota 60.9 68.5 7.6Mississippi 51.3 61.1 9.8Missouri 78.8 77.7 –1.2Montana 39.7 54.3 14.6Nebraska 53.8 56.6 2.8Nevada 48.5 71.6 23.1New Hampshire 61.6 66.5 5.0New Jersey 74.3 72.9 –1.4New Mexico 62.0 31.8 –30.2New York 70.2 75.9 5.6North Carolina 65.3 71.5 6.2North Dakota 64.5 52.7 –11.8Ohio 77.5 78.5 1.0Oklahoma 73.9 74.5 0.6Oregon 62.1 71.4 9.2Pennsylvania 81.9 85.3 3.4Rhode Island 64.0 65.9 1.9South Carolina 59.5 70.1 10.6South Dakota 62.4 65.2 2.8Tennessee 78.2 84.3 6.0Texas 74.7 82.0 7.3Utah 55.5 63.2 7.6Vermont 52.1 35.0 –17.0Virginia 67.4 73.1 5.7Washington 48.5 41.3 –7.2West Virginia 37.0 34.3 –2.6Wisconsin 67.8 73.0 5.2Wyoming 29.6 17.5 –12.1

* Increases of 5 or more index points are shown in bold; declines of 5 or more index points are shown in italic.

Table 4



index points. The increase was largest in Nevada—roughly 23 points. Seven other states—Alaska,Hawaii, Illinois, Massachusetts, Michigan,Montana, and South Carolina—experiencedincreases of 10 or more index points. On the otherhand, only six states experienced declines of 5 ormore points. Of these states, New Mexico had thelargest decline, about 30 points.

A Geographic View of ManufacturingExport Growth

The importance of specific destinations hasalso changed over time and varies across states.Table 5 separates the world into eight regions: thethree leading countries for U.S. exports (Canada,Mexico, and Japan) and five areas (Africa, otherAsia, Europe, Oceania, and other WesternHemisphere).6 The data show that during 1988-98U.S. export growth ranged from 14.8 percent inMexico to 4.9 percent in Africa. If the geographicmix of exports was identical across states, thendifferences in growth rates of exports by geo-graphic area at the national level would notexplain differences in export growth at the statelevel.

Using an export-similarity index based onexport destination, Table 6 reveals a range ofexport similarity in 1988 from 34.1 for Alaska to92.5 for Alabama.7 Three other states—Mississippi,North Carolina, and Pennsylvania—had index val-ues exceeding 90.0. Numerous other states, 24 tobe precise, had index values between 80.0 and90.0. Consequently, the geographic distributions ofexports for nearly half the states matched veryclosely with the national distribution of exports.

Table 6 also provides information on how thegeographic concentration of each state’s exportshave changed during 1988-98. Although there isclear evidence that the industry mix of exports bystates became more similar to the national distri-bution, there is little evidence that the distributionof exports based on destination changed much.Most states, 32 to be precise, experienced a changein this index in the range of –5 to 5 index points.Only eight states experienced increases of morethan 5 points, with Delaware having the largest(roughly 38 points). On the other hand, 11 statesexperienced declines of 5 or more points, withNew Mexico’s index declining the most (roughly46 points).

THE BASICS OF SHIFT-SHAREANALYSIS

Shift-share analysis is a method of separatinga change, in our case the change in a state’s man-ufacturing exports between 1988 and 1998, intomeaningful components. The insert discusses thedifference between this accounting explanationand an economic explanation of the change in astate’s manufacturing exports. An economic expla-nation identifies factors that interact to determinethe pattern and level of trade flows. Various inter-national trade theories provide guidance as to thepotential determinants. Similarly, the existence ofalternative shift-share formulations reflects differ-ences of opinion as to exactly which componentsare most useful.8

The Classic Shift-Share Model

Using the classic shift-share model, the changein a state’s manufacturing exports is separated intoa national growth effect, an industry mix effect,and a competitive effect. The national growth

Manufacturing Export Growth by ForeignMarket 1988-98

Compound annual rateRegion (percent)

Canada 9.7Mexico 14.8Other Western Hemisphere 10.8Japan 5.3Other Asia 8.4Africa 4.9Europe 7.1Oceania 5.6World 8.7

Table 5

6 The Middle East is included in other Asia.

7 This export-similarity index is also calculated quite easily: For aspecific state, calculate the state’s share of its total exports that areshipped to a specific region for each of the eight regions and thecorresponding national share. For each destination, compare thestate share with the national share, take the minimum, and thensum these eight values; next, multiply by 100. The range of thisindex is from zero, indicating complete dissimilarity, to 100, indi-cating the state’s regional distribution of exports is identical to thenational distribution.

8 See Loveridge and Selting (1998) for a review of seven shift-sharemodels.


R E V I E W

Export-Similarity Index on a Destination Basis

Difference betweenState 1988 1998 1998 and 1988*

Alabama 92.5 87.9 –4.6Alaska 34.1 61.4 27.3Arizona 76.3 73.9 –2.3Arkansas 89.2 88.3 –0.9California 81.5 82.4 0.9Colorado 77.2 77.1 –0.1Connecticut 84.9 82.6 –2.3Delaware 45.7 84.1 38.4District of Columbia 53.5 66.1 12.6Florida 57.6 50.7 –6.9Georgia 85.2 86.9 1.8Hawaii 49.2 34.2 –15.0Idaho 77.4 76.2 –1.2Illinois 87.3 84.9 –2.4Indiana 83.1 74.5 –8.6Iowa 78.7 77.7 –1.0Kansas 86.6 82.6 –4.0Kentucky 78.8 80.2 1.4Louisiana 78.8 73.8 –5.0Maine 80.9 69.6 –11.3Maryland 82.8 80.5 –2.4Massachusetts 76.8 81.4 4.6Michigan 49.2 63.9 14.8Minnesota 84.8 84.6 –0.1Mississippi 90.5 83.4 –7.1Missouri 85.0 82.5 –2.5Montana 55.2 72.9 17.7Nebraska 79.2 73.4 –5.7Nevada 71.8 78.9 7.1New Hampshire 82.1 81.9 –0.2New Jersey 85.0 89.8 4.9New Mexico 80.8 35.1 –45.7New York 85.1 87.9 2.8North Carolina 91.8 90.9 –0.9North Dakota 47.7 61.3 13.6Ohio 79.4 72.9 –6.5Oklahoma 84.2 85.5 1.3Oregon 77.9 76.9 –1.1Pennsylvania 91.2 88.9 –2.3Rhode Island 87.0 87.5 0.5South Carolina 86.8 87.7 0.9South Dakota 76.5 79.6 3.1Tennessee 87.3 89.4 2.0Texas 72.5 67.5 –4.9Utah 86.7 69.5 –17.2Vermont 52.3 54.9 2.7Virginia 80.2 81.3 1.1Washington 81.7 66.5 –15.2West Virginia 82.4 89.1 6.8Wisconsin 83.1 82.7 –0.3Wyoming 72.9 71.6 –1.4

* Increases of 5 or more index points are shown in bold; declines of 5 or more index points are shown in italic.

Table 6



CONNECTING SHIFT-SHARE ANALYSIS TO INTERNATIONAL TRADE FLOWSIn the present analysis, shift-share analysis is

an accounting tool to separate the change in astate’s manufacturing exports into potentiallymeaningful components. The analysis, however,does not provide an economic explanation as towhy a state’s exports grew faster or slower thanthe national average. The following discussionelaborates on the distinction between an ac-counting and an economic explanation.

International trade occurs in response to dif-ferences in prices for the same good betweencountries.1 If a potential U.S. consumer can pur-chase a product at a lower price from a producerin Mexico than in the United States, then anincentive exists to engage in international trade.Assuming the costs stemming from trade barriers,including governmental policies such as tariffsand natural barriers such as transportation costs,are not so large as to eliminate the price advan-tage, then the product will be exported fromMexico to the United States.

A focal point of international trade theory isto identify the reasons why prices differ acrosscountries. Differences across countries in termsof labor forces, stocks of public and privatecapital, technologies, tastes—even labor laws andenvironmental standards—are some of the manyreasons that might cause prices to differ. More-over, the economic size of trading partners islikely to be a key determinant of the magnitudeof the trade flows between two countries. Largereconomic size is likely associated with largertrade flows.

All of these factors—trade policies, transpor-tation costs, productive resources, technology,tastes, and income—and more interact to deter-mine the pattern and level of trade flows. Con-sequently, changes in these factors will likelyaffect how trade patterns and levels change overtime. In the present study numerous factors areundoubtedly responsible for the export perform-ance of a state between 1988 and 1998.

The shift-share analysis we perform does notallow us to identify which factors determined therelative export performance of a state. At best itsuggests which factors deserve scrutiny. Forexample, assume a state was found to have afavorable destination effect. In 1988 a specificstate might have exported a relatively larger por-tion of its manufactured exports to Mexico than

other states. This initial situation reflects the ad-vantages possessed by the state with respect toexporting to Mexico. Transportation costs mighthave played a key role in this initial condition.Between 1988 and 1998, faster U.S. exportgrowth to Mexico than to other regionsthroughout the world could be expected tobenefit states with relatively larger dependenceon the Mexican market. The shift-share analysisprovides information as to the extent to whichthe state is likely to be affected by thisdevelopment. For the estimate of the destinationeffect to be plausible, one must anticipate that theadvantages under-lying the state’s initial exportrelationship with Mexico are not altered substan-tially during the period under consideration. Theshift-share analysis, however, does not providethe economic reasons for the actual change in astate’s exports to Mexico. Two possible economicreasons that come to mind immediately are theimplementation of the North American FreeTrade Agreement and faster economic growth inMexico than in other world areas.

Similar comments can be made concerningthe insights revolving around the industry mixresults. A state with a favorable industry mix isone whose exports initially were relatively moreconcentrated in industries that experienced rela-tively rapid growth between 1988 and 1998. Theshift-share analysis provides information on theextent to which the state is likely to be affectedby the rapid export growth in a specific industry.Additional analysis of economic factors is neededto determine the reasons for the actual change ina state’s exports in a specific industry. Some re-source or technological change may cause therelatively rapid export growth.

In summary, the shift-share analysis providessome basic information to begin the analysis as towhy the export performance of a specific statewas above or below the national average. Lookingat a state’s industry mix and geographic distribu-tion of exports is a reasonable first step in tryingto provide an economic explanation for a state’sexport performance.

1 Trade may also result from a difference across countries in thequality or variety of goods.


R E V I E W

effect is the amount that a state’s exports wouldhave increased (or decreased) had they grown atthe same rate as the nation’s exports. Because thefocus of many studies is how well a specific statehas performed relative to the nation, frequentlythis national growth effect is simply subtractedfrom the change in a state’s manufacturingexports to yield a state’s net relative change. Thenthe analysis focuses on the reasons that a state’sperformance differs from the nation’s perform-ance.

Regardless of the handling of the nationalgrowth effect in the classic shift-share model, dif-ferential state performance is accounted for by anindustry mix effect and a competitive effect. Theindustry mix effect is the amount of change attrib-utable to differences in the initial industry makeupof the state relative to the nation. A positive (nega-tive) industry mix effect indicates that a state’sexports were initially relatively more concentratedin industries whose exports expanded faster (slow-er) than the overall national average. The competi-tive effect measures state economic changes notattributable to national growth or industry mixeffects; it captures how much the state deviatesfrom what would be expected if state exportgrowth were due solely to national export expan-sion and the state’s industry makeup. In the classicshift-share model, a competitive effect indicatesthe quantitative difference between a state’sexports and those of the nation caused exclusivelyby the difference in the growth rate of that state’sindustries compared with that of the nation. Thus,the competitive effect is a residual. It captures theeffect of changes in various factors operating atthe state level, such as endowments of humancapital and, possibly, export promotion expendi-tures.9

Because a state’s net relative change is simplythe sum of the industry mix and competitiveeffects, this relationship can be expressed mathe-matically in a straightforward manner. Using thesame notation as Gazel and Schwer (1998), therelationship can be expressed:

(1)

where NRC stands for net relative change; s is asuperscript designating a specific state; X is thedollar value of exports; x is the growth rate of

exports over the entire period of the study; o is asubscript designating the first year of the period ofstudy; i is a subscript designating a specific two-digit manufacturing industry; and n is a super-script designating the nation.

The first term on the right side of equation (1)is the industry mix effect. The second term is thecompetitive effect. In terms of the notation, theindustry mix effect for state s is the summationover the i two-digit manufacturing industries(Σ

i) of the difference between the growth national-

ly of exports of industry i (xni ) and the overall

national growth rate of exports (xn), multiplied bythe level of the state’s exports of industry i at thebeginning of the period (Xs

i,o) . The competitiveeffect for state s is the summation over the i two-digit manufacturing industries (Σ

i) of the differ-

ence between the state’s growth rate of exports ofindustry i (xs

i ) and the corresponding nationalgrowth rate of exports of industry i (xn

i ), multipliedby the level of the state’s exports of industry i atthe beginning of the period (Xs

i,o) .

INCORPORATING DESTINATION INTOSHIFT-SHARE ANALYSIS

Gazel and Schwer’s incorporation of the destina-tion of a state’s exports into a shift-share model isstraightforward. The destination effect is the amountof the net relative change attributable to differencesin the state’s initial export destinations relative tothose of the nation. A positive (negative) destinationeffect indicates that a state’s exports were relativelymore concentrated in foreign markets whose pur-chases from the United States expanded faster (slow-er) than the overall national increase in exports.10

A state’s net relative change is now the sum ofthe industry, competitive, and destination effects.The industry mix effect (the first term) is unchangedfrom equation (1), while the competitive effect fromthe formula is decomposed into a new competitiveeffect (the second term) and the destination effect

NRC X x x X x xs

ii os

in n

ii os

is

in= − + −Σ Σ, ,( ) ( )

9 Coughlin and Mandelbaum (1990) found that the percentage changein human capital per worker was a statistically significant determi-nant of the competitive effect for the change in state exports from1976 to 1986.

10 Cronovich and Gazel (1998) used state-specific trade weights tocreate a measure of trade-weighted foreign income and foundthis measure to be a positive, statistically significant determinantof state-level manufacturing exports.



(the third term). Thus, a state’s net relative changebecomes:

(2)

where j is a subscript designating a foreign market.Focusing on changes from equation (1), it is

easier to begin with the destination effect. Interms of the notation, this effect is the summationover the j foreign markets (Σ

j) of the difference

between the growth rate nationally of exports toforeign market j, (x

nj ), and the overall national

growth rate of exports (xn), multiplied by the levelof the state’s exports to foreign market j at thebeginning of the period (X s

j ,o) . The redefined com-petitive effect is simply the competitive effectdefined in equation (1) less the destination effect.By removing the destination effect from the com-petitive effect, Gazel and Schwer eliminate one ofthe explanations as to why exports from individ-ual state industries grow at a rate different fromthat of the corresponding national industries.Gazel and Schwer argue that their modification isespecially appealing because it retains the tradi-tional shift-share approach, while highlighting fac-tors that can be viewed as proxies for the supplyside (the industry mix effect) and the demand side(the destination effect). The competitive effect, theresidual, still may be picking up both unmodeledsupply and demand factors.

SHIFT-SHARE ANALYSES OFMANUFACTURING EXPORTS

We present and discuss two sets of results. Weexamine the results using classic shift-share analy-sis and then the results using Gazel and Schwer’sextension.

States in which exports grew more rapidlythan the national average between 1988 and 1998exhibit a positive net relative change, while thosestates whose exports grew less rapidly exhibit anegative net relative change. Over this period,exports of 29 states grew more rapidly than thenational average, while exports of the remaining22 states grew less rapidly. The map in Figure 2distinguishes states on the basis of their net rela-tive change. States with positive values for net rel-ative change can be found throughout the conti-

nental United States; however, the states along theAtlantic Ocean northward from Virginia tend tohave negative values for net relative change.Selected states in the South, Midwest, and theRocky Mountains, as well as Alaska and Hawaii,also show slower export growth than the nationalaverage.

The classic shift-share results are presented inTable 7. To increase comparability among states,each state’s net relative change and the individualcomponents accounting for the net relative changeare expressed as percentages of the export levelsthat would have been achieved in 1998 had thestate’s exports grown at the national rate between1988 and 1998. Thus, California’s exports in 1998were 4.1 percent higher than if its exports hadgrown at the national rate between 1988 and1998. This normalization allows us to highlight theexport performance of those states that have sub-stantially over- and under-performed the nation asa whole. Export growth in 11 states was more than30 percent higher than would have been the casehad their exports grown at the national ratebetween 1988 and 1998. This difference was mostpronounced in New Mexico. Meanwhile, sevenstates, with Alaska leading the way, had substan-tially slower export growth than that of the nation.

The net relative change in state exports wasdriven by the competitive effect, a result similar tothat of Coughlin and Mandelbaum (1990).11 For

NRC X x x

X x x X x x

X x x

s

ii os

in n

ii os

is

in

jj os

jn n

jj os

jn n

= − +

− − −

+

−

Σ

Σ Σ

Σ

,

, ,

,

( )

( ) ( )

( )

Net Relative Change

Positive (29) Negative (22)

Figure 2

11To check whether our results were sensitive to the time period, wealso examined the periods 1988-93, 1993-98, and 1996-98 sepa-rately and found results similar to those reported in Table 7.


