Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Asset Substitution, Money Demand, and the Inflation Process in Brazil Author(s): Charles W. Calomiris and Ian Domowitz Source: Journal of Money, Credit and Banking, Vol. 21, No. 1 (Feb., 1989), pp. 78-89 Published by: Ohio State University Press Stable URL: http://www.jstor.org/stable/1992579 Accessed: 27-02-2016 20:32 UTC Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Wiley and Ohio State University Press are collaborating with JSTOR to digitize, preserve and extend access to Journal of Money, Credit and Banking. http://www.jstor.org This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions CHARLESW. CALOMIRIS IAN DOMOWITZ Asset Substitution,MoneyDemand, and the InflationProcessin Brazil THERELATIVE SOPHISTICATION OFFINANCIAL MARKETS in Brazil sets it apart from other developing economies. In the last twenty years, Brazilhas provedfertileground for Elnancialinnovationsin responseto high ratesof inflation, reserverequirements,and low or zero nominal interestrateceilingson conventional bankaccounts. Theseinnovationsincludethe rapidgrowthof relativelyunregulated Elnancecompanies, the emergenceof a fledglingequitiesmarket,reductionsin bank transactingcosts through computerization,and the use of bank repurchaseagreements as substitutesfor conventional deposits (Gelb et al. 1980). Primaryand secondary markets for government securities,which also provide the basis for bank repurchaseagreements,emergedin the early 1970sas well. These securitiesprovided a convenient source of funds for the government in the face of shrinkingmoney demandas they permittedagents to maintainrelativelyhigh yields and liquiditywithout resortto nonElnancialor foreignassets. Though the availability of treasurybills and indexed bonds contributedto the fallingdemandfor money, the minimum denomination and transaction costs associated with holding and trading these instrumentsensured that some agents would retain funds in zero or low-interestaccounts. At the sametime, those who mighthavetransferredfunds out of the domestic Elnancialsystem, and beyond the reach of the government, were offered relativelyattractive alternatives.Thus one can view the creation of these Thisresearchwassupportedin partby NSF grantSES 85-20097to the secondauthorandby the Centerfor UrbanAffairsand PolicyResearch,Northwestern University.TheauthorsthankPatrice RobitailleandGordonPhillipsforhelpwithdata,andtwo anonymousrefereesforcommentsgreatly improvingthe focusandexpositionof the paper. CHARLES W. CALOMIRIS is assistantprofessor of economics, Northwestern University. IANDOMOWITZ is associate professor of economics and is associated with the Centerfor UrbanAffairs and Policy Research, Northwestern University. Journal of Money, Credit,and Banking, Vol. 21, No. 1 (February 1989) Copyright i' 1989by the Ohio State University Press This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions CHARLESW. CALOMIRISAND IAN DOMOWITZ : 79 securitiesas a discriminatoryfinancial taxation policy in which the most inelastic asset demands receivethe highest taxation rates. The peculiar features of Brazil's financial system have been recognized as a stumbling block to estimatingthe demand for money and for isolating the role of money in the inflation process (Gelbet al. 1980).Below we presentevidencethat the availabilityof alternativeassetsas a sourceof financefor the governmentand a store of value for agents has had importantquantifiableeffects on the demandfor money and the relationships among government deficits, money growth, and inflation. Takingaccount of alternativeassets allows one to explain much of the observedrise in monetary velocity (see Figure 1) and the puzzling fact that Brazil, unlike other high-deficitdevelopingeconomiesexperiencinghigh inflation,shows money growth innovationsfollowing ratherthan precedingthose in inflation (Hanson 1980). We first develop a steady-state model of money-market equilibrium, which stressesthe long-term connections between governmentfinancial policy, inflation, and real asset supplies and demands. This equilibrium model is nested within a short-termdynamic model allowing deviations from equilibriumin section 2. In estimatingand testing the model we pay particularattention to problems of model specificationcommonly encounteredin the estimationof money demandequations. Wefind that-contrary to the resultsof previousstudies Brazilianmoney demand appearsresponsiveand stable. Section 3 turnsto the issue of the connectionbetween government deficits and inflation. There we argue that the peculiar pattern of intertemporal"causeand effect"between money and prices in Brazil is consistent with the predictionsof section 1. Specifically,innovations in deficitscan be seen as the 'Yorcingprocess"to which other nominal variablesrespond. 1. ESTIMATING MONEY DEMAND The estimation of money demand provides a short-termindicator of economic activityand a long-termguide to inflationtargeting.In Brazil,selectivegovernment intervention in asset markets (e.g., interest rate ceilings, time-varying inflation indexation rulesand treasurybill supply policy) along with high and changingrates of inflationmake interestratesunappealingshort-termindicators.Data on GNP is generallyconsideredpoor and is subjectto extensive revisions.From the long-term perspective,the persistenceof high and varying rates of inflation and the government'srelianceon the inflationtax furthermotivatean interestin money demand,in order to connect long-termgovernmentpolicy with the time path of inflation. Money demand estimation in the Brazilian context entails a unique set of potential advantages and pitfalls. The potential problems include an active and expanding market for repurchaseagreements backed by treasury securities, the computerizationof banking in the 1970s,and the existence of a black market for dollars. The latter provides a potential alternativeto cruzeiro holdings, while the former two imply a change in transactioncost due to improvementsin transaction technology and the availablerangeof accessiblealternativeassets. Dornbuschet al. (1983)find that the dollarblack marketin Brazilappearsto be unrelatedto currency substitution.Below we reportevidence consistent with this view. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions 80 : MONEY,CREDIT,AND BANKING 2.8 2.6 - _ 2.4 - _ 2.2 < - 2- I .8 \ ' 1974 I 1975 | 1976 . 1977 . 1978 . 1979 . 1980 1981 FIG. 1. Real Money Balances Measures the Log of the Ratio of Nominal Money to the Consumer Price Index (Sources are given in the Data Appendix.) On the positive side, high and varying rates of inflation magnify incentives for active short-termportfolio adjustment,allowing identificationof such adjustments on an empiricallevel. Moreover, Brazilianasset marketsenjoyjust the rightdegree of regulation, variety, and sophistication for the purposes of money demand estimation;moneyholdersmay choose to hold any of severalalternativeassets with observablerates of returnwhich vary independentlyof one another mainlybecause of regulation. The existing literatureon money demandin Brazilbypassessome of these unique opportunities.Indeed, many studiesignorethe existence of interest-bearingdomestic assets in Brazil, and focus instead on the expected rate of inflation. Moreover, existing models have not emphasizedthe role of the treasurybill marketin influencing money demand.1 Domestic assets in Brazilfor the period 1972-1981 may be classified usefully by the determinantsof their rates of return:cash and demand deposits that earn zero nominal interest, assets with pre-fixed interest whose principalreceivesa government-determinedrate of indexation2(retirementaccounts, governmentbonds, and passbook savings accounts), governmenttreasurybills which also serveas backing for repurchaseagreements,"billsof exchange"issued by finance companies which earn a market-determinedrate of return,3and bank time deposits which earn pre-fixed, regulatedrates of return. tSee Calomirisand Domowitz(1987)for a literaturereview. 