R E V I E W

Classic Shift-Share Results

State Net relative change* Industry mix effect Competitive effect

Alabama 19.5 –3.2 22.7Alaska –86.3 –29.0 –57.3Arizona 53.3 18.5 34.8Arkansas 59.2 –8.9 68.1California 4.1 6.2 –2.1Colorado 20.0 1.8 18.1Connecticut –7.1 3.0 –10.2Delaware –14.7 –0.8 –13.9District of Columbia –40.2 8.9 –49.1Florida –3.2 1.2 –4.4Georgia 43.3 0.5 42.9Hawaii –25.7 –9.8 –16.0Idaho 11.1 4.0 7.1Illinois 19.6 0.8 18.7Indiana 22.7 –0.4 23.1Iowa 5.9 –3.9 9.8Kansas 4.9 –7.8 12.8Kentucky 44.1 –4.4 48.5Louisiana –33.9 –19.9 –14.0Maine –0.1 –5.3 5.2Maryland –6.5 1.9 –8.5Massachusetts –22.9 1.8 –24.7Michigan –35.0 1.9 –36.9Minnesota –15.3 –0.6 –14.6Mississippi –17.7 –8.8 –8.9Missouri 5.2 –1.0 6.3Montana –39.0 –16.1 –22.9Nebraska 9.4 –8.4 17.8Nevada 34.7 –1.8 36.5New Hampshire –19.2 5.4 –24.7New Jersey –8.4 –1.4 –7.1New Mexico 420.9 –5.1 426.0New York –31.2 3.2 –34.4North Carolina 18.3 –0.6 18.9North Dakota 46.5 –7.1 53.5Ohio 1.7 0.1 1.6Oklahoma –9.2 –2.1 –7.1Oregon 16.8 –17.2 34.0Pennsylvania –0.8 3.5 –4.3Rhode Island 4.4 6.4 –2.0South Carolina 28.5 –0.1 28.7South Dakota 128.2 –2.3 130.5Tennessee 52.0 –0.2 52.2Texas 17.1 –0.5 17.5Utah 61.7 12.2 49.6Vermont 62.7 23.1 39.6Virginia –6.8 –7.0 0.2Washington 9.9 –9.5 19.4West Virginia –33.2 –12.7 –20.6Wisconsin 7.4 –0.8 8.2Wyoming –0.4 –17.7 17.3

* Net relative increases of 30 percent or more are shown in bold; decreases of 30 percent or more are shown in italic.

Table 7



example, for the 29 states with a positive net rela-tive change, the competitive effect is positive for27 states, whereas the industry mix effect is posi-tive for only 10 states. For the 22 states with anegative net relative change, the competitive effectis negative for 19 states, whereas the industry mixeffect is negative for 13 states.

Results based on equation (2) are presented inTable 8. The competitive effect remains the mostimportant factor accounting for a state’s net rela-tive change. The industry mix and destinationeffects are similar in importance, with the destina-tion effect arguably being slightly more important.Using the absolute values of the individual effects,one finds that the competitive effect is the largesteffect for 45 of the 51 states, whereas the industrymix and destination effects are the largest for 4and 2 states, respectively.12 The destination effectis the second largest effect for 28 states, whereasthe industry mix effect is the second largest effectfor 20 states. Finally, the industry mix effect is thesmallest effect for 27 states, whereas the destina-tion effect is the smallest for 21 states.

California, Louisiana, Montana, and RhodeIsland were the four states for whom the industrymix effect dominated the shift-share results. ForCalifornia and Rhode Island the industry mixeffects were positive, indicating that their exportswere relatively more concentrated in industrieswhose exports were rising faster than the nationalaverage for all manufacturing exports. ForCalifornia this was primarily the electrical andelectronic machinery industry (SIC 36) and forRhode Island these were the miscellaneous manu-factured goods industry (SIC 39) and the electricaland electronic machinery industry. As Table 3shows, exports of both of these industries grewfaster than the average for all manufacturingexports.

For Louisiana and Montana the industry mixeffects were negative, indicating that their exportswere relatively more concentrated in industrieswhose exports were rising less than the nationalaverage for all manufacturing exports. Food (SIC20) and chemicals (SIC 28) were the key industriesfor Louisiana, whereas primary metals (SIC 33),lumber and wood products (SIC 24), and chemi-cals were the key industries for Montana. All ofthese industries had export growth rates below thenational average, as shown in Table 3.

Hawaii and Texas were the two states forwhom the foreign destination effect dominatedthe shift-share results. For Hawaii this effect was

negative, indicating that its exports were relativelymore concentrated in markets whose purchasesfrom the United States expanded less than thenational increase in manufacturing exports. Japanaccounted for over 50 percent of Hawaii’s manu-facturing exports in 1988. National exports toJapan rose more slowly than did exports to allcountries, as shown in Table 5.

For Texas the destination effect was positive,indicating that its exports were relatively moreconcentrated in markets whose purchases fromthe United States expanded faster than the nation-al increase in manufacturing exports. Mexicoaccounted for 28 percent of Texas’s manufacturingexports in 1988. U.S. exports to Mexico rose muchfaster than exports to all countries, as shown inTable 5.

The absolute values of these three effects arenot the entire story. For the 29 states with fasterexport growth rates than the national average, thecompetitive effect was positive in all 29 states. Forthese states, the industry mix effect was positivefor 10 states and the destination effect was posi-tive for 6 states. Thus, the industry mix and for-eign destination effects were more likely to benegative than positive for these states. For the 22states with slower growth rates than the nationalaverage, the competitive effect was negative for 18of them. The industry mix and destination effectsalso tended to be negative. The industry mix effectwas negative for 13 of the 22 states, whereas thedestination effect was negative for 17 of the 22states.13

In view of the conflicting results concerningthe relationship between net relative change andthe destination effect and between net relative

12 Our discussion of the results makes no attempt to differentiatestates on the basis of size; however, two of the six states in whichthe competitive effect is not the largest shift-share componentwere California and Texas. These two were the leading export statesin 1998.

13 When we examined the periods 1988-93 and 1993-98 separately,the results for these periods were very similar to the results report-ed in Table 8. For 1996-98 the competitive effect is the most impor-tant factor; however, it is not as dominant as in the other periods.The industry mix effect is the second most important factor,whereas the foreign destination effect is the least important factor.As Gazel and Schwer (1998) note, the results might be sensitive tothe level of data aggregation. The use of a two-digit SIC aggregationrather than a four-digit SIC aggregation might result in a smallerindustry mix effect; however, the absence of data precludes explor-ing this possibility. To see if our results were sensitive to the level ofgeographic aggregation, we recalculated the model using 20 foreigndestinations with the same 20 manufacturing industries. Theresults are virtually identical to those reported in Table 8.


R E V I E W

Gazel-Schwer Shift-Share Results*

State Net relative change Industry mix effect Competitive effect Destination effect

Alabama 19.5 –3.2 25.9 –3.2Alaska –86.3 –29.0 –35.4 –21.9Arizona 53.3 18.5 24.9 9.9Arkansas 59.2 –8.9 68.4 –0.3California 4.1 6.2 1.6 –3.6Colorado 20.0 1.8 26.7 –8.6Connecticut –7.1 3.0 –5.3 –4.9Delaware –14.7 –0.8 –21.1 7.2District of Columbia –40.3 8.9 –39.5 –9.6Florida –3.2 1.2 –12.1 7.7Georgia 43.3 0.5 45.2 –2.3Hawaii –25.7 –9.8 1.7 –17.7Idaho 11.1 4.0 13.5 –6.4Illinois 19.6 0.8 21.3 –2.6Indiana 22.7 –0.4 23.7 –0.6Iowa 5.9 –3.9 12.5 –2.7Kansas 4.9 –7.8 17.6 –4.8Kentucky 44.1 –4.4 51.1 –2.7Louisiana –33.9 –19.9 –11.5 –2.4Maine –0.1 –5.3 7.5 –2.3Maryland –6.5 1.9 –7.1 –1.3Massachusetts –22.9 1.8 –15.4 –9.3Michigan –35.0 1.9 –46.3 9.4Minnesota –15.3 –0.6 –8.3 –6.4Mississippi –17.7 –8.8 –10.1 1.2Missouri 5.2 –1.0 4.4 1.9Montana –39.0 –16.1 –16.0 –6.9Nebraska 9.4 –8.4 21.8 –3.9Nevada 34.7 –1.8 44.1 –7.6New Hampshire –19.2 5.4 –18.1 –6.5New Jersey –8.4 –1.4 –3.8 –3.2New Mexico 420.9 –5.1 420.4 5.6New York –31.2 3.2 –31.1 –3.3North Carolina 18.3 –0.6 23.6 –4.7North Dakota 46.4 –7.1 43.4 10.2Ohio 1.7 0.1 2.2 –0.6Oklahoma –9.2 –2.1 –7.1 –0.0Oregon 16.8 –17.2 45.2 –11.1Pennsylvania –0.8 3.5 –4.6 0.4Rhode Island 4.4 6.4 0.3 –2.3South Carolina 28.5 –0.1 32.8 –4.2South Dakota 129.0 –2.3 133.1 –2.1Tennessee 52.0 –0.2 52.5 –0.3Texas 17.1 –0.5 0.6 16.9Utah 61.7 12.2 50.2 –0.6Vermont 62.7 23.1 36.3 3.3Virginia –6.8 –7.0 7.1 –6.9Washington 9.9 –9.5 30.1 –10.7West Virginia –33.2 –12.7 –17.8 –2.8Wisconsin 7.4 –0.8 11.0 –2.8Wyoming –0.4 –17.7 22.7 –5.4

* The largest effect for each state is shown in bold.

Table 8



change and the industry mix effect, we calculatedsome simple correlation coefficients between netrelative change and each of the shift-share com-ponents. Not surprisingly, a large positive and sta-tistically significant correlation existed betweennet relative change and the competitive effect(0.87). Albeit positive, the correlation between netrelative change and the destination effect (0.20)was not statistically significant. No statistically sig-nificant correlation between net relative changeand the industry mix effect was found either, and,in fact, a weak negative correlation existed (–0.05).

The preceding results are similar to those gen-erated by Gazel and Schwer using state exportdata for 1989-92; however, noteworthy differencesexist. As with our results, they found that thecompetitive effect tended to be the most impor-tant factor in accounting for a state’s net relativechange. Using the absolute values of the individualeffects, their results revealed that the competitiveeffect was the largest effect for 39 of the 51 states,whereas the industry mix and destination effectswere the largest effects for 5 and 7 states, respec-tively. Furthermore, for the 35 states with fasterexport growth than the nation as a whole, theircompetitive effect was positive for 32 of them.

Meanwhile, somewhat contrary to our find-ings, their industry mix effect was positive for 21of these states and the foreign destination effectwas positive for 17 states. For the states with slow-er export growth than the nation as a whole, thesigns of the shift-share components were moresimilar than those for the states with relatively fastexport growth. For these 16 states, the competitiveeffect was negative for 12 states, whereas theindustry mix and destination effects were negativefor 10 states (not all of which were the samestates). Finally, contrary to our findings, theyfound that both the industry mix and destinationeffects were correlated positively to a statisticallysignificant extent with net relative change. Thedifferences in their results and ours are due pri-marily to the dissimilarity in industry coverage.Our study is limited to manufacturing industries,whereas Gazel and Schwer include agriculturaland natural resources.

CONCLUSION

State export performance over 1988-98 showsmuch variation. To account for this variation, thepresent study used both a classic shift-shareanalysis and an extension proposed by Gazel and

Schwer. In the classic shift-share analysis, differ-ences in state export growth relative to the nationare accounted for by differences in industry mixand competitive effects. The former is positive(negative) if a state’s exports were relatively moreconcentrated in industries whose exports expand-ed faster (slower) than the national average. Thelatter effect captures differences accounted for bydifferences in industry mix. The Gazel and Schwerextension incorporates a destination effect. It ispositive (negative) if a state’s exports were relative-ly more concentrated in foreign markets whosepurchases from the U.S. expanded faster (slower)than the overall national increase in exports. Thecompetitive effect, now captures differences unac-counted for by the industry mix and destinationeffects. It is important to stress that these shift-share models are accounting identities and are noteconomic explanations of, in this case, relativestate export growth.

Regardless of the shift-share formula, thecompetitive effect is the key determinant ofwhether a state’s exports grew more or less rapid-ly than the national average. Increased knowledgeof the factors determining this effect is essentialfor understanding the relative export performanceacross states. Prior research suggests one possibleeconomic explanation for this result: that thosestates with larger increases in human capital perworker have seen their industries outperform thecorresponding national industries in terms ofexport growth.

Generally speaking, the destination and indus-try mix effects were equally important but notnecessarily in the ways one might expect. Forexample, for those states whose exports grewmore rapidly than the national average, both theindustry mix and foreign destination effects werenegative. For those states with slower exportgrowth rates than the national average, the indus-try mix and destination effects were negative, asexpected. Overall, the destination effect was corre-lated positively with net relative change; however,this relationship was not statistically significant.Consequently, the results associated with the for-eign destination effect, while enriching the shift-share formula, reveal that this effect is, at most, asmall piece of the puzzle for understanding therelative manufacturing export performance acrossstates between 1988 and 1998.

Looking forward, because the industry distri-bution of most states became more similar to thenation’s between 1988 and 1998, the industry mix


R E V I E W

effect is likely to become less important inaccounting for state manufacturing export growth.Because the export similarity on a geographicbasis has changed little, the importance of the for-eign destination effect is likely to increase inimportance relative to the industry mix effect.

REFERENCES

Coughlin, Cletus C. and Pollard, Patricia S. “State Exportsand the Asian Crisis.” Federal Reserve Bank of St. LouisReview, January/February 2000, 82(1), pp. 3-14.

Coughlin, Cletus C. and Mandelbaum, Thomas B.“Accounting for Changes in Manufactured Exports at theState Level: 1976-86.” Federal Reserve Bank of St. LouisReview, September/October 1990, 71(5), pp. 3-14.

______. “Measuring State Exports: Is There a Better Way?”Federal Reserve Bank of St. Louis Review, July/August1991, 72(4), pp. 65-79.

Cronovich, Ron and Gazel, Ricardo C. “Do Exchange Ratesand Foreign Incomes Matter for Exports at the StateLevel?” Journal of Regional Science, November 1998, pp.639-57.

______. “How Reliable Are the MISER Foreign Trade Data?”Unpublished manuscript, May 1999.

Erickson, Rodney A. and Hayward, David J. “TheInternational Flows of Industrial Exports from U.S.Regions.” Annals of the Association of AmericanGeographers, 1991, pp. 371-90.

Finger, J. Michael and Kreinin, Mordechai E. “A Measure of‘Export Similarity’and Its Possible Uses.” EconomicJournal, December 1979, pp. 905-12.

Gazel, Ricardo and Schwer, R. Keith. “Growth ofInternational Exports among the States: Can a ModifiedShift-Share Analysis Explain It?” International RegionalScience Review, 1998, 21(2), pp. 185-204.

Loveridge, Scott and Selting, Anne C. “A Review andComparison of Shift-Share Identities.” InternationalRegional Science Review, 1998, 21(1), pp. 37-58.

Pollard, Patricia S. and Coughlin, Cletus C. “Going Down:The Asian Crisis and U.S. Exports.” Federal Reserve Bankof St. Louis Review, March/April 1999, 81(2), pp. 33-45.



Membership Structure,Competition, andOccupational CreditUnion Deposit RatesWilliam R. Emmons andFrank A. Schmid

How do credit unions set deposit rates? Aswe will show, the answer to this questiondepends on (i) who actually makes busi-

ness decisions in credit unions (who is in control),and (ii) whether local deposit market conditionsare important. Although members legally controlall credit unions, an occupational credit unionrequires a sponsor (i.e., an employer) that couldwithdraw its support from the credit union orapply pressure on members if the sponsor’s inter-ests were not being served. Thus, the question ofwho effectively controls an occupational creditunion is an empirical question.

If members effectively control an occupationalcredit union, then the array of services offeredand the pricing of these services may be skewedtoward the interests of a subset of members. Thisis because, in a one-member, one-vote governancesystem, there is a strong “winner-takes-all” incen-tive to gain (and exploit) control. Conversely, if thecredit union’s sponsor (i.e., the employer) controlsthe institution, then there is likely to be a morebalanced distribution of surplus to all members.This is because employers sponsor occupationalcredit unions to maximize the total surplus thecredit union creates for all employees rather thanto maximize the surplus for a particular group ofemployees. Finally, if local deposit-market compe-tition tends to constrain rate-setting behavior, thenthe credit union’s membership structure shouldn’tmatter at all in the determination of deposit rates.

This paper explores the member-control, thesponsor-control, and the market-control hypothe-ses and then tests a key prediction made by each.The test is motivated by our theoretical examina-

tion of the pricing of deposits that would beexpected under each of the three regimes.Because we present a simple model in whichthere is no risk, we restrict our empirical analysisto deposit rates. An examination of loan rates, byway of contrast, would require controls for theriskiness of individual borrowers and is rife withcomplications that would only obscure our simpleobjective. Furthermore, we assume that each cred-it union member is either a borrower or a deposi-tor but not both (again, for simplicity).

We hypothesize that, if depositors are in con-trol, they will maximize their own surplus by set-ting high deposit and loan rates, taking intoaccount local market conditions that constraintheir actions.1 We will show that the larger themajority that depositors enjoy, the lower depositrates should be. This is because the number ofmembers that the depositors are able to “exploit”is lower. Thus, we expect a positive relationshipbetween the borrower-to-member ratio anddeposit rates, conditional on depositor control ofthe credit union. We also expect that the loan ratein this regime is independent of the membershipdistribution, corresponding to the familiar text-book monopoly case for constant marginal costs.