2Duringourperiodalmostall indexedassetsreceived"monetary correction," whilea fewreceived the moregenerous"exchangecorrection." Theindexationformulaearerelatedto pastinilation,but subjectto frequentanddrasticchange.Forexample,a ceilingof 50 percentwasplacedon monetary correctionfor 1980,whenthe inflationratewasnearlytwicethat. 3 Ratesof returnon billsof exchangewerelimitedby regulation before1975.Weusetheconsumer creditratein RiodeJaneiroas ourmeasureof thereturnon billsof exchange.Thisallowsusto capture nonpecuniary "convenience" servicesorothermeansbywhichcompetingElnanceiras transferred excess proElts whichcamefrominterestrateceilings. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions CHARLESW. CALOMIRISAND IAN DOMOWITZ : 81 0.8 0.4_ SJ 0.4- J A/ A 1.2 - TJ - 1.6 N | l | 1974 1975 1976 l 1977 I | l I 1978 1979 1980 1981 FIG. 2. T-bill Velocity Measuresthe Log of the Ratio of Total Transactionin the Secondary Market for T-bills to the Public's Holdings Stock of T-bills (Sources are given in the Data Appendix.) We allow relative rates of returnas well as income and transactingcost to enter the demand for zero-interestmoney: (M/ pd wU1 /12, = /j3 u(ils i2, i3, wU4 < 0 u5 7re, y, wU6 > F), (1) Os where Mdenotes public holdings of cash and demand deposits, Pthe price level, il the rate of indexation ("monetarycorrection"),i2 the yield on bills of exchange, i3 the yield on T-bills, Re the expected rate of inflation, Yrealincome, and Fthe cost of transactingin Elnancialmarkets. Bank repurchaseagreements act as a money substitute. The growth of transactions in the secondary market for treasury bills, and the rising transactions "velocity"of treasury bills (i.e., the ratio of secondary market transactions to existing T-bill stock) are evidence for the increasing use of T-bill repurchase agreements as a substitute for cash. Figure 2 shows that T-bill velocity increased tenfold from 1972to 1981. Though repurchaseagreementsare not separablefrom other T-bill transactions,the increasein T-bill velocity at least in part reflectsthe increasingattractivenessof repurchaseagreements.Weemploy T-billvelocity( F) as a proxy for the (negativeof the) cost of transactingin these instruments.Wewrite(1) as: (M/P) = wU(ils i2, i3, 1rS YS V) wU6 < O. (2) Brazilianfinancial institutions allow some a priori identification of exogeneity and endogeneity of variableswhich enter money demand. The rate of indexation il is pre-Elxed,while the other variables are simultaneously determined with the This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions 82 : MONEY,CREDIT,AND BANKING supplies of money and other securities. The nominal supplies of money, treasury bills and bills of exchange, and the pricelevel are not predeterminedwith respectto money demand but depend crucially on deficit policy and asset demands. The budgetconstraintof the governmentrequiresthat nominaltaxes (X) net of transfers (R) and expenditures(G) be financed by the net creation of governmentliabilities (D), which consist of nonmonetary liabilities of the monetary authority (NM), outside money (C), treasurybills (p, and indexed bonds (IB): R + G-X (3) = 1vD = 1vC+ 1vT+ 1vIB + 1vNM. The government divides exogenous real debt creation among treasury bills, treasury bonds, and liabilities of the monetary authority. Often liabilities of the monetaryauthoritycorresponddirectlyto governmentloan items or pass-throughs (subsidiesto commercial banks for particularloans). Some direct loans and passthroughs are backed by bona fide loans, while others have been made with little expectation of repayment. Thus it is difficult to measure the implicit transfer accomplished through central bank programs. When the monetary authority createsliabilitiesthrough direct loans or loan pass-throughsto commercialbanks, this does not determine the composition of monetary authority liabilities. The extent to which monetary authority liabilities take the form of outside money is determinedby the relativedemandsfor various types of depositoryand nondepository accounts by the public. The key exogenous variablesset by the government, therefore, are the increase in total real debt and its composition among treasury bills, bonds, and other liabilities, not the supply of outside money. Total money demand, along with the money multiplier,determinesthe level of outside money. Finally, expected inflation is determined simultaneouslywith equilibrium real balances and real government debt. In the steady state, given exogenous real debt D = creation, p -Z,andthecondition p , we have D = 7i . AND TESTING 2. MODELESTIMATION Based on the discussionin the previoussection, we posit a long-termrelationship for desired real money balances of the form m* = k + aop + aly + a2Tre + a3il + a4i2 + aSi3 + a6v, (4) where all lowercase letters denote logarithms of variables; m* is desired real balances;p is the price level;y is income; Tre iS expected inflation;v is the logarithm of T-billvelocity,and il, i2, i3, arethe logarithmsof the interestratesdescribedin the precedingsection. The price level is included in orderto test the zero-homogeneity assumption (aO= 0) This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions CHARLESW. CALOMIRISAND IAN DOMOWITZ : 83 Based on (4), the error-correctionrule for real balances iS4 Amt = ,Bo+ ,BlApt + 25Tt + 7(m-y)t-l + 12Vt-l + + + 8Tt-1 3tilst + + 9ilst-l 4ti2,t + + loi2st-l 5ti3,t + + 6AVt lli3st-l (t (S) in which we have imposed long-term price homogeneity (aO= 0) and a long-term unit income elasticity(a1 = 1).The validityofthese restrictionsis testablesimplyby adding the terms a1p,-1 and a2y,-1 to the right-hand side of (5) and testing the hypotheses a1 = 0 and a2 = 0. Our data are monthly, extending from March 1972 to the end of 1981. A complete descriptionof data sourcesis given in a Data Appendix,availablefrom the authors. The choice of estimationperiod is dictatedto some extent by our interestin the role of repurchaseagreementsin treasurybills. Prior to 1972 such repurchase agreements were not very common. In recent years the growth of repurchase agreementsin government bonds as well as bills makes treasurybill velocity a less appealing proxy for repurchaseagreements. It also is the case that measuringthe rate of indexation has become more difficult due to the growth in the number of assets receiving"exchange"correction instead of "monetary"correction. We have been unable to gather the right kind of information to sort out the potential confusion over the types of correctionsin the most recent time periods. Estimates of the coefficients in the money-demand relation (5) are reported in Table 1. Instrumental variables estimation with respect to expected inflation is common to both sets of estimates reported.5Version II estimates, however, are correctedfor the potential endogeneityof the yields on bills of exchange and T-bills and T-bill velocity, as suggestedabove.6 A series of diagnostic tests of model specificationare given in Table 2. Tests of nonconstant residualvarianceare mixed. There is no evidence of dynamic ARCH effects (Engle 1982),but a White (1980) test for heteroskedasticityoverwhelmingly rejectsthe null of constant variance.As a consequence,all standarderrorsin Table 1 are corrected for general forms of heteroskedasticity,in order to draw proper inferences.The heteroskedasticity-robusttest of Domowitz and Hakkio (1985)fails to reject the null of no serial correlation in the regressionerrors, confirming the 4SeeDomowitzand Elbadawi(1987)and Domowitzand Hakkio(1986)for derivationsbasedon single-period andexpectedmultiperiodloss functions,respectively. Thepreciserelationship between the long-termcoefficientsin (4)andthemodelcoefficientsin (5)is givenin CalomirisandDomowitz (1987). 5 Allvariables weresubjectedto twelfth-differencing priorto estimationin orderto removea strong stochasticseasonal.Seasonaldummieswerenot sufficient.The instrumental variablesfor expected inflationincludedlags of the inflationrate,of oil priceinflation,of (blackmarket)exchangerate depreciation, andof nominalT-bondgrowth,in additionto a quadratictimetrend. 