If borrowers are in control, on the other hand,we hypothesize that the credit union chooses rela-tively low deposit and loan rates to maximize thesurplus of borrowers, taking due account of localmarket conditions. In this regime, the deposit rateis independent of the number of borrowers in themembership, corresponding to the textbook“monopsony” result (monopoly of the buyer,rather than the seller). Meanwhile, a higher num-ber of borrowers implies a higher loan rate,because the number of “exploitable” members islower.

In summary, the member-control hypothesisimplies that, as we examine the membershipstructure of credit unions in cross-section, depositrates will rise as the number of borrowers in themembership rises as long as the borrowers are inthe minority. Once the majority changes fromdepositors to borrowers, the deposit rate drops

William R. Emmons is an economist and Frank A. Schmid is a sen-ior economist at the Federal Reserve Bank of St. Louis. WilliamBock and Judith Hazen provided research assistance.

1 Technically, members buy “shares” or “share drafts” in creditunions rather than make deposits. Instead of interest on deposits,credit union members receive “dividends.” We use the termsdeposits and interest to avoid any confusion with shares and divi-dends issued by for-profit firms.


R E V I E W

and remains at a low level, independent of howbig the borrower majority is.

The sponsor-control hypothesis of occupation-al credit unions, by way of contrast, presumes thatthe welfare of each member is weighted equally.The only reason why deposit rates would be asso-ciated with the membership structure in a givencredit union would be if deposits were an impor-tant marginal source of loanable funds. That is, byvirtue of having a high number of borrowers, acredit union would have high loan demand. Thishigher demand would be met by attracting addi-tional deposit funds from members (existing ornew). To do this, a credit union would need tooffer higher deposit rates. Thus, when examiningthe cross-section of occupational credit unions,deposit rates would be higher for credit unionswith a high proportion of borrowers in the mem-bership.

Finally, the market-control hypothesisdescribes a situation in which local deposit-market competition constrains credit union rate-setting behavior completely. In other words, nosingle credit union could set its own deposit ratesbelow local deposit rates without suffering adeposit outflow. Likewise, it could not set ratesabove the local market without attracting a hugeand unusable deposit inflow (limited by its poten-tial membership, of course). The important pointfor our purposes is that local market competition,rather than the credit union’s number of borrow-ers or lenders in the membership, would dictaterate-setting behavior.

We analyze a large sample of occupationalcredit unions that reported financial results at theend of 1997. We characterize the membership dis-tribution by calculating the ratio of the number ofloans to the number of members. Our assumptionis that this ratio can serve as a proxy for the num-ber of borrowers in the membership. Our samplecontains some occupational credit unions with rel-atively low loan-to-member ratios, which is con-sistent with depositor majorities. It also containsobservations with loan-to-member ratios so highthat borrower majorities can be assumed. We findthat observed deposit rates are positively associat-ed with the loan-to-member ratio over virtuallythe entire range of loan-to-member ratios. Thisresult is consistent with the sponsor-controlhypothesis of occupational credit unions but isinconsistent with both the member-control andthe market-control hypotheses.

BACKGROUND AND RESEARCH ONOCCUPATIONAL CREDIT UNIONS

Credit unions numbered 11,392 at year-end1996, serving some 70 million individual mem-bers (U.S. Treasury, 1997, p. 15). Credit-unionassets were $327 billion, compared with $5,606billion held by commercial banks and thrifts (U.S.Treasury, 1997, p. 21). Of the 7,068 federally char-tered institutions at year-end 1996, about threefourths were occupational credit unions (U.S.Treasury, 1997, p. 19).2 One or more firms spon-sor each occupational credit union, sometimesproviding office space, paid time off for volunteerworkers, and other forms of support. The remain-ing federal credit unions are as follows: (i) single-group credit unions of the associational or com-munity type, or (ii) multiple-group credit unionswith predominantly associational, community, ormore than one type of membership (i.e., severaltypes of membership groups that span the usualclassifications). According to a credit union surveyin 1987, 79 percent of all Americans who were eli-gible to join a credit union had done so (AmericanBankers Association, 1989, p. 29). Given theprominent role of occupational credit unions, amajority of members are in the prime workingages of 25 to 44 (American Bankers Association,1989, p. 30).

Most members of occupational credit unionseasily could (and often do) obtain financial servic-es from a for-profit financial intermediary such asa commercial bank or a thrift institution. Whythen do so many employers sponsor creditunions? Hansmann (1996) suggests that occupa-tional credit unions continue to thrive todaybecause employers (sponsors) benefit from them:

Employers can also benefit from having acredit union for their employees. Thecredit union ties the employees moretightly to the employer, improves theemployees’ financial situation (and conse-quently their effective wage), and helpskeep the employees out of financial diffi-culties that may interfere with their workor create bother for the employer (such asgarnishment of wages). For these reasons

2 We concentrate on federally chartered credit unions because theNational Credit Union Association does not vouch for the accuracyof data provided by state-chartered credit unions, which reportdirectly to their state's regulatory authorities.



employers have often helped promote theformation of credit unions, for example, byproviding free office space and free timeoff to the employees who administerthem. (Hansmann, 1996, pp. 259-60)

Most credit union sponsors are in non-financial businesses, so they may not be well-suited—or particularly eager—to operate a finan-cial institution. At the same time, credit unions arelegally governed by their members on a one-member, one-vote basis. Do the members there-fore control occupational credit unions? Or dolocal market conditions dictate many of the busi-ness decisions an occupational credit union mustmake, such as setting deposit rates?

The possibility of member control is importantbecause there may be (roughly speaking) two dis-tinct types of members: those whose primary rea-son for joining the credit union is to borrowmoney and those whose primary reason for join-ing is to save in the form of insured deposits. Ifmembers (or member coalitions) effectively con-trol occupational credit unions, then the relativesize of the two groups of members determineshow deposit and loan terms are set. Gaining effec-tive control of the credit union would allow onegroup or the other to skew the credit union’sterms in their own favor.

The question of who controls cooperativefinancial institutions, such as occupational creditunions, is an old one (Taylor, 1971; Flannery, 1974;Smith, Cargill, and Meyer, 1981; Smith, 1984; Hartand Moore, 1996, 1998; Emmons and Mueller,1997; Emmons and Schmid, 1999a, 2000b). Moststudies of this type assume that one group ofmembers, such as depositors or borrowers, con-trols the cooperative firm. Empirical evidenceabout credit unions is mixed; some studies finddepositor domination and others find borrowerdomination. Emmons and Schmid (1999a) offer anexception to the usual approach that memberscontrol the cooperative financial institution. Theyassume the sponsor exercises control. This studyrevisits this question in an attempt to find evi-dence consistent with the member-control, thesponsor-control, or the market-control hypothesis.

A MODEL OF CONTROL INOCCUPATIONAL CREDIT UNIONS

We examine a simple credit union that gener-ates surplus for its members by presenting oppor-tunities to borrow and to deposit funds in an

insured account. The supply and demand func-tions are reduced-form equations that result fromhousehold optimization subject to the localdeposit-market conditions. To keep the model sim-ple, we do not explicitly model these householddecisions but instead postulate the existence ofdemand and supply functions.

We present three (mutually exclusive)hypotheses regarding effective control of an occu-pational credit union: (i) the member-controlhypothesis, under which a controlling group ofmembers sets deposit and loan rates to maximizeits own surplus, (ii) the sponsor-control hypothe-sis, under which the sponsoring firm sets depositand loan rates to maximize the total surplus for allmembers, and (iii) the market-control hypothesis,under which local deposit-market competition dic-tates the rates every credit union sets.

In our model, all potential credit-union mem-bers are identical except that some are borrowerswhile others are depositors.3 We assume that thenumber of loans is proportional to the number ofborrowers, and likewise for depositors. Morespecifically, we assume that each borrowing mem-ber wants one loan and responds to changes inthe loan rate by adjusting the size of the loan.4The aggregated demand curve for loans and theaggregated supply curve of deposits are obtainedby adding up the individual quantities for eachprice (loan rate and deposit rate, respectively) foreach credit union. We analyze aggregated demandand supply curves to derive hypotheses aboutoptimal rate-setting under the three controlregimes analyzed here.

We assume the credit union operates under azero-profit constraint, given its not-for-profit char-acter. The credit union can lend or borrow in theinterbank market at the same rate, r. For simplici-ty, we assume that the demand for loans and thesupply of deposits are both linear in the borrowingand the deposit rates, respectively. In indirectform, the loan-demand and deposit-supply sched-ules facing the credit union are as follows:

3 Assuming identical potential members implies that all potentialmembers are also actual members (or the credit union would nothave formed). See Emmons and Schmid (1999b, 2000a) for modelswith endogenous membership decisions.

4 The specific proportionality assumption—that each borrower takesout one loan—allows us to use the ratio of loans to members as aproxy for the number of borrowers in the credit union’s member-ship in the empirical section of the article.

(1a)

(1b)

where p and q denote the loan rate and thedeposit rate, respectively; x and y are the dollaramounts of loans made and deposits accepted bythe bank, respectively; and a, b, c, and d are fixedparameters reflecting member preferences. In par-ticular, we assume that the maximum loan ratethe borrowers are willing to pay, a, exceeds the

interbank rate, r. Likewise, the minimum depositrate the depositors are willing to accept, c, isbelow the interbank rate, r. The parameters b andd reflect the sensitivities of the loan demand anddeposit supply to the credit union’s deposit andloan rates, respectively. For simplicity, we assumethe credit union has no operating costs. This doesnot affect the results qualitatively and simplifiesthe analysis.

Sponsor ControlSponsor control implies that the objective of

an occupational credit union is to maximize thesurplus enjoyed by all members. Figure 1 showsthat a sponsor-controlled credit union optimallyequates the deposit and loan rates at the level dic-tated by the intersection of members’ demand andsupply curves. This result follows from the basicprinciples of welfare economics. The surplus ofborrowers is the area under the demand curve andabove the optimal loan rate, p0 (triangle ECD).Intuitively, the surplus equals the differencebetween the borrowers’ aggregate willingness topay for loans (indicated by the demand curve) andthe amount that they actually pay. Similarly, thesurplus of the depositors is equal to the differencebetween the amount of interest they receive andthe amount of interest they are willing to accept,which is reflected in the deposit supply curve.This surplus is the area between the supply curveand the optimal deposit rate, q0 (triangle BCE).

Figure 2 compares two credit unions with thesame number of depositors (the same deposit sup-ply curve) but with different numbers of borrow-ers. The credit union with the higher number ofborrowers (and thus the higher borrower-to-member ratio) has a deposit and loan rate equal toq1, compared with the lower value q0, which is therate of the credit union with the smaller borrower-to-member ratio.5 This simple analysis leads toour first hypothesis:

Hypothesis 1: Sponsor Control. The depositrate strictly increases as the number of memberswho are borrowers increases.

A later section discusses our strategy for test-ing this hypothesis.

q y c dy c d( ) , , ,= + >0

p x a bx a b( ) , ,= − >0

Sponsor Control: Credit Union MaximizesTotal Surplus

D

E

B

p,q

q0

=p0

x,y

C

y(q)

x(p)

A 0

Figure 1

Sponsor Control: Number of Borrowersand Deposit Rates

D

E

B

p,q

q0

=π0

A 0 x,y

C

y(q)

C’

x(p)

E’q1

=π1

Figure 2


R E V I E W

5 The demand curve rotates upward—rather than tracing out a paral-lel upward shift—because each individual borrower’s willingness topay is identical and the credit union’s aggregate demand curve rep-resents a horizontal summation of individual loan demands.



Member ControlMember control of a cooperative firm

inevitably leads to conflicts of interest if there ismore than one type of member. Control ofdecision-making determines how much surplusthe credit union will generate and how this sur-plus will be distributed among the members. Thekey difference between a member-controlledcredit union and one that is controlled by a spon-sor is that, in the former case, the dominantmember group will act like a monopoly withrespect to minority members. The latter will actlike a “benevolent dictator,” resulting in the com-petitive market outcome.

Depositor Control. If depositors are in controlthey will maximize their surplus by solving the fol-lowing optimization problem:

(2a)

subject to

(2b)

(2c)

The quantity to be maximized is the areaabove the deposit-supply curve, which is deter-mined by the choice of the deposit rate, q.Constraint (2b) sets out the credit union’s zero-profit constraint and reflects the fact that inter-bank borrowing or lending is used to offset imbal-ances in the amount of loanable funds demandedand supplied by members at the chosen interestrates. Constraint (2c) excludes profitable roundtriptransactions in which members finance high-ratedeposits by borrowing low-rate funds from thecredit union.

Solving the maximization problem shownabove, the optimal deposit rate, q*, and the opti-mal loan rate, p*, are as follows6:

(3a)

(3b)

Equation (3a) shows that, if we compare twocredit unions that differ in the number of borrow-ers (i.e., in the demand parameter b) but are other-wise comparable, the credit union with the highernumber of borrowers (lower value of b) has the

higher deposit rate, q*. This reflects the fact thatwith a higher borrower-to-member ratio there aremore minority members who can be “exploited”in terms of high loan rates, which, in turn, allowsthe credit union to pay higher deposit rates. Adepositor-controlled credit union chooses themonopoly loan rate, p*, and pays q* on deposits,producing surplus for depositors. The credit unionhas excess deposits, in the amount y (q*) – x(p*),which are lent in the interbank market. In effect,the monopoly profit obtained through lending theamount x(p*) is distributed to depositors in theform of deposit rates that are higher than inter-bank rates. Note that the loan rate changes inresponse to changes in the membership structureonly if control of the credit union also switchesfrom depositors to borrowers.

Borrower Control. Borrowers will control thecredit union if they constitute a large enough frac-tion of the membership. Under borrower control,the credit union’s maximization program is thefollowing:(4)

subject to

(2b)

(2c)

In this case, the quantity to be maximized isthe area below the loan demand curve, which isdetermined by the choice of the loan rate, p. Theoptimal deposit rate, q*, and the optimal loan rate,p*, are as follows7:

(5a)

(5b) pa r a r

bd

r c*

( ) ( ).=

+ − − + − 2 2

2

qc r* = +

2

p q≥ .

0 = ⋅ − ⋅ + ⋅ −[ ]p x p q y q r y q x p( ) ( ) ( ) ( )

Maxp q

a px p

,( )

−2

pa r* .= +

2

qc r c r

db

a r*

( ) ( )=

+ + − + − 2 2

2

p q≥ .

0 = ⋅ − ⋅ + ⋅ −[ ]p x p q y q r y q x p( ) ( ) ( ) ( )

Maxp q

q cy q

2,( )

−

6 The second solution to the deposit is

q*= (c+r– [(c–r)2+d

b (a–r ) 2 ]12 )/2.

The solution represents a minimum in the depositors’ surplus. Weassume the solution is interior (constraint (2c) is not binding).

7 The second solution to the loan rate is

p*= (a+r+ [(a–r)2+ b

d (r–c) 2 ]12 )/ 2.

The solution represents a minimum in the borrowers’ surplus.Again, we assume p < q.

In a way analogous to the case consideredabove in which the loan rate was invariant, thedeposit rate is unaffected by changes in the num-ber of borrowers (represented by variations in theloan demand parameter, b). Holding all else equal,the optimal loan rate increases with the number ofborrowers because progressively more borrowersmust share the surplus generated by low-priceddeposits. A borrower-controlled credit union sets arelatively low lending rate, p*, and attracts moreloan demand, x(p*), than there is deposit supply,y(q*). The shortfall in deposit funding is met byborrowing in the interbank market at rate r. Ineffect, the monopsony profits earned by thecredit union on deposits collected at rate q* aredistributed to borrowers in the form of a below-interbank market lending rate of p*. In this case itis the deposit rate that is insensitive to smallchanges in the membership structure as long asthese changes do not transfer control away fromborrowers.

Implications of Member Control. When com-bining the two variations on the theme of membercontrol, two features are noteworthy. First, thedeposit rate is greater throughout the depositor-dominated regime than it is anywhere in theborrower-dominated regime. This makes intuitivesense because depositors who exercise controlover the credit union will capitalize on this factand pay themselves a higher deposit rate. Second,

the fact that control must shift from depositors toborrowers at some point as the percentage of bor-rowers in the membership increases implies thatthere is a discontinuity in the schedule of optimaldeposit rates. Figure 3 displays these two features,where β is the fraction of members who are bor-rowers. The results of our analysis of the case ofmember control lead us to the following hypothesis:

Hypothesis 2: Member Control. The depositrate strictly increases as the number of borrowersincreases if a credit union is under depositor control.The deposit rate is constant and takes on its globalminimum value if a credit union is under borrowercontrol.

Market Control

Thus far, we have focused only on the internalcontrol features of occupational credit unions. It isclear, however, that external factors may constrainthe ability of a controlling group of credit unionmembers to extract surplus. If price competition isstrong in local deposit or loan markets, minoritymembers may threaten to leave the credit union toimprove the terms they receive. This leads to ourthird testable hypothesis:

Hypothesis 3: Market Control. Deposit rates areindependent of credit-union membership structure(the fraction of members who are borrowers).