6Thesevariablesareinstrumented usingfourlagseachof i2, i3, and v, as wellas fourlagsof the exchangeratedepreciation andoil priceinflation. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions TABLE 1 MONEY DEMAND ESTIMATES 1 11 constant 0.078 (0.048) 0.072 (0.050) Ap - 0.159 (0.332) - 0.094 (0.368) re 0.946 (0.374) 0.963 (0.426) Ail - 0.034 (0.014) - 0.027 (0.017) Ai2 - 0.067 (0.035) - 0.056 (0.087) Ai3 - 0.048 (0.035) 0.039 (0.175) -0.012 (0.011) 0.016 (0.039) A A V (m - y) l - 0.130 (0.037) - 0.124 (0037) -1 - 2.019 (0.597) - 2.116 (0.600) il -t - 0.007 - 0.006 (0.010) (0.009) 0.036 (0.033) - 0.028 (0034) 0.034 (0.026) - 0.033 (0.027) V-l - 0.024 (0.012) - 0.018 (0.013) R2 0.463 0.391 SEE 0.026 0.027 i2 -I - i3,-I - ' NOTE: Dependent variableis the change in the log of realbalances, with a standarddeviation of 0.035. Estimatesbased on monthlydata, from March 1972through December 1981. Heteroskedasticity-robuststandarderrorsare in parentheses.VersionI estimatesare obtained from ordinaryleast squares; Version 11estimates are based on instrumentalvariables procedures. TABLE 2 MODEL DIAGNOSTICS: MARGINAL SIGNIFICANCE LEVELS IN VERSIONS 1 Heteroskedasticity (White) Heteroskedasticity (ARCH) Serial correlation Hausman statistic Chow statistic Zero exchange rate effect Price homogeneity Ullit income elasticity 0.003 0.84 0.34 0.12 0.45 0.94 0.31 0.22 I AND II 11 0.001 0.80 0.46 0.56 0.77 0.97 0.96 This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions CHARLESW. CALOMIRISAND IAN DOMOWITZ : 85 appropriatenessof the dynamic specification.7One cannot rejectthe stabilityof the model parametersat any reasonable level of statistical significance, based on a Chow (1960) test. A Hausman(1978)statisticis used to examine the potentialendogeneityof T-bill yield and velocity, as well as the yield on bills of exchange. The marginalsignificance level associated with a test of the equivalenceof parameterestimatesacrossversions I and II is 0.12, and we would fail to rejectthe exogeneity of these variablesat the 10 percent significancelevel. A closer examination of the components of the statistic suggest that it is the exogeneity of the T-bill rate that may be questionable. T-bill velocity is clearly exogenous on statisticalgrounds. The price homogeneity and long-term unit income elasticityrestrictionscannot be rejectedat reasonablelevels of statisticalsignificance.Point estimates of unconstrainedincome and price elasticitiesare 0.93 and 0.97, respectively,based on the version I results.Instrumentalvariablesestimatesare 0.90 and 0.95 for income and price effects. The point estimates suggest that any violation of the unit restrictions also is unimportantin economic, as well as statistical,terms. These resultsare quite similar to the income elasticities reported in Blejer (1978), Vinals and van Beek (1979), Cardoso (1983), and Khan (1979, 1980). Leiderman(1980) reportsincome elasticitiestwice as large. We also impose a zero restrictionwith respectto the effect of exchange rates on the money demand function, as suggested by our analysis of target balances in section 1. Dornbusch et al. (1983) arguethat foreign currencysubstitutionis not an important element in money demand.8In order to test this hypothesis we add the rate of exchange rate depreciationas an opportunitycost to the set of interestrates in the model. As Table 2 shows, the hypothesis of zero explanatory contribution cannot be rejected. Point estimates of exchange rate effects are quite small in magnitudeas well. Blejer(1978)found a largeand significanteffect of exchange rate depreciationin his study of Brazilianmoney demand. Blejer'suse of annualdata and his assumption of adaptive inflation expectations may explain the difference between his finding and ours. If currency depreciation predicts inflation (as our rational expectations forecasts of monthly inflation indicate it does), then by constrainingannual expectations of inflation to depend only on the past rate of inflation,Blejermay have misinterpretedthe indirectrole of exchangeratedepreciation (an inflation forecaster)as a direct currency-substitutioneffect. The coefficients reported in Table 1 are not directly interpretablesince they representcombinations of structuralcoefficients, including long-term elasticities and relativeadjustmentcosts. We simply note that the estimatedcoefficientsare of the expected sign, with the exception of those on the change in the T-billrateand the change in T-bill velocity in version II, which are estimated with standard errors 7StandardLagrangemultipliertestsfor serialcorrelationresultedin evenhighermarginalsignificancelevels.All otherstatisticsreportedsubsequentlyhavebeencorrectedfor non-constanterror variances. 8Empiricalevidenceon this point is given in Domowitzand Hakkio(1986)for industrialized countriesandin DomowitzandElbadawi(1987)fora countrywithrelativelycrudefinancialmarkets. Thesereferences all supporttheviewthatforeigncurrencysubstitution is notimportantformodeling domesticmoneydemand. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions 86 : MONEY,CREDIT,AND BANKING TABLE 3 EQuATIoN, IN REALMoNEY-DEMAND COEFFICIENTS LoNG-TERM LAGS ANDMEANANDMEDIANADJUSTMENT version I II Variable IndustrialProduction Index (Y) IndustrialProduction Index I Expected Monthly Inflation (re) II Expected Monthly Inflation Long-Term Coefficient la la .155 (0.037) -0.171 MeanAdj.Lag 8.5 months 8.8 months Median Adj.Lag 2.8 months 2.7 months 7.1 months 2.7 months 7.5 months 2.7 months 1.9 months 1.4 months 2.2 months 1.5 months (0.040) I II I II I II Monthly Rate of Indexation (il) Monthly Rate of Indexation .054 (0.072) -0.048 (0.066) Monthly Bill of Exchange Rate (i2) -0.277 4.8 months 2.3 months Monthly Bill of Exchange Rate (Instrumented) (0.270) d.226 (0.302) 5.1 months 2.3 months 5.3 months 2.4 months 8.2 months 2.8 months Annual T-Bill Yield (i3) Annual T-Bill Yield (Instrumented) d.262 (0.153) -0.266 (O. 160) 2.7 months 7.2 months -0.185 (0.078) 2.8 months 8.0 months d.145 T-Bill Velocity (Instrumented) II (0.094) inflation. except expected in logarithms, areexpressed NOTE: Allvariables fromunityatthelo percent different were0.93ando.so,insignificantly estimates tobeunity.Unrestricted a Coefficients arerestricted level. significance I T-Bill Velocity (v) exceeding the coefficients. A detailed derivation of the long-termelasticities the coefficientsin equation (4)- and the mean and medianadjustmentlags for disturbances to each of the argumentsin equation (4) is given in Calomirisand Domowitz (1987). These are reportedin Table 3. Our long-term elasticities with respect to expected inflation are smaller than many of those reported in other studies. For example, Blejer (1978) reports a long-termcoefficientof-0.35, comparedto our estimatesof-0. 16and-0. 17.This may be due to the inclusion of asset yields and the use of forecastedinflation rather than lagged or actual inflation. Withrespectto the dynamics,previousstudies'mean and medianadjustmentlags have been large relative to those reported in Table 3. This is because estimates of speeds of adjustmenthistoricallyhave been based on partialadjustmentmodels (see Hendry 1980for discussion).In contrast,mean lag estimateshereare quitelow, with adjustmentlags ranging from 8.5 months for an income shock to 2 months for a shock to the rate of indexation.The lag distributionis skewed,however,and median lags are much shorter. Fifty percent of the adjustmentto an income shock takes place within less than 3 months, for example. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions CHARLESW. CALOMIRISAND IAN DOMOWITZ : 87 3. EVIDENCE OF THE ENDOGENEITY OF NOMINAL VARIABLES TO DEFICITS Previous studies of the relationshipbetween money and inflation (e.g., Hanson 1980)document a unique feature of the Brazilianeconomy. Changes in money do not predict changes in the price level, but changes in the price level do predict changes in money. This finding and the large increasesin the growth rate of prices relative to money (rising velocity) have led the Brazilianauthorities to claim that "cost-push"influences (like the oil price rises of the 1970s)have been the source of high Brazilianinflation. Ourmodel is consistentwith a laggingresponseof nominalbalancesto changesin prices, and the rising velocity of money, but our interpretationof the source of Brazilian inflation is very different from the cost-push view. Nominal money in Brazilis endogenous to real asset demands and nominal governmentdeElcits.Real money demand has been stable and responsive.Rising deficitsare mainly to blame for secular rises in money, interest rates, prices, and exchange rates; thus money should not predict or cause changes in other nominal variables,as it would if the money stock were supplyZetermined. In Table 4 we report the results of a VAR (Vector Autoregressive)model for Brazil, which includes money, prices, the bill-ofwxchangerate, the industrialproduction index, and public holdings of T-bills and indexed bonds. Innovations in nominal government debt aggregates serve as proxies for innovations in government deElcits. The exogenous variablesare time, time squared,eleven monthly dummy variables, a constant term, and lagged oil prices. All others are lagged endogenous variables. Five-month lag structureswere used in the estimations. Treasurybills, TABLE VAR 4 RESULTS F-Tests for Inclusion of Lags Dependent Variables: Lagged Values of: T IB E i2 Y P M T 0.00 0.24 0.96 0.52 0.42 0.44 0.11 IB 0.90 0.00 0.89 0.21 0.90 0.55 0.50 E 0.85 0.01 0.01 0.25 0.21 0.28 0.38 t2 0.09 0.35 0.54 0.01 0.87 0.04 0.08 Y 0.33 0.29 0.76 0.99 0.03 0.76 0.51 P 0.48 0.05 0.10 0.71 0.04 0.00 0.47 M 0.21 0.00 0.01 0.00 0.06 0.01 0.00 Forecast VarianceDecompositlon (2Smonth horizon): Forecast Varianceof: T IB E i2 Y P M T IB E i2 36.4 3.5 2.1 6.1 12.4 2.0 17.5 10.5 53.9 55.4 14.5 10.4 35.2 18.5 17.6 19.7 29.0 11.0 16.3 10.4 6.3 12.4 5.4 1.9 41.2 6.1 9.7 25.3 Y 3.8 6.1 3.6 5.6 44.0 7.5 16.1 P M 7.2 5.3 6.6 10.8 5.0 27.1 3.1 12.1 6.2 1.5 10.8 5.9 8.0 13.3 NoTE: Variablesare as defined in the text. All variablesexcept i2 are expressed in logarithms. Five monthly lags of each endogenous variables are included in the estimation equations, as well as five lags of oil prices, monthly dummies and a quadratic time trend. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions 88 : MONEY,CREDIT,AND BANKING whichplay an importantrole in repurchaseagreements,are separatedfrom treasury bonds. All variablesexcept the bill-ofwxchange rate are in log levels. Table 4 gives an account of the avenues of predictionin the model. We Elndthat money does not predict prices, while bonds are one of the strongest predictorsof prices. The bond supply is also a signiElcantpredictorof the money supply and the exchange rate. At the same time, no other variableis important in predictingbonds. One way to measurethe importanceof the effects of each variablein the system on every other variableis the decomposition of forecastvariance.This describesthe percentageof a variable'sforecastvariance,over increasingtime horizons, that can be attributed to innovations from each variable in the system. In order to run simulations of the system's responses to shocks and derive forecast-variance decompositions,it is necessaryto ordercontemporaneousshocks when orthogonalizing the system. We order the variablesfrom most "exogenous"to most "endogenous" as follows: T, IB, E, i2, Y, P, M.9 E is the black-marketexchange rate. Table 4 illustratesthe importance of public debt for E, i2, P, and M. Shocks to bonds produce positive responses to all nominal variables. Bond innovations account for 35 percent of the long-term (20 month) forecast variance of the price level and 55 percent of the exchange rate, while money innovationsaccount for 1.5 percentand 8 percent of the respectiveforecast variancesof the exchange rate and price level. It is interestingto