EMPIRICAL EVIDENCE

We use the loan-to-member ratio as a proxyfor the borrower-to-member ratio. Our sampleincludes occupational credit unions withloan-to-member ratios so low that depositorsalmost certainly are in the majority. The samplealso contains credit unions in which the loan-to-member ratio is so high that a borrower majori-ty appears inevitable. Given the wide range ofmembership structures that we observe and ourmodel’s predictions, a deposit rate that strictlyincreases as the loan-to-member ratio increaseswould support Hypothesis 1, namely, sponsor con-trol of occupational credit unions. A deposit ratethat initially increases, drops discontinuously, andthen remains constant as the loan-to-member ratioincreases would support Hypothesis 2, namely,member control. A pattern of deposit rates that iscompletely unrelated to the loan-to-member ratiowould support Hypothesis 3, namely, market con-trol. Note that these qualitative predictions aboutthe dataset do not require us to identify whichcredit unions are depositor-controlled and whichare borrower-controlled (if any).

Member Control: Loan-to-Member Ratioand Deposit Rates

q*

0

Lender-dominated

0.5 1

r

Borrower-dominated


R E V I E W

Figure 3



We use a semi-parametric estimation methodto address the possibility that the deposit rate isnonlinear in the loan-to-member ratio (as indeedit appears to be). This ratio is included in thenonparametric part of the regression along with aconstant regressor, whereas the parametric part ofthe regression contains a set of normalizingregressors. The normalizing regressors are fourzero-one indicator variables that represent thetype of membership (TOM) categories shown inTable 1, with educational credit unions serving asthe numeraire unit (the excluded category). Thenonparametric methodology allows the relation-ship between the deposit rate and the loan-to-member ratio to take on a (smooth) functionalform. The assumption of a smooth relationshipbetween the loan-to-member ratio and the depositrate is not restrictive if the relationship betweenthe loan-to-member ratio and the borrower-to-depositor ratio is stochastic. For example, thenumber of loans per borrower might not be thesame for all credit unions in the sample. Fordetails on the econometric methodology, see theAppendix.

We examine a subset of all federally charteredand federally insured occupational credit unionsthat reported financial information at the end of1997 (see the Appendix for details on construction

of the dataset and the variables used). Table 1 pro-vides a breakdown of our sample according to thetype of membership group characterizing eachcredit union. The Table distinguishes betweencredit unions with a single common bond andthose with multiple common bonds. Overall, 2,211credit unions in our sample had a single commonbond (43 percent of the sample), whereas 2,923credit unions (57 percent of the sample) had mul-tiple common bonds among their membership.

Membership Structure andDeposit Rates

The estimated relationship between depositrates and the loan-to-member ratio in our sampleis displayed in Figure 4 by the solid line. Inessence, the figure depicts the expected depositrate (vertical axis) for an occupational credit unionwith a given loan-to-member ratio (horizontalaxis). Ninety percent confidence intervals for thepoint estimates are shown as dashed lines.8

Table 1Distribution of Credit Unions by Type of Membership

Descriptive Statistics

Minimum Median Mean Maximum Standard deviation

Deposit rate 0.004 0.038 0.038 0.089 0.008Loan-to-member ratio 0.009 0.447 0.471 1.859 0.177Herfindahl index 0.053 0.208 0.208 1 0.095

Table 2

Table 1Distribution of Credit Unions by Type of Membership

Distribution of Credit Unions by Type of Membership

Type of membership (TOM)

EducationalMilitaryFederal, state, local governmentManufacturingServicesTotals

* National Credit Union Association (NCUA), Instruction No. 6010.2, July 28, 1995.

TOM codes *Multiple-

Single-group group creditcredit unions unions

4 345 356 36

10-15 40-4920-23 50-53

Number of credit unionsMultiple-

Single-group group creditcredit unions unions

327 49439 130417 649843 871585 779

2,211 2,923

Table 1

8 For econometric reasons we discarded 627 observations with zerovalues for the deposit rates. We also discarded 55 observations ofloan-to-member ratios greater than 2 because these seemed to beaberrations or reporting errors. For instance, one credit unionreported a ratio of loans to members that equaled 365.


R E V I E W

Figure 4 indicates a positive associationbetween the loan-to-member ratio and creditunion deposit rates over the bulk of observedloan-to-member ratios. The intercept is not identi-fied in regressions of this type, so only verticaldistances in the figure are meaningful (not thelevel itself). The tight confidence intervals over thecentral region in the figure (where the overwhelm-ing number of credit-union observations lie) implythat the slope of the relationship is reliably posi-tive. Confidence intervals widen dramatically andthe point estimates appear erratic at extreme val-ues of the loan-to-member ratio because there arevery few observations in those ranges. Figure 4therefore provides visual evidence to support thesponsor-control hypothesis of occupational credit

unions. That is, it does not appear that membercontrol switches from depositors to borrowers asthe loan-to-member ratio increases in thecross-section of credit unions. Nor does it appearthat loan-to-member ratios and deposit rates arecompletely unrelated, as the market-controlhypothesis requires. Instead, the upward-slopingrelationship is consistent with the sponsor-controlregime where the membership structure anddeposit rates are related because deposits are animportant marginal source of funding for loans.9

Table 3 displays the coefficients from the para-metric part of our semi-parametric regressionequation. As mentioned above, the regressors arezero-one indicator variables that represent thetype of membership with educational creditunions serving as the numeraire unit. The signifi-cant coefficients on TOM codes indicate that thereare variations in the ability or willingness of occu-pational credit unions to pay higher deposit ratesamong different types of sponsors.

CONCLUSIONS

This article seeks to answer the question ofhow occupational credit unions set deposit rates.We investigate three potential control regimesunder which occupational credit union depositrates might be set, namely, member control, spon-sor control, and market control.

If members control occupational creditunions, we would expect a positive relationshipbetween the extent of loan demand and thedeposit rate, conditional on depositor control ofthe credit union. If borrowers are in control, how-ever, we expect the deposit rate to take on a valuethat is lower than what we would observe in thedepositor-dominated regime; also, conditional onborrower control, we expect the deposit rate to beindependent of the membership structure. Thus,the member-control hypothesis implies that, in across-sectional snapshot of many different creditunions, the deposit rate would be higher whenev-er the borrower-to-member ratio is higher so longas depositors are in control. For sufficiently highvalues of the borrower-to-member ratio, however,borrowers would be in the majority. For these

Partial impact on deposit rates

-3.8

-3.6

-3.4

-3.2

-3.0

-2.8

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00

Estimated relationship between loan-to-member ratio and deposit rates

Limits of a 90 percent confidence interval

Loan-to-member ratio

Impact of Loan-to-Member Ratio onDeposit Rates

Figure 4

9 We experimented with other empirical specifications including onein which a county-level Herfindahl index was used as a condition-ing variable to represent the intensity of deposit-market competi-tion. Details on the construction of this variable are in theAppendix. This variable was never significant, however, so we donot report results from models in which it was used.

Parametric Variables from theSemi-Parametric RegressionType of membership(TOM) codes Coefficient t-Statistic

Military 2.385 × 10-2 1.795*Government 2.087 × 10-2 2.239**Manufacturing 4.560 × 10-2 5.053***Services 4.178 × 10-2 4.480***Number of 5,134observations*/**/*** Significant at 10/5/1 percent levels (two-tailed tests).

Table 3



credit unions, the deposit rate would take on itsminimum value in the sample and would be com-pletely independent of the membership structure.

The sponsor-control hypothesis of occupation-al credit unions, by way of contrast, predicts thatdeposit rates would be higher, the greater the pro-portion of borrowers in the membership. This isbecause deposits are an important marginalsource of funds for meeting loan demand, andhigher deposit rates are necessary to attract addi-tional loanable funds.

Finally, the market-control hypothesis suggeststhat occupational credit union deposit rates andthe loan-to-member ratios should be completelyunrelated. Local competition would dictate depositrates and there would be no statistical relationshipbetween membership structure and deposit rateswhatsoever.

Using data from a large sample of occupation-al credit unions observed at the end of 1997, wefind that deposit rates are positively associatedwith a measure of the credit union’s loan demandthroughout most of the range of observedloan-to-member ratios. This finding is consistentwith the sponsor-control hypothesis of occupa-tional credit unions but not with the member-control or the market-control hypothesis. Thisresult supports the theoretical approach taken inEmmons and Schmid (1999a), where effectivesponsor control of occupational credit unions wasassumed.

Thus, it appears that Hansmann’s (1996) sug-gestion—that employers sponsor and operateoccupational credit unions to provide surplus toall of their employees—is a better description inpractice than the view that they are subject to cap-ture by a subset of members or that local competi-tion dictates credit union rates completely.

REFERENCES

American Bankers Association. The Credit Union Industry:Trends, Structure, and Competitiveness, 1989.

Emmons, William R. and Mueller, Willy. “Conflict ofInterest Between Borrowers and Lenders in CreditCooperatives: The Case of German Cooperative Banks.”Working Paper, Federal Reserve Bank of St. Louis, 1997.

__________ and Schmid, Frank A. “Wages and Risk-Takingin Occupational Credit Unions: Theory and Evidence.”Federal Reserve Bank of St. Louis Review, March/April1999a, 81(2), pp. 13-31.

__________ and __________. “Credit Unions and theCommon Bond.” Federal Reserve Bank of St. LouisReview, September/October 1999b, 81(5), pp. 41-64.

__________ and __________. “Bank Competition andConcentration: Do Credit Unions Matter?” FederalReserve Bank of St. Louis Review, May/June 2000a, 82(3),pp. 29-42.

__________ and __________. “Pricing and Dividend Policiesin Open Credit Cooperatives.” Working Paper 2000-008A,Federal Reserve Bank of St. Louis, March 2000b.

Flannery, Mark J. “An Economic Evaluation of CreditUnions in the United States.” Research Report No. 54,Federal Reserve Bank of Boston, 1974.

Hansmann, Henry. The Ownership of Enterprise.Cambridge, MA: Harvard University Press, 1996.

Hart, Oliver and Moore, John. “Cooperatives vs. OutsideOwnership.” Working Paper, Harvard University, 1998.

__________ and __________. “The Governance ofExchanges: Members’ Cooperatives versus OutsideOwnership.” Oxford Review of Economic Policy, Winter1996, pp. 53-69.

Smith, Donald J. “A Theoretic Framework for the Analysisof Credit Union Decision Making.” Journal of Finance,September 1984, 39(4), pp. 1155-68.

__________; Cargill, Thomas F. and Meyer, Robert A. “AnEconomic Theory of a Credit Union.” Journal of Finance,May 1981, pp. 519-28.

Taylor, Ryland A. “The Credit Union as a CooperativeInstitution.” Review of Social Economy, September 1971,pp. 207-17.

U.S. Treasury Department. Credit Unions. Washington, DC:U.S. Government Printing Office, 1997.

DATASET AND VARIABLESThe Dataset

We analyze a dataset comprising all federallychartered and federally insured credit unions atthe end of 1997. The dataset was obtained fromthe Report of Condition and Income for CreditUnions (NCUA 5300, 5300S) and produced by the

Appendix


R E V I E W

National Credit Union Administration (NCUA).These reports are issued semi-annually in Juneand December. We used the December data. Theflows in the December income statements includethe entire year of 1997.

We concentrate on the following types ofmembership (TOM) groups among occupationallybased credit unions: educational; military; federal,state, and local government; manufacturing; andservices. Thus, we do not include communitycredit unions, associational credit unions, or cor-porate credit unions. Lists of TOM classificationcodes are from the NCUA (Instruction No. 6010.2,July 28, 1995).

We excluded observations for any of the fol-lowing reasons:

• Missing TOM codes.• Activity codes other than “active.”• Number of members or potential mem-

bers not greater than one (applies to actualand to lagged values).

• Number of loans to members equal to zero(applies to actual and to lagged values).

• Ratio of lagged number of loans to lagged number of members greater than two.

• Nonpositive values for total assets or lagged total assets.

• Dividend on shares equal to zero.

We calculated county-specific Herfindahlindexes as measures of concentration of the localbanking market. A Herfindahl index is defined asthe sum of squared market shares. We measuredmarket shares as the fraction of totalcommercial-bank and Bank Insurance Fund(BIF)-insured thrift deposits (as of June 30) withina county (or independent city) based on FDICSummary of Deposits data. These data are avail-able online at <http://www2.fdic.gov/sod/>. Table2 provides an overview of the dataset.

Definition of Variables

Definitions of variables and underlying datasources are listed below. Relevant item numbersare in brackets for data taken from the Report ofCondition and Income for Credit Unions, producedby the National Credit Union Administration.

Dependent Variable. Deposit rate: Natural log-arithm of the ratio of dividends on shares[CUSA6091] and total shares and deposits[CUSA6091]. We use the natural logarithm becausethe deposit rate has a lower bound at zero.

Independent Variables1. Loan-to-member ratio: Number of loans to

members divided by number of members. Thenumber of loans to members was obtained asthe difference between the number of totalloans [CUSA1262] and the number of loansthat were purchased or extended tonon-members [CUSA1205]. The loan-to-member ratio was lagged by one year.

2. TOM code variables: Equal to one if the creditunion is of a specific type (military, govern-ment, manufacturing, or services). Because weuse an intercept in the nonparametric part ofthe semi-parametric regression, the TOM codevariable for the educational credit union wasdropped.

3. Herfindahl index: Sum of squared marketshares of total commercial-bank andBIF-insured thrift deposits. By definition, theHerfindahl index is greater than zero; its maxi-mum value is one. The Herfindahl index waslagged by one year. Results using theHerfindahl index are not reported in the text;see footnote 7.

Econometric Method

We use a semi-parametric model to allow theinfluence of the loan-to-member ratio on thedependent variable to be nonlinear. The parame-tric part of the model contains zero-one variablesthat indicate the TOM code. In particular, we use asemi-parametric model of a credit union’s partici-pation rate of the form:

where yi is the ith observation of the dependentvariable; xi is a vector consisting of the ith observa-tion of the explanatory variables in the nonpara-metric part of the model, the loan-to-memberratio, and (in unreported versions of the model)the Herfindahl index; xpi is a row vector consistingof the ith observation of the explanatory variablesof the linear (parametric) part of the model; βp is acolumn vector of the parameters of the linear partof the model; and εi is the ith realization of theerror term. For details on this econometricapproach, see the Appendix in Emmons andSchmid (1999a).

y f x x i ni i pi p i= + ⋅ + =( ) , ,..., ,β ε 1

Retail Sweep Programsand Bank Reserves,1994-1999Richard G. Anderson andRobert H. Rasche

In January 1994, the Federal Reserve Board per-mitted a commercial bank to begin using a newtype of computer software that dynamically

reclassifies balances in its customer accounts fromtransaction deposits to a type of personal-savingdeposit, the money market deposit account(MMDA).1 This reclassification reduces the bank’sstatutory required reserves while leavingunchanged its customers’ perceived holdings oftransaction deposits.

The use of deposit-sweeping software spreadslowly between January 1994 and April 1995, butrapidly thereafter. Estimates of the amounts oftransaction deposits reclassified as MMDAs at allU.S. depository institutions, prepared by the Boardof Governors’ staff, are shown in Figure 1.2 By late1999, the amount was approximately $372 billion.In contrast, the aggregate amount of transactiondeposits (demand plus other checkable deposits) inthe published M1 monetary aggregate, as ofDecember 1999, was $599.2 billion.

In this analysis, we interpret the effects ofdeposit-sweeping software on bank balance sheetsto be economically equivalent to a reduction instatutory reserve-requirement ratios. We seek tomeasure the amount by which such deposit-sweeping activity has reduced bank reserves (vaultcash and deposits at Federal Reserve Banks).Currently, transaction deposits are subject to a 10percent statutory reserve-requirement ratio onamounts over the low-reserve tranche ($44.3 mil-lion during 2000, $42.8 million during 2001),whereas personal-saving accounts, includingMMDAs, are subject to a zero ratio.3

To be useful in policy analysis and empiricalstudies, aggregate quantity data on bank reservesmust be adjusted for the effects of changes in

statutory reserve requirements on the quantity ofreserves held by banks.4 In the past, suchadjustments were straightforward because changesin statutory reserve requirements applied simulta-neously and uniformly to groups of depository



1O’Sullivan (1998, p. 88) identifies this bank as First Union NationalBank: “The trend started almost four years ago when First NationalBank applied to the Federal Reserve to do something that becameknown as ‘the reserve sweep’.” First Union’s idea is reminiscent ofthe automatic-transfer-from-savings (ATS) account of the 1970s. Itdiffers because transaction deposits are reclassified as MMDA de-posits, a category first created in 1982 by the Garn–St. GermainAct.

2 These data are updated monthly, with a one-month lag, and are avail-able on this Bank’s Web site at<www.stls.frb.org/research/swdata.html>.

3 The Monetary Control Act of 1980 imposed Federal reserve require-ments on net transaction deposits, which equals the sum of check-able deposits due to individuals, partnerships, corporations (includ-ing other depository institutions), the Treasury, state and local gov-ernments, and government agencies, minus the sum of cash itemsin process of collection and demand deposits due from other depos-itory institutions. So far as we are aware, data on net transactiondeposits have not been published by the Federal Reserve Boardsince implementation of the Monetary Control Act. The statutoryreserve requirements applicable to transaction deposits are tiered,with a zero rate applied to the reserve-exemption amount, a 3 per-cent rate applied to the low-reserve tranche, and since April 1992 a10 percent rate applied to amounts that exceed the tranche. Thereserve-exemption amount and the low-reserve tranche are adjustedeach year using a formula set by law. For 1992-99, the reserve-exemption amounts were $3.6, 3.8, 4.0, 4.2, 4.3, 4.4, 4.7, and 4.9million and the low-reserve tranche amounts were $42.2, 46.8, 51.9,54.0, 52.0, 49.3, 47.8, and 46.5 million. Prior to April 1992, themarginal ratio applicable to transaction deposits was 12 percent.

4 A broad cross-country study illustrating the importance of suchadjustments is McCallum and Hargraves (1995).

Richard G. Anderson is a vice president and economist and RobertH. Rasche is a senior vice president and director of research at theFederal Reserve Bank of St. Louis. Marcela M. Williams providedresearch assistance.

Sweeps of Transaction Depositsinto MMDAs

1994 1995 1996 1997 1998 1999 20000

50

100

150

200

250

300

350

400

450

Board of Governors Staff Estimates, Jan. 1994 - Oct. 2000

Bill

ion

s o

f Do

llars

Figure 1


R E V I E W

institutions within only a small number of broadclasses. The effective date for changes in statutoryrequirements varied slightly among depositoriesthat report data to the Federal Reserve weekly(larger banks), those that report quarterly (smallerbanks), and those that report annually (very smallbanks). Within each group, however, the effectivedate was the same for all institutions. During the1980s, the only changes in statutory requirementswere due to the phase-in and indexationprovisions of the Monetary Control Act. During the1990s, the reserve-requirement ratio applicable tononpersonal savings and time deposits wasreduced from 3 percent to zero (December 1990)and the highest marginal ratio applicable to trans-action deposits was reduced from 12 percent to 10percent (April 1992).5

The economic effect of deposit-sweeping soft-ware is unlike these previous changes. The essenceof deposit-sweeping software is that it permitsbanks to change the share of their transactiondeposits that are subject to a non-zero statutoryreserve-requirement ratio (see the insert “HowDeposit-Sweeping Software Reduces RequiredReserves”). Each bank is free to decide when andhow to implement the software, subject toconstraints discussed below. In this way, in part,banks’ effective reserve requirements are “homebrewed.” As a result, the economic effects ofdeposit-sweeping software must be analyzed andmeasured bank-by-bank.

Our analysis suggests that required and totalreserves in December 1999, measured by thereserve adjustment magnitude (RAM) developed inthis article, were lower by $34.1 billion and $25.8billion, respectively, as a result of deposit-sweeping activity. In addition, many depositoryinstitutions have reduced their required reservesto such an extent that the lower requirementplaces no constraint on the bank because it is lessthan the amount of reserves (vault cash anddeposits at the Federal Reserve) that the bankrequires for its ordinary day-to-day business. Forthese banks, the economic burden of statutoryreserve requirements has been reduced to zero.

DEPOSIT-SWEEPING SOFTWARE,REQUIRED RESERVES, AND RAM

The effectiveness of deposit-sweeping soft-ware hinges on the use of the MMDA. This depositinstrument was created in 1982 by a provision inthe Garn–St. Germain Act. At that time, many

banks perceived extreme competitive pressuresfrom money market mutual funds. The MMDAallowed them to offer a type of deposit that wasfully competitive with money market mutual fundshares. The MMDA was not subject to RegulationQ interest rate controls, and, so long as no morethan six withdrawals were made by check or pre-authorized transfer during a month, it was notsubject to the statutory reserve requirementsapplicable to transactions deposits. (If a bank per-mitted more than six such withdrawals, theMMDA was reservable as a transaction deposit.)The Monetary Control Act specified three cate-gories of deposits subject to reserve requirements:net transaction deposits, savings deposits (person-al and nonpersonal), and time deposits (with aminimum maturity of seven days). The act set thereserve-requirement ratio for personal-savingdeposits to zero, and the Board of Governors setthe ratio for nonpersonal-saving deposits to zeroin December 1990. Because MMDAs are not timedeposits and the Garn–St. Germain Act prohibitedthe Federal Reserve from imposing transaction-deposit reserve requirements, they are classifiedas savings deposits for reserve-requirement pur-poses.6

At its start, deposit-sweeping software createsa “shadow” MMDA deposit for each customeraccount. These MMDAs are not visible to the cus-tomer, that is, the customer can make neitherdeposits to nor withdrawals from the MMDA. Todepositors, it appears as if their transaction-account deposits are unaltered; to the FederalReserve, it appears as if the bank’s level of reserv-able transaction deposits has decreased sharply.Although computer software varies, the objectiveis the same: to minimize a bank’s level of reserv-

5For banks that reported deposit data weekly to the Federal Reserve,the reserve-requirement ratio applicable to nontransaction depositswas reduced from 3.0 percent to 1.5 percent as of December 13,1990, and to zero as of December 27, 1990. For banks that reporteddeposit data quarterly, the ratio was reduced to zero as of January17, 1991. The latter change applied to all banks as of April 2, 1992.

6 Banks have attempted other combinations of transaction and savingdeposits. In one case, a bank suggested that customers maintainseveral MMDA accounts and simply shift all funds among theaccounts as necessary to avoid making more than six third-partypayments (or transfers to other accounts) during any given month(12 CFR 204.133). In another, a bank reclassified transactiondeposits as seven-day large-time deposits, staggering the maturityso as to be able to pay, each day, all checks presented (12 CFR204.134). In these cases, the Board of Governors reclassified thesaving and large-time deposits as transaction deposits and imposedtransaction-deposit reserve requirements. See the Board ofGovernors Regulation D, 12 CFR Chap. 11.



Let us consider a hypothetical $1 billionbank with $200 million in transaction deposits.We focus on two constraints faced by the bank:(i) to satisfy the Federal Reserve’s statutoryreserve requirements and (ii) to convert depositsinto currency and settle interbank debits (relatedto check clearing and wire transfers) usingdeposits at the Federal Reserve. Fortunately, theassets involved—vault cash and deposits at theFederal Reserve—do double duty.

The Bank Before Deposit-Sweeping Software.A bank’s statutory required reserves are calculat-ed from close-of-business data. Excluding anyspecial adjustments, the bank’s required reservesas of late January 1999 would be as follows:

• 0% on the first $4.9 million of transactiondeposits (the reserve-exemption amount);

• 3% on the next $41.6 million of transactiondeposits (up to the low-reserve tranche of $46.5 million), equal to $1.248 million;

• 10% on the amount in excess of $46.5 mil-lion, or $15.350 million.Banking industry data suggest that such a

bank might choose to hold vault cash equal toapproximately 5 percent of its transactiondeposits, or $10 million. If all vault cash is“applied” to satisfy reserve requirements, thebank would need to maintain at least $6.598 mil-lion on deposit at the Federal Reserve to satisfyits statutory reserve requirement. Its balancesheet might look like Table A (see page 55).

Payments Activity and the Reserve-Requirement Tax. Banking-industry data used inour analysis suggest that a typical bank, in theabsence of statutory reserve requirements, wouldtend to maintain deposits at the Federal Reserve

equal to approximately 1 percent of its transac-tion deposits (in the example, $2 million). Thedata also suggest that deposit-sweeping activitydoes not affect the amount of vault cash held,relative to the sum of transaction deposits plusthe amount of deposits being reclassified asMMDA. For the example bank, the reserve-requirement tax is the interest foregone by main-taining $7 million, rather than $2 million, ondeposit at the Federal Reserve.

Overnight Repurchase Agreement–BasedDeposit Sweeping. During the 1970s, manybanks began “sweeping” customer deposits intoovernight repurchase agreements (RPs). Let ussuppose that the bank in this example wishes, atthe behest of its large business customers, tosweep half its deposits each night. To do so, itmaintains an inventory of high-quality liquidsecurities, such as Treasury bills. Its balancesheet at 3 p.m., prior to sweeping, might looklike Table B1. At 6 p.m. after sweeping, it mightappear as Table B2.

This example includes the sale (lending) of$5 million in the federal funds market; the bankis assumed to retain $2 million in deposits toservice customer accounts and reduce the risk ofan overnight overdraft at the Federal Reserve. Ifthe bank’s customers routinely desire to engagein overnight RPs, the bank likely will reduce itsbalance at the Federal Reserve and this lendingwill vanish.

1990s MMDA-Based Sweeping. Our examina-tion of banking data suggests that MMDA-basedsweeping may reduce transaction deposits at atypical bank by two thirds or more. If the bank in

HOW DEPOSIT-SWEEPING SOFTWARE REDUCES REQUIRED RESERVES

able transaction deposits, subject to several con-straints. The general parameters of this optimiza-tion problem are as follows:

• The Federal Reserve calculates a bank’srequired reserves based on a 14-day average ofthe close-of-day level of its transactiondeposits.7

• Each calendar month, an unlimited number oftransfers may be made from a customer’stransaction deposit account into the shadowMMDA. However, only six transfers may bemade out of the shadow MMDA to thecustomer’s transaction-deposit account.

• Checks presented to the bank for payment are

only debited against the customer’s transaction-deposit account, not against the MMDA. If the

7 The computation of statutory required reserves involves two legallydefined time periods: the reserve computation period and the re-serve maintenance period. The former are 14-day periods that endevery other Monday; the latter are 14-day periods that end everyother Wednesday. Prior to August 1998, a bank’s required reserves, tobe maintained during a reserve maintenance period, were based ona bank’s deposits during the reserve computation period ending2 days prior to the end of the reserve maintenance period. As ofAugust 1998, the required reserves have been based on deposits dur-ing the reserve computation period ending 30 days before the end ofthe reserve maintenance period. Required reserves must be satisfiedby eligible vault cash and deposits held during the maintenance peri-od at Federal Reserve Banks. Eligible vault cash is vault cash held bya bank during the reserve computation period ending 30 days beforethe end of the maintenance period.

(Continued on p. 54)

(Continued from p. 53)our example does so, its required reserves willdecrease by more than 80 percent, to $3.298 mil-lion ($1.248 million on the first $46.5 million ofdeposits, plus $2.050 million on the next $20.5million). Its vault cash—$10 million—is nowmore than adequate to fully satisfy its new, lowerrequired reserves. After sweeping, its balancesheet might look like Table C.

Although the bank no longer needs depositsat the Federal Reserve to satisfy reserve require-ments, we assume that the bank retains $2 mil-lion to service customer accounts and reduce therisk of an overnight overdraft. In a recentbanking-industry journal article, a seller ofMMDA-based sweep software says that an aggres-sive deposit-sweeping bank can reduce its hold-ings of reserves (vault cash and deposits at theFederal Reserve) to less than 1 percent of its totalassets.1 If our example bank reduces slightly itsholdings of vault cash, it will attain that target.

In our example, both RP-based and MMDA-based sweeps reduce to zero the “burden” ofstatutory reserve requirements—the bank holdsno more reserves than are necessary for its day-to-day operations. In addition, both types ofsweeps reduce the bank’s required reserves byenough that they are fully satisfied with vaultcash. But, the two types of sweeps differ in otheraspects. Note that RP-based sweeps constrain thebank’s balance sheet—the bank must hold aninventory of suitable liquid securities, as collater-al—but MMDA-based sweeps do not. Also, RP-based sweeps typically are conducted only withlarge business customers, often in amounts ofseveral million dollars. These customers are eco-nomically equivalent to partners with the bank inthe RP-based sweeps and hence are likely toreceive a significant share of the earnings. In con-trast, MMDA-based sweeps may be implementedfor most, if not all, transaction-deposit customersand may be invisible to the customers. Finally,MMDA-based sweeps do not directly change thebank’s total assets, liabilities, or deposits. Rather,like changes in statutory reserve requirements,they allow the bank to deploy funds from non-interest-bearing deposits at Federal ReserveBanks into loans and other investments.

The Role of Clearing Balance Contracts. Theanalysis above excludes one additional effect ofMMDA-based deposit-sweeping activity: anincrease in clearing balance contracts. A clearing

balance contract is an agreement between a bankand the Federal Reserve wherein the bank agreesto maintain a certain amount of deposits at theFederal Reserve above and beyond any amountnecessary to satisfy statutory reserve require-ments. As compensation for (and an incentive toenter into) the contract, the bank receives earn-ings credits from the Federal Reserve. Earningscredits accrue at a rate slightly less than the fed-eral funds rate and may only be used to defraythe cost of financial services, such as check clear-ing, purchased from the Federal Reserve.

Kohn (1996, p. 48) notes that, through 1996,the aggregate amount of clearing balance con-tracts had tended to increase by 16 to 17 centsfor each dollar that required reserves decreaseddue to deposit-sweep activity. Let us, therefore,reconsider our example bank. Suppose that thisbank incurs an annual cost of $200,000 due tocheck clearing and wire transfers through theFederal Reserve, on behalf of customers. In ourexample, MMDA-based deposit-sweeping softwarereduced required reserves by more than $13 mil-lion and freed the bank from using its remaining$2 million at the Fed to satisfy required reserves.If this bank were typical of Kohn’s average, itmight sign a $2 million clearing-balance contract.This clearing-balance contract does not requirethe bank to increase its deposit at the Fed beyondthe initial $2 million, nor does it infringe in anyway on the bank’s ability to use its $2 milliondeposit for routine business activity. If the federalfunds rate were to be (say) 5 percent, the bankwould receive approximately $100,000 per yearin earnings credits. The deposit-sweeping soft-ware has done double duty—it eliminated thereserve-requirement tax and, at no cost to thebank, reduced by one-half its payments to theFederal Reserve for purchased services. (Tablesshown on p. 55)


R E V I E W

1See O’Sullivan (1998). A bank consultant, quoted in this article,estimates that almost all banks with over $750 million in assetswere using deposit-sweeping software at the end of 1997 versusabout 100 banks at the end of 1996. He also estimates that even-tually bank profits likely will be increased between $1 billion and$3 billion by deposit-sweeping activity. The quoted consultant fur-ther suggests that most banks could reduce their vault-cash hold-ings by 25 to 50 percent after implementing deposit-sweepingsoftware. To us, this seems unlikely because the deposit-sweepactivity does not change the amount of deposits that the bank’scustomers perceive themselves to hold. In fact, we find that theimpact of MMDA-based deposit-sweeping activity on vault-cashratios (when the estimated amount of swept deposits is includedin the denominator) at the banks in our sample is near zero.

Assets * Liabilities *Table A: A Bank with No Sweeping Activity

Vault cash $10,000 Transaction deposits $200,000Deposits at Federal Reserve 7,000 Savings deposits 400,000Other assets 983,000 Time deposits 300,000

Other liabilities and capital 100,000Total assets 1,000,000 Total liabilities 1,000,000

Memo: Required reserves 16,598Applied vault cash 10,000Surplus vault cash 0Applied Federal Reserve deposits 6,598Excess reserves (excl. vault cash) 402

Table B1: A Bank Preparing for RP-Based Sweep

Vault cash $10,000 Transaction deposits $200,000Deposits at Federal Reserve 7,000 Savings deposits 400,000Treasury bills 100,000 Time deposits 300,000Other assets 883,000 Other liabilities and capital 100,000Total assets 1,000,000 Total liabilities 1,000,000

Memo: Required reserves 16,598Applied vault cash 10,000Surplus vault cash 0Excess reserves (excl. vault cash) 402

Table B2: A Bank After RP-Based Sweep

Vault cash $10,000 Transaction deposits $100,000Deposits at Federal Reserve 2,000 Savings deposits 400,000Treasury bills 0 Time deposits 300,000Federal funds sold 5,000 Other liabilities and capital 100,000Other assets 883,000Total assets 900,000 Total liabilities 900,000

Memo: Required reserves 6,598Applied vault cash 6,598Surplus vault cash 3,402Excess reserves (excl. vault cash) 2,000

Table C: A Bank After MMDA-Based Sweep

Vault cash $10,000 Transaction deposits $67,000Deposits at Federal Reserve 2,000 Savings deposits, including MMDA 533,000Other assets 988,000 Time deposits 300,000

Other liabilities and capital 100,000Total assets 1,000,000 Total liabilities 1,000,00

Memo: Required reserves 3,298Applied vault cash 3,298 Surplus vault cash 6,702Excess reserves (excl. vault cash) 2,000

*As of close of business; dollar amounts are in thousands.



DEPOSIT-SWEEPING SOFTWARE AND BANK BALANCE SHEETS


R E V I E W

amount of funds in the transaction deposit isinadequate, a transfer must be made from theMMDA.

• On the sixth transfer, all funds remaining inthe MMDA are moved to the transactiondeposit. (A seventh transfer would cause theMMDA to be subject to the reserverequirements applicable to transactiondeposits.)Because no debits are made to customer trans-

action deposits between just before the close ofbusiness on Friday and just before the opening ofbusiness on Monday, some early software simplyreclassified transaction deposits as shadow MMDAsprior to the close of business on Friday. Thisreduced a bank’s weekly average level of requiredreserves by 3

7 : its transaction deposit liabilities forFriday, Saturday, and Sunday, as reported to theFederal Reserve, were zero. About ten times eachyear, a Monday holiday allowed delaying the returnof funds to transaction deposits out of the MMDAuntil the opening of business on Tuesday. Latersoftware is more sophisticated and analyzes thereceipt and payment patterns of customers.8 Ofcourse, regardless of the efficiency of the software,the bank faces two additional constraints that limithow much it can reduce its reserves. It must keepon hand sufficient vault cash so as to be able toredeem customer deposits into currency, and itmust maintain sufficient deposits at FederalReserve Banks to avoid both excessive daylightoverdrafts and overdrawing its account at the endof the day.

To measure the effect of deposit-sweeping soft-ware on bank reserves, we need a benchmark, oralternative. RAM furnishes one such measurebecause it was designed to measure the changes inbank reserves caused by differences in statutoryreserve requirements—specifically, the differencesbetween those requirements in effect during thecurrent period and those for a specific benchmark,or base, period.9 The view that deposit-sweepingactivity should be analyzed as a change instatutory reserve requirements, and henceincluded within the framework of RAM, is not uni-versally held, however. The Board of Governors’staff, for example, does not publish reserve aggre-gates adjusted for the effects of deposit-sweepingactivity, apparently believing that the impact ofsuch activity is not to be interpreted aseconomically equivalent to a change in statutoryrequirements.10

In its economic aspects, deposit-sweeping soft-ware programs of the 1990s differ distinctly fromthe collateralized overnight-loan sweep programsof the 1970s—to borrow a phrase, they are not“your father’s Oldsmobile.” The business-orientedsweep programs of the 1970s essentially wereovernight collateralized loans to mutual funds andbanks, initiated by depositors (see Kohn, 1994,Chap. 9; or Stigum, 1990, Chap. 13). These loanswere made with the full participation of depositors,who received directly the lion’s share of the invest-ment return; the bank’s net earnings arose frombeing a middleman. Although such sweepsreduced banks’ required reserves, their primarypurpose was to simulate a legally prohibitedinterest-bearing demand deposit.11

The retail-oriented deposit-sweeping activity ofthe 1990s differs. First, except for competitivemarket pressures, it seems unlikely that bankshave directly passed along the earnings fromdeposit-sweeping activity to transaction-accountcustomers.12 In part, this may be due to few retaildepositors understanding the process, despitemany banks notifying customers via monthlystatement inserts (containing phrases such as“…your deposit may be reclassified for purposes ofcompliance with Federal Reserve Regulation D…”).Banks’ answers to question 12 of the FederalReserve’s May 1998 Senior Financial Officer Surveyare illustrative. On that question, banks responded

8O’Sullivan (1998) includes a description of one learning mecha-nism in recent software.

9 For further discussion of RAM, see Anderson and Rasche (1999,1996a, b) and earlier references therein.

10 Alternative measures of adjusted reserves currently are publishedby the Board of Governors of the Federal Reserve System and bythe Federal Reserve Bank of St. Louis. The measures differ withrespect to both the items included and the adjustment for changesin reserve requirements. See, for example, the annual benchmarkrelease Reserves of Depository Institutions (Division of MonetaryAffairs, Board of Governors of the Federal Reserve System).

11 We emphasize the economic effects of sweep activity. From theviewpoint of a bank manager, both RP- and MMDA-based sweepsfurnish a synthetic interest-bearing demand deposit for its cus-tomers; see, for example, Coyle (2000). Note that MMDA-basedsweeping may be very profitable for a bank if its customers areunaware of the practice and do not demand a share of the earn-ings. Some analysts have estimated that profit margins may be ashigh as 90 percent (O’Sullivan, 1998).

12 To test this hypothesis, we have examined scatter plots of bankoffering rates on other checkable deposits and time deposits, rela-tive to market yields on both short- and long-term Treasury issues.In monthly data, no change is apparent during the last decade.

that even if they were permitted to pay interest ondemand deposits and if the Fed paid interest ondeposits at Federal Reserve Banks, they most likelywould tier rates paid on demand deposits and thatthe highest rate “would still be considerably belowthe level of market interest rates.” Second, thesweeps of the 1970s required banks to maintain asignificant amount of high-quality liquid collateralfor use in repurchase agreements with largebusiness customers. The retail sweeps of the 1990sallow a bank to deploy into higher-earning assets,as it sees fit, the funds released by reducedrequired reserves. In the boxed insert “HowDeposit-Sweeping Software Reduces RequiredReserves,” for example, the bank’s earning assetsincrease with no increase in total deposits orfunding costs.

Linkages among retail deposit-sweepprograms, the Depression-era prohibition of thepayment of (explicit) interest on demand deposits,and the payment by the Fed of interest on depositsat Federal Reserve Banks have been discussed byFederal Reserve Governor Lawrence Meyer inrecent Congressional testimony.13 An importantissue is whether banks would reduce or eliminatethe use of deposit-sweeping software if the FederalReserve paid interest on reserve balances. Becausethe economic effects of deposit-sweeping softwareare similar to reductions in statutory reserverequirements, in our opinion such an outcome isunlikely. First, as noted above, it seems unlikelythat banks have passed much of the benefit from1990s-style deposit-sweeping activity on to theirtransaction-deposit customers. Second, becausenewly released funds may be invested as the banksees fit, including in consumer and business loans,it seems unlikely that deposits at Federal ReserveBanks, earning interest at the federal funds rate,would be an attractive investment. In question 10of the May 1998 Senior Financial Officer Survey,banks were asked whether they would dismantlesweep programs if the Federal Reserve paidinterest on deposits. In their summary of thesurvey, Board staff noted that “several” banks saidthat they would, or might, dismantle sweepprograms. More than half of the respondents, how-ever, said that interest paid at the federal funds ratewould be unattractive, relative to the higher returnsavailable on alternative investments. The staffsummary also notes, on page 8, that “the resultson this question seem qualitatively different from

the responses to a similar question on the May1996 Senior Financial Officer Survey. On thatsurvey, two thirds of the respondents indicated thatthey would dismantle their retail sweep programseither immediately or over time if interest werepaid on Fed account balances held to meet reserverequirements.” In our opinion, retail deposit-sweeping software is here to stay for the sameeconomic reasons that cause banks to preferdecreases, rather than increases, in statutoryreserve requirements.

Reserve-Requirement Ratios andEconomically Bound Banks

To measure the effect of deposit-sweepingsoftware on the amount of reserves held by banks,we need to separate banks wherein the quantity ofreserves demanded is sensitive to changes inreserve-requirement ratios from those in which itis not.14 When reserve requirements are “low,” adepository institution’s demand for reserves maybe largely, or even entirely, determined by its busi-ness needs (converting customer deposits into cur-rency, originating interbank wire transfers, settlinginterbank check collection debits) rather than bystatutory requirements. In the United States, thelevel of reserves held in the absence of statutory



13 See Meyer (1998, 2000). Meyer’s 2000 testimony was in regard toHouse Resolution 4209, a bill that would “require the payment ofinterest on reserves maintained at Federal reserve [sic] banks…”(106th Congress, 2nd Session, as reported with amendments onOctober 17, 2000). The text of the bill is available on the Library ofCongress’ “Thomas” legislative Web site. Although the FederalReserve does not pay explicit interest on deposits at FederalReserve Banks, banks that enter into clearing balance contracts docurrently receive at approximately the federal funds rate “earningscredits” on those deposits that are obligated under clearing balancecontracts. A clearing-balance contract is a contractual agreementbetween a depository institution and a Federal Reserve Bank. In thecontract, the depository agrees to maintain a specific amount ofdeposits at the Federal Reserve Bank above and beyond theamount, if any, necessary to satisfy its statutory reserve require-ment. In turn, the depository institution accrues earnings creditswhich may be used to defray the cost of services purchased fromthe Federal Reserve such as check clearing and wire transfers.Earnings credits may not be converted to cash and have no cashvalue except in exchange for Federal Reserves services. Penaltiesapply for entering into such a contract and subsequently not hold-ing sufficient deposits (Stevens, 1993). A bill now pending inCongress (H.R. 4209) would eliminate earnings credits in favor ofcash interest payments.

14 Note that in this analysis the term “total reserves” includes all vaultcash held by depository institutions. In Board of Governors’ publi-cations, however, reserves includes only the amount of vault cashthat is applied to satisfy statutory reserve requirements—any “sur-plus” amount is excluded.


R E V I E W

reserve requirements might be very small indeedbecause banks are permitted daylight overdraftson their deposit accounts at the Federal ReserveBanks (Emmons, 1997; Furfine, 2000). Whenstatutory reserve requirements are “high,” theamount of reserves held will be approximatelyequal to its required reserves. (This statementassumes that all base money held by depositoryinstitutions can be used to satisfy reserve require-ments. In the United States, member banks couldnot apply vault cash to satisfy reserve require-ments between 1917 and 1959.) Hence, measuringRAM requires a model of banks’ demand forreserves that includes an explicit role for statutoryrequirements.

Let us denote a depository institution’sreserve demand function as TRD(D,rr), where Ddenotes the institution’s deposit liabilities and rrthe statutory reserve-requirement ratio. Further,omitting all tiering of reserve requirements, let usdenote its required reserves as RR(D,rr) = rr D.Then, when rr is relatively large,

such that statutory reserve requirements are, at themargin, the binding constraint that determines theamount of reserves held.15 When rr is relativelysmall, we assume that

such that the bank’s business needs, rather thanstatutory requirements, are the binding constraint.In Anderson and Rasche (1996b), we introducedthe term economically nonbound to describe bankswhere

and economically bound to describe banks where

To measure RAM, we must know (or infer) thesign of the derivative

at all dates and for all banks in our sample. To bespecific, for an individual bank, let RR(Dt,rr0) andRR(Dt,rrt) and denote the period t levels of

required reserves when the statutory requirementsof a base period, 0, and of period t, respectively,are in effect. For all cases, assume that sufficientdata on reservable liabilities during period t existso as to permit calculation of the quantityRR(Dt,rr0). Then, consider four cases:

Case 1: If rr0=rrt , that is, reserve requirementshave not changed, RAM = 0.

Case 2: If

that is, if the business needs of the bank were thebinding constraint in both the base period 0 andperiod t, then RAM = 0.

Case 3: If both

that is, if the statutory requirements were abinding constraint on the bank in both the baseperiod 0 and period t, then the RAM adjustmentfor period t (conditional on the choice of period 0as the base period) is

Case 4: If

that is, if the statutory requirements were bindingin the base period but not in period t, then tomeasure RAM we must find the smallest reserve-requirement ratio, say rr*, for which

Then,

The above analysis is applicable to cases where theonly change in statutory reserve requirements be-tween two periods is the reserve-requirement ratio,rr, and data exist to calculate the counterfactual level

RAM RR D rr RR D rrt t t= ( ) − ( ), , .*0

∂ ( )∂

>==

TR D rr

rr

D

D Drr rr

t

t

,.

*

0

∂ ( )∂

>∂ ( )

∂=

==

==

TR D rr

rr

TR D rr

rr

D

D Drr rr

D

D Drr rr

tt

, ,

00

0 0but ,

RAM RR D rr RR D rrt t t t= ( ) − ( ), , .0

∂ ( )∂

>∂ ( )

∂>

==

==

TR D rr

rr

TR D rr

rr

D

D Drr rr

D

D Drr rr

tt

, ,

00

0 0and ,

∂ ( )∂

=∂ ( )

∂=

==

==

TR D rr

rr

TR D rr

rr

D

D Drr rr

D

D Drr rr

tt

, ,

00

0 0and ,

∂ ( )∂

TR D rr

rr

D ,

∂ ( )∂

>TR D rr

rr

D ,.0

∂ ( )∂

=TR D rr

rr

D ,0

∂ ( )∂

=TR D rr

rr

D ,,0

∂ ( )∂

≅∂ ( )

∂= >

TR D rr

rr

RR D rr

rrD

D , ,,0

×

15 Throughout this analysis, we assume that the response of excessreserves at economically bound banks to changes in the statutoryreserve-requirement ratio is zero (see Anderson and Rasche,1996b). Excess reserves at economically nonbound banks typicallyare positive and an inverse function of the statutory reserve-requirement ratio.



of required reserves, RR(Dt,rr0). For analysis of othercases, see Anderson and Rasche (1999).

An empirical criterion for measuring RAM inCase 4, for dates beginning November 1980, wasdeveloped by Anderson and Rasche (1996b) basedon statistical analysis of a large panel dataset. Thatanalysis suggested that a bank was economicallybound during a reserve maintenance period if thebank was legally bound and had more than $135million in net transaction deposits.

Empirical Analysis

Because the design, implementation, andoperation of sweep software is idiosyncratic, ouranalysis focuses on a longitudinal panel of 1231depository institutions between January 1991 andDecember 1999. A depository institution is includ-ed in the panel if, during at least one reserve-maintenance period, it was either economically orlegally bound.16 Our panel is a subset of a largerdataset containing more than 7500 depositoryinstitutions, which, in turn, is an updated versionof the dataset used in Anderson and Rasche(1996b). For some banks, data begin after January1991 because the bank opened for business at thatpoint, was created by the merger of existingbanks, or only then began reporting data to theFederal Reserve. For others, the data stop beforeDecember 1999 because the bank failed, mergedwith another bank, or was dropped from thereporting panel. For each such bank, we use theFederal Reserve’s bank structure database to tracepredecessors and successors. When a bank withdeposit-sweeping activity is acquired by anotherbank, we add the amount of activity at theacquired bank to the amount at the acquiringbank. In all cases, we focus special attention onthose institutions where deposit-sweeping soft-ware has reduced the level of transaction depositsto such an extent that the level of the depository’srequired reserves is less than the amount ofreserves (vault cash and deposits at the FederalReserve) that the bank requires for its ordinaryday-to-day business.

RAM 1991-93

We begin by re-examining RAM from January1991 through December 1993. Our previous meas-ure of RAM was based on the statistical models ofAnderson and Rasche (1996b). Those results sug-gested that economically bound banks (the onlyones included in that measure of RAM) were char-

acterized by two features: (i) a level of requiredreserves that exceeded their vault cash and (ii)having more than $135 million in net transactiondeposits. This framework allowed us to classifybanks into broad groups without tedious examina-tion of time-series data for individual banks.

In this analysis, we revise our measure for1991-93 for two reasons. First, because deposit-sweeping software allows banks to home-brewreserve requirements, our analysis for 1994-99must necessarily be based on the examination ofdata for individual banks. It is important to assesswhether this change in procedure—from usingaggregated data for groups of banks to usingindividual-bank data—has any effect on measuredRAM. The 1991-93 period provides an exper-imental control for this change in procedure.Second, we seek to reduce the number of occur-rences when a bank, as well as its deposits, movesfrom being included in RAM to being excluded. Itseems unlikely that a bank’s responsiveness topossible changes in a statutory reserve-requirement ratio fluctuates very much fromperiod to period. Absent changes in statutoryreserve requirements (or a merger), we assumethat a typical bank switches infrequently betweeneconomically nonbound and bound.

The most reliable indicator of a bank’seconomically bound or nonbound status is itsresponse to a change in statutory reserverequirements. An economically bound bank willreduce its holdings of reserves, following a reduc-tion in reserve requirements, by approximately thesame amount as the decrease in its requiredreserves. An economically nonbound bank willnot, although it might reduce its holdings by asmaller amount. Between January 1991 andDecember 1993, there were only two ways that thestatutory reserve-requirement ratio for a bankcould change:

• In April 1992, the Federal Reserve reduced thestatutory reserve-requirement ratio on transac-tion deposits above the low-reserve tranchefrom 12 percent to 10 percent. If a bankreduced its deposits at Federal Reserve Banks(relative to transaction deposits) following theApril 1992 reduction in reserve requirementsand did not begin or increase the size of aclearing-balance contract, we classified the

16 A bank is said to be legally bound if the amount of its requiredreserves exceeds the amount of vault cash that is legally permittedto apply to satisfy its reserve requirement.

bank as economically bound from January1991 to December 1993. If the bank began orincreased the amount of a clearing balancecontract, we classified the bank aseconomically nonbound from the date of thatincrease through the end of December 1993.An increase in the bank’s clearing balance con-tract at the time of the reduction indicated thatpayments activity, not statutory reserverequirements, had been determining the levelof reserves held by the bank.

• The second change affected only banks thatacquired another bank. Federal Reserve regula-tions permit an acquiring bank to “amortize”over eight quarters the reserve exemptionamount and low-reserve tranche of theacquired bank.17 For an acquirer with transac-tion deposits greater than the low-reservetranche, the amortization reduces theacquirer’s required reserves. If an acquirer didnot reduce its holdings of reserves so as tomatch the reduced required reserves, we clas-sified the bank as nonbound beginning in thatmaintenance period.Finally, we also classified a bank as

economically nonbound in a reserve-maintenanceperiod if it is legally nonbound (that is, if itseligible vault cash exceeds its required reserves).Because some banks alternate between legallybound and nonbound, we modify this presump-tion by judgmentally smoothing changes in status.

Figure 2 compares two measures of RAM for1991-93. One is based on our 1996 method, andthe other on the method outlined above. The twomeasures, for all practical purposes, are the same.

RAM 1994-99Deposit-sweeping activity by banks substan-

tially complicates measuring RAM for 1994-99. Tocope, we follow a three-step procedure. First, weidentify the dates (reserve-maintenance periods)affected by new or expanded deposit-sweepingactivity and estimate the amounts of transactiondeposits relabeled as MMDA. Second, we classifyeach bank, for each reserve-maintenance periodbetween January 1994 and December 1999, aseconomically bound or nonbound. This procedureis similar to our revised measure for RAM during1991-93 and relies heavily on the observedresponse of the bank to changes in reserve-requirement ratios and the effects of implement-ing its deposit-sweeping software. Finally, we cal-culate RAM based on the framework of Cases 1, 2,3, and 4 introduced above.

Sweep Dates and AmountsOur first task is to identify the dates on which

banks either began or changed their deposit-sweeping activity. Although the date of the firstsuch deposit-sweep program is known (January1994), banks are not required to notify the FederalReserve when a program is implemented, expand-ed, or discontinued; nor are they required toreport the amount of deposits affected.18 To identi-fy those dates when deposit-sweeping activityeither began or was expanded, we visually ana-lyzed time-series data for each bank. The variablesexamined were changes in the levels of transac-tion and savings deposits, changes in the size of aclearing-balance contract, and the ratios of vaultcash and deposits at Federal Reserve Banks totransaction deposits.19 For a typical bank, the data


R E V I E W

Bill

ion

s o

f Do

llars

RAM, 1991-93

1991 1992 1993-1.6

0.0

1.6

3.2

4.8

6.4

8.0

9.6

11.2

RAM, 1996 Method

RAM, 2000 Method

Figure 2

17 Under Federal Reserve Regulation D, when a bank acquires anoth-er, the required reserves of the survivor are reduced by a trancheloss adjustment. The initial value of the adjustment equals the dif-ference, during the reserve maintenance period immediately pre-ceding the merger, between the required reserves of the survivorbank (computed as if the merger had been completed) and therequired reserves of the acquired bank(s). The reduction is phasedout over eight quarters: During the first quarter, the survivor’srequired reserves are reduced by seven eigths of the adjustment,during the next quarter by three quarters of the adjustment, etc.

18 The staff of the Board of Governors maintains a database of sweepdates and amounts at individual banks, gleaned from deposit-reportdata and interviews with staff of individual banks. This databasewas not available for our research.

19 We also experimented with statistical methods, including vectorautoregressions containing transactions and savings deposits. Theidentification error rates from these methods, in our opinion, wereunacceptably high.

signature of deposit-sweeping activity consists oftwo simultaneous changes:

• The level of transaction deposits decreases andthe level of savings deposits increases, duringthe same reserve-maintenance period and byapproximately the same dollar amount, whilethe bank’s level of total deposits is approx-imately unchanged. It is important to conditionthe analysis on the level of total depositsbecause, in some cases, mergers of banks withdifferent mixtures of deposits otherwise createfalse signals.

• The ratio of vault cash to reported transactiondeposits (that is, transaction deposits notreclassified as MMDAs) increases sharply. Thismost likely occurs because the amount of vaultcash held by a bank depends on its customers’perceived amount of transaction deposits, notthe amount of reservable transaction depositsreported by the bank to the Federal Reserve.

For each so-identified maintenance period, ourestimate of the amount of deposits affected is thesmaller of the increase in savings deposits and (theabsolute value of) the decrease in transaction

deposits. For some identified periods, however,transaction deposits increased, savings depositsdecreased, and the ratio of vault cash totransaction deposits fell. We interpret thesechanges to indicate that deposit sweeping was dis-continued or reduced in amount. The amount ofthe change in deposit-sweeping activity iscalculated as the negative of the abovecalculations, but is capped at the maximumamount that we estimate the bank previously hadbeen sweeping.

ResultsOverall, we observed deposit-sweeping activity

at 680 of the 1231 banks in our panel dataset.20

Due to mergers, acquisitions, and liquidations, as of

1994 1995 1996 1997 1998 19990.5

0.6

0.7

0.8

0.9

1.0

St. Louis Sweeps / BOG Sweeps

1994 1995 1996 1997 1998 19990

50

100

150

200

250

300

350

400

St. Louis, Sample of 1231 Banks

BOG, All Banks

A. Sweeps of Transaction Deposits into MMDAs

Bill

ion

s o

f Do

llars

B. Ratio of St. Louis to Board of Governors (BOG) Staff Sweeps Estimates



Figure 3

20 For 671 banks, we have identified specific reserve-maintenanceperiods in which we believe deposit-sweeping activity began or wasexpanded. For 9 banks, deposit-sweeping activity is inferredbecause they acquired other banks with deposit-sweeping activity.Among the 671 banks, there were 425 acquisitions of banks by oth-ers between January 1994 and December 1999. (This numberincludes acquired banks that later were acquired by others amongthe 671.) Bank mergers and acquisitions are handled by assumingthat the acquirer continues to sweep the same amount of depositsas was being swept by the acquired bank.


R E V I E W

December 1999 our panel includes only 649 activebanks. Of these, we estimate that 463 banks wereoperating deposit-sweeping software, affecting$255.9 billion of transaction deposits. Figure 3Ashows estimates of the amount of deposit-sweeping activity at our panel of banks during1994-99. For comparison, the figure also showsthe Board of Governors staff’s estimate of theamount of deposit-sweeping activity at all deposi-tory institutions. As of December 1999, the Boardstaff estimate is $371.8 billion. Figure 3B showsthe ratio of the two estimates. Prior to mid-1995,the aggregate amount measured in our panel ofbanks is approximately 85 to 90 percent that of theBoard staff’s; since mid-1997, our measure hasbeen approximately 65 to 70 percent of their total.The difference between the amounts may be dueto one or more of three factors:

1. Deposit-Sweeping Programs at SmallerBanks. Our panel includes only 1231 banks,those which are relevant to measuring RAM;the Board staff’s estimate seeks to include allbanks. For our purpose—measuring the reduc-tion in total and required reserves due todeposit-sweeping activity—the difference isunlikely to be important. Our previous analy-sis (Anderson and Rasche, 1996b) suggests thatthe smaller banks omitted from our panel areunlikely to change their holdings of reserves in

response to a change in statutory reserve-requirement ratios.2. Overlooked Deposit-Sweeping Programs.Repeated re-examinations of our data suggestthat we have overlooked few, if any, bankswithin our sample that are operating deposit-sweeping programs. We have searched notonly for the implementation of deposit-sweep-ing software but also for subsequent changesin the level of sweep activity.3. Inaccurate Estimates of the Amount ofReclassified Deposits. Our estimated amountsare the smaller of the absolute values of thechange in transaction and savings deposits,subject to caveats explained above. As a fur-ther check, we visually examined two ratiosfor each bank: vault cash (VC) divided byreported (reservable) transaction deposits (NT),VC/NT, and vault cash divided by the sum ofnet transaction deposits plus the estimatedamounts of deposits reclassified as MMDA(SWP), VC/(NT+SWP). These data suggest thatour estimates of the amounts being reclassi-fied are quite accurate. The ratio VC/NT almostalways increases sharply when deposit-sweep-ing activity begins or expands. On these samedates, the ratio VC/(NT+SWP) shows no suchjumps.

Figure 4 displays total transaction deposits atthe banks in our panel. The smaller series is theamount of transaction deposits reported by ourpanel of banks to the Federal Reserve, NT, andhence subject to statutory reserve requirements.The larger series is the sum of NT plus SWP. Thedifference, of course, is deposit-sweeping activity.

E-Bound Status

We emphasize that our purpose in this analy-sis is not to estimate either the total number ofbanks using deposit-sweep software or the totalamount of deposits involved. Rather, we wish toidentify how deposit-sweeping activity at econom-ically bound banks has reduced the quantity ofreserves held during each reserve maintenanceperiod during 1994-99. The concept and calcula-tion of RAM focuses on deposits, not on banks. Itis the derived demand for reserves, arising fromthe level of deposits and the characteristics of thebanks, that is of primary interest to us. Our nextstep, therefore, is to classify, for each biweeklyreserve-maintenance period, a bank (and its

Transaction Deposits

1991 1992 1993 1994 1995 1996 1997 1998 1999300

350

400

450

500

550

600

650

Reported Deposits Plus Sweeps into MMDAs

Reported Deposits

Bill

ion

s o

f Do

llars

Figure 4



deposits) as either economically nonbound oreconomically bound. To do so, we visually ana-lyzed time-series data, for individual banks, on aperiod-by-period basis from 1994-99. Similar to1991-93, we believe that banks should not (and donot) alternate often between bound and non-bound status.

The most important indicator of the bank’sbound and nonbound status is the change in itsholdings of reserves, relative to the change in itsrequired reserves.

• If a bank acquired another bank, did it makeuse of the reduction in required reserves asprovided for by Federal Reserve regulations? Ifnot, then the acquiring bank is revealed to beeconomically nonbound during those periods.In most cases, such a bank is classified as eco-nomically nonbound in all subsequentperiods.

• If a bank implemented a sweep program, didits ratio of reserves to reported transactiondeposits (after subtracting required reservesagainst the low-reserve tranche from thenumerator and the deposit-amount of the low-reserve tranche from the denominator)increase above 10 percent? If so, the bank isrevealed to be economically nonboundbecause it holds more reserves than isnecessary to satisfy statutory reserve require-ments. In most cases, such a bank is classifiedas economically nonbound for all subsequentperiods.

• If a bank implemented a sweep program, did itincrease its required clearing-balancecontract? If so, the bank is revealed to be eco-nomically nonbound because it voluntarilyincreased its reserves above the amount neces-sary to satisfy legal requirements. In mostcases, the bank is classified as economicallynonbound for all subsequent periods unless itreduces or eliminates its clearing balance con-tract.

• If a bank implemented a sweep program, didits required reserves decrease below its vaultcash (that is, did the bank become legally non-bound)? If so, the bank is revealed to beeconomically nonbound because the amountof vault cash necessary for its ordinarybusiness exceeds its required reserves. Thebank is classified as economically nonboundfor all periods in which it is legally nonbound.

In general, if a bank is reclassified to economi-cally nonbound from economically bound, itremains nonbound through to the end of thesample. We observed, however, that a few bankssubsequently sharply reduced their excessreserves and began responding to changes inreserve requirements. Although the reasons forsuch changes in behavior are unknown to us, wereclassified these banks as economically boundbeginning at the date of the change.

Banks that neither implemented a deposit-sweep program nor acquired another bank during1994-99 experienced no change in their statutoryrequired-reserve ratio. Hence, we use different cri-teria to classify these as economically bound ornonbound. For most banks, their status as ofDecember 1993 is extended forward throughDecember 1999. A bank’s status might be changedif it significantly changes its level of excessreserves, enters into a clearing-balance contract,or experiences a major change in its level or mix-ture of deposits. In our sample, there are 551 suchbanks; 88 of these had their classification changedbetween January 1994 and December 1999.

Figure 5 shows the numbers of banks in ourpanel classified as economically bound andnonbound and the amounts of their reservabletransaction deposits. Changes in the numbers ofbanks should not be over-interpreted because ofthe large number of bank mergers andacquisitions since 1995. Regardless, the figureshows clearly that a major shift has occurred sincedeposit-sweeping computer software began tospread rapidly through the U.S. banking industry.In late 1994, for example, deposits in our panel’seconomically bound banks totaled approximately$500 billion, whereas deposits in economicallynonbound banks were less than $100 billion. Bylate 1999, reported transaction deposits (subject tostatutory reserve requirements) in economicallybound banks totaled less than $100 billion, andreported transaction deposits in economicallynonbound banks were approximately $250 billion.

Estimate rr*

The above analysis allows us to classify eachbank in our dataset, during each reserve-maintenance period, as being in Case 1, 2, 3, or 4.For those banks in Cases 1 and 2, RAM=0. Forthose in Case 3, calculation of RAM is straightfor-ward, as shown above. For banks in Case 4, itremains to estimate rr*.


R E V I E W

We remind the reader that rr* does not equalthe marginal reserve-requirement ratio againsttransaction deposits but, rather, is the smallest(counterfactual) reserve-requirement ratio forwhich

when

rr0>rrt , and rr0 and rrt are, respectively, the baseperiod and period t reserve-requirement ratios. Inthis case, RAM t = RR(Dt , rr0) −RR(Dt,rr*). Anestimate of rr* is calculated only once, for themaintenance period before the sweep activity thatallows the bank to become economically non-bound. This mimics in spirit the pre-1994 Federal Reserve statutory regime in which reserve-requirement ratios changed by specific amounts atspecific dates and then remained fixed at the newvalues until the next change. RAM is calculated for

all subsequent periods in the dataset as if thecalculated value for rr* were the applicable statu-tory ratio.

In what follows, we treat rrt and rr* as ratios ofreserves (vault cash, VC, plus deposits at FederalReserve Banks, RB) divided by the sum (NT +SWP).21 Complications regarding tiering (thereserve-exemption amount and low-reservetranche) are omitted because all aspects of thestatutory reserve requirement system, includingtiering, are irrelevant to Case 4 banks. To estimaterr*, recall that the amount of reserves held by aneconomically nonbound Case 4 bank is deter-mined by its day-to-day business needs, not bystatutory reserve requirements. Hence, the amountis less than the product of rr* times the sum

∂ ( )∂

>∂ ( )

∂=

==

==

TR D rr

rr

TR D rr

rr

D

D Drr rr

D

D Drr rr

tt

, ,

00

0 0, ,

∂ ( )∂

>==

TR D rr

rr

D

D Drr rr

t

t

,

*

0 ,

21 In this analysis, “RB” refers to total deposits held by banks atFederal Reserve Banks, including amounts held to satisfy clearing-balance contracts (line 25, Table 1.18, Federal Reserve Bulletin,August 2000). Note that this differs from the concept of reservebalances published by the Board of Governors staff, which equalsdeposits at Federal Reserve Banks minus the amount of clearing-balance contracts (line 1, Table 1.12, Federal Reserve Bulletin,August 2000). Also, “VC” includes all vault cash held by depositoryinstitutions. In Board of Governors’ publications, total reservesincludes only the amount of vault cash that is applied to satisfystatutory reserve requirements—any “surplus” amount is excluded.

1991 1992 1993 1994 1995 1996 1997 1998 19990

80

160

240

320

400

480

560

E-nonbound Banks

E-bound Banks

Deposits

1991 1992 1993 1994 1995 1996 1997 1998 19990

250

500

750

1000

E-nonbound Banks

E-bound Banks

Number of Active Banks in Panel

Bill

ion

s o

f Do

llars

Figure 5



(NT + SWP):

Note that the opposite is true for an economicallybound Case 3 bank where, at the margin, theamount of reserves is determined by the statutoryreserve-requirement ratio rrt: 22

These two relationships are not sufficient, how-ever, to provide an estimator for rr*. To do so, weimpose one additional condition: We assume thatthe amount of a bank’s vault cash is determined byits day-to-day retail business needs and is notaffected by statutory reserve requirements ordeposit-sweeping activity. Conditional on this

assumption, we examine separately the ratiosVC/(NT + SWP) and RB/(NT + SWP). From theseratios, we infer upper and lower boundaries for rr*and, thereafter, a value for rr* itself.

We begin by comparing the reserves held bybanks before and after they implemented deposit-sweeping software. Selection of the appropriate“before” and “after” reserve-maintenance periodsrequires some judgement. Our data suggest thatat many banks deposit-sweeping activity wasphased-in during a number of reserve-

VC RB rr NT SWPt+ ≥ × +( )( ) .

VC RB rr NT SWP+ < × +( )( *) .

0.0

0.1

0.2

0.3

0.4

0.5

-4 -2 0 2 4Normalized Value

After Sweep

Normal (0,1)

0.0

0.1

0.2

0.3

0.4

0.5

0.000 0.025 0.050 0.075 0.100 0.125 0.150 Value

After Sweep

Before Sweep

0.0

0.1

0.2

0.3

0.4

0.5

-4 -2 0 2 4Normalized Value

Before Sweep

Normal (0,1)

Panel B

0.000

0.025

0.050

0.075

0.100

0.125

0.150

0.000 0.025 0.050 0.075 0.100 0.125 0.150After Sweep

Panel D

Panel CPanel A

Den

sity

Den

sity

Den

sity

Bef

ore

Sw

eep

Vault Cash/(Reported Net Transaction Deposits + Sweeps)Banks E-bound Before to E-nonbound After Sweep, Obs=458

Figure 6

22 This analysis ignores the carryover provision of Federal Reserveaccounting. That provision allows a bank’s required reserves duringa reserve maintenance period to exceed the sum of its eligible vaultcash and deposits at Federal Reserve Banks so long as the deficien-cy is offset (made up) during the following period.

maintenance periods. Also, we observed somebanks increasing their sweep activity at later dates,often a year or more after the initial implemen-tation.23 For each deposit-sweeping bank, wevisually searched the data to select the first(“before”) reserve-maintenance and last (“after”)reserve-maintenance periods affected by changesin the intensity of deposit-sweeping activity—thatis, the period before sweep activity began and theperiod during which the bank’s transaction andsavings deposits later settled down to new levels ortrends. We inferred from the ratios VC/(NT + SWP)and RB/(NT + SWP) whether the economicallybound status of the bank was changed byimplementation of deposit-sweeping software. Weclassified 458 banks that were economically bound

before implementing deposit-sweeping software aseconomically nonbound after (Case 4); 155 banksthat were economically bound before as remainingbound after (Case 3); 53 banks that were economi-cally nonbound before as remaining nonboundafter; and 2 banks that were economicallynonbound before as bound after.24

Data for the 458 banks that changed status


R E V I E W

0.00

0.08

0.16

0.24

0.32

0.40

0.48

0.56

0.64

-4 -2 0 2 40.00

0.08

0.16

0.24

0.32

0.40

0.48

0.56

0.64

0.000 0.025 0.050 0.075 0.100 0.125 0.150

0.00

0.08

0.16

0.24

0.32

0.40

0.48

0.56

0.64

-4 -2 0 2 40.000

0.025

0.050

0.075

0.100

0.125

0.150

0.000 0.025 0.050 0.075 0.100 0.125 0.150

Deposits at the Federal Reserve/(Reported Net Transaction Deposits + Sweeps)Banks E-bound Before to E-nonbound After Sweep, Obs=458

Normalized Value Value

Normalized Value

Panel B

After Sweep

Panel CPanel A

Den

sity

Den

sity

Den

sity

Bef

ore

Sw

eep

Panel D

After Sweep

Normal (0,1)After Sweep

Before Sweep

Before Sweep

Normal (0,1)

Figure 7

23 The ongoing tuning and expansion of deposit-sweep programs isdiscussed in O’Sullivan (1998).

24 Note that these banks differ with respect to the number of periodsbetween the “before” and “after” dates, and the first and last peri-ods in which data were reported. Twelve of the 680 identifiedsweeping banks are omitted (the figures in the text sum to 668)because data were not available for periods before and after theimplementation of their sweep programs.

from economically bound to economically non-bound provide evidence for a lower boundary forrr*. Scatter plots and smoothed density functionsof the ratios VC/(NT + SWP) and RB/(NT + SWP)are shown in Figures 6 and 7, respectively, for thereserve-maintenance periods immediately beforeand after implementation of deposit-sweepingactivity (the normal density is included forreference).25

• For vault cash, Figure 6, the similarity of the“before” and “after” distributions is striking.The means are approximately the same inboth cases—4.5 percent for the “before” distri-bution and 4.6 percent for the “after”distribution—and the densities have similar

dispersion. The Kolmogorov-Smirnov statisticfails to reject both the normality of the two dis-tributions and their equality. For normality, thevalues of the test statistics are (458)1/2 × 0.0446= 0.954 for the “before” distribution and(458)1/2 × 0.0241 = 0.516 for the “after” distri-bution. For equality, the value of the statistic is[(458 × 458)/(458+458)]1/2 × 0.0611 = 0.925. Inall cases, the 5 percent critical value is 1.36.



0.00

0.10

0.20

0.30

0.40

0.50

-4 -2 0 2 40.00

0.10

0.20

0.30

0.40

0.50

0.000 0.025 0.050 0.075 0.100 0.125 0.150

0.00

0.10

0.20

0.30

0.40

0.50

-4 -2 0 2 40.000

0.025

0.050

0.075

0.100

0.125

0.150

0.000 0.025 0.050 0.075 0.100 0.125 0.150

After Sweep

Normal (0,1)After Sweep

Before Sweep

Before Sweep

Normal (0,1)

Normalized Value Value

Normalized Value

Panel B

After Sweep

Panel CPanel A

Den

sity

Den

sity

Den

sity

Bef

ore

Sw

eep

Panel D

Vault Cash/(Reported Net Transaction Deposits + Sweeps)Banks E-bound Before and After Sweep, Obs=155

Figure 8

25 The densities are calculated by the RATS program KERNEL.SRC,which computes a nonparametric estimate of the unconditionaldistribution using the Epanechnikov kernel. We also have exam-ined these ratios during the 5th, 10th, 15th, 20th, and 25th periodsafter the intensity of sweep activity stabilized. Those densities andscatter plots are nearly identical to the ones shown.


R E V I E W

The Jarque-Bera test also suggests normality.Values of the statistic for the “before” and“after” distributions, respectively, are 5.54 (p-value 0.0628) and 4.89 (p-value 0.0866).

• For deposits at Federal Reserve Banks, Figure7, the difference between the “before” and“after” distributions is striking.26 The meansdiffer: 4.2 percent for the “before” distributionand 1.2 percent for the “after” distribution (themedian for the “after” distribution is 0.8percent). The Kolmogorov-Smirnov statisticfails to reject normality of the “before”distribution, with a value of (458)1/2 × 0.031=0.663. Normality of the “after” distribution iseasily rejected, with a value of (458)1/2 × 0.173

= 3.70. (In all cases, the 5 percent criticalvalue is 1.36.) The Jarque-Bera statistic yieldssimilar results for normality, with values of2.03 (p-value 0.363) and 5059.7 (p-value of 0).

These distributions suggest that the mean of rr*likely is not less than 5.8 (= 4.6 + 1.2) percent.

Data for the 155 economically bound banksthat remained bound after implementing sweepprograms provide evidence in favor of an upperboundary for rr*. Scatter plots and smoothed den-sity functions of the ratios VC/(NT + SWP) and

0.00

0.08

0.16

0.24

0.32

0.40

0.48

0.56

0.64

-4 -2 0 2 40.00

0.08

0.16

0.24

0.32

0.40

0.48

0.56

0.64

0.000 0.025 0.050 0.075 0.100 0.125 0.150

0.00

0.08

0.16

0.24

0.32

0.40

0.48

0.56

0.64

-4 -2 0 2 40.000

0.025

0.050

0.075

0.100

0.125

0.150

0.000 0.025 0.050 0.075 0.100 0.125 0.150

Normalized Value

After Sweep

Normal (0,1)

Value

After Sweep

Before Sweep

Normalized Value

Before Sweep

Normal (0,1)

Panel B

After Sweep

Panel D

Panel CPanel A

Den

sity

Den

sity

Den

sity

Bef

ore

Sw

eep

Deposits at the Federal Reserve/(Reported Net Transaction Deposits + Sweeps)Banks E-bound Before and After Sweep, Obs=155

Figure 9

26Note that the inclusion of deposits at the Federal Reserve used tosatisfy required clearing balance contracts gives the distributionsthick right tails.



RB/(NT + SWP) are shown in Figures 8 and 9,respectively, for the reserve-maintenance periodsimmediately before and after implementation ofdeposit-sweeping activity.

• For the vault-cash ratio, the means of the“before” and “after” distributions are equal, at4.1 percent. Normality of the distributions is notrejected, with Kolmogorov-Smirnov teststatistics equal to 1.03 and 0.980, respectively,and equality is not rejected with a value of0.040. (In all cases, the 5 percent critical value is1.36.) The Jarque-Bera statistic also does notreject normality, with values of 2.04 (p-value0.361) and 0.719 (p-value 0.698).

• For deposits at Federal Reserve Banks, themeans of the “before” and “after” distributionsare 5.0 and 2.1 percent, respectively. Normalityof the “before” distribution is not rejected with astatistic of 1.11, but normality of the “after” dis-tribution is rejected with a value of 1.74.Normality of both distributions is rejected by theJarque-Bera statistic, however, with values of1752.4 (p-value 0) and 4187.2 (p-value 0).

These data suggest that the mean of rr* likely isnot more than 6.2 (= 4.1+ 2.1) percent.

These statistics suggest that, in the absence of

statutory reserve requirements, a typical bank inour sample likely would maintain approximately a1 percent ratio of deposits at Federal ReserveBanks (including deposits used to satisfy clearing-balance contracts) to total net transaction deposits(including any amounts reclassified as MMDA).Hence, we conclude that a reasonable estimator forrr* for a Case 4 bank is the sum of 1 percent plusthe bank’s vault-cash ratio during the reserve-maintenance period immediately before the period(or sequence of periods) during which the bankbegan (or changed the intensity of) its deposit-sweeping activity. Applying this rule to our sampleof 680 identified sweeping banks, we estimate rr*for 454 banks where, during the reserve-mainten-ance period immediately prior to beginning sweepactivity, (i) the bank is classified as economicallybound and (ii) the level of required reserves is lessthan the sum of vault cash plus 1 percent of trans-action deposits plus sweeps:

where RR denotes the bank’s required reserves, VCits vault cash, and SWP the estimated amountswept. The mean of these rr* estimates is 5.79

RRNT SWP

VCNT SWP+

<+

+τ τ

0 01. ,

1991 1992 1993 1994 1995 1996 1997 1998 1999-5

0

5

10

15

20

25

30

35

1. RAM, 2000 Method

2. 2000 Method, No rr* Adjustment

3. RAM, 1996 Method

4. 1996 Method, No rr* Adjustment

RAM, 1991-99

1.

2.

3.

4.

Bill

ion

s o

f Do

llars

Figure 10

percent, exactly the lower boundary discussedabove, with a standard deviation of 1.96.

As of December 1999, after numerous mer-gers, our panel contained 649 active banks. Ofthese, 269 were classified as Case 4, 199 as Case 3,60 as Case 2, and 121 as Case 1. The mean of theestimated rr* for the 269 banks classified as Case4 is 5.62 percent.

RAMTo illustrate the importance of our adjustments

for deposit-sweeping activity and for banks fallingbelow frictional levels of reserve demand (the rr*correction, in Case 4), four alternative RAM seriesare shown in Figure 10.27

• Our preferred measure, which includes theeffects of deposit-sweeping activity and rr*, islabeled “1. RAM, 2000 method.” This measuresuggests that bank reserves in December 1999were lower by $25.8 billion, relative to whatmight be expected in the absence of deposit-sweeping software.

• The series labeled “2.” is the same calculationas “1.” except that it ignores the rr* adjust-ment. That is, it assumes for each bank thatthe amount of reserves freed by deposit-sweeping activity equals the reduction inrequired reserves. Our analysis shows, howev-er, that deposit-sweeping software often is ableto reduce a bank’s required reserves to a level

below the reserves necessary for the bank’sday-to-day business. This measure suggeststhat banks’ required reserves in December1999 were lower by $34.1 billion, relative towhat might be expected in the absence ofsuch software.

• The series labeled “3.” is RAM according to themethod of Anderson and Rasche (1996b). Thisseries ignores deposit-sweeping activity.

• The series labeled “4.” is the same as “3.”except that it adjusts for rr*-type behavior.The very small difference between series “3.”and “4.” ($1.9 billion in December 1999)emphasizes that a correct RAM adjustmentmust include the effects of interaction betweenreductions in reserve requirements and banksrealizing that their required reserves have fall-en below the amount necessary for day-to-daybusiness.

SUMMARY AND CONCLUSIONS

This analysis has examined the extraordinary


R E V I E W

27 These estimates differ from reserve measures published by theFederal Reserve Board. In that data, total and required reserves(adjusted for changes in reserve requirements and seasonal varia-tion) both decreased from January 1994 to December 1999 by $19billion. Note that the Board’s measure does not include “surplus”vault cash, that is, vault cash held by depositories but not used tosatisfy reserve requirements. The Board’s measures also include anadjustment for the effect of changes in the low-reserve tranchebetween January 1995 and December 1999 of approximately $700million (during this period, decreases in the size of the trancheincreased required reserves).

During the 1990s, Federal Reservepublications have documented the spread ofdeposit-sweeping software through the U.S.banking industry. The July 1994 Humphrey-Hawkins Act monetary-policy report introduceddeposit-sweep programs in a single sentence. TheJuly 1995 report noted that approximately $12billion of deposits were involved in sweep activityand, as a result, that deposits at Federal ReserveBanks had decreased by about $1.2 billion. It alsoraised concern regarding an increase in federalfunds rate volatility if deposits decreased further.The July 1996 report included a special appendixon the operation of sweep programs. TheFebruary 1997 report noted that the aggregateamount of deposits affected by sweep programs

had increased to approximately $116 billion,compared with $45 billion in 1995. The July1997 report noted the introduction of deposit-sweep programs for household demand depositsand noted that some banks were increasing thesize of their clearing balance contracts whensweep programs reduced their required reserves.Subsequent reports have repeated these themes,along with an appeal that the Congress allow theFederal Reserve to pay interest on reservebalances (Meyer, 1998, 2000). In addition,deposit-sweep activity also has been highlightedin the annual reports of the Open Market Desk(see, for example, Hilton 1999, and Bennett andHilton, 1997) and in the Federal Reserve Bulletin(Edwards, 1997).

FEDERAL RESERVE PUBLICATIONS AND DEPOSIT-SWEEP PROGRAMS



unwinding of statutory reserve requirements inthe United States since January 1994. Based on thestatistical results in Anderson and Rasche (1996b),we selected a panel of 1231 banks whose demandfor reserves likely is responsive to changes instatutory reserve requirements. For these banks,we estimate that deposit-sweeping activity hasreduced required reserves in December 1999 by$34.1 billion. Adjusting for banks where the newlower level of required reserves is less than thebank’s necessary day-to-day operating balances,we estimate that deposit-sweeping activity hasreduced total bank reserves, as of December 1999,by $25.8 billion relative to the amount that wouldhave been held in the absence of such activity.

Our analysis suggests that the willingness ofbank regulators to permit use of deposit-sweepingsoftware has made statutory reserve requirementsa “voluntary constraint” for most banks. That is,with adequately intelligent software, many banksseem easily to be able to reduce their transactiondeposits by a large enough amount that the level oftheir required reserves is less than the amount ofreserves that they require for day-to-day operationof the bank. For these banks at least, the economicburden of statutory reserve requirements is zero.

REFERENCES

Anderson, Richard G. and Rasche, Robert H. “Eighty Yearsof Observations on the Adjusted Monetary Base:1918–1997.” Federal Reserve Bank of St. Louis Review,January/February 1999, 81(1), pp. 3-22.

__________ and __________. “A Revised Measure of the St.Louis Adjusted Monetary Base.” Federal Reserve Bank ofSt. Louis Review, March/April 1996a, 78(2), pp. 3-13.

__________ and __________. “Measuring the AdjustedMonetary Base in an Era of Financial Change.” FederalReserve Bank of St. Louis Review, November/December1996b, 78(6), pp. 3-37.

Bennett, Paul and Hilton, Spence. “Falling Reserve Balancesand the Federal Funds Rate.” Federal Reserve Bank ofNew York Current Issues in Economics and Finance, April1997, 3(5).

Board of Governors of the Federal Reserve System. FederalReserve Bulletin, August 2000.

__________. Division of Monetary Affairs. Reserves ofDepository Institutions, March 2000.

__________. Regulation D: Reserves of DepositoryInstitutions (12 CFR 204).

__________. May 1998 Senior Financial Officer Survey.

__________. May 1996 Senior Financial Officer Survey.

Coyle, Tom. “Managing Sweeps.” Community Banker,February 2000, pp. 27-29.

DeGroot, Morris H. Probability and Statistics. Reading, MA:Addison Wesley, 1975.

Edwards, Cheryl L. “Open Market Operations in the 1990s.”Federal Reserve Bulletin, November 1997, pp. 859-74.

Emmons, William R. “Recent Developments in WholesalePayments Systems.” Federal Reserve Bank of St. LouisReview, November/December 1997, 79(6), pp. 23-43.

Furfine, Craig H. “Interbank Payments and the Daily FederalFunds Rate.” Journal of Monetary Economics, 2000, 46(2),pp. 535-53.

Hilton, Spence. “Highlights of Domestic Open Market Oper-ations During 1988.” Federal Reserve Bulletin, April 1999,85(4), pp. 217-35.

Kohn, Donald. “Commentary [on Anderson and Rasche,1996b].” Federal Reserve Bank of St. Louis Review,November/December 1996, 78(6), pp. 45-9.

Kohn, Meir. Financial Institutions and Markets. New York:McGraw Hill, 1994.

McCallum, Bennett T. and Hargraves, Monica. “A MonetaryImpulse Measure for Medium-Term Policy Analysis.”International Monetary Fund Staff Studies for the WorldEconomic Outlook, September 1995, pp. 52-70.

Meyer, Lawrence H. “Payment of Interest on Reserves andFed Surplus.” Testimony Before the Committee onBanking and Financial Services, U.S. House ofRepresentatives, May 3, 2000. Federal Reserve Bulletin,July 2000, pp. 454-62.

__________. “The Payment of Interest on Demand Depositsand on Required Reserve Balances.” Testimony Beforethe Committee on Banking, Housing and Urban Affairs,United States Senate, March 3, 1998. Federal Reserve Bul-letin, May 1998, pp. 326-30.


R E V I E W

O’Sullivan, Orla. “Counting Cash in the Dark.” ABA BankingJournal, February 1998, pp. 86-92.

Stevens, Edward. “Required Clearing Balance Contracts.”Federal Reserve Bank of Cleveland Economic Review,1993, pp. 2-14.

Stigum, Marcia. The Money Market. 3rd Ed. Homewood, IL:Dow Jones Irwin, 1990.

1. FRED (Federal Reserve EconomicData), an electronic bulletin boardservice. You can access FRED eitherthrough the Internet’s World Wide Web or by dialing (314) 621-1824through a modem-equipped personalcomputer. FRED’s Internet address is www.stls.frb.org/fred. For a free brochure on FRED, please call (314) 444-8809.

2. Research Department, FederalReserve Bank of St. Louis, Post OfficeBox 442, St. Louis, MO 63166-0442.You can request data and programs on either disk or hard copy. Pleaseinclude the author, title, issue date,and page numbers with your request.

3. Inter-university Consortium forPolitical and Social Research(ICPSR). Member institutions canrequest data through the CDNet Orderfacility. Nonmembers should write to:ICPSR, Institute for Social Research,P.O. Box 1248, Ann Arbor, Michigan48106, or call (734) 998-9900.

All nonproprietary and nonconfidential data and programs for the articles writ-ten by Federal Reserve Bank of St. Louis staff and published in Review are avail-able to our readers and can be obtained from three sources:

www.stls.frb.org

Federal Reserve Bank of St. LouisPost Office Box 442St. Louis, Missouri 63166-0442

Documents

RESERVE BANK OF S EVIEW JANUARY EBRUARY OLUME UMBER · Michael R. Pakko Patricia S. Pollard Macroeconomics Richard G. Anderson James B. Bullard Michael J. Dueker Hui Guo Kevin L