note that price innovations are neitherstatisticallysigniElcant nor economically important in accounting for changes in bonds. This may seem surprising,given that bonds are indexed partiallyto inflation.This resultreflectsthe fact that past levels of bonds incorporate previous indexation adjustments. The importance of exchange rate innovations for bonds and the price level reflects expectations of inflation, and consequently,indexation. These resultsare consistent with our model in which governmentdeElcitpolicies and real asset demand functions together determine nominal money balances, which adjust with a lag to innovations in deElcits,interestrates and inflation. The central role of deElcitsin generatinginflation may help explain the recent collapse of the "CruzadoPlan."In the absenceof consistentreductionsin the deElcit 'Yorcingprocess,"pricecontrols alone will not be successfulfor reducinginflationin the long term. Only if government spending is reduced and direct taxation is increasedwill Brazil be able to stem its inflationarytide. LITERATURE CITED Blejer, Mario. "Black Market Exchange-Rate Expectation and the Domestic Demand for Money." Journal of Monetary Economics 4 (1978), 767-73. Calomiris,CharlesW., and Ian Domowitz. "Inflation,TreasuryBill 'Velocity,'and Dynamic Money Demand in an Economy with Growing Deficits and an Endogenous Money Supply: The Case of Brazil, 1972-1981." Manuscript,Northwestern University, 1987. 9The resultsare quite robust to variations in the ordering. In particular,when money was switched from last to first, the only important difference in the simulation results was the percentage of each other's forecast variance which Af and i2 impulses explain. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions CHARLESW. CALOMIRISAND IAN DOMOWITZ : 89 Cardoso, Eliana. "A Money Demand Equation for Brazil." Journal of Development Economics 12 (February 1983), 183-93. Chow, Gregory C. "Testsof Equality between Sets of Coefficients in Two Linear Regressions." Econometrica 28 (July 1960), 591-605. Domowitz, Ian, and Ibrahim Elbadawi. "An Error-Correction Approach to Money Demand: The Case of Sudan." Journal oJnDevelopmentEconomics 26 (1987), 257-75. Domowitz, Ian, and Craig S. Hakkio. "Testingfor Serial Correlationand Common Factor Dynamics in the Presenceof Heteroskedasticity."Manuscript,NorthwesternUniversity, 1985. ."Error Correction, Forward-Looking Behavior, and Dynamic International Money Demand." Manuscript, Northwestern University, 1986. Dornbusch, Rudiger, et al. "The Black Marketfor Dollars in Brazil."QuarterlyJournal of Economics 98 (February 1983), 25X0. Engle, Robert F. "AutoregressiveConditional Heteroscedasticitywith Estimates of the Varianceof United Kingdom Inflation."Econometrica 50 (July 1982), 987-1007. Gelb, Alan H., et al. Brazil:Financial Systems Review. WorldBank Report No. 2790a, 1980. Hanson, James A. "The Short-Run Relation between Growth and Inflation in Latin America:A Quasi-Rationalor Consistent ExpectationsApproach."American Economic Review 70 (December 1980), 972-89. Hausman, Jerry A. "Specification Tests in Econometrica."Econometrica 46 (November 1978), 1251-72. Hendry, David F. "The Simple Analytics of Single Dynamic Econometric Equations." Manuscript, London School of Economics, 1980. Khan, Mohsin S. "MonetaryShocks and the Dynamics of Inflation."InternationalMonetary Fund Workingpaper, 1979. . "Formulating Monetary Targetsin a Dynamic Setting." International Monetary Fund Working paper, 1980. Leiderman,Leonardo. "TheDemand for Money under Rational Expectations of Inflation: FIML Estimates for Brazil."Manuscript, 1980. Vinals, Jose, and Fritz van Beek. "The Demand for Money in Latin American Countries, 1964-78." International Monetary Fund Workingpaper, 1979. White, Halbert. "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity."Econometrica 48 (May 1980), 817-38. This content downloaded from 128.59.83.236 on Sat, 27 Feb 2016 20:32:01 UTC All use subject to JSTOR Terms and